lkml.org 
[lkml]   [2015]   [Dec]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
Date
This commit adds RCU requirements as published in a 2015 LWN series.
Bringing these requirements in-tree allows them to be updated as changes
are discovered.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
.../RCU/Design/Requirements/2013-08-is-it-dead.png | Bin 0 -> 100825 bytes
.../Design/Requirements/GPpartitionReaders1.svg | 374 +++
.../RCU/Design/Requirements/RCUApplicability.svg | 237 ++
.../Design/Requirements/ReadersPartitionGP1.svg | 639 +++++
.../RCU/Design/Requirements/Requirements.html | 2799 ++++++++++++++++++++
.../RCU/Design/Requirements/Requirements.htmlx | 2643 ++++++++++++++++++
Documentation/RCU/Design/htmlqqz.sh | 108 +
7 files changed, 6800 insertions(+)
create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx
create mode 100755 Documentation/RCU/Design/htmlqqz.sh

diff --git a/Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png b/Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
new file mode 100644
index 0000000000000000000000000000000000000000..7496a55e4e7b41becdb658a2bd34765fbe80013f
GIT binary patch
literal 100825
zcmaI8c|2C@8#TTWl}t}j$kbCx3K5Zr>M<lKr3@7^WS)o2o(2+1GBhYeGG+=PQ>lnb
z5(*hAnadoRertEW@8|b<|9Ly7Pv__v_P+1yzOHqxwXS{jsE!6ZD?cki5bOsvX~zhH
z$%G&n9aga5CvPs*`Qi@-w__UniR||R1Na{nvqKs*Vu}1Gu{<>jKUwLlY3PQxx01gZ
z_KU@N;fKrJ4<1%q{*{%ohGTi>PcaUH5FrlI_UU=G4SzlFclgiJ^3m2e#`{@TYrXrd
zCi=iA#kMd-Ho7gN&_g6+I_)pN#K@6M>l2Up)zT$SjzqR}Zcwik%8!cc`13nLA@xt)
z;Y}2aviOJ^|6<4YehF?)@3$qmsR%REXmrLzMZ$@gVkl;GX6c@+t=zSphv)bJVMfpx
zLUzr_^H+HgTNv)K+<1A6LeceHcbcX*Z@^^D-pxaFh9-Gk)jW87jggU2l3w~TXIIyY
zj~`2H(|EPhJ0*4F=FKd+y1F-S-@e`a`@JmDC(M!;MR+n6Hw;oJl<IZ5V}6zLOdk&E
zMrQ|~%{*P=`fXq!-OI-(NjF(N?dSu>d-v{j|M;=Y*Vp&t^;)`4>H&9mSu-;;E+L`o
zPpleVbae?YE-n_k53K&!)YN=%ikN0FSZTVEu5dCUBxIdAO|PJwFilGmQ>Oh;Bb;4a
zB7c2-w#}if;6iHZmX5Dq4MmguZrr{dveW8)cYkBDLAj6I;=-S<aFe3Yo7<^z4D)QJ
zQrpyMFGtmAYKK-5CgPzo$7uFhSy>McT;b44*y}&qa?0uX$B#F8s6KU)#$_Qq+fJJH
z*2NT`{~iA~UzpxVywBfH!>^z?X8MNEq{|FkTefUjkni%_h(*~?YtZoVDIM-8InFQd
zxSd$WC+M}E`DnbbA&vHG`KgMK(AZl<7W2`jx;m}BbHh6=JBnSzPGoRzJD{hUb?Pa{
z_t)2|udSC)6w|sH63#rTyzKx@Z}QHQWi&N)eKp-8M(5v+$;)>+v^^|aTAa(xJNtW-
z#(eO;qQ?%4TCU>a;_jKT?`z+@dE>;8#%j7;#z%V-HHjK3VL}T^;G+)^uiHL<uKo5Z
zET*&E_lY<ysOnvud}E%u#_3{b24bF}Y`I8C%m%923^nOu4L{!*gUT}x_A`Geu!$NU
zua_d<pt*lP<KO9#F!uGkRCS}BDc-B9RI=D8x{cNC99LPNCNQ`=be8R&9{!T7mwECU
z&jxV|EbQpw-0$iaFBs>hN1o_&r!djG7+yY6q|w++XnH&RSWWF0{!UqTRrs^+X%-PL
z<5qh<lv8>A(<eOvMfbF2bp1$a1vkaa1OIlcLI04N?#a^PVljN!ajjdD??SnFg5*8x
zV-@`W%4Qpv|7E1R5{qUhLPA100{X(zMqQ3__r~72bLUAyLc;j#^-j4|nwqAcpOUe6
zFHebc&qc<>J#@J&R@{Q=>(yL5icD%RU$wU0dLEXx!qwTiYOpzzM@^S7-M(!<P0xL_
z=s?noSI2at3tL)RL|Y!8O8wm4{sf<R^p1?>r*_VIMun9k^b;GX;`5)!)%1!dxg*CH
z=Kko{MQ^RP<>EPO5-^+c*PShMA5BkobPsl=-q6i$;#T&VXTjnpGIIH8Y8~sH$}TW1
z=OC);2_1YgEoeQ0hLv+Y;jv!MR&V>ITgEw3(_09;qx0?~Z72K(U)gd|r2Qz3Pmc4@
z4W|a1S9Ju7ii)27l5cs$VHT0#)f=Oq8@ROCSDFwI5D;?ps<-5j&cTEGnjRg_vuY5k
zSQARQ^vBoZ<fV|1p0JR3K~5s=<YS`??YWJ~8p$S=0R~StJ1Lu)6gzKdPe921R`4F*
z<TEum)iFJWSo0cs{PaWm@zm7(`I=_aucaOeF5i~<Ob@>~zhByHG~kb^Tu2CSMM(db
zbPnDH(Xy&y|G8a+h=hbh>x86X{xXEo%lG&9Jh}H10Z9;h_B`t=ium*E>j57h#b2G}
zalh-0i|qOMWv%ItPd#1R&;Q|xex8|{!KqVrzXPfN;?$~Ikf@8<rr*75yR<ZgFtxLb
z8}hblh`&})AS<xfPoduCMRhd;A*vN!+cf>|{+@MReiQH3_*)Iy1a#RVPBo7lsd;eV
ziqI*h<;$02F`A@ocJ7RpIJGu3sTli#bHjW{RnJ<C7<KM$X5_1|x3?d6cbQc<vRdip
z&6_)HKZy~leedt<mbmsRjU<OCHLn`3S-EPJZxW#uG;=K{*W8Rs4VI!NSqW5(uU6W=
zeS3FL4+FlcxU}@Aw{XI_`dcL>-q-D_Vx_46UT8K`*W_?z_evp7A*EWk(cFT9m%a73
z9<&afJsbHaqO9-hIM4d^aeagq5)gTQXb&GgBnTX`c&*Wph={<X?z894#U@W#Xrwk-
z`dJ(OSgJXv!<c5oROE-;%ws(4d?_T}0Lx1cKlvz?qO^MT>Uc4~J4W%PZ|7Iswyigb
zR|!ejM@U?ER}@T2N(wh##m?URUZ09PxP~+m?bNeP{_%eE*Wu4EE$963#Oeqe?+zj|
zGV->JC9779I5SdVwD<{DK}C1^ojZn)P9|KwyliH6_AHL~=gL48Y(`Ii{|YB3r?=P(
z%erW3qt-W3TXdf~p85Rc%Z3db7(9krf>KgM7iar~$$WTA^349DN4b#K$9_Hg{cWn%
zq@h46Jw2UB^qhM<HZgH2B7)PlHT&9{Ejm9k9L;O4tD%^bdX9v0s?2jC9~-fM(zwDQ
zXzJ(~zm>ju^JcS#c;(MeorI@H+OOp1?od!ve9_dj#;xytP+VNx8xOpKtUswo5?+6K
zV%0M+a5Y}pH_777%_Dw7_jbFnWS)HdqTKgSV$;BfVi$Aw;m^|TJ(T?i4v<IbbcG+^
z*|TTUOe*)5m6!MQ_Qv!oF)%PFc#VoknpA{iA;=nN)AT6K@X4b_cL}3{Dr9<Re8%ZA
zFN%r(6pv24($$YChmwyZ?7h9yiet|c5d;M_s)9*m?$7UA7hg9wpCD?RGxgnM7jXT|
zbm9hX_-aV)<Umtl>!+HlYqrp6FHu;CTB`*A&DbQ$mMy8;O&JmSmi1bGeq}}3I7X~Q
zI=(+7<mKzv_)p{VQdygiFMr7J%h_&PxpJj?yVChzrx`?Y%Em*d+mQVV9Wo8_%(88Z
zeE&E)_+u$oQz8wYoU?3;-KcRrBqXG~yqvhvQ+=hr*ya1nFJE@o^CPsia7-R~WSdkP
z<80ot@i}$s)Ey;n*Z7DfTzzk0tMSefiO{6l#C@0YJA~G)TNgFq@w-#GN&3Uk)cDBL
zuF%55!Z%aew|DFV0w^3n;@d<#JoVJk454ikKhTtV=<MjXFsaZau?O=jHX+%1wQry(
z=XDhMsnKXvZEgH0*VSL2op<?BOL_R{QIczK9dN=0gjL4L#~Fp4{qBm}PZx9flnXsM
z_ijG|hNyb?j{On~`-_)X*XU-Rlpy#dwr1+*w6vrY-A$iJyMNKk*LS11IIE|p=jimb
zIu3l{+Pi^^v&;SGrsdksq4=S`eaO_8a{t{ai0I6H`m{zka6U5gu9~{~3tZ5zU%ytc
zvc8DisBz=oy>R4uYq>PN^qae~<8)H15vs(YLx<kpmOk+8ezx`<8D0fvk<B{yLKHm)
zt*6eExTZAz-McV(+<o99^Pk^cHT6=J;beGMR8&}Z6ieTXijwvi)bpF^xl%INEG;U^
z9JWsCHC~i;t?8VlWf0JUDL%WuDV1}0WF*hJky^I+cSuu5r{(9@a$osU_koMGwVEho
z$=KKak11hAoVa)OXPzVBii&sSorKh{@drgma|2!^p2$4e|0#WqxPCSp<vGGm+P(j%
zgHNHTw(|15^P>mszn1c#cr7Ee@5-)R&LPmf{x(7+L}qKQgo)^(SZ=~~sO7Q1;hff?
z!oGKR<<<}~jvWfgHlNnAl^aVDn-mp|_Abr4F^Cu&8|V8>*&aN6xcO0~`@qeXq{hm?
zz~q*##J8V6AC;f8vDtbeLoc#HW$E;<juP{N&Y)H6Z0lFDvVJs9P;f~?fjCqjD{;-J
zbMRC8#&wdXYdzkv*^Q$5uOMa-INB|$w3i(nr(e8y@v^m5-|qgYLfaT5QtPJE&pZ#e
zXQLR|<p}TfGvHQ>TrH|q`}*4YoeuTLc(2~QJ51Exm9uj#<HUVN+csys3Fq6nj9y^Z
zs)KNMls!zNUH+7D;t)}rnh=mt-ia*1AcAm=+NPh)`|gSRz%d7vfQ2c23gv1<#4Q=i
zI_o(hAt6V-dX!r`3DcSg;dPYAt;Zj&Ado1oWtmiVTK{$YT6#4oC<x{3Hc~Kt>3O}l
z#jnoD#N?*zgatj!!WpBJx`s$<8n8e~%eQS7Pwr0A!UjuzW_Mzr2l{2E7e9Mum$O?@
zk^k$jTg?v9!TT(K)JA^9s*a6~MV_q=zIy4>CBLa=eW3c8ms7rTBTt#>;k?q+dK*I1
z%q+<dKgjo+b&ZnyFyP)c;(;A%8#)}fi-#CTq+68CcLmHHCNzQM+{(TJkAA$^`8m)0
zuwxYeA`ya*-Xtpe=2h6bLUTF0mMr^*uv^AH2WhmH>@%6~o;bAcikdS9(qs@(_L=0O
zMB?b^5}Fe~-yP20C@FdUDRNHZ&Q~1rA^mSuG=(}HZ{NzxMJO}Ox5msRBSoCGvAK$@
zkbEpf3yA;9n}Mvyk9&UnxZK}w-BIds$oI_G&DGa=hKGl_)~^q{eVaF`-qp?R0FAb6
z|AD1vo!25FG&D6=;61vy2B0b)z%<JWcV#CltL22|p+oigETKu=aZVK%o;};+^XJ$7
zUKeK%4>bdW+b5aIjU&iB(An9Uo10tp=y06&`}GtG1EJtMy&ZKyx5)lWjpNhg<c&l(
zqRODc&pWnB&Y_Ll(a~|m%9YOn^$LU}frlD%OjM#K{808rCnl<AC;BKnJVA+xB7|xX
zE6<1QGkc=u_WI2@BiPiR#e8`6ikaYAyY`%`;BZmGKZKR`L09UL+l2oEjY}GyJ%>`8
z&LZ>;ik;6E{kvCp_kwA`gy9$KM!TwBMI7@zq*uA+QE<*sYYw?+yn8&i)0P1r7@!V6
z65^WJLm!;-30blE@ZHP80W&>$aw}G?YO>|x<z0dLUSQkI`n&v(@%19)!969cCWb-k
zF-c-#CKXEi_wPp+mn|4|Q}oce!8^yA6ZbKV%m9yW*tBUGakS93dHb`U`hB|37WI9d
zWtFb~55(A4w|>8P<*c>O>1|P1VpL0e*<Z2gqrK=7i4~lj@(a&roRx#W?@RBguc=vn
zrqum;_w!3*ttNpz_s@4e|M@PiaG^G=hWTi<Ltsa-HP%+){4W9Ev&7aR3O|47jT>Az
zJiI_G?;EFo$ZBVN?acf-YNf$$ZDV88HvVLlgYEJygVSj>o3>1()$2$2pY#y@4_@xq
z5IntG>Qd|8U$J|tbH|FidxnRz95l4FSP?FT4l?!<y4Ct8pC+eI@cxSM2}_EKTEoI2
zumVv5P-*tNv)pTG;ZL4reXIaG@BLfE-CO)9EvkUMmkSH!E?&Ci(p`1IPq0<XVOEAs
z>rqFE+j3x6W&+LAVY`vgYu8>OtQ&LBR%=hpPc*2sxvgKnUe@;03c`8p8%iCms<~ON
z-W$a>e3AFYEfrQS3PqWH*fu^S{`TLmeAU@66E=CF!DKoR%(hQYPcIvbTy?Dap3k$p
zJt8k`Ri)>3Zf%${JHsAR75Fqe{kg+v_w$*`r8yO%>bl^b9rmAdm0BD0>h(=cMS?j5
zLqkGVh=nEvJ*p{OR5oHlO_xO=x8|7CWEnmgpDNCjFbRI@y>}4p_)PaDPJ%$={L(I`
zoQ0E<Gg?fS5#<&owvPKcS+IO2tqE213q9T4i9tCysNwu_brw}|fIal^-?w`=aj&M3
zeT{A&tI42&F=~k_&L|S`m9|1V^PU$$yz-9sbh~{{kM6K<);&;rf6sZMs{fgqExk_M
zFI~bo`0475@3wqJ+u4Qp3ZPIDbhWUxXUD$1^3F&=h!NcgAo2iSUb*Hd3jX|L(-GdC
z=fZVI-ySATBt!_S2!$q@zPP+{rfX@@DnCXV$n?-HNwuerUvrU;PY2}y&e7D=Hpt3y
zqhF~;2a;?5MTP{!Q$wvadFC}hoY#@UZT7i-dwG?hA5Yck`Fg(V6~Hb|cFKau$PSB{
z(H?>h(EI>+kDvqQn`8Zpy~e22jT`gM|28gg>Jqwo_3DnDJF9POI6%y|=G*X=T_p>u
zYf_b`Tvx`4Z44slgX^ctq@|^gW$0~X5J6vRbpG4nyRwm}=@iQ2xyeGi)^pFYuqk#E
zTSP^#hG^m)%ya;iGjWc?pYs@10ag!7_BHF9AeRF*Sm%8C@?|9f_<Q8hSgD_%-!eL&
z`5RoG+nGzn#R@K-p6~UTq@<*9C0vw7W0MkB%cA<hug{+DsywSdSRKsXWc+7z^oCUB
z{4qd}O*_xMy>Y(FUw<e-nC{;HVS`3{xyyHT1`$As=AUv?&YqrYRa7dqJ?z?x9Ev6Y
z|0Ry!S5i{Szi-u6VAD1kwdF|5YvyflKQ}8OK~bvpAC1LpU#pB>GiX{on0)4$=W1bL
zC+>{XXU_EE9=}MYYD5UN$$WZ#WjPz~4I6<xAl}fDl9B?)uWdhESvZBZpFDXITe5Pb
z5D@whD)Tj#KPN;tZw|e6>lSh2&Q7bJ=`1FJ3x&GuMo%3bxZ{!UCEmY(A0b#<>eN-y
zCc|czcHK86So3(qcyE3E72$x}-`(CeG#nu`N7@Q^O4Pr9uY)zScooXCT2SzQFRx=)
z;8OY8ismcH$;ofy<d1y(_)+HFn)U0qk??%mwoL8z3dgUC$u@zrz3Ui604gkINYI~@
zM5%D>s*oVx-O?7lZqNB&(SQi`h@Zn8iNn8tFMYO5+gA^;|IGQw;lnW~s}`j<R9t^(
z7PYVE=C+tzijI!9uQoxE!s%51)m53$$3NUze#+0^|Jwe9p1+5if9fJFPzW=CDewfF
zPkjM)*huac7NHIa+yVvI6({Dj)9<nGH9!!>#kDz=i{kB1oo*qV51N`LzVWaUm6JOG
zmQ3tak$oiEsDrk)gXVn)H8pPpE-mPz`cYf6&(yop`*3HPXgjOdk-NxqZS=%ZG$2}t
zMN5ztuZmazL&nkJ)rInGi~n1SPfRORQuZ39ZxJEiqRjH_?v^BKGVSW>&-s>z^0F_K
zmn-jb>YRT&_*hF*^FwL;7AzVXZEBqc(tp~ir;^8x9gDZVVe`KqC)pbq8JPi|pb<Em
zNnXUik3B4B`^l6-o~V#Z+t-knMkbyA>(75LB!BYXrT_nb{=Z(xvzP4r0vG)v+jFoG
z>z+Dw{bbk-$ZO~Pt7G@ti~q4Av%E@PJB<COPRz{Bt-Is@acHXl>oa-qhT&CJszMfw
z1RIhUK$aR<x3GOQnzz~W)m*wKEB+2{lePXppvm4b>qdEw7$NBnGSSD!$E%rD9yoBo
zW#FSIQ3dV^%!Fk;+cL^Nd<H6_sgu*v%lc<-eH?&~$V4#%l=bUZ0kEK&sL})&vb=+z
z#c_zRsD9cQp-HS`Q0nH}_m6p-wLs^5Th?YZfENWv&h;*U)Weh%20}nUV7wq6=>`G9
zo#u~h-x_~QO!wlJ>YaLA4a;Nh7`}c-RlG0RIx*<3NH)<pp|8*hIZyWMc04;TB_Ec{
z@Zm8%dOv5k0KJ;vP@ey@x54<|Zq{v_n$tl)WvH`e?b>PD2FuPeZ(ccD?w@b(TmqTQ
zQ#%^ddh}pP-|ycM`X>H7kBtgn6gjlBc1+)ii|bSLyRYoKcGcSLRT(ET!yWbA-Q8(%
zyBtP89+G+v0EI){R*gTeU%w8Da5iOIxlm(cV@P;7O;0a++R55lOy{1$S<k<d11tjr
z1MCD_Ajl-5YeTY!WQ^6$e__zuvwvsm22rV=AwUo=%Ccq4aQ{~ItrHY(?#m!BPge$%
zt2&zKoT@l>{P-Afvgz}S%nCk}V*mP0lq_|yaE%2v$3LYX*EnT-mRndj0v&x1YSYZz
z?@AXGh+t~Iwr;c`b)1PFR9mbT`7dz9R9@&?{0MA4>20$CeLJ^S?Dnft6*EbyBhPvQ
zr2UlljI=+k`SB*IvBa&9aUs4|(hsaGAL#8@3v+zudyp&GkfurMWV?h!*xNi1ovP>~
z8cRLwh$<v{X;1)J26;*LT3U7PYPzpOYj+U!Ugy?O7Os>Fu8thwydp4vqI2|_srSmQ
zG2$m)pb-SCV~{v~=Bt#Oo7?BlpGPgc)IoObu&m?$gl1B;a*K9Q4)DWde<KsRgZ%~u
zyue8zNEqbJ_Jc`{WHo+-#%Yzx()>>Jbjf4~xog)hPa=RRmzBP;!STO<P}ec4V5_V?
z7G1GEC0MG+herw!uGo8G^ZR?d&GYxnW*Jp1&QBb{f-WbUSlp(|_>Y6zjSAALf@@RL
z(OlnF^bprc*!#F_e!QOea6mVDC!ypWOreYw9iaNqfb5czlOylm4FzZNqN77#>(;Hg
zMMdvr9*TvUeVXz_3WJ(pLZY$Uyu1rxVXHa==0<8*O6^*+&o0i-0+tAXtcr|}zuw(#
z`tsGQ2j|}1HS|<<{`2e1Jw*>zz$J}i$5w+~6A%>a`S$H1_75}&O9|_-!Jt3M7mlSi
z^?-yx>QF+z(*5h#b<|Lc5&<B}(NX8i!NEyLfuxACeY@+QyiDNR)vFmn9z;u?Q8bD0
z0995OD`A&^;wQS*nZH9h2mvcuG}-OFh|a2yA6Juc@+LvWYm|8K-~oEM(b3Ubmu*Mw
z?YE<O36?xlns_j3Gl`6X+1ESwykI6l2*efd;r#q|jhK+`cBjf$W!@77?aUG&GeMA{
zckH>uxtER(^-XLJN+%E6H)$}EJIr6Rg$pSww$>DjW#}fe08$38T(g;Mq|&p*>WnEi
zq2Z2$B%Wa{`%Tm?F3k1X@GCec5;xFRgoS9@+GY@F*w(waxkVvy)&U)>(Jau`FsdTu
z&}bF_UgMp<zYo5-DauT5Oh!U*UHI#f?D6j2?&GMzKTzPQ)bU0Q;g!TdYfb{mm%n}Y
zu5h+KX4@8&zR1s?4ICDM;|9JwNxgmNj-^pTPc;ha->IQ3$OOAHp3@nameI*>9zCWd
zLg^#=u)HTv>I;UU_8dpEBO_ssE?T{vQx(DfvZ?7fvQO;DU-Y}ZKvE<<a5W|-W>@9?
z`}e^W>Vj_=ceih9YU%|!a1%89)v&NI;sz+cu$-J6w5s*iZIw%lr{2DO`?9ms7}tNR
zR|X8I_|Km|KLS%~tL$;^rn~>@_+l-L@JRtB25KutyXiONCZj@IQNU!h&w_hAR}kr-
zm*!{t622h}^%_V3HtFO4$p3V%>4mb{pK)z_W49hx2N)=vczZ)9H3~GRZN1;vt98Av
z)|to}9YLw;g;4aZr)LB4<HLgkRBFD(o7KA9YRENPNtQy!p)pB~QI%V*WL2VoN*%e4
zJwqF&E*1hBsQ1*5HWoRYqEPtucxI7oOpbB+L*sIv&6gsQ#M&k}_s&#VpBYs9<y5(_
znMzIY8^21Bcu}TaWD4TjK_jEPAt5B6rLN-U;Sp2nF;s`3*$OBZi{rw(F#kP5<-IOD
zWC&3X4i1vUNI#ZRXa1+YwN<{J%g)hJk|dkZpd`=YQk`T3{LxpNIXK+!U3~WJnH6i^
zZkO3{OPh}m560Ge;0us*`QGicYuU~qQdCs*t*`Hv%^jpB2tO}DxuQ@Owr-?SE&4w^
zP?z(WJZo;gQDtE=s&V}9pWjl%!}Gs8snpfHykf*doP}daw3Q4Zz#zw-c{(_J0m1n3
zCrM%bJB0p9F5fOUDpt<@+B3qr0zj(OxDpx^$sWbEK7YQ?*k?eLK?GROVze&B<@@XO
zLM0~pp9!==l<HeP`xe0D)DM4o(hH6>_Gzcjp_VMejUfACcf0+pMZOnRP%xm;DrUZ|
z;@j=|aM|+Z>(A)@8EQR(Lbd_it}Aj-`q2j$TNqQC1`0e*WEtoZwMcUXy$m;S<p)td
zN!;p+gt^<j|HjPUzqUg&1E~p{0P*b3`u5#F{{tbh`HvN<R$Gt{@PM1%C4laF<bxl`
zxatzSs;_X~B2nP6V>h8G?n84UN<1`s6?!K&^7d^p2v9)lIfc3QT@`0mT}D)^?^Dfb
zwd-M`&&<q_m+}L5=eo4$4=V2_mdCEX{u=5iNm#eseHyUfvyzQ18YdRSe-uuKorE1u
zhYVi`*#R?t1|RUa@aM?aXZpSi2-%Mn{=UhHtzc!zyV<mPa~%T60-J7z06BN=Ts}B^
z2V*(yKLbq~nO7e_eyp%(PZ+uY?S#E&(4r#OpmSqX1>O<ez1t9QvY-uo)v1;o6DhQT
zAE(E_Jv2^_p$(u?fvl+L@_C4SvsbK9g>%p;3fDfBI8{`;N~%m`X|8Lj;OM*V7`Jkt
zsbdXSgXR?gSJko(t@%G*06<qJ-0wPqq_}%!lhv=?1k$gE)Ijq2IHuqus7>yJBbSMA
zNBQFizTjJNfzTBhRedJ=$=x+9+U(K)Vc&Iu-ANKnU0o{e{x)p|=`9y;u0fL&lMuLQ
zcwJaU66Zz6|F1g|Prd9JprHbX9aC<{zkU1m5SfHgb@P$<tnz^Q85ajb2ofhMgzEWr
zo&y)zl640}`rUiyYHw^XI_e`^Txi|6e^p{%L&7a^Rd#)^!Si2xgf#R7iS$i`%A#0y
zcD7-B4teLx=p>!t`xs&(qbkJX%yMKtIR+7Ax@V2^DBt;DMGcFcAL#Phcb1jtR>=e0
ziX=PZTq#Eda(jIuvjp)2I3eF_%xt9WYk|$0H76v&*J^@LnhZ^P$dL9X!ksF9ede(#
zRp#J`>VceA{2MpyvsYPJ5O=KL_LGfN@{y14oQ{RiGd+*%O+!thhhOI)tOad8x3w`6
zNYAELR<~?+v-8Um#4;8Zeh-GF;_FJEdi&f*J0~aI!I5ti6I+3f<`VMAn*lbHN}nll
zq@;%nOCTN<oV$tl2`UkIC7O*FEiG#Sj`o8`CpDgq3jb4a#)9tpIju-Hf+)N6OP83)
z8vpGZqrSdADF~+5V_R=kzkbaEG*<7rxZyx>H!8&oRKHOi{pv4Y_y|*Qo#2niDqS3~
zp!l;PAp&8|NUYs%Kt~eimq-ObfQ~}?9<brpfqn0u?gE9h0~!G^K==IYM4En%@uLX_
z;`Hg$TwGkHvRM~=-p)-8r7i?QHU_MD4vyu?@TeI!+n~Uj60o#z-tW)X^N<YQH}3xQ
z)XBuYtzad95vaKu$SjR1ht{FIMGAA5jOHF+odPKIlQ_rN|1yK?U(;mD9e}Ra)Y3A>
z<}Uh_xADsBr>Cb^u(7>$H4N^lcP74Pz0k+C6xBmHJpjJ5pjGO0aU$~*4fN&TQ1;b(
zB)0SnJ!#l`0A=Zk;HZ4&0ttJ6zP}&W>xVP)8Z4Q#*~=9pa!^*Y&OBR7N;&A3T>v+=
zCk)|X!3~r2&RSDp=%PKKk@~wTRdAQLM%)Lb!AAtItsttfyx<_J!CQ|(lQ9J~M@p6~
z?EKL-y9aMOpoOXLP6;kRIU+T}EjssVW&G0H&l+@Tq&#{=-LZoU?Un{U!L2Of8o%5N
zoP0VMT#|XjW!JlQ;#lyW|J6b2CL}wJgEMh}%GnvX<PUl2JMy<6Lm5l=j?Df4l-*%I
zPr;m~0qtNZ%04eI1=QU;Bl&EnLfC4K_)MCeDtC?u4V5Vo@K8qcE_S9=7O0aVAP|n*
zCRq^>)bC^)P*anFN&sm3th<^QGEka}n)rgHJSw48T=$;%$v`N24zB>0^Xc)abp#iv
z(hnbwZrZd-1m`AI=e~@QG4ktZ-L@Q3+9y#q*);O;U7eZnQUZd+34I0dk_YBXJ8l>I
z_nL^n;R>=1MHPk9;5j%Xw}K_y@&%;TF{xFnR*8i&s_EuJN`%nz?nl|6wD#e{{=+_S
z1QqPC81y3&*<DO+J0wo1suCn!4+bUEe{yn?MqUQ!v$N<7z66s!#&BEVs)Gj)-j=hA
zZOlvUxU<XQ?%yl)3P|Flv1i%$*c^@JF>+-ha`W=rkeJ`;&Kx_n8wXrlQX{98sn+@q
zBb~BQy*o^*g5(vAitMlM_5Yg)IS=COW~9ShpdEl$qqxe`CHJOWA9tF~7LVO!ld4*?
zW=&!47OM5<+zli$cdbAgThAk|_hGu`=<Im!HKUh8>}=akW>q~r7`5Z<i)DIQ1{8$o
zd(ZV}aN5?by8_U*K|w)5V{Xa%wvMKzD%P3g1d2;adcY<Tgsf9%M}Lb-^=M4<Zga2g
z4GAhvdRE7K4k@=hvb8LZuvK_BvNr%g%>d2_T)|}|1B)*oOQ6jND0y+k#>OHOEmskV
zuL5LRsiC29_1ZQ6xMq>0Z?D5eMUpx$=lu9#@z_#46&P*K5KGwRHsz~iI~3fudc6BN
z6SAs>(RH5<*~Ske_~lsl?gcYB(br&(lazKMb1i9L=<g4~(pf(ij|i?BR(cbVK7=;H
zuv6U1!QmRfrbKS99#DjIz`PephZ4EgjY&1{kC(dJpcCYTvkO|nrMS3ubK${}gbY&0
zgp7dRRB(q=r_$s|`#ROve~KiHWrcR?3%>p9GB=IBSp@O{Dt6n*jfi;8;F{A5@|n&i
zImXKOl)P8>^z@XpX`DZQ{&7k^7&-+gl8<@=Ay<(Z5E%UfXi_i}qfFmc8W<QPI<yzD
z%DkS>XbAn6oOpbLp9-qeXasRG#~7#9wi{J_pO#kWS{<R*{T81Mue4=Q)~reX%3**z
zAa?wXk4$fonqoZ79*hezs0>gh)jTgRFN?bzgi(R@gRj%Kn@+uQ<*#5nWvz!+_hOP*
zv2zb&r{BcksQZ3tQ#(BDI>|GGB4A&wj@Sd;eYKYi@EhnE(fMqYm6fd@pe?&4WwOiG
z@-Kp76f2MjfM#U1cGWsb8o&YRZxd4X;bpdz^HXly7c8!zz>Cw@Y7~J>xUkOHi$PVr
zKE7u*k!4KdQubCb4K35snXye-(|K4TGWWsvLHgXz0^*R)sl~-PCqS?&sG)I1h|qcG
z+U38?Ca(XDeoQKkdLqhA*+122q-pB$oMYiL7XmnH&DHDIW2Mbs|43W)hh+QW&I6B7
zsjlD=wZArWRR)^)jsO<kiG5;Ozm`D+#jws*q=ys=Uq=W_9e<=93I9VVluLd6*6p`!
zv?7F*m(k1Sem(o5^9NYYru`WXVRCEp^9%+}tfU3l>U>~863w9i9V8o`&?H_7DKRlk
zbWF?W(NZSHq^Isn?%7NCR055W<Rk$(+YAigH9!({D^NkMJO;X-K?Lj%i6(FdEi9gU
z&;Ppj7ENP@evTyZ5JCakMrg|+up!{2l8%dOvB;K(9q!3h?|UTg)S1Q3vhZe00_x*|
zALvD5(a?xA(VLOxqhKm=FrHFanE4*jW(STNq&iBs)w2lls2QA-l$8AUy;rX`mZqrq
zPG^zdJ5;)OO}?{MwrVQ*lIQxpXj0X5xgG9ITHp?{TRbIH+R$F5iQHvX1-H>)<$0k1
zQLMf00s8VrmcRFr1Bn~45~t2Ry9d*w^~905T@p?C$w^5CtE@d#f8e!u1a|Pzk>llS
z<`6n!YuB!2R3-gY$=-Wse?)12KYS-Ccpoti-`zP@Cg>b7_%x_b*L2y@&wC()@DO)i
zfBoBg%jD(PugdjY^eas3Gd<p>BOI+6w9lx6S1n|?%q|T1_QrBfA?tl=g3IP~p)R(*
zo>O%Ff!=Vq{pk@R<jNI|*mjfjW1@F<oW0ms0Wr}|WeE#d(AqrW>#ihy9PlZJIk#ry
zd-C@?=?wNr_s4MT929r%-02O?@gdBl_`!hHbmysZACFj%z{z=XNqRvZ%;tB3gZu)y
z<9Jskbc({)%V!#O5WLK~M+h0K_xG&+t{X&k3_~-nm!WqMP<J~pj5^b(Ct7n5AJHQ+
zY^O>d9ZAS$FK$EcFpDiz%V}NDZru(N#e1}S-=kjC##-3kUV-rU9YF^ogDhwG)bW0!
ztKPHozgr|<vOiVclb4tG5#8@f4vzOO3Jd7f@*QHnJatO&TbOG7SnhkC4<yIwa-S1W
zIkY)^?qNenDC{@>X3Mvc5ecw!-0(-)d@U_}duL}~UVF^fpub=@MPV4?+Oua*UQtmj
zVB#TQoFC<XI-{6Zt<_XjrH6!E4hlL*Rv?y(yDVRZP{sj3YAgKbXf(1nV^q!1OQ(YE
z=0dgiMs`er=p+M3xEtL-(WQOzb}iZ$FJ3HYjg>NyM81dxi6`*;dx1^ULqz_a>BgEG
z4M0p0dB=`gPd<?K`EGqWU-Hba08)k+=d6QDF2MP5;UQ?ajbv{H3|Ld~^pgs=m$$d_
zn2(3zePEohBryd4x73OR00bZ6*8lb=Ql7w)bnmQG6Wg_`#Ul4jI$|gUkO3&V4hs3W
zM?1bckvWkx4WM@r1-E`PK3?+q^XDHSFf)iCZ`b`PUB=9O5L!YC@b5p_;$2*P{Nib+
zxw<!FvOH|UeSrLJW8inUkIvm*gZDD3D!TWNY8UyUk!4hk1<eAT@FogqZ^7W>o7e`w
zZ&R;pYV2x$B}eS_8Jsgr(c;nN77Z01q3PYr&Gb%s@tvQD5>!xfu-(Gy-HWnFLIDsC
z8W2t<Jle@r45v>wxM5O`hAVQVNOFzReVtT~tPgOQ`Q;qKcDbP00vKXP+`exv3%-9t
ziSv=-D(~o&l-@P>w~>yl`)S-S{ok%EWQwqoE4tNopbxc!>?oqQ=}Xr4EdL}q!R|Zt
zc3ZXa=EGCF?sfr4JL3Y5&0B9t`gR$7Esu~iadBZ`L2X1dcSl}f;VW27BY)RFMmfiB
zT>=x&Kec%4=FKGL97vmIK{_9?th{&sej_q`{C$?Qs_YKI)vVm-2E8Y<7@sH>RKv&+
z2^SMwDi&W3rSMk>$!8nsptf2o`|S-Ioyv`8?UEyxs5dOlc~(?X((zl8T3C2#rQ|QI
zXt4|E=GLy+tX&1ut21;s7(<>nH}jxXropmf_|OW>mk_d!+@w#r?w_S-H`{G3C=*AY
zoI4D`$P8s?bB)LGrS2QT5+<tZL;5g|7ab6U@J-=am%^iV(DnTJ*1m#?6;vM?tnpex
z#`;5|Rel9@MVqejyDEVd1Q+Pi5|5#ik8DUDhCu`_#5cD*aI#hr105w3$GaNPA7suQ
zr0LoJ^ZSaPrp2->U%svWmXwuTNT^8l$t={~Do6*4_c@@K)yh6rtnv;h^h>{FsPg;w
zrq%y+>wRMq!~)diA6Ts{7F~+@%z^~+9SLxVQIM!`M7*>%nws;N9lI~pU8y4vn=EOS
z6{}_^j>9>^BP^U?HU7$r>NDL_kB$&>G<If(yu603t$4uvm_}63wo0RlPqDI7CjYz&
zR8vLz)RrUhFF_|8uqRP@D?Xh}L(C@wX%KW6r1+<(>C-y5Pt1e@nj>M3H62lT-rBkj
z3JJZkQUwylVVxY$g0oIe4NUuJliZ_6PSJv*i2mh?f5FTQ>I2B4fq3`S<yuJptnQ^q
z{w%^2dFFGm3&=5QuIGr+<WOtF4AVZDKc~j}5(3T8Kq2$jPVTg_vVsW7f<ASZ&lz?X
zfOOgs(hJX;vy2erQCqn<n3+SG#5eBU8-HShI2F*erLxu7mmP8zL5CYhv`Hx<I0uYK
zbxlo;#V8{^ElsQpR3Qv54;qK;Xa6&6WPM#rIYQLP+kd9RxgsJ$I<dtWd+Kw0R}rMI
zjHn`2ClIFFt6QOvNW1;i9$6w=MdTPRC^j%Onj*f)H3rR3A_)jMXX0W>(-p2cx_pi>
zB4(z*fsIa1MuSngbLrt~p1jfKLOXpNMh*f^!%ciP(ruw7na*)}8}7<GT}Qf3h8F<J
zL&mN^Y@hmdf%JEo0glnz4R0=o76{{Y{td?NjYPG%x%m{7QW~x**RECaWN_Mz;2aHK
zfdcq|!A27VY-F;3nZnM_$oO?}3h(Xsp$LKc8qy3wpp0EW;v{BrdNVPzp~Us?swa6>
zLNx`6ILXt+bkC8t2j~q+aRKuNwJ%zeB}|A>2vgM%I$*wq^{9T*4N?O@48hm6W@2=7
z$3lJATebxMX+C;*OLnqGglz-g{{poyo6(C?bth4Sj|~F6`VP{o28%-x6nw`@;V~Wj
z?EmORE7=8~@AO_y<f21KUW;bd!s20z&k2x?_N`61XAg{cySmEImy9aV1bT`5cLvsR
zFj=!EhJ^_1aqs6s1AZ9+7z(*+mt#lTS6h^HjgwCWGoa?P5ud>ZAqNz+PWhDYQa8}y
zy}k~nM`l7yKU=`w!C?$la8#9xqRcFSX@LIr(NSGNRj2zptwR8KS@Q_f{kq(M?^X5n
zt2#S7N5{waW9?1N&8ssg%6+4j8Hk;>B8M<AnLSXc0Qk$NOG6TYL&%;Rc4o+@aU(wX
zkDWm|5YAt`dv^`n`Oi!L!Gnk*v{^>^>5kNoKj5XE!hnKpp=+<!$i;>)UkrT}cJH=n
zaBy%ie8rdzh61YfuQgJAAOgrUwTjic>w&mfPyO4sY)mX{Re&8XV8oI~nF*3zK&O!B
zKj(>}{{W&E_|@3hU~qNTOH!wbwiAf}aZ23)KeQu}Jj4;=0I{9HnV;`#>lQ|?y{xTd
zJ9kr9J)HNs&)#ufS?e20t6-|45E#_C(<_VGNSXrhrCd0<3?e{DRH_Wr9EZ3#SsSX0
zi_4v#3Cg}jyAI<*3R;^!e*9r*(@~sP%8nTq7MyD_9{N&sO>!1Pw2W8oKxTJ7ORfV*
zv!9sm3HTPoH+f@Lh<@A5xZRcr2sr0dTT|f2K#z-CPjV3;?aTgj`Wob0Tx}VbgPma>
z;YoHVB$oxm-veER1{M|6z%>l7m=AnRY0JlHUq%#ziW_Ws90t<C9BIi<gVgLPmfv>}
zC4ozeYo2Ps)qj#(?W);924ae+MNc)aZFCuXyN!5=!}&$$DfnQLH;3(MAE5~bd-@4b
z45=;&x8Lm>8p?d<26GC7NZ48lpUravy}hEuq*8fEn!;J9D>Rx~4F_>Vsn`?BCbIVi
zs+Hj}LRE|a39k|r1GDPNs=%)5IxW;VQuSLe?^yNzJ;%9oOAA`iMijG8ms|u^+pim~
zJpm-bk7NE4Qh7nEvEP{5+XNM1VgJd-onIP&aCdZ*J_X&gQ!>OXY*z7UYNV&#w!ZfK
z_N{~}07_xN!rz96YGM)+hhJY8+y<4*cenrsgjo2)>kz}3w!mr4N500;N+@aUgyOLU
zpmn_LOMCke!|5Lbq_>O~B)93Mx0jbvquLlaxaQI31kvUWcNZ5*oiJQUgS(3&QrA*=
zwv>(B2ubVbqO2_QpDRkcFZCm6pVj)Zkx`X>7z+pq{dUH6wg)1?VABrFn2<#-Ew?lV
zXyfwb%StvgTFaeQ?sR7)YWCJ|@%{7bM8YBlW&W{-JZ*?{5bbqMZ{Zw+iI<CLB{n^|
zclQL5d+r_oNvbgV!2|xK_nSrj89&57^$Y=#d*9tHB0`|gWA&=<Nq$chyLVSHTn-x3
zDcM4*x}aiM>3d9v#Q*s`bT_PJCdf&Prnev=n&R`L@64m-TnA&6xB7@FHcL}mTd(S*
z6gg`GPtItGqr%VQj=4{+*-M<@77c(H>Du>R@Z?5LD1_23-)JJqsnw!cEB^cNvyX@!
zyK^0N?kQ(lN*4K9dYaH)Z{D;?4BF#axp7l~zhb{x8BBw9lVk+Br@ILHcEkKlt>#i-
z6t%9d76rRa%MUk{VR9CrXBicSGI{p8+UKzHi1!2PFdq#{T}|O>HNlvWp8JoI-l5T+
z>X5Lo<1~vc^|wWp=?vh;5jPrUA1oCu^<~QeTPCYO*z!<jDOHD02K{e4o-h+IBAsq1
zh7osa;AE`ysXn!bzTiL*k<*Bcnoy<JwUHb1Y??L`3M{4q7=^+wwdeAL#nR$JNrglf
zG;T%_cKVw7+Z)ets~vn&RMY~V;r<_<b7w&Y{8I@zs$x>v$rH)L6G+@Z{QnqaoCVOj
z7%tS$k$<^_(-dvPIMgN#DkwFz93Y#hWDQO^uOOzhx~3oiiIC8nYeJJWwY0RdC8MD{
z<OJtP8N4{{dt5jA)IX8wFKQdlT(`&Ir>*Hmg&B)IyFEBmCE_=eSF88wG8nerzh)(P
zk?l;zyVXChD$|#kZg8lgpQNI0Zufk!O+-YMG}`Co<{rJ&8bR25`TW}#T{gl;N-F2{
znH^g;Z6a9#q2UXXDrKzQAFI1J;Oinm<sQ=3){f*jLk?;{=nloinNQr=Wpo`SLGsem
z(rlSj;i0^nxvH;n@$Vsrsk;rUf3Dg|I5Etu;uE|b0{N)|P9GS;xxbYAJ_Wm)c=XC+
zD1o)88<wAQ&yuvo)I5c_#lXg3TGK=i5!<;cmL8H;Yt2BQGpudahJK`<kRLWX|6qCU
z_nu@%`URW<;gQbrc7Nmmz5&z`px{@Bq{^3J<N{)s9VhvxqT;zdC*y*Gn3a^3C(14$
z`C~}Svt>_W>qfZ6|DEB`kBjW=?5{BA<&0>re$&J-%rGiQKf`%|m)ud^NKbA$tO-9e
zc#`)Au_lRWF%ovMqv;teiSjZP%crD-ib8v*C|q1O`eHNTC*II7gu}y8<w!&^6fkyI
z9XLdb-bsQuasFWjFKBG*Nk~4XjXfsi1%@h4Ro40$J!U1ptQ`vHPDDy)-(5pL#?k<j
zea1khBK;|9OPI^|@m<DwQBhH6O;s$-Zn_hBM3S&KEkv`bI^mzOi|3H{-wG0&HLtxE
zVfbQywV>)>tVMKRY!1L+W{tmM{XyE~@a?>1i0%EWM9z2p(ZX;TV8Zs$q)kkpt5&b&
ze@<{D>^aY&pMCmG6C;~Zg@B>Xf37p1x*j`dHkk1^Mzdg-aon2x!d#Gm6tU~b&i3~9
z%p+g6ZTPoTj@c;;4_nfYZ_$4mUue^$UOts2eaN7{m_aiKWKq21o(_l-KEqad(KK2E
z2Wn{uYo$Je!Ous{2u_sC0K%ESYGrKhhm{P@{lJx^iZb<<Wnvlm1`=3DyGhVHzPf^Q
z+#j-n*lsb)`nZW?_U{hy9l7^m+>@x+B+o;1n4OK9?k4cL8c*-RpndhplPAWy(FPVB
z`I<pZdPF`#<@8rq?C%D+aD3Ese_cB89x&?o{0C90c!w<`=3u7PQGMNHt09PR!P6>Y
zUY5QLS60{bbsIR!{jRS07sF3B$i~971W45Hz4b!)KF9l!jqK<5RVw?65#$&O`hL~x
z=Wg$Id-V5{FsGB@!~Jj86BsYK_ni+(B?yU_p^AyHVjUY>N4V7Xr|e|qW7@y!$V#es
zKR3cZ^uah$CS+XZ!;l6e@a_wN(MkI^!>jA*8Sz^>0B!K-*w_V3Oom!&F6Wxq5zeeK
zU;Sb$#-2O64foTSnWbb<zr~mm*VAJ;ne=y83%7mwLN3KOIM1ha;iv^fUmg0QQdjqR
zZ+G{N<ZdYE-Ef;7**p8NcG&myX#u!}zz<)7gLwkKZ{c?zs1sY_ZhF(!(AA>UR6<}`
zF2AZsg8<dX*z3eTnp%t?{W{^pV14OIpgn+lG^WDDQFM+9!<s|QRMVqfzH+6S975B%
ze-VAz{K5isovcLzLVzK6H6POnH)ofB<;aIIhS1jIPRR^WAXEXTKRi3{M2-l7uCxjY
z3B7}IdKN2y=`Ch}r+62JhhmD9FN~&&LZ0nM{>dcKJ5z{oeeT}llss41AW)^Z`Tw0H
z=cqAmOUSVBn&7=Bvrk`*lXqg|dwe#N2w(1eh38+$CuJEj@MehBZKAk{pD8WSj#n}S
zi6a6-nV}%H?jyRRtKccVA$uL=7FDBFEFV>|aPR!jOK6eEN<T~t!e@LHz84vI!9la;
z%Fv=WLJgh)%SighR<g67;8t6?h9Yd?Cm$XnW<pLu>uDAq<9@|i6^8444w5pOLr}4*
zt}ZlYLy`bE!b`aHU^c#_kQE%ld+l??w48_y#G++Th?q)VZ&(OxvMzDX5SG|?2`VOh
z@Lr;l&6=Ug_rn1Muhv##EiuEU%8#LI93?%01;W$-Ax>I<v|!{gpit}darXC-lFiiM
z3F7mrZU-VA^Uum#cOn>BBOegK%}8GmIoOP8lX{=B3^A8@@1GJwI5EzspyZA-;ZUdL
zT4#&Zbz=15&g|UWwi6lCJiK^jz<<vL&^xOZ(x2`#+RXrGb}vLqR{8}Z;2cJpaf(7q
z*YT*0>2trbMAFm3I3g~}XEXW=R&a(;^$)Fy{7GET6Q@P%lfaDi($X4At5d~n@N*3J
zrjz={pRPcCZnZo;ZcCoe$7LzX3;}d5-F^D2s{TOsJ_<9NJaAh4ZrC`<fkxmiXGUkf
zDrWN1FQOCH`{2)@ibjKN+df*u?@Rq*Qfw+wAtB?JyJj&pKFqA9%LwplA28A;_p0OG
zwJF-<JiOQZzSF}D^diC&BhAsLkn58!W+j9e!1NzVUf7;XEjgTq>||IcB|YLGJhB9-
zNqvJL#bGOBCO)0R7%_SC{G5>f^s-Oii4jTadykQF0P71$VnSuX|6)@77`NJ3nn4~F
z?lS{6syG)IRuSqhLOh@UU%aaY#f2#7<rv?`+ZlN(guY!_^N*{E)MX{meU6vWA3ofK
zIOz3W#Nn}N$bm|V)?gpa%OXD)j8T?x`Q0F~ZnBle3)fO3z-K+!TL@O@UxIr8f+88N
z9aMq3sWR-%esB$WP#Fk%PL34#=5SD{JoQGKw`@`Iyo&*mBsgKoVbmiPjJCj^q^8Kf
z%f8m^-$Xa1Hv#ag2RZ(8`JO-3WAH}sDc;u+Y+4Rwh=qWkW$iP?M=2=}@T%K;{e5Da
zKu&k#3=>r+V;G0gA7XT=8nhx-;@=4J{BaPNL={*VS7j%1=rIwEBSD9)kWharz_T9b
zfTZ@ML6vV$)g3zCVH<t(=0$LC)Xke89ktI%5+ex81)bDwx$Wqv0!BE6{;N2z#QX-_
z6-U0cJ%9e-bF=i8rvhtRN9C;E3&5zxhe9B(*RzTMyR_=ZE*2J+B&2lfPid4&^Mn8D
zRghY9#)%1m3Je<;jPd|XY`+vV9rj{)tWun<THl2=4OB4^pJmP)EM*Se0t3-Bm=q#!
z-h9C+Llr-^-GGrQp7AL*r>d&zbc5aJ@szz%***0g9ZEG*;*N6sd;JuUg06rymiFt7
z0&pgu39;1B>3>Dd<#8af^gh4pqpwub3avt7wQpFtU3lIrgQVBi+7!3L;ahFZwo9u=
zd0pc#*EPQ;&yT-cHhG0vjhV9V;mV7XrMYv9iw2&O+YS|8s7UDhqH*v3?wu3whlk62
z-`^w0$!jfSDsD)>R+u#l&wCW@m-6soB7VJ>lhI09jgbVi@e@anid6ie?%I=e=v`|V
z|3-g*fAE7(-`G1H*Q_8!cI>zzWg4E>H?b$?KozgAaY=Eps(8%Bkin_Z07u8{nXEU}
zZ`QT?HQ)1Fd6C$-QD`wFO6oPQd7JaQb>_pf<y8k$$T2m*X~&M@3veQgl>_FAY}>X9
zBGieyyHZk9!vgmxDq?%p=L36<?xS{*f+`~M5{-HxJ$);*5RSmWCH|+|@Fu_)a#$Xc
zSHWg?&<n3H^qU7IzkhJ>8-(jB@VYFZ3&Jrv11U16X#VY6Z2%3}c9!$XSZH2~dH}!W
z=~DNVe0+THVz50twiQ5`1*U-8ad9L81?K&RR*Km6?aK&WcnsucSc4cLi@w0A!gn(;
zF)^XAc0=;Gl$>mS?cYNTFv7>p%1lmNojG&n+MgILQF3hd<Huw6?M3Y1&Pjb1hNz3M
z(`(;X<N<GxQ8bUEi|bOQMkaz6xS_Rv(E~aIE7tz;ZwdvfSQrduBrE;-GcRm>qf=9l
z^N+}Z19^@ipBrFVU||u9(JHWRWJJ;3j|w|<TBp}Y+haTie+dH|37gTqz%8Q=NJx4z
z(0G$pV&dq@lff9JIG~}i92%<{Jhw2Dk&bT|0(g<6Hv>sM%Bu(uXQOFoe8*YDf?vtY
zlZLgU=lgdiJG&vpSF+d&&~WPTndD*6lavn7+pEL3jATa+<o6E@j4?+@BM^ou1uUu<
zGkKsfhQXdS1nbI`jkhZ>S4*Xq;sD230)I)t^R*crQh|952cs$>ws~{#$t;6?P%_t~
zo(LEi9Q=qLt<bR^)}l0=T>gpme0+@n(F534{)y?q<}l&|xZT?3=E6FkfWHH&;=NAo
zMb|nyI%I$Dl$L$~O6(X`vD~~BBq^gRh<_1k68=`aoiv$KskkN>OuM^PA&+veUcDbj
z+{oPG+&Oh*q4fNU@o`t95?9eLZEd!1o$Txmk=9~3YHi-;NCsjE%);LObg#m_J?C}I
z14>ToASy)>s@J>)1q4u-DiX)X$9-!0QAR}U?Ccc1$1P$!;Do<ZR#sN<IuQtW%BTT0
z=Ym)!dN8N34u+^a#H}!rvWyN1)&`aT6Cl;c#>Vs_dlM5A4AjD`nsV%*y1M!@`dSGi
zT5xbM|HR<XP!lFw<yz!{V;=X`S;&y<>|-pX;7kw41Yg5202$_~K7alIV|Qk>rJ31&
zq}t@vUhwZHa(op`9`TunOTT`jFA6b1gpf6@BCZWUf_sDmWK`xQZE0blchF3{OT_6*
zeh?~~HAF58E2~tO?^f2<n#dQaINL6FF<+#fbkxX53uvWDb1yuDZ>p;g-22(vs|KGA
z&#gOhc54|Dv2AMzS6A00zyK5N5DOVTa6McUJSl;K!L=jX>tDRMI6pt11`+&Q0qiA5
z(QcQUrB;*#j1;zJ9Ioi}8Iqt<@9sG->Fwk55vlbcyldt{SoOgV4=$-wKf)b$%>$|K
zDArMKbv788DA=JDNcUfKLV}cJ6CO8njdyl-wh^w;+V=K1;s#96UiR<~9zc6zW{4Sk
zZZ59G$bJ}G!b3wt2YEII1_m;!!Z+|(VFdiuiJNW^YsM!hH^KB_i^t6zj#nx%9xB1$
z+#3rS+pnc^<U2_i7)eu9R6KdGyAUn4JVPQX-!Ukqx69aslx~yb>TRVS+eSJ{q5vc}
z$EyIfeH$M%C(Vb9su0anP_6MWfX6EcIj7Eh`T6-TXDb(`Pkn4|PJzt@g_^e<BLojo
z+8zHOD`wb#N3C8-Ff%h>9b2YK#$vjj7NJRvypfWUq8?5Tc<EH|l3N{bIX*=`=0LvE
z7A^Xw6^#Dy^L)beg%R|O*D`Pqi4bJyQK=Y#Tf;7>7#$J<qtr@*T=S=00f%<)-u=xr
zr{Zsh*G4ZmnO*+EQL>WQy=PAo2DG%aQjYma$;jM<;<$CQrfrx1pZmq^c(7115~FnF
z@h;iGiFfie8i*lb3V})JWFQEe;d)pP)8XA+M#p0Z^m0B21qUCqu-N3#Q5?y}##Y<b
zb{8f-GCu(RsKV|czm=O6k)Z6m%OA&{d)>M%uM?5VU`^QN(vB^CwM<iA|JD!JlHzQA
zWA7m1D2$N(u)ckB#}NW5=5qSg<@KF?tQ`40Oxg&9uY$6iTnwzdD+to)Orwz}fl(Dx
zP9M@J$Hlna@iYghds~Rf>2{~qZR&#ycwPZ0B|OAHKP%nc!y^SsQ4lUtsQ&O$_0_Y6
zhJ60GQM3F#Za?1|)zAD6TslPbV^kqAV(;G&4gm#K^u0!X{s7vAj$iK*A5hS)hFF7D
zTt+v>1h`R!Ux|VfX5VkKDWCp6GwO-z)<lZ`6BZck<^LTBIsaQ8m5LLxI_f(_kVeRs
zqK<B83a8-QCvKFNm(P4D1;yv^wKN8StARl+<ptOk+~FnpKcQeTJ_QjXRNi~=ZUQuC
zE@FIaYy*msjJ$lS_2MMv8z_VizUkFw6Kvr|Y7&C=BStk|K1`PJC?TxjB{fST69KZ4
z`}f!EkJ}B79iw*6FM&7bm*XZaBAP9qCh%B<i>$1yVi>1~pRV@p+i*-5J}#WUc#+Z7
z)3ZfWeei&PwIedkh!Q*`4>7t+j)ct4%zQu>SMXlm=Hn)5bxcL@zf(F@V7(8mo6)Uj
zzJDb7czJK=iKn1sQmKDiD;L!WYe;A(em1;&mx3}C`tI{l-Dd_RKeO^t@glRcC2>Sc
zG243I{QJ<*A;KC%ll(R81a8peNa5y3Am<9-`(jGeWv)POA?sI780w{+-!32#W6Mrh
z0&OFn?HUwj7IIuYkBL<Z!OFq0HcFMSfT|`mR|gjMl|OREYV(gtYOBVlrc#imZ(uUz
zH7E6+9}XVRFXv0r)2An<rg)YrurrvVARpaAQ~{=d>bd|);X9OpmjE;?RGYsrB@RkZ
zh`Rdv2bq~+D1RH2l*-x|JVr)FUR6S>V<491Cphtaqc8(iH8xUkE{$%De9qfQItXzP
z*!X2v0M-z6nAoog4H75*`~(5bLXeX{t|bC!l6NDaXGm1S{|3D98sj4$uxdW9K79DV
zMbD@X8WiJZjgE~q1Azy9!?sX)lcIux^~*HWE`@-3g-xqhR?LqdB0iLIK6>;Bo?=y;
zSW=DyA&F5IR>B#I(c=ls0B{n{FdFpt_jkkOqweip(lYK~Z;#Tw6UOBA$ko-*q|c;&
zPV;@GC!CD@23VJrmBj{^)^j8`Q)}xL8#ivmC`UD(W&(6@z}eZk{Q7gW8+p(vfu}^I
zq&R(id_FWa0iJCnpVxx`{0=*RVeIjesEE7QQ;m=S@F2Z6Z-}GjsITwjO26K!?=>j2
z<ssc?B#;5%D=jVEv113R29c^83fqsqKD%q_rEXkPsx98$-eTh78TrOgiTa0zk_!sB
zF&qfZ?ggHA6Wp+OE?kNpE=``SVVnG?Pp{yCFG$<Y$dH(2pkh>;*nwv#@UCBPPVyqK
z2f{(T>W+3W+(N|C`YqZIj+z3o5p>wO5?_0yr`N1}2G6Up?}(g$f-`|Zi);_yHLF(>
zs@fI!TvghM)jJA7jhI0OAZ19HRF3wIC(y<VLB7FF*A3rDhJH90qA|orH(&^fa>~so
z5p4W@5!bK3#>u}9FQw{(qe?Bj%J=+AN=r$<Gp|a35;RRrnkzo8XI{3992>;gqu%48
znc*i5qnPTk^Z$*d4+#z3ucO1UcX7rX?);X|Q5L;FG9}Nv%k&j^TVKyaj9s~c@Ipj3
zXWp7I^ff?c!oriI<)CXyUf%{*S#gn|X536m$AR7;ElmM!?bm1??lS(0*Sz5-N*n1A
zK`GlujiiTkcX!hf87nT*Nn<*0f*cOnfkrQ{rS9w3axFu-Yu+bz=N(<zA2>Jd0Jla`
zaWOA2a)!sLW;}5ra7tf{`=}Uqvu98M{yxt$OR`t~;UPy4{VCeAp#8e$Y_{vy{C{M9
zcRbha9{0~IiHIU2qe93kD>HlVkTOCNDkBZMDIv-VDKjfOWm7guGAqeUMx-)Rp7*!=
zJpVk;>vhh3?sM+bkKgxuU7zdod9O?Fco_ChWPCi0w6yd>Q>93cxZr^Uc^)-*M{qPJ
zV-tD{T4b9^$*h|2n-gFRR{ZGL*uBWepxIeB=AAo7(c;K>PcuW&bpr=aM&F+Uf`WvE
zo<xFgr5>t(0C?LlcBE%!?t#vk2rwWVv<(evj;Bub%C6?<c37C1F_Y>b@)}#|i{1Me
zo>iFT7%0gQ9!3WVbfZMiM}GhuL;!$<L~+xDE!Rk<rl#XxzL+PT=$+U^-o!dK9*S~v
z4GWlXykcT+LTMo*VW00fKL^nr6Cu_ma(e2nQZ(AG_O@C~+(`W)qrI)9v~<UYd4%u8
zw{PKSLQY~qUPg7Csp^Pvb#~FI0VY(0=6x1^dxKH|x|Eutqn`&|bPIdBU36=5Lc7qY
zGwj;6e<-=1U!%0w=?awP|FJ?$t_QhT&bG!K4w-wKLILShvUfZL>-{q3i602G?%_iY
zXn5?fx7nJ(V$-2(-%L+`OOH<#TriHIYPNP$ep{35X9kkCIR0O(?8@Po`LNy0yL$-w
ziBYJtdoOo_V*r$iCk5f~(S!w;&}{5Mv>1|2!>%pmby!phH;7qwsQTACT~hMG>t6hF
zlJ`p>M?ra^Km$xl_IcCL+N!-LCF?BrP}_X2t3ywxlRqATeX;T9@!Tiu*50c`Kuc1>
ztfbn+kp!WBVX*O3fWv&<#MhAHFjHzhF)`t-L)kMSYW3`EQM1OU^?XH&W3<<Tp#N}@
z=Q^4MG3^ls%JK2>+xoVp#?srK9!N3i_cMUy)U&**$*LaD0uPu4aK&Xr;>_&u@_x4q
z;&{BTKG($3Mzf^kPql%yO`W>l9bRbOiLu+at+gNhJy;H6Dg2u5)pd9TK$rX)H=z_V
zFfhO%x#Rr!v{kNYxw&4o8!|Nj?FB=~62>l_|AAgST4x_X%P=(lsi2?SZ1%<IG(Nfg
zfk1Q-wnH$5pt;Sy9TZ2EgwL-IFWLxNuY#3$h=G55;piW%4dg<-J_7MPp_u@pMct_N
zXBpJpNyI?mUGLeuH~qsChsMUnff|3gySo+cV7^*7Cd$MV#nvzD^qTcSLV`Wc0+>FG
zOH23Q++67l{Ciz#W4**cQUKkOy@XM@hA1`-q%sZYIUDZ^?v4dt3TagT{=aA=sTfDl
z*fw3GqHTB*6qD^5>f(AWJAYLW(=+-ABFAR^^TwaYeOG`h-X0ybhiyQAP|5o>w5F~n
zI#E(iLPU<Ik3Mk|Ja(orRzizXN)P>D1ThcEUR$l1cB$!Uwe!Dkhxs;4d`T%HBln6V
zN<~{sOENT0cdk<_xgoZL7-!KtM>RdrCVqRFWY4o>VkxQZGx#m;bQhc`=e^rVOr2=8
zlkp1Bk?j7nT)UBhq!>>y+;N|P$`O+yDKb1<KkxU&zw1}-YwaDadqq@$#)bxrMO!d=
zMG)o<&kGlr+bu4=wBaIlck}jx($XyLCYPB0zLA?}AWfs5=Q#vt$K}Zz^k5oloWhjz
z>|{MP+D6nz87NKVI*|P3Pz_P#Ar#X*3MtnT{Bw`@_<y5fCE~%~gfLINcKrt}C_;S>
zD8zu>^jkUU>4o)qd;)T&#X#vHHVk=W2n6rW@M=6m#D)(9vkgZcK<+xjya5SLa9CJa
z!(yr1`}_D%4VW%zFc94(4iA_2#rrsw0;&u?B3Qz3k46|b2}6nZw+}H0|JejkV&@<D
zoncInUs%^gw;P9))Wc*9&DFWuNEM*tkTH^%-nX5`8y+C1&PV##CQm?Sb+Op`9U&dM
zWoop|5-0q8Xtzeh?!2yvKlMG2AK!ZTaj;jO7={KM3~JzFGTd_W8sS|0Q0|r4cLjdB
zuh2~bWZ@-L$A|8-W0tyl^{QlkT6(%#m2AKY-vV?pY&5aY<IsLWOZX07#u+-~qtrI}
zt=Q)+KqH%eEdKhH1(5Xu`X`@eQFL!QS1@}sef|uA4Ouc=yd_PZL<-mhPLk|}A^gbJ
zhKBSee=!M(R8%z+ZEYI7oV2#OcF9#J@mt=$Eo}OM#d-!_(4$S}-Cr>Bje5o~G4rN&
zXBwKK2)UlHKuZqnK-XK58vWoe$eZ86FwsC3-dj*yEU{(lR_5lA-2FTzqC^@P6p4z6
z7dZ1t5yp~=%L}P}9v_@uKQNyBb~iLM4Y-0SRPkCsAB1ZGNk>OTsge#+`+$4P_Y44w
z@tE`$o2E2B4d)_W8D@ER9Nn%xW%YDT1GGfq41qEs!{9m~u)LojefJzZ$OK030R-4G
z3V@x^zYuwLt|v4nCiQm*ETtoWsO;i5wx#*cgbWO(WLg4th1WeioDQc(5Ny05`d<mh
z49?pyPPyj>zx(09C8Tn~N+Kfk=y&Z`MXP^phcD6C)vIV4y4;h{2@pIKfEOZ#jZi0z
z{)U)P{o+L_APRYG?OlC`OwaxpZ@s)a5r<GDC^l~LNJ{QNoDK1#5%xjE)Zs|K-FW~V
z#TGmO$%`Ad?~k9azKOGA3=aXel6urQ$+r2vJa&w*>0Nv5^obG+#+eLl<>=3=zy`eQ
zfk-hRX91>`J@WD#r%s&`lawS9CjkUe8ya@{a8iIsf-0X0bI=Ww>;?ef?l|l5{WyVI
z5b1aDmb(ZOVR=LgPE=9EEWB!I*^Cv?j88!@*DY*;SHKUk9_w-b!(^9Sw2t3p<ja>^
zk&%&y&tHOws1RLZ(JlvIn1mY(l^C#>Jzf=CCpmgM7Nx%{d$3XY42~B&(rl%pBjk#C
zGx(OOaD$;aZfB7rYG4ud<)@OihVAhAwG0eA0PL^%R1jDgD#so`GXSl=1Rat90+DVc
zAj5FtXCtexO4%1QC%e-C0dWB@IE*MOSyn3o$A#HY#mR}UIg8PyMaCVQbNa5YoMTu~
zb6C`tyNv%(Di-3e___aD#D9NF3|9YOjsKdSe}7A0-v9lt1~oF2|NGmf*8l$x|2sSX
zedNE6>Hqn||NcZ#(|%OI>u4VS`=PViZEbBTIHtsruR>4lJBzs%a<Go8cK`e5s~T!6
z7(P<W1ap%nar%`7ZYW~>ZYFbx_=NTk*5FNQ8XBHrtPcNvb>9H!CpLrMdj*wYPSPaS
z6N!Y~NYCm!OEi%fkx|2gufP27`wjdgq=S^Cz5M(Za)ufj)TFwBft8zgI)G=r1;`0v
z=10g6O3b;WAeEG1JF%t&;r+ou;!T!kkwD^-q?zPb?JQ{vyMP9QAONP2@R}5lJz~Jn
zPs%Eq68hiQeA*hN<#Jx&9$Mky#@y8eSudcEykg+X6^$Qp=i>qYdj+#Mjp7mt*4O<=
zqBb4<nq2-3{4J(+lD6Ot$dvH59iA?Uh92(m+ri9S71_$-d$>o^miXHL-q6yCT-}bi
z!>jN5X2s;>*fHO0X=`6SArk2JaS;I|p_1oFXeL7-!q_7!8nNwzmk)j4#0kg$W~`?n
z52Vhr5ydL=N(4Z5$j%6%*XT0<Nogon5Xs&sFPFtA*p+F>MMT?KMP0+0k5I$iFl|}&
ziv!;w4ti8cwr;e4MTF-LrZa?PvWnYNq8nNtaG?gkk4kbCFVF$@qT6@x-inK3!GV1j
zP25Rf4VU3ey_b|kM<SSm@3jd9C@w_Y983ZT?zr{hg*2gMue##3jx(q?e;M0wGytdI
zcARr{AmMW4XG+e0RwSZ!aDJ=n>1_q$-t|PE9a($SAfGHtoWlElh$6w2Xy@>Uf{-7E
zlPnx8hqS;sbYmNTmuq1)HO`wq?dWI+y&$WC&mov8-r|Q699R^e{wq`wwhJc+cQ77P
z3Xbu@l9KC5Nu1C;-iDBV7>YbPK`$V?tcreRD}y%rm>RnP0iq6`;Rpp40Oj*0L{}j$
zO9P^~@8~U{%hYz2#YFk_;|ozF;-oG9fqN4g>}^C0goE|!`hR;c!qAm;2+f_(2Mj?P
zBmo|tl)fvP;cV)puNcq&p}Y*FJW(LzTafiD;DED|G6=v(Hcr~HYDkc7w?9cw2S$AI
zBoBs|2(02~`Dv)$#2l-6w;r%rqr^5&1bV9}0aT4yBp}t>w;{gOW}r+!Vi2-jt3R-2
z*%I*4jF7N!I=Za;IvG}z^FT^YIXT_=85|tUL~{A~`~Y%$EaGhiW8T5h+qnAk=g%qF
z@pwYiQcKq%UNb@)U2T5>As-&AAW1ydQ17SEZ}L%4Tz6Rxe4BiP?yv}Y(SqlCA!;h<
zaI~x2Ik1w*k_qs*?+Vx?9g;I9oY7|h9W5&LCE~*Zur38BvN|)c6kM!3H*xtXlbj)1
zQ&VGRW>zB|x;s_!9b<0J^4Wht{bD2MIkBbzRh$7tqzi03JMekA9yW_X1u>H!tF9a*
zns`)UBJdnT^#G7(&?5XqZAV`U5GQAaHUMeJCmkJ$h#-R1Jzu>TYuSgfm5dPhxJQ>j
zvGYGIz=aExB&2YX!&MSoQ=>>oVuFH5;K8_oW*~Cw2;`=RFU^JF#e>9>lA*_0O)Lk<
zFgzyOb`XORN<R_K3@RitE^ZUN4+JPhGP1F`3n3aU=Gp<E;H2IXdqQLk?=$rO5l9fR
z14Lr~kZ=R#pYO>sCQ^%jEiMv=5RCD!U`RmJ@d}U#kGVlvVv*y&65K4tRWcHwEx=Jo
z^<gK_Da&KW{@DzX3=S?S7==#-1Q?;a$5Cu&XBUM3NX4jISX}%H;$5UA?j-OrK!NJc
z&dy1$7Pqz5k26q`a`<8i)d$!Vp3UU{YiZyvI)V=ap}IYU5_{A6t;ZGQR1n`3qyco@
zfL4a*Yd4yJnuj98R@kPB6%&?5ZNQ&t?>F<GYq_Ai&hxH(mh=;S;dS&G=o}~k^a5)b
z#%PMoPlqQbhHMZx+!@h&Y>$KZm5hWRe8#%7z5~!E&QbtQtyi2RZK0v33_=$hYBgC}
zSjdf@6_Y3cO(bd*J=d{5o(`=10y!bpdG}xl!4%}mTz=oLY8;5q%@z8D#<%S-sF^IN
zj{osXzXt3jfLD<#qZ4_0aW~o$M4ct$@OcRCwEEA|M}Y2`;)Wg`MV5OA-o;}LGvQ_u
zXksHQ&zNFpC9;9KQYxo8H3Pl}Sor6A{kC*7!Nw`<@ib)d-2PaJr%kR5_YI7{>4Fo;
z1qwowhb;h--TSUCwEAUA(;u4Pn`0so%m-3VGcOJ~IXW7FfOLJTpYsN5orZ*z#j3t5
zDEj&&0SIOLuAIlfg5(9tAhh-c3pRM(n@Jd`j?EOKZ1R(yq11=7O@I+54l>DOL|@{t
z4HFPT3E+%-cyVd=7_b~(G^kk7q@b!Qc~D;Uh~G_P3PN%(0XQO$nUtKFxd|r+MiUMN
z-zCi7{O-_bhkL(Ckfddoa#Kb%$NuIbhG<BO2v~?@x8mOGKLkS^RfyU(ps~O@^4gv!
z#>byT<^DQUEe8Q&C`@&?s~Z%yAW4m+eC*h<(w_ig0UXwU`osyG*vj>gqeQ$V`(*(6
z9)vrl;SY$YD=`ybX^2zo!c3q3pePt?tbHCd+-PI%JD5nwj~I)y4d@cEwXk4819QEk
z#4F|p(q9*_ksRuR$OueQab=i7E2A_}-6eOmxTuJX1bSRG+l&`eCGZH~tG>dMr>|U*
zB}jcJZMc<`xc-N`mXsXi(|Zuq($XS0X`%XN6-IBu9)OJmWSAin+EsA1#4Ju=W`ty@
z`~<y%f&yd;2%i{@v9KX>A9CHI){cqcZjQW<>G`V?5)#`xMT1-csS-`YfEVsYc+nQ_
zj3yC`>0NNv)L6k}q;DMsH5-3@5pcMTAXVCJDlY6NjM306>p1uzXQbiDHT+ct6}S+1
zJNR>0vXj9!841SVNY}u8K)4x*T$q`|z>CPT5^5Wpo9jP*{GGkewxD{6jpzZu1z5;E
z#wIN;DS?<<kWUP}<=6fKF2eePn$=Tgi(Uc>=<AS`4LtpUlG>YZC=$0vN{Wbt*q$iw
zk|si6emFF(qUtL4+7hHYQLw>ZV2!BzVbJ2jbcC7X2%@wI_c&gz4`(nQv>k{cwvd^6
zZ8MyMfO<)QEGHk4;Q{i*=n(KD)+q3k(N{NV3DBqrHV@))1mz?y16ssgDYunWIOHyF
zMK_>=8$$YuaRlGQpYa2G#?}qOjEqEtcgC4s8h+J5zw3X`8#pd4N*g>EEgc;<RZ{9O
z&fF$?Ap|oL?H7#CIeUqd^#O>ue-RcqLZHpObwY$NF@X(pN1P6DdxD%i+9!uX2Bj{M
zOiQ%LMKE!a>K=$YaBn7uBOJw;1R2@W6XnGQzlk7d%lo3<-&D`G_G4o~UO3Oz4p?0i
z+96OPJaO%K2ZU|{;5s6Hc6bko;_)psRXxd{0679AhfZ4hg^RCpk9mM<Q~|gRLJhbA
zgpLR&Y-n!Yg8>hx6Csp<1&HvF0kr@cMh|s3^CpJ;j_CcSurlfs`>wNd8@?E#P>HJ{
zK=XLu@WsL%pO_%VL<m}d3L6@31NcaQ;+Uz34iO|avHIcM-%q%tK#cIn%I?C*ht`yt
zDGO399;hiWuzp`&4u1TY3mjtOSwR7TppX#lm`AL=N=9bVzm!qC2&3Wxt5A#%2$b7O
zlwwV+{I)=!IK&Pex&e?Jd96guFj^&UtTe1@btDYoB?Tk#qfF-~p?M}FVF(PuSm<)V
z7Z0C6_(37$R7u0)f>pySDrz`uM@}NR2JwZ~H*bjR2Npop^qsNoNivbY`vAw^`CjAx
z;-0<}1)c{>UcFL@j)~#n=idaEWB$*8Y$hG=<?oVbEx>x>@er!)$seCSeVRG+`1v@k
zkz^FgYUxT~hDiwiBvDxv6?YRki~x5CqDmm9&Baw*NDj_0ya%iX7@A6@W}@3CzV_y_
zAPp%u&p*!qAq^1v-h_=u1^a{m&e6W#i;B9AxGg*MG)Ld8;&J1d>aL5w>gjo=Q;z+K
zhE@jlmQ?Xiz(H~D)Z+jrQt3Q=pg8~m6N*+()zHvTPY)k&mH8wY(;_h_N{SO80v@f$
zm)E-pT@q;+VPs2!%SZ{mQ#+pfIM|rq44l0rAGzGc*|&!y^!Vzb?IK(XL<Ec)B@~P>
zCzFgM>AZl@C4S9L^jFBcVOIU8bf0ahq<8M1+taM9%}8HIPkrufKz3qk?;(viC~m=-
z5ML9id<uO|u2`WPzg6;h#3Qx#iiwL~#qmENG>kwU#9`ki8YSS2(2t%DpD~dReO6ME
zyCl0GMS*Bx)Xe5zKD9-rm)rYdDhOJRKoWiVDaH{3!ADMts{CZHr0w+C9|!}wdi6-@
z<V~LwFWU5ZDuH!CmPeEiyd!ly*?WnJq0ndS!SY4VyNQHh9g#nxHYX<8RGL^_pI(kU
z)x_GJil&?t1TupY@G6M}`}igZZj@21A>!RnFTsug@8Q5C`*m>8gmtpdvse0l?<ZEn
zhcb+$@9$4P1}F|h$9(J$fyd)ufqJzk%R8I!R1x7r#HA1`^7_`<iJa#7`mnTfq0fM}
z+^S$)mv{BEMT8F*BGDo+Fyroq6r0~ikvTnc7|0-TWe~Jyb{OCy7UfZW8JQ;AIz}+Z
z=j8n{%N<+@UaQJ#g|%xd3I{?(M}Yb5-1H*g!C{)G4_W&rAJzi?CZwsPoXJvj6x(=T
zVbV4GCL4$xY$B$`)zjw6XTkH`1fhl|h#;Qu^9TzSfwEv!cffE!kZ0U+?Tr(%3;!k~
zU(ZL!EPkq~;f>l$_k%wlpPv`Kn3R{6ri#`Dxqm_uQc`?3Rgk<&VBgCFq?YdPC}4z|
zz{oOyeu44^Z{1o~7X+7QzIU}w#(hYev_#@eAsflrss;pJ53Ux<AQU45Ol>(PO@NA=
z@e~}PWdc0=Nahn-n!)EdvV9uIuzKy6r=XN&tYqJ$jNoGo{OH#qeF@y~4?k56!|P}<
z_*1C&5jRF3J%}s;czLsn2SN9q0SWo<Q)_;UFfxKj+)lEp4e$@235$r}M~g(RjOID}
zT~f74c6#97)nJebYM5@Hfm=M>cnTU!`~HoXjz<c|ISvo!7SyC-j0&BkVeG=h*Lac=
z!`u|sgtygaX!f-Dq8LdM2M?yfq8m<}@;Z2-!E$eVd-p+xhjHXY_1z+q?7d)VeQ**^
zWyIO=q8DoG?99aYj}wGtQ!2GGVl<fJ2yBbynlh<nXy_3Uk+KG~^ecW4z$!v&0(lb6
zHSPd<LE3HGGBGtK$awJL5}&v?j}&zAus^&wZ%FzI2rCs!ZW^r{)C8fAB<EZn5iR*c
zKLs*aI-~Itz)WNidDIX8en~w44t(}k3_3k%{sh~TAdG+?R~rbS{e7|^%s5jW(x20j
zT>&qNY3U5E-`K~<=nt2i)#uM}=3J%_*-TDFP5lm-&>6IB_T=dS*qOMRjF7r00J~rd
zQMS%Le@m#=2H6>cLPBsy<4KS!qaV#GdWb>f_z6WM>Nz|usF;iM2@}H!O0JB<HQ)R9
z#?Qwah=JKn#*O(EPMkw{meU_!6g*hTw#9UlWvYjQsY?=oPszatqaV7RGtp=4?35vz
zazrelK{r}-z>xhjG7tm5MXw;xbo>vP3&->44b|1vdvFN*#ObOLVT5S&8UD=JZV<P!
zg><8s8)CW!P;QMh*te{vFnQy&J_OgNgt&Ou03T)<WU%l8S;Wwc6K1CBz_fUKPwRv7
z@>Nq#>F|HjQ$SLS%mk|dO@uS--xt91{x_v#Z>Qr$CDb5$(j^X&Ey>V1yh9dQHiic9
zuIgCsPY@>(iI%vuU0qYt6s<dcV9tjtXmpQ2xw$<mq)ik$Gd5^Pp=t?1Am8oV#Csh5
zO_!$tD7NfDI=Za4kZ(NoV5-4fgR+{8{m9mC+`ok^8M4j1-x&ruE+0f8Kq96P96EUv
z{H|UYlwlk_9@D`3Po;;8Y@9I+A|bu{aLlEZ?<etXnOsMA?>1VioUiFlke71p3j=L3
zrpYE|nQdl*W&;D=9YR5fx4$>46`uF{`s-lMP3I*Lex;2qOGq1Q5%M8ponrJB?3gpU
z|3ld6E^&}RFtf4g5S$6teH40l*n(~W^qTtmwMRC8noEiF7Tde=Y6&~V|1?9b{Y)lM
z4ogkuJWWr(fs7Q>dCaIk1jNOQ`q*0`^2(d(ea;E@R3=)HF3?o_`S~?x{D@wcEE#%s
z1rvFUnO}k4pgyNB`k(D++x;ANP=sK_#zLgGq4hfh*y({ZJ}`yo6*P#Cg1f9gL99>h
z(}iLwh%Ta&lB_ewJ{Vz9p%2OE5-zL(p_K*5XyzB3or0w{rTc;Ozz$~$k}@ey)^mI_
z8MmrxJvFB-@ioF@>UiwJBu$)1o@haCVa<9S+D*pz5~f)6c?fZ0aFlp|@beRFtu&xz
zc+Y#dl3C2q-C#VUMVSGs@Uo$S5@O;NAV5Uth-ugX41y$m*dO64OvlqF_f=5H)N9`V
z*H6r@mMPP`^2cD42-t<bX-`XQ>r3!5?hqU8PL7pi!zbaI!<>sN<cMh?fD?4%;VzZG
z!JXh3Bn2r1T~NILAM)858c4bzA|Sv}dHD1qv;vg8qpZam{09!uVa9m{GAa^MBDD6b
zY;5+#xq}?ulc!Jje|`QQNf%fRSJWJvG@%Sfn=^am$H5?J^jvR|7;~4}VK!b+Q1BWw
zzqbWAl*Ke}h@0EWw@5fV6|bbsqW*a*tmyPvtjp;mb$evk!isJUCx|v#IymueR=E%9
z#u=EF?cCg6GoB1nU{a$50z$ABxY+N+Se97RbSN>#g&$L7^nV-`wFfrmt7$J;eVs7>
zJT<7I_ZkIpB=Bz>zxezbHz73oiV=)<ZEcN%v1!1R5b&aJP$FFt?_~9*Bqny(**T`J
zFL<0Ic`t@WoyD3SRRH_VWhw}i!3nFdg_EAl;VF#_h22+3r`__<R&l&eLVNe<E;heC
zjzQ9Ql5!xaoMT6mrK(4XYLlgDmkPs;(Cr}&>wr0!iHqcj?_vx&HB1$$c%=IntG5m_
zYu*%J5-5Am<M1^5Hh*dt1dmu>T8Du!;fXCZ5LzOl<O3+UdV6TBAyj-x#{J*Ndu~!T
zIdf(z5^xg@4ty0jd1M5^Z@8c_g-U%hw6>c`u@shV#wY*5>%NTE@XgmnqFk^KfeHgg
z76cKS<}L9hE$mT@fkAMqZv)EXNirop9c>Z&2dr1O%#jaJB{yUBaso@FfX@;pn1ns`
ztgexOr%z&(TtVl+vsp!uagoCP2L6HX!<BfvOVgOm$p?e-@P!EooETzI89W_X9lUEJ
z_wv~zt_u%>fXT>)_cBtE%IdoCe9jGi<Z<YXwQ1C?9nG+q`m9G<mQnNGzSWK@DE}I_
z-TKPH%Gb1r;faaHdnd!r#J~m(r6Hm8Ftf4}+<c8eU5$%;<$78@L2wj3X=kg;CD~9j
zkm_~^oy+LtI<u;0(x^+r%~>az^~kG&G)x`zthvm8JLwx~h0LAGj+&cH*|O1RLSLXy
zl}wmon2H4G9sD9W+~>4V>MAnr@_jimIq5*$0SC4lx17;rn9v+ym}BhXB`HxDxkvv)
z7y5pck*^f;P8A1)P2@yLPzC)a{Vf>1H9{G5-zPea!ftO+1nF7!7pvr_E6Zd-bflVG
z%hL{-6m}dztt1~317K;S>TnSR0!~h11inkJ^rLOy=iS@xIEG-V8-vycenV1BbaYn1
zr8qJ-szOF(Mor*%*ogX}ly#J($0X-f=KkELm#xVP>FbV=j`hCI-DF1=w2kzj=irnn
zNyerqFfP$(yF6K)8p-}V2M`&EmAO%jcn}PVlkXOoM3^e=TwK~tOE~E3VaCCn7a0?C
z6Vu@^umK`76G8<6+|2f)+F(o2SyI5&oE>EUNFPfky^5wIi?8qfIXJY_+qALLsMq}4
zwHzqq%SZ#;j&?U|8q*I;lMa$ON~er5Z%)3{SbLQ;_^e}ZjLyk(=L}H2uYqu(yeb-{
z--zTF<Z{Moikt?iFi`5svgz5q_V#vFFs}H8xZ^B79zIN0*LbIYS60G#p&57mOH^;S
zD3|E;oBMp7Z_}?I6Us03RFwEv@O(WhlzZJW)c1T+R9C-tfm)fXr{H<+^SuRSykx&K
zLUzl|+}SIf%qw0>X}9pU_K9DT+hJrNEM#3jH#s?pjQyKHbXLoxm)Kd42ymwgXSuAE
z2>tQhmh)jXlyu*9e-o|pCeouFt>!{mRw6BIh1zX*78@<9Mg{+H(s4hIU;{naCw7a}
zXYTj@9qqc{_31cA2gQ8@2cJ=q&?t1LAr>NLZP2Fqr2n~6vcfW6o}8sMzI$3)nwr7D
zJZY{Ojp-^B+`VhyJrP5(dDC~vpuc_{j<b!8wX@>pfmcZvDV*P=kOa>O-%rV*(i?X#
zJUvVb;sGgNSXz44od<9M5)7#@1Me{E4cP1M{5W@JNoa@YrEqqhBSGAoxp^^+5%VzI
z0EBc=kMFu3UvB9tI$*UkXR=oWMMN|pBN+FTg~Dg=_7E8^#uVONm89YMI~J_ZF0GA_
zj-Gue7o?KqvN%VkMP*bM98>f}vyI2@sf~}aBy|{L@Mz$&h>_&SO0wRAM@KleFGU|C
z!4}uJXGbwqE|_g75;}E?pXBWcDj;3xk9F71FexhGB#znwHQsx041t^0a<%7+Yd5yC
zx`P@DotlbBobJBu+L{?A@9*7Kyf%HT==+^GfAv=rAqk0hSj_g?Pm<QR94&Y;;n%Xu
zbzNFNtf*qw=A)WSYaDl6TFI(xJbtzfbbHogZpnqE>E-7$YLrS!N*HM=j3jkK#|x#-
z{3XSbSbw;`%E-;7Mb`#tnKECLLR2_}+yVjuSr5Y?DKyL9)vYz#Bk85KcMcQaQNhN)
zs=JkY1;;E%8_Hf=wg15^lKTtD2k7$;<cE4@J#w{dNwG83&H-GeYHweEy!tOtETV9e
zIId0y@9aBN8Z><+zi8T*!@Jnt#bpb2+38sK-q<kIpU_rT#v^=DFYD?^1d!U_AB<OJ
zCa8{{0)r?q7zWaA*<ys_{0Ku&tL65fqokTm{kQ(B4zHO>N=JJCh~<t|t2OG+3XybV
zIc90v(P9>s!5Mg+mKAnVm{BRS5_UfR5-Fu!K(WG@Lw}Ey<q{6^9+@9Mv^usVUV{Pz
zKsF)5$jHcP&o{{qqY|Ub*?s;W-q-iFgQ_q=Nr(<tOrs2_GX`H*I5LGgU81k0<vb9g
zLy2*)4HM%YUQPPbQEA(T<w)H^{MI<4;q<y$v=G`~gnZ!rM8%1XeR3|pEOYet?C!@R
zVG76Bbrl_Yk8d76ZcxRhM;hLA)X--AA&2k%Hdj{17v`|OpdVtqEfA*qe*92yPNs5r
zLwjp04dy3N$5AR0@e|zLgMg8txN+ap(omHg8sJbo_3y-Jauzo0q$|`ktRpAmpi&ZT
ze)LeH{KU}k8!B)EnSX!E2N!GaqU^P6`$l#<=?tz9o4$98<Ffhf%@|*Z3t{iv*=^`R
zmTh(iqjf!cZHah2Y~PyeIb#CWivIuoh}3;SfnKb;XS|=V2Yp^msDBjcxp+{7&N#c=
zUO2e8m351q63I*X=tsC~$dR1Ix9VnLW3$`d{v0nPDH8FsMf1>iGXky{o}Jwqq9#F}
zBi{6S^^iTGKX@eRpH(p}IbtNg<f<P?l6%=9^M1_2%G|tf_$D}%-ay4+BPp6cWV1)q
zzlFmY2LKvgtK_t_?(vIHl9LgG%SWOl{9p)`G<35DeoRo`C^+m#80<-93sVgQT&SQc
zMD}b<OulHQLx99C%RDQt3*>(X-qmi&)wXdXDWB@(oP~F${5Aj|@UWfl%b6M(AsGEK
z_D*U>aPyvR%r-1&T8N_!IA@+3ZwU3<5JMrN<Cs0hN@FJ#JQ8B-#doCl9*skUE;$Jk
zYUUAIi<|@EqefCL)+7!RH$EMsBCe-Ae_t*tK0X8l3_^%r0^2c}0<mBY2o+*Oz?#w8
zd+#4KTGVO(_xLlL=<4|B>rGQk?~Qu~^UZpc{}yOPKwsRNAs79rv<b?inv5FL4ef}$
zzI#{yyd#NM#onF>MJFh5a3Bhav_*I65ikLkFlEm@Edy0aWl4uwnz5l~R#OG(LfZB3
z3*82Ff2p{+>OAQ;`Kr-G-1mcYH8jvufc;x_n+Qn?pAJDgMpXTCnmW3=jr2B9!lo2#
zxM_|+ZVIs~l*JVgh%@SByxXBH^4K#>L6+!IzwZd5=eTv+Q?-=2PMDs`SM3FfTSDT8
zag=EOs>{O~>bJ*|gb4<utn5&Ex)54;;1W0D<9FduS$xe@(1sL0!g^?WgdzWGWy<N<
zkWriFflv_!WtqU8kFwGAk*x=-zGO>ip}iJ42F3-%q1pV+kj-$_L68u(Y52{Eo0*1k
z?;!O2*$<Inb-|lGFhE(Y1!wWelVkwsQ!+DYPoHLakSP5|V<+SUz)S^%gesNQ*01j+
zSb!z9hc~Br&MnO{6&8_}JOwN%IH}%>Vp$)s>e=XufyDSk)n#LvUNI(hjyD=FY3k@m
zPC+^4DUd0Ej>@%8hKwIl&dBx&JfcXg8-gzs)N~lK+j7?lYugW?1Co3eU{LWd{B-;k
zh-YNGuCa~zt%rjrR#tt|e(G`-dbCTGRY+Y2OqoVs26Ev)5<r^lK~pBT1MW-v*=XZf
z;6B=$m9H)*c+O{=my9LFL+ZEl@CwrD)DTzQ0ji@3vLJICahKYXEBJ3AS}_&iJi_3X
zQkcweYT}f0f8REnJo%BYcH7upg}_=|yN!Ar_UaWc^A47@!a{~Ct5kS!ngK_6e*FA-
zFDWXtYS0fd-c;^dhMVf@+`OzUj)Vh2P4gP`RMD;NX01skH2(eyy~!=q)L7O*nU(B5
z%hR_EM5$8zj|WD_#no3@vkZKcJG9n7YqWR`=1C&#z~HE0j2zj+;qJ(yW=u^3p*md3
zhl0Lv+|kI`TBrQ?)oHr7jLHK9+6?r<N7N6TBf&&xn+>6|2Fav%n$5TMlf~;%b3NWr
zxUZ?!0>RR`bM#O1gvB@hy#Lylu}5BE_xaqecbHp#{)mi<qAjo5+SH^0+2M_%dyHGR
zI+luT%vIeS(7U^S{r_qKmOrZ&y4rslfyjhgR<>L#+bqH2MBKlFV@d0a>LP=*>qDA>
zT+-k66uZTB#ZX{D-n#nwJ0S0xv92Qme1JNWSSj<b`p5nvsgXNVWw}%nd(ttT?N(k@
z6$ge6caMt=(-WcM$+q(vm3t2OA6CxE-R9sh#&?MUh6E!^%gYaMuPy9v9T-ci=dlyz
zK2GvY&iUhV_)_dO+&HpD!5XB@TI8-4?(Vpk%1^e%>v3IEQ@(O#^{1yb8x=QW-BCsB
zV+ySMW7>vLhM=W0t!6pb_ms%7rX<}ZJ@=G(9M3><=;>F0KL1}P|DDiKB3@%ip>lLE
zyZ*cL`?BCCAC^BWMMghNZ4+4h#l2a7Qu;g;*N%>kWlUY(Syp$%nszt;;Rp_@4(g+L
z&^5qxAVe()qWY@iFI1X`l~Ik9Z4G0br)DhHb2B#qsn^OdX=wcXQt9M~jNdQAmiD&D
z$F^^KB1~n2u8{h^?{A-OAyc*opqmbsjjo~#fjV-)b||uZRH_?z{HJG04HROOX4ULC
zBUJ=rQo$(z>QyP2vvyy9`-rdTp!kb(4*f%)J%%IF${0B*aH8Oj1j+b>Lc)y=%*HI+
z%!*84ygQ0;uH=d;kr<tCl-8JJGxwRtKKRpT7RGv*VaO(r&Fn7dzmz?J_Knl?9#cP#
zMhh_u2>Jfv`#rt;C1%mvk3NivO&EpOKfi5gZLay76gF7oun9BR@%fuj=iC4pN|EDK
zut(;h;U^0tx;pw0HF5|Ypd|^q?GE*LnWWdtMcvcfF;dlkUu~fc{oa08!LoRkJ-TfI
z%8YZ{wJo6}h1AYK5O2@VM}iod7QH!?A2$_?udU@c38ZP#)z8M8zU3+W<8WZZK!vx)
zww6ZePZb?X3KiAO#Sc=PoSbD9G7ie0hd{7iE05omZo>I=HGtxrTBCfP*nW~5NsK+{
zzTru8T2Yi+bO+j4bd>01eGcfU!Hm#^!HLwn;*}_yZWvAzwAbvHJDB7V*g1|cL`Fy7
zSYMY<Z_s>|dhc%3QKF*BEzdSTy`*@f{HE;Fyt8bJOR+IABT!mF0^wiRv27PI?)tE9
z9kz2AfPj$mvQL@AP-ThOmzWW3M8DLTsP>70$_?%ss|8M=Y@zqxnFui75&Ve;8+8UT
zk@Y>&O%9`FUCv!q@qH438)E_*5s7P-XmC7h$%oAw#lyx-<9qGtf-VuMQhrPCPcPcY
zpl}1GPJ=$ogm(+Mdy?v!*$t8<J@z68CnsQ{U=Vaf(rNeFdGRx^9!xr&W}4$W`MqVK
zrR@z`57lsoI|{Bq;NT69$kBo-5=Vr&kS1wkQ{8HOz%V6Qeel-0`<*)vl^q*NtJr>p
zb7N?Q>!BjkosSDN0-~b9ckj{*A9yVuFe`M~;g#$6<)8Vc$9BbUd)v0L&++u6VqQm_
z>QUOI1N<kSQh&L|J=oXXU9y#ymg&Qbg23<};8wmPtP>y&EAJMP()^F~DY7zCkGxxI
zyhnh2p)-WGcALW<)xAv`@|WbStpuB2%l>p9j&N{yzi>m3&o%mEwwWNV5TLoK(nVKy
zOwsz^OGscPm!aSuK6W+;pJSvf$01in>*Vw#e46%XwLe`<R3h#CbL+_1M|U%q1z4F$
z7s(&gWlHBZ<&rLf=z!EZj;&{L9X9mupbyCFq)1FyAU$Rz_TK(IfZoyJpfnTYvy8`O
zE}x6@>w|OafzRh(JzQ40mX;PZkP+bE1(qBNn7!b5S@28w-tbkSC-Wr>up3wsW`dGq
z*OzKJ+R09iGiD)iYkZT&i4_kL<41H>J{P_y@lE(7Faswhex;Y^NJ870m~Mg@(RV%g
z@dFY=*nl;utKV0OQC(9j=;n#x$8DNqBq$h0hV7P5@Y|5SudRp=)Vk6lB7^-~BoVjt
zwyVp7S}eiv7}452J#0I+Ysu2d*)C|_+&`osPVsZn;`<Mld)6O26h1HZOQ5loc|7!V
zCoO4x6{+b2wzL?aXpDgroWd<2i8R*=ch|YoPceye?4S^&y++(MUL;))0i9pz<q=xO
z8#W3R_HT}x-n%$&Gxj=?iVSTmeuLF}|NVTu#O%vo`+G81_nmTcyz!^<h5d(dmb6n3
zOQQC?GT^+m+MWAJt}Rn0_a=GHOP~7X+4t9pWE1pLHdN4G*#WJ*hHmvP_hw(xx{~iQ
z=aAWZMfdf@Ua6^6^uGtvW~(y=RMn)&+Zfbix4x`)uP(b{yEwUXhq0M5!s65&TO+Ga
z#aMqZshE20oZt}<m>!<TEWl~>@wb=ti4N^IZArmVgOl=;0dw~YV=Ld^j7Wl{EhzU#
z`g=F0S?gZ@N&i!C+<0gDdG_tgFqyN3?S?I!kKyZ{4)e|0H8%ui&4+J2OzL)#gsQF3
zzx(`P!0~yP9ltc__N;o<7p$zV*+g`ljdc>y;if2~8>m9?Ah^tSuz2}rsKCUY25}|`
z9wA&OV@*UUg=r#<<!veV0-UFf7-UG5pYBGykUpH0zqT%5n!7snyFi@H3xig6^lxCp
zIq#ego!9N&k$NG0_*T}P1ai5nLF@~SyR3w6d^+CGz-UuxvS~kmvzDihpmyv2z56ng
z1-ASR2v}KN9i5*{-00tS{(QaUrG0}aQT}TYb&e7iJ#&-fK4+#!*)ES%Y;((cX=en!
zDA(s}N^t(vqh?Tf2QPoOYjl&6yGm0hy}eK3_r3M}5tq^g=*&&Tq!JtLU)oD;_xH0c
zrEObNdM@VqT8&NeT$n~$w%PA5?lLldD}~)t7jBVfzq@fWa$Hv{afs?dIU{?)$b4rF
z=ggUb<iB+KSzGSsjE}2Cn4ajyFP)zLBxU&rtd}>eczAlY&@`QU5fzi9fAJ#yvG=l|
zeehqx>qn9uem}o3{~;u?;!=ulzrl^ZE#=cSZtp9uAE}>ja(iypaQqYbuh=glIv#!x
zFftFBF*m+|u#LE`3}aVngAgHML~O2^9$yr>9(7G@VXPHHWKiyFdCaeeT>D=C0Tv+m
zsq*4*rTXKe?ZboqpAS9%HaIDwXcFJh&;Z6w(bhP6*Sj#{TbZ3-9?m&;`O_6&_JOi5
zniUtO-K}5vZcs!NdHgiLZ))<%VrXzf#X8XSCZ(-}^}Bp|Sy_#xS-F-spM2k!`*KxY
zoNxa8IWDonyVbDeaQBm=q4u3`L^c*!_ze76OUefeg2KrIBJo}JbF8qkKIxu{#&r!x
z7C}wNu52`bU1b%s3j)}bM(P}4W3{|jXj8K3>b95Oy}1w$Q&O&fI*Hti%U`OOl{ALL
z%dDs;U%&Ty_U#?P=bBaI+U^X!3ASP!K|$y1_Uj+R8dqCP-(*1E@4t2X^_O+UjgQ40
z-+t@=qyJdu3L6v>E;lL1zMakb{%!xASFNU}2g-eI|JGJNJzn~-@a<^l8%60!Z^f0%
zN3L8vB2RlLma8&V{{H0{oyB{p0z5!F&57m3cF5T%q=J{YPXNA90qE=mhfmw_M(*RO
z=JUO3K-FsTGP!M-?fUo2X3YC1BjEcsJxg(Dspib=K<Q(+A#`4x$H-$T9P1l&K4U*E
zPitJ(BO8)U^$`2wlGJxl-yyq9`P=y2;~mcqb6zT0?fhn$*XOyaKrS5Ba>sK!sftdI
zJj=GWhDfOGg=DDgL)~jA35jees8Lc(vfl{ak=QOH%yZygY@M<Djt`qRZ^j7t?vFjh
z`#dkN&JI>pSZ$gjp5u&<P`J=YwJnRN%{aJ8WBTB<j6<h+bPRGh@8rCqz5K=Jd{6Vo
z#ADt3x})Ayf-!e?=(22Q+-?4aEa|GVUAfxP=(MdTmd|-p#o`42@aW2yUC`0Q;6{rG
z{3g~HU*ZN#T#5S%!VF{Pj1--W4^&}sk+Ch5%i9=B6NZO}c|{ZSG@j^{Wq#O{ZCtN)
zYwNV;T4QL;xz8>=Sy#KK+!8ka2;C5vn&+jtn#IdDJbm|~wfFSfyVW22wSVbyONl%(
z=REh)St90?<hw?7vuAd8cXTp*l#Qb+NMHd^Ln9u>A$uQ5N%ky6un0&u3*nuiTUMAa
zQ?^98@6@Guxak$?nbiTDSl+eHrNs4w`(&lIh>CW)a6LVEy0fe1V;-pEba?~2m|RdI
zLQW-DWWI+Ua^=y{$oHc$)`3L1YGplY`|a&H7Sh)&Y<;NDpPzr1Z?E({NzKHp&}=vm
zd8Xu-NHg@i<+G5;69Z14AbD_AjW)Q*dpB`0NGJF+Ga`yCXjj>-O6NTz!sfNt8gw&E
ztUY^$U<8>m<<lCCDZu0TP<mG1+<c-k1F)-6Mwj#MXhZGW_oIYtQw5LFZLP9lr=`tP
zUw^L}GpFZx%IU?&5VyXN6<V793TNTBoLbQWci5A~^XG1;1}#y+$DtwN*usE_P9l#N
zdP(FYh2oAaA|xG4X!cHO75!`V{A)X&Gxt{~X7#!N7Z}-X`F`qxCt@k=4d(c(E<R8b
z{Cqz{N$mHk1x>DXc~itQt2|hQk0J5@f&#P3eNkgNb&}{<MO?DVyx!8n(wEY>M-%rp
zWgh)tC&NK6QBtgJ*vYN1=itV9>O16zlXfN<1REr7SY-bGEt|A6&pgy17qNhG-#?t}
zmv8y8<KVa?JL|y!K@EqHN-igoLeSW|2K9Xm^joQKWtruCWLEE@z?BO`pr7&L*RJd6
z+&m$q+8wy=L0s%8c;e<hzOkG+i%+X$t>d5h)B9&a!%AWi_!XCV*HrVOcKQ}^HvOKQ
z{$i8ZBqgHoJC%4-)3@;NyRp|c)yE}lCLnZVRVlhH36Xmt;+mUbtWg<KHu2mtX+K(M
zDQqcR{@~|^m2j*|gqL{Y0ZY#VP{2n$x_89KhmWPZ8{f3SSz=^zM%T`c?ix7-Vl26n
zHRyU=c1R_1CU)HFc(l>)gzufmi6gY*T)m@2)J5^6wZ-YTSx5GUvuV65Gp0a*xWVMz
zZ#wcK@01p+&a%o~ny)|i!uD*R{IHZ=?jcuWnk!^4Mk{!wShHpVUbgF=>B+h`jhTJ!
zCr_PR9S8Rdwq`f#?=)rkKD|gb)jHY=Q|tG>K-<xg_+{OhwcwF7{_p)GS-IvTRYNdt
z7~*DY{yV9!vB-$d;n}>ws_hcf{$^i}L}^!*o|k8LJRh?aJ=T;-Tl30JEV<to2j45Z
zw`u(|UA|XR`dRdRPrhY`&~Tp^H@8yPmc3$y)Huv1-tBwm(lO@tOV%JM<V(*~$-~3`
zvHqL>@~)i_R;ai%(l_+N-E}_k%;)FcAum|G*Y`><?$$HExYX-pO&=EfMELJMHZp~)
z$A4@{m5nE-n<SYqPlya_Nrrck{ir<m*UPQo=~H10QDRRTaaSu(-V~R4o=tPlF++yk
zyN$D#gyiHBp%F+b{rD|>+;lsZD9@H><U<r|vEJ%ddABmT0&n-*2IO09`epA!owzah
z=E;GN1J6#$g@m*|wJDE2XMNbl`+aoJAt(7?dtLj_zjAwi;!K~?RhxHyjkRW%Q`Ba6
zo^Uu$cBomB+t0h3aUkyT4cwEd1y|fB(MR%E?!UEq`B1UqxRrspn-sg`o&@9C0LN@!
zPP3i>P6eNyFBcB)bJ&@>w|37y%G3mzJGrdQd?%K-#Uva#Z}&FnV~)S_b#jU_uN8(l
z4n@kK@o2MkKY!h4Zd&g8U*wnfvP++($baA`8T~})Z==X?&f8-~83!4ht@W8iNz3}h
zsNJYN%e2U0d(h$ZL_hZAI}uzBCw^I3>Ei5=ZFa?OTvlx+7{iuz-E}h=Ps0wacq!wg
zfglOn{9;St3)626QdhT?s$A=u(Y5*B=Jwp^+Tq=jdy7-IXm5IM*1bGOF=)NAsq&?r
zd+h@ZrW$D@Zx~f|tFJ~(2DYZ2uPJzUghSDd`(kC@yKiwe1r`*X<F#cB_x)%V3<F06
zkK4FdB^k_V3T)zw+BGp<qMdv*i>-;gX1JmsFoe|{i||fbs1<thm<+DB_+$(zFP}?V
zu^S>^qpJxm=@LKn;aIL@#Jt5*$GSrCs=rN@F?_dQeEJae>(^Fh^=vb){D_DMFy@Qs
zYO$A9m?zba?WEuOJ#KR>Mag>=nqtv#HJ#gnDJ^9jFx<{H0~JEKd!kpBsmJd8`MG4}
zi~N&#{9lJ$xzAVqIjJp}vvZ+;Bu#$3!AtFsTRn@de|^J(!<Er{ZKe&PV%$O0_ZG$C
z^<t*$7ZDhnXTBWcB+}5^s$iRXSVy4!6%WtL?Js|5)d)L1ykknvqt?r^^@7L)Y?Dh7
zM`Hj`qa_8NSE(5x{PWf!Ho6FfL)Ybkw~eGUKX#Z)obO7pX0Py`(&`uCoC2sOd7WHa
zOON-j3NH)%AhnUoetx{{(&zUzJ3Ef96mbkJZt~`RC3H$GcpGO+|7Slu2g8i_{J8HR
zRp8#c9-{fXJsLgr?&xL92Wo}#_eD#nGRJaS;~Y<`tp7gsWy97y*Gv$oGg-g#cRUrD
zB!g+S8>Y23NuD3=J{@L!dZOm`LykSZ%Uq7jLll1H&ifDU_PhTpOi$&S>vkKtu-o6o
z&<)()T+p#;{`JI{#ERIN<R3duey<DgTWTC_TXZnV#kEhOXQoxOe(Y28xYkzeUhTuU
z{iUa!M0}APhehH6L*9K2wU6kus9!Ow^S}?FZcN;4NgqrhYG}x3$ZEmEzO=a5$`$&3
z=dzbUX32>r<tpQf3oa4w?VFrRjwamy$retnZhR<6WaH;fl@qrHf9}gkGg<C^>r&cN
zU}3>)N&kAMuJ&AWOQaT!@xBHYc}UK2eG7fb`><4jEJ>H9yKEkR0#>p%J}LR*U07FM
z0Qs0%{4I`YbL5kMYiZro^9;+cA3yd!Grz4?$N_csNZDFUfn`WE!_ilL*ETto#N%?z
zhXxh}qlXzsdUAynw&XXyJ@Ti>DlcYcQ-)REFk8F9W^{lI>d`(7THzY1bb(4CcCxr&
zZZ2_7XtZRzikYAxsXoZ?p~0jmRY;t9?#Z5(liQPAFSr&Y7FoWIae0jXCA#Q%EwP(;
zS}hoHw=qXrkx5kuUX8|B_yOsM(GOoe{&;cwCB+)ERtuVnjRq~R9jR?AJ057Kyu4gZ
zH}q?&SC!W*-0FqP-B0J5opR-0Se!Y;w!=tOk&`E1Z_7DU|NN_;ocIqO{9Q5U=48I}
z<MVl^=KN|7MYhKHlF72jNG+R3bn3?Lhm(48*<XH@o45OM!{gAoo@AA!dZ&`2v(4uM
zPOP%eNV)ON`N<2$Y%$zsC|thNCHSUi|Lv5m>=Hq9)^gU1Ik5)02ef~b9!<z)II2S$
zo^x<+?Z`MDXQ1vR!_KLSlXh`hgBTJAFCq0G1$6iBjUn;Dqk=n#b7OqUOT9_2^4>#(
zO!9FZCKjT!|J|SejP}cWH%F&aF~8~Gf4fM}Dce;3K*q%6otCyfaRxpAHlS~BA^{3$
ze4O>uvevrkuX4-3^VJcn!(&wP%fHg(e~#Dhn|f!#+cjyw+S$G_mT4QHn%iqH5g$4m
z+&H?HufW38?91ifv@W`H#9f9}>M}`GabvH3;vqrB)pt`yw&m$wuwAH_$x5?4a_&i1
zq`tn}{it%a>!735MmeXbq3F{q=4{d~bq5U<UuWH`-%?Q;^FC!y^MgZ+C+X*gRto6H
zv0P+-^!<);3gn;<zQ(Nm@6@_uJS}8PAAG&%S?EMDQ*eY@(XZcJ$0`GujQI_4sl}8*
zgQr)X`gV@o#pxaW-!95stV}9DU!#f1nyX6Th1-!kwRK)zHm4gqa|_?k?#tG@-QnbQ
z>H4w1W1{QpGcMG>Dq{>4{X(raSRYztcBfYSh$)y1ayyf+Vz%FUi{Z2L{nYN>Q#B4f
zm!HoMs-DY_89b5uTSoS5ACvsAfs)q&KTbLgDv{ZKE~XQyI@2yyadFC0<UAi7wX`(a
zj;BsX#l+}4IX(Yflq%4Zdpx*h%{$w+05Z6m3APr+YM<}xx?VyZn-Xo6web@7u5KUQ
zz7Rwn?XtHcI=aW?ybOPH%QaD!gY(@t0TI}h{P}dcb}x;&_JfX{I|Ke4@b_IFznem#
zQ+hW%@r8unFG@U!jGD}3?p^ENi-!%8WZP@Lu_HoBINHy8p)HPF`2G_?wHPL=%c&PF
z50K2)tL}6J&beDC_fB27%)V7-2fs!dO<GrjZb5Y`@$YA?%M0J1W~=pyek$ZjeC~F7
z@WH)|01Az?kk9>pUXGrR``P?N$vC&y`4F2~mXVRW498s5Q7Zmk;n)+;tmsarlsXS2
zD_+{VtL3&FHRDdJ+@9ZERmYevkz4MnFv1WknNL=Fli13}H+E=I=^Q<%_|4uYTdKlw
zYMcD}ikIcjrw=+}xA<S@i;90S^e4-#wg0rO@MuxyhH0ulaP`Z+?03G6W(jO@DsdGq
zVB+r-JNN2>x!>}>ob?;JncWp!EPLn+wM=)UPV?CwPKq5KP9TM`$%fIfZ#cHTo5(T~
zthhMW;$^6GM<eY^$a_)G>DR|DT}?6E>DPLimIc2q(#bT$3j=Mfq|U==oU)gO{P>G+
z_J1`uEJiusw6*a~H|(e9l5JWR3OM=u%_jtX-%s*i^mx7{f4Kbq-u^#sr_&q;id;n_
zI{$79XP30YAvGxLC$KovTmHpvX|^l8zS&?q^96^{tzSL|KBX>B4<V*whDPyNsa2iM
zvgb^`MIH-!2}@#l@F319&IY8=WOj!bWMWH}J*K$SkA_;(C5Da4V!KzKMY2xR-hfIs
zs))`<Uhj#4YGh{SdiPYiZYDkYnb^KOtGDQ?<rOv*f4Y=@VPs-&Mr}yF`?%rbN9Ed@
z>Y|U|zFw$U=d!5~+Gnez&9oy`Cui}}kBK8j!=D{~1gwo5|5cG8muGb-Q-5aX+~o2%
z%TA&fNHeSsYIVH0yd6cWe&&2tvmBS74z>u}-+*UsU17$1f;~3UeFvKpUeWJ9tfHmW
z!E5o*er)vEi<3<6Wo73?`a-l@XO-M<n`Wko$9J&)^5HOmVO(Qbd`{<=k0^@$oz&zm
zXA#+3EPL;3MkEcdvP}LMKPcK&<bS-z>N`)Km6VW&Sn$@0q(6b2<s)1<9Zsrw=Esxb
zD#ecx2Tj(}x^AVN)0fhufn1ENbh&}UDk?;ElVusx(J>R;p{O{7bq|Y^(LsU8;bCr<
zFOOfoe)EQ>*?I4tJyEKfcL`XC{;xN4)cYGSrt?LW5Bx5~W2e6Uh?vyYhZG;YAVp2h
z7diO(RkgO4penwddbV{skbT*(A}+w+*Lv~Ga(<HBi3M+qAKvfdGTKuZjz(R2Nuh3h
zXRkQgR1b$a;idG>KT&fTr$U{!vP#{;Np>f-Ebtf0UdjF0XVTr?H`P!CF8r7|zwl*{
zWtY-9vwKnUSP_|*I#V;gJr48`QCj~+Ymk}i^AzY^L9MclO>4zXmJ4|`zQ42{`MB-0
z%8MQ2H}E(#O$m5*e0-|pgqBu=qlAOlR}UGS;qO_!`rU2f4<_=3v)h=8{r*)EpZUml
z^;TdOUldKLmA{eazp9iv7=^24U27+G^UO{8c^~3{|5Y*Z#MLGM9S(7J<bTUdFTV8o
z2mzL1j<_A75P|M4AuhOn#BSy75&6F@TMv6XU+VbiA@gv~chkA4u9!2Xb7{7pZSVpn
z0t~MP8#M1IF1Y?q?Pk_}5z+w?+F&$W<79lX&M$Ff7P+?1|H2#?f9!qS0rKc5hrOtb
zuQeax4!^{Em$r&T`<)#ds+<>;*r>krxv{f0U&M2C%4&6OZFpj;)nQww-SqfE&(^id
zb>FmG_qK8l*0!vdmz$5iX-@0+HI@^8*3)~2T{27!r(LPlr{%a$n_=(1Q$J!U(sBHD
zuBrBBHOcYt+>a--zg{@}ZqzW^cOs{ad%r9dR@->6s`~rF1w)PSL_H_-=NySQvJ{7?
zcZ)u7?k$$pTsm9+eD(m(y+<$T##B^Bk5+ChR+Y=)p7{%gpA{XMZ4azOemJQ<9U$g0
zt?E~6pW*v8Y<pn*Wp?MmLb8JDU&0i>mk+cK+?+T!cw?8_=h6t^NbE}~u*>(HE48X5
z--N2+amodV(JhylQzxs3Lc33=<vQA_xZXSw*Ff7+tEuNH%##zJc9Z{2^r$5pI#mj}
zjpVa4v8=z=+TWZO`x-Cl;CXKn^&voeluB&AS~_dQ_6I#9R(QgZt9}bpz8r(q_g+%`
z{uJ3jYxf~~;elQg`_k2l3WG0af>hsC{jaY+6U5S(?Y^e>Osy#F#3$ArG&dM}b?;3x
zxRl;f_?`d9KBTsm%hqpgH(TD50m0y*A?jP(I2+JB33RPhRntBFv*`Ou#Wkhh*LtDs
zl-O6y7x4{Ae?x=VBrmT~fA;E~iA}`5dZ72j)Ao|~+(_G-HyUYOSwV6cS7<%FJv-yF
z?jFXsk9z-lfhn`#;~OjCaz{s}oxabPd=C5bB}5J<to;5R)4m2i+(_iH@6<h$pDJeR
znjUlw?8Ap^j$NLPk9%Kya$-ls)y_9p=-jPOVQch!U*_L)Fbcy-`l+8!<mDzMmxq7h
zsq9)d9PcRiy7QF(pb%}6`K9;wf8`yQob0)0trx!_P@8WPYM#4~zZo}NX8sPyITO|V
z^Q&t2r{=NG{eR&)t)rvM(#c5eKOkM?+Ia-yM*F@IB6|2h;sbBy|EmRn(!7I+wtyEo
zu6ka2|NaojK%gbDxg^8!S@(bKUbfe<+6R9R`~9l;Hk*1rcHv_J-TIIEQFgXxzI=*x
zQ4t-&3-RN*e73uH$C)t;r3)1vOUPk3Y9R8R{<~nA;O@DN83x}a&Sc%pQK9slUYF_q
z-0NF*V1x+Qs4stD8DLeOKvDTO`J$x&#ktxTMh8AmsWlE<Z`1jp@|JD!qs&MC@u+L=
zsP3YbUMQo&$BTn#X+mZPt0U?)Q?r<i6QAg_b>-<c>u!%cXDui3=OzEKQXfvX@LRJF
zIjZQW*(BZSdRmubbLvdad{wB<zaW1=yz-L)qWN%7Ctyfc@GjumWO27990Jvje_!kM
zzZC-e10GLpblP~|zHP+P+bmtM&b4&PbJTIFz|UGsYrNx<qUYn62TlwPOkF^47*Y@+
z{o>cxTl=KrWDBZZemE-q*Kzdb4YdTRbChNI*EnS_a%6V54wL=KAJCsHZjH<2H<&2?
z+;vF@Lt4xt*>_4KiaNLF-*M(FOy07+^ZLS<S=sGT_qBBlUaC$W`E+>X*}2|pKHhss
zmsgcaJ~cl=74^J4?}Rh%48yT`iDr(7+?`(q7_jXsHMUOVGpZKIe8v;>YL2`*`=La1
zW8j<W-L95=h2LcxSDQD}<hQ<4|CxQ9KUIZddTPR0q%{sFT4pyW)v17dI9&f(j9V|X
z#+1U-jCr%mr4&eROox+e_~35>yi;L*$0Qc*y>EeKyUxbGsVVLHmbJNqWiF*{lF6Qb
zm2!J^oXn-T{MV?An+=P;^q+d>+C6=1)44ZZhVC;#_r6}Rg7)p}Kn-2&;0Y%MiNuhx
zW(kLBH85-k7S`nDFIZIm*mol<q4?^ht8p*A7ia2hgJlConL83CJvw^=2P|JZ(U!X0
zB$D6)j=zWp<OXfDdTuQC%L)UZVcAD(0oAmw6H*K0-+z5?NH~4q=bOyaj)iC1FK%aw
z($>@A6W>pT62=$K_BP!^HQ~^?mimL4KfYe?z7jZedTaGh8~V8qmMK)Se-B@HdP1k;
z<F_UI_MuHwR8Su^^0rz4@yEcmvA)8-l^V)BE4;*ikEE|c-h9+urnTku-eiq%SDWWf
zM%>#v6MTACos<?B3^IS-I{J1kb86Ypt|j@S@&BReE5oYXzos{!pn!rPU4oQIhten|
zDIg#!DJaq*-5^~8B3;tbozf}NEhXLE4e#3L|2`kjb$Qr(-}f)pnwd3o-`j6zd}~<`
z%9KJ2jwz=*3o)6LOZ^y`QM|Wp_~SpD;H5Y^QjHF?iCEYyUoBRYZ7r;a_tw!odPEY}
zEeQ&2us{9)p>2?^>jN|_SP?fwX|pbrEhUf$7~Q}TVo#Tvdws!?xu8~;PrK!eXEl8m
zdUAqQ3f65XXy`~CKg`UYxLhUySQr3$GoL%Fy+hK+#a8><1dilSW4g%sKL>%BGRi^o
zN!%T{fK2uX2nfI&?1oVpX~|d@a$NB5&i3<UK&DZn(+%N%c(n<$&aHPD=jS(rlL~ME
z$cOWAcF|^$%MS|UuC9%MYn<$SjcbW!V_yR48pPUg@C?A$AA8HooUfZ@Y`0n7E^;d1
z;E+cZq|tdYV=0w=L@~yshcdtJ{EFc?(7bM+TE7H-{?gSsOn{@oH1so=2Y6e&K$ep7
z6(8?)lrWm)UdJbGO%OwNvz$j1kl+aSbcAPuA<=s<RI51}H=eYcc5yzLeUbb2SLDiy
zl}4q72W7?NpN9(F$(p0;4(n46A3myf*|kx{U8VWdyHz=B1H<^9BhFvMEfJ0cI35d;
zH7q|^qf5)n(>;9}1;!eR0Qb|VH9;@=?Yat+I{#S@Yv7u`NFihW=2LWjCnqxd!}L!p
zQzaOQl$0$kiOVfnT+CNXPEZ$%Kc{Bvwe1LCjtqF*mdS{T1b)_!H2<2|!r^sxye<~r
zcE{Wv0o^7b({QuXKr#0Oifrg_)<*nw&NHSJvt*vwM)~?;yv$7fZN(Ik{p0Z*sbRLL
zibFk0Di;0MBtdycNp!75En0dyqL_Z*8V**{y}yFC-T{UxqhS~lNY4lS^pQ`slax0s
zK^Yeq6x7AJX#rh3^!95n=W<^^270&RpJO0M1Ez$MV7eKMu5?Fo-)?|OkHfxj3Ou>D
zH5Qalr&a1|v(+vx$f`S4Xlh2b6DBpV$Xu^c%HD}hp4vyUT*s7=J-MgB6%D3vFO!7j
zVPYRloVK=nZkjxaLrs${Aw)t90nYUZe1!G`0Qv^YlznM57h|6qH#J<jIBX=Z!mrrx
z)9WpES*HB<O*z2X|EckKrH0+ciNA7<h&hUH2<7sdAxn^Hz|iy4@Yq<H?%3xw=lxSF
zwKc>XCO>j&J_lxuvKWZuUj)mmms=hDy$wH727GF6-?{~5+0MxZ(eD1?#{~*cVm@bf
ziirxAetyt*6;4oWwZpS7H+^S$@Ym9}<Ac^Cc~w8A?c)o*V$6olLC=#D@$pi#MkujA
zHa5`#YPCr9P%wu7aBr$?eJHE>VTj-#G#o+q*QvAh1$BxW9GSUOcL%vRS*w~rKnQuf
zYagRlK@|ozS)V|x9aLwBH_i4|G5_}8ZR4l7uW~lT!hrmERj;M#Xl=Kt4%I%gqUPk}
z%9<{HSm)|WjExRU@-`+WQ*rLFQR}S5-_!O7<D!1D(iLJ#nnLI|)@p0)kMl-Z1(e?y
zu(-ad4t=^@x8mBP)3?9y5a+vk)U%SYCN8!b$?dC$21BI;+coFH#v{4uF6Vcd-<A4-
zm-o(;T_-WG1@5)eygYYe-2n;vmt&E$+$K-!g%wN3L$aRhe~B+@(d8ZHvc|xmvo^SK
zrW@YF+j>{D%X6HWtjtgIDoz|{{|uqoyrsWCT!5La!o*FAY;uHa&6Zk{6E?>81gIN%
zR2x{3<@w&LM=FsTv)lP2fd1U45GaK+{m0-c(n|m)Igyg=z}7E=tF-+5vET;41yin&
zO!RxeoUr8cDUFufX_fBNrR4nwy|2p`a+G~-;R?!@Qo;nfTE8dWVfZ5kHgQ8Uj!4t*
zqHyRUv{_LEtxt#@o(KSL_3>Ru^Xvl_mWbY^T9*!E6dogXt8rQwR`w4HN(1mbsw+G!
z45bj;LaKyt?$5#>s0)n{AvwcFX9smLPx90?Y&_gaub}hCMi&~)oNV7<pkl2^a99#4
zz_Swehp{zeo<|BJhzf#W*uHAH<IkTt5FIUn^$aqpR4$g=ADJdhO8RQU{>dWsl_+ZM
zInFXnK<mSv1022X>dZJpF6I@ZCHFoIxkDT}urs>$$n7lbLOP$>rQ15-tv2N&nY!5T
ztH1NdEz$Nmx(XV#!|U4P-32gE6iLg%R7%fsuP-6K$xW)ZN1(yw7tvX-f8HZR&TH?d
z*RHU8q89=ah|-cU_is#jZ??>E@+S&Fa34;eVyz4%PZ2!T?=G2JRXPfApLRZcgys{x
zco4&7?{p6Kbwn@-`vOL9<zu;%!C(y_3*C}2DL5^usXf5c6Vi-mupQapQtbPP;aK;Y
zs_PtETYH)_uX)w_Td0KxwDbL`PEoqc4hPAr&L_LeQ5w5UZ-nXSTDPqE>qF^T#5?Xf
zi>f?jLt#5RJu7xx9O!f%yI-z1s`c-Bg-c&9;aQT_)2NBBNER&a{O;n(F8&r<8O*;Y
zfE$CRZ43kq^n<fI(hA^ezsZXHZ&k9!(c)!@uwDy==j43<B-bhmDYsD%Ot7ydW`ds^
zx!@aTmLRYPj|9yQG;W4}3sG|1&W^JE^XwFbkWdiX+0>e=86frtWX@{(FTZj;Lsr4D
ziK4^12aZCSUpH#fekwRP9M+rbm5e!j?AhQfkX5=r$3|<hX3d@1D{ggWQhIJ_GU)(t
z1Qa)oJjQsWq@tre6w$n{U8?vk#$Zk-wuznxpL`FLKq~7<RGp8{|M{i4-t5J301KD4
z&c(vpiz-v1Mrvw_L=mT>+;0g9wIi$6cfZrK09O*hrbH2ZLrvt8HDNr;zyqWuOBrSf
zoh~nIH^&q}UlkZHe{nJ&Qc&5&kmTHiG>Cbb35F>ZlShkj+pqgdr60n50~oRE)uNrQ
zuCQf&9VA<H(rY1*vChcQInZa_7`nx%G<ydq=Y9jGJq_Sh72^Y_aPMk_E&&{<y6|Id
zeO~juKaqRC0SzN#p~gu1tE`{+_g*IWS)3g`8kZfPwFpfJSI$nB%5RMdjDplqyVjP+
zVCjtHtur4J-y-pC_UJcgF?pib)$zV(Uty=^q1yHnhu*&er+WJ%G`*TN_F>@f&z(8w
z5QUUnNN%0t4a>9lVgAtbEl(F#f1~C2Z!!Uls*vHYa`b9=1KGn77%FDN*#y9a2_F4T
z4XXcjQMYY0j7iT&H+<PV2Mo&|c(zt!h_dGYk`AMKSyu;_KXL=+j%WbJY;m|Cn?tuw
zMRz`I#4|gaZ1f|YTa7xqw9z{}=E^(YUCoO;^<I)_djDg9|FU*kux2)oMH;8C_a=a@
zfPkR6_0;z0aEXJl5^{cPKbT*~y}#1y$whB<)ul}*Q-gTNZT_X-fO$_tRP@W#s`XcN
zxq*0ZMv|&My6iZ0lnCI>n|-huZG8@cuu;+S6+$Y2-ri|t*xu{f$mF#-V{UBZ0UX$M
z_S?DyY{R|IKM`Fq`&5+C{9%vqw;xrw+`gT^`mzpDcd_AOM#UKm#ZgpSFlm3T0~aWA
zp20X|Q76PUc!3vgA;dJ5o1H9X>!o|F{~7g7a_(t4ZzEjY`X*v4r-_8I4Vy73LF~qE
zKVwQz&5_y<t~Lr0w<djp-E@l<@bEsLGWmkC2=rO+FS8f7mN#sFt+#7mJ^KUfiUN^y
z_qX&;|KPw()40f*s{yN0Q>)dg0>FR%?wNqF0jMI7^74{@RnaNo{`en&{rL=GZ@=Pq
z3+M5kC}_b3shqz><hW*nrsJQ}I*-|==B~uC4`6oGo@VW<)_C|%i2zW&FCrfMwLQS<
z3PLX3r6DInk?lJ`nnbR<g0PI9?{s@12YS&L$JhR2CcuAik&9SjwkAUDn?7u0&`<jf
zdd$7uUj-5N)6k|wvk?9}H>f+`$Vp+07a+n4rSFE4pp(qt75n?LCDoJN5d&_>T+*cb
za=iMbfSox1LW`M@E8&Xc;KlO+)t!V<w-tl?60^fZN$Jl&2#w@U{5JL6`ml%-$hK-!
z@G1*zT3e0EblHg^uJ$whbhWhqgiW|=c-TKgRnt0?2UkA~2`K{^C@(W<`NZtspTE0<
ze~Vq|piin|707s<1I=LwP^ZL#g;W?g6tnwmQHA<blA!sD&I)f=TY1n4ROZ{m^J>||
zr^E-P<`c0;rs`Z~T?sWiUHHytr)th6);Aj)pUcJA<AgH=rRqQ0O5f>ou$agbaya%p
z734G~1=#Nr183HPA0;Kd3RCIhlxVU#6+BQp4X&S}Powp?@f4e{wW)Z0n;WY|p5T*=
z%>?Yj>>AfuzCeJBrqX@-$?FFR{N?kP*G^NKb8aI+9l%KU!0+I-)mf)g!&Dv|FzEY1
z4IM~ozX1lJtnz;qVtxx&!-pp&F70NL!PbY4r>CRS9bH{Z6-jjxIm4VLzY$vUGWsK_
zy@i(DAx}TNn5Z!AhCuL2vk5Z(6I)8Bvzwrk`8n%$O2x>c&b`=dchxCiUi0e6X-Q?q
zf)_UCmYFZWzzwU`@7x~_Xn6lD_PH;pm@|w``M03yDX|M{Bs|osLk|Hm3eT50zvgzI
zLahLece=~ONNe5ImT%HG5SlDbyIOWcM~Gn@f0`Yg-kbk39c{5qe7p}W@tie36ygxK
zkn$>Pg!<U=wq-X`y=m6txt=@oN7Z;r!F6}|so+V&cN$GjAZ6RbQkBGiRX3<Qz(u?|
zStKNKz31e=#1^7U2h&O*bzJ(}Xu7rLM-mhBck{?zLoXkbI-B0k<_?uPnvmJrt7Nj0
zF%#O%wAThat?R%K=c~va>AoW+Eeoi)I)7DE@Pb05!)@9*NkY5p@m_*jG@xj|jy4__
zcke3eL#?vG|MJrZ*o5t`4BL>ra?5Zon|)~P(J_lecPzIp&emJrDq@>u_Bze-l<^y%
zx-^^L9yRxlnmRc+Ji$gsgOYBl7c4qqPiTHe5<^#;vs+@D7QCx{`SRXoI~sE2HtXUV
zpCgv1r3#Ll^h=1W1iK${P@~`^M1qD@tNgJ&^dkJW2C^S6J4d<KppW9pl($td2XSZg
z-~FLn5-6r_s^W~w{QuR_rrFx0+hHSedPK&CT+UHlf<_DMt{0~jM)>hcc6ap#BTY5r
zdMkwi_wcEBuDwuIw*EAkQ6`96l{!1O&HY8A*VOC}rRLw5Q!}ev{RDNv>*@N6xUkm2
zgbg}=$V7YyCQerU{c@Fd+~>BrnXjdGK1OZ-O=daAA*Xr5dR4qHM=k<vNbIl7QS-?!
zTIPrc;IdWrjsh+1|CVrDf-019pFanw#q$-r3gO{}XUkiFl5!A0$P>p7yU~wMPvY`^
z<O(7HpC{*c#!j;?!c~3ys|g>ZOOXbON5#B0OU%KfD!~h#I(=3K+w^DCBY``GxIor-
z=2+=l(Hr4dX**nHFCH|D8gra?_*Q+I3)1!Fe$<maygu2!1|)jxsx|9(eP$$j@m-wo
z^cZiow&riKO*_G-tBX!LY*GvLGR(e*B^kT@4B9i+s^qyl3pqPKGLxK-=$OjJZ|8<0
z-=-=p7AE%36BPdn(`osNe2N@7#KR`z5{s*I6Rl&eX|~h&4*WLFU87MEq21qTFhn5#
z>hmAf<xAIAnYnp<5LrG~x(~HAML>!(L-(8i1l6}0a53_Wo3&Zq+K!Qo=1;eP++{g<
zbXg?_ac5(i2x@rOOH)m6^G>sh`T2F3I%tC&))O2+{+aii7>MDIl->+Af<lk2_26z{
zQMc>M9PP$Cb%y>8c+rBthqy?YqhU$n18!#Q2pLt20yR|5`+i7@(DIHell{*5_?@&*
z_ubqEm$g}E52KQDc8WeR%}4DaAH`KoZ@$>xVnciTu*6cmd7ZXJ*(QZy3S55@G8?cP
zRae@eqm<knPafWS<I3Y$dljI$9sW(ZGJRVS70t76UGsQJ3uf&s`4gyQ54@0i5#!md
zfec=Pe~zEDinuoamPM&MTIWQf#KLiLmqNHLEVcbH`)+aStE@gJhi|dAVIjhresoWd
zaX~ey1hoduyo$Y}>PAF?$P?3u9Ifj@omt(U)|P=XD_>o0g<sJxLxeq*vOt_(olOr(
zZUFbrD&Cin>H3cc9Q`0*LEP{yuJc8b>vLP%mtiEp?WSqUG0+I+zC-`5)5m3b@n)&J
zpwcNjZ$uET;Y7;Bnu{}aUW&W!4CxM-O10nE5?-n&lc^V5^F8#r^X(^E%-gp?aqOPQ
z%PwuP+L5!%%T67Fa57C!e!BJK*Pu{7(K8d1+!V$?0t<FiME~hjq#reC9D6CxdWI0)
zk7?}e3<pqm5A;C8Po(g-D4L5GMfBi4{e9L=h{d09x9nXvkb;3cY~~Dw&to%jRkgy5
zgg^Bp<`mx9{O5!Sq>K6|<fL-`!wq=KJpKb(@Zxisc-oh1I4>L0sgEUeez*BU2OTYK
z%cb+{*MyR(9GXhs`39W-qZkI>+>J+PkLj*C+!BNbA)!w!Y>iuRy**___5(UQ#d<9Y
zS4~Zhd4slu5RiHmf139A!ZrPQSk`ZC)_T^zgdp14rF!OZ#!E`Jdl;;=v>d>6oXl)i
z*=uIxET43gNcH|sfeun4F(f$HhBu+7$X%7@(SQ5T2meI7(Mu*>_mV0uuxsaHqn#Ju
z<ttm=S;jcJVbci*^Odc`O*(BN#rxjcY6SrON8i{8)qf;jW|6Yv)nU?Rz^(eMIPCWy
z?NN4kWB$B5LwY5ZdMoaoU$c%HYi^p1m3zKRPL8Sp4eM}25}pLIhDu>3oM(!Wh3cu(
zjrrX&QYGcRKfSy=W|eJv+zh1QhvqgGSPX2(HtlT*FH%PeZ7=EF=?2KXe?AFN*#k|Y
z<=#Qep28D=>0<^Nl(-lbX<P|99%Tt$4`?(s8Ku6~J{x3rW7qc;m-%mFW0CFXH4=3e
z)aS9Be6d`In=HwBhn`T@tPEQbG<UI2O`ERJeE+yiK~FS4n=F<&jOxtNt>kC)@Qy`N
zKM0th9;?=PT4;053Nn*!VAYxWSv@{!*++n*u^^ksQmoAX<P9l;m27r=1m2ezZ1S9K
z!ByeF?Ws;6Zen0qa@1ZFI4AH^z{~*c#PH(IPi_h7><GP^ICAt>Rd<cU62lt5lZ5Q`
z#1-hUuh0#sS+4AGASDbxe#ExcC4z^ZWNHV9n~)^>;(M;pJ32)Rd-J{Z(&AUtjO?L-
zM4i`8Pgz+Lb2pgpOPsz;VHPi&uiJ)h?{)rJt;XxvsHpbe9O0FEzY$xc94aZ!1aL$-
zG4!VCck6JI$`|cN%h9_};j#SpdIbNv^B|tY1)E3Zh5B7^onm6n=sKP<X#00`7z~lR
zOW=K^&Q73dCDYZN_p1+8T#9&^aLW7K1Eq%VSi;}pHWCoa$|9OVdql(leu}_08PmS|
zZ^JpfeY^93k1qIcrM<rtHB$2S%f3isG$zit)ucT1Q%Q`6=YM}&weqDa3Q*%Gr=!K1
zmyVlm+ZGoshLO&v9UmW`sWN1~Ols9H(L6fb-zBn#iscq7cj9Pz`<ez^k}#3_;lIz~
zdna}k-108i2dF9Pdyi;oE9s@BA1zk*fXt8>+Yv-qg#LxN!>{P`&8FRrEv6-=owe5@
zST3(#)Sax6xn5e{cC3vOCn6TqB^Nrwem|Oc*;*(3@hla58MVrP`%IZJ(9^eNG5!g3
z2Q8X7U%$*PuLl%rZbxxJ^!S}_8XZAd__R!?Ks?FHZJd7ATn5_M;x7eapRP$%MO_5t
z5|bVdSxLY-0O!#dlqolxZk<=?oeh~*Q+!POCMCAOJSow7eIU#dKO-Ev@TeY&mGP>F
zpc8_(@_>sg&hc<vyEs0D5zNv;zkNe3lves~M;9ylovMnJ&8u_2!59Ip@rTU#R!GuJ
z`I58#N4^##%y#y}=_MmE?>7J;QyBy97RrA2{uyb6wtKT)0&u+O<9}p9R8QApUvqv~
z5SkQnym1^<D@*gvZ_K1or?2GW`dy%bzl0P}IXJvlY-k@P0=oj)NVrmvK~doy){U#y
z|4=rfq%f80(k7B|S#z)`p;tQ58yu0<|K0Vc`U834+7GQM{<KdYbh%^yd~|`IqxtJN
zl~E>^R!daH`1bwml&Ek$q=03#^vlKu`N(zA%?KISo(tvYF_)VH$1~}p_@KfyFd$BT
z|Gt%8D<B1FZ`%5SGtZVk;gD|kKS34Kfc7(*%C>kCm+edE+7BcXb`U0?G~octXrR!6
zsP3JHLrw|<U1s6jvTE+z3Rr;atE<o78039kcdXoQpth#ucP~gAyREc&sbW9Zn<pUa
zoU#n^X6sEp1Q<&KeNee2a5baXyVHY_qysm^8p3?Ch5!erH>W2G9Q(Ai{yone<Rj+e
zphq6DD(v1D-}YZ-;)C?4IYgvA^xix%ha;+a?fE^bI)Nx4p<e$<m<EbHwtN+=@!gu%
z%E<n>T-vOASqWc>R^=K``N)GL9bqnAf}b*eUbTZft(z`!v}}^~lo(bWCfF&R4ERej
z6lQC29v>a`gPQo&tHA)l<KMZ4CR62s=-mRlO9E5?<tRTwtFfEbws&&Uwxk5I4DunT
zW-X+q<^OAJs~sqKRoG!LT?|Mx5F)^5lJk?R8g>$0ADuhWPYU>K#<|XWRxC%JpKaP=
z03M_<ayZDA(*f2z=&#Z`;CUr9soGK^WX)JXAh<MtipBP;;q(LxyGD<ByY7%_Fk`YE
zKSZ!$HdP7kxf;xg4jhrIc)NesX}_&NLNt@PU9Ufty+|I&|Gb$n0V&aY%v+_x_r>Xq
znKXDQpxs<T3H+9AnF97uEndc^{Z}{D#@{q+;z^Xo43&4i?HJeHR!rrZT`J&e6_S5Z
zY;@<gzcfMdRqlofgiQRYp#+Wa&~7+lBCYB+^`1Op4wD`>tAfxu=Q$1~lkJivzW&Y(
znsOyTmJO;u_13BCi>=2(1g{R_F{_^1EpNHNpjXx3(Rbwre@2tsTCNhvc!}^|Ot99D
zmFwB1-`e=9>9o6ByH_MYO^uQ>>?o!5{b~9mOyr3@8pn6-`;4W5%07R_8-#xF8{a;%
zJL{Ph0};5Q<G&}llbuayK>5sXFRLvbnp{~4II^*Yp4WQey2)ZlDjW2Dgs1B18v@e=
zL@}0^RO-KfcP(VK>k~K#!@(9edfDYse|NB2S>ft@<b2VVS>w8$@QC6h5P$<M%erbC
znm2ok?KQh+<Xhwd{t@;0RkI#b7t2SwfnkgR|HY<!r2tD~9suyMn}#p7ZYec>h-ahD
z-@ux=gNW<y0pU{<QBs}SBfEpmr&?{zVlTo+rdIm$d0fs9zzsPL4s;Nk;FA%B)F-iN
z3fQF?6OG!lg&nR(GN1nGX`8O!=onSs?s@oF;6*aAO-k$kX#o^%q^Z3cR1bF^&D4Hb
zKsMK)eXMdDsNIUuaz4DHWxuqm`^)|O^t9M;Qvu1o!{PU1o#oQu%pU_OqllU^0L>?i
zZ2-9}4ScnP4>2FWPn~WT-VRBu3(%Rf>=NrjeVC#(bqgJb3Y?O0TpXvIo4=ebA8AT#
z9fKncov?5s#CCyTzGK#+2Bn6nTfuMrMH^@p|5#V%`xv4Clq${fVt0Q(uT&r&uIaY!
ziNCO2LADBNw`4fZK!NNF!lqxxlsdh2Xd7crp~$QrK2yBoe?osWBTteBSTpcnW%6YN
z7Ofm=rT;4!{FGuik^YoWc8l+a2R93~I7c~(5GH*rk7?I#eL0#1EB)6Vp|h~hSG-1t
z!vb3^ZLNb<FGP@tFR=Yfu2;Z1F~2=Sn|3}%gI*W#z-RCjIEt2mLQ1yXRQwjM+`nF1
zd75F?<+8}PHd1wD6i}c0uKBUmwEpzh!>6>NZ{Lo2jhKVSG>c-E3YgYfLP!{NWoBsq
z()wy`MdA0uOqAuvhgjDjfK)~5!9xUj^9Hsobe<vXf{FiVKJ+0xKb!HmM2y<2gDRx`
z1>kZU%i1471}MmGN+VDDl$rCoeV>TaMCzK8P748YdnB5&z?_3Sw<pXR@7&tXU~wXK
z-}GI@gGzl|I2zPaTCm`5xK(#`@vv=DAOa|_Zy+wc@D>Lpn_UQHW5MIPWSg9&ty2OC
z>P=mKm_)sSJ^~UBgP4E6`x;?MoT~k;ba-5GRJT}-@fq%G7>m9CvPKRVaKf1%++5UV
z<bYqcL|op*2G#iA&HVQYO~$^KfOtlf@4YvS=XHp6*e5DJnOL;mob-S>IWCwsG?}Uk
zyqC&QY&*=!Z)c*tcBm~Sl_s5I1{mDK8Q)3u2VRVc47fN$H~I{?p(~e*sdoDG@neZ?
zGd-YDuPR>h0@PJ7vFo(-P+>4z!b<;TlJyiC-*UwD?(nzm-Q$AiDk|RrU@b@cDvghi
zWFeVKu%2x;bwm{u;5?3Wl?L%*_GBCnsXX)9lG9SRw$#h7VIVOf>~Tv%>cR#c&8av8
z;4*(-A@C287^Y9>Rr#gDodJs;c-&&;R9WZN9L_59Kg?%`o82dtsrl#SP6s4%GVjpA
z$oOS=2e-P?_v`SC|AYi8`v2VTJ)oia0xap#)>I$}d1ROGnw-z8m5ygmSlP`l<$;v&
z`)UADW!qu3`5to+eK^4_RJnmtFI=@gQeF*qn2GA_VaYM<uUW0%3eLT-xdY5i+S|dl
zQJY_I|4q-JLqN4V@E~LNC3ew1ryJh3oR`jWpk^&lYV?GBjr~3`9F+WX)UV{hDbq7b
zAdh7_Uozixk&VHJSs;@ww;2vNuNzFHYrJsjEOK{>nzDGu+;(>A_kt%mtQ%|;Y7oF}
zi*7G^_Zx`ky*7_%kwupOCr>Th^uRqf5}5=+zQM#}@<jb_?I)3^ha}>X`WfPDoEahq
zAD(}4M7P(bq48owVc1b?^y)XzUCyt?i!d%+iTk*6=m;b36nA~~>cqwW`<tV*+UTAh
z5NHfCoDML2l!N5L8PXdEEggf_E4E}X#6kD&x^QolC-=;h0w#4Z*~Q^+SE6HJDws@E
zsnfg?1%cj@Tiva=sjsc73o9@D1CSTmZh>^2H9rMVSST|Yab~)ZNdoSFXi#O`bEEb+
znmYTuH$>w9rmi$_RWAMvm%fB##&jl4T;i6}_`>?ZD&V>SX^|~8r+}OAca*%_-P@Bx
z*}x;9X2N+?N!_w&islGE1!+P2=1ugucV#87S^cH$%S7wN`yj8NqB0-3E{|tT;8!_s
zxI<!zL-xe9(WpDHk>^M8g%O^pM&Of%<S)0W%*`QZQDKliIKQ$F?)R>1=X(%u2V-=e
z?(U>FhISZW9PZfa-#;HFvQkgquNGEG)Nl6p09t)aVO}lI5!03D&S>~bROK$`?_%7j
zQDP0%R7@}vfMDQ{kj#NhIEO$C<PYfSNmjTHyC~gwY*wFVS9WK-qSN)m{s;`wI=q~D
zf?NjaxHsQe2A_M&<mp~FimA!hfqYd<E3u`W;WN;~0MNj+9xVvnf`XXgKZ?dS^1YZv
zIC9h(zC^Ob4irmpf&5tpG?UkVWXi|J_^>I3@0PSSrJZkO7=}v<9yDBL;|OaN_L-Dh
zhtt!EFFsI%x%)(_(3^@muUWt&Dj)%}vN*uM6R5{!*Ovvyy0aHJ!~#b4-mV>5pw|)=
z#rxklPXnm30CWRKETl>WHaAkGV(+9}U`D}jV<M|_b!kLE$MGm8BkxzZ9ye;ujT&Zo
z>03%NPVPMjz>j<>E0d0=U2D4@m%ZEH3onuz;NtZ651xNBB%Sgyo_6iz5s-R)Ki|``
z#PX!O{2IZosbzzs_2)q&sdf95$z)ALwavr7^?GRVu>e8-c$sP&(5jn-5g$eWzrFvO
z#Um8~mY>UDycLzuuV2FPzUf9JtL8gelyjp--=MnvuXv_iW))U*3nU=R`(A&SVzHq4
z#mRb~Qm%u;jhpDW_lczx6%jl}CnZLQ{Ge@ygU|mKPUp3H9RIyCScjl!7601r2OJ#O
z0)Q$E5KeMIYp)*-2vE+x$o#qq^ezj<cttW&{&F)<;d<6&x{4^S+ZNA&zFX>+5_%Hq
zAQt_2g>kn>f}6^Tq?J`(yr7NGPke^4lH&*Ud2d5nzjeZmL<}R8XGg34jXw%hJRE^9
zxL97io>Ph67%$F{D&}49A7j2aoA_8&HBvN?_8$)RCmXAf><93Q;u4xwziLt#1<zJe
zop%~Byz$tDrdeWp3#rjqW3FeCADT=TwI`S)2)g(Hpl3LmL0~#jnx+{yQffSb33E)W
z8JWmY&(c!smr4HmjZuNH-k2{PuXqxIRquPlr9Wmj!*)c)+jZsX%WJ=K3lQpk{NnN;
zGAjm{CXgX0yop7$e==yr9xG@20M9Q<pVo`f)6o5wT=l<BEM-S#rs3-nyOH-w`liJ7
zzQoZqv;?@~ij5C$JVzn>g=Tzb#8zI#q`z$BrHLj%>+6w`GR3i0R6Wi6O3xHgBFv&4
z>O^pbmApuzpKJ7PS1*13@XzYsexhTq%bWv~+sdBc99>U*y%*Q=Rm;?QvW4OJ4Nd-X
zvE{E@>|8vj88M}fmxNvFmDo@@{H^-P;)6q`vT<CPIcs5U8kG<h)`2mB-lN05+p43Y
z^bWNW9RZ<XZD3NKhRF>{cOd2X8I2IIT(Pq@7$rr7#((>=3L1ui#r2hcWoRa3Jnnh&
zXhRj*7(6`rL;eC5E5!5)ae=|cQOR!8iHojMG@>W_M3q!$xW`)}Ezx0OO6|qvUAyA-
z0TM1wYT%xP+Mt3kenSA96+sKivoKFK;UwP|c4$fylN)jLOgubyqVhG`W2~YquI32d
zyo;V68Rm*e{2CvxflaZC)f0b?4>%z)=WjG8J2R%;cb(`j@9)wKP&5#{&LhUAuOkl1
zmSRABLiqvpQl;x<^$^OZu4oUCrpyeYG_TH|hkcL#BCWjqLoKA|O|`{cFVW}Tc)TA$
zQ0PSIVTFQGZzy^1UVNNg!)4zK+`v{G#NVQ|W*zngcT#t2XYRWh$<$5xdFy?dpGKL&
zpFytA@1JTBthc7yTAORFv}=NbL|?ytGG6+*IZN-nzMA6m;Tg-DLi^U>gM%Q#+;<$r
z(}YXQfsKqYH0mv)$%+~7V1C~YijKgmq?j1*M9OCN_D_$#o^3bz49h-duAvlm{BnsA
zqC1hHIxUzwVmj{Q*4#u*7-)r3S(1(XR^ouKdS7?^{@nuNkL-~LGSXQC-lNkM+sSpc
zWQpD{V78nrye*{3TiI7u$oEO1Hrv8-peAAnWybqU0>wEBHu{gwY*Nq079!8xyhpEI
z(1b}XHGm_1`%3@K<rW7)n;<Fy@{ld97s?F2m<$_hhjA-?c~s?Q?{4DjrJBb|_clFb
zH*Yitz!KrUi}?4yFZKJfiA}eN_~~`^WwU<EIgURuzpmi8^@ev^+6VBPn?2qnD>Pf2
z<CLFQd&=!DpL+I((Zar-^rrec2FhPwRNs-a+-wfw2(<nER-NzI^)z!>2=IdZk)wPk
z^Xqn0;0i>-G-uny6@5IXj2A2>>X9dP8^p#7Y+V_i#(z6DhS+)PPLVs#PK?cmk2<s5
zbbYZ;yW!i3D0&mrJz=X?-unuf<-))~G<@fd#oli%UcPZzirz}4#FM8RB%E9&4{&cG
zjqJcm6zs*L;R|2Kg@wJxeh{b1hb~rB7{P`<FDs5Xv#iqinVl`FE;;=M3Wl8na`bC7
z*IT#iXozNQxEP6Td-5#Lw_GkL?6=~u*bpZ+t46Q`Jj7`_BpXTi&^;R!y-t|u#jhOp
z4tyO>$SSUUKB9f<imkt~x=I4REfP-0uPcv7lQ?g)A$}@NA;ghE^$_9cjzcB<01SX3
zP|O!T_G@+ZndR##V(ixD*Rb;E!tW)L-~GfB(muE`UiNqmy!Nw03VNlbq@ERM2`M!z
z;#B0v=NGa?<M>NQ3mc>4Yt{ST$%%?ID6b2R3hVa^aP{u$Vuq<bnEAxvA1+lIY#~0(
zTmQq8DoQ2X2G1Os5s64de0zKNo}CyIPx}#q8|8lG81H9PS;C$x1(AcY2KAk{rKr@&
zA)!iry-hWdk4Ey}sH{2tp#AnOd{)Tm9;;TJ51ce9kcQUkN_6W+rS&QvL<{kwvfsdM
z&!+H9=8Y~BLr-;ILfB$@i^_sK8|Vk|;e3*W0ANsx`GRA{_bOCE3W{JT=Nham+`8fx
zRo`oKJ6+s;KT^{QG&Y;_lAQ0e2DZbqGO~)ft}2nn<^6``-G;?%n%OBQ!7B#rZp}E|
z`SFi^gTq-ClIa;4$;o0?EiH8@_cU+xd8ksMdnP|SN^W4k0fC!8(RFDkFP14e+1PMi
ze7#0Vl6Z_ub)CP%%yQyPssA$0Z&kTXOjA=ngZG$95Ph4kK1I9!3K?sig`S3vLm33R
z<Bgu~Zi=34^8dD3GNTd74#BM(6sOL@NHaPdoDf|iY!c2J$f#V1gj0LyaD64Dp{dDk
zIC(3Gf+tt>c0iy)!*5G1`+fSmPPQiE&FHTQ0@%2?u<BVgz2AGDr)p_-Vz$zSN%d8W
zu}#)lQ8}ER;@-NU041;+3hwW0aAQ%#)kKMBx(IpJ(7s7Vv{;vNod;^y)}_sQz2SG-
z;L=Or&kEq)KcQ-!!#(exhPuqk@w2_!IvXt{QoUVGBY6`#pVZ1RLVa<lnp#`wxE?+f
zYk=RA%=;Id#6tpZH5jo*oyYes+Z`MowdJT4)ZuC&fJ(JTW)9-{r?vP&x>sM76~n>y
zU~N;=`ij4Dp_Nf^o9&ZImwi!hZz4r@<Fko52LeL)-x`exx4Z@2-+(=rA}&bTl<OT2
z*VMRqc$!g4KMe;o>M**?GX8bE^+U(oU^($}`ch0nqFtxU%kby>f4G5>d6LjvL%sDQ
zqpfx4z2{EvB^mEpy)6A~1k`_g!m|C7wE~XEo4)necOzx&QQRSG*h@5--(52I4+pWF
zGY9cr?*2nw@|W02y~k@~C7GW=%Z$u{v9USCLt9y2e+cQNkWz3xR{<augc6Y*Jvl7W
z**f^l!1RcZ#M39(yElb^*F{%^uq)w8b+Azz-y=Z3-ArU-!VMw>Vj!E{J3Jg88EF<B
zP5cpJtYnpy2TMpvN&69w?s?f<#Mq}{o(iw)K3nY<nl1}_iMZ`*3U_V(qFx!&L*$xj
z7CIG(z##=%_B#ni<Iw_xSpBof$U8naFHL)_v#YCVZ<y@;SRaWw-nX5ZMloc{m5faB
zoqc$U>1in`dt5o6v3=%r&}iF2#IX<p!Xl%pq`z!8n;@YRM7Ceye7TRWBkl#(S`WB{
zZoW9$4n3t)lln09n3vla3oNRcqt}BiQ-uPxGVHdzXB=8&;VXJ9vuVvah|_9GN<IgZ
z<E1h=rC*UG%toVM(!KdeCnkO<p&7$r`A29T1SlRcRzweBUd71B$R5(zh(2NX-GbD~
zcK8CVkUn-BY<;xAqA|2EyIqH!NlFNTgJR)<y4)7LNjlq;Cp#@M=YPvBzRLZyp0&O6
zHzNpMJ^T=eNW!IP&iB|pS3?T=59NVk{ae?GBl|j63{=R42K~Bk-bT2!w;&UtV10T)
zKqK1g#s8HHd#|kXWgp<`K#nm4TAF8}%+1ZQ>$cnis9pzz<3Ck%2*KUd65_$$ydyN3
zhAhlSteU(bSebX9%Ll9Hd(qW6;TWr{!FX-5<l&mZoB>YKnAYDSCy}Eqk)UcDX2puH
zH^B!FOSr82z0xnj7F`Y+Ey+1}aN;$7ErC%=w{E+5`9*~pyqU(l8*KXi?F6|EvvH~!
z1y(fiC}R|FS8s27^zA{*69WV+>%KY{CuG9s>E5Ejc<HKIa{`Dt;V0s~c=3WEGc(f(
zjTifYr>X?HrxFSwO0@b3m4`ddv`zi{6@|+jlP_QiS+w1t#QdfP6>01zd+qjgWxVUQ
zGDC=1MbGBPtt5z0I-(Nvxf!U|RQ!^H`EBw2yLSofjK2XACoeBA4OV#^?iI+bf{iXl
zD2^DhsG4;4c=inPRMNozgDJ4t7biS9nVLnTG8BSB0<M5^B}XQ+-n$2K+yGD$JUVI>
zOUd>6?D7g<>zjYky{anSa9L#r)MunL_YD2K7B+bjQ<}v-eC6WWC77yq^cl(1kkgaA
zn?Axa^}FS--wjmZ!;KODyQij^K{)-dS|d)L#K-ape0$cwfPh>YG042&HX34tctIlR
zX0Yq~JvF7-HNX2A(9&Ks1GMs~m`0uNI5voTh@U$T@i149y0I3wd<hPv{oS5-5_Z#>
zvdH6M-jrBXef^p~<lj_hg1)6BIfcl9z+3g<)C~EARrt%a=HMhG4Iw1PZ*26S%<w6_
zpSUhFyMH<a>Ew>EPGZ>NK?6d{u7h~1u09CnZS^NYkTC>8IYuEN@Rr&CkzU{P=s1vX
z?d`3%!#QH9J~+wgfkfY=7}MVA^BA5l@qCXtk=a7?Sg5=OO>`89HC_;w8TbGCo0+a9
zU_PP(UlDXcl8E2ncn`w{etrgEz0Ym4rV^vIL?!I}1FlOz<5cUOJ@`Gde;S8)Q<D8h
zhy6p^R3_9?yQhNV<%X&exSHmvSJVa4deRSO@RNEUN4ELb2c`YaJnkge7CAN_07|Um
z?{7cwibUF{rZIvsr5LKm0}OYj95ZYvU>E}l5ls>__h2|Nf-yyA4s5JGn3%|cSeeDf
zCRZUz7|JXO!m{p10(|8fZ2ex*?AfMflIv<L$%JOzKh;85tvBa7WXQ-b_}$Q}xJ~yY
z&O~Y)B!L-WBIe$?5;2(htQzP=bgv&mTYry?L=v*R6Nj)l&icR2O--=&4b#9-u^=n!
z^c^%gSniZL)vzRGmQrV!4V6UbaCZ!M8tT4><DJWktDVg2kaGU^v|n432!D-`Qp+S|
zi55usSMU0omAXQ|wcx|22|uF|?qmqNP=>4{Rdfh_htV!;e^nlIv8zHZ?VxMSCB&1C
z&-qvgi0I)k9p~1lw!RIiI?rJyK(_V@CSV}Xlk8}7(hHQc9r3>b|0*?_MoSQ=Zq7EN
z;I=FcA9D*7ToJI~-7q0b@vsrnsIx^Qx~}p0xk3yT=0gYvL_EN|&0#a}CSK6x9uB1-
zGv@hRD;Co08}db3L0SoMD?u=m^FzDd^`};C^^hT?#6f?<4>n(-kP;U0Gb`&dJU+gL
zvOyZrkt@SR9Q11CZUfb`uww_x5DYXI&A($$z9i<~Jvg*Z&da;aA}?=XXecVHjCSVe
zSol*@R{s;!5I|hVJ+CNoywumIzXvX-&hQS=5OT((dTVKx#KI<_a+2I|``}ufkzZVd
z%pOghgUu0!4>m~vydL-FtQzVG@cNovSb)Ugp$^1I$y(cgZRUL%)IzX@eyO6Ws+!25
zhn}I3bfsFdoAep{jIk;Bkx5Q%u(Rq5|Eb{th}N9tK40lL>(zBTLW*`}Y8s!Pf4I=9
zbKE@EFsk6+vuTf;BVHi-3ydis@`rV0djq5tSl40<J148MCKKhgD|SV1ux)H?z_1yC
z$k(pF3jsC21UEpKFwC)tgOWKJ(lcxvp)9dDJ21R}yl!hSruBzvIvoBV8#zA?czlI7
zEbh8X_@j~E#mLUiR#a8>E-uCcg`1Qpu%lnfBk7SBoO~*}!B7LK&H9j}jToArxTv3(
zsGoQfTaZpmE5n0130h<im&$O}kl>U&^SMguJPpTmsE_oY)_i|~LP$Y1*6DX>AuF9k
z?%Qn(H6)xT+wbgfK#&o;{SNNIN<W$)<#yc1E#$ogRdWy=pwu}=$la%i%Koi+Htp68
z)V1k)ky@AY*V&u*VJmXO;)I`-0s}`<wRiNIeeS3XwnS_BOg$bg|9*R3Qqk8o?-AQ(
z_fwIm(5NGy8_uppR`{aH?oi6UF>&frKoc|on<niAZyz6YI%ty4)nC56d-KK(#d_CE
zkhBFN1Lhz;F$Jc{V#<F38PN}Rzgb{9hX)msv#IH(FG3K-pvb6Khh13A9W{!O?_`5H
zx|B*d<V9>OX&N<Nn0qe+WJ16g6qtJ(8u&?RYlpyM@#|%)+!{0;^G<@(VMU{%5$}BB
zMD2PYk%G+dz%O~I)|lb5(cH5P9(gHE%><)i&WAW5b>Q&Ngo%xv29G6M#=yuZ042KE
zs7>cLPWfKz38mp}onKP4#igaCNVh{Bb8|M;28gc)#F7}Wvv%0D$*HLiRV!ggZGI)J
zxzhejv8wzw^h6{WcPZP0J2M~L$IKR|KO))Tf|rOIB0SaKlt-cw`@{p93)|qTh=(r|
zgdy#f^#1uTfB)Oyv#W@_iRWJ`D6~_#U7*0CySO}_hSL)t{Vw?}J3G6Wi_rAU%z1F-
zlX$EPZ&n%%QDhKsa}{Sa!1E&Mtwdu-(Yf`E%AM7H91-u1It}qIMiS~doi_M)F=)bW
z!KI_9q~t*wh4qo(qlKz!<KpM!jJBImoj0R$0)~dTk!ub}Z?OIK%*i3kRx7v*j`Cmt
z+5}M)s2)FLJ{dtSt_1`I4I*4oaGSG9kUhzKdlR{Pfg<BZ!}2~jH3SN~oJkR|!=FJ0
zRYT|~pryW7m+AXh;HS?Hd2U`llg(q{7|@W*FNr7dTcPMNIL`mmc1x*FzPNb3w7mQf
zhS4F|H0dHPsHVo5m~8l&JpRLIVlt2R#|~Ial=Whj7OPBfg|qD*=Zu=`4ji1?Uz{Yd
zxf=KVK!Ie718>I1{!oX`z}doUzw@A}sR<%s&9RBu5#YYhned+RzAhlo;Wsca0;@E(
zPGR+^BFu_28g<y~nm!<mQJz7#034wR-{$A`_Pj8fy1OyPCFpuV3^r5H^MQix36xTy
zK(RC9B`YH6=sqIPS$LCQz_R!T>hpeIzfd>{fo6o(2pNYgnp`OLA)>ukS9PW@S&W#+
z=sw)KD|*`NI5WiVIJrjubq-g*ZXuY@4m6@*T6$<GO@RFpJ{8&7fpcp!SPd$sO5Vw8
zT3lL!q%IFB=`(kHUstx#fZ>vzTa{(20U}>dle3ghDRh>Wmgd$D)p;G($XT`Th&TN@
z{@v5#0mn^gS(&1|I@WNup_sZX78BDSKbK4N0$r_#^hZ2is3=CU#6*4^_^Bz|G|9N(
z8QsrM=4aS$D(m7+3LLk#NGr2L4i}rNIhde>w#*v>-a8J+A&mpPR@pBuFL$@51P->R
zsUUYM)N!#Gra<k%{<U#*RP6%N;ts$(e;Pt2O%Yl}OJF)D{2s?W*?aCWHFa`JARMq?
z`AR;`jeF0-2?wAVgtdsnI3f5fvOAn<43(F*rAP$2{_F~uqD3aRLn9BTC>&B?;=G5V
zal+Vgg_!V$4M2ST$Pvw+4fobhs2*M=t<%szWB|ljA!B$`;fnZE+AGAv6Y;LXOyctF
zh`C~Q?)wRmN<lu|=HHr($yppI3SC_sL(3xx%S^UJk(I`$kgE*%H?P|9uh+Dz$*ldq
zQk^A?<ZFm4(+4F(yW4j)U4a{tsl}W>IW$^o+0Dtx&HVsmG6hQtkejy(B@F?;EG`7o
z?JxH-fSYl)dIc*0dhW2MK7NImYcI(VN_m<vMs0O<0qh1HXI+%wMxBek$@X@uw-Bz8
zencl5#hffX|03#d$kE0>Z&Xy&l^3+^?4htVUxK55I*f4PCG|4gxf9&9Z$^SRLR-1<
z%v<#-9PlTuMF<PA2e_dzF(i-z`RAY*@`jPJz%Yn#ueIM5%_zKm`}R;}k<cZlKCeI5
z+#kAl+L&E9C^1L{jH0{Vw~|<vnl0)MWM#-}@|UG*s*gsj-LIlaqQ84&efbI@%{jli
zb48&tQA1c}K8+mpl<j^9qFH{hV4T$2vfKWb{zy4PF_@GG8DCuLcxV(5BNspOe_DXx
z;D|?GA@+>lVUHGa1SOG0_D$4ALiX$T*li(xqzEU(L&O46Kmmtd+xKx_zgAtNf0Mzi
ziNbpl+I$AjwD*pWGpPt5%BQ%n@CO9DfsH~27=0?k-GV&e#c7n1ab*#OBsV{8M|<ic
zpT|o@j6!i$RUU;0xP9jJmy*!qan{n>ZvT^abaX7$ZN&r#kTbBH--g+6wuXcNj|Xen
zAv`XSYcZoP^3fzzwBN#0h008)Y9?P3d^`;z;rreeyhrQXX1lPm@(kdp2&N>-AmZoH
z0!E!<!3@SAT)Kt}zr&!;6oYA?H1I0z2LR+lz%p2zayuRv%<Z36p?WkB-Ne`?!ESTI
zXF&*|oS@p@dVrJk_MwWJS|;4skz;n^l9I1#t8Ex}P3?>7J}>lo8p4$E!Nb7{^JzBQ
zFfcY7+?x6YeT%=nAm#40xx~}k?$hm`-wWJ>9xF9qfTTek8Y~E?OaZ_kMhhES<RJ*i
z16%J@=$bn(bp0WAm5WPxyv9|+`$qTLB31{WZmsoaYHRF#kBYe6j<^R#2h-v8tHySm
zQpAmhdel);@O}1zgs340m_U^PtG`z(fx-+7zAqE_C7@k_d_k7Afpo0X^YhLR0jF?T
zasz2Lw`*VJEbPXG;2P;74%(*VLF6HrQ7p*ZPppbSi6TXd7bZnZ16Ml)(L$u9P}W&6
zG+=!r!H)wEBLIX>fxmOzH)lN4;8AKknqP1QWiuHh#|n*2OoUfg^WXMvYTElP?$O&u
zjVOJ4_>rAB$Xr1lL39HZWf;gxR;?Pwec4%v;rw7=AZ4}Gl?<ydrg?d3=^F$;u|%(H
zQgd)YZ&X~KEnCD!DMLXI9b}o=sw8spS14VpWY-s*yJ^!K8$XWOH*WlL6R3_08K+ff
zng?MQ{)DtVsAiBqKNDIQrI#<is?3`ItC1F@41ub3AJ(=bV)#vo9(x@bFJc2Bhhech
zuXBY}l%AaYT!S?V#!AJJ$)cVZIIp0G06W3rva<B9F1f6;K){P&w$v0J19UejA3Gy2
z3!B!V>PUmxCy=(k(gYa9X}s3yX_-O)m!aWd5ut~X7&Dxha-(EL+1+q1Mqm*&kP{F@
zXb2}Mz)6$|iNG;hw?GsC4E=*_<;=SvS&cex<un?SS{uwEgRR}^(q??|cXmO?T)j)$
z@@-OX8nMwtHQ$|`6^YYT^T$(TpD%LUq!kp@mM-4aIrBjRj$8Qi23Xz(P<{`=;3Fiq
zBtzFR0GmK|N_TB-O$Ihb=E{lzDdz{jh54AmGK6?838(OpL<XcOlp^913NUY>V55Of
z-0Axa`CG>O-=N-$iH#kAl+zX9De3ZqGSkz;f`fzQF6CT|AODecQXW!c-HVY1U(n0<
zt!hx27(n5&h#40dDGA>sgH^K{H>CX`;0|DzaO2sRFA=iaA>oDpRlg7NP{grrpc;)8
z)73hkzFN9_30AYn_7h^LLmeecXh9#@F_J<3(b<nR%DqF@Nz4=DC_S=UlaXgOU9Krd
zb0TsG>Th$J8pg*SS9fY9=-b*jm&YK>&JbG5Y3FCS`1oS5F=!yhGg~PYT|S1>8+@nL
zH)3IAJS;2Aki(LYghUz)M#FF2_>valY1$Uz&523A-GjQrP^Td+jWUDw%=4pax*m?s
zV|SJB7G?l3g+h7ZU~eB<Uth2Cm;)}D@VC5UW8;eYb2+*b9kn$zu2Qo$M;mXm!DG5>
zxj>c%OjD8UKlCiLVN&9;v9SQMXfU>zCF@AYCGnuasTfK%j1s@Ff0~w=sm3~33kG!D
zwi|D<-<Hs8);U*+vV8%!7S3#?1i)1Ou;`SVX-Fft1FVLu=7r4q^sk!}iql$C0p0np
zkCsDH&wo1Fh{f!i0-B&|_Fxa{m%ZKHWI&}up<U`u5=nxYgy*0Lxm>eNhPS3vW;DF#
zB|7{1KC#pL3%h=FcG8B`kh!(ai2EMyrQAGH%3_}4H<wlC8uS+)qfxsnA-0mehFM?{
z!&#aRCh>!?m_KfjK|t`c>#Oqux%cE0q(pIY6E&_kgDC`-eDSqlo=+w7B}Mh;#em8A
z9>{eDa%wP|L+>Y~CZ)m7gM!!#6)PeNVUODPq1zKmz&|2`8|k^Z@vZZ_NR;4vSeO!P
z(TDf%pTmr9Iy85f>LVpZhT=9XEKIaIaA9&{F`5_b4&|G6nlsgGO;$^xpE=Ur$6SH0
z$GJEa%VUg8(N?u>uz({imfwbn0!A>cZEPM<Q_tp%LiW-tuyvpnP2OQ~Kg-R!%A84Q
z7$3il;7h{UICZAnwO>n9mX8xWv(Bs)c-kMa-~Nd91QW^?A8VWJuxAYlAGvlh#RwXP
zdRq$v0x)+^oxy+m-s$P7((Bi+tIofioSbl8Iz@;pf6CM)#HlcSCy^={d@<WH>UA3*
zqJPC^kKxQkwyt{z2goG?II_1oz7^sIVg*B=Ck!4(S_X;xvS@N>d_ux1czG<wy2L^M
zO+rdK&=JnCI51Z5w#0{=Y(j}^nLdKa5va!$fS^{J<34@%SNKMv05{0@Bi3)ifXtZF
zPz~lRC#Rn=IuV_J4+=Kwi^0b*`2E8bS)s-3LAuUc)w!K2ahgYE!c#R4;gD3$s;c4Y
z>e{lrU!tfZ5xnc7=ldnk*gpx2_0FApw{N1ZO*ST!pXX=zG1$pWk!}o>-iGI8{x<SY
z!d*n-T!|_xTzs(F(is#If2W81t+afY_ohg6$pW@fEAz+}JVm&$uf%S%nVXw`wzbWG
zMnI<49Maelz=s3U6J(*qvhwZ0Kv-Y$7D3fY9Ms!)bEbEGS4v$y4vfLy<bUOb05Z}i
zCQ=|_VYay`MhN-1$c7ZwfjB?^_-B04fs@^V!Qte-3AIj&_Rd15Qs)lm&(?(x0mzs`
zv@Cv>W<+l5C#p}mABdp|;rZVf%22feD+;CrEZ#RJgcVj8cCkDzcuKEP6RziR`2++y
zzPpL}U{i~T)H4;)Nz&rE*brDlm(FcCK$|s>g^dlb@q12gu3$9J^fsO`qFvF#HO)Yg
zZDwCuQ$84s$fZ$U9Ca+Xv2AM74(Aw4Lx`{M+oL}>u@IR%V{hTX-hoB0?ESh*rm3St
zTEBu${J-$^jnOKJL!ITwGH&OSUXh@kSOGg$9>}|H5|Cq(fdLl^5!ZNr+l`Ydv1L01
zA)-%D9+(z(1RX*TV#Wp>1Sw%ZmTPc*f~8sdp#u*J^6)==8O%9!P#OII3XzbnS7U)o
zcDCv9qb<I-@1nz}x4JofgI#t*UDX9cfC32h@SqH)<oFI1;Hx#hlFAhZYv5k@68SwQ
zHRAnymz8VzBZs=i!T?iY8sCcsXaWM#8JU=5)&xxkGr$7YHcti{Z*ppCHALi!arY@$
zGL4sf3X$7|z_BpM3S$jvk86(Iv*Y<O9ing{1Y-<yyF(r0*{r7j3g@sgtBRdz5+|vJ
zLl~Yn29Ty+9JDh`h(J86U+?lDloXF1J(AvXX@(R*!;_u8yw_Egm5SZ5^rf!DcqzFz
z#%oQ5931bS_Xg%e&4GlodElniUp@+;hj!%R;^HUBQILQvs)sw`1KMAdh5#NmbkE(?
z0pbP`*n7xK``KpS#V3g~3k%{9DV_}FW=9zs*a9=i%Lmhx?Dbw($)BB7Td#I0dEYL$
zIH(x<S4-gF6dc93VLLti;YYy>d^Zwl54S6&HQS~E;i{~HRkJ?2*e4F;-3feUf6eD5
zujM+UuIX(k-@SSBy;C%u<_+M2YHAh8j|PB-vz>&9=mnfs4uE7O=1Qv>g3wIdWBW%+
z#C^(;7v1p`NF@BTadV)HewZSopYB^NGaommuim26z%na={7m%B8!mr1os_ERuXOeF
zI{N#=Qc_Z&HH~z%UKdw?lAo63N_vn)s09%VaLkGW-v7#vIFbRF>da}^^I$-~F)Qd~
z5b<bzK0M&(v77qircNf@Y2spHo?s^>0Ysrj&MU~Se*mG|<CT_rRdb=}EG+UeUjrFQ
z-`t$8eCC6GW_@<z>g~r?;IEfu^};U3OL-&EcL=<hVN0e$vll7L|M`^`zMAS`V10o`
zwR>)EnpJsxT%4@4a1!LW<+RB1^72N*)cVdai~pi!wPq!cs<JW{(KFeRnjCT&#1cU2
zpU}BCOvwfS|M_4<w5^*o(8kmG(=>;0f$xKk|E!g=c6ePPE6Pe!v9><_Y<S)ec<E4V
z5R5QDq5mx}k43ZIwN_&mYCZ)4+YQFG7vLl!4dpEiG@mEk*MyE6*{eXGnFc7mtD0eN
zXXiseil8B+^cPPNHTDldV50~S0d9KLj13-+kDvhcxkD!D@9)33CazuUC;<xy<O>Wy
zLd1v)n0)%ga3-+hT+}%1IWhV(!~fpl@rv~9XFP%L5zp$)r=K0Jy?Z5dftA%v4lKs%
z+FEFD{Oad2zvzhjVj8R!e;pz@QU#^8wUg`x6ICYU0BZqs*->?AF)k@d0p5=G$@%R&
zcj%m)&M{jl?%x*&1_0}s&`4Jfb{it!jj0+4nc<3jS+_(XdS!Lq8aE7^)uFGYrQEBT
z9xf|TQ5b-B<%5xt9DZByMat%7oeP@3c+-v7H#j*eDxWP40=7UfLIpyL(6TZfxHU>9
zC$@5PbL$G*oKE*XLWgzD_rH<|9uFukip$G05(MlbJMk=4XF!8Gl&>iO{|L@uEKP|n
zRK;K(F<4pJ$tdpU@LG@tKXzM_kL)t5KJ`)}L4?*oBDFsO0h!C}u%})8(TXEF36Uog
zOZ@Wx<LSHOv2Ne6Z%TtwnWZ8^wo=H<sAQ$=WTv4MG9qrPl8h)LNk&GpNwQZcLPjKe
z@9e$b<9>ec`~LNOdW!r0Uf=6Fuk$>P<2+6SlMEXiKcV`TnCTy->YieDHU9USP6my?
zNl=SaQ+=siDk@M9Mj0SzV4K~T$`65c#a{xcw#(DKLURJgj=5pFymp*no<W{1xKnES
zwRV(s)y`f08bDVRkL@QCj@?k*4D0t*<2#u6=@ZcV*yB4NDRu66k{TWt>A8pf!2^D4
z%l3f9iP^X>_OB~aNtBd<M0N~AYAu(*s$#}O+4qaWy9lQ%*^Wv{GaAl*)?O2KSsgvm
zhI#I}Wh|)ZIMpKb`z$eFg6{C)waEPjuIn>9#m=661&*5(-*Es&0X;V5+T2OhPGM4g
zds5bd;RJ6Ow(4&TRX`|YjG(F}Pao|s--}F&n7vG>9G-Fbf{FJitp}xDn`0FA-2+Ak
zY*<@wxPZ7~{Z4vtd^qhJ%T5hd+89cSlYdZYMw#D<qo!RRW<hzQytGx6{AvPuHcQ|;
zrVt1{Muszr>)p@&s$aQYYP+d@@@2%?P?yb>I|;KEV@(NKW4H3*y1^v&IcHzQZd%UD
z2UN9gWZT%u-HmseFLnpD5oo5pD9S!wrk-CNJf6VEFxmAr0=LVRqty^yC8&k=BBT3a
z)wKtAHfy(Cwa?c*EMn<_=h;kzc>RP9kiRHBF+7dBB6hlnq@DgZF>x2!oh?`hh>8Na
ziOBei^5&x9@@EShw<&EaMRpp}uy+9K=*T%5F8TO#dCCSN+!^qiq%#Z<IS^J?g98IZ
z`?|2VZ!e*Y<hSnGgGLPbKW6|7EM1ZyN}dEhzi4gm3pci7^78Vti;ExOK%z7eq=teU
z3Av!%^ey8kg@;$S-Y5?&PjxCG2eq9_JBM|oA$owZ8EvZwq&#34P~*Tft8wmLSEAkP
zp4Iux=DJ8<-<Q=j`M6zl%x+&hAn7pecCRJX;Gbn`YAWfma?7^>>^0WcS8g*dd$%Pj
zyHn<@v{<|F>~QIYi#F05&!l!mul}(kWifYEO@%}fW%cx!nzHRhSLgm^(3AxYGDqiL
z*SL4zQr_OefL-yn<h4H{CkVyqH@JhZEtdzaqqzHd{|<IgNHsfhJ!w*Q4B3l{-B{BH
z?El|bpHa9~#XWbL9oU_lpYOV`gfsYid8WTu;y^%XsN&}4ChnBu_d0TBbW+ORe4&`b
z{K%#}JM+etmJ`V99Pu9Pm*9_@J1{(~z`($OY*K?@mN`@fv^Tr8h6{dk)(;<Y>nOZM
zGZ(7wkKmE2A6V%}JlLI8YaT2<l6bZ370r!rHyOkvMAAMmHSZmrDLkz1sp#hu^^WtR
zZQnF=?(P&)KC(4cZR@Z_y?U$0O6YWoBB!wG$aJd@hj-ds_7lbKk|>oEe@Ux~;)q8y
zCMhXN8DknRY_$OEMZC-HM5cNaCCLFYx3WBUrD0;Pj3F-fj`lC>i8D-ftlLJiE9SU(
zxt|*r6VIPIu#aoQS<|DvsmP#6v-3(M=S!9gC-zOTzP|MKr+1ttGvk}kOY8wZ$BcnW
zkDw~Go??N|>Urfie5{8LA0C4T7*2FSdGLE|=K8ea&DCZH!y=+0CPNCPeRQTG9d(R=
zKZ7A$jeq`p1pU#7<iGl8cjlcsl6e~nm;`J|+u4CC9#dOICV;Ud3K@5l0!sad1cziV
zAz(Zns>^06N({wUJl0o+BX8cgA!qvC_rv`ByyfZ*z(Ga}fx+*l1J0ah3*B%Kwtt{x
zH?CWGKiq0sX5YJ$77n>PpFeC+8v)38#$vusv%Ae3jE`2*M^#{ou>i#D@mvXBfwcU3
zS&K>E#@ADf$LizEL$5F30}=;vJ&_bNfRgok!6hmAo{yh@P+%JAOa(@0$6OP=l3><D
zt1%0ygvNKz5p{i@f)DS5bUl)pPyD!WrEW6(Q(PETXy_kyLYeMN(513(v+CN~AFsw-
z3~ns~QB&F2sBXewK?MYrRZvh6ov1t_Uh@N0v;b{5q@)VfYr91sv;`p_)VX!xyOHb0
z1z?jWz@=cTv65`4bk=XOVr~yQFO(51IF27bu9t2s8+>l3{p^6Ekf>-pE>Scv_*3rn
zmE0Ni?Hy15^%5Rtd4>*xFYT`F2hX`kGD|wR0cc|7<lJG|T|}bW@Yqp+vS1|Fl&D4J
z=jR8Xk;<ByXW$cn8>ib0U$z~o?+ufFfzjS)&YZt=S<(GN-X>%5otw9x2)=r7Lq%DG
zHnZ?VS)nQw7xLMolMW?^Itp35UlpfSt{?fa?N`a22Fik$KAw;g>V3WM@!FI#%Azx`
zXU03;budxulR^CDgqhZ}XZt!<=I{-PUd6@EYYZr9aHD9YTE7bm>mRHRA-I7<``cR8
z!Xf~_7Tm$Xs2g(NX6wB6qm)B_=8PawXZLnu&3lmp#LbrWkBNzWhUS2R&bmz7*KFB|
zeLHR_bF9a*wz9ILdh8qH7ru42{ku-$$%M9ny4{QVmfSPuug5J32?@+EH0dk%!i=O6
zXk<?T>q0$jJoq#C@0Yu@5z@}ykdz!gTML0qSE18~<YZr@2GKZ%TAZ0%8%xMs*7|!&
zpF#lvN4jGe3~X}yHs_L4LQ;}x<ttXwEO%yU;tTK!(eNL`bT^9czK2MeD0ji0I1aoP
zNcYZT66sf<<020z_1!u#bRNTG835Ta;()b(9{OqA=W|u->dtg!MMbZe*w|;lGx||i
zpA;5`>4|Vdx`{l1nSNl34x8&@_y#l6z4GjwoY5fcF?d@Bv*MV~x$Jo`kyzZR7ynMA
zt5|LOfvS17yHj0CAJ+;VANGW|O5gm<M$b%B9QD%3Oz~SuXXXHP^$eFXPLQ98nx#vC
z4hMz@W%J+R<+VwOxKE7*<Ags=OdMk@#yF+@U%!6U`dov5>rq*Gi6BF==4?sjUvkRH
z%aha5(Fux(=wc{|Ouqh!&inmGC3c!G4?5SwIG@8R`*od4Q+BH7>;8|QV3cjlW__bV
zb5GJY!O+l9e0i>gzSvmy_u_Vx5LBgp*^XSX4+Ao-Y;DhjbR$Fo;(^!=R;@s(%YOXW
z31F@Se-C2`GIyD2RgVf}Nj+0j{`e&%W{-@F%<ONvNnBhhNP1|jN)a3$_LPv-yceVG
z|NgQJTww3$=x&@%%K*{y&j7)jb`?CvIDu~v;o;ru&*w)(Fym+&o0|US1f_S@%^<l-
z=<V_i%?mCR##Q|@u*&0EC=p>}Bgbc<=G%!X=HbGa);#X1pXZVj6M2|~)4zWWcRF=@
z;G{?A9Y_ifj*1<mUToa`j-;2(H$HSZyzqqwm1_k$i1?a$+Ht<G<X`R-O8Z^k^dtie
z@YAQgLqm%dQy?E-t$7C&U~;PhmdgE;^V|-*pSpv*f)JT$08{!CjVpt*h{yAfhY^4>
zlN<WN+6HS&6J&FG!VMqgr#MFs9g-$6e2!jKCj!%RujcP;nJ98zt5484F)}u$eUA--
zrc<|T4x($W4@+~Q@7+kXM-3p_804P7$99l!+je08em9ueY29<ArKNrN@9(K2XRQdB
zi$3x@exHa%$1X<35X-zaU|pv`X81BoF_o5<lI9yC)HWbT^gVR?W4XOo^b23#^6B2P
z*MWgg4~tq5GoGv@EO%wz*@tVs0I_m|i}2`q;MKca8GOc<WTgm(lFa+?!UFGJDE;=0
z9M3Bc9Vft+T`Vjtwn*{JoOHFmTV@Mtr*!VAYoL8SJz3b%f6lRwj?M7A$z%<;x&g}u
z4}<D;+b1&eJHZ2%y*O|@wySG-q~ZA6KSCj6dST!Fa$siYRjg$<qoteHRpjgg@SSTw
z8nB8iW?5__9RDNCbwrBpJqtM);2(<*5;gmd{CkjbW0W}UFm{?u^G?M=-12H$&yoiE
z%$_&I6xUw;In)26pR|zV8E0k&|NYxTIC#cvKZx>}=`VxB!~KgB?YbE^C=lvkxJ89k
z|JL4q81z6X5}e+#F}+(~SIFElGf(3up5mOL$}|RbgkgcwSm_s#8+vqxyhukTjS;aQ
zFi|q+@!aiDyEK5wX6;{7c>I5u(%vOCFKGFB5Q2b9oP@2WwL)X^wbRd@KmS%+yBF8N
zS-4jQyO^0L<vBPkiSJq1TVMY{jDi39Q~RlNl^m0?12<LPiSLxiPUg4KK1hg=cJdo~
zkZ;mvYniN<MalO1{-%q(oV*EVoY_~K`%mT=1%j5_A=gVvUS73r@5L-~I?kf=YzKgw
zB_&5k3q{!N^=7H=jj>bQ`+660)FeH@D;ykV=6A^mj5WLMS=$JEC_PlJU8lw>>?W9X
zt+(GlF=(BuzJOg+YGrNB!}JAse`#6SHn`?lD=S-A2;g+4+m5P#vFv6R7Z*P$<#@)h
z*ySZy1~x?7ozzEUpgj2yt)3++aM#@26p6XR>O40@($BD9r8PBsj$iaG0jD<Ln1)t3
z?4{iben(0X2~Itpy_s-=6Q$m6Vc{%J!j8`$tN)A4=@j)Lb;G9Eb@b9{afylE*D`PS
z&JI=|6tiZ>m6|wTMzx*Dcm2%HrL8>Qzb$E3ySwNYRohFP7Ueot$1i)Gj%_^o;5+&J
zSi>PeB-I5XLSyuLb|W6$Z_-LoG5xeF22b(fjYphj%OA_XyL#&)Sh=|q{&t^+dU18Z
zya{)U(wpF~Uqt|-y?Xoh)0G(t0y8u7eoBw0x=T@3e3ZqPE|E)wJqOacA?5hZrjo$n
z?&cay&*-M6T?pO_`&x6X0@J&n#mC<XRwOF&8Ud+tm>b$llJ?K*i&}KJrKSo1k$M6g
zt_K90diQY%)zu;-OObtK7?+aIPqZ`s`SYj#&!6WQ8i8@2q*k%>3JU!|g>*r_K^DBA
zTK87#8JOi>9COl=*VNWF#H+E8HZf`mq#(&xI^xwLFXC8ZuX~r2$RI&c{y2@7_!mM;
zFaVbcKa*F4c(y$n^0~_3L$D%I{}3YOUbu%b(-4P7Au6G(K^7vSL|aZMLQ1}>CS}|z
zQ|h34>^8@kcB6yclHKLfH??hReJv|p>a`K%C1@qcAx`Yr%yrGIF+Y7evp%!?WX;=C
zv`{E;>9kecIAwVLeALH;P}}e)2;FUMy|^(D+SD4lsb1<<-l)oGtgiZd5aY96`1r^#
zjCr?H4E6Or0V5aV?`$SL&PjK%!LSvs6-{Rn-bqM_CL4O69u_!s>*h^rvh=B)f|=2U
zhF}*kk(_6tZ*6JHw5KGoaKvi(7qZ!o9Nq38gVq^1#PEhD184p12?Z4tdYE5Lh4_q%
zl@U-p24};wmj^=p*KgmxzzX&ue<I}uNa~@da&~ZVh{CMZPe=oUjT#bn?bxvcy_6yV
zjJ&XB#*p7kLr-7gz73K7E!#uhy|@ZU|0kGN2d>rx1%)5h@wj{qbqgJZr>2UxA{5S3
z>1J17Idgx|=PJ`LLLSs%H4#$Rhf*wVk#Z4I5>L7@u`PKvdcptJ$jK=x?p8hWulMvF
zCg#pUziLLm-qyKclGYpwmpG@_-BcBC_VOO136}B^BmTR2=5t2HKviJ%#`|IAr%+uV
zBPgDi)x>s_wTof8!d>dv+hW%Amu$F>vD@+BfdgbVHgh}%CSpgZ)z#IZruy*l<Hz(h
z=ng+ze93f#u$UZnh3kUTEjznJt!uq#8ulGj9+H=5*Q@YO&WS5){M8BBfaK#@Qg4J-
z;=UE);5g#&h3nMkNQSF_aR(vzQ$a&N|F4x6oBWEpI*)ZNl6(@Hz8zVYCrWc{`I-4q
z4U$_u7L`wLWmS+Wt*$!;jQTg!L<g9dC_#7U<mbl##Kkx`h0B*OC-nLsmDpi!ZjR3f
zHl=yVf&R1gpc0tAL&ix)zDjN>#8XP9yRW2c!O+BuGS?m86)Ex~kL<;{!ubxKH3oGB
zJKLwJm$N7<G}8Py{8pyE*O_9#`t-wHXVn*Dmmsi>^>*^_W__k`o|S7PjWVvGyPM@@
z*x72j@863~{oylQ?(`BwkJ-{xH?M8A2`f^nLG<+`F&^kNG4SZC2noGb%M*SDI^|xm
zd(qSc1HzPRj}R}75puCOi(smK`xH>r4N7O|TvehmLaZ0mxN=rPg7MIx@g)W1rx|7q
zY{;mO9Xm!qgeAS$Q}_R#L?)2;);$dqWPMrAG2{GTQ9M!Zb~HBNI%-NYGTK00l<8%5
zfcMaIxf9298BP|-nG;ed$jOLAVv*HfUNyk<4Ikups)P!~H8Kl`i!*V4zi1t*sH3fY
z(e(OZ&EEl$Tc1hSRj4#03Q`R4aLnxoku}|3%#zYpJ=D~e{6-8Zy(iXwN{Q%&nRRqV
z^^J{VKl3O(iXv+$w=}vF{k~aVIuD0iexn<~LB?iLf@H?)PeZQ1UgoLpECUV@yLkBE
zp+jEaj<_UeIId9_*bGzCYEfRiaDf7oN=tDP575zVhfeg)ojdyZkM_#AOOOLqMOICT
z$qv@^h8{%yjdy#$u2$eGnR)qE`>ewj&JVpyCaISYbq^-usl-X@-@M6wFEsQghX7ub
zfz!UI>mk(E%H7G2m3#Weo<lif0!|y6s6p>xJW&tTcJ;7OK_)jY?$W9XdH}>SkXP>6
zyB8=2mB(3uHzz!Jjl3nNv|b>WK#gIZnJV|N-<Xi-vF*m=td_JH9!>^}2YEM^ik{NH
z^xoPQ651C3Bp06&XUOOCdl98*FDf4IG~37r6288^&wxAQmtXTqW=B$&0Sym^P?sQ^
zz+8*BV1RHTpTsQVPe0Q=3>i8)I<A0fwIdwY<vvNun?-ID+Rz0T11V5FYQE*YXj%Gd
z(TOR&Y1UMGTPs4m^!vo^dxe&X(x3piLY6JM_m;L&g*=?@kp<zu6NLT~^tzzXJ!Yft
z&n)%imO+wrT#U|^yKd&~y{Nfgy?<Y$T_ha?EgDLee)4)_gXcHAX=RwMq_Zuso>Ke}
zU3e>>d6{=vz@m#%K<9`nMR$c~dF)Ms>sHCFj~ZuFuV`pI_3*fmz!(pgOg-VzuP%{i
zACe<QNS)uwGf}`OpYaLB9tuW$plSkj;9ynqwalo51hJ4o>8D<T7jK;=cMHJo^VlO*
zsRZorN};12EOtMsmxqKV5X)OU%h{m-?9)<7*6*8(nB#rJzqS`DOhmD>uA7pu$R_3E
zFzBhi`ECgb<jGT~d3D59%tOm6V>yA@?_iU=af1!bBT2Ru5Her7bm<e_45Tu~78bu`
z-6eKBzi~-mJ$jeIu*bhP8^TnVvs>Ofrg!3EK}lV;Z|9QOd3l}42&?YZ%JI`aiHY|@
zzbRe2b}g~$E9Qw3KGCr}jZ7nYJbM;2n&kW8AOoK+?2gWL6g;NAJxKJgQp&dIguc#q
zy6UmU4m|7g=&NKWj}Wjaes+kkS>mLogU5qthr%Z<<P%5ywsmL6M;Ai;CX9!hm;&SL
zN|XM!4El|y_2fIuiEL73&O(Gx4iMz^t5;WUa;P&+b`(<CRL@2R9}>BJ1So<B6oPz4
zMfxV~QUlyY*y=jmedB?pQ4-@KxM;0gFiPv})_Y%H6^FVNdQ>-81{1$E$S#s?&bLy{
zps;-nW>LardNy$}A)QB=2&_MWGw|~Jv+e*=oC9|{A}vWyNx7z+d=tUP6Ody6>cV*N
zpGd|(RfxI|LH)i-S0`!BKk?;lJ4v@|yY<322S0AQ=U|KdSk9#y&u|o<HX<iufh?>)
z@>STC{=~Ab1mQ_fX+9l&Wu4?7^P^o?v^Qr9((nVMV2)^O^6ziPNJ|>aWOwmaa}M^t
zx92|QC_k;0oz%_d2SLK@1mJd1)IvWvtpFXbK$C*4ho}f>nP8=(G`N=i!C^)#Jw|Pl
z?^@B<%F5@!ZT>yDWi#+^*p;sR(WI)fs<Judv8A1YX>3{87Ev)V8e<hOlPEQE4SrCY
zB~;Ne%xi>#E%<Zi%NbWvp#z+^ANfFiOH11P8ty0tfMHUldo2GuuQV~R8D!}GIWq{9
zKLe|xwBKk}*ubE(o=_c9QwH)nJ^KN>&9Z|3)`gL>9|y&4%@SI&9<(#~vz&X5c!SKJ
z3Wxx1%H+R`;h#Qnt<P2qo)k&AW}zH>{u%$cb_{&k5pgddZ@~it7y=R!ETYjCCA-Zk
zS6_rb^{h9Yj<t1S$8j|krEiYiN0B;`N7g-E3NpH>*IDQ+y4{phrLL~B^8SvVqr9nG
zK7tv#`a?f;Am%-A@ZgK}^|Uo^SG;Ik*|=&ll^Bh1hmx{gpcfTNUuo8+ak-k}omkn}
zta08&Y9^PWm!(N&dz0fEg7gQodt_$Sy;d-nZ&BaK<N}q_eq2)W2p|VauXFtTd(kLT
z-|}1Q(mE4UG)*miNQGsRsPXIox4PKQp&5s_LboVE|DQw~&Hav?1CZhB=2~;$yTF`s
zCz=UtM?2#?>RyMw_FKz-`fz>uA->{{$;n9+gh`F`x^PLvu7o*FGnND1P^0Oo@+NdC
z#JG=fRZ_E=!-DdP2~jE5C8aqHjn^MJZw&E^+f51n!oZVlJuKRq1ZlCk;E)iVp`Rhx
zf6%P&KYDbd*&I9955R9wl3EH^UF10hTqepLUL{xdAcEsoc`LEh-c~R@KR>X`Lrzee
zjZ{K<e`F;10Zv?9n%b(R`4AmV4pk>ge;L*p#)Ivj^r)|!-Z1K~<8?oH`nK==y}F9K
zA$~Y-Y<vtA(KL{alJfHHNC!Lu0-zkW<F<bneLSY5e<lj$6VL=%b{Z1|R4qf|?(O90
z{!+?~bay91z8^$5DbYm9V1%=6?cY(iwX5;>>8RAy=U2--W#r|>?-Wm+V~<8Szoe|@
zy0_;P??*Q8H+k09)=_bBTDRTs!0Ks9|K<!MQ;AZ}t3eOMl1V1I)9seUBn`5F^jTgx
zv~8gG-JtbsC`9Y>4YoVopo4>ya59W@AU@LN8;MKgEE*^rOpi|P{<1A1;+#A?O|S$t
zki>tT`AhvW9^148vgI->vQv#pV-*7lbh-lh)`OI!!;ZP!eaHs2K9dg*56f$4WU5c0
z53!1@{+S&MCF{y>_t1h|=M%9bYhSn<!|xIgGw$=eJ^8|go=Qur`JKL2^GSk~?p1n5
zBcSKEQjz&0-;GK!`22Jg#|V?zKS1=?JzxTM)%ez}3Z<-ozXgv^qscevnL90tMI<T-
z1hZ15=cEU_Izn!?gMh1?B691&#xrp-F<I;_oP2Bl{6w!bsQmx6y6&PqaXCnt`26{E
zz`&`~HLN?iICm^8EF7Src>;-43_fpqW+ow^FG*;<(b*e)4umha4PMAyTHTU4TKV(m
zo9pE`zlT_KenU1Mb8+D_qhO3tMfAhWuOv~qek#w!y)5KT`4{y+T(LUnb_$2pQX=R!
z8@HI)A-2~eNzTncIU7>0i+=j_NwaMmv1qoUTxSrYAly-#+yF4A#i@!MY-oCAojci<
z)K^P);SS-a+=u-reG+MQkNwR!4up!Hy+|cd<-29~geNG9Z+~^AO+}$#85sOebuZRB
zdpH@M=6~DT`p_Ionm~+sFniJIctp&4I|O5vKsrGdr~QiNrG!q;%*^bJslkgI6^tR(
z7p#@MJJ>EOs(Lj32tE|UbSv_}9@kBBKY_^?{;n{t5I|p{Zx$~@Z`$~9PeZ;$$dCS`
z#uW=&!Gd?c9^vHVeB&ta4<uE0c{%aJ@{Kpty#6S#Ntx4LrZb;}?#@n66FkdIQqTYQ
zXcz6Hj0dBjY1Lwr6%Y{EMM)V@yzIE5ZgG2Pu=LMSr&EtVuXDCMlVjZ_bNP`-Kw!W?
zgdu1^)Yr#Op4<=PN8KDNRtVZCkcz%X0s}?aD(A}EhzM%zE<{0mrW@>QD8}XJ<Ql%p
z)6?H#KT)klylZ~<gR+7>seg<oZ(iJK&0^Pk7+1h>0j+Zn=Su{9r`HqA)ce3Ym5_B1
z4zBkMYJC~jdG}};n<-82yswW>8E*96NyA*z?r$k==bwSci@{yKWja;RACuT9y%}!U
zNTqm*5O^i<irfM6O9Zm3?<g;6?B|wi3U<EQZD-d>Z`Yk7az80YK;YP?GY9GkA3VBt
zKf5ePCH$<ZO66mr>z+`G(%@&@_P2j6&GnWA)fC2eI8CmE4~x|upOo8o(rNMjN3{@b
zq7TpqH1~ZVo1~Xw-`}U)bMH&ZHqsg35Yc9u^IV(E_s1V{4T*}C6VubLO?8vCLpU?3
zckd?4@0@Uol#`Ql7z(|kn{9CvOs(X~Kp=5-G#)rp1c@F4KQ6u8vn~5X&7)t9%E~ZR
zXYQU0hR75eu7lKSs_y0wQ<*CVt$+SBI6O1mAL<H!c<ZtGYlh3UQ<hSX?{ce2NxfWS
zAqF*mwzk#dhsph|lBdp=tNA+`u1ZQf=BOlS7SQx;(ErmkvA1G@aF6E55n}rkgP>sz
zxNEqVF*6!r3@|#f>Z4?zI>wLf?+^o^@Wbp(BZS;l!4qC5B3|9iYBTaNq}esYVI}PB
zcFaSr^cAZ-yT&KXR)g1Fwfrx}2=NP_&+{2oyXw}KxIfi*tZ6G`IF*s1Uu4@bVfVtz
zixs`4=)r|BytT+#0-rVzs3zSTtYb(*yqND+lUO6Bmp>|WAWvpxWfeMhZ4ZaQaO2Cs
zzUmNzFeP3Ak>{2}zvR+>&-|)!tuI`XSsWg;S*|=Ky)j&@c)#^EMYq6r<4@l$vT2VT
zDNl<Gle^ZMBKG6E-(%mXsHjt9T}GEV>`qvCA_;mgVL$n`+j*{5)c4@2Z|3x3kN&+)
zcYOQ;gHas$(?$eOp}Sg>r+I&aGPzDPo$9e3$EcUksfOqo>&vE$e@F_@2bXy$Mcd{F
zYg}ioH%rTOScs9`=&3<#F{QWwZT{<B!ONfZ9xVK{{EdqAov@8V?_(hy{kwO$d4k9M
zIGs~JB2oE_18ckOvUhKJ-1{`=tEP#)`VABzk;Zs+Vvnp2mno<jSAPwUwI6+{mvS*i
zyPcPZx7y9httaZ(?YdPrS$URWJw<L%vggyb{Iz18T7Laa{(32Dw-u+%S~YjkWckEN
z%tFK}8)xZRGMPi2UI~2OVT84AI432G7%t`X#%VHFp~Kw*6VXq>RR>oc`?VsZBr(57
z=RsQEl_?rI<*qz*H+uJW8S%4=PIe>pvg#BzS>q<MvSd|?UICxNsx-?|(PG*9JL~m*
zULKwo(2L<E^N+VQubn;Nck}~Pand#iLf*WI*5r}0SRq%iy_~DxrHuC)|86|hv-#rj
zJ9YHAgX}YV`$;1o!dg;{*Z*GS3Rb%5S9`8k^fHak?KIX5MN^n^)XcJIcYYo@WhQej
zSB2AwM`lfF$61~0E*63<FJVFdPUOK)5bKt!L`=om!*r&yU%*q;da^@|hK8p6=*Vr6
z`+sHh_20*)kPI(^Fy-jL!j)s?AtiBNcK+C5s&nnF{)MH<P6M7IN-8SF%K_i-4Pnk8
z;E{)qbzKV${Z%+)kZ$Yqr$&ZptE%p$_~HLMPMI+>_l2N-PLJ?RFh+ENM{zn8l;bg%
z)&F}Y(8EINYPPS_O3R(o++wc2uC?zJBfF&P?`GlQ5??l#e?JptWaRnXF!NDAe=*Bx
zYwY+^+HGGV=d*0Q^Ey0TH^z-ZB_9i+1;@^1NHwHPuUhBo)z{T*^^`6V*rrOH-)8G(
zFzN;aB`$h*(d1C}KH_>o!9*UgC!rnUgkO>8>X2n_CLOK@X6p6gMHDVv+ljVdaJcPx
z1>hn5zVDt1_|!g>B9Svy#GD#QHoEY4;t{D!ptoxEacQWz3Rf@&TfMB)uO#&mr?og$
zOq@J*>VSt_gYKRa7PLlp2>=WiDt8Mr3CpuESJr#(?3xX)y?vZJV|VdmyYKwn3x&le
zI0L!K6e<@22;(Q>t3N2k<@G<C0Y2I@k=>0JTj_t@#Vo-}6=-i5=`JqTsc6BYJzKUW
zV-(Og2G*Y4sb+D8=~|vGw`JkNo7nMstT`ZoC(x9NljmPZqnu<gS)uuo=G<dD&RM8=
z!8(^DH%P~$Ms9?mIT$^^<-?J#zouNboE(=XRngc5xX5<0Bh+YPOos>#Rv+g$rs0J)
z?_SxATkQg99-W-+|1<Onod*gD>Q^6}++OZcfSEA}aXNfPkhs0A9lCPreU8V^HpU8_
z!n8XB)?(q}U;6vvc;EA#wfk+am-5{ASNP46D8j!J@yqI5boP8j1d==sK@Bn1gNjiI
zG;x~XIvqXz?reERlHu^?=2CI@Vm=U`-x;O~Q2TPNII2MW^zrwoA*~b->ADe`p{Ju`
zzs8tp)V3sy!#X@ZJ#7W)eKm&J)q-I3^6@dP3g94Fjh(UVdY|3x66Q+W<nZ^eyRrD5
z;{DlcBXqnGj7Ab2Y(<wL=>f@#wL&Z7E7xOn=FYGguE8&wSe!$Z08}T#=~(&Zgjr+k
z?&juZo>L$5D9k6Q!fJWE$c2?&4CivwKHiF{vskgHsr>M{{?)79a>Qo-PY`p<q_xc9
z%{mJI{E$w}b;;;M^sWgLRht#?%p&=GgsLpQI@nK<DwA`R^8K1-TXx9Mn_zTr&W*-L
zZbmPwCpAkhG-;H7|K3>ax)s);jj<bOqC##Q0um+DdF@e=LBXCZJp@F~Gwg5w+)Wv2
zjydc+S;x*F$8j&K?SY;6qjh$GW91bUx$A#nbO#cXuw(C$O5CucO&n$B`;s9l$=#3y
zAkKE`)a%(<>$IGs80d=onF~rz!Y^-=3aP&8ds&JfuUymk9@sVLVz@Rzb>@ZNYMtA=
zt{u5@T{O$g%XGTT$2RD!W84z_mw5yQhXH9xPM2&mOo6jCu$SlcS4?Uig)BqLdIR_-
zab++>`l>(NfC65<iV}BOU*1;YAAN_y-p=R+=PrjvG8N81_9deG@fjOg+0qBas5Zmq
zBycRCU9woScPxxRFEeTl67)`LJITv?cU0ohGzL1&j?|mz=jwg;-Es;`mD;)ANzJ@3
zu>Do5{Vu_+N!33O{@S*8NH5$f%~pO4|GCcN|NM^13)pwGLHgQvX_KtD_ywRd)%eTd
zNYUoJ2dR#+vdZrGLND*VRvr>h^Qe3GOPapXrVmHkzfR?i&b{5S_heY@)wBBhinoz-
z$R=q#Xq{LT_emQYyQiJ3OG*bi@}?J*D{l&snjWe@_?(3>?d9R_yrUu*Fo_oENSYjt
zk1DUx3slVGsD<d;W5M#Kw&b<tg@mAwtY^f;Lcxx7|B;%;%rfgD3$X3s-<o;oD8Wx0
zI(>W3RMDCMsmCWe`U#4QlMuLl{v2^3X<x6AP0^6M?D3rVTM9Rr;v0*qHzYUyQSs@|
zN?jNpkFc*aJ2TS|Z#3OuQ*5p51n#o_Yl{4~4yUQY_hQcP%sF(%McT94UR4o?<pcx;
z(W>@4sW0j7?~lJK_P9HhhiJ+5QAI_6rU}(wq@AIES#?vrS=zMu-a>v!@wT4Fhizgm
z6$MHfZ5^V0X?BJwE&oUtR}miGoFoBR>a_3Xj~pHts0UjWnOF_pJ6)I0fGlg|n)`mE
zC##Lr3Kk5Co^&na3+K<DzoMx*PwQ?x7-m^SCKLaA2E#$~2Z+!jp&wV4H1!iSsW;?%
zSM~L4KHsWd?lezp5aM|Gej)6Tlm#XC=+c4UdT)jYLJt3Cyw*+s-1j>owP~a?axQd1
zur1qiN?X}<qFg~BgqDd$>6ibxrqLye6MnA)0>(}=&SLzkL!Z|P3<Wi;-WwDU@I+rg
zjlRsgIw92=Q$snl(nB_f#W+<W9tCRY$N!FgB|)<TlVnf9L22G`L-!b?!r4cz9A6hb
zs%|C*DrX@OAVFQ8HUVHWzTh47<_%?a2vvm+Al7v`qVWB{ufUmdxg8*#fv11x=>L96
zqZd{8wRJe2mFg*)dJ)pAH+W95ZvR84zdsn>M@4n~^yxqlV-c>Kk4rKH(Lo-GbDUxN
zgBl|8Z{94MZkT%TkBpK!TC4i@&90VUblEn3wPR|3^sg^%ig?~)lG8t{^KXV)y2<a`
z&;1|w`MCy0hA5z3Cyw1nHIQ0jKzrkM3W}`cPoMT5_c0gKSc@)*CdY_Y+oS~%|2-}*
zT$Yy3r?qs>KQ=SlHAo+fM&wAfa~nbU3PM&rB^JTK!8;#_T-WIdU|9^*c+lKDQEw&1
zW{$SGs|lKhDd$rLl@l$Ni@u|-!N#gpjWj593`S{#35>lq3#7@3p!T<}8yZUM%%?^b
z?WzvhZ-hjlYDfEW!gAj0!v@sXv+fPf1S<tAU2t?1$4?Gp)Z%Tp|DXpG+J(w<TPtec
zrNknbs0!9f@7!E;e*0YU!NKXugKWoRmPcazA%h=++So!c-qbF`;=s|p!So2Tt4$U9
z8XEmtGr^$$l;UWr2UCZeUM_Wb_%t^1%gXx24v9uL-`ll!@1T!h5#h*z>-Lbr-@ku9
zKq*d~ayiDMiMwcwX&wqHN}r0ptPaC-aRmkXx<&e2w_P1wlkpZ&7u6*>t&LTqD;<75
ziWKM1Fvu1jcq7PK1Yeq8QgTIXMZOo0`!EU?Nw4g_0UkVn+gx3p2AzX&-Dp~AMR#9o
zwj~SFpj%LC@WHib-?{buNXT#nh={zwU3Thb<*P84VM;Vz*EW=S7T!38?;96eemXZ4
z)&vEsC)HF`$}w5p$E3o!PZ>G;YK<68UtAdDh~8c9L3%B|>tFwU-0(y<)d@jCe^?7V
zgK&{(1ha@^#=+;j(_o*_H0rtIDH-M!TU&o-FPFP@ivt4NzF#(ESq;>d#Kq<4Y7>5-
zpioj(XFq5cIr#R3qwbZoHMujgCDB8xW9DFB#KV8Zm$!fb<`%U7mzl1YvWG=lzj4qX
z9n(mI3YPnP_zd!bz%{8G8vcU*X1Skb>k=}cR+y4ORbY$}$pTXL$X|L*O!T?Lhi4oS
zxx>MTxa~J00kXy?&C2Whq-SnrwsJ|@={0(UuO@BXLT&*srmk57TiY|F$cc|iL@mVP
zerMw<v02#TVoLjm>_zjgR7tzt2icZy&Db3;Lb`&UaegXlY8SovRGAlovIch_8MtAp
zHuuUmoZBEYk&1VBk66FO-ler$<`xh4LnsXa!1YeBX!!}!E~MNYLSzj)DNKYyB@KU|
z%-GrU#+ZAu<fhPLE#loQJy@X2Y&%tu6U26GUsQB-KpF$zoBOiEzl<s~Xg$_P)m%5z
zTB;D@-a^i-lGft7i`Zk?H~#2D#nxez!~qA4ghG8Ghl(VAy!5&)wmA#)BAx@?cpWjB
z?`{G$f^Wr*s{J-*q!^pm_x{*L1%<U|%ZXV;;Fp=`Z6M&f2bk*JZL6fE6?}a*0zJ+)
zn;l|lP27BZzEECfNB(GcDkAjt_upC`c44-hyD;ICHEzS5`$*NAqUM}!5Tomw_}q~5
zf{zaYUxhSp5E<9uLc^^iez7S<M~d|Jo_8y21gWkY&w}~dX%~4m?sjWxYg5h)CMpi8
z@fiUEi^m(x{<m$81IP)c!@1Lp@Hy(@Qj9X}O9D0?Ab1~djWf^-4~k0Nx4~?C$&HMY
zH!JrSkJ7<jWq2Ugel<OZI1{Zih8{wa&PR%^sG!h&Z8aj8`i8?pu{k#;LayAejx+|P
zM5~ZAy3-DSX-Pw)V?TD~E=HhJ5XX*_i|bRd3TOLuUSs0^JGRW2kbNh+3@SlWkuZ+H
z>q*hkaxVl$z0x3BY(QBNnbv}nJcufmyP=GnwP?Ft<LN9xtdnz0NkhYXbhV*tc$**n
zPe7e5d5$ZR|BbvGVVG%-eQ27>U^Me?8AY0o%!@QAj&2HFGvA*2@!)b-hRoGkiPPsw
z?`0`P8()m^@?<&uBep7VJZtHT>(=J|*3XazR%*tnKc>61Si1Qzkv~h%;Lc2z>y``F
zuD~L9Q!SFIv(NolIU^kdLqJf_z`j1J@XMkq^Xa7h8(#4uL6iCaX#ql%;%0YOlcqQE
zA(`R?MPik6<np^SRhHthu+K@+9)I)V84`W{+<~LjA=*iBU;I8XJhK=fX!Xwy&<uaY
zbRWa!_P+#gALZqq{<=cLWBHxGG&CaOFI3YRJ|1j#WQj+q2w$StK|xJ2F%1LTI*G}a
zH(bQETRh++uw&1`V_x2vkAqD}Y~#N?)~8H)m4E6DU@emjT{gn&KpD~|upq^%7BRkb
z(TW!AuVMX}udVk<Nv6whR`kSM_=vZDzNP#tQfgdS*4i|k&Ysmn$_RSzPnwAbak4x*
zOFRi$JGP}2ZMwcMwrWVCaNI~KzG~Y`nR&ajaKEp=KNOx)@VYfyaC8=!&e^x{a*p*z
zhVBcmgm!*3BC_|F8ViMvy^mM18tzMqiYgK0GXk^$+N|2WH#&quj<|b_y(ma2PHlU~
zCwe3FWZO`Wpi59bPJKny6KH$R19Fl$%==e(A@^8N{0Q7W(N7$cm)ADYSKa<54ZT_e
z){oyFH0=03Wp(}>rlFn|GtX<G6tXOp8*(muTj3Lu^Dhrk2ye*q!10Tf3yvH5J*q~u
zMi3U-xqL*+yW&M+b5ql<nL*AO^x~6J*w7T};&SMoa07(%IbI{J<F9?nMhA>A`TN^9
zxnXuhL@$?S(gNbP&mH{hSjZe!`@Hr3qC+8K5~4b6v$*kNwYRC{dcF!%_UKU0+%j>V
zWN@kcHRHf){bz9YTXk9{afLF(Ka$ep38_pcWo9D_tKDS&D~)rwX}W^+S1-SMH>kY5
z-3sLaAKyTP)8x$IkYAq6j=N$S{!;E=IZ9f;$@>wzs9U6r9!f}}89|kH0!vbLKhM4@
z`E&G>o*v`uqecmujwyx6ep`!Oi`!4As#{5X(dFxt{b%^5XHSud@%ryN4TA^sf2%Te
z(oAcX6ZrJLyai5~J0U4pye<Sit4xCCBaE@pigc4jEA|5_H94Z90+BRd_SyEWRyKe<
zq?ru^GL{X$NrxGw>R3NpA3yj8o4{A-i9e0n1Kc;sk-W=kO6pIlf%3zL8ETI%J${_4
zTprSV@4|;G9pVz+ULOV$?b|Tzq;!3{Y@dacrzW2fXP_eMbH?nsBUdP{T)SFjetL7@
zbzthUiLlz`P5SCL&T4>LI$esuh3oM#RO=bh7%e<(4Gd`i&|o$nN}*`f^Zj9Nbo4j=
z24Z0qIxcRBhg*1$+$Du{2?nFvnXM<*+w0zoJ*~QFT?AncjxMR4AFv0m`#lsC6a_Qi
z55qL)350m7^P^={f(0d@{4rc+?oxSSKD|-uE=!nLsc338@kfgKnpb8e=@hDqhF>18
zsU528Q%-9+ZnNC9t3_UP-|mtrvOUes#M##S)GT5{e`p0ZHg|GLoem$Y)*%5B**f*N
zGn)m@K3H_$%;6P%#GQL|wed-;;FW9FM)frT|IbUzEMQq;u5dB$m*Ph?KkZ{<BD2r0
zP@|~FwKG<cR>ZPP1w1?sQGa&vM#`K7z*UK<f-<nA*Pwvo;NXC$#)r|+!5zs+I&?nJ
zfy+a~j6S<wOxhk6bX}o3w?4h?!i5U~38oh=cql)c2AE)Zc<dV951rl5K<xI1Xjdy{
zgpq$fO{VD5oU|2PiJ8#Mthr_G#CfU^V%_QAE*K=fehJl}q<Sv2NkEpo-OY>Gk>cGh
z(k3caOyD^cKk^m%Rop1G$nK^)SbbPFvO_UU<N>Bt=pDN-pMz*ydQ^&QJnvLn7}861
zEHJ173mQ6z=JPnlOz@wBFdZcAP9x_9dfe3LObHUR8LBUKbxqaJ{gyvfq>L^VFQi4d
z9aX|aUb=41MUo75pO*H_tm6KrJ#gS*%a4S_#Jq`tQr4n^%wcnEyZhZmohh=SvFXE9
z?=5t^ZoGXcq1_{xMLX0iHqOl@IWZeZTv%ZHShy^K^8q`SydON-CS^O4wa5L77F81J
zBAD^k+U|>F$F1)=6Qsn!^peDd6QHK81n;w+8Ue28P4(dHiSzzc&z5P{Mc*gQ7JbP;
zb2gn%g`AcaJ#6;2A=Y<1Y?99+eSkQmi2Z<_Bp}-8gY&}?Y2sTQBj;e`LY`>R#oTi_
z5KU%qhwztLJZM?8!rAS-bQG<^76w_o+}yb-`B6zP%nvFkX4rRNR3JaN2U1X-IC4Al
z8co8^)O<b;FO%=bYx*5W9{W^$3v+e+ack$-_KBM1{_d5DsV#qDb8EBCyetLYuh=iZ
z%l@p$u$Z|UTW#0*<i8w%V^(`RQ03aXJ0dzT;v)pGI~AbDa^PcN)I5{%gZVf+`_1H$
z2mc*78l$2i!vlv8S1)zD*1_)Z{S&_9$NN+L4#9&0HAtwmvp7=M+qhAJkYPc#K?1z&
z?1bSqM(QMZ=HUTzID0@C%Rnq!zuXP9(NY`hOS&pY-lCN1m@?Hdv1pdv&e1F9BvpE!
z_Ee3-Go~H`&t>TIvz5>4*lwoo(=#+IJ$^ib{Z1Lfqsfj7O5S&*yaBKL{d|iG$;cKV
zt9-5G?n0y48N18re8Mxgvh+Nbty$}&*Z8Oy_y{tbfejGCKukSkPyXBN3@3q2_I;1v
z6>rX6m!4R>-@itKuGgB#bKNUZk!S7(v_F%BDiZaIFO%@kmB)bXfkNRz&#>&4-rSf%
z?HH-1sY!=hzHrnCZ+^U~2r^Wgq{E-o6Uy(M!~^j$P$|W2?GtOEeZl#{<h$wmHih}J
z^FGZ#|L(5)Ea0SAxbM;Y-`&iIX4sdOMytOqbuD=dFH{Q_v!l)hKF$7K%IQ&l#f?$I
zCA&;t4Pt_3qT|kvS3l*Vn(x_#wfBM>{tF5PMXTtBvXT-TH+Pj_=TQxHbuKP0t0GIZ
zqvL4Z#JO<MPzb-QfI0#L3UZ&i<=(i&00JlH=w+XH{mJX!>qzkg?To`~5C=KUulM%$
zqFzAP+vqxJI0HI`;<>J;8<V9K%u)i9J;uQ?$Z&PhCXiO=f<(|~E^aQaV`5^h{69jx
z2YIEXnRE4-r&4-2$zaa5TNeprYp#=^P0{L`4AZfQu;n%SRu#wF;;yT|BSTM`JkjAX
z&2C@mdAe^Iu@Uti*deLHoE!&_P79V}b~79y)#LOZ4V0XU_!xZJf6ii*nH9bd(0P1n
z4>0Xy-3%bX<iFmF$=|#gtKqw$`5x0Smv374LHcGh(I$LHhSc`<C(UTe{%Jjsmz&5x
z!t}0QWi1aO#i<7m-<c9?XK$Th3NJkgD~QXeFAEJhLsc$f*P6kE3<}P|yK<x!4=h4~
z!zq&fJN|dWN7ci)kw?J~<Ne2h$HA6gGdS3Fc(gV11e(L6&@AQRY$I$|PYA!vsH3ID
zagnL+^|l&vP93d74vQ_tA6$#O%FpDS=Jf@8a(HXWesiNaoA3H!p@Ll1))B-DxdhEd
z;6(%o^DjZ7#tejUYmbB8&%SEhGs;8??{*V2GcF*a@c==xiFZ))@$f`PMY&-)us+uc
z9kb?~;c~pxVN<92<ljYi&gMIw`T{37I6NS+{Ed!eJirP-fprT46B>^Ix=s0-KD?e>
zu$vUy3YKSuMbXUli_DKm(sKm}n4>`g8u{#hiV~o>angs>Tvr2s$5ok~xslnr-ZtBe
zOs`mfW}W+g2`Jb0h@lHTtWmEU%=!}IBkxoMq`Ax}9JbCrIhY$_783Cb0M9r*PiHWC
zX99tTh;?V*Yxag#0(g-`E*gXQrax*{=)hMeydtF2;>S%P#qWc^?u$5G*!6SJnn~0`
z1<@7DJ^X;;UP?vsVmHFgdpXtk-<^NLPXR^525?%q){O^6Ceu%6L{j13&RlsvT<aPG
zDI(BeZmg5Uw#}>rRU~^Q!u<sr9xJZ~$AOT%mwZfPmGj*9h*;x^!^Mh$oNsGy*W8{{
z_3{K~r~<`vMe{Ya*uF;x;qR`(NkuQ!_weUK-RIixYH$Cs?5K&3t3I??8P&e|uqw;E
zV7^^cH{XPo#bx;b)M@jM6j00jf;{XRG85IvJ2!8h1e6b@@xt>`<Fh;0l@t4tQV-Zd
z%Gz)*D<HwQ_5PcIF5W^zHMG1!?2)~(iJL^!D-&1+DQ4pta$!ynfG{sb!W7uV^=(pQ
zlCzb^3s#20NI*K`?g6BDCMF-)C3W<xaA*c0!ruIi3>x_--gU<lwY;M)ph0z0{3Nxc
z9Otdpwv~mGU`9wyextsCMe^Tb*K;IvZ0F9ki3{=S2^KVAn@XrmMZ#2{DbG=#BV@YI
z?2s3xjo*A{7~1LOT_EP#RX{=Ycr;iP4HSEzu~6pxTQ;6n{40oC!-DrBx^ilO7yL*=
z?(E96=lPt>*QWB2%kfEbr!OdhC+MqYj5+$fqT&GjAP`k8S5g6}Lx3)pauaDCQi0zq
z(uk7acyd$jqobyGp~dJ9Y9SaI!qZg~?8U}qq7-T5%3|xCeQ17t@tRasmz*naK>`?l
z3x#iDok&sbZIVunL;)>OR<A4nC~}yIuJ#|NoGe^2G<05Q2Cjmu_GQWS^?lQX{YC~8
z`;C>0S7-tgo~IQVmgln&y|v@-tjyA*FHly<!w{`EkR%Oe;>Kw$CJ}dtxq*cni|xF)
zrNqhC;@FE)PaI6kTn<qG8Hhx+m`XSelkL&d$FKJ#XJTp6kS?(5=6dfwS7Ri1q-R0w
z0zXp@RDC@$Jl}ToN%OtEKMqWMy|ky|Nu>P13gxuRQm~W}<ij_yD=)kER-4>|C#A1p
zn!S$<LdMS6P|19x_b8OEY7Ru)2?*+-et9v5+&#&QB7*!mVTmjyP0e~<X3zA=={v@U
zT5JyeX?lCD@V1niio=^&?eMp23-$#|n>(Rq(@_5g!xvvjkY-!&L~cay4@LhLt|?8F
zJr86|xxLa>eI?YrZFgO|w-p*c{j7Z$nWPF}GGI=yYI_MtXe&&&YH)tMMU7hWIjO$w
zIkEhH&>T9PU!|&X>d5(tUuPf)CC!lhQJ18hl%dOKnBs&S7h7O%Jd9Ke?@-?@-=oCi
zrsg)~rr$%bqE`5{qFkchBPz2`hCi_C2-*FcLTB>!qtAsTSw^N(JixQqs4Un3E=j^~
zySVuuJ`Cv2(n}FhBX0@SAU2=S^=j&%-e{QRptvb?$X{}H7s=ZLhAD{rE4QQ+UNS*Q
zxO}sr|H;;KD+#*3>Jyau<JO;kkHShh|7P-$1rtcrvlpt<D$LGUBfAFmIIxb6mmCSD
z#l)8Jt4YlWCbJB#tI#I<dc;poD{|XUrrB_Syt#?u2d+eIX_o2uOJu$@R6$<7-)|js
z+j~}@5-0;0ZG0!7CrQoTN^!6TdiCm+4Dzed<|IZ^E*i^`^9J^x+Nt_gkQ`1ei-OLB
znO!5wscHN@^T|+^TC%|G3LS9XYCdV$aRb)5b#cZs^jHIA&$I0>JPAnh<MZ<JIHG3y
zq_|O+9J@6$NM+%Pw>AuE(dC&1V9cw@)8r&7!^mZY18M4zDtXdkWb1u^uYYpg#RUZX
zP@8wWCQ)~A7r7x4X34wDx4GF~^LzQgM{J$4LAxm!Fr#^EsOm(d?Mg-*d*yL9Hl&$K
zLRKn!L<KnVwTYp%x3$v@!L0NF9102!25v^eKAkzORFPT(ZydrGNu6`yetA(IuNF6-
zC2^=GbV55RYg<H}U@ZB!Qj=D*j+}>uZ8AJgMetu!bJDsLe)zaI8iTFJnn-%hjYK~Z
zMEmjMPb_`=PF-$C18l=6hDb``bBLw!>2Hyfs8OW>Mg+g20;+-})`5DHGzbJYh8RUi
ze!wlr+0>n-WETV*@(jeOZoBr*3PC(4wiLH=8do2LseFoigaz(sO;90fKYlE>Pip>^
znl?azS!i7v(T?f3lNoYMQ{);zi?ewJ1Pt$Hx3kR8#~H54rD(4ew0|fy-j%aq*cC0(
z_o3KJ&<)KZRS;Od1HwVNoR<`Hjzvu$st#$rKZKb2NI%(H3~a#C=yT-Z5|FlF<8(QW
z^8<1a!wiRC&P$yN$WS{u3rb9c&xB6aP_doi3xbo^!}zAa@xc1;(H#1}WP7+|&~*Sy
z;0!hGZx-37(XPhB%NvR|P=ljL>>^+U15CG(Wx1nF>#kMIa-#W@TM1r*34=%G?yL)s
z>^Lhoj^b$Tf*`3lT8>>Nd1dD09JFFTaamDu2Hy=Szx}vzHf|7pj_fj6x4Jo;YRd5_
zZeFy<3x(^%bS|Yzq;xn|=rdX;o5u3%l<E_;c%c<)7+;`uKj$D%tQxeAj)@s35)yKc
z0!I~_8BCKEMcH!U%9U3Pmk#__Sz9R=J4vi|3SRy1oD#*eb@-7`-6(M*BXjD7ozz>L
zk35dIo<UHmC?sf#(8<4adQOo#Tz@L87Oh)l@swNN4_l@`n$QEI*{klOcJbm46cFIG
zi>XznzbumzyJk7EDt$IpfSec4B6F>#r>t=JAAFK^ww7h`a;k$UsR+8?DpZnj?goCT
zbms_$U*k~D{xx+EBaa1_ft(`pi*^&qu*g>2fREh%%C-VBzls=y>kRW2fk>`pl>Vr*
zs38_xP`b3POd4Xh)aA4f2(wsyHv~~PGL}6s64@w%m7l-%&`dqcA!*`Sh^9=>tC8M8
z$JxgZ3qB>hzR!xZeT2>vKWkYiVx#PZ(DtEa9GKxl3Zc+{eudP{O#BQ!t^9rZDcYv?
znAL(?A|@v0Di_`BT^$vC#x@hr9>e83AiJ*^zhiyE($y4<_5x*<OT0>iWGJH%@hSry
z-4}ARDef7h(&NHk6}hWErxPFhA+Fsimv@at{r6VoqeH5~Hu_!oW^XNt#!A6$9foMa
zl$@(ME!{nWi<5)9%86_mdo0I#rSa%S3Wh~q^Lc}Id({s91v<d}eCNMBYfs;%aqq`x
zfYWn;NmK(UQoyg{%b|lTZ0}(a2lmocv;?DdGFC3F#ggougB+1N_@|@CUD1Pk0@~&{
z%0q^y@3fHWbKEooC1ug^BpA+u#z-A<C<tQkI|$VFZxomxmk%G_2npcL_kR2KZF?#w
zCp)`U*7DDzEb(`J7*Gf1y41g}o;o`gE$Abd)Xei}GCzdala>`YB-%|31kR{PKMlkR
zGLu8$mw!OsAjPEZ1oJsGfoUZD9-X!79?RDt!cbwgpyz;vXGBHhQwsu`n%<#_CE<Tx
zs$^#e^>A&NXYFly&Onl{K(WIyi>`uDw4&y1NFp86FL0Q7Q+L$D(|&UDL`ZdKpZ^vT
zdhu-uIW#yPf0|&@mlKxlMw?~YN~<Qx8v2AZ@B$1fL^r?i(t2msuj0nwP=50YSHs5!
zGit(QCG$Tzqt~w&<~|4OmFzWAte+6<tHxK`bL8y)%+`+AhuKu9w#yMI1Ol>&&3Q+5
zP*9K<>NVoikf5L|`uhJ4REIPnncs`ZJvT0nHm0i@8pjz;liN>`Ym^=cx&t!wS4P=&
zC`=3;dMRAy!n$E_T?^lYLT!fs1$Kn5mqAK{)S~+SmuIxP3zOBUgav~?Z}(oFy;F^w
z><A>5TI6iQr25VM3%?NxI+A~an<<NIq_i`!s1%O1mw_?8s;=N?+{H!b-L@fU5-r?u
znR?p4&1?O1ouoW8R63NOalXavCu>mjw!+jD_sNoDKayok7f=Qde;qQF7@vJCkNeMO
z7Cli=vhe&iJ1B;lNz8iS#dY$NiXJE;s!)8o@H<{7rIvSBgu-eAc=acKW?_~#Zao)P
zU27W~MOD>>1R+m_i((hj1+VC2E7{Z1(2yUqc7FVvF?Z5M*TA6G)L&|-!JO0*1&aYU
zQi3&>l=d01Bl}rEnjpFtAoNE=%PW@>Mghq$ze9xB*uD){gy4L=!TCZnz`95uYe~9E
z8DW(m)X)zQ_t`Re?46a*t@nuw%{p#nV}q(8M|D1<#)rfEh{mm=;w^c=#FhMPRY#qw
zfZxY{d+F)nLB{Ug`~A2knJHOP#SIva_#^n_ZNvglRC#~)HCjMVS@p<}+PzL(4<I#J
zz=wsFj&7c=C$Xj}FOpBBsj#1&;a=Bor#bSYuH8$)aJo@Z@nr&|g$2?v?$^{;uYMUA
z7+41V6G5S=2qqRQMAP>BBGgCDmY^<iPVCT2*^$(|z?OiL>xh(N)8G%S1kHfY1BUx#
z>bJYc|J|{k`$%pJtCKe<z73-89{b|8qa-&`s0lz$&HT%(i&jry@03BI%BP!7K%{gs
zo;6^;O!jFK;AK@Cd-PF9W>;!aqiQ4*0>4LXf1kHRsHD8lZoZOeC)RW?OC@}a<$&JY
z+#L2HE8NTFqZO6y?6PjQ$3VvT-**xlisdXy-FfT<W7$iEP_lms;g|M-rjDahL32<6
zx=~1oGg!6HV5V>jr#MpWvpk8@+vU3D4537PiCg~Jpx@_XHF?G#oC<jNUjVP@#_5+_
z{A<b;{IfJVn-r2}C@xjc;=<Pj>d=$UpvZlHzO!d^;3u~KAkIrR{R#CEOC%WB>d>6~
zD*GpC@^Ekz!7hTVZ@+`%{lipmABY^j^Xq-@Z?$uUN!wtG`kKU1qMkTuTtc9uq<~Mt
zz+BzA6VhFxSqngu97f|pZpzBaR=ZK#^A_PB=snnIY=b_HwP?L_wi)uNc|RXXRT9FR
zR&pW0!V~h0IIv|_x(ZMX4luqf#14bM*ignC8on@z!mfIl1X*fn-^kJ{?Be7J{yDf-
z6w4-He)n%WE5f<<9#Q7u{Vg#hH8N?>tAl$PR$K-|S|Ptr>O9~#z%4!D_FcUg?Xo`O
z?;4BdSLyCca1s*#zm*4nQ=QZz=Zu}`8^PX32XnT|mCL&!KK|6=X+9nO1M7%L<@?CN
z)B^MXU-8}hY8*P76=4PfCnVXP%zkc^pgnzW@1y)$egR*kHy=T94fZK3wb-DvMTu~6
zkEmhq-l!u-Qkx+?{NuV|3i==RQAWfW3KC+AqgfL!5rRN)Ra3JZ5mlDtvKm<4u2u6+
z*e62D>7K24Cs|Z#FXfaTvGu-4&25_pn3DV=ZB=i2`@-@4Y|Ms?Mkt@I{`XCH&kVl#
zWV6r<`w@~Af|YK6EYqpvR5azIV_0tZSK}_**k-eYQ=P5S=ScAKkITvWxs8K^Ik=Ns
zv&=(`eA3|Qpvj{Zd(fgl?$mUA+wKKV78eJN8&kk>GLTV`LN3Y07D!HKo?SosMC~sw
z6Nkg?QKNq`%Kw1m>J8sDz@GO$U5pXi)pUVE?C?d-z~-dz*rB02LZQ!=XUW~ur;?B!
z6MA3l%K0zS1b8pQRESJFi5{pKI_E*VE%-d|f_zQwgpQw=`N(YHwH_<shj#4Z+}?s7
zatFLit{3clPvPm|3T+oui=OD=)%#*rs-Y%z3z)#t<tcP1k&403Xx7U@N1;!rMSMZQ
zdl;bDpr5JOlKuZ*6f;V3);dRrDD-q*2y*AHgsN~>>GqO)$-Hpj3aoM%c&&hXhZr#O
z5csvbm(6c~y2UL5lB%6<VAg=X_lfQeN3_>_H#M2@(z4?%Nfx-CL+SBtuqC32+=5ig
zWVc`Vu03V4-7r_6=pVozzH695_8nd4Es04<5^q96%rW?x?f*3O-tk!X{rk94RwXOS
zrbs192_+<>NTRaJC?nZqXH_IbJ4GmxBqOqCAxR~~sbpto@2v0f>b^d|$M3rTxvwjo
z&i8md$MHOl<2kl*%&}wF7)wU0L+(i<Bcf_NJHo$16-pW?(L{uW<Fk*6)|8aT<yTCI
zBtHM3?Bf-z#85yz1|*!FoBQwme_xSWqU5>R{C=0;422$}?j`<U>n*(?CY-~UsvkH$
zxA&$87}(EliOd|1C;eJ4<x#a)q?eMTV7)`GI`zRTxlVxzrLC2^%<>F>FeoN=4xneu
zF6a;vLp8q0mMzaE$1II2xK2opizFTircz7L$>Z}r>N8*1B826O#h~^%2Q;q{|A&r_
z>hf40es*FPHdP@<`0veZ)PQk`_NZyt&4{kSMz%OsQ{7A-q=$3+q8~66h42}yzu(j;
zlUPPuR>HHo8KziEC2wHibQGH`yo>#K$9iKzB@0_6elf3}CYGeYAs2ir4uj)JyIx+f
z>JIZ^<Wa%0j3cT(;1ou)?a+?;tvK=xjVzX&1%5nI`{69asdGO<;^S7^-g`S23PLmN
zOqu7LQx<lLC@FRT81Dgk=Q>ayg!*v7w{nk^)Ld`0<F}#dQrg%Oc0!B^Y<c#y?R%ui
z{n7v90;tK$`~H2~PH{ud#$N0TAtqd1nk=42K1^U^fM{O;4=1PS<Cg(xW*E(ZBXW7M
zh=hulFZ*-aSFff&X}=4@XKF-CEV~vwwpLEz=<aXyvUYC_(B3k^X0a1+ouBDA*v(Rq
zVj}@41{@kexq0aTM;&vq`qMuPWHpXj!fr;EcaKmT<J0(TXzf8Ggi+!j{qR4C`5wy0
zu5HxNOzc$ax8Y(()8(|Kh_0*Rc6ZiUx)J&aLvc<p`;VYaiIg;*SqRlex{fsQK1zym
zj{e`w<08_zhZE|=#7#|zBah&W@4&P-Sk4mkDqlguT6JZ6Nzs70z0~8mnHI;G(W-Es
zKK<|_%Now3Tl7oryzCeIRP$*_(E7W9xK(6W6lTHsi-$<I+2BOW+jLg&C@%U-<q*xE
z=apDhJwHhp=3QOqOVj5F;D<>(2cABAZtqr?s2i{KY>UO!aPhB`lUciJG*^LeYPL8x
zjvKGdRV;z$pK-e)HXY*i6Gns0^zHC?PtWVm4}*b(u94ApIT?-ovuId+)0(W096N4>
zGE-uwI!Zt}I3tUQq-DK7S~ub5TWV^)`}dr@s%*SyEUGdnJtDxY$e;R)?Fqg!o1i58
z-tStlq*O7y4Bf*s-Q+=$lIQ110Y_8^Qg2jG-jJM~KrMY<>TwipU3_2A9NQ}9<28}T
zL??~X(v2d^Aq0u_>fKao(*f~rJ*)xK4}K&n0I3}(CiS*K2UeBXb5e|haFD<IYKC^D
zK$J+u=(o@B<rOPg8r@^G3iSJEf;18{ftheib8Cg`rAUJlbxkj2C&eaP#3*$p9w8Xh
zSLT$)!P74Ji?g!7esC!0j%HC-H$O(X4vm=>K#Z6*NBj;Py<KCNX+c&JKI0DW6<m4a
z`NPpC;`;?5ZA?x;L-jxr+Jk1qPhdR5{p%joBKgp`HKkLj1|DK|B2$((MyX;Pu`w$M
z8Dz9v_NGe~>X%X*Ge=SVh+DUko<Cm&a^KyIC0*CP$qnNhczD*76!&RT{|ehY_4k@m
zs9FLekV~noA2*es3BtP;AszO$N=T!aCczsBaTLups{XzenIZ?bf%-&0V3v!%7*iXk
z>Iuj^KGy!zE@#7{53|HJC#K*vUDsg_5r{6zd#az1ni$tjw=Yd%APY1C+@s(*K;M5d
z4aDuQM1R-w75#!Z6k^8HCIX*qP~yqG0fyL4k%9Rg3RSJ0(5S0Um4G%t#ctY`=|e<6
z&|*_zYstX4)1MK%($W@|UR$H=P_R~2zGX4ZO#=JsqebRMfiqvUiX?LAXngN$Sl&J>
zq5kXy3;~V-P#|I^MD~4A-NqpYuDvS*T7Hs=iOJ6&^b-A=nwnvo?7vE25C)gAbwpMC
z?ECFTozrF^B`ps^G|b8De!dMs8Y(R={Vz*k^fip=g@n>F!4tQ6`i;+~wRBkFBC@$8
zu&1W*yXYC_RpX#SN+QL}e&hVeUMz{WW93b!)22IKnm&!aGfa$(<55SPqcgLO<Xro2
zBmd4+zdM$E|H)ivM1)#GT?Kcz+R-6BUcJ=iGPB3WnpDv4XB&~t9Eip*w4$2|P89H>
zq}0%0f`JcJC^+4?pr50N<W0CbMHJ(gk@haov{OG|yO?Dz`~F+H40rf$_(LTkO`%QE
z(XCrJtL5V(k4Yn1-+T|uOh1OVH(t2%48mMMK8QBQ#~;MS)k1T&Sds?b97_Fac*)FT
z`r)ba?-`YqF^LTop7)Xgk1G@AHAyEn$vQnXmHvYm`sC>5#Oo$Wm<G-FtWHI(-^Y(y
zKg1|C;_l12P}&z%hDxv15)LKm)?x62s^SpJ<D&D28V6jlHsj92hoX4ZCQv)N!W*nM
z65%5&<<u)JwS@7*W+Eqsdu9&!7zd>8;SR4uE&t1ZW$E>q+%%JO6yG31+>eRTLW{Y7
zQ(E8JBA%Ma`!hZ$jz>W``GlKC^P&WVbPGU|v`%>*XDT<4rf}s=W2k_3`b?uyF*G7b
zc(a9#9}qF_-rw@2ziS=0%1ING-rO$D&^qP~OhMoG(OOcK(x^e@m#F;gDc(Q~V(Vd)
zepK}`2G$S}M-byb2qHjQ8_5Q!!jI8lCh!euT!FdSh7;Y1!u+fL(%6QpnOneN9*Z9~
zZ{1pzo+^f$<nuqJ+f^fKz4n8qx{T6bbF~p<9vtSqW`iyNL~6D#U@^<X<g+6!pzM&!
zfwgB(buh60**Ts_8G4MNc)Ro;@KOo%9jU5Ei%ri?HGRN5Pd$vJ3>auU__S5H;Hd2P
zB94bM(q*z{TQg3l>fHha{t5gEta_}(J^&0=A3ra*8=Ceeyb74PYyb>;a(V>&;hUSA
zMGz5{QyL{`uTu+|i2d~Ay(6HBN!X7<56Sfkb5Ad6r<8(ftwg+sOwwn=tb<WW#bMyX
zETF*4!n)qpYb3u9(CR{Qzc1dj<y&;OsU<?L>vOsPX?_QqHdXNDI)I^l9^r+^ZffBZ
z4ikUz=v+V!I-@`taD8Mm*z5kNEqPB^q&RC=N-TlcJR2gnZS<s7e>5eDIGu#>0!xP~
zP@z9!bmS}5taaoF1JXQvHZ|ulYfZj9Jtz$juG2P1*%{Nwo*IixQ(CQq2G0-5&U}6V
z9w_-P_^Tow2MeM?*tgz|j;?{vpX%Gg?~G4E*@^m;`eYvfbx`kg=O6QW)2yiE@zII?
z<k|W<_IcA_qFxG_oVSa-6_=2Bfr$+0&qs4^y4jC^Z?c#5{(WPLeh(wvFRWpDk723p
z6F%l5`Yiy20aa|2_slRA*+b8;e*M<Xo3BAo0<L}+o<TCCFx~|2_ovqY_mRRG)?;h7
zW@O#DQd&=jkNY;ReFv|@bFiZ%qcVQe-xFKlpxc2tSAxO<cVbSTra?)Ow*9wo5L>ni
z4x(UE1J2WfiPYsbhum5geI+wNaa7=6dB_dTfZtcy9gErIq}YkV>(lNombo{i^Bvc+
z>#og1yLbq8v_<?;vkulwi&Q=-C&bvk+ejm0|JE=4`=FgMJube^8j?IqI9g{$n)(^;
z#{GN=!;lLALR=7|AE+f5uDeP7%W|ZJW-Zk#92UeP!Xvj0`E&57?`xu`JzpEy!8r~o
z<FkNQm-O^Fp%$tQX<N@5*t26-U7M`oIN4EnHs57+fT&i~H;h0_ZJR;Z0lR!>+Ac}K
zQVBNuebJNI6ncY1HYkzmtk<so#W$crTjaA8r<baC@6n@=(9kaKryi@7;QnZVU9UGV
z;H}GbG_}M$0Nkdb%iv`!V3cF4pL{Av^xnXw!q$>i>iwRP>uAf&%v8Zhc7TZjgsE2@
zOPBPdD2b<79D=XLv01aOr^bLh?5xBs8eJz9oegsfKB{{UnT1JmZ8=(O?y&Z4r(=Gx
z4IJdwfm+0-|0wcyS4sL%h=+NrPZYhAO50`RTie6R%EVj_Ho1)0<#Isz@ZkZh|HB9&
zvK075C<2ePZA;E@+*>325~EZJJu!|^YGdc{2QGYLyUgH0NmjnVxsCpmAT_v{?}txt
ze=p#pV*?iwXj;u%qf{~})9k2{ZDl?<oskTn2AnQh-jQHH`!j`V4|H8D^UwOV^qU){
za%TPL%H_)8_$T-cD0Erhe<eM@4$0|Qs$K(F!Z@C4h!Cuci*EHrUYGcNr*wWe`c3lm
zKo=E{RRmHy_wIE~&`5lxn|#WYmNx&~vse7D>8{gA<1rI<2%2+7YIj;vZa<V9#nkl^
z-~pNu1w=D8lvU?;OQPEEDE0g`>;%o|Dg6U3@$Tp!u*Is>_B$F-Bw*XEEhcbUy$-me
zb;1jD*_gv$7T9dEja-CEu=2p}&-KF>s22k2xWnm+O=^M7z)L#Ul-|=xI(sB!z5Eyz
zxE1<1rKc&VVp%~lYTw2Y$`r1aMZHmTX(nVTVDnfA2oIqc-AOrjsnqpT?f?}6ycY*}
z55)n^_O5fO{dV=wLzr07_5d1MH1-*A-GA$e6L>EHn^oSH77%Zn(^Ayy0G`A1XK`no
zI*?bx&eAT;qy{<Fl2b63qjhrYPa7z!Qo02208K105`hkqVI%_P4`>Fy0vSzsq~Kcs
z+-mtTxX)P_YFLJ=UE26aPkr~&!Cg!l;I2<V*(U|2YJlSf3?(6edO1|cVp?)Ykgk==
z6SH<p93_7ieGJY44va*_#|wjdvtITrGw2WJO6|U*emf%S_@g!-z0>M*d-WEEgHjs?
zAOpF(y}|9!h{pIdeTTi1L0UsY!=_zk0nIHvbO>BHsMjvVHE+HpU%y>idfGVK$QO0<
zNZGl@0`}>Qjh?(2K#-&y+vK4Qnh$U|vrWW+t|}IYkQm?i2J3|qD&m=2ZDF&%;zkOA
zjRn5y9Nbn?@^^5D-98@FAef$N4qh43cz*lr{PCAqT&Iz|1ppDE&XE!a?_v{|hv^j+
zDz5uuQf7RL*A*?F>=YUMd{qr~5@Z>-mqu5X)8ROOXm+2?ck2?3#O<e38sn(!{W@9N
zsk*7oU<SVOms09EUdct9b8sEqHH(?cS2BR`_4o>J-n>cL=KisyhMgPqulR3f*Al$Q
zQ8wg~UzBc?>TWoLu)xo>ca{2<9Mn#^dR@T5P`3(3)}Uyc)g%lz5TTv|6}u}|i_(xf
zcuiyY#&9)sgDThZ3i6X47#CjGOGs`l{hc`^tf0`A=%5Zn5N^(Dx;vV|`~}mw+Ag<c
z%}B2-8bjxO(L4XBbxwafygo#yz6L|sH&DHz<gm>BVaJ)$Xe1n2Q)4Ohc=&=Jn^kU0
zciEeEaro2s_xA_b%4Gfc4BhS#4UM}hUZG)OcWQbh__|k3m;Y}pPS%k)x#43tI5^C5
zTWV?C+`BQJHf<iBDGg0I0)qb1ZQ85{8a>T<d3gaII;;#dkMVJLBO~c$uZDtOafM#M
z=fvKdfl#NBvJk2-FkiqJdxlequ6*Hiz1tNf!PedVpp>K(zkNHUg`vz$1mn2KOZ8Hp
zv{T!+VhdRSz*Ax|Sht<#5Y@!{HIiePM^iC}=_CIU?XmSn#*@azTBw|e+Bx~uL}B|<
z1c@N5qAU!*KEX`{qxVP_Gn~P}x~(fG#5lC!^N|IZ00&Pcg<%!-N7LmAF8CmJa+qME
zz{M#kknr;54Llkk<rTODB{j9{)*M%qI`pWUvMy*}HZ*+v@8d>pug7uAf(a0;yy%|S
zW+@2<5N+AYeB26kQJ;-9xb54vb)rA`x&oXAYj4V|vd(22QL8(|q9j4?!y2h=xF2F1
zOm!hLPW|76Am#4|8_j>d#<U_nN=cA~<``c6Ri641kt_S9r0xR6xclgln30MggA%IN
za=cq_T88=AO+bt%A`bWvIxR1Hg@%SgsYC|cE2?U0Y@iZ3Iqh=E=M7T60$dLD2c$A-
zCdHpn%H)BFgia~}vpk3O{oR`A_lPtPz}@40_S}-iQ+HXCv5K&j<_{PRe+#TxQKRKf
z=;xnZ)&<Ff(B?2YidS7lBUIGi{3hYdtS8(NZUWdpm)k<R6d|E2I0j@968<TUFg$&l
z*fSGi0)}A(Nkl(Ymro)2W?{5+EBx4TG8lI}QWA7Zq1v%?r*FAc79+JNh$nQ*00>PO
zF@#5n47^_X?F}WCfW?LO6&Q5Hf+yys;z<qu<wKa5Ap;ou+2Qp@Dr&~02m*(D#|)ge
z;WhpNe;i<SR+#Gvn?{pcEUv_|%-q)JhUt+J3PT{JgdK=$x=LjM9-92ZpDg|jCdA*^
zp(gCb2{D|Pcue<+rVYEiz0Wi(t^<sEPZU+<kq$@y1s=d_uzo(1C&aXhCX#K_rfRTi
z#6ibAq<ybCu+^bm(5SMeyVRN{ec}}i5q6;WhT{J&8l8X7wav48-ST3eBJES%Wb!gV
ziy8-JFX~OWpglpd4pYfdd<HEsWVOtgQ^JD-HUTW*Fgrg>o~TXvx7v&!c6(I97{)i;
zSd<)##4X0v9)>(wH4N{&8hj8y4<xkNZABzGV?wr+Spr1H9P~UGOaWY}SUtv$mywv!
zp{x=<5rXHDM0L5<Dyvw?NaaSlQkQ>#2co<|w#^1|Xm$#-tlF5Em<amaT~u9YoS@vm
zuHq-4^ewm*b4i*ULu~`Cq`J~<Y32EI*pdkC7ai^!rh`N$hvJi0-t#t6&J|z>VLYmW
zEDY4W7e1=8dp_y22_g!-pfNTEn%e|mm>yJx=`Un0xUyKkzZXlnOsK#cQp6QPsIuQ|
zVFfp(2tGcDl_BUMCcdB4*1?;HI-T%80lrI;AICs{gG?B%Y#UAhd8;yE2v6y8nIBaR
z^W=Y%Maeowtx!KSok;}#5De4_(;N!G$1y_+RE7o{5>XPMaL?J`$;vVf6bM{zvIIHc
zSn|2O+;EH029Ak43K8B>96=1KEkO66B4``x1jL*}Xa@88VlP*OZ^q+OgWwR%xI0(`
z`!urJA~Q^oH&)f4M80ZO2Ya!J<QZsO&=mfAGyj<e%ynGj;NnWqOF4S{xD_A_%q0SB
zf-x<g8Ex0bBQ7(hRa~Cite5ft<ssntCvjjoJ;Hg!5x-hzy#>Tlp~KgAK_vpdV8s{m
z6+$xtkmN>h(psTh`GM`aixQ!al?3N3OXj-P#Cmoy-c(E2bAJ&C;fd`Js41KfnxIU2
z<2KY3@M;#(|06c(0h7mc@7w%5sw2dquit4vBcdaL0z|S~_>E*Wx-&a$zC4&RDC9lu
zk-SN_oY*k%yWMAU9a1>QzxPSNcZ#+4;4H6uAs^(kK+Xn$x=I;#e2}+k$=Qkc56TO@
ziZ4cH5W^?~KR~fw#_DQbInhHngyKedn830ijkz?$=7E++x>sNr2+FK75NkQ*6LZoS
zadC0G>m;Dh)83-K<hDP?XL)fRkx8Az7c-y5@v5S!dd)I_7tavYjcC!(>*|6;&N7X8
z%ns;>32_s>8Wp*xt=1L`i}UwQ3qSmI8zP4ng$gDkEB==~`1RdlMx7Vk^kjp3GDe?@
zUlqJj_~Ksv?Wmo45f5duUSIs4({jo@tM=DphcCyz>^!3uU4LwiDxX%vS?Bg|_bXPP
zIHaVbbX%*l%+POk{q(U_BX(VV^PW>TvX;J%&K_(R2+nP9pPh6o%0VdzchfNx1g&*J
z8i6c36OeB0+KLQm9tEbiv~4b<x%Td9tfIdHse?FKyO@Eh@cT`tm{J3aN63z>iSQSM
zM3*2gbKS5;v;+@n?(bhVKE8DU@nZMkU6FN5L1WaR^~jlwfScFoF?D@cSs4VQHM)eW
zL`tT^L2%XqS!D)BGq#e7Tk;IzQXAJs!2#|G0)S>1gL5$Ay;=S9qXwyZr?zi^RuZw_
zl)IO{z<pQ)OJ0K&zk;Jp?F4Cs+(->LN%605vR<R3Hq+-g6*(dR>a$hWW3cF|CQonQ
zKu$7X)Ym4(M*DBEDB~(6ovVGzN>Q}Un6CWgB2=!oH&{HL;$J8zWvIB+i$#w;a#H7{
z<VC@8nnSe{ae;4UBV6zZVHf<CbxqD$MyXSwVVi}wtO{W9_vc({UwLzB!5uX>vyzdK
z(KUMjzgQ9YPE)d21<kLkppZ%!-L7!8$paTz8BK><i_4^NmlrTA+62?Xf4?3`(^#Xv
zsqvyniXMq3Fv@&>E(-kYYT$GS{)|Hu#P~p?(d_#gL7A<=ffB~tzh*mb3t`<a82Lb7
z*I?ZNVg0?kgvw!Ta<V%$x9#rN&4Jc+RJ|i3%NNMK{+acxj0uXHz3^V;_g{3r$SO2)
z!hdC<9bhpP#uM9eE^UBKIHHv6NAr~ba;}sV8`gwXErrJ42Hs<<eXwzCjbW-)McEbb
z;%*{ZKw*a@5S|l#|9%Bva;V<#GH(gA9;bStm*T%;r*LOVaXya$=H5^6z>GzYKM~t0
zT){U8kRn1y5+@g?C2ehO^}9UG=6+_Jqs2`5jUUAlqUH}Z7Vp1QCf3>V^D8{LS|pDP
zL*~N7{%zeQc-RfJ<Vu3~T=MVk((k%gJ;n9!BJW)(<Hi~(&xPS!Fm^M^-Oc#@+oI}a
zq)B=bSp4;36b4gNM0_K8;B6nrC=1RtWi>S^@63M~8ooUJN|fTI$-OH8Pvj|O2oqr;
z85v4loSztq6@+^8HwMqy8HJ4Y1{fRBA0wy6{h2^U6WJn;v^+T(w{jLK@W3wwM_$Dp
zp~@;MfqOqmzt>minSYg*W(7AF><5wYs=in^&`|XE`$?_qlBSZ91yvP*u8%?qP6eDY
zJQlNA_~;=nud5HiA>EFpkgR2}?OC;p(bWEhOh%j8t~}x|(-*{L^6!{>!2DKyz!mos
z)LNaC9?G9aY}(M(*QfmM|2q7#q=eWD#s5Q^l<!feG}&1htSsyQ`jGBFvuEZ-B3Zn!
zmwO$6jKVGf6%=3O*QRoo!BJTxpj9~evd<-NDjr1l1;jqDo(KBRRRdX2-^ltk4NZdq
zm8CVCl#_$j|0W#c3mp#>wvNpC!%qrmF&uF)PSoL8P5#?@tXuEs?7VX8)_QE{gnkW)
zuw`l^cBC1Cmx(n;pr37u6zz2-o1#w!1$}??MgP~={&&}$y=RRv#)i)Z*teP0eU+7!
zyFvZ6v9;a3ajgptlaP@}z&gJDQ<6Xln(cB{9_ZZ}Yvg7NIF(S>KOSbfcooM8u|)_U
zfYL1E(A*$s<Iy7x@W%;HPnSfUJ-p8ewaR*^eqpeQu_@SmPhw)mCnkgi1p|opyp1#l
zg5hf28;`zDOCvZPSa&*L`+)ocrCeTqJ`6k6QLrHLV>U+xq2N&r@KE+(wX%-BesH`P
zI-w32*;iY<w~*@W=?TJy9q;1hf(V#plGYT&DZ-4!$6A8gsLwktf9daD#f1TF#;NJz
zu6>ARw3nxzzAoT~TPhkd9gf60F>J@-!STVVxQ~WbjxmUZt!#^ro9vr1j0gIsyTT+V
zTG~IoOkwGYPcQlV>5a()lk^K1(`%WyG#Q~0RE3@kFp;DF@F^XgLm+5CtcIY{{pSzS
zE~0yZQFq7iux~?{4s5h_Pn=LfZ3Ui*#r#AM`qI~WJK+9^_}?`&6joSx0EP+xBk-A|
z<=F2Qin-Qjt37Zr<ti1-{z<ojvH`CdXIRkRzur|V*8l|I6a%yXmvF}jAs{PC-cpob
z+I#NG75#$`=U$3gnMBC|58MrT1;wKA7%W9mjBN2?6#BV6U>n2=oc#C48imgXe8J`R
z^(n5%5g`(JIjNfM_AP^9BI_{ZTCi`yyxAc)3lK5hG%tAc@oiSLPtD1BhyxA&(qU9z
zU~-CLM~jEYtkvc9_o`Ktm9Hk82m&7$T@-n4qoe+ZCy4?KB30*s-kudoV`Jx^KQJ>~
z1MfxL0TT}%Si*Gl^nN%<(V?E?x+gkb8??sR%|ks=7u@4t7IQ8h9x$WOKCUdJ61^$|
zl}?eNv2iNi*4)I?TYRLGUTVIpal~q@qUndLe1t0e#SJq9k;gi^UQ(bJ`1$<X2b^I8
zlzI4t#N)hLPTk(JGIRshUhaZt3$U~L?V|&Da*)4j73(E6JZl($Wvy+_(l&|h=YQQ`
z?f2W8sc&QwQG2VmC;8NNywNa$?04M{V;|s2wr{nt-4roVV(T-OcrBV{l-j3JaL%`5
zY_ui7SYTVZn|){GReiD*$qOVCa+}wIFGCs6YIWGEpipwU-BgDIREkI7xZ-`bcpu0(
zyxr3<RZm^18I2V>pLP7{I6E+rEuXztf0ym=@=Hgbm?({p8RzVlMh;Z@a@pAO#ii6)
zyPoC$!GUk1KE|bdBZt8ZYE)gIzDwz93G>-+KW0~dq%;OmpSh`KJ_fSg2NVdBJB0(S
zVWp69p?2Kim(~Aq0r;!zuf6;B>OminwQ>+SuN_Y^$1T<}iE3)Xn}{F;@B!n3i;y{L
zVY-RBrlPeq1~3=g8II*P8wkh8qpTae-MH{~df?hm>n*u?8x`+`DEcq%p43m<@<x6A
z^@EdZ9ZMFa1|C0t-If!6xF&MXb5ljmRKsiWV)|=M;p-2vMs<i-p!FU=g^_&$)eOiQ
zXV_e}c?Kz<JoK5WQ?D0RzM+P9Gr4%LE;S1$XVdo4Gp44su%D<wQM76I`BjkdSm)Y5
z*G=9iZNmXFga^kf_^A?unW@3XHDQ$yl8Ah?h%YI6;_tV(n|gWK)7$&}#F(IhQ0J>4
zaR)AlP0lU6F}Eu%a)?aYs;w~1L3iy6hThA?mTNAAhVG(J?qzyQJC;nbX*Q*ztv256
zV=Q4L9Ef=0u%7$kRngTZOeU<D*$86bZWyh%lU@0(DY2T6mfnyjtNjw072!O)|1v+y
z0=%afv_Fz{76y7aToQ9klpXKG^R%=G&<LucjG=u-Yr>AP!tq1Yx3-A9Su9^au!;S!
zg_)q6w6ssQ=Wcaoffq$Zyg^|bmiNCOwJ5&F4pBd{o01Y$-R|8Wc93;9de2{Ub$7li
z;5&duLM-$>hA=$#=4b@~bnV6-5V=i3EcTzOs)FhHJ(aS}-d#P%8J2il=l*d0_VA<e
z)}L2&$3lVDaiD~?y%#d6e50itm}$2`drft9$Zqowdkr@}lf&nN0wL!%$bnk%L`v;W
zTM5`h6EY$ok2uNA513N*!roPfNW|>Fvp8GI?t3w2ud2?G+RqkCX~yCvxe@iBH0WA=
zWiVCaOH+g>!gW6%F}^Xk+1?p*;L_mKMdY<P=ug}G7sN2c0cRN#q#aumSmKT0?_&k1
zR0Bh2M@*4T6IHqIo~pYo$we6tCDd^6)`C_^3RRpoHenxh39wFX--K8TcKS6O*%!xS
zyj$qfA;80^^8?rd1R=-z371vkP&YTuPg?LE)VQTC&~r<`|7Ly~GwyDBo@|@vjA4o9
z(N0tUIojZv(F<QM#G$E;po@Dq{Po9=O<-!F8Q&Cpzv7zthmNi;mr5^Fko-W%V2=JG
z`tM8U`7p*uN4J~-b3wmecwP0A^p!C>T?{aP9i07j7`Q>9p@2F=ou8rjL%Q5E!aj`@
z%4D6oKxcP%IRcAgO}iFKARho5B!Li=0Bsr-ia%JnS)xCNrE~z&3L-IQM{BL(g{JpG
zi+(@%jtIy<8q-YQwA)E=Gaa)<>(?0hUH06#=?Km4FZ~h$m!Lm%a^9H^gK5mXdA|5!
zDh$I8C}$j7lq3TC^Qi4`uN{q}Xhk=pTIL>Gg*n;vK+LFNM<V=Alfl{8&vW2OyqH*M
z{hv{?!~)q5n24=MX*3=oyi6b@1q~n42E<hS7cogmLUV&o3XdETh2D4c0<fEsX}*r~
z2Z6={6E9c_2STtzw8vWZJEgO;i;%O%QfAw_CuQ0#eto>d)zWI!eNQ#)$c<a0PVa`Q
zii(Rt{3<?Q7Y>X7ww_x%H5uwgXqo|~oF+gP)eP<3U1hG@Jf{f@2@|`+NexF+7=+Sl
zuquZ8bryP@-y0h*GUFX4rwE8q_vmQ*NQ7xp!yD|00>L>sJv}KS<2LHw6FNFqOwy~`
z+v8yTvd6Xm1U8MObMrUaNMn5_DEimn68qL@Cx-3`FLF2`SkFC^n>k%L`ZCwRU~D+|
zbxU!y^R0tTos#R<GcX9^-G-DE>w`no5+DoagX=J6d9Wgclh_0l!=D(p1zPK)@yP==
za~fbJuY{`X79Z+7>Y=rChua}(heOAN+KTXwKo<Qh2r7VMtUt>;@8af`2w3h*&Jrpd
zCVVO;bBzuy@HoDWJG%!17;sJwp_)Nq29eTL)N*iGa`22+OSpp|0=On1E=~Z7iVnMr
z+vbzBv-~I$CAFX8GhZ=DI*z(;(uhWh!0znqOzawP6ad#$qq--S2Z%}6-YH>f0Plm$
zv)Mv{)}z#)UyFT;IFNsE<hKo5S{W7*^(|xeg2@$G^C5RQkVWQq)rMKrZ_%*84+}y$
z<6hd4fp35^=ym>EL8nX5yH`|I5xEYAm4R`kid^7tzE1mKlci%~;-@~f2UQReG$L&q
zl=#TuBC8r3H&7v6VkbB-w+BydE8lQW#yuefjvlOr%TY`4I{cmzQ^oMw)!m(+LqMSR
z)#((AI~qbNS%K-u*j6d>`X^2V98c;-x4W+QVYWJ`4M`YbtzN_V)?oS_=s!ez^x>Ak
zEfk#I#U+)LQ?}Z;uUGU^dQc&SH!Th}rSbqQc*w5H;9R&WV}Pw>gqN;^XY1Ca%b+kr
zzU)@GqHigrval?5aZK}cYA;gPm&{@KAa$JYDFj56J>-u4UGS@mEh{g-2K$H;Cr*45
z9!=0qMsZRO5A2jiZ}lL_aJ6eVKDb%oy`oi>m7svAf&TyGKBBi!z|!>bb!Hw$b1pO`
z?k+CeoP2y=AU?yhyR+vIgL4C!i-PLV?QU#LVzB)~`@(_(bVbl5l0i5^%Tr(fXD=Ix
zYy<qRafnbSL8=8yb==Bd09;`m9aiM?jo}{-yef?Wqz+2(z5>&93O3!9V|Kl5YvMnQ
zA>IAZdhy1g>CG4i!Rw>A+uKVhFmHpl{J6@s7R<@MH#P0koXcz(F00HgA!<K-$urNN
z_+BC#RwYa3gaMLS0&?`3sNRYQ@%+ck0`3-CbSV2>)u*#?(>_~enV%aJ&;+Us>cZ`R
zcYSq+sm9!(r)$~yViU#vK+8d-KnIuDw?-(c5_MA3(@#%~QhUL-$}KX+{vz0)cISZN
zhgq&U>LX;$>`_Us>?m~gNFJY1bT-4N3<9#<?fxspn@_Z@Rn;N-Iiu8v5D&EKEWO6=
z$$)?WI^V8fqP!po9~QIU!y9mBRR(_f1LT4Bk>^zITfaYVq}hgMr!ymC7mQJE1_uXE
zdE)KEqd5n|zKH1(4DNuJVQ<*%L)x48mUycog!y5viee`V%zK;|MkXd3%A_6|Aq52m
zQubiCAMn(2<Tuz|&&|)bj&U0U`Rju60`MhvEP3N*5x0J$ok&vpc0;fRMdDDLuh#xu
z!J+*2GAdC#YolzZ&i7lwDs>aYJ{@N?&0QJy1>E=nj+Uflz`m8|Nw|YCdMTDrBuBKM
z<H7lCciA;DJ4^I0sI*)#SLx{LzxBTX3Z9<x%pXHU%NE~8J)BDtfY4(=*Z`>_1Pxei
zRGK!O{KuV`JpdjA&R*AY-nj?o2M+#ZITl8FXs_NH4QtsCA$6laO1XX<cs|(p#N!OF
z`#<>1=mGbmq=o^}RjXEIz@v_!M!{=<!2?tQq0aTbA=n5=M#!EH!;w>oOdja)&4C3&
z>2VVKWW@r9;lc_sId&3#AEAiveS8eeTo!)*uNa_$7`+n`x3aRbl<OJr`GnC2YR^-J
zX)Z6tn84#VTaKUqQ~jfyb(%Hvk0Gzz<v{zI-0`KlQ8`n$*o=-09;VwU*?~HzT$l9l
z@PJiw8O9XjQ&acZPcjHmAA7H#a)SC2E^!M0CbAJRE-r4$4p13n@|c-TEX}qbSf-MM
z2?D5S_?R<4bI>rumkw=u7PLM{9{3hcub1590*UW6GOR9oZY3q9LIK80ND9Djv2pIw
z?~knbbTwC3*Lw^0*+xLa)zT?@0I%=(df7$;;|{!F=w6l!)S4ZFARwQtjJxM2!FTD(
zrpvGRh8$B3;xf+1*r|FQekb1ydFln(j9x0%XRH)sHyIfjE*Nnb7&Jd9T|Bj&Y>0tU
z2)(<@c_twT3YGh_1iFj}1E<5Q7c~>=<cEibYM>h%pPh~5-x+A_tbK6$R00I-uMHc-
zTSkBfqdkCu;1eVzq7jFZquuO2+#%#X)%(wN1|u>9z5je>yHg!E9`r`YQt_hoKn#+Y
zk&#h%3B^D#h9}tn7*Qvzssj_18tEOHe!#(i`0h30Ny*<9`u=XCrdX)t`-#M!u9%hC
zgsm3?+RO|uKI_un;In__WL3{|3cFtV1-^99B_2u`Z{PZyr7MS>hv#xWLsdH@7`A-l
z8xM4Sq$o;rf8oHIY;t3roHC{?gfJ{+WF(vXFwl@&-iYnSNkEo?inLD5l)rxuhEgBm
zlo_B(q6n2~^St)+rx83Z`xy)))!##xD(AN-i^WtRfYV{qle_Vg%@}<J1W=VMPG}Ae
z4dJe-r&AKd2$2ybIhX@o8>b<NBFF~@a##@f^S9z#ODiiEAYmwmC|#E-+=p9L3+_f(
z=CA%rV@hRMcIs(z&qwbo+f9QSSd(@<ytN{P)kS3Zb?n&Z(-fIYE-vRW>sg$LEWJ-+
zSWCV<&BMEQT^zB?#>*@+Fnz}Mx>2fM9Tr&wfNphukP*mYpROIJ0I6c(JCQ4ZPJoUj
z8Rm=*4r)MZjHaY$y6FrW%!y#d#VgxRzBr+;UxkW`Z}O`IJxBDH571*Z!v_=OQS9gJ
z93F0!Sfh6~Ht8)H(vSSA#&|k8Elm{M9OM}w?8BaEFv#4#9rkMnvS4(C(iIcWORGMk
z9>yZ3Z`&v15JlHSN<MdR=37OjS{50wT9)pn<6k$8DCLe2CE@bT>GS-c=&0(TT78Q1
zq3;t5JO~{VSzeGC`(O6roW$>04oXVeiXTZtc&NV=#FF|IJ2f#h1@n)07Bzc!Pfw%W
zC2a&_?6n01cn@{RWR>Fb9rVs7e)E>gb5)9LV1L192j)|MYX#;b)yj#J7XT!pMF(gI
zms5Fpc?WCe)0(h^vqByK=mxqX{Jemm5rAJ>?rMK;uNnsAc}|@)Scom{<jmm$%mcc<
zI*@NT+xq~Q-5I_cZvVGZgj$TImqt3$&o3c_E0umpa{4B()I@>0+OYcpWQ78&M~d!$
zHWHv`%NCX68bTHJd}W*xXFVN#eSL!&&G;<8Qd(d;Bz>>C0+V69v61{c(6mFG<bBNs
zPE*rR@qF6N-E<bXU_R!nL|!E(AwfDc;?Rw+L@H<;LqaBOwhjr$v=P|`zxsS9niQh4
zZG3fF>ua{n8e~umi93J)4vZHAQ1#@=lQg5w6RCQH%1sI_l;OlgN7id%3$D-pfp43?
zo#J#YWu=c<yJJFY&sgUzliN&x$Gu~D7q(ova)q*_fGWIjghnEga-AKe&}kSU0;W5a
zUsza*3d-ccSfj$CVWRHRC$Ql`fY>3Sunh<=VALR>B3bWzQHsKVB=_S2T0C$r`Ii?;
zS9p-RV!a2qp=^3AcS#gJV!Mz(y2&{Pb#R?p9*63)v@~YC6xc!VNXC($csavh2?c%5
zMINAw*@vFp9x_g-YiPKKCK5gZ=q)awOIasTZnsS}DXb&7a`ge~d(>P{ISqXCHwU;E
zaF%z|i7M<Lzb<i0(R)e^hMEMer1+%8i{)1Cw^n9wrltPV8o^TLFk8dypRSv{j8yRf
z=Y0SR3a3s8sT6+0dKk>C`4W!)Z}|laQrr2kV(bQ>p`{(bRxN;<Bxrz9HxB!ur(#TS
z*{z+&)6qnVwzRYaG9fc(j$yWrsj2q4b3!C*CnX{Bpy-e(-ZhrWTLVcaj0tHaIC<dE
zUYJIa!0a7^PU4&lZ98#yGchsQnbXt_Y`FPx4*Np8->K6ReouLifY7}9ssAzU=8#>n
z{3}z^v+6n=sN-dQ=63DSdo2h@Z796v3M5i+>o1@|({<Hgs1rgD>w@|;1#hctJZhGx
zmN_<+M*Bs%MU6z=;$egvAh#G*)b9ym873~BoA$ekp&@gQL+1jQ75m23PFSf&9EnjH
z5p=|J%nZ#1S(#0g^jHIkTk^J=GC1vxq6B&9mm9@6b$_rm7>$6#=~Q*{VEUVGTLB;K
zdl*!cZB^<Q=g}h)b;07&Qo5|;A3Z`c`LQiaF3yv`#%|4`la1+ymls~uef`$e-CKUJ
zcoa0H-q4o)a2q1dTYY^e(Q?#TKSzMU%zOHJ+Mh>w%kan}Qw{~P@L|>`ig)ECtX53)
zIeJ#M_+-aY0z8b3jmZs0PgT!@tF1>`^cv1y5kHMOetc}q?C0ByJ4;4Rog^*!*RP`J
zI>FW>#6!F!>bPD+6+X>};++^kg_nw-+K%2Fcq+MGY)Ic{VE$SJ7knLaS!gvt=IBZ*
z94Y91vxLEp1=@gcPT?G*RN`HV+)J!L|ES6$=0<zii4H4q&#YvchiAsu)E|;kQCGd>
z(O$^Nz+gR*4ZI8c{chkWgFSHrDkpNugWm^Uq--de{CWTYZpX+7fnCXfGX2bi)UI+I
zOz=K(G&bxYorB`SkJ=;0kGubB8gK=3XS~_`R`!4&qVT=QNOg>`(9L7b1H`~b5X&KG
zCOj4UT{a#+dg?Non+?$8emftT^99;QFd#d+ES~_kC>)%mq@D?Gb-3{TrH;N2JRKN;
z(7GZDtIp&Dx40I(Uf1wUNbtLBu?M~PZF-7X%lt^;sH|&03qU4R!0ZA7{*?jln}P5P
z_>3Kju5Cf&(PJ=6-xfL%8<_rZ&$_uJz~gwq;!R8a>K6i8hE444cjB26lb&@npV5e<
z%SAOr!N^V)*&mz)40qU2)Fa@?p>Mx7e=ah)(FWB!jvm<$YAAVBeEatEsQDwt`UVF-
zL5_r*a5hc!$Ir<IhwsRjWUB;9eex(563IfCEtuZms8?5S4eZZ|;FWui62-BQ%Kbo4
zdS8GxCR&3qX1}QKpTQthk&SnfL|~-XzC7)u2C|?Wc*T#A847&U;5b5zo`A<<K3ztk
zM1sjlcbcJ7F>(tXJ^dbt>TnJm2Z~|M1ojoUMo6=~=cZxhxD~3BYTwl4WGtek!;B0A
zpw`x6<5XjJLwE`T+a`t*5GU?in89BLwGnUNi%PZGK)eqIRqIW)X)oc6+MxOP1+sg6
zOCU>b47w})V1r;0%6oK$#8K$ybO|Oyq~4X1l0tj}@`L$Qt5_h$(dW;f_w2Yr@8s@Y
z(m9))lCoJur1A}?;nAZZ-Mi5V!fNXS32Nfv&(ZXwoIu5e!5Uf+<htb7uVLiM&cdPu
zu|z-Gd~U54uG!So)SheW6*}M@WdG(<7-83-8w8O;u9JYqTYOsYhm>(ak63I`eb;@3
zK9oiI?3b-^KJT#xU~psIm9T{UM6>!VnBOKcL0W6|8z2t%(9Giv!RxgyLOrE1zF~l9
zK5s-sl#C2>a&g_mLJLGx)KypoNQ^iinbRS>lPCt03{NkwbGtU;3LU;X;1h_mAGizH
zrlE<RD*DzcSpJeWjodZi7D;h&3-}a)`}^zL-*d|U`VordE$V8zYlmX)-c{Dsy*nj=
z3c(Qr+r>W@6cb8H+MMH#CvAq6G$Xs{+XXpI#rq)oqZ)!cUztO$dDahZ-ji@C#p_)-
z_|pnqKG2stAS5BVgDF+iy*u0J3u%C6XRBIT7GAs`Ahja2;Anx-tQx!VgvQINH6lqN
ziaE)zOvHbe*%MDchWm@fMK{zQY>eK);txQeX!-9DYG~u*QbH9myk#b*IeL5<O=TR!
zZ=c__aW5K5(5!U5AF#xiH-m$M;T~|DPtI|ADU~fP(Gcf!0<}HErwwZu!gK?o3Z6yP
zkw<zds)@SzGi6)@Fb?Z%BlW~e!1bVDpf=z?=#Aa*;#lt2{!|No$Pc0=^8F0aJRy|6
zv)((=5V!oct{W<rUL+W#WHh-al20Kfaj3p)ZRG=OjGx6rl3$+dmdRdE$U7_3rYntq
z!t5|tBnDJ<>86TsBI1$DC$G<|SANgLKHZ$jyA|dB)7?Bk70=}W_`nE23IkfoH<%NV
z)(A|4+F!2s_>W<CL?J*rl@?O1E}fxGzyFQFI0)pobt^rX&0nm_9dZvuqyX6jAPdxu
z)K+9zJnRmL#0$0xq{IQe176y}nAlQtNmfIIP*Gmq{cr}cfEAb-9yV}>aE$`Y{s_~H
zV9;TeI&x@Ob<dp9#1-PkVZ}@5^(L7eVq#`Eu%sw-Y*&CDrQ!8Z0sxM2EJa~sVJX9K
z4LTYZ7e=8Gga1&72Y2sEW#a$SSbMOWwV2Np2+ZZaGyC=hMkp{DKH_~BkJRtzO}+yU
zarp6GuQ^-v_KQ^(*$Ihy5i~#R@yeUg_mENHvWiL?bp`j?>vM;(F$Fn*c!Ypew^6|-
z&%AB`MQs!i0xB-a18d>NyldCk!&xm5^x>xoGzGl7v)5|qwf>E5;B%y_r*}PG3}zN(
ztR;Azfk1q6+iB#3Jxst34d9Z7(2c}FmUqHV!r9&ZLv8Ko!&|C494G=S?}}*lLydsu
zE^nYlD2OFK!8f-J&IQB!0<v>!nAK@^z}smVTf;6+)UyZ$_FJPH^je<pGt99wOvnK5
zC?A|W9F|&potxdkq%i`$&I>>xru1N>f`EeI*i&4y{#q&6N{J%0vDS|i3`u~4Q)jo3
zjcy9hJI!<cX}lOlaj>t)<V+a!HiYs9n2V%1<Ra+do?<pZc3lYxJ$py<{l6ZC5V&&R
zu3(HWZc_9Jc^k%K{tqQhBi@<aW<Sax<VRd9aelK4HCb93K2n@IctEkiNKb$y>7(ha
z%D-3<LqYTr2NjAq+UhuUO^6x5bpUN2l$31~zml3bczHj=J`wc>*_N>Qd#hq6>JDLf
z`NdZWUN}`&2(J+>+1c6P$ff_7*@2{n7oU}nuhru6t5e(Y3v@Vw@Lq%43w?SI-ZoST
z(%Wz7Vu%RH!4QuKS#fbHNl^r_;HOU-&{usXe}EP40BKOLgE{PZGT{_zW&*Fc4(MV=
z;sx$3_BE&7^kfuLVF$SUvxmXp^8AJGmjfE%JUs^Y=-n6Vw+!r7zHvYoq<F`oaXK<f
z<6uC_#!(^|92=X<8D4E?=iTVr0cHL!koZAMP;2;DJN*KAgfyOCj1)kjT9yHih*%;~
z)6(r_I)z%qXK9*(IP~A1C6qKGkbfc)VMi(CwLPfd>{<&R0M7^bcnr`i6sXS&3b<pA
z2g6YHv%s~mgvU^;lHVZnwiGn^b3h-!Zo(4#GVC?@gh<sB$f|t9*4lBi*|3n$I+`X3
zG`sOlJYq5991y`}wV`VC>I~unbJ3CGJ!(aqPjXrO{B0w*ERjKo*(LDPg@RZ|;<Fy|
z?B&ak&Yy9!<;N@7zvO7+A*0i!$Cv;lbc~}QAjV%l4k{TW8;I+yC{<|8VD^cHLl}RN
zG7$qRL>BkL*73(+)6><^pA9*llz<ZI7u*!|ePc0f2P6Q|Aq3QjYv!nU7i(7**CW>+
zXG66(PBjx6FDCG-y3>{_13M)02nqqhfU0*ea<EJ1`_;tuTVZ+590!dDgNomQ)fVS5
zi~@lmu~o0&4EE3B@sLWIvJ{0Gbad%rkK{U<8t~a%TN5r&i1H7V+ue_kA$6e6Cg(6!
z5A-B+=EfdP#i7AL-qnXB^j(U$K?Z@TDF{|ip=~4vmg|}%VbTJS5xmlmYLRA5O--4I
zk2usJjW&s(DZy@!BpPyj<#q?MbnRm{v)@$mmnIo6+%ODE*$L}>iV33@V$e)v7{qCo
z?=-7F?|Ukc_WNO_2g{RTn-y#|35wgcUDT0^ilO?b%Ar$n0&xJ;-AZ4;of^MJR^zMf
zUT$cwYQR^GvWz1nB&<+!3qNIC(^Ef<(+ZK<eXI_-{peBiQ-%-dlgkmLepsIyu8BKi
zW+sTUg4W^^HsVzMynqT7w>XGma#^d_9mj>9!OVU`i8oC&Dz^cd()o25PQ99<!{fOQ
z;7eg;@e+9mkUa1l^vmBLpOv)-_@(IX-Qy1_RW{713CSj_(&g!NnW+L)X8Rs3U~d=^
z<nW}{x$Wuw8lws82J_whLkMRONNxrv6B9Ca*lU8tfJ93<9pZvggydTYO!of$1-1hU
zj3(6~825o}DNj0p)Q8+G$~R_4P>7Rv%bv6Br?8J3{XU*>wOZ&NB8lC4{Te(4y6<Q;
zP<<77O@cA+aj&+{dM|Kmpf@+d!$tP*kHY?ueffKxL5{_%;@^oaRwd6UeBf-O8I{Sa
zMK=SZ^UjC8@-Yg&QJ~|s`AljdV<TT8QQ>1C9g#h7z<k%nUEJZo(^2_>#Ay4_Wdsm!
z#}|x3;ZH^m!!6XZsNl*0j!H&yF$t**t;$%@<LD`m;4LX2gph>U9qC2yNBw3G%ufZe
zbaA2<$%H-;UV)^exL2DmFZ@D`31e_!xE#W9$SPXVO&Tz_|6;cQIObz7I5n2PIQo@?
zQqnH?6bKcekYniT=y@5~0Ks5A+Ps1hJ_^1l<f8w*jkQWY?7%((v+ENCmprf6hz7F6
zcdyQ9T5D>2B^7@%ixE;^t^abC&!93x-A%%}a&eJD;dC)msR|DG%&I!5T164lF&qJH
zo&i1Mj9W#?(##nUCu$-kO9#egfCfQ2-MixBZ2U^X6U1jMpNbnQ5eLwE3pKG6Xxe}P
z1^Lt%gti1^dsEq_jy0}sw%W_R6bC%1&2Pbe{w-2(5E3TI$%v`IRS1BdH~32wlE>po
zAbo%Zj@!N&qI+=X$zJA(iGnmaEyaiMc!KK%YRh*G4U;c!5tRrEpiCTUH6b<$iRS}D
zLz-BHhf;PO(@FdasLY9pH<2dcHY2reYm~rU(V{W14NSkG`EEx)%Aj8C(R!e5$RK2-
z1CS87kzI4f75r?~Uqk&RmlMfg$N#T@1mFq|05)r2#qIm{_La~pn0}E04Dd5vhCNX|
z6D7dfh#L-`bNbwqngv^_#~|h5uLDkm@g@p7(##Pod7GIKzKISM9JWc0GsVl3&?4vK
zR>D!FOepLa3<$-gu_(WNgH{=h6p+#mw5#Y8;8a~b<eh4g4o25AF(#xIxZ#7`ABy#G
zSX2P}MN~sC(H3EB=s1&Kj0)lczMdB6>pV69&3?0m?YlA5MVxe{!gs)FGm`QN0<9(n
zNu~;~R?o=|7~H-pD*DA*ArYL3!4<6RrOP*Vyo~qsqVBp7>vbIjoQ@X~`bGRe?sygb
zp5QyeWb+H6G0*E~Jt%rn&l9>3HnLru$+tsev!A%!yTFDg$C}DEd2E9ZXh(`z=s;H0
z?wnaIj+<eL+jHmf<7!YHWSu&wz{eNev18~}I%cH@FetG;C0vqt`qZi0adEd`U6-nd
zfxr%O_yVmxs9%e)=}>W{aJze<y_{Q|tQ8@}0=+XRRah%jfh(I|UKj;7<@|Xy{8o4{
zK!2bKC_woLm2Ki!PgSK`Q9%S$v1Ed-#W}@+Z;U^@F~eCFwZ~_kDjK{Ic=}&|gEKr&
z(I$U}A6S)P#cXO)CJJiT65!Hv5PG6cn6LQXpR(loQXPP3Qq#R~se6W{L_XjYe$W<A
z5!;x)zZnC{{qL8!22$x68!yGji>=tzjP)?vLA3}}XSgPG!wz1{;GqBgSo~(Z0cJ?l
zxx8f#B0l*vLoqQi^q4*pBZ{MjrvLx1B=K+FUx;ZiDXhMJ{W=ly?>CV%DQ!L{MU{f;
z`TB>dDl)%9tny*|?{W`p3n+dn_UZG(wWeRoWDpgP5;lNoDL)q%*ZivgE{m-!YPXLZ
zwdoKHrcg;>l-SkVE4owie}8<Ns_AB@COsKDuLqWp2@M;W{Z{^W8Ej=~djgEL=jv&~
zroETS!7{P}IRyNdix>ZQ1r>awc+E-gD2IE8@?+^0^)cUY`-=e<F~?_}kj9ae?ska$
z_hlkcw-Cp@2Pr8-RW{k)KO19H3e?7;(#bRugHuc+5j7Lt{`>V^>VCUuW{O%J;u^nS
zO4~b;NT&cn3_D^B;zSrYO#l1XysH8?1C~yv>S-6ZjV013G}>fq|6N!hV^|+w-$d#E
z`!r6PM+G4r(kC%h>jrNKEenF?ZN2|{d(NyiQ;eOQz609BJ2btR)?gO=bMQcl*ryGS
zrUQM^zOO}c>h47JxVL=x@^&ev{zd8=-;0iN>wn9WhrYZSs3Bz90B~Ly$I{M}k4^KB
zF$G{@Z>nB<J=a2Z?B$)?AG|s(VEe`X*XPNfe~Y8L-G-fchR2IlT9yyB@Q$`dwdDPn
zwHX<1|LFI&_j2xxIX+f}efyPnN*{Vk#a8H!MY^;8_w21am^QwmS*BSWVba!38=L&1
u@ZuL;OJw)`|L=cAkL}58q<?8*g}UkjqrR0^z8e+(IjVYGB~97v>i+|+b-vyJ

literal 0
HcmV?d00001

diff --git a/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg b/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
new file mode 100644
index 000000000000..4b4014fda770
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
@@ -0,0 +1,374 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+ xmlns:dc="http://purl.org/dc/elements/1.1/"
+ xmlns:cc="http://creativecommons.org/ns#"
+ xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+ xmlns:svg="http://www.w3.org/2000/svg"
+ xmlns="http://www.w3.org/2000/svg"
+ xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+ xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+ width="447.99197"
+ height="428.19299"
+ id="svg2"
+ version="1.1"
+ inkscape:version="0.48.3.1 r9886"
+ sodipodi:docname="GPpartitionReaders1.svg">
+ <defs
+ id="defs4">
+ <marker
+ inkscape:stockid="Arrow2Lend"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lend"
+ style="overflow:visible">
+ <path
+ id="path3792"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ <marker
+ inkscape:stockid="Arrow2Lstart"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lstart"
+ style="overflow:visible">
+ <path
+ id="path3789"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(1.1,0,0,1.1,1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ </defs>
+ <sodipodi:namedview
+ id="base"
+ pagecolor="#ffffff"
+ bordercolor="#666666"
+ borderopacity="1.0"
+ inkscape:pageopacity="0.0"
+ inkscape:pageshadow="2"
+ inkscape:zoom="1.6184291"
+ inkscape:cx="223.99599"
+ inkscape:cy="214.0965"
+ inkscape:document-units="px"
+ inkscape:current-layer="layer1"
+ showgrid="false"
+ inkscape:window-width="979"
+ inkscape:window-height="836"
+ inkscape:window-x="571"
+ inkscape:window-y="335"
+ inkscape:window-maximized="0"
+ fit-margin-top="5"
+ fit-margin-left="5"
+ fit-margin-right="5"
+ fit-margin-bottom="5" />
+ <metadata
+ id="metadata7">
+ <rdf:RDF>
+ <cc:Work
+ rdf:about="">
+ <dc:format>image/svg+xml</dc:format>
+ <dc:type
+ rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+ <dc:title></dc:title>
+ </cc:Work>
+ </rdf:RDF>
+ </metadata>
+ <g
+ inkscape:label="Layer 1"
+ inkscape:groupmode="layer"
+ id="layer1"
+ transform="translate(-28.441125,-185.60612)">
+ <flowRoot
+ xml:space="preserve"
+ id="flowRoot2985"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"><flowRegion
+ id="flowRegion2987"><rect
+ id="rect2989"
+ width="82.85714"
+ height="11.428572"
+ x="240"
+ y="492.36218" /></flowRegion><flowPara
+ id="flowPara2991"></flowPara></flowRoot> <g
+ id="g4433"
+ transform="translate(2,0)">
+ <text
+ sodipodi:linespacing="125%"
+ id="text2993"
+ y="-261.66608"
+ x="412.12299"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ xml:space="preserve"
+ transform="matrix(0,1,-1,0,0,0)"><tspan
+ y="-261.66608"
+ x="412.12299"
+ id="tspan2995"
+ sodipodi:role="line">synchronize_rcu()</tspan></text>
+ <g
+ id="g4417"
+ transform="matrix(0,1,-1,0,730.90257,222.4928)">
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+ d="m 97.580736,477.4048 183.140664,0"
+ id="path2997"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 96.752718,465.38398 0,22.62742"
+ id="path4397"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 281.54942,465.38397 0,22.62742"
+ id="path4397-5"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ </g>
+ </g>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.04738"
+ y="268.18076"
+ id="text4429"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431"
+ x="112.04738"
+ y="268.18076">WRITE_ONCE(a, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.04738"
+ y="439.13766"
+ id="text4441"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4443"
+ x="112.04738"
+ y="439.13766">WRITE_ONCE(b, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="255.60869"
+ y="309.29346"
+ id="text4445"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4447"
+ x="255.60869"
+ y="309.29346">r1 = READ_ONCE(a);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="255.14423"
+ y="520.61786"
+ id="text4449"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4451"
+ x="255.14423"
+ y="520.61786">WRITE_ONCE(c, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="384.71124"
+ id="text4453"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4455"
+ x="396.10254"
+ y="384.71124">r2 = READ_ONCE(b);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="582.13617"
+ id="text4457"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4459"
+ x="396.10254"
+ y="582.13617">r3 = READ_ONCE(c);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.08231"
+ y="213.91006"
+ id="text4461"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463"
+ x="112.08231"
+ y="213.91006">thread0()</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="252.34512"
+ y="213.91006"
+ id="text4461-6"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-0"
+ x="252.34512"
+ y="213.91006">thread1()</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.42557"
+ y="213.91006"
+ id="text4461-2"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-2"
+ x="396.42557"
+ y="213.91006">thread2()</tspan></text>
+ <rect
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="rect4495"
+ width="436.28488"
+ height="416.4859"
+ x="34.648232"
+ y="191.10612" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 183.14066,191.10612 0,417.193 -0.70711,0"
+ id="path4497"
+ inkscape:connector-curvature="0" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 325.13867,191.10612 0,417.193 -0.70711,0"
+ id="path4497-5"
+ inkscape:connector-curvature="0" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="111.75929"
+ y="251.53981"
+ id="text4429-8"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9"
+ x="111.75929"
+ y="251.53981">rcu_read_lock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="367.91556"
+ id="text4429-8-9"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4"
+ x="396.10254"
+ y="367.91556">rcu_read_lock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="597.40289"
+ id="text4429-8-9-3"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-4"
+ x="396.10254"
+ y="597.40289">rcu_read_unlock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="111.75929"
+ y="453.15311"
+ id="text4429-8-9-3-1"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-4-6"
+ x="111.75929"
+ y="453.15311">rcu_read_unlock();</tspan></text>
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 33.941125,227.87568 436.284885,0 0,0.7071"
+ id="path4608"
+ inkscape:connector-curvature="0" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="394.94427"
+ y="345.66351"
+ id="text4648"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650"
+ x="394.94427"
+ y="345.66351">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(36.441125,199.60612)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.11968"
+ y="475.77856"
+ id="text4648-4"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-4"
+ x="112.11968"
+ y="475.77856">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-7"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(-246.38346,329.72117)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <path
+ sodipodi:type="arc"
+ style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-7-7"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(-103.65246,202.90878)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="254.85066"
+ y="348.96619"
+ id="text4648-4-3"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-4-5"
+ x="254.85066"
+ y="348.96619">QS</tspan></text>
+ </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/RCUApplicability.svg b/Documentation/RCU/Design/Requirements/RCUApplicability.svg
new file mode 100644
index 000000000000..ebcbeee391ed
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/RCUApplicability.svg
@@ -0,0 +1,237 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Creator: fig2dev Version 3.2 Patchlevel 5d -->
+
+<!-- CreationDate: Tue Mar 4 18:34:25 2014 -->
+
+<!-- Magnification: 3.000 -->
+
+<svg
+ xmlns:dc="http://purl.org/dc/elements/1.1/"
+ xmlns:cc="http://creativecommons.org/ns#"
+ xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+ xmlns:svg="http://www.w3.org/2000/svg"
+ xmlns="http://www.w3.org/2000/svg"
+ xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+ xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+ width="1089.1382"
+ height="668.21368"
+ viewBox="-2121 -36 14554.634 8876.4061"
+ id="svg2"
+ version="1.1"
+ inkscape:version="0.48.3.1 r9886"
+ sodipodi:docname="RCUApplicability.svg">
+ <metadata
+ id="metadata40">
+ <rdf:RDF>
+ <cc:Work
+ rdf:about="">
+ <dc:format>image/svg+xml</dc:format>
+ <dc:type
+ rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+ <dc:title />
+ </cc:Work>
+ </rdf:RDF>
+ </metadata>
+ <defs
+ id="defs38" />
+ <sodipodi:namedview
+ pagecolor="#ffffff"
+ bordercolor="#666666"
+ borderopacity="1"
+ objecttolerance="10"
+ gridtolerance="10"
+ guidetolerance="10"
+ inkscape:pageopacity="0"
+ inkscape:pageshadow="2"
+ inkscape:window-width="849"
+ inkscape:window-height="639"
+ id="namedview36"
+ showgrid="false"
+ inkscape:zoom="0.51326165"
+ inkscape:cx="544.56912"
+ inkscape:cy="334.10686"
+ inkscape:window-x="149"
+ inkscape:window-y="448"
+ inkscape:window-maximized="0"
+ inkscape:current-layer="g4"
+ fit-margin-top="5"
+ fit-margin-left="5"
+ fit-margin-right="5"
+ fit-margin-bottom="5" />
+ <g
+ style="fill:none;stroke-width:0.025in"
+ id="g4"
+ transform="translate(-2043.6828,14.791398)">
+ <!-- Line: box -->
+ <rect
+ x="0"
+ y="0"
+ width="14400"
+ height="8775"
+ rx="0"
+ style="fill:#ffa1a1;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+ id="rect6" />
+ <!-- Line: box -->
+ <rect
+ x="1350"
+ y="0"
+ width="11700"
+ height="6075"
+ rx="0"
+ style="fill:#ffff00;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+ id="rect8" />
+ <!-- Line: box -->
+ <rect
+ x="2700"
+ y="0"
+ width="9000"
+ height="4275"
+ rx="0"
+ style="fill:#00ff00;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+ id="rect10" />
+ <!-- Line: box -->
+ <rect
+ x="4050"
+ y="0"
+ width="6300"
+ height="2475"
+ rx="0"
+ style="fill:#87cfff;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+ id="rect12" />
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="900"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text14"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3017">Read-Mostly, Stale &amp;</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="1350"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text16"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3019">Inconsistent Data OK</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="1800"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text18"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3021">(RCU Works Great!!!)</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="3825"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text20"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3023">(RCU Works Well)</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="3375"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text22"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3025">Read-Mostly, Need Consistent Data</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="5175"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text24"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3027">Read-Write, Need Consistent Data</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="6975"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text26"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ sodipodi:linespacing="125%">Update-Mostly, Need Consistent Data</text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="5625"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text28"
+ sodipodi:linespacing="125%"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ id="tspan3029">(RCU Might Be OK...)</tspan></text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="7875"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text30"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ sodipodi:linespacing="125%">(1) Provide Existence Guarantees For Update-Friendly Mechanisms</text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="8325"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text32"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ sodipodi:linespacing="125%">(2) Provide Wait-Free Read-Side Primitives for Real-Time Use)</text>
+ <!-- Text -->
+ <text
+ xml:space="preserve"
+ x="7200"
+ y="7425"
+ font-style="normal"
+ font-weight="normal"
+ font-size="324"
+ id="text34"
+ style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+ sodipodi:linespacing="125%">(RCU is Very Unlikely to be the Right Tool For The Job, But it Can:</text>
+ </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg b/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
new file mode 100644
index 000000000000..48cd1623d4d4
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
@@ -0,0 +1,639 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+ xmlns:dc="http://purl.org/dc/elements/1.1/"
+ xmlns:cc="http://creativecommons.org/ns#"
+ xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+ xmlns:svg="http://www.w3.org/2000/svg"
+ xmlns="http://www.w3.org/2000/svg"
+ xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+ xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+ width="735.25"
+ height="516.21875"
+ id="svg2"
+ version="1.1"
+ inkscape:version="0.48.3.1 r9886"
+ sodipodi:docname="ReadersPartitionGP1.svg">
+ <defs
+ id="defs4">
+ <marker
+ inkscape:stockid="Arrow2Lend"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lend"
+ style="overflow:visible">
+ <path
+ id="path3792"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ <marker
+ inkscape:stockid="Arrow2Lstart"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lstart"
+ style="overflow:visible">
+ <path
+ id="path3789"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(1.1,0,0,1.1,1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ <marker
+ inkscape:stockid="Arrow2Lstart"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lstart-4"
+ style="overflow:visible">
+ <path
+ id="path3789-9"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(1.1,0,0,1.1,1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ <marker
+ inkscape:stockid="Arrow2Lend"
+ orient="auto"
+ refY="0"
+ refX="0"
+ id="Arrow2Lend-4"
+ style="overflow:visible">
+ <path
+ id="path3792-4"
+ style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+ d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+ transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+ inkscape:connector-curvature="0" />
+ </marker>
+ </defs>
+ <sodipodi:namedview
+ id="base"
+ pagecolor="#ffffff"
+ bordercolor="#666666"
+ borderopacity="1.0"
+ inkscape:pageopacity="0.0"
+ inkscape:pageshadow="2"
+ inkscape:zoom="1.3670394"
+ inkscape:cx="367.26465"
+ inkscape:cy="258.46182"
+ inkscape:document-units="px"
+ inkscape:current-layer="g4433-6"
+ showgrid="false"
+ inkscape:window-width="1351"
+ inkscape:window-height="836"
+ inkscape:window-x="438"
+ inkscape:window-y="335"
+ inkscape:window-maximized="0"
+ fit-margin-top="5"
+ fit-margin-left="5"
+ fit-margin-right="5"
+ fit-margin-bottom="5" />
+ <metadata
+ id="metadata7">
+ <rdf:RDF>
+ <cc:Work
+ rdf:about="">
+ <dc:format>image/svg+xml</dc:format>
+ <dc:type
+ rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+ <dc:title />
+ </cc:Work>
+ </rdf:RDF>
+ </metadata>
+ <g
+ inkscape:label="Layer 1"
+ inkscape:groupmode="layer"
+ id="layer1"
+ transform="translate(-29.15625,-185.59375)">
+ <flowRoot
+ xml:space="preserve"
+ id="flowRoot2985"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"><flowRegion
+ id="flowRegion2987"><rect
+ id="rect2989"
+ width="82.85714"
+ height="11.428572"
+ x="240"
+ y="492.36218" /></flowRegion><flowPara
+ id="flowPara2991" /></flowRoot> <g
+ id="g4433"
+ transform="translate(2,-12)">
+ <text
+ sodipodi:linespacing="125%"
+ id="text2993"
+ y="-261.66608"
+ x="436.12299"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ xml:space="preserve"
+ transform="matrix(0,1,-1,0,0,0)"><tspan
+ y="-261.66608"
+ x="436.12299"
+ id="tspan2995"
+ sodipodi:role="line">synchronize_rcu()</tspan></text>
+ <g
+ id="g4417"
+ transform="matrix(0,1,-1,0,730.90257,222.4928)">
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+ d="M 97.580736,477.4048 327.57913,476.09759"
+ id="path2997"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 96.752718,465.38398 0,22.62742"
+ id="path4397"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 328.40703,465.38397 0,22.62742"
+ id="path4397-5"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ </g>
+ </g>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.04738"
+ y="268.18076"
+ id="text4429"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431"
+ x="112.04738"
+ y="268.18076">WRITE_ONCE(a, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.04738"
+ y="487.13766"
+ id="text4441"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4443"
+ x="112.04738"
+ y="487.13766">WRITE_ONCE(b, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="255.60869"
+ y="297.29346"
+ id="text4445"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4447"
+ x="255.60869"
+ y="297.29346">r1 = READ_ONCE(a);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="255.14423"
+ y="554.61786"
+ id="text4449"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4451"
+ x="255.14423"
+ y="554.61786">WRITE_ONCE(c, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="370.71124"
+ id="text4453"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4455"
+ x="396.10254"
+ y="370.71124">WRITE_ONCE(d, 1);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="572.13617"
+ id="text4457"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4459"
+ x="396.10254"
+ y="572.13617">r2 = READ_ONCE(c);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.08231"
+ y="213.91006"
+ id="text4461"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463"
+ x="112.08231"
+ y="213.91006">thread0()</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="252.34512"
+ y="213.91006"
+ id="text4461-6"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-0"
+ x="252.34512"
+ y="213.91006">thread1()</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.42557"
+ y="213.91006"
+ id="text4461-2"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-2"
+ x="396.42557"
+ y="213.91006">thread2()</tspan></text>
+ <rect
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="rect4495"
+ width="724.25244"
+ height="505.21201"
+ x="34.648232"
+ y="191.10612" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 183.14066,191.10612 0,504.24243"
+ id="path4497"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 325.13867,191.10612 0,504.24243"
+ id="path4497-5"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="111.75929"
+ y="251.53981"
+ id="text4429-8"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9"
+ x="111.75929"
+ y="251.53981">rcu_read_lock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="353.91556"
+ id="text4429-8-9"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4"
+ x="396.10254"
+ y="353.91556">rcu_read_lock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="396.10254"
+ y="587.40289"
+ id="text4429-8-9-3"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-4"
+ x="396.10254"
+ y="587.40289">rcu_read_unlock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="111.75929"
+ y="501.15311"
+ id="text4429-8-9-3-1"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-4-6"
+ x="111.75929"
+ y="501.15311">rcu_read_unlock();</tspan></text>
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 33.941125,227.87568 724.941765,0"
+ id="path4608"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="394.94427"
+ y="331.66351"
+ id="text4648"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650"
+ x="394.94427"
+ y="331.66351">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(36.441125,185.60612)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="112.11968"
+ y="523.77856"
+ id="text4648-4"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-4"
+ x="112.11968"
+ y="523.77856">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-7"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(-246.38346,377.72117)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <path
+ sodipodi:type="arc"
+ style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-7-7"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(-103.65246,190.90878)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="254.85066"
+ y="336.96619"
+ id="text4648-4-3"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-4-5"
+ x="254.85066"
+ y="336.96619">QS</tspan></text>
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 470.93311,190.39903 0,504.24243"
+ id="path4497-5-6"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ d="m 616.22755,190.38323 0,504.24243"
+ id="path4497-5-2"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <g
+ id="g4433-6"
+ transform="translate(288.0964,78.32827)">
+ <text
+ sodipodi:linespacing="125%"
+ id="text2993-7"
+ y="-261.66608"
+ x="440.12299"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ xml:space="preserve"
+ transform="matrix(0,1,-1,0,0,0)"><tspan
+ y="-261.66608"
+ x="440.12299"
+ id="tspan2995-1"
+ sodipodi:role="line">synchronize_rcu()</tspan></text>
+ <g
+ id="g4417-1"
+ transform="matrix(0,1,-1,0,730.90257,222.4928)">
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+ d="M 97.580736,477.4048 328.5624,477.07246"
+ id="path2997-2"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 96.752718,465.38398 0,22.62742"
+ id="path4397-3"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ <path
+ style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+ d="m 329.39039,465.38397 0,22.62742"
+ id="path4397-5-4"
+ inkscape:connector-curvature="0"
+ sodipodi:nodetypes="cc" />
+ </g>
+ </g>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="541.70508"
+ y="387.6217"
+ id="text4445-0"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4447-5"
+ x="541.70508"
+ y="387.6217">r3 = READ_ONCE(d);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="541.2406"
+ y="646.94611"
+ id="text4449-6"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4451-6"
+ x="541.2406"
+ y="646.94611">WRITE_ONCE(e, 1);</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-7-7-5"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(182.44393,281.23704)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="540.94702"
+ y="427.29443"
+ id="text4648-4-3-1"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-4-5-7"
+ x="540.94702"
+ y="427.29443">QS</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="686.27747"
+ y="461.83929"
+ id="text4453-7"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4455-1"
+ x="686.27747"
+ y="461.83929">r4 = READ_ONCE(b);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="686.27747"
+ y="669.26422"
+ id="text4457-9"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4459-2"
+ x="686.27747"
+ y="669.26422">r5 = READ_ONCE(e);</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="686.27747"
+ y="445.04358"
+ id="text4429-8-9-33"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-2"
+ x="686.27747"
+ y="445.04358">rcu_read_lock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="686.27747"
+ y="684.53094"
+ id="text4429-8-9-3-8"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4431-9-4-4-5"
+ x="686.27747"
+ y="684.53094">rcu_read_unlock();</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="685.11914"
+ y="422.79153"
+ id="text4648-9"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-7"
+ x="685.11914"
+ y="422.79153">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-8"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(326.61602,276.73415)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="397.85934"
+ y="609.59003"
+ id="text4648-5"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-77"
+ x="397.85934"
+ y="609.59003">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-80"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(39.356201,463.53264)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="256.75986"
+ y="586.99133"
+ id="text4648-5-2"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4650-77-7"
+ x="256.75986"
+ y="586.99133">QS</tspan></text>
+ <path
+ sodipodi:type="arc"
+ style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+ id="path4652-80-5"
+ sodipodi:cx="358.85669"
+ sodipodi:cy="142.87541"
+ sodipodi:rx="10.960155"
+ sodipodi:ry="10.253048"
+ d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+ transform="translate(-101.74328,440.93395)"
+ sodipodi:start="4.7135481"
+ sodipodi:end="10.994651"
+ sodipodi:open="true" />
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="546.22791"
+ y="213.91006"
+ id="text4461-2-5"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-2-6"
+ x="546.22791"
+ y="213.91006">thread3()</tspan></text>
+ <text
+ xml:space="preserve"
+ style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+ x="684.00067"
+ y="213.91006"
+ id="text4461-2-1"
+ sodipodi:linespacing="125%"><tspan
+ sodipodi:role="line"
+ id="tspan4463-2-0"
+ x="684.00067"
+ y="213.91006">thread4()</tspan></text>
+ </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
new file mode 100644
index 000000000000..95ef7b5de951
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -0,0 +1,2799 @@
+<!-- DO NOT HAND EDIT. -->
+<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
+<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
+ "http://www.w3.org/TR/html4/loose.dtd">
+ <html>
+ <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
+ <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
+
+<h1>A Tour Through RCU's Requirements</h1>
+
+<p>Copyright IBM Corporation, 2015</p>
+<p>Author: Paul E.&nbsp;McKenney</p>
+<p><i>The initial version of this document appeared in the
+<a href="http://lwn.net/">LWN</a> articles
+<a href="http://lwn.net/Articles/652156/">here</a>,
+<a href="http://lwn.net/Articles/652677/">here</a>, and
+<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
+
+<h2>Introduction</h2>
+
+<p>
+Read-copy update (RCU) is a synchronization mechanism that is often
+used as a replacement for reader-writer locking.
+RCU is unusual in that updaters do not block readers,
+which means that RCU's read-side primitives can be exceedingly fast
+and scalable.
+In addition, updaters can make useful forward progress concurrently
+with readers.
+However, all this concurrency between RCU readers and updaters does raise
+the question of exactly what RCU readers are doing, which in turn
+raises the question of exactly what RCU's requirements are.
+
+<p>
+This document therefore summarizes RCU's requirements, and can be thought
+of as an informal, high-level specification for RCU.
+It is important to understand that RCU's specification is primarily
+empirical in nature;
+in fact, I learned about many of these requirements the hard way.
+This situation might cause some consternation, however, not only
+has this learning process been a lot of fun, but it has also been
+a great privilege to work with so many people willing to apply
+technologies in interesting new ways.
+
+<p>
+All that aside, here are the categories of currently known RCU requirements:
+</p>
+
+<ol>
+<li> <a href="#Fundamental Requirements">
+ Fundamental Requirements</a>
+<li> <a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
+<li> <a href="#Parallelism Facts of Life">
+ Parallelism Facts of Life</a>
+<li> <a href="#Quality-of-Implementation Requirements">
+ Quality-of-Implementation Requirements</a>
+<li> <a href="#Linux Kernel Complications">
+ Linux Kernel Complications</a>
+<li> <a href="#Software-Engineering Requirements">
+ Software-Engineering Requirements</a>
+<li> <a href="#Other RCU Flavors">
+ Other RCU Flavors</a>
+<li> <a href="#Possible Future Changes">
+ Possible Future Changes</a>
+</ol>
+
+<p>
+This is followed by a <a href="#Summary">summary</a>,
+which is in turn followed by the inevitable
+<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.
+
+<h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>
+
+<p>
+RCU's fundamental requirements are the closest thing RCU has to hard
+mathematical requirements.
+These are:
+
+<ol>
+<li> <a href="#Grace-Period Guarantee">
+ Grace-Period Guarantee</a>
+<li> <a href="#Publish-Subscribe Guarantee">
+ Publish-Subscribe Guarantee</a>
+<li> <a href="#RCU Primitives Guaranteed to Execute Unconditionally">
+ RCU Primitives Guaranteed to Execute Unconditionally</a>
+<li> <a href="#Guaranteed Read-to-Write Upgrade">
+ Guaranteed Read-to-Write Upgrade</a>
+</ol>
+
+<h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>
+
+<p>
+RCU's grace-period guarantee is unusual in being premeditated:
+Jack Slingwine and I had this guarantee firmly in mind when we started
+work on RCU (then called &ldquo;rclock&rdquo;) in the early 1990s.
+That said, the past two decades of experience with RCU have produced
+a much more detailed understanding of this guarantee.
+
+<p>
+RCU's grace-period guarantee allows updaters to wait for the completion
+of all pre-existing RCU read-side critical sections.
+An RCU read-side critical section
+begins with the marker <tt>rcu_read_lock()</tt> and ends with
+the marker <tt>rcu_read_unlock()</tt>.
+These markers may be nested, and RCU treats a nested set as one
+big RCU read-side critical section.
+Production-quality implementations of <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> are extremely lightweight, and in
+fact have exactly zero overhead in Linux kernels built for production
+use with <tt>CONFIG_PREEMPT=n</tt>.
+
+<p>
+This guarantee allows ordering to be enforced with extremely low
+overhead to readers, for example:
+
+<blockquote>
+<pre>
+ 1 int x, y;
+ 2
+ 3 void thread0(void)
+ 4 {
+ 5 rcu_read_lock();
+ 6 r1 = READ_ONCE(x);
+ 7 r2 = READ_ONCE(y);
+ 8 rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13 WRITE_ONCE(x, 1);
+14 synchronize_rcu();
+15 WRITE_ONCE(y, 1);
+16 }
+</pre>
+</blockquote>
+
+<p>
+Because the <tt>synchronize_rcu()</tt> on line&nbsp;14 waits for
+all pre-existing readers, any instance of <tt>thread0()</tt> that
+loads a value of zero from <tt>x</tt> must complete before
+<tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
+also load a value of zero from <tt>y</tt>.
+Similarly, any instance of <tt>thread0()</tt> that loads a value of
+one from <tt>y</tt> must have started after the
+<tt>synchronize_rcu()</tt> started, and must therefore also load
+a value of one from <tt>x</tt>.
+Therefore, the outcome:
+<blockquote>
+<pre>
+(r1 == 0 &amp;&amp; r2 == 1)
+</pre>
+</blockquote>
+cannot happen.
+
+<p><a name="Quick Quiz 1"><b>Quick Quiz 1</b>:</a>
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+<br><a href="#qq1answer">Answer</a>
+
+<p>
+This scenario resembles one of the first uses of RCU in
+<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
+which managed a distributed lock manager's transition into
+a state suitable for handling recovery from node failure,
+more or less as follows:
+
+<blockquote>
+<pre>
+ 1 #define STATE_NORMAL 0
+ 2 #define STATE_WANT_RECOVERY 1
+ 3 #define STATE_RECOVERING 2
+ 4 #define STATE_WANT_NORMAL 3
+ 5
+ 6 int state = STATE_NORMAL;
+ 7
+ 8 void do_something_dlm(void)
+ 9 {
+10 int state_snap;
+11
+12 rcu_read_lock();
+13 state_snap = READ_ONCE(state);
+14 if (state_snap == STATE_NORMAL)
+15 do_something();
+16 else
+17 do_something_carefully();
+18 rcu_read_unlock();
+19 }
+20
+21 void start_recovery(void)
+22 {
+23 WRITE_ONCE(state, STATE_WANT_RECOVERY);
+24 synchronize_rcu();
+25 WRITE_ONCE(state, STATE_RECOVERING);
+26 recovery();
+27 WRITE_ONCE(state, STATE_WANT_NORMAL);
+28 synchronize_rcu();
+29 WRITE_ONCE(state, STATE_NORMAL);
+30 }
+</pre>
+</blockquote>
+
+<p>
+The RCU read-side critical section in <tt>do_something_dlm()</tt>
+works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
+to guarantee that <tt>do_something()</tt> never runs concurrently
+with <tt>recovery()</tt>, but with little or no synchronization
+overhead in <tt>do_something_dlm()</tt>.
+
+<p><a name="Quick Quiz 2"><b>Quick Quiz 2</b>:</a>
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+<br><a href="#qq2answer">Answer</a>
+
+<p>
+In order to avoid fatal problems such as deadlocks,
+an RCU read-side critical section must not contain calls to
+<tt>synchronize_rcu()</tt>.
+Similarly, an RCU read-side critical section must not
+contain anything that waits, directly or indirectly, on completion of
+an invocation of <tt>synchronize_rcu()</tt>.
+
+<p>
+Although RCU's grace-period guarantee is useful in and of itself, with
+<a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
+it would be good to be able to use RCU to coordinate read-side
+access to linked data structures.
+For this, the grace-period guarantee is not sufficient, as can
+be seen in function <tt>add_gp_buggy()</tt> below.
+We will look at the reader's code later, but in the meantime, just think of
+the reader as locklessly picking up the <tt>gp</tt> pointer,
+and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
+<tt>-&gt;a</tt> and <tt>-&gt;b</tt> fields.
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 p-&gt;a = a;
+12 p-&gt;b = a;
+13 gp = p; /* ORDERING BUG */
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The problem is that both the compiler and weakly ordered CPUs are within
+their rights to reorder this code as follows:
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy_optimized(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+<b>11 gp = p; /* ORDERING BUG */
+12 p-&gt;a = a;
+13 p-&gt;b = a;</b>
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+If an RCU reader fetches <tt>gp</tt> just after
+<tt>add_gp_buggy_optimized</tt> executes line&nbsp;11,
+it will see garbage in the <tt>-&gt;a</tt> and <tt>-&gt;b</tt>
+fields.
+And this is but one of many ways in which compiler and hardware optimizations
+could cause trouble.
+Therefore, we clearly need some way to prevent the compiler and the CPU from
+reordering in this manner, which brings us to the publish-subscribe
+guarantee discussed in the next section.
+
+<h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>
+
+<p>
+RCU's publish-subscribe guarantee allows data to be inserted
+into a linked data structure without disrupting RCU readers.
+The updater uses <tt>rcu_assign_pointer()</tt> to insert the
+new data, and readers use <tt>rcu_dereference()</tt> to
+access data, whether new or old.
+The following shows an example of insertion:
+
+<blockquote>
+<pre>
+ 1 bool add_gp(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 p-&gt;a = a;
+12 p-&gt;b = a;
+13 rcu_assign_pointer(gp, p);
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_assign_pointer()</tt> on line&nbsp;13 is conceptually
+equivalent to a simple assignment statement, but also guarantees
+that its assignment will
+happen after the two assignments in lines&nbsp;11 and&nbsp;12,
+similar to the C11 <tt>memory_order_release</tt> store operation.
+It also prevents any number of &ldquo;interesting&rdquo; compiler
+optimizations, for example, the use of <tt>gp</tt> as a scratch
+location immediately preceding the assignment.
+
+<p><a name="Quick Quiz 3"><b>Quick Quiz 3</b>:</a>
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+<br><a href="#qq3answer">Answer</a>
+
+<p>
+It is tempting to assume that the reader need not do anything special
+to control its accesses to the RCU-protected data,
+as shown in <tt>do_something_gp_buggy()</tt> below:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 p = gp; /* OPTIMIZATIONS GALORE!!! */
+ 5 if (p) {
+ 6 do_something(p-&gt;a, p-&gt;b);
+ 7 rcu_read_unlock();
+ 8 return true;
+ 9 }
+10 rcu_read_unlock();
+11 return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+However, this temptation must be resisted because there are a
+surprisingly large number of ways that the compiler
+(to say nothing of
+<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
+can trip this code up.
+For but one example, if the compiler were short of registers, it
+might choose to refetch from <tt>gp</tt> rather than keeping
+a separate copy in <tt>p</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy_optimized(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */
+<b> 5 do_something(gp-&gt;a, gp-&gt;b);</b>
+ 6 rcu_read_unlock();
+ 7 return true;
+ 8 }
+ 9 rcu_read_unlock();
+10 return false;
+11 }
+</pre>
+</blockquote>
+
+<p>
+If this function ran concurrently with a series of updates that
+replaced the current structure with a new one,
+the fetches of <tt>gp-&gt;a</tt>
+and <tt>gp-&gt;b</tt> might well come from two different structures,
+which could cause serious confusion.
+To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
+<tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 p = rcu_dereference(gp);
+ 5 if (p) {
+ 6 do_something(p-&gt;a, p-&gt;b);
+ 7 rcu_read_unlock();
+ 8 return true;
+ 9 }
+10 rcu_read_unlock();
+11 return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
+memory barriers in the Linux kernel.
+Should a
+<a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
+ever appear, then <tt>rcu_dereference()</tt> could be implemented
+as a <tt>memory_order_consume</tt> load.
+Regardless of the exact implementation, a pointer fetched by
+<tt>rcu_dereference()</tt> may not be used outside of the
+outermost RCU read-side critical section containing that
+<tt>rcu_dereference()</tt>, unless protection of
+the corresponding data element has been passed from RCU to some
+other synchronization mechanism, most commonly locking or
+<a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.
+
+<p>
+In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
+use <tt>rcu_dereference()</tt>, and these two RCU API elements
+work together to ensure that readers have a consistent view of
+newly added data elements.
+
+<p>
+Of course, it is also necessary to remove elements from RCU-protected
+data structures, for example, using the following process:
+
+<ol>
+<li> Remove the data element from the enclosing structure.
+<li> Wait for all pre-existing RCU read-side critical sections
+ to complete (because only pre-existing readers can possibly have
+ a reference to the newly removed data element).
+<li> At this point, only the updater has a reference to the
+ newly removed data element, so it can safely reclaim
+ the data element, for example, by passing it to <tt>kfree()</tt>.
+</ol>
+
+This process is implemented by <tt>remove_gp_synchronous()</tt>:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_synchronous(void)
+ 2 {
+ 3 struct foo *p;
+ 4
+ 5 spin_lock(&amp;gp_lock);
+ 6 p = rcu_access_pointer(gp);
+ 7 if (!p) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 rcu_assign_pointer(gp, NULL);
+12 spin_unlock(&amp;gp_lock);
+13 synchronize_rcu();
+14 kfree(p);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+This function is straightforward, with line&nbsp;13 waiting for a grace
+period before line&nbsp;14 frees the old data element.
+This waiting ensures that readers will reach line&nbsp;7 of
+<tt>do_something_gp()</tt> before the data element referenced by
+<tt>p</tt> is freed.
+The <tt>rcu_access_pointer()</tt> on line&nbsp;6 is similar to
+<tt>rcu_dereference()</tt>, except that:
+
+<ol>
+<li> The value returned by <tt>rcu_access_pointer()</tt>
+ cannot be dereferenced.
+ If you want to access the value pointed to as well as
+ the pointer itself, use <tt>rcu_dereference()</tt>
+ instead of <tt>rcu_access_pointer()</tt>.
+<li> The call to <tt>rcu_access_pointer()</tt> need not be
+ protected.
+ In contrast, <tt>rcu_dereference()</tt> must either be
+ within an RCU read-side critical section or in a code
+ segment where the pointer cannot change, for example, in
+ code protected by the corresponding update-side lock.
+</ol>
+
+<p><a name="Quick Quiz 4"><b>Quick Quiz 4</b>:</a>
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+<br><a href="#qq4answer">Answer</a>
+
+<p>
+This simple linked-data-structure scenario clearly demonstrates the need
+for RCU's stringent memory-ordering guarantees on systems with more than
+one CPU:
+
+<ol>
+<li> Each CPU that has an RCU read-side critical section that
+ begins before <tt>synchronize_rcu()</tt> starts is
+ guaranteed to execute a full memory barrier between the time
+ that the RCU read-side critical section ends and the time that
+ <tt>synchronize_rcu()</tt> returns.
+ Without this guarantee, a pre-existing RCU read-side critical section
+ might hold a reference to the newly removed <tt>struct foo</tt>
+ after the <tt>kfree()</tt> on line&nbsp;14 of
+ <tt>remove_gp_synchronous()</tt>.
+<li> Each CPU that has an RCU read-side critical section that ends
+ after <tt>synchronize_rcu()</tt> returns is guaranteed
+ to execute a full memory barrier between the time that
+ <tt>synchronize_rcu()</tt> begins and the time that the RCU
+ read-side critical section begins.
+ Without this guarantee, a later RCU read-side critical section
+ running after the <tt>kfree()</tt> on line&nbsp;14 of
+ <tt>remove_gp_synchronous()</tt> might
+ later run <tt>do_something_gp()</tt> and find the
+ newly deleted <tt>struct foo</tt>.
+<li> If the task invoking <tt>synchronize_rcu()</tt> remains
+ on a given CPU, then that CPU is guaranteed to execute a full
+ memory barrier sometime during the execution of
+ <tt>synchronize_rcu()</tt>.
+ This guarantee ensures that the <tt>kfree()</tt> on
+ line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+ execute after the removal on line&nbsp;11.
+<li> If the task invoking <tt>synchronize_rcu()</tt> migrates
+ among a group of CPUs during that invocation, then each of the
+ CPUs in that group is guaranteed to execute a full memory barrier
+ sometime during the execution of <tt>synchronize_rcu()</tt>.
+ This guarantee also ensures that the <tt>kfree()</tt> on
+ line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+ execute after the removal on
+ line&nbsp;11, but also in the case where the thread executing the
+ <tt>synchronize_rcu()</tt> migrates in the meantime.
+</ol>
+
+<p><a name="Quick Quiz 5"><b>Quick Quiz 5</b>:</a>
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+<br><a href="#qq5answer">Answer</a>
+
+<p><a name="Quick Quiz 6"><b>Quick Quiz 6</b>:</a>
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+<br><a href="#qq6answer">Answer</a>
+
+<p>
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
+
+<p>
+The common-case RCU primitives are unconditional.
+They are invoked, they do their job, and they return, with no possibility
+of error, and no need to retry.
+This is a key RCU design philosophy.
+
+<p>
+However, this philosophy is pragmatic rather than pigheaded.
+If someone comes up with a good justification for a particular conditional
+RCU primitive, it might well be implemented and added.
+After all, this guarantee was reverse-engineered, not premeditated.
+The unconditional nature of the RCU primitives was initially an
+accident of implementation, and later experience with synchronization
+primitives with conditional primitives caused me to elevate this
+accident to a guarantee.
+Therefore, the justification for adding a conditional primitive to
+RCU would need to be based on detailed and compelling use cases.
+
+<h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
+
+<p>
+As far as RCU is concerned, it is always possible to carry out an
+update within an RCU read-side critical section.
+For example, that RCU read-side critical section might search for
+a given data element, and then might acquire the update-side
+spinlock in order to update that element, all while remaining
+in that RCU read-side critical section.
+Of course, it is necessary to exit the RCU read-side critical section
+before invoking <tt>synchronize_rcu()</tt>, however, this
+inconvenience can be avoided through use of the
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
+described later in this document.
+
+<p><a name="Quick Quiz 7"><b>Quick Quiz 7</b>:</a>
+But how does the upgrade-to-write operation exclude other readers?
+<br><a href="#qq7answer">Answer</a>
+
+<p>
+This guarantee allows lookup code to be shared between read-side
+and update-side code, and was premeditated, appearing in the earliest
+DYNIX/ptx RCU documentation.
+
+<h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
+
+<p>
+RCU provides extremely lightweight readers, and its read-side guarantees,
+though quite useful, are correspondingly lightweight.
+It is therefore all too easy to assume that RCU is guaranteeing more
+than it really is.
+Of course, the list of things that RCU does not guarantee is infinitely
+long, however, the following sections list a few non-guarantees that
+have caused confusion.
+Except where otherwise noted, these non-guarantees were premeditated.
+
+<ol>
+<li> <a href="#Readers Impose Minimal Ordering">
+ Readers Impose Minimal Ordering</a>
+<li> <a href="#Readers Do Not Exclude Updaters">
+ Readers Do Not Exclude Updaters</a>
+<li> <a href="#Updaters Only Wait For Old Readers">
+ Updaters Only Wait For Old Readers</a>
+<li> <a href="#Grace Periods Don't Partition Read-Side Critical Sections">
+ Grace Periods Don't Partition Read-Side Critical Sections</a>
+<li> <a href="#Read-Side Critical Sections Don't Partition Grace Periods">
+ Read-Side Critical Sections Don't Partition Grace Periods</a>
+<li> <a href="#Disabling Preemption Does Not Block Grace Periods">
+ Disabling Preemption Does Not Block Grace Periods</a>
+</ol>
+
+<h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
+
+<p>
+Reader-side markers such as <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
+except through their interaction with the grace-period APIs such as
+<tt>synchronize_rcu()</tt>.
+To see this, consider the following pair of threads:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(x, 1);
+ 5 rcu_read_unlock();
+ 6 rcu_read_lock();
+ 7 WRITE_ONCE(y, 1);
+ 8 rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13 rcu_read_lock();
+14 r1 = READ_ONCE(y);
+15 rcu_read_unlock();
+16 rcu_read_lock();
+17 r2 = READ_ONCE(x);
+18 rcu_read_unlock();
+19 }
+</pre>
+</blockquote>
+
+<p>
+After <tt>thread0()</tt> and <tt>thread1()</tt> execute
+concurrently, it is quite possible to have
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0)
+</pre>
+</blockquote>
+
+(that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
+which would not be possible if <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> had much in the way of ordering
+properties.
+But they do not, so the CPU is within its rights
+to do significant reordering.
+This is by design: Any significant ordering constraints would slow down
+these fast-path APIs.
+
+<p><a name="Quick Quiz 8"><b>Quick Quiz 8</b>:</a>
+Can't the compiler also reorder this code?
+<br><a href="#qq8answer">Answer</a>
+
+<h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
+
+<p>
+Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
+exclude updates.
+All they do is to prevent grace periods from ending.
+The following example illustrates this:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 r1 = READ_ONCE(y);
+ 5 if (r1) {
+ 6 do_something_with_nonzero_x();
+ 7 r2 = READ_ONCE(x);
+ 8 WARN_ON(!r2); /* BUG!!! */
+ 9 }
+10 rcu_read_unlock();
+11 }
+12
+13 void thread1(void)
+14 {
+15 spin_lock(&amp;my_lock);
+16 WRITE_ONCE(x, 1);
+17 WRITE_ONCE(y, 1);
+18 spin_unlock(&amp;my_lock);
+19 }
+</pre>
+</blockquote>
+
+<p>
+If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
+excluded the <tt>thread1()</tt> function's update,
+the <tt>WARN_ON()</tt> could never fire.
+But the fact is that <tt>rcu_read_lock()</tt> does not exclude
+much of anything aside from subsequent grace periods, of which
+<tt>thread1()</tt> has none, so the
+<tt>WARN_ON()</tt> can and does fire.
+
+<h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
+
+<p>
+It might be tempting to assume that after <tt>synchronize_rcu()</tt>
+completes, there are no readers executing.
+This temptation must be avoided because
+new readers can start immediately after <tt>synchronize_rcu()</tt>
+starts, and <tt>synchronize_rcu()</tt> is under no
+obligation to wait for these new readers.
+
+<p><a name="Quick Quiz 9"><b>Quick Quiz 9</b>:</a>
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+<br><a href="#qq9answer">Answer</a>
+
+<h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
+Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
+
+<p>
+It is tempting to assume that if any part of one RCU read-side critical
+section precedes a given grace period, and if any part of another RCU
+read-side critical section follows that same grace period, then all of
+the first RCU read-side critical section must precede all of the second.
+However, this just isn't the case: A single grace period does not
+partition the set of RCU read-side critical sections.
+An example of this situation can be illustrated as follows, where
+<tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 rcu_read_lock();
+19 r2 = READ_ONCE(b);
+20 r3 = READ_ONCE(c);
+21 rcu_read_unlock();
+22 }
+</pre>
+</blockquote>
+
+<p>
+It turns out that the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0 &amp;&amp; r3 == 1)
+</pre>
+</blockquote>
+
+is entirely possible.
+The following figure show how this can happen, with each circled
+<tt>QS</tt> indicating the point at which RCU recorded a
+<i>quiescent state</i> for each thread, that is, a state in which
+RCU knows that the thread cannot be in the midst of an RCU read-side
+critical section that started before the current grace period:
+
+<p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
+
+<p>
+If it is necessary to partition RCU read-side critical sections in this
+manner, it is necessary to use two grace periods, where the first
+grace period is known to end before the second grace period starts:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 r2 = READ_ONCE(c);
+19 synchronize_rcu();
+20 WRITE_ONCE(d, 1);
+21 }
+22
+23 void thread3(void)
+24 {
+25 rcu_read_lock();
+26 r3 = READ_ONCE(b);
+27 r4 = READ_ONCE(d);
+28 rcu_read_unlock();
+29 }
+</pre>
+</blockquote>
+
+<p>
+Here, if <tt>(r1 == 1)</tt>, then
+<tt>thread0()</tt>'s write to <tt>b</tt> must happen
+before the end of <tt>thread1()</tt>'s grace period.
+If in addition <tt>(r4 == 1)</tt>, then
+<tt>thread3()</tt>'s read from <tt>b</tt> must happen
+after the beginning of <tt>thread2()</tt>'s grace period.
+If it is also the case that <tt>(r2 == 1)</tt>, then the
+end of <tt>thread1()</tt>'s grace period must precede the
+beginning of <tt>thread2()</tt>'s grace period.
+This mean that the two RCU read-side critical sections cannot overlap,
+guaranteeing that <tt>(r3 == 1)</tt>.
+As a result, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 0 &amp;&amp; r4 == 1)
+</pre>
+</blockquote>
+
+cannot happen.
+
+<p>
+This non-requirement was also non-premeditated, but became apparent
+when studying RCU's interaction with memory ordering.
+
+<h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
+Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
+
+<p>
+It is also tempting to assume that if an RCU read-side critical section
+happens between a pair of grace periods, then those grace periods cannot
+overlap.
+However, this temptation leads nowhere good, as can be illustrated by
+the following, with all variables initially zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 rcu_read_lock();
+19 WRITE_ONCE(d, 1);
+20 r2 = READ_ONCE(c);
+21 rcu_read_unlock();
+22 }
+23
+24 void thread3(void)
+25 {
+26 r3 = READ_ONCE(d);
+27 synchronize_rcu();
+28 WRITE_ONCE(e, 1);
+29 }
+30
+31 void thread4(void)
+32 {
+33 rcu_read_lock();
+34 r4 = READ_ONCE(b);
+35 r5 = READ_ONCE(e);
+36 rcu_read_unlock();
+37 }
+</pre>
+</blockquote>
+
+<p>
+In this case, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 1 &amp;&amp; r4 == 0 &amp&amp; r5 == 1)
+</pre>
+</blockquote>
+
+is entirely possible, as illustrated below:
+
+<p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
+
+<p>
+Again, an RCU read-side critical section can overlap almost all of a
+given grace period, just so long as it does not overlap the entire
+grace period.
+As a result, an RCU read-side critical section cannot partition a pair
+of RCU grace periods.
+
+<p><a name="Quick Quiz 10"><b>Quick Quiz 10</b>:</a>
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+<br><a href="#qq10answer">Answer</a>
+
+<h3><a name="Disabling Preemption Does Not Block Grace Periods">
+Disabling Preemption Does Not Block Grace Periods</a></h3>
+
+<p>
+There was a time when disabling preemption on any given CPU would block
+subsequent grace periods.
+However, this was an accident of implementation and is not a requirement.
+And in the current Linux-kernel implementation, disabling preemption
+on a given CPU in fact does not block grace periods, as Oleg Nesterov
+<a href="http://marc.info/?l=linux-kernel&amp;m=143431078202914&amp;w=2">demonstrated</a>.
+
+<p>
+If you need a preempt-disable region to block grace periods, you need to add
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
+as follows:
+
+<blockquote>
+<pre>
+ 1 preempt_disable();
+ 2 rcu_read_lock();
+ 3 do_something();
+ 4 rcu_read_unlock();
+ 5 preempt_enable();
+ 6
+ 7 /* Spinlocks implicitly disable preemption. */
+ 8 spin_lock(&amp;mylock);
+ 9 rcu_read_lock();
+10 do_something();
+11 rcu_read_unlock();
+12 spin_unlock(&amp;mylock);
+</pre>
+</blockquote>
+
+<p>
+In theory, you could enter the RCU read-side critical section first,
+but it is more efficient to keep the entire RCU read-side critical
+section contained in the preempt-disable region as shown above.
+Of course, RCU read-side critical sections that extend outside of
+preempt-disable regions will work correctly, but such critical sections
+can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
+more work.
+And no, this is <i>not</i> an invitation to enclose all of your RCU
+read-side critical sections within preempt-disable regions, because
+doing so would degrade real-time response.
+
+<p>
+This non-requirement appeared with preemptible RCU.
+If you need a grace period that waits on non-preemptible code regions, use
+<a href="#Sched Flavor">RCU-sched</a>.
+
+<h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
+
+<p>
+These parallelism facts of life are by no means specific to RCU, but
+the RCU implementation must abide by them.
+They therefore bear repeating:
+
+<ol>
+<li> Any CPU or task may be delayed at any time,
+ and any attempts to avoid these delays by disabling
+ preemption, interrupts, or whatever are completely futile.
+ This is most obvious in preemptible user-level
+ environments and in virtualized environments (where
+ a given guest OS's VCPUs can be preempted at any time by
+ the underlying hypervisor), but can also happen in bare-metal
+ environments due to ECC errors, NMIs, and other hardware
+ events.
+ Although a delay of more than about 20 seconds can result
+ in splats, the RCU implementation is obligated to use
+ algorithms that can tolerate extremely long delays, but where
+ &ldquo;extremely long&rdquo; is not long enough to allow
+ wrap-around when incrementing a 64-bit counter.
+<li> Both the compiler and the CPU can reorder memory accesses.
+ Where it matters, RCU must use compiler directives and
+ memory-barrier instructions to preserve ordering.
+<li> Conflicting writes to memory locations in any given cache line
+ will result in expensive cache misses.
+ Greater numbers of concurrent writes and more-frequent
+ concurrent writes will result in more dramatic slowdowns.
+ RCU is therefore obligated to use algorithms that have
+ sufficient locality to avoid significant performance and
+ scalability problems.
+<li> As a rough rule of thumb, only one CPU's worth of processing
+ may be carried out under the protection of any given exclusive
+ lock.
+ RCU must therefore use scalable locking designs.
+<li> Counters are finite, especially on 32-bit systems.
+ RCU's use of counters must therefore tolerate counter wrap,
+ or be designed such that counter wrap would take way more
+ time than a single system is likely to run.
+ An uptime of ten years is quite possible, a runtime
+ of a century much less so.
+ As an example of the latter, RCU's dyntick-idle nesting counter
+ allows 54 bits for interrupt nesting level (this counter
+ is 64 bits even on a 32-bit system).
+ Overflowing this counter requires 2<sup>54</sup>
+ half-interrupts on a given CPU without that CPU ever going idle.
+ If a half-interrupt happened every microsecond, it would take
+ 570 years of runtime to overflow this counter, which is currently
+ believed to be an acceptably long time.
+<li> Linux systems can have thousands of CPUs running a single
+ Linux kernel in a single shared-memory environment.
+ RCU must therefore pay close attention to high-end scalability.
+</ol>
+
+<p>
+This last parallelism fact of life means that RCU must pay special
+attention to the preceding facts of life.
+The idea that Linux might scale to systems with thousands of CPUs would
+have been met with some skepticism in the 1990s, but these requirements
+would have otherwise have been unsurprising, even in the early 1990s.
+
+<h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
+
+<p>
+These sections list quality-of-implementation requirements.
+Although an RCU implementation that ignores these requirements could
+still be used, it would likely be subject to limitations that would
+make it inappropriate for industrial-strength production use.
+Classes of quality-of-implementation requirements are as follows:
+
+<ol>
+<li> <a href="#Specialization">Specialization</a>
+<li> <a href="#Performance and Scalability">Performance and Scalability</a>
+<li> <a href="#Composability">Composability</a>
+<li> <a href="#Corner Cases">Corner Cases</a>
+</ol>
+
+<p>
+These classes is covered in the following sections.
+
+<h3><a name="Specialization">Specialization</a></h3>
+
+<p>
+RCU is and always has been intended primarily for read-mostly situations, as
+illustrated by the following figure.
+This means that RCU's read-side primitives are optimized, often at the
+expense of its update-side primitives.
+
+<p><img src="RCUApplicability.svg" alt="RCUApplicability.svg" width="70%"></p>
+
+<p>
+This focus on read-mostly situations means that RCU must interoperate
+with other synchronization primitives.
+For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
+examples discussed earlier use RCU to protect readers and locking to
+coordinate updaters.
+However, the need extends much farther, requiring that a variety of
+synchronization primitives be legal within RCU read-side critical sections,
+including spinlocks, sequence locks, atomic operations, reference
+counters, and memory barriers.
+
+<p><a name="Quick Quiz 11"><b>Quick Quiz 11</b>:</a>
+What about sleeping locks?
+<br><a href="#qq11answer">Answer</a>
+
+<p>
+It often comes as a surprise that many algorithms do not require a
+consistent view of data, but many can function in that mode,
+with network routing being the poster child.
+Internet routing algorithms take significant time to propagate
+updates, so that by the time an update arrives at a given system,
+that system has been sending network traffic the wrong way for
+a considerable length of time.
+Having a few threads continue to send traffic the wrong way for a
+few more milliseconds is clearly not a problem: In the worst case,
+TCP retransmissions will eventually get the data where it needs to go.
+In general, when tracking the state of the universe outside of the
+computer, some level of inconsistency must be tolerated due to
+speed-of-light delays if nothing else.
+
+<p>
+Furthermore, uncertainty about external state is inherent in many cases.
+For example, a pair of veternarians might use heartbeat to determine
+whether or not a given cat was alive.
+But how long should they wait after the last heartbeat to decide that
+the cat is in fact dead?
+Waiting less than 400 milliseconds makes no sense because this would
+mean that a relaxed cat would be considered to cycle between death
+and life more than 100 times per minute.
+Moreover, just as with human beings, a cat's heart might stop for
+some period of time, so the exact wait period is a judgment call.
+One of our pair of veternarians might wait 30 seconds before pronouncing
+the cat dead, while the other might insist on waiting a full minute.
+The two veternarians would then disagree on the state of the cat during
+the final 30 seconds of the minute following the last heartbeat, as
+fancifully illustrated below:
+
+<p><img src="2013-08-is-it-dead.png" alt="2013-08-is-it-dead.png" width="431"></p>
+
+<p>
+Interestingly enough, this same situation applies to hardware.
+When push comes to shove, how do we tell whether or not some
+external server has failed?
+We send messages to it periodically, and declare it failed if we
+don't receive a response within a given period of time.
+Policy decisions can usually tolerate short
+periods of inconsistency.
+The policy was decided some time ago, and is only now being put into
+effect, so a few milliseconds of delay is normally inconsequential.
+
+<p>
+However, there are algorithms that absolutely must see consistent data.
+For example, the translation between a user-level SystemV semaphore
+ID to the corresponding in-kernel data structure is protected by RCU,
+but it is absolutely forbidden to update a semaphore that has just been
+removed.
+In the Linux kernel, this need for consistency is accommodated by acquiring
+spinlocks located in the in-kernel data structure from within
+the RCU read-side critical section, and this is indicated by the
+green box in the figure above.
+Many other techniques may be used, and are in fact used within the
+Linux kernel.
+
+<p>
+In short, RCU is not required to maintain consistency, and other
+mechanisms may be used in concert with RCU when consistency is required.
+RCU's specialization allows it to do its job extremely well, and its
+ability to interoperate with other synchronization mechanisms allows
+the right mix of synchronization tools to be used for a given job.
+
+<h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
+
+<p>
+Energy efficiency is a critical component of performance today,
+and Linux-kernel RCU implementations must therefore avoid unnecessarily
+awakening idle CPUs.
+I cannot claim that this requirement was premeditated.
+In fact, I learned of it during a telephone conversation in which I
+was given &ldquo;frank and open&rdquo; feedback on the importance
+of energy efficiency in battery-powered systems and on specific
+energy-efficiency shortcomings of the Linux-kernel RCU implementation.
+In my experience, the battery-powered embedded community will consider
+any unnecessary wakeups to be extremely unfriendly acts.
+So much so that mere Linux-kernel-mailing-list posts are
+insufficient to vent their ire.
+
+<p>
+Memory consumption is not particularly important for in most
+situations, and has become decreasingly
+so as memory sizes have expanded and memory
+costs have plummeted.
+However, as I learned from Matt Mackall's
+<a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
+efforts, memory footprint is critically important on single-CPU systems with
+non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
+<a href="http://lkml.org/lkml/2009/1/14/449">tiny RCU</a>
+was born.
+Josh Triplett has since taken over the small-memory banner with his
+<a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
+project, which resulted in
+<a href="#Sleepable RCU">SRCU</a>
+becoming optional for those kernels not needing it.
+
+<p>
+The remaining performance requirements are, for the most part,
+unsurprising.
+For example, in keeping with RCU's read-side specialization,
+<tt>rcu_dereference()</tt> should have negligible overhead (for
+example, suppression of a few minor compiler optimizations).
+Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have exactly zero overhead.
+
+<p>
+In preemptible environments, in the case where the RCU read-side
+critical section was not preempted (as will be the case for the
+highest-priority real-time process), <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have minimal overhead.
+In particular, they should not contain atomic read-modify-write
+operations, memory-barrier instructions, preemption disabling,
+interrupt disabling, or backwards branches.
+However, in the case where the RCU read-side critical section was preempted,
+<tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
+This is why it is better to nest an RCU read-side critical section
+within a preempt-disable region than vice versa, at least in cases
+where that critical section is short enough to avoid unduly degrading
+real-time latencies.
+
+<p>
+The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
+optimized for throughput.
+It may therefore incur several milliseconds of latency in addition to
+the duration of the longest RCU read-side critical section.
+On the other hand, multiple concurrent invocations of
+<tt>synchronize_rcu()</tt> are required to use batching optimizations
+so that they can be satisfied by a single underlying grace-period-wait
+operation.
+For example, in the Linux kernel, it is not unusual for a single
+grace-period-wait operation to serve more than
+<a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
+of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
+overhead down to nearly zero.
+However, the grace-period optimization is also required to avoid
+measurable degradation of real-time scheduling and interrupt latencies.
+
+<p>
+In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
+latencies are unacceptable.
+In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
+instead, reducing the grace-period latency down to a few tens of
+microseconds on small systems, at least in cases where the RCU read-side
+critical sections are short.
+There are currently no special latency requirements for
+<tt>synchronize_rcu_expedited()</tt> on large systems, but,
+consistent with the empirical nature of the RCU specification,
+that is subject to change.
+However, there most definitely are scalability requirements:
+A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
+CPUs should at least make reasonable forward progress.
+In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
+is permitted to impose modest degradation of real-time latency
+on non-idle online CPUs.
+That said, it will likely be necessary to take further steps to reduce this
+degradation, hopefully to roughly that of a scheduling-clock interrupt.
+
+<p>
+There are a number of situations where even
+<tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
+latency is unacceptable.
+In these situations, the asynchronous <tt>call_rcu()</tt> can be
+used in place of <tt>synchronize_rcu()</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2 int a;
+ 3 int b;
+ 4 struct rcu_head rh;
+ 5 };
+ 6
+ 7 static void remove_gp_cb(struct rcu_head *rhp)
+ 8 {
+ 9 struct foo *p = container_of(rhp, struct foo, rh);
+10
+11 kfree(p);
+12 }
+13
+14 bool remove_gp_asynchronous(void)
+15 {
+16 struct foo *p;
+17
+18 spin_lock(&amp;gp_lock);
+19 p = rcu_dereference(gp);
+20 if (!p) {
+21 spin_unlock(&amp;gp_lock);
+22 return false;
+23 }
+24 rcu_assign_pointer(gp, NULL);
+25 call_rcu(&amp;p-&gt;rh, remove_gp_cb);
+26 spin_unlock(&amp;gp_lock);
+27 return true;
+28 }
+</pre>
+</blockquote>
+
+<p>
+A definition of <tt>struct foo</tt> is finally needed, and appears
+on lines&nbsp;1-5.
+The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
+on line&nbsp;25, and will be invoked after the end of a subsequent
+grace period.
+This gets the same effect as <tt>remove_gp_synchronous()</tt>,
+but without forcing the updater to wait for a grace period to elapse.
+The <tt>call_rcu()</tt> function may be used in a number of
+situations where neither <tt>synchronize_rcu()</tt> nor
+<tt>synchronize_rcu_expedited()</tt> would be legal,
+including within preempt-disable code, <tt>local_bh_disable()</tt> code,
+interrupt-disable code, and interrupt handlers.
+However, even <tt>call_rcu()</tt> is illegal within NMI handlers.
+The callback function (<tt>remove_gp_cb()</tt> in this case) will be
+executed within softirq (software interrupt) environment within the
+Linux kernel,
+either within a real softirq handler or under the protection
+of <tt>local_bh_disable()</tt>.
+In both the Linux kernel and in userspace, it is bad practice to
+write an RCU callback function that takes too long.
+Long-running operations should be relegated to separate threads or
+(in the Linux kernel) workqueues.
+
+<p><a name="Quick Quiz 12"><b>Quick Quiz 12</b>:</a>
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+<br><a href="#qq12answer">Answer</a>
+
+<p>
+However, all that <tt>remove_gp_cb()</tt> is doing is
+invoking <tt>kfree()</tt> on the data element.
+This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
+which allows &ldquo;fire and forget&rdquo; operation as shown below:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2 int a;
+ 3 int b;
+ 4 struct rcu_head rh;
+ 5 };
+ 6
+ 7 bool remove_gp_faf(void)
+ 8 {
+ 9 struct foo *p;
+10
+11 spin_lock(&amp;gp_lock);
+12 p = rcu_dereference(gp);
+13 if (!p) {
+14 spin_unlock(&amp;gp_lock);
+15 return false;
+16 }
+17 rcu_assign_pointer(gp, NULL);
+18 kfree_rcu(p, rh);
+19 spin_unlock(&amp;gp_lock);
+20 return true;
+21 }
+</pre>
+</blockquote>
+
+<p>
+Note that <tt>remove_gp_faf()</tt> simply invokes
+<tt>kfree_rcu()</tt> and proceeds, without any need to pay any
+further attention to the subsequent grace period and <tt>kfree()</tt>.
+It is permissible to invoke <tt>kfree_rcu()</tt> from the same
+environments as for <tt>call_rcu()</tt>.
+Interestingly enough, DYNIX/ptx had the equivalents of
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
+<tt>synchronize_rcu()</tt>.
+This was due to the fact that RCU was not heavily used within DYNIX/ptx,
+so the very few places that needed something like
+<tt>synchronize_rcu()</tt> simply open-coded it.
+
+<p><a name="Quick Quiz 13"><b>Quick Quiz 13</b>:</a>
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+<br><a href="#qq13answer">Answer</a>
+
+<p>
+But what if the updater must wait for the completion of code to be
+executed after the end of the grace period, but has other tasks
+that can be carried out in the meantime?
+The polling-style <tt>get_state_synchronize_rcu()</tt> and
+<tt>cond_synchronize_rcu()</tt> functions may be used for this
+purpose, as shown below:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_poll(void)
+ 2 {
+ 3 struct foo *p;
+ 4 unsigned long s;
+ 5
+ 6 spin_lock(&amp;gp_lock);
+ 7 p = rcu_access_pointer(gp);
+ 8 if (!p) {
+ 9 spin_unlock(&amp;gp_lock);
+10 return false;
+11 }
+12 rcu_assign_pointer(gp, NULL);
+13 spin_unlock(&amp;gp_lock);
+14 s = get_state_synchronize_rcu();
+15 do_something_while_waiting();
+16 cond_synchronize_rcu(s);
+17 kfree(p);
+18 return true;
+19 }
+</pre>
+</blockquote>
+
+<p>
+On line&nbsp;14, <tt>get_state_synchronize_rcu()</tt> obtains a
+&ldquo;cookie&rdquo; from RCU,
+then line&nbsp;15 carries out other tasks,
+and finally, line&nbsp;16 returns immediately if a grace period has
+elapsed in the meantime, but otherwise waits as required.
+The need for <tt>get_state_synchronize_rcu</tt> and
+<tt>cond_synchronize_rcu()</tt> has appeared quite recently,
+so it is too early to tell whether they will stand the test of time.
+
+<p>
+RCU thus provides a range of tools to allow updaters to strike the
+required tradeoff between latency, flexibility and CPU overhead.
+
+<h3><a name="Composability">Composability</a></h3>
+
+<p>
+Composability has received much attention in recent years, perhaps in part
+due to the collision of multicore hardware with object-oriented techniques
+designed in single-threaded environments for single-threaded use.
+And in theory, RCU read-side critical sections may be composed, and in
+fact may be nested arbitrarily deeply.
+In practice, as with all real-world implementations of composable
+constructs, there are limitations.
+
+<p>
+Implementations of RCU for which <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> generate no code, such as
+Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
+nested arbitrarily deeply.
+After all, there is no overhead.
+Except that if all these instances of <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> are visible to the compiler,
+compilation will eventually fail due to exhausting memory,
+mass storage, or user patience, whichever comes first.
+If the nesting is not visible to the compiler, as is the case with
+mutually recursive functions each in its own translation unit,
+stack overflow will result.
+If the nesting takes the form of loops, either the control variable
+will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
+Nevertheless, this class of RCU implementations is one
+of the most composable constructs in existence.
+
+<p>
+RCU implementations that explicitly track nesting depth
+are limited by the nesting-depth counter.
+For example, the Linux kernel's preemptible RCU limits nesting to
+<tt>INT_MAX</tt>.
+This should suffice for almost all practical purposes.
+That said, a consecutive pair of RCU read-side critical sections
+between which there is an operation that waits for a grace period
+cannot be enclosed in another RCU read-side critical section.
+This is because it is not legal to wait for a grace period within
+an RCU read-side critical section: To do so would result either
+in deadlock or
+in RCU implicitly splitting the enclosing RCU read-side critical
+section, neither of which is conducive to a long-lived and prosperous
+kernel.
+
+<p>
+In short, although RCU read-side critical sections are highly composable,
+care is required in some situations, just as is the case for any other
+composable synchronization mechanism.
+
+<h3><a name="Corner Cases">Corner Cases</a></h3>
+
+<p>
+A given RCU workload might have an endless and intense stream of
+RCU read-side critical sections, perhaps even so intense that there
+was never a point in time during which there was not at least one
+RCU read-side critical section in flight.
+RCU cannot allow this situation to block grace periods: As long as
+all the RCU read-side critical sections are finite, grace periods
+must also be finite.
+
+<p>
+That said, preemptible RCU implementations could potentially result
+in RCU read-side critical sections being preempted for long durations,
+which has the effect of creating a long-duration RCU read-side
+critical section.
+This situation can arise only in heavily loaded systems, but systems using
+real-time priorities are of course more vulnerable.
+Therefore, RCU priority boosting is provided to help deal with this
+case.
+That said, the exact requirements on RCU priority boosting will likely
+evolve as more experience accumulates.
+
+<p>
+Other workloads might have very high update rates.
+Although one can argue that such workloads should instead use
+something other than RCU, the fact remains that RCU must
+handle such workloads gracefully.
+This requirement is another factor driving batching of grace periods,
+but it is also the driving force behind the checks for large numbers
+of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
+Finally, high update rates should not delay RCU read-side critical
+sections, although some read-side delays can occur when using
+<tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
+of <tt>try_stop_cpus()</tt>.
+(In the future, <tt>synchronize_rcu_expedited()</tt> will be
+converted to use lighter-weight inter-processor interrupts (IPIs),
+but this will still disturb readers, though to a much smaller degree.)
+
+<p>
+Although all three of these corner cases were understood in the early
+1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
+in a tight loop
+in the early 2000s suddenly provided a much deeper appreciation of the
+high-update-rate corner case.
+This test also motivated addition of some RCU code to react to high update
+rates, for example, if a given CPU finds itself with more than 10,000
+RCU callbacks queued, it will cause RCU to take evasive action by
+more aggressively starting grace periods and more aggressively forcing
+completion of grace-period processing.
+This evasive action causes the grace period to complete more quickly,
+but at the cost of restricting RCU's batching optimizations, thus
+increasing the CPU overhead incurred by that grace period.
+
+<h2><a name="Software-Engineering Requirements">
+Software-Engineering Requirements</a></h2>
+
+<p>
+Between Murphy's Law and &ldquo;To err is human&rdquo;, it is necessary to
+guard against mishaps and misuse:
+
+<ol>
+<li> It is all too easy to forget to use <tt>rcu_read_lock()</tt>
+ everywhere that it is needed, so kernels built with
+ <tt>CONFIG_PROVE_RCU=y</tt> will spat if
+ <tt>rcu_dereference()</tt> is used outside of an
+ RCU read-side critical section.
+ Update-side code can use <tt>rcu_dereference_protected()</tt>,
+ which takes a
+ <a href="https://lwn.net/Articles/371986/">lockdep expression</a>
+ to indicate what is providing the protection.
+ If the indicated protection is not provided, a lockdep splat
+ is emitted.
+
+ <p>
+ Code shared between readers and updaters can use
+ <tt>rcu_dereference_check()</tt>, which also takes a
+ lockdep expression, and emits a lockdep splat if neither
+ <tt>rcu_read_lock()</tt> nor the indicated protection
+ is in place.
+ In addition, <tt>rcu_dereference_raw()</tt> is used in those
+ (hopefully rare) cases where the required protection cannot
+ be easily described.
+ Finally, <tt>rcu_read_lock_held()</tt> is provided to
+ allow a function to verify that it has been invoked within
+ an RCU read-side critical section.
+ I was made aware of this set of requirements shortly after Thomas
+ Gleixner audited a number of RCU uses.
+<li> A given function might wish to check for RCU-related preconditions
+ upon entry, before using any other RCU API.
+ The <tt>rcu_lockdep_assert()</tt> does this job,
+ asserting the expression in kernels having lockdep enabled
+ and doing nothing otherwise.
+<li> It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
+ and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
+ substituting a simple assignment.
+ To catch this sort of error, a given RCU-protected pointer may be
+ tagged with <tt>__rcu</tt>, after which running sparse
+ with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
+ about simple-assignment accesses to that pointer.
+ Arnd Bergmann made me aware of this requirement, and also
+ supplied the needed
+ <a href="http://lwn.net/Articles/376011/">patch series</a>.
+<li> Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
+ will splat if a data element is passed to <tt>call_rcu()</tt>
+ twice in a row, without a grace period in between.
+ (This error is similar to a double free.)
+ The corresponding <tt>rcu_head</tt> structures that are
+ dynamically allocated are automatically tracked, but
+ <tt>rcu_head</tt> structures allocated on the stack
+ must be initialized with <tt>init_rcu_head_on_stack()</tt>
+ and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
+ Similarly, statically allocated non-stack <tt>rcu_head</tt>
+ structures must be initialized with <tt>init_rcu_head()</tt>
+ and cleaned up with <tt>destroy_rcu_head()</tt>.
+ Mathieu Desnoyers made me aware of this requirement, and also
+ supplied the needed
+ <a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
+<li> An infinite loop in an RCU read-side critical section will
+ eventually trigger an RCU CPU stall warning splat.
+ However, RCU is not obligated to produce this splat
+ unless there is a grace period waiting on that particular
+ RCU read-side critical section.
+ This requirement made itself known in the early 1990s, pretty
+ much the first time that it was necessary to debug a CPU stall.
+<li> Although it would be very good to detect pointers leaking out
+ of RCU read-side critical sections, there is currently no
+ good way of doing this.
+ One complication is the need to distinguish between pointers
+ leaking and pointers that have been handed off from RCU to
+ some other synchronization mechanism, for example, reference
+ counting.
+<li> In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
+ information is provided via both debugfs and event tracing.
+<li> Open-coded use of <tt>rcu_assign_pointer()</tt> and
+ <tt>rcu_dereference()</tt> to create typical linked
+ data structures can be surprisingly error-prone.
+ Therefore, RCU-protected
+ <a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
+ and, more recently, RCU-protected
+ <a href="https://lwn.net/Articles/612100/">hash tables</a>
+ are available.
+ Many other special-purpose RCU-protected data structures are
+ available in the Linux kernel and the userspace RCU library.
+<li> Some linked structures are created at compile time, but still
+ require <tt>__rcu</tt> checking.
+ The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
+ purpose.
+<li> It is not necessary to use <tt>rcu_assign_pointer()</tt>
+ when creating linked structures that are to be published via
+ a single external pointer.
+ The <tt>RCU_INIT_POINTER()</tt> macro is provided for
+ this task and also for assigning <tt>NULL</tt> pointers
+ at runtime.
+</ol>
+
+<p>
+This not a hard-and-fast list: RCU's diagnostic capabilities will
+continue to be guided by the number and type of usage bugs found
+in real-world RCU usage.
+
+<h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
+
+<p>
+The Linux kernel provides an interesting environment for all kinds of
+software, including RCU.
+Some of the relevant points of interest are as follows:
+
+<ol>
+<li> <a href="#Configuration">Configuration</a>.
+<li> <a href="#Firmware Interface">Firmware Interface</a>.
+<li> <a href="#Early Boot">Early Boot</a>.
+<li> <a href="#Interrupts and NMIs">
+ Interrupts and non-maskable interrupts (NMIs)</a>.
+<li> <a href="#Loadable Modules">Loadable Modules</a>.
+<li> <a href="#Hotplug CPU">Hotplug CPU</a>.
+<li> <a href="#Scheduler and RCU">Scheduler and RCU</a>.
+<li> <a href="#Tracing and RCU">Tracing and RCU</a>.
+<li> <a href="#Energy Efficiency">Energy Efficiency</a>.
+<li> <a href="#Performance, Scalability, Response Time, and Reliability">
+ Performance, Scalability, Response Time, and Reliability</a>.
+</ol>
+
+<p>
+This list is probably incomplete, but it does give a feel for the
+most notable Linux-kernel complications.
+Each of the following sections covers one of the above topics.
+
+<h3><a name="Configuration">Configuration</a></h3>
+
+<p>
+RCU's goal is automatic configuration, so that almost nobody
+needs to worry about RCU's <tt>Kconfig</tt> options.
+And for almost all users, RCU does in fact work well
+&ldquo;out of the box.&rdquo;
+
+<p>
+However, there are specialized use cases that are handled by
+kernel boot parameters and <tt>Kconfig</tt> options.
+Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
+about new <tt>Kconfig</tt> options, which requires almost all of them
+be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
+
+<p>
+This all should be quite obvious, but the fact remains that
+Linus Torvalds recently had to
+<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
+me of this requirement.
+
+<h3><a name="Firmware Interface">Firmware Interface</a></h3>
+
+<p>
+In many cases, kernel obtains information about the system from the
+firmware, and sometimes things are lost in translation.
+Or the translation is accurate, but the original message is bogus.
+
+<p>
+For example, some systems' firmware overreports the number of CPUs,
+sometimes by a large factor.
+If RCU naively believed the firmware, as it used to do,
+it would create too many per-CPU kthreads.
+Although the resulting system will still run correctly, the extra
+kthreads needlessly consume memory and can cause confusion
+when they show up in <tt>ps</tt> listings.
+
+<p>
+RCU must therefore wait for a given CPU to actually come online before
+it can allow itself to believe that the CPU actually exists.
+The resulting &ldquo;ghost CPUs&rdquo; (which are never going to
+come online) cause a number of
+<a href="http://paulmck.livejournal.com/37494.html">interesting complications</a>.
+
+<h3><a name="Early Boot">Early Boot</a></h3>
+
+<p>
+The Linux kernel's boot sequence is an interesting process,
+and RCU is used early, even before <tt>rcu_init()</tt>
+is invoked.
+In fact, a number of RCU's primitives can be used as soon as the
+initial task's <tt>task_struct</tt> is available and the
+boot CPU's per-CPU variables are set up.
+The read-side primitives (<tt>rcu_read_lock()</tt>,
+<tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
+and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
+as will <tt>rcu_assign_pointer()</tt>.
+
+<p>
+Although <tt>call_rcu()</tt> may be invoked at any
+time during boot, callbacks are not guaranteed to be invoked until after
+the scheduler is fully up and running.
+This delay in callback invocation is due to the fact that RCU does not
+invoke callbacks until it is fully initialized, and this full initialization
+cannot occur until after the scheduler has initialized itself to the
+point where RCU can spawn and run its kthreads.
+In theory, it would be possible to invoke callbacks earlier,
+however, this is not a panacea because there would be severe restrictions
+on what operations those callbacks could invoke.
+
+<p>
+Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
+<a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
+(<a href="#Bottom-Half Flavor">discussed below</a>),
+and
+<a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
+will all operate normally
+during very early boot, the reason being that there is only one CPU
+and preemption is disabled.
+This means that the call <tt>synchronize_rcu()</tt> (or friends)
+itself is a quiescent
+state and thus a grace period, so the early-boot implementation can
+be a no-op.
+
+<p>
+Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
+continue to operate normally through the remainder of boot, courtesy
+of the fact that preemption is disabled across their RCU read-side
+critical sections and also courtesy of the fact that there is still
+only one CPU.
+However, once the scheduler starts initializing, preemption is enabled.
+There is still only a single CPU, but the fact that preemption is enabled
+means that the no-op implementation of <tt>synchronize_rcu()</tt> no
+longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
+Therefore, as soon as the scheduler starts initializing, the early-boot
+fastpath is disabled.
+This means that <tt>synchronize_rcu()</tt> switches to its runtime
+mode of operation where it posts callbacks, which in turn means that
+any call to <tt>synchronize_rcu()</tt> will block until the corresponding
+callback is invoked.
+Unfortunately, the callback cannot be invoked until RCU's runtime
+grace-period machinery is up and running, which cannot happen until
+the scheduler has initialized itself sufficiently to allow RCU's
+kthreads to be spawned.
+Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
+initialization can result in deadlock.
+
+<p><a name="Quick Quiz 14"><b>Quick Quiz 14</b>:</a>
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+<br><a href="#qq14answer">Answer</a>
+
+<p>
+I learned of these boot-time requirements as a result of a series of
+system hangs.
+
+<h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
+
+<p>
+The Linux kernel has interrupts, and RCU read-side critical sections are
+legal within interrupt handlers and within interrupt-disabled regions
+of code, as are invocations of <tt>call_rcu()</tt>.
+
+<p>
+Some Linux-kernel architectures can enter an interrupt handler from
+non-idle process context, and then just never leave it, instead stealthily
+transitioning back to process context.
+This trick is sometimes used to invoke system calls from inside the kernel.
+These &ldquo;half-interrupts&rdquo; mean that RCU has to be very careful
+about how it counts interrupt nesting levels.
+I learned of this requirement the hard way during a rewrite
+of RCU's dyntick-idle code.
+
+<p>
+The Linux kernel has non-maskable interrupts (NMIs), and
+RCU read-side critical sections are legal within NMI handlers.
+Thankfully, RCU update-side primitives, including
+<tt>call_rcu()</tt>, are prohibited within NMI handlers.
+
+<p>
+The name notwithstanding, some Linux-kernel architectures
+can have nested NMIs, which RCU must handle correctly.
+Andy Lutomirski
+<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
+with this requirement;
+he also kindly surprised me with
+<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
+that meets this requirement.
+
+<h3><a name="Loadable Modules">Loadable Modules</a></h3>
+
+<p>
+The Linux kernel has loadable modules, and these modules can
+also be unloaded.
+After a given module has been unloaded, any attempt to call
+one of its functions results in a segmentation fault.
+The module-unload functions must therefore cancel any
+delayed calls to loadable-module functions, for example,
+any outstanding <tt>mod_timer()</tt> must be dealt with
+via <tt>del_timer_sync()</tt> or similar.
+
+<p>
+Unfortunately, there is no way to cancel an RCU callback;
+once you invoke <tt>call_rcu()</tt>, the callback function is
+going to eventually be invoked, unless the system goes down first.
+Because it is normally considered socially irresponsible to crash the system
+in response to a module unload request, we need some other way
+to deal with in-flight RCU callbacks.
+
+<p>
+RCU therefore provides
+<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
+which waits until all in-flight RCU callbacks have been invoked.
+If a module uses <tt>call_rcu()</tt>, its exit function should therefore
+prevent any future invocation of <tt>call_rcu()</tt>, then invoke
+<tt>rcu_barrier()</tt>.
+In theory, the underlying module-unload code could invoke
+<tt>rcu_barrier()</tt> unconditionally, but in practice this would
+incur unacceptable latencies.
+
+<p>
+Nikita Danilov noted this requirement for an analogous filesystem-unmount
+situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
+The need for <tt>rcu_barrier()</tt> for module unloading became
+apparent later.
+
+<h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
+
+<p>
+The Linux kernel supports CPU hotplug, which means that CPUs
+can come and go.
+It is of course illegal to use any RCU API member from an offline CPU.
+This requirement was present from day one in DYNIX/ptx, but
+on the other hand, the Linux kernel's CPU-hotplug implementation
+is &ldquo;interesting.&rdquo;
+
+<p>
+The Linux-kernel CPU-hotplug implementation has notifiers that
+are used to allow the various kernel subsystems (including RCU)
+to respond appropriately to a given CPU-hotplug operation.
+Most RCU operations may be invoked from CPU-hotplug notifiers,
+including even normal synchronous grace-period operations
+such as <tt>synchronize_rcu()</tt>.
+However, expedited grace-period operations such as
+<tt>synchronize_rcu_expedited()</tt> are not supported,
+due to the fact that current implementations block CPU-hotplug
+operations, which could result in deadlock.
+
+<p>
+In addition, all-callback-wait operations such as
+<tt>rcu_barrier()</tt> are also not supported, due to the
+fact that there are phases of CPU-hotplug operations where
+the outgoing CPU's callbacks will not be invoked until after
+the CPU-hotplug operation ends, which could also result in deadlock.
+
+<h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
+
+<p>
+RCU depends on the scheduler, and the scheduler uses RCU to
+protect some of its data structures.
+This means the scheduler is forbidden from acquiring
+the runqueue locks and the priority-inheritance locks
+in the middle of an outermost RCU read-side critical section unless
+it also releases them before exiting that same
+RCU read-side critical section.
+This same prohibition also applies to any lock that is acquired
+while holding any lock to which this prohibition applies.
+Violating this rule results in deadlock.
+
+<p>
+For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
+implementation must be written carefully to avoid similar deadlocks.
+In particular, <tt>rcu_read_unlock()</tt> must tolerate an
+interrupt where the interrupt handler invokes both
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+This possibility requires <tt>rcu_read_unlock()</tt> to use
+negative nesting levels to avoid destructive recursion via
+interrupt handler's use of RCU.
+
+<p>
+This pair of mutual scheduler-RCU requirements came as a
+<a href="http://lwn.net/Articles/453002/">complete surprise</a>.
+
+<p>
+As noted above, RCU makes use of kthreads, and it is necessary to
+avoid excessive CPU-time accumulation by these kthreads.
+This requirement was no surprise, but RCU's violation of it
+when running context-switch-heavy workloads when built with
+<tt>CONFIG_NO_HZ_FULL=y</tt>
+<a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
+RCU has made good progress towards meeting this requirement, even
+for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
+but there is room for further improvement.
+
+<h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
+
+<p>
+It is possible to use tracing on RCU code, but tracing itself
+uses RCU.
+For this reason, <tt>rcu_dereference_raw_notrace()</tt>
+is provided for use by tracing, which avoids the destructive
+recursion that could otherwise ensue.
+This API is also used by virtualization in some architectures,
+where RCU readers execute in environments in which tracing
+cannot be used.
+The tracing folks both located the requirement and provided the
+needed fix, so this surprise requirement was relatively painless.
+
+<h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
+
+<p>
+Interrupting idle CPUs is considered socially unacceptable,
+especially by people with battery-powered embedded systems.
+RCU therefore conserves energy by detecting which CPUs are
+idle, including tracking CPUs that have been interrupted from idle.
+This is a large part of the energy-efficiency requirement,
+so I learned of this via an irate phone call.
+
+<p>
+Because RCU avoids interrupting idle CPUs, it is illegal to
+execute an RCU read-side critical section on an idle CPU.
+(Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
+if you try it.)
+The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
+event tracing is provided to work around this restriction.
+In addition, <tt>rcu_is_watching()</tt> may be used to
+test whether or not it is currently legal to run RCU read-side
+critical sections on this CPU.
+I learned of the need for diagnostics on the one hand
+and <tt>RCU_NONIDLE()</tt> on the other while inspecting
+idle-loop code.
+Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
+which is used quite heavily in the idle loop.
+
+<p>
+It is similarly socially unacceptable to interrupt an
+<tt>nohz_full</tt> CPU running in userspace.
+RCU must therefore track <tt>nohz_full</tt> userspace
+execution.
+And in
+<a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
+kernels, RCU must separately track idle CPUs on the one hand and
+CPUs that are either idle or executing in userspace on the other.
+In both cases, RCU must be able to sample state at two points in
+time, and be able to determine whether or not some other CPU spent
+any time idle and/or executing in userspace.
+
+<p>
+These energy-efficiency requirements have proven quite difficult to
+understand and to meet, for example, there have been more than five
+clean-sheet rewrites of RCU's energy-efficiency code, the last of
+which was finally able to demonstrate
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
+As noted earlier,
+I learned of many of these requirements via angry phone calls:
+Flaming me on the Linux-kernel mailing list was apparently not
+sufficient to fully vent their ire at RCU's energy-efficiency bugs!
+
+<h3><a name="Performance, Scalability, Response Time, and Reliability">
+Performance, Scalability, Response Time, and Reliability</a></h3>
+
+<p>
+Expanding on the
+<a href="#Performance and Scalability">earlier discussion</a>,
+RCU is used heavily by hot code paths in performance-critical
+portions of the Linux kernel's networking, security, virtualization,
+and scheduling code paths.
+RCU must therefore use efficient implementations, especially in its
+read-side primitives.
+To that end, it would be good if preemptible RCU's implementation
+of <tt>rcu_read_lock()</tt> could be inlined, however, doing
+this requires resolving <tt>#include</tt> issues with the
+<tt>task_struct</tt> structure.
+
+<p>
+The Linux kernel supports hardware configurations with up to
+4096 CPUs, which means that RCU must be extremely scalable.
+Algorithms that involve frequent acquisitions of global locks or
+frequent atomic operations on global variables simply cannot be
+tolerated within the RCU implementation.
+RCU therefore makes heavy use of a combining tree based on the
+<tt>rcu_node</tt> structure.
+RCU is required to tolerate all CPUs continuously invoking any
+combination of RCU's runtime primitives with minimal per-operation
+overhead.
+In fact, in many cases, increasing load must <i>decrease</i> the
+per-operation overhead, witness the batching optimizations for
+<tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
+<tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
+As a general rule, RCU must cheerfully accept whatever the
+rest of the Linux kernel decides to throw at it.
+
+<p>
+The Linux kernel is used for real-time workloads, especially
+in conjunction with the
+<a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
+The real-time-latency response requirements are such that the
+traditional approach of disabling preemption across RCU
+read-side critical sections is inappropriate.
+Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
+use an RCU implementation that allows RCU read-side critical
+sections to be preempted.
+This requirement made its presence known after users made it
+clear that an earlier
+<a href="http://lwn.net/Articles/107930/">real-time patch</a>
+did not meet their needs, in conjunction with some
+<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>
+encountered by a very early version of the -rt patchset.
+
+<p>
+In addition, RCU must make do with a sub-100-microsecond real-time latency
+budget.
+In fact, on smaller systems with the -rt patchset, the Linux kernel
+provides sub-20-microsecond real-time latencies for the whole kernel,
+including RCU.
+RCU's scalability and latency must therefore be sufficient for
+these sorts of configurations.
+To my surprise, the sub-100-microsecond real-time latency budget
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
+applies to even the largest systems [PDF]</a>,
+up to and including systems with 4096 CPUs.
+This real-time requirement motivated the grace-period kthread, which
+also simplified handling of a number of race conditions.
+
+<p>
+Finally, RCU's status as a synchronization primitive means that
+any RCU failure can result in arbitrary memory corruption that can be
+extremely difficult to debug.
+This means that RCU must be extremely reliable, which in
+practice also means that RCU must have an aggressive stress-test
+suite.
+This stress-test suite is called <tt>rcutorture</tt>.
+
+<p>
+Although the need for <tt>rcutorture</tt> was no surprise,
+the current immense popularity of the Linux kernel is posing
+interesting&mdash;and perhaps unprecedented&mdash;validation
+challenges.
+To see this, keep in mind that there are well over one billion
+instances of the Linux kernel running today, given Android
+smartphones, Linux-powered televisions, and servers.
+This number can be expected to increase sharply with the advent of
+the celebrated Internet of Things.
+
+<p>
+Suppose that RCU contains a race condition that manifests on average
+once per million years of runtime.
+This bug will be occurring about three times per <i>day</i> across
+the installed base.
+RCU could simply hide behind hardware error rates, given that no one
+should really expect their smartphone to last for a million years.
+However, anyone taking too much comfort from this thought should
+consider the fact that in most jurisdictions, a successful multi-year
+test of a given mechanism, which might include a Linux kernel,
+suffices for a number of types of safety-critical certifications.
+In fact, rumor has it that the Linux kernel is already being used
+in production for safety-critical applications.
+I don't know about you, but I would feel quite bad if a bug in RCU
+killed someone.
+Which might explain my recent focus on validation and verification.
+
+<h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
+
+<p>
+One of the more surprising things about RCU is that there are now
+no fewer than five <i>flavors</i>, or API families.
+In addition, the primary flavor that has been the sole focus up to
+this point has two different implementations, non-preemptible and
+preemptible.
+The other four flavors are listed below, with requirements for each
+described in a separate section.
+
+<ol>
+<li> <a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
+<li> <a href="#Sched Flavor">Sched Flavor</a>
+<li> <a href="#Sleepable RCU">Sleepable RCU</a>
+<li> <a href="#Tasks RCU">Tasks RCU</a>
+</ol>
+
+<h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>
+
+<p>
+The softirq-disable (AKA &ldquo;bottom-half&rdquo;,
+hence the &ldquo;_bh&rdquo; abbreviations)
+flavor of RCU, or <i>RCU-bh</i>, was developed by
+Dipankar Sarma to provide a flavor of RCU that could withstand the
+network-based denial-of-service attacks researched by Robert
+Olsson.
+These attacks placed so much networking load on the system
+that some of the CPUs never exited softirq execution,
+which in turn prevented those CPUs from ever executing a context switch,
+which, in the RCU implementation of that time, prevented grace periods
+from ever ending.
+The result was an out-of-memory condition and a system hang.
+
+<p>
+The solution was the creation of RCU-bh, which does
+<tt>local_bh_disable()</tt>
+across its read-side critical sections, and which uses the transition
+from one type of softirq processing to another as a quiescent state
+in addition to context switch, idle, user mode, and offline.
+This means that RCU-bh grace periods can complete even when some of
+the CPUs execute in softirq indefinitely, thus allowing algorithms
+based on RCU-bh to withstand network-based denial-of-service attacks.
+
+<p>
+Because
+<tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
+disable and re-enable softirq handlers, any attempt to start a softirq
+handlers during the
+RCU-bh read-side critical section will be deferred.
+In this case, <tt>rcu_read_unlock_bh()</tt>
+will invoke softirq processing, which can take considerable time.
+One can of course argue that this softirq overhead should be associated
+with the code following the RCU-bh read-side critical section rather
+than <tt>rcu_read_unlock_bh()</tt>, but the fact
+is that most profiling tools cannot be expected to make this sort
+of fine distinction.
+For example, suppose that a three-millisecond-long RCU-bh read-side
+critical section executes during a time of heavy networking load.
+There will very likely be an attempt to invoke at least one softirq
+handler during that three milliseconds, but any such invocation will
+be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
+This can of course make it appear at first glance as if
+<tt>rcu_read_unlock_bh()</tt> was executing very slowly.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
+includes
+<tt>rcu_read_lock_bh()</tt>,
+<tt>rcu_read_unlock_bh()</tt>,
+<tt>rcu_dereference_bh()</tt>,
+<tt>rcu_dereference_bh_check()</tt>,
+<tt>synchronize_rcu_bh()</tt>,
+<tt>synchronize_rcu_bh_expedited()</tt>,
+<tt>call_rcu_bh()</tt>,
+<tt>rcu_barrier_bh()</tt>, and
+<tt>rcu_read_lock_bh_held()</tt>.
+
+<h3><a name="Sched Flavor">Sched Flavor</a></h3>
+
+<p>
+Before preemptible RCU, waiting for an RCU grace period had the
+side effect of also waiting for all pre-existing interrupt
+and NMI handlers.
+However, there are legitimate preemptible-RCU implementations that
+do not have this property, given that any point in the code outside
+of an RCU read-side critical section can be a quiescent state.
+Therefore, <i>RCU-sched</i> was created, which follows &ldquo;classic&rdquo;
+RCU in that an RCU-sched grace period waits for for pre-existing
+interrupt and NMI handlers.
+In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
+APIs have identical implementations, while kernels built with
+<tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
+
+<p>
+Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
+<tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
+disable and re-enable preemption, respectively.
+This means that if there was a preemption attempt during the
+RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
+will enter the scheduler, with all the latency and overhead entailed.
+Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
+as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
+However, the highest-priority task won't be preempted, so that task
+will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
+includes
+<tt>rcu_read_lock_sched()</tt>,
+<tt>rcu_read_unlock_sched()</tt>,
+<tt>rcu_read_lock_sched_notrace()</tt>,
+<tt>rcu_read_unlock_sched_notrace()</tt>,
+<tt>rcu_dereference_sched()</tt>,
+<tt>rcu_dereference_sched_check()</tt>,
+<tt>synchronize_sched()</tt>,
+<tt>synchronize_rcu_sched_expedited()</tt>,
+<tt>call_rcu_sched()</tt>,
+<tt>rcu_barrier_sched()</tt>, and
+<tt>rcu_read_lock_sched_held()</tt>.
+However, anything that disables preemption also marks an RCU-sched
+read-side critical section, including
+<tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
+<tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
+and so on.
+
+<h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
+
+<p>
+For well over a decade, someone saying &ldquo;I need to block within
+an RCU read-side critical section&rdquo; was a reliable indication
+that this someone did not understand RCU.
+After all, if you are always blocking in an RCU read-side critical
+section, you can probably afford to use a higher-overhead synchronization
+mechanism.
+However, that changed with the advent of the Linux kernel's notifiers,
+whose RCU read-side critical
+sections almost never sleep, but sometimes need to.
+This resulted in the introduction of
+<a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
+or <i>SRCU</i>.
+
+<p>
+SRCU allows different domains to be defined, with each such domain
+defined by an instance of an <tt>srcu_struct</tt> structure.
+A pointer to this structure must be passed in to each SRCU function,
+for example, <tt>synchronize_srcu(&amp;ss)</tt>, where
+<tt>ss</tt> is the <tt>srcu_struct</tt> structure.
+The key benefit of these domains is that a slow SRCU reader in one
+domain does not delay an SRCU grace period in some other domain.
+That said, one consequence of these domains is that read-side code
+must pass a &ldquo;cookie&rdquo; from <tt>srcu_read_lock()</tt>
+to <tt>srcu_read_unlock()</tt>, for example, as follows:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+As noted above, it is legal to block within SRCU read-side critical sections,
+however, with great power comes great responsibility.
+If you block forever in one of a given domain's SRCU read-side critical
+sections, then that domain's grace periods will also be blocked forever.
+Of course, one good way to block forever is to deadlock, which can
+happen if any operation in a given domain's SRCU read-side critical
+section can block waiting, either directly or indirectly, for that domain's
+grace period to elapse.
+For example, this results in a self-deadlock:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 synchronize_srcu(&amp;ss);
+ 6 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+However, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
+deadlock would still be possible.
+Furthermore, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
+and if an <tt>ss1</tt>-domain SRCU read-side critical section
+acquired another mutex that was held across as <tt>ss</tt>-domain
+<tt>synchronize_srcu()</tt>,
+deadlock would again be possible.
+Such a deadlock cycle could extend across an arbitrarily large number
+of different SRCU domains.
+Again, with great power comes great responsibility.
+
+<p>
+Unlike the other RCU flavors, SRCU read-side critical sections can
+run on idle and even offline CPUs.
+This ability requires that <tt>srcu_read_lock()</tt> and
+<tt>srcu_read_unlock()</tt> contain memory barriers, which means
+that SRCU readers will run a bit slower than would RCU readers.
+It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
+API, which, in combination with <tt>srcu_read_unlock()</tt>,
+guarantees a full memory barrier.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
+includes
+<tt>srcu_read_lock()</tt>,
+<tt>srcu_read_unlock()</tt>,
+<tt>srcu_dereference()</tt>,
+<tt>srcu_dereference_check()</tt>,
+<tt>synchronize_srcu()</tt>,
+<tt>synchronize_srcu_expedited()</tt>,
+<tt>call_srcu()</tt>,
+<tt>srcu_barrier()</tt>, and
+<tt>srcu_read_lock_held()</tt>.
+It also includes
+<tt>DEFINE_SRCU()</tt>,
+<tt>DEFINE_STATIC_SRCU()</tt>, and
+<tt>init_srcu_struct()</tt>
+APIs for defining and initializing <tt>srcu_struct</tt> structures.
+
+<h3><a name="Tasks RCU">Tasks RCU</a></h3>
+
+<p>
+Some forms of tracing use &ldquo;tramopolines&rdquo; to handle the
+binary rewriting required to install different types of probes.
+It would be good to be able to free old trampolines, which sounds
+like a job for some form of RCU.
+However, because it is necessary to be able to install a trace
+anywhere in the code, it is not possible to use read-side markers
+such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+In addition, it does not work to have these markers in the trampoline
+itself, because there would need to be instructions following
+<tt>rcu_read_unlock()</tt>.
+Although <tt>synchronize_rcu()</tt> would guarantee that execution
+reached the <tt>rcu_read_unlock()</tt>, it would not be able to
+guarantee that execution had completely left the trampoline.
+
+<p>
+The solution, in the form of
+<a href="http://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
+is to have implicit
+read-side critical sections that are delimited by voluntary context
+switches, that is, calls to <tt>schedule()</tt>,
+<tt>cond_resched_rcu_qs()</tt>, and
+<tt>synchronize_rcu_tasks()</tt>.
+In addition, transitions to and from userspace execution also delimit
+tasks-RCU read-side critical sections.
+
+<p>
+The tasks-RCU API is quite compact, consisting only of
+<tt>call_rcu_tasks()</tt>,
+<tt>synchronize_rcu_tasks()</tt>, and
+<tt>rcu_barrier_tasks()</tt>.
+
+<h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
+
+<p>
+One of the tricks that RCU uses to attain update-side scalability is
+to increase grace-period latency with increasing numbers of CPUs.
+If this becomes a serious problem, it will be necessary to rework the
+grace-period state machine so as to avoid the need for the additional
+latency.
+
+<p>
+Expedited grace periods scan the CPUs, so their latency and overhead
+increases with increasing numbers of CPUs.
+If this becomes a serious problem on large systems, it will be necessary
+to do some redesign to avoid this scalability problem.
+
+<p>
+RCU disables CPU hotplug in a few places, perhaps most notably in the
+expedited grace-period and <tt>rcu_barrier()</tt> operations.
+If there is a strong reason to use expedited grace periods in CPU-hotplug
+notifiers, it will be necessary to avoid disabling CPU hotplug.
+This would introduce some complexity, so there had better be a <i>very</i>
+good reason.
+
+<p>
+The tradeoff between grace-period latency on the one hand and interruptions
+of other CPUs on the other hand may need to be re-examined.
+The desire is of course for zero grace-period latency as well as zero
+interprocessor interrupts undertaken during an expedited grace period
+operation.
+While this ideal is unlikely to be achievable, it is quite possible that
+further improvements can be made.
+
+<p>
+The multiprocessor implementations of RCU use a combining tree that
+groups CPUs so as to reduce lock contention and increase cache locality.
+However, this combining tree does not spread its memory across NUMA
+nodes nor does it align the CPU groups with hardware features such
+as sockets or cores.
+Such spreading and alignment is currently believed to be unnecessary
+because the hotpath read-side primitives do not access the combining
+tree, nor does <tt>call_rcu()</tt> in the common case.
+If you believe that your architecture needs such spreading and alignment,
+then your architecture should also benefit from the
+<tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
+to the number of CPUs in a socket, NUMA node, or whatever.
+If the number of CPUs is too large, use a fraction of the number of
+CPUs.
+If the number of CPUs is a large prime number, well, that certainly
+is an &ldquo;interesting&rdquo; architectural choice!
+More flexible arrangements might be considered, but only if
+<tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
+if the inadequacy has been demonstrated by a carefully run and
+realistic system-level workload.
+
+<p>
+Please note that arrangements that require RCU to remap CPU numbers will
+require extremely good demonstration of need and full exploration of
+alternatives.
+
+<p>
+There is an embarrassingly large number of flavors of RCU, and this
+number has been increasing over time.
+Perhaps it will be possible to combine some at some future date.
+
+<p>
+RCU's various kthreads are reasonably recent additions.
+It is quite likely that adjustments will be required to more gracefully
+handle extreme loads.
+It might also be necessary to be able to relate CPU utilization by
+RCU's kthreads and softirq handlers to the code that instigated this
+CPU utilization.
+For example, RCU callback overhead might be charged back to the
+originating <tt>call_rcu()</tt> instance, though probably not
+in production kernels.
+
+<h2><a name="Summary">Summary</a></h2>
+
+<p>
+This document has presented more than two decade's worth of RCU
+requirements.
+Given that the requirements keep changing, this will not be the last
+word on this subject, but at least it serves to get an important
+subset of the requirements set forth.
+
+<h2><a name="Acknowledgments">Acknowledgments</a></h2>
+
+I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
+Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
+Andy Lutomirski for their help in rendering
+this article human readable, and to Michelle Rankin for her support
+of this effort.
+Other contributions are acknowledged in the Linux kernel's git archive.
+The cartoon is copyright (c) 2013 by Melissa Broussard,
+and is provided
+under the terms of the Creative Commons Attribution-Share Alike 3.0
+United States license.
+
+<h3><a name="Answers to Quick Quizzes">
+Answers to Quick Quizzes</a></h3>
+
+<a name="qq1answer"></a>
+<p><b>Quick Quiz 1</b>:
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+
+
+</p><p><b>Answer</b>:
+First, if updaters do not wish to be blocked by readers, they can use
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
+be discussed later.
+Second, even when using <tt>synchronize_rcu()</tt>, the other
+update-side code does run concurrently with readers, whether pre-existing
+or not.
+
+
+</p><p><a href="#Quick%20Quiz%201"><b>Back to Quick Quiz 1</b>.</a>
+
+<a name="qq2answer"></a>
+<p><b>Quick Quiz 2</b>:
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+
+
+</p><p><b>Answer</b>:
+Without that extra grace period, memory reordering could result in
+<tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
+concurrently with the last bits of <tt>recovery()</tt>.
+
+
+</p><p><a href="#Quick%20Quiz%202"><b>Back to Quick Quiz 2</b>.</a>
+
+<a name="qq3answer"></a>
+<p><b>Quick Quiz 3</b>:
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+
+
+</p><p><b>Answer</b>:
+No, it cannot.
+The readers cannot see either of these two fields until
+the assignment to <tt>gp</tt>, by which time both fields are
+fully initialized.
+So reordering the assignments
+to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt> cannot possibly
+cause any problems.
+
+
+</p><p><a href="#Quick%20Quiz%203"><b>Back to Quick Quiz 3</b>.</a>
+
+<a name="qq4answer"></a>
+<p><b>Quick Quiz 4</b>:
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+
+
+</p><p><b>Answer</b>:
+Let's start with what happens to <tt>do_something_gp()</tt>
+if it fails to use <tt>rcu_dereference()</tt>.
+It could reuse a value formerly fetched from this same pointer.
+It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
+manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
+mash-up of two distince pointer values.
+It might even use value-speculation optimizations, where it makes a wrong
+guess, but by the time it gets around to checking the value, an update
+has changed the pointer to match the wrong guess.
+Too bad about any dereferences that returned pre-initialization garbage
+in the meantime!
+
+<p>
+For <tt>remove_gp_synchronous()</tt>, as long as all modifications
+to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
+the above optimizations are harmless.
+However,
+with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
+<tt>sparse</tt> will complain if you
+define <tt>gp</tt> with <tt>__rcu</tt> and then
+access it without using
+either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
+
+
+</p><p><a href="#Quick%20Quiz%204"><b>Back to Quick Quiz 4</b>.</a>
+
+<a name="qq5answer"></a>
+<p><b>Quick Quiz 5</b>:
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+
+
+</p><p><b>Answer</b>:
+If RCU cannot tell whether or not a given
+RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>,
+then it must assume that the RCU read-side critical section
+started first.
+In other words, a given instance of <tt>synchronize_rcu()</tt>
+can avoid waiting on a given RCU read-side critical section only
+if it can prove that <tt>synchronize_rcu()</tt> started first.
+
+
+</p><p><a href="#Quick%20Quiz%205"><b>Back to Quick Quiz 5</b>.</a>
+
+<a name="qq6answer"></a>
+<p><b>Quick Quiz 6</b>:
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+
+
+</p><p><b>Answer</b>:
+Yes, they really are required.
+To see why the first guarantee is required, consider the following
+sequence of events:
+
+<ol>
+<li> CPU 1: <tt>rcu_read_lock()</tt>
+<li> CPU 1: <tt>q = rcu_dereference(gp);
+ /* Very likely to return p. */</tt>
+<li> CPU 0: <tt>list_del_rcu(p);</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li> CPU 1: <tt>do_something_with(q-&gt;a);
+ /* No smp_mb(), so might happen after kfree(). */</tt>
+<li> CPU 1: <tt>rcu_read_unlock()</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li> CPU 0: <tt>kfree(p);</tt>
+</ol>
+
+<p>
+Therefore, there absolutely must be a full memory barrier between the
+end of the RCU read-side critical section and the end of the
+grace period.
+
+<p>
+The sequence of events demonstrating the necessity of the second rule
+is roughly similar:
+
+<ol>
+<li> CPU 0: <tt>list_del_rcu(p);</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li> CPU 1: <tt>rcu_read_lock()</tt>
+<li> CPU 1: <tt>q = rcu_dereference(gp);
+ /* Might return p if no memory barrier. */</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li> CPU 0: <tt>kfree(p);</tt>
+<li> CPU 1: <tt>do_something_with(q-&gt;a); /* Boom!!! */</tt>
+<li> CPU 1: <tt>rcu_read_unlock()</tt>
+</ol>
+
+<p>
+And similarly, without a memory barrier between the beginning of the
+grace period and the beginning of the RCU read-side critical section,
+CPU&nbsp;1 might end up accessing the freelist.
+
+<p>
+The &ldquo;as if&rdquo; rule of course applies, so that any implementation
+that acts as if the appropriate memory barriers were in place is a
+correct implementation.
+That said, it is much easier to fool yourself into believing that you have
+adhered to the as-if rule than it is to actually adhere to it!
+
+
+</p><p><a href="#Quick%20Quiz%206"><b>Back to Quick Quiz 6</b>.</a>
+
+<a name="qq7answer"></a>
+<p><b>Quick Quiz 7</b>:
+But how does the upgrade-to-write operation exclude other readers?
+
+
+</p><p><b>Answer</b>:
+It doesn't, just like normal RCU updates, which also do not exclude
+RCU readers.
+
+
+</p><p><a href="#Quick%20Quiz%207"><b>Back to Quick Quiz 7</b>.</a>
+
+<a name="qq8answer"></a>
+<p><b>Quick Quiz 8</b>:
+Can't the compiler also reorder this code?
+
+
+</p><p><b>Answer</b>:
+No, the volatile casts in <tt>READ_ONCE()</tt> and
+<tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
+this particular case.
+
+
+</p><p><a href="#Quick%20Quiz%208"><b>Back to Quick Quiz 8</b>.</a>
+
+<a name="qq9answer"></a>
+<p><b>Quick Quiz 9</b>:
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+
+
+</p><p><b>Answer</b>:
+No.
+Even if <tt>synchronize_rcu()</tt> were to wait until
+all readers had completed, a new reader might start immediately after
+<tt>synchronize_rcu()</tt> completed.
+Therefore, the code following
+<tt>synchronize_rcu()</tt> cannot rely on there being no readers
+in any case.
+
+
+</p><p><a href="#Quick%20Quiz%209"><b>Back to Quick Quiz 9</b>.</a>
+
+<a name="qq10answer"></a>
+<p><b>Quick Quiz 10</b>:
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+
+
+</p><p><b>Answer</b>:
+In theory, an infinite number.
+In practice, an unknown number that is sensitive to both implementation
+details and timing considerations.
+Therefore, even in practice, RCU users must abide by the theoretical rather
+than the practical answer.
+
+
+</p><p><a href="#Quick%20Quiz%2010"><b>Back to Quick Quiz 10</b>.</a>
+
+<a name="qq11answer"></a>
+<p><b>Quick Quiz 11</b>:
+What about sleeping locks?
+
+
+</p><p><b>Answer</b>:
+These are forbidden within Linux-kernel RCU read-side critical sections
+because it is not legal to place a quiescent state (in this case,
+voluntary context switch) within an RCU read-side critical section.
+However, sleeping locks may be used within userspace RCU read-side critical
+sections, and also within Linux-kernel sleepable RCU
+<a href="#Sleepable RCU">(SRCU)</a>
+read-side critical sections.
+In addition, the -rt patchset turns spinlocks into a sleeping locks so
+that the corresponding critical sections can be preempted, which
+also means that these sleeplockified spinlocks (but not other sleeping locks!)
+may be acquire within -rt-Linux-kernel RCU read-side critical sections.
+
+<p>
+Note that it <i>is</i> legal for a normal RCU read-side critical section
+to conditionally acquire a sleeping locks (as in <tt>mutex_trylock()</tt>),
+but only as long as it does not loop indefinitely attempting to
+conditionally acquire that sleeping locks.
+The key point is that things like <tt>mutex_trylock()</tt>
+either return with the mutex held, or return an error indication if
+the mutex was not immediately available.
+Either way, <tt>mutex_trylock()</tt> returns immediately without sleeping.
+
+
+</p><p><a href="#Quick%20Quiz%2011"><b>Back to Quick Quiz 11</b>.</a>
+
+<a name="qq12answer"></a>
+<p><b>Quick Quiz 12</b>:
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+
+
+</p><p><b>Answer</b>:
+Presumably the <tt>-&gt;gp_lock</tt> acquired on line&nbsp;18 excludes
+any changes, including any insertions that <tt>rcu_dereference()</tt>
+would protect against.
+Therefore, any insertions will be delayed until after <tt>-&gt;gp_lock</tt>
+is released on line&nbsp;25, which in turn means that
+<tt>rcu_access_pointer()</tt> suffices.
+
+
+</p><p><a href="#Quick%20Quiz%2012"><b>Back to Quick Quiz 12</b>.</a>
+
+<a name="qq13answer"></a>
+<p><b>Quick Quiz 13</b>:
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+
+
+</p><p><b>Answer</b>:
+We could define things this way, but keep in mind that this sort of
+definition would say that updates in garbage-collected languages
+cannot complete until the next time the garbage collector runs,
+which does not seem at all reasonable.
+The key point is that in most cases, an updater using either
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
+next update as soon as it has invoked <tt>call_rcu()</tt> or
+<tt>kfree_rcu()</tt>, without having to wait for a subsequent
+grace period.
+
+
+</p><p><a href="#Quick%20Quiz%2013"><b>Back to Quick Quiz 13</b>.</a>
+
+<a name="qq14answer"></a>
+<p><b>Quick Quiz 14</b>:
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+
+
+</p><p><b>Answer</b>:
+In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
+maps directly to <tt>synchronize_sched()</tt>.
+Therefore, <tt>synchronize_rcu()</tt> works normally throughout
+boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
+However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
+so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
+during scheduler initialization.
+
+
+</p><p><a href="#Quick%20Quiz%2014"><b>Back to Quick Quiz 14</b>.</a>
+
+
+</body></html>
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
new file mode 100644
index 000000000000..c5bb5bddd7ee
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -0,0 +1,2643 @@
+<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
+ "http://www.w3.org/TR/html4/loose.dtd">
+ <html>
+ <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
+ <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
+
+<h1>A Tour Through RCU's Requirements</h1>
+
+<p>Copyright IBM Corporation, 2015</p>
+<p>Author: Paul E.&nbsp;McKenney</p>
+<p><i>The initial version of this document appeared in the
+<a href="http://lwn.net/">LWN</a> articles
+<a href="http://lwn.net/Articles/652156/">here</a>,
+<a href="http://lwn.net/Articles/652677/">here</a>, and
+<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
+
+<h2>Introduction</h2>
+
+<p>
+Read-copy update (RCU) is a synchronization mechanism that is often
+used as a replacement for reader-writer locking.
+RCU is unusual in that updaters do not block readers,
+which means that RCU's read-side primitives can be exceedingly fast
+and scalable.
+In addition, updaters can make useful forward progress concurrently
+with readers.
+However, all this concurrency between RCU readers and updaters does raise
+the question of exactly what RCU readers are doing, which in turn
+raises the question of exactly what RCU's requirements are.
+
+<p>
+This document therefore summarizes RCU's requirements, and can be thought
+of as an informal, high-level specification for RCU.
+It is important to understand that RCU's specification is primarily
+empirical in nature;
+in fact, I learned about many of these requirements the hard way.
+This situation might cause some consternation, however, not only
+has this learning process been a lot of fun, but it has also been
+a great privilege to work with so many people willing to apply
+technologies in interesting new ways.
+
+<p>
+All that aside, here are the categories of currently known RCU requirements:
+</p>
+
+<ol>
+<li> <a href="#Fundamental Requirements">
+ Fundamental Requirements</a>
+<li> <a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
+<li> <a href="#Parallelism Facts of Life">
+ Parallelism Facts of Life</a>
+<li> <a href="#Quality-of-Implementation Requirements">
+ Quality-of-Implementation Requirements</a>
+<li> <a href="#Linux Kernel Complications">
+ Linux Kernel Complications</a>
+<li> <a href="#Software-Engineering Requirements">
+ Software-Engineering Requirements</a>
+<li> <a href="#Other RCU Flavors">
+ Other RCU Flavors</a>
+<li> <a href="#Possible Future Changes">
+ Possible Future Changes</a>
+</ol>
+
+<p>
+This is followed by a <a href="#Summary">summary</a>,
+which is in turn followed by the inevitable
+<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.
+
+<h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>
+
+<p>
+RCU's fundamental requirements are the closest thing RCU has to hard
+mathematical requirements.
+These are:
+
+<ol>
+<li> <a href="#Grace-Period Guarantee">
+ Grace-Period Guarantee</a>
+<li> <a href="#Publish-Subscribe Guarantee">
+ Publish-Subscribe Guarantee</a>
+<li> <a href="#RCU Primitives Guaranteed to Execute Unconditionally">
+ RCU Primitives Guaranteed to Execute Unconditionally</a>
+<li> <a href="#Guaranteed Read-to-Write Upgrade">
+ Guaranteed Read-to-Write Upgrade</a>
+</ol>
+
+<h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>
+
+<p>
+RCU's grace-period guarantee is unusual in being premeditated:
+Jack Slingwine and I had this guarantee firmly in mind when we started
+work on RCU (then called &ldquo;rclock&rdquo;) in the early 1990s.
+That said, the past two decades of experience with RCU have produced
+a much more detailed understanding of this guarantee.
+
+<p>
+RCU's grace-period guarantee allows updaters to wait for the completion
+of all pre-existing RCU read-side critical sections.
+An RCU read-side critical section
+begins with the marker <tt>rcu_read_lock()</tt> and ends with
+the marker <tt>rcu_read_unlock()</tt>.
+These markers may be nested, and RCU treats a nested set as one
+big RCU read-side critical section.
+Production-quality implementations of <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> are extremely lightweight, and in
+fact have exactly zero overhead in Linux kernels built for production
+use with <tt>CONFIG_PREEMPT=n</tt>.
+
+<p>
+This guarantee allows ordering to be enforced with extremely low
+overhead to readers, for example:
+
+<blockquote>
+<pre>
+ 1 int x, y;
+ 2
+ 3 void thread0(void)
+ 4 {
+ 5 rcu_read_lock();
+ 6 r1 = READ_ONCE(x);
+ 7 r2 = READ_ONCE(y);
+ 8 rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13 WRITE_ONCE(x, 1);
+14 synchronize_rcu();
+15 WRITE_ONCE(y, 1);
+16 }
+</pre>
+</blockquote>
+
+<p>
+Because the <tt>synchronize_rcu()</tt> on line&nbsp;14 waits for
+all pre-existing readers, any instance of <tt>thread0()</tt> that
+loads a value of zero from <tt>x</tt> must complete before
+<tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
+also load a value of zero from <tt>y</tt>.
+Similarly, any instance of <tt>thread0()</tt> that loads a value of
+one from <tt>y</tt> must have started after the
+<tt>synchronize_rcu()</tt> started, and must therefore also load
+a value of one from <tt>x</tt>.
+Therefore, the outcome:
+<blockquote>
+<pre>
+(r1 == 0 &amp;&amp; r2 == 1)
+</pre>
+</blockquote>
+cannot happen.
+
+<p>@@QQ@@
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+<p>@@QQA@@
+First, if updaters do not wish to be blocked by readers, they can use
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
+be discussed later.
+Second, even when using <tt>synchronize_rcu()</tt>, the other
+update-side code does run concurrently with readers, whether pre-existing
+or not.
+<p>@@QQE@@
+
+<p>
+This scenario resembles one of the first uses of RCU in
+<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
+which managed a distributed lock manager's transition into
+a state suitable for handling recovery from node failure,
+more or less as follows:
+
+<blockquote>
+<pre>
+ 1 #define STATE_NORMAL 0
+ 2 #define STATE_WANT_RECOVERY 1
+ 3 #define STATE_RECOVERING 2
+ 4 #define STATE_WANT_NORMAL 3
+ 5
+ 6 int state = STATE_NORMAL;
+ 7
+ 8 void do_something_dlm(void)
+ 9 {
+10 int state_snap;
+11
+12 rcu_read_lock();
+13 state_snap = READ_ONCE(state);
+14 if (state_snap == STATE_NORMAL)
+15 do_something();
+16 else
+17 do_something_carefully();
+18 rcu_read_unlock();
+19 }
+20
+21 void start_recovery(void)
+22 {
+23 WRITE_ONCE(state, STATE_WANT_RECOVERY);
+24 synchronize_rcu();
+25 WRITE_ONCE(state, STATE_RECOVERING);
+26 recovery();
+27 WRITE_ONCE(state, STATE_WANT_NORMAL);
+28 synchronize_rcu();
+29 WRITE_ONCE(state, STATE_NORMAL);
+30 }
+</pre>
+</blockquote>
+
+<p>
+The RCU read-side critical section in <tt>do_something_dlm()</tt>
+works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
+to guarantee that <tt>do_something()</tt> never runs concurrently
+with <tt>recovery()</tt>, but with little or no synchronization
+overhead in <tt>do_something_dlm()</tt>.
+
+<p>@@QQ@@
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+<p>@@QQA@@
+Without that extra grace period, memory reordering could result in
+<tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
+concurrently with the last bits of <tt>recovery()</tt>.
+<p>@@QQE@@
+
+<p>
+In order to avoid fatal problems such as deadlocks,
+an RCU read-side critical section must not contain calls to
+<tt>synchronize_rcu()</tt>.
+Similarly, an RCU read-side critical section must not
+contain anything that waits, directly or indirectly, on completion of
+an invocation of <tt>synchronize_rcu()</tt>.
+
+<p>
+Although RCU's grace-period guarantee is useful in and of itself, with
+<a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
+it would be good to be able to use RCU to coordinate read-side
+access to linked data structures.
+For this, the grace-period guarantee is not sufficient, as can
+be seen in function <tt>add_gp_buggy()</tt> below.
+We will look at the reader's code later, but in the meantime, just think of
+the reader as locklessly picking up the <tt>gp</tt> pointer,
+and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
+<tt>-&gt;a</tt> and <tt>-&gt;b</tt> fields.
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 p-&gt;a = a;
+12 p-&gt;b = a;
+13 gp = p; /* ORDERING BUG */
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The problem is that both the compiler and weakly ordered CPUs are within
+their rights to reorder this code as follows:
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy_optimized(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+<b>11 gp = p; /* ORDERING BUG */
+12 p-&gt;a = a;
+13 p-&gt;b = a;</b>
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+If an RCU reader fetches <tt>gp</tt> just after
+<tt>add_gp_buggy_optimized</tt> executes line&nbsp;11,
+it will see garbage in the <tt>-&gt;a</tt> and <tt>-&gt;b</tt>
+fields.
+And this is but one of many ways in which compiler and hardware optimizations
+could cause trouble.
+Therefore, we clearly need some way to prevent the compiler and the CPU from
+reordering in this manner, which brings us to the publish-subscribe
+guarantee discussed in the next section.
+
+<h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>
+
+<p>
+RCU's publish-subscribe guarantee allows data to be inserted
+into a linked data structure without disrupting RCU readers.
+The updater uses <tt>rcu_assign_pointer()</tt> to insert the
+new data, and readers use <tt>rcu_dereference()</tt> to
+access data, whether new or old.
+The following shows an example of insertion:
+
+<blockquote>
+<pre>
+ 1 bool add_gp(int a, int b)
+ 2 {
+ 3 p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4 if (!p)
+ 5 return -ENOMEM;
+ 6 spin_lock(&amp;gp_lock);
+ 7 if (rcu_access_pointer(gp)) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 p-&gt;a = a;
+12 p-&gt;b = a;
+13 rcu_assign_pointer(gp, p);
+14 spin_unlock(&amp;gp_lock);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_assign_pointer()</tt> on line&nbsp;13 is conceptually
+equivalent to a simple assignment statement, but also guarantees
+that its assignment will
+happen after the two assignments in lines&nbsp;11 and&nbsp;12,
+similar to the C11 <tt>memory_order_release</tt> store operation.
+It also prevents any number of &ldquo;interesting&rdquo; compiler
+optimizations, for example, the use of <tt>gp</tt> as a scratch
+location immediately preceding the assignment.
+
+<p>@@QQ@@
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+<p>@@QQA@@
+No, it cannot.
+The readers cannot see either of these two fields until
+the assignment to <tt>gp</tt>, by which time both fields are
+fully initialized.
+So reordering the assignments
+to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt> cannot possibly
+cause any problems.
+<p>@@QQE@@
+
+<p>
+It is tempting to assume that the reader need not do anything special
+to control its accesses to the RCU-protected data,
+as shown in <tt>do_something_gp_buggy()</tt> below:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 p = gp; /* OPTIMIZATIONS GALORE!!! */
+ 5 if (p) {
+ 6 do_something(p-&gt;a, p-&gt;b);
+ 7 rcu_read_unlock();
+ 8 return true;
+ 9 }
+10 rcu_read_unlock();
+11 return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+However, this temptation must be resisted because there are a
+surprisingly large number of ways that the compiler
+(to say nothing of
+<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
+can trip this code up.
+For but one example, if the compiler were short of registers, it
+might choose to refetch from <tt>gp</tt> rather than keeping
+a separate copy in <tt>p</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy_optimized(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */
+<b> 5 do_something(gp-&gt;a, gp-&gt;b);</b>
+ 6 rcu_read_unlock();
+ 7 return true;
+ 8 }
+ 9 rcu_read_unlock();
+10 return false;
+11 }
+</pre>
+</blockquote>
+
+<p>
+If this function ran concurrently with a series of updates that
+replaced the current structure with a new one,
+the fetches of <tt>gp-&gt;a</tt>
+and <tt>gp-&gt;b</tt> might well come from two different structures,
+which could cause serious confusion.
+To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
+<tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 p = rcu_dereference(gp);
+ 5 if (p) {
+ 6 do_something(p-&gt;a, p-&gt;b);
+ 7 rcu_read_unlock();
+ 8 return true;
+ 9 }
+10 rcu_read_unlock();
+11 return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
+memory barriers in the Linux kernel.
+Should a
+<a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
+ever appear, then <tt>rcu_dereference()</tt> could be implemented
+as a <tt>memory_order_consume</tt> load.
+Regardless of the exact implementation, a pointer fetched by
+<tt>rcu_dereference()</tt> may not be used outside of the
+outermost RCU read-side critical section containing that
+<tt>rcu_dereference()</tt>, unless protection of
+the corresponding data element has been passed from RCU to some
+other synchronization mechanism, most commonly locking or
+<a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.
+
+<p>
+In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
+use <tt>rcu_dereference()</tt>, and these two RCU API elements
+work together to ensure that readers have a consistent view of
+newly added data elements.
+
+<p>
+Of course, it is also necessary to remove elements from RCU-protected
+data structures, for example, using the following process:
+
+<ol>
+<li> Remove the data element from the enclosing structure.
+<li> Wait for all pre-existing RCU read-side critical sections
+ to complete (because only pre-existing readers can possibly have
+ a reference to the newly removed data element).
+<li> At this point, only the updater has a reference to the
+ newly removed data element, so it can safely reclaim
+ the data element, for example, by passing it to <tt>kfree()</tt>.
+</ol>
+
+This process is implemented by <tt>remove_gp_synchronous()</tt>:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_synchronous(void)
+ 2 {
+ 3 struct foo *p;
+ 4
+ 5 spin_lock(&amp;gp_lock);
+ 6 p = rcu_access_pointer(gp);
+ 7 if (!p) {
+ 8 spin_unlock(&amp;gp_lock);
+ 9 return false;
+10 }
+11 rcu_assign_pointer(gp, NULL);
+12 spin_unlock(&amp;gp_lock);
+13 synchronize_rcu();
+14 kfree(p);
+15 return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+This function is straightforward, with line&nbsp;13 waiting for a grace
+period before line&nbsp;14 frees the old data element.
+This waiting ensures that readers will reach line&nbsp;7 of
+<tt>do_something_gp()</tt> before the data element referenced by
+<tt>p</tt> is freed.
+The <tt>rcu_access_pointer()</tt> on line&nbsp;6 is similar to
+<tt>rcu_dereference()</tt>, except that:
+
+<ol>
+<li> The value returned by <tt>rcu_access_pointer()</tt>
+ cannot be dereferenced.
+ If you want to access the value pointed to as well as
+ the pointer itself, use <tt>rcu_dereference()</tt>
+ instead of <tt>rcu_access_pointer()</tt>.
+<li> The call to <tt>rcu_access_pointer()</tt> need not be
+ protected.
+ In contrast, <tt>rcu_dereference()</tt> must either be
+ within an RCU read-side critical section or in a code
+ segment where the pointer cannot change, for example, in
+ code protected by the corresponding update-side lock.
+</ol>
+
+<p>@@QQ@@
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+<p>@@QQA@@
+Let's start with what happens to <tt>do_something_gp()</tt>
+if it fails to use <tt>rcu_dereference()</tt>.
+It could reuse a value formerly fetched from this same pointer.
+It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
+manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
+mash-up of two distince pointer values.
+It might even use value-speculation optimizations, where it makes a wrong
+guess, but by the time it gets around to checking the value, an update
+has changed the pointer to match the wrong guess.
+Too bad about any dereferences that returned pre-initialization garbage
+in the meantime!
+
+<p>
+For <tt>remove_gp_synchronous()</tt>, as long as all modifications
+to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
+the above optimizations are harmless.
+However,
+with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
+<tt>sparse</tt> will complain if you
+define <tt>gp</tt> with <tt>__rcu</tt> and then
+access it without using
+either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
+<p>@@QQE@@
+
+<p>
+This simple linked-data-structure scenario clearly demonstrates the need
+for RCU's stringent memory-ordering guarantees on systems with more than
+one CPU:
+
+<ol>
+<li> Each CPU that has an RCU read-side critical section that
+ begins before <tt>synchronize_rcu()</tt> starts is
+ guaranteed to execute a full memory barrier between the time
+ that the RCU read-side critical section ends and the time that
+ <tt>synchronize_rcu()</tt> returns.
+ Without this guarantee, a pre-existing RCU read-side critical section
+ might hold a reference to the newly removed <tt>struct foo</tt>
+ after the <tt>kfree()</tt> on line&nbsp;14 of
+ <tt>remove_gp_synchronous()</tt>.
+<li> Each CPU that has an RCU read-side critical section that ends
+ after <tt>synchronize_rcu()</tt> returns is guaranteed
+ to execute a full memory barrier between the time that
+ <tt>synchronize_rcu()</tt> begins and the time that the RCU
+ read-side critical section begins.
+ Without this guarantee, a later RCU read-side critical section
+ running after the <tt>kfree()</tt> on line&nbsp;14 of
+ <tt>remove_gp_synchronous()</tt> might
+ later run <tt>do_something_gp()</tt> and find the
+ newly deleted <tt>struct foo</tt>.
+<li> If the task invoking <tt>synchronize_rcu()</tt> remains
+ on a given CPU, then that CPU is guaranteed to execute a full
+ memory barrier sometime during the execution of
+ <tt>synchronize_rcu()</tt>.
+ This guarantee ensures that the <tt>kfree()</tt> on
+ line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+ execute after the removal on line&nbsp;11.
+<li> If the task invoking <tt>synchronize_rcu()</tt> migrates
+ among a group of CPUs during that invocation, then each of the
+ CPUs in that group is guaranteed to execute a full memory barrier
+ sometime during the execution of <tt>synchronize_rcu()</tt>.
+ This guarantee also ensures that the <tt>kfree()</tt> on
+ line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+ execute after the removal on
+ line&nbsp;11, but also in the case where the thread executing the
+ <tt>synchronize_rcu()</tt> migrates in the meantime.
+</ol>
+
+<p>@@QQ@@
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+<p>@@QQA@@
+If RCU cannot tell whether or not a given
+RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>,
+then it must assume that the RCU read-side critical section
+started first.
+In other words, a given instance of <tt>synchronize_rcu()</tt>
+can avoid waiting on a given RCU read-side critical section only
+if it can prove that <tt>synchronize_rcu()</tt> started first.
+<p>@@QQE@@
+
+<p>@@QQ@@
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+<p>@@QQA@@
+Yes, they really are required.
+To see why the first guarantee is required, consider the following
+sequence of events:
+
+<ol>
+<li> CPU 1: <tt>rcu_read_lock()</tt>
+<li> CPU 1: <tt>q = rcu_dereference(gp);
+ /* Very likely to return p. */</tt>
+<li> CPU 0: <tt>list_del_rcu(p);</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li> CPU 1: <tt>do_something_with(q-&gt;a);
+ /* No smp_mb(), so might happen after kfree(). */</tt>
+<li> CPU 1: <tt>rcu_read_unlock()</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li> CPU 0: <tt>kfree(p);</tt>
+</ol>
+
+<p>
+Therefore, there absolutely must be a full memory barrier between the
+end of the RCU read-side critical section and the end of the
+grace period.
+
+<p>
+The sequence of events demonstrating the necessity of the second rule
+is roughly similar:
+
+<ol>
+<li> CPU 0: <tt>list_del_rcu(p);</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li> CPU 1: <tt>rcu_read_lock()</tt>
+<li> CPU 1: <tt>q = rcu_dereference(gp);
+ /* Might return p if no memory barrier. */</tt>
+<li> CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li> CPU 0: <tt>kfree(p);</tt>
+<li> CPU 1: <tt>do_something_with(q-&gt;a); /* Boom!!! */</tt>
+<li> CPU 1: <tt>rcu_read_unlock()</tt>
+</ol>
+
+<p>
+And similarly, without a memory barrier between the beginning of the
+grace period and the beginning of the RCU read-side critical section,
+CPU&nbsp;1 might end up accessing the freelist.
+
+<p>
+The &ldquo;as if&rdquo; rule of course applies, so that any implementation
+that acts as if the appropriate memory barriers were in place is a
+correct implementation.
+That said, it is much easier to fool yourself into believing that you have
+adhered to the as-if rule than it is to actually adhere to it!
+<p>@@QQE@@
+
+<p>
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
+
+<p>
+The common-case RCU primitives are unconditional.
+They are invoked, they do their job, and they return, with no possibility
+of error, and no need to retry.
+This is a key RCU design philosophy.
+
+<p>
+However, this philosophy is pragmatic rather than pigheaded.
+If someone comes up with a good justification for a particular conditional
+RCU primitive, it might well be implemented and added.
+After all, this guarantee was reverse-engineered, not premeditated.
+The unconditional nature of the RCU primitives was initially an
+accident of implementation, and later experience with synchronization
+primitives with conditional primitives caused me to elevate this
+accident to a guarantee.
+Therefore, the justification for adding a conditional primitive to
+RCU would need to be based on detailed and compelling use cases.
+
+<h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
+
+<p>
+As far as RCU is concerned, it is always possible to carry out an
+update within an RCU read-side critical section.
+For example, that RCU read-side critical section might search for
+a given data element, and then might acquire the update-side
+spinlock in order to update that element, all while remaining
+in that RCU read-side critical section.
+Of course, it is necessary to exit the RCU read-side critical section
+before invoking <tt>synchronize_rcu()</tt>, however, this
+inconvenience can be avoided through use of the
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
+described later in this document.
+
+<p>@@QQ@@
+But how does the upgrade-to-write operation exclude other readers?
+<p>@@QQA@@
+It doesn't, just like normal RCU updates, which also do not exclude
+RCU readers.
+<p>@@QQE@@
+
+<p>
+This guarantee allows lookup code to be shared between read-side
+and update-side code, and was premeditated, appearing in the earliest
+DYNIX/ptx RCU documentation.
+
+<h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
+
+<p>
+RCU provides extremely lightweight readers, and its read-side guarantees,
+though quite useful, are correspondingly lightweight.
+It is therefore all too easy to assume that RCU is guaranteeing more
+than it really is.
+Of course, the list of things that RCU does not guarantee is infinitely
+long, however, the following sections list a few non-guarantees that
+have caused confusion.
+Except where otherwise noted, these non-guarantees were premeditated.
+
+<ol>
+<li> <a href="#Readers Impose Minimal Ordering">
+ Readers Impose Minimal Ordering</a>
+<li> <a href="#Readers Do Not Exclude Updaters">
+ Readers Do Not Exclude Updaters</a>
+<li> <a href="#Updaters Only Wait For Old Readers">
+ Updaters Only Wait For Old Readers</a>
+<li> <a href="#Grace Periods Don't Partition Read-Side Critical Sections">
+ Grace Periods Don't Partition Read-Side Critical Sections</a>
+<li> <a href="#Read-Side Critical Sections Don't Partition Grace Periods">
+ Read-Side Critical Sections Don't Partition Grace Periods</a>
+<li> <a href="#Disabling Preemption Does Not Block Grace Periods">
+ Disabling Preemption Does Not Block Grace Periods</a>
+</ol>
+
+<h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
+
+<p>
+Reader-side markers such as <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
+except through their interaction with the grace-period APIs such as
+<tt>synchronize_rcu()</tt>.
+To see this, consider the following pair of threads:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(x, 1);
+ 5 rcu_read_unlock();
+ 6 rcu_read_lock();
+ 7 WRITE_ONCE(y, 1);
+ 8 rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13 rcu_read_lock();
+14 r1 = READ_ONCE(y);
+15 rcu_read_unlock();
+16 rcu_read_lock();
+17 r2 = READ_ONCE(x);
+18 rcu_read_unlock();
+19 }
+</pre>
+</blockquote>
+
+<p>
+After <tt>thread0()</tt> and <tt>thread1()</tt> execute
+concurrently, it is quite possible to have
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0)
+</pre>
+</blockquote>
+
+(that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
+which would not be possible if <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> had much in the way of ordering
+properties.
+But they do not, so the CPU is within its rights
+to do significant reordering.
+This is by design: Any significant ordering constraints would slow down
+these fast-path APIs.
+
+<p>@@QQ@@
+Can't the compiler also reorder this code?
+<p>@@QQA@@
+No, the volatile casts in <tt>READ_ONCE()</tt> and
+<tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
+this particular case.
+<p>@@QQE@@
+
+<h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
+
+<p>
+Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
+exclude updates.
+All they do is to prevent grace periods from ending.
+The following example illustrates this:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 r1 = READ_ONCE(y);
+ 5 if (r1) {
+ 6 do_something_with_nonzero_x();
+ 7 r2 = READ_ONCE(x);
+ 8 WARN_ON(!r2); /* BUG!!! */
+ 9 }
+10 rcu_read_unlock();
+11 }
+12
+13 void thread1(void)
+14 {
+15 spin_lock(&amp;my_lock);
+16 WRITE_ONCE(x, 1);
+17 WRITE_ONCE(y, 1);
+18 spin_unlock(&amp;my_lock);
+19 }
+</pre>
+</blockquote>
+
+<p>
+If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
+excluded the <tt>thread1()</tt> function's update,
+the <tt>WARN_ON()</tt> could never fire.
+But the fact is that <tt>rcu_read_lock()</tt> does not exclude
+much of anything aside from subsequent grace periods, of which
+<tt>thread1()</tt> has none, so the
+<tt>WARN_ON()</tt> can and does fire.
+
+<h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
+
+<p>
+It might be tempting to assume that after <tt>synchronize_rcu()</tt>
+completes, there are no readers executing.
+This temptation must be avoided because
+new readers can start immediately after <tt>synchronize_rcu()</tt>
+starts, and <tt>synchronize_rcu()</tt> is under no
+obligation to wait for these new readers.
+
+<p>@@QQ@@
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+<p>@@QQA@@
+No.
+Even if <tt>synchronize_rcu()</tt> were to wait until
+all readers had completed, a new reader might start immediately after
+<tt>synchronize_rcu()</tt> completed.
+Therefore, the code following
+<tt>synchronize_rcu()</tt> cannot rely on there being no readers
+in any case.
+<p>@@QQE@@
+
+<h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
+Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
+
+<p>
+It is tempting to assume that if any part of one RCU read-side critical
+section precedes a given grace period, and if any part of another RCU
+read-side critical section follows that same grace period, then all of
+the first RCU read-side critical section must precede all of the second.
+However, this just isn't the case: A single grace period does not
+partition the set of RCU read-side critical sections.
+An example of this situation can be illustrated as follows, where
+<tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 rcu_read_lock();
+19 r2 = READ_ONCE(b);
+20 r3 = READ_ONCE(c);
+21 rcu_read_unlock();
+22 }
+</pre>
+</blockquote>
+
+<p>
+It turns out that the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0 &amp;&amp; r3 == 1)
+</pre>
+</blockquote>
+
+is entirely possible.
+The following figure show how this can happen, with each circled
+<tt>QS</tt> indicating the point at which RCU recorded a
+<i>quiescent state</i> for each thread, that is, a state in which
+RCU knows that the thread cannot be in the midst of an RCU read-side
+critical section that started before the current grace period:
+
+<p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
+
+<p>
+If it is necessary to partition RCU read-side critical sections in this
+manner, it is necessary to use two grace periods, where the first
+grace period is known to end before the second grace period starts:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 r2 = READ_ONCE(c);
+19 synchronize_rcu();
+20 WRITE_ONCE(d, 1);
+21 }
+22
+23 void thread3(void)
+24 {
+25 rcu_read_lock();
+26 r3 = READ_ONCE(b);
+27 r4 = READ_ONCE(d);
+28 rcu_read_unlock();
+29 }
+</pre>
+</blockquote>
+
+<p>
+Here, if <tt>(r1 == 1)</tt>, then
+<tt>thread0()</tt>'s write to <tt>b</tt> must happen
+before the end of <tt>thread1()</tt>'s grace period.
+If in addition <tt>(r4 == 1)</tt>, then
+<tt>thread3()</tt>'s read from <tt>b</tt> must happen
+after the beginning of <tt>thread2()</tt>'s grace period.
+If it is also the case that <tt>(r2 == 1)</tt>, then the
+end of <tt>thread1()</tt>'s grace period must precede the
+beginning of <tt>thread2()</tt>'s grace period.
+This mean that the two RCU read-side critical sections cannot overlap,
+guaranteeing that <tt>(r3 == 1)</tt>.
+As a result, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 0 &amp;&amp; r4 == 1)
+</pre>
+</blockquote>
+
+cannot happen.
+
+<p>
+This non-requirement was also non-premeditated, but became apparent
+when studying RCU's interaction with memory ordering.
+
+<h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
+Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
+
+<p>
+It is also tempting to assume that if an RCU read-side critical section
+happens between a pair of grace periods, then those grace periods cannot
+overlap.
+However, this temptation leads nowhere good, as can be illustrated by
+the following, with all variables initially zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3 rcu_read_lock();
+ 4 WRITE_ONCE(a, 1);
+ 5 WRITE_ONCE(b, 1);
+ 6 rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11 r1 = READ_ONCE(a);
+12 synchronize_rcu();
+13 WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18 rcu_read_lock();
+19 WRITE_ONCE(d, 1);
+20 r2 = READ_ONCE(c);
+21 rcu_read_unlock();
+22 }
+23
+24 void thread3(void)
+25 {
+26 r3 = READ_ONCE(d);
+27 synchronize_rcu();
+28 WRITE_ONCE(e, 1);
+29 }
+30
+31 void thread4(void)
+32 {
+33 rcu_read_lock();
+34 r4 = READ_ONCE(b);
+35 r5 = READ_ONCE(e);
+36 rcu_read_unlock();
+37 }
+</pre>
+</blockquote>
+
+<p>
+In this case, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 1 &amp;&amp; r4 == 0 &amp&amp; r5 == 1)
+</pre>
+</blockquote>
+
+is entirely possible, as illustrated below:
+
+<p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
+
+<p>
+Again, an RCU read-side critical section can overlap almost all of a
+given grace period, just so long as it does not overlap the entire
+grace period.
+As a result, an RCU read-side critical section cannot partition a pair
+of RCU grace periods.
+
+<p>@@QQ@@
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+<p>@@QQA@@
+In theory, an infinite number.
+In practice, an unknown number that is sensitive to both implementation
+details and timing considerations.
+Therefore, even in practice, RCU users must abide by the theoretical rather
+than the practical answer.
+<p>@@QQE@@
+
+<h3><a name="Disabling Preemption Does Not Block Grace Periods">
+Disabling Preemption Does Not Block Grace Periods</a></h3>
+
+<p>
+There was a time when disabling preemption on any given CPU would block
+subsequent grace periods.
+However, this was an accident of implementation and is not a requirement.
+And in the current Linux-kernel implementation, disabling preemption
+on a given CPU in fact does not block grace periods, as Oleg Nesterov
+<a href="http://marc.info/?l=linux-kernel&amp;m=143431078202914&amp;w=2">demonstrated</a>.
+
+<p>
+If you need a preempt-disable region to block grace periods, you need to add
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
+as follows:
+
+<blockquote>
+<pre>
+ 1 preempt_disable();
+ 2 rcu_read_lock();
+ 3 do_something();
+ 4 rcu_read_unlock();
+ 5 preempt_enable();
+ 6
+ 7 /* Spinlocks implicitly disable preemption. */
+ 8 spin_lock(&amp;mylock);
+ 9 rcu_read_lock();
+10 do_something();
+11 rcu_read_unlock();
+12 spin_unlock(&amp;mylock);
+</pre>
+</blockquote>
+
+<p>
+In theory, you could enter the RCU read-side critical section first,
+but it is more efficient to keep the entire RCU read-side critical
+section contained in the preempt-disable region as shown above.
+Of course, RCU read-side critical sections that extend outside of
+preempt-disable regions will work correctly, but such critical sections
+can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
+more work.
+And no, this is <i>not</i> an invitation to enclose all of your RCU
+read-side critical sections within preempt-disable regions, because
+doing so would degrade real-time response.
+
+<p>
+This non-requirement appeared with preemptible RCU.
+If you need a grace period that waits on non-preemptible code regions, use
+<a href="#Sched Flavor">RCU-sched</a>.
+
+<h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
+
+<p>
+These parallelism facts of life are by no means specific to RCU, but
+the RCU implementation must abide by them.
+They therefore bear repeating:
+
+<ol>
+<li> Any CPU or task may be delayed at any time,
+ and any attempts to avoid these delays by disabling
+ preemption, interrupts, or whatever are completely futile.
+ This is most obvious in preemptible user-level
+ environments and in virtualized environments (where
+ a given guest OS's VCPUs can be preempted at any time by
+ the underlying hypervisor), but can also happen in bare-metal
+ environments due to ECC errors, NMIs, and other hardware
+ events.
+ Although a delay of more than about 20 seconds can result
+ in splats, the RCU implementation is obligated to use
+ algorithms that can tolerate extremely long delays, but where
+ &ldquo;extremely long&rdquo; is not long enough to allow
+ wrap-around when incrementing a 64-bit counter.
+<li> Both the compiler and the CPU can reorder memory accesses.
+ Where it matters, RCU must use compiler directives and
+ memory-barrier instructions to preserve ordering.
+<li> Conflicting writes to memory locations in any given cache line
+ will result in expensive cache misses.
+ Greater numbers of concurrent writes and more-frequent
+ concurrent writes will result in more dramatic slowdowns.
+ RCU is therefore obligated to use algorithms that have
+ sufficient locality to avoid significant performance and
+ scalability problems.
+<li> As a rough rule of thumb, only one CPU's worth of processing
+ may be carried out under the protection of any given exclusive
+ lock.
+ RCU must therefore use scalable locking designs.
+<li> Counters are finite, especially on 32-bit systems.
+ RCU's use of counters must therefore tolerate counter wrap,
+ or be designed such that counter wrap would take way more
+ time than a single system is likely to run.
+ An uptime of ten years is quite possible, a runtime
+ of a century much less so.
+ As an example of the latter, RCU's dyntick-idle nesting counter
+ allows 54 bits for interrupt nesting level (this counter
+ is 64 bits even on a 32-bit system).
+ Overflowing this counter requires 2<sup>54</sup>
+ half-interrupts on a given CPU without that CPU ever going idle.
+ If a half-interrupt happened every microsecond, it would take
+ 570 years of runtime to overflow this counter, which is currently
+ believed to be an acceptably long time.
+<li> Linux systems can have thousands of CPUs running a single
+ Linux kernel in a single shared-memory environment.
+ RCU must therefore pay close attention to high-end scalability.
+</ol>
+
+<p>
+This last parallelism fact of life means that RCU must pay special
+attention to the preceding facts of life.
+The idea that Linux might scale to systems with thousands of CPUs would
+have been met with some skepticism in the 1990s, but these requirements
+would have otherwise have been unsurprising, even in the early 1990s.
+
+<h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
+
+<p>
+These sections list quality-of-implementation requirements.
+Although an RCU implementation that ignores these requirements could
+still be used, it would likely be subject to limitations that would
+make it inappropriate for industrial-strength production use.
+Classes of quality-of-implementation requirements are as follows:
+
+<ol>
+<li> <a href="#Specialization">Specialization</a>
+<li> <a href="#Performance and Scalability">Performance and Scalability</a>
+<li> <a href="#Composability">Composability</a>
+<li> <a href="#Corner Cases">Corner Cases</a>
+</ol>
+
+<p>
+These classes is covered in the following sections.
+
+<h3><a name="Specialization">Specialization</a></h3>
+
+<p>
+RCU is and always has been intended primarily for read-mostly situations, as
+illustrated by the following figure.
+This means that RCU's read-side primitives are optimized, often at the
+expense of its update-side primitives.
+
+<p><img src="RCUApplicability.svg" alt="RCUApplicability.svg" width="70%"></p>
+
+<p>
+This focus on read-mostly situations means that RCU must interoperate
+with other synchronization primitives.
+For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
+examples discussed earlier use RCU to protect readers and locking to
+coordinate updaters.
+However, the need extends much farther, requiring that a variety of
+synchronization primitives be legal within RCU read-side critical sections,
+including spinlocks, sequence locks, atomic operations, reference
+counters, and memory barriers.
+
+<p>@@QQ@@
+What about sleeping locks?
+<p>@@QQA@@
+These are forbidden within Linux-kernel RCU read-side critical sections
+because it is not legal to place a quiescent state (in this case,
+voluntary context switch) within an RCU read-side critical section.
+However, sleeping locks may be used within userspace RCU read-side critical
+sections, and also within Linux-kernel sleepable RCU
+<a href="#Sleepable RCU">(SRCU)</a>
+read-side critical sections.
+In addition, the -rt patchset turns spinlocks into a sleeping locks so
+that the corresponding critical sections can be preempted, which
+also means that these sleeplockified spinlocks (but not other sleeping locks!)
+may be acquire within -rt-Linux-kernel RCU read-side critical sections.
+
+<p>
+Note that it <i>is</i> legal for a normal RCU read-side critical section
+to conditionally acquire a sleeping locks (as in <tt>mutex_trylock()</tt>),
+but only as long as it does not loop indefinitely attempting to
+conditionally acquire that sleeping locks.
+The key point is that things like <tt>mutex_trylock()</tt>
+either return with the mutex held, or return an error indication if
+the mutex was not immediately available.
+Either way, <tt>mutex_trylock()</tt> returns immediately without sleeping.
+<p>@@QQE@@
+
+<p>
+It often comes as a surprise that many algorithms do not require a
+consistent view of data, but many can function in that mode,
+with network routing being the poster child.
+Internet routing algorithms take significant time to propagate
+updates, so that by the time an update arrives at a given system,
+that system has been sending network traffic the wrong way for
+a considerable length of time.
+Having a few threads continue to send traffic the wrong way for a
+few more milliseconds is clearly not a problem: In the worst case,
+TCP retransmissions will eventually get the data where it needs to go.
+In general, when tracking the state of the universe outside of the
+computer, some level of inconsistency must be tolerated due to
+speed-of-light delays if nothing else.
+
+<p>
+Furthermore, uncertainty about external state is inherent in many cases.
+For example, a pair of veternarians might use heartbeat to determine
+whether or not a given cat was alive.
+But how long should they wait after the last heartbeat to decide that
+the cat is in fact dead?
+Waiting less than 400 milliseconds makes no sense because this would
+mean that a relaxed cat would be considered to cycle between death
+and life more than 100 times per minute.
+Moreover, just as with human beings, a cat's heart might stop for
+some period of time, so the exact wait period is a judgment call.
+One of our pair of veternarians might wait 30 seconds before pronouncing
+the cat dead, while the other might insist on waiting a full minute.
+The two veternarians would then disagree on the state of the cat during
+the final 30 seconds of the minute following the last heartbeat, as
+fancifully illustrated below:
+
+<p><img src="2013-08-is-it-dead.png" alt="2013-08-is-it-dead.png" width="431"></p>
+
+<p>
+Interestingly enough, this same situation applies to hardware.
+When push comes to shove, how do we tell whether or not some
+external server has failed?
+We send messages to it periodically, and declare it failed if we
+don't receive a response within a given period of time.
+Policy decisions can usually tolerate short
+periods of inconsistency.
+The policy was decided some time ago, and is only now being put into
+effect, so a few milliseconds of delay is normally inconsequential.
+
+<p>
+However, there are algorithms that absolutely must see consistent data.
+For example, the translation between a user-level SystemV semaphore
+ID to the corresponding in-kernel data structure is protected by RCU,
+but it is absolutely forbidden to update a semaphore that has just been
+removed.
+In the Linux kernel, this need for consistency is accommodated by acquiring
+spinlocks located in the in-kernel data structure from within
+the RCU read-side critical section, and this is indicated by the
+green box in the figure above.
+Many other techniques may be used, and are in fact used within the
+Linux kernel.
+
+<p>
+In short, RCU is not required to maintain consistency, and other
+mechanisms may be used in concert with RCU when consistency is required.
+RCU's specialization allows it to do its job extremely well, and its
+ability to interoperate with other synchronization mechanisms allows
+the right mix of synchronization tools to be used for a given job.
+
+<h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
+
+<p>
+Energy efficiency is a critical component of performance today,
+and Linux-kernel RCU implementations must therefore avoid unnecessarily
+awakening idle CPUs.
+I cannot claim that this requirement was premeditated.
+In fact, I learned of it during a telephone conversation in which I
+was given &ldquo;frank and open&rdquo; feedback on the importance
+of energy efficiency in battery-powered systems and on specific
+energy-efficiency shortcomings of the Linux-kernel RCU implementation.
+In my experience, the battery-powered embedded community will consider
+any unnecessary wakeups to be extremely unfriendly acts.
+So much so that mere Linux-kernel-mailing-list posts are
+insufficient to vent their ire.
+
+<p>
+Memory consumption is not particularly important for in most
+situations, and has become decreasingly
+so as memory sizes have expanded and memory
+costs have plummeted.
+However, as I learned from Matt Mackall's
+<a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
+efforts, memory footprint is critically important on single-CPU systems with
+non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
+<a href="http://lkml.org/lkml/2009/1/14/449">tiny RCU</a>
+was born.
+Josh Triplett has since taken over the small-memory banner with his
+<a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
+project, which resulted in
+<a href="#Sleepable RCU">SRCU</a>
+becoming optional for those kernels not needing it.
+
+<p>
+The remaining performance requirements are, for the most part,
+unsurprising.
+For example, in keeping with RCU's read-side specialization,
+<tt>rcu_dereference()</tt> should have negligible overhead (for
+example, suppression of a few minor compiler optimizations).
+Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have exactly zero overhead.
+
+<p>
+In preemptible environments, in the case where the RCU read-side
+critical section was not preempted (as will be the case for the
+highest-priority real-time process), <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have minimal overhead.
+In particular, they should not contain atomic read-modify-write
+operations, memory-barrier instructions, preemption disabling,
+interrupt disabling, or backwards branches.
+However, in the case where the RCU read-side critical section was preempted,
+<tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
+This is why it is better to nest an RCU read-side critical section
+within a preempt-disable region than vice versa, at least in cases
+where that critical section is short enough to avoid unduly degrading
+real-time latencies.
+
+<p>
+The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
+optimized for throughput.
+It may therefore incur several milliseconds of latency in addition to
+the duration of the longest RCU read-side critical section.
+On the other hand, multiple concurrent invocations of
+<tt>synchronize_rcu()</tt> are required to use batching optimizations
+so that they can be satisfied by a single underlying grace-period-wait
+operation.
+For example, in the Linux kernel, it is not unusual for a single
+grace-period-wait operation to serve more than
+<a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
+of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
+overhead down to nearly zero.
+However, the grace-period optimization is also required to avoid
+measurable degradation of real-time scheduling and interrupt latencies.
+
+<p>
+In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
+latencies are unacceptable.
+In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
+instead, reducing the grace-period latency down to a few tens of
+microseconds on small systems, at least in cases where the RCU read-side
+critical sections are short.
+There are currently no special latency requirements for
+<tt>synchronize_rcu_expedited()</tt> on large systems, but,
+consistent with the empirical nature of the RCU specification,
+that is subject to change.
+However, there most definitely are scalability requirements:
+A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
+CPUs should at least make reasonable forward progress.
+In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
+is permitted to impose modest degradation of real-time latency
+on non-idle online CPUs.
+That said, it will likely be necessary to take further steps to reduce this
+degradation, hopefully to roughly that of a scheduling-clock interrupt.
+
+<p>
+There are a number of situations where even
+<tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
+latency is unacceptable.
+In these situations, the asynchronous <tt>call_rcu()</tt> can be
+used in place of <tt>synchronize_rcu()</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2 int a;
+ 3 int b;
+ 4 struct rcu_head rh;
+ 5 };
+ 6
+ 7 static void remove_gp_cb(struct rcu_head *rhp)
+ 8 {
+ 9 struct foo *p = container_of(rhp, struct foo, rh);
+10
+11 kfree(p);
+12 }
+13
+14 bool remove_gp_asynchronous(void)
+15 {
+16 struct foo *p;
+17
+18 spin_lock(&amp;gp_lock);
+19 p = rcu_dereference(gp);
+20 if (!p) {
+21 spin_unlock(&amp;gp_lock);
+22 return false;
+23 }
+24 rcu_assign_pointer(gp, NULL);
+25 call_rcu(&amp;p-&gt;rh, remove_gp_cb);
+26 spin_unlock(&amp;gp_lock);
+27 return true;
+28 }
+</pre>
+</blockquote>
+
+<p>
+A definition of <tt>struct foo</tt> is finally needed, and appears
+on lines&nbsp;1-5.
+The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
+on line&nbsp;25, and will be invoked after the end of a subsequent
+grace period.
+This gets the same effect as <tt>remove_gp_synchronous()</tt>,
+but without forcing the updater to wait for a grace period to elapse.
+The <tt>call_rcu()</tt> function may be used in a number of
+situations where neither <tt>synchronize_rcu()</tt> nor
+<tt>synchronize_rcu_expedited()</tt> would be legal,
+including within preempt-disable code, <tt>local_bh_disable()</tt> code,
+interrupt-disable code, and interrupt handlers.
+However, even <tt>call_rcu()</tt> is illegal within NMI handlers.
+The callback function (<tt>remove_gp_cb()</tt> in this case) will be
+executed within softirq (software interrupt) environment within the
+Linux kernel,
+either within a real softirq handler or under the protection
+of <tt>local_bh_disable()</tt>.
+In both the Linux kernel and in userspace, it is bad practice to
+write an RCU callback function that takes too long.
+Long-running operations should be relegated to separate threads or
+(in the Linux kernel) workqueues.
+
+<p>@@QQ@@
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+<p>@@QQA@@
+Presumably the <tt>-&gt;gp_lock</tt> acquired on line&nbsp;18 excludes
+any changes, including any insertions that <tt>rcu_dereference()</tt>
+would protect against.
+Therefore, any insertions will be delayed until after <tt>-&gt;gp_lock</tt>
+is released on line&nbsp;25, which in turn means that
+<tt>rcu_access_pointer()</tt> suffices.
+<p>@@QQE@@
+
+<p>
+However, all that <tt>remove_gp_cb()</tt> is doing is
+invoking <tt>kfree()</tt> on the data element.
+This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
+which allows &ldquo;fire and forget&rdquo; operation as shown below:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2 int a;
+ 3 int b;
+ 4 struct rcu_head rh;
+ 5 };
+ 6
+ 7 bool remove_gp_faf(void)
+ 8 {
+ 9 struct foo *p;
+10
+11 spin_lock(&amp;gp_lock);
+12 p = rcu_dereference(gp);
+13 if (!p) {
+14 spin_unlock(&amp;gp_lock);
+15 return false;
+16 }
+17 rcu_assign_pointer(gp, NULL);
+18 kfree_rcu(p, rh);
+19 spin_unlock(&amp;gp_lock);
+20 return true;
+21 }
+</pre>
+</blockquote>
+
+<p>
+Note that <tt>remove_gp_faf()</tt> simply invokes
+<tt>kfree_rcu()</tt> and proceeds, without any need to pay any
+further attention to the subsequent grace period and <tt>kfree()</tt>.
+It is permissible to invoke <tt>kfree_rcu()</tt> from the same
+environments as for <tt>call_rcu()</tt>.
+Interestingly enough, DYNIX/ptx had the equivalents of
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
+<tt>synchronize_rcu()</tt>.
+This was due to the fact that RCU was not heavily used within DYNIX/ptx,
+so the very few places that needed something like
+<tt>synchronize_rcu()</tt> simply open-coded it.
+
+<p>@@QQ@@
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+<p>@@QQA@@
+We could define things this way, but keep in mind that this sort of
+definition would say that updates in garbage-collected languages
+cannot complete until the next time the garbage collector runs,
+which does not seem at all reasonable.
+The key point is that in most cases, an updater using either
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
+next update as soon as it has invoked <tt>call_rcu()</tt> or
+<tt>kfree_rcu()</tt>, without having to wait for a subsequent
+grace period.
+<p>@@QQE@@
+
+<p>
+But what if the updater must wait for the completion of code to be
+executed after the end of the grace period, but has other tasks
+that can be carried out in the meantime?
+The polling-style <tt>get_state_synchronize_rcu()</tt> and
+<tt>cond_synchronize_rcu()</tt> functions may be used for this
+purpose, as shown below:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_poll(void)
+ 2 {
+ 3 struct foo *p;
+ 4 unsigned long s;
+ 5
+ 6 spin_lock(&amp;gp_lock);
+ 7 p = rcu_access_pointer(gp);
+ 8 if (!p) {
+ 9 spin_unlock(&amp;gp_lock);
+10 return false;
+11 }
+12 rcu_assign_pointer(gp, NULL);
+13 spin_unlock(&amp;gp_lock);
+14 s = get_state_synchronize_rcu();
+15 do_something_while_waiting();
+16 cond_synchronize_rcu(s);
+17 kfree(p);
+18 return true;
+19 }
+</pre>
+</blockquote>
+
+<p>
+On line&nbsp;14, <tt>get_state_synchronize_rcu()</tt> obtains a
+&ldquo;cookie&rdquo; from RCU,
+then line&nbsp;15 carries out other tasks,
+and finally, line&nbsp;16 returns immediately if a grace period has
+elapsed in the meantime, but otherwise waits as required.
+The need for <tt>get_state_synchronize_rcu</tt> and
+<tt>cond_synchronize_rcu()</tt> has appeared quite recently,
+so it is too early to tell whether they will stand the test of time.
+
+<p>
+RCU thus provides a range of tools to allow updaters to strike the
+required tradeoff between latency, flexibility and CPU overhead.
+
+<h3><a name="Composability">Composability</a></h3>
+
+<p>
+Composability has received much attention in recent years, perhaps in part
+due to the collision of multicore hardware with object-oriented techniques
+designed in single-threaded environments for single-threaded use.
+And in theory, RCU read-side critical sections may be composed, and in
+fact may be nested arbitrarily deeply.
+In practice, as with all real-world implementations of composable
+constructs, there are limitations.
+
+<p>
+Implementations of RCU for which <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> generate no code, such as
+Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
+nested arbitrarily deeply.
+After all, there is no overhead.
+Except that if all these instances of <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> are visible to the compiler,
+compilation will eventually fail due to exhausting memory,
+mass storage, or user patience, whichever comes first.
+If the nesting is not visible to the compiler, as is the case with
+mutually recursive functions each in its own translation unit,
+stack overflow will result.
+If the nesting takes the form of loops, either the control variable
+will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
+Nevertheless, this class of RCU implementations is one
+of the most composable constructs in existence.
+
+<p>
+RCU implementations that explicitly track nesting depth
+are limited by the nesting-depth counter.
+For example, the Linux kernel's preemptible RCU limits nesting to
+<tt>INT_MAX</tt>.
+This should suffice for almost all practical purposes.
+That said, a consecutive pair of RCU read-side critical sections
+between which there is an operation that waits for a grace period
+cannot be enclosed in another RCU read-side critical section.
+This is because it is not legal to wait for a grace period within
+an RCU read-side critical section: To do so would result either
+in deadlock or
+in RCU implicitly splitting the enclosing RCU read-side critical
+section, neither of which is conducive to a long-lived and prosperous
+kernel.
+
+<p>
+In short, although RCU read-side critical sections are highly composable,
+care is required in some situations, just as is the case for any other
+composable synchronization mechanism.
+
+<h3><a name="Corner Cases">Corner Cases</a></h3>
+
+<p>
+A given RCU workload might have an endless and intense stream of
+RCU read-side critical sections, perhaps even so intense that there
+was never a point in time during which there was not at least one
+RCU read-side critical section in flight.
+RCU cannot allow this situation to block grace periods: As long as
+all the RCU read-side critical sections are finite, grace periods
+must also be finite.
+
+<p>
+That said, preemptible RCU implementations could potentially result
+in RCU read-side critical sections being preempted for long durations,
+which has the effect of creating a long-duration RCU read-side
+critical section.
+This situation can arise only in heavily loaded systems, but systems using
+real-time priorities are of course more vulnerable.
+Therefore, RCU priority boosting is provided to help deal with this
+case.
+That said, the exact requirements on RCU priority boosting will likely
+evolve as more experience accumulates.
+
+<p>
+Other workloads might have very high update rates.
+Although one can argue that such workloads should instead use
+something other than RCU, the fact remains that RCU must
+handle such workloads gracefully.
+This requirement is another factor driving batching of grace periods,
+but it is also the driving force behind the checks for large numbers
+of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
+Finally, high update rates should not delay RCU read-side critical
+sections, although some read-side delays can occur when using
+<tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
+of <tt>try_stop_cpus()</tt>.
+(In the future, <tt>synchronize_rcu_expedited()</tt> will be
+converted to use lighter-weight inter-processor interrupts (IPIs),
+but this will still disturb readers, though to a much smaller degree.)
+
+<p>
+Although all three of these corner cases were understood in the early
+1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
+in a tight loop
+in the early 2000s suddenly provided a much deeper appreciation of the
+high-update-rate corner case.
+This test also motivated addition of some RCU code to react to high update
+rates, for example, if a given CPU finds itself with more than 10,000
+RCU callbacks queued, it will cause RCU to take evasive action by
+more aggressively starting grace periods and more aggressively forcing
+completion of grace-period processing.
+This evasive action causes the grace period to complete more quickly,
+but at the cost of restricting RCU's batching optimizations, thus
+increasing the CPU overhead incurred by that grace period.
+
+<h2><a name="Software-Engineering Requirements">
+Software-Engineering Requirements</a></h2>
+
+<p>
+Between Murphy's Law and &ldquo;To err is human&rdquo;, it is necessary to
+guard against mishaps and misuse:
+
+<ol>
+<li> It is all too easy to forget to use <tt>rcu_read_lock()</tt>
+ everywhere that it is needed, so kernels built with
+ <tt>CONFIG_PROVE_RCU=y</tt> will spat if
+ <tt>rcu_dereference()</tt> is used outside of an
+ RCU read-side critical section.
+ Update-side code can use <tt>rcu_dereference_protected()</tt>,
+ which takes a
+ <a href="https://lwn.net/Articles/371986/">lockdep expression</a>
+ to indicate what is providing the protection.
+ If the indicated protection is not provided, a lockdep splat
+ is emitted.
+
+ <p>
+ Code shared between readers and updaters can use
+ <tt>rcu_dereference_check()</tt>, which also takes a
+ lockdep expression, and emits a lockdep splat if neither
+ <tt>rcu_read_lock()</tt> nor the indicated protection
+ is in place.
+ In addition, <tt>rcu_dereference_raw()</tt> is used in those
+ (hopefully rare) cases where the required protection cannot
+ be easily described.
+ Finally, <tt>rcu_read_lock_held()</tt> is provided to
+ allow a function to verify that it has been invoked within
+ an RCU read-side critical section.
+ I was made aware of this set of requirements shortly after Thomas
+ Gleixner audited a number of RCU uses.
+<li> A given function might wish to check for RCU-related preconditions
+ upon entry, before using any other RCU API.
+ The <tt>rcu_lockdep_assert()</tt> does this job,
+ asserting the expression in kernels having lockdep enabled
+ and doing nothing otherwise.
+<li> It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
+ and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
+ substituting a simple assignment.
+ To catch this sort of error, a given RCU-protected pointer may be
+ tagged with <tt>__rcu</tt>, after which running sparse
+ with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
+ about simple-assignment accesses to that pointer.
+ Arnd Bergmann made me aware of this requirement, and also
+ supplied the needed
+ <a href="http://lwn.net/Articles/376011/">patch series</a>.
+<li> Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
+ will splat if a data element is passed to <tt>call_rcu()</tt>
+ twice in a row, without a grace period in between.
+ (This error is similar to a double free.)
+ The corresponding <tt>rcu_head</tt> structures that are
+ dynamically allocated are automatically tracked, but
+ <tt>rcu_head</tt> structures allocated on the stack
+ must be initialized with <tt>init_rcu_head_on_stack()</tt>
+ and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
+ Similarly, statically allocated non-stack <tt>rcu_head</tt>
+ structures must be initialized with <tt>init_rcu_head()</tt>
+ and cleaned up with <tt>destroy_rcu_head()</tt>.
+ Mathieu Desnoyers made me aware of this requirement, and also
+ supplied the needed
+ <a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
+<li> An infinite loop in an RCU read-side critical section will
+ eventually trigger an RCU CPU stall warning splat.
+ However, RCU is not obligated to produce this splat
+ unless there is a grace period waiting on that particular
+ RCU read-side critical section.
+ This requirement made itself known in the early 1990s, pretty
+ much the first time that it was necessary to debug a CPU stall.
+<li> Although it would be very good to detect pointers leaking out
+ of RCU read-side critical sections, there is currently no
+ good way of doing this.
+ One complication is the need to distinguish between pointers
+ leaking and pointers that have been handed off from RCU to
+ some other synchronization mechanism, for example, reference
+ counting.
+<li> In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
+ information is provided via both debugfs and event tracing.
+<li> Open-coded use of <tt>rcu_assign_pointer()</tt> and
+ <tt>rcu_dereference()</tt> to create typical linked
+ data structures can be surprisingly error-prone.
+ Therefore, RCU-protected
+ <a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
+ and, more recently, RCU-protected
+ <a href="https://lwn.net/Articles/612100/">hash tables</a>
+ are available.
+ Many other special-purpose RCU-protected data structures are
+ available in the Linux kernel and the userspace RCU library.
+<li> Some linked structures are created at compile time, but still
+ require <tt>__rcu</tt> checking.
+ The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
+ purpose.
+<li> It is not necessary to use <tt>rcu_assign_pointer()</tt>
+ when creating linked structures that are to be published via
+ a single external pointer.
+ The <tt>RCU_INIT_POINTER()</tt> macro is provided for
+ this task and also for assigning <tt>NULL</tt> pointers
+ at runtime.
+</ol>
+
+<p>
+This not a hard-and-fast list: RCU's diagnostic capabilities will
+continue to be guided by the number and type of usage bugs found
+in real-world RCU usage.
+
+<h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
+
+<p>
+The Linux kernel provides an interesting environment for all kinds of
+software, including RCU.
+Some of the relevant points of interest are as follows:
+
+<ol>
+<li> <a href="#Configuration">Configuration</a>.
+<li> <a href="#Firmware Interface">Firmware Interface</a>.
+<li> <a href="#Early Boot">Early Boot</a>.
+<li> <a href="#Interrupts and NMIs">
+ Interrupts and non-maskable interrupts (NMIs)</a>.
+<li> <a href="#Loadable Modules">Loadable Modules</a>.
+<li> <a href="#Hotplug CPU">Hotplug CPU</a>.
+<li> <a href="#Scheduler and RCU">Scheduler and RCU</a>.
+<li> <a href="#Tracing and RCU">Tracing and RCU</a>.
+<li> <a href="#Energy Efficiency">Energy Efficiency</a>.
+<li> <a href="#Performance, Scalability, Response Time, and Reliability">
+ Performance, Scalability, Response Time, and Reliability</a>.
+</ol>
+
+<p>
+This list is probably incomplete, but it does give a feel for the
+most notable Linux-kernel complications.
+Each of the following sections covers one of the above topics.
+
+<h3><a name="Configuration">Configuration</a></h3>
+
+<p>
+RCU's goal is automatic configuration, so that almost nobody
+needs to worry about RCU's <tt>Kconfig</tt> options.
+And for almost all users, RCU does in fact work well
+&ldquo;out of the box.&rdquo;
+
+<p>
+However, there are specialized use cases that are handled by
+kernel boot parameters and <tt>Kconfig</tt> options.
+Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
+about new <tt>Kconfig</tt> options, which requires almost all of them
+be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
+
+<p>
+This all should be quite obvious, but the fact remains that
+Linus Torvalds recently had to
+<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
+me of this requirement.
+
+<h3><a name="Firmware Interface">Firmware Interface</a></h3>
+
+<p>
+In many cases, kernel obtains information about the system from the
+firmware, and sometimes things are lost in translation.
+Or the translation is accurate, but the original message is bogus.
+
+<p>
+For example, some systems' firmware overreports the number of CPUs,
+sometimes by a large factor.
+If RCU naively believed the firmware, as it used to do,
+it would create too many per-CPU kthreads.
+Although the resulting system will still run correctly, the extra
+kthreads needlessly consume memory and can cause confusion
+when they show up in <tt>ps</tt> listings.
+
+<p>
+RCU must therefore wait for a given CPU to actually come online before
+it can allow itself to believe that the CPU actually exists.
+The resulting &ldquo;ghost CPUs&rdquo; (which are never going to
+come online) cause a number of
+<a href="http://paulmck.livejournal.com/37494.html">interesting complications</a>.
+
+<h3><a name="Early Boot">Early Boot</a></h3>
+
+<p>
+The Linux kernel's boot sequence is an interesting process,
+and RCU is used early, even before <tt>rcu_init()</tt>
+is invoked.
+In fact, a number of RCU's primitives can be used as soon as the
+initial task's <tt>task_struct</tt> is available and the
+boot CPU's per-CPU variables are set up.
+The read-side primitives (<tt>rcu_read_lock()</tt>,
+<tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
+and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
+as will <tt>rcu_assign_pointer()</tt>.
+
+<p>
+Although <tt>call_rcu()</tt> may be invoked at any
+time during boot, callbacks are not guaranteed to be invoked until after
+the scheduler is fully up and running.
+This delay in callback invocation is due to the fact that RCU does not
+invoke callbacks until it is fully initialized, and this full initialization
+cannot occur until after the scheduler has initialized itself to the
+point where RCU can spawn and run its kthreads.
+In theory, it would be possible to invoke callbacks earlier,
+however, this is not a panacea because there would be severe restrictions
+on what operations those callbacks could invoke.
+
+<p>
+Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
+<a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
+(<a href="#Bottom-Half Flavor">discussed below</a>),
+and
+<a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
+will all operate normally
+during very early boot, the reason being that there is only one CPU
+and preemption is disabled.
+This means that the call <tt>synchronize_rcu()</tt> (or friends)
+itself is a quiescent
+state and thus a grace period, so the early-boot implementation can
+be a no-op.
+
+<p>
+Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
+continue to operate normally through the remainder of boot, courtesy
+of the fact that preemption is disabled across their RCU read-side
+critical sections and also courtesy of the fact that there is still
+only one CPU.
+However, once the scheduler starts initializing, preemption is enabled.
+There is still only a single CPU, but the fact that preemption is enabled
+means that the no-op implementation of <tt>synchronize_rcu()</tt> no
+longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
+Therefore, as soon as the scheduler starts initializing, the early-boot
+fastpath is disabled.
+This means that <tt>synchronize_rcu()</tt> switches to its runtime
+mode of operation where it posts callbacks, which in turn means that
+any call to <tt>synchronize_rcu()</tt> will block until the corresponding
+callback is invoked.
+Unfortunately, the callback cannot be invoked until RCU's runtime
+grace-period machinery is up and running, which cannot happen until
+the scheduler has initialized itself sufficiently to allow RCU's
+kthreads to be spawned.
+Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
+initialization can result in deadlock.
+
+<p>@@QQ@@
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+<p>@@QQA@@
+In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
+maps directly to <tt>synchronize_sched()</tt>.
+Therefore, <tt>synchronize_rcu()</tt> works normally throughout
+boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
+However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
+so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
+during scheduler initialization.
+<p>@@QQE@@
+
+<p>
+I learned of these boot-time requirements as a result of a series of
+system hangs.
+
+<h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
+
+<p>
+The Linux kernel has interrupts, and RCU read-side critical sections are
+legal within interrupt handlers and within interrupt-disabled regions
+of code, as are invocations of <tt>call_rcu()</tt>.
+
+<p>
+Some Linux-kernel architectures can enter an interrupt handler from
+non-idle process context, and then just never leave it, instead stealthily
+transitioning back to process context.
+This trick is sometimes used to invoke system calls from inside the kernel.
+These &ldquo;half-interrupts&rdquo; mean that RCU has to be very careful
+about how it counts interrupt nesting levels.
+I learned of this requirement the hard way during a rewrite
+of RCU's dyntick-idle code.
+
+<p>
+The Linux kernel has non-maskable interrupts (NMIs), and
+RCU read-side critical sections are legal within NMI handlers.
+Thankfully, RCU update-side primitives, including
+<tt>call_rcu()</tt>, are prohibited within NMI handlers.
+
+<p>
+The name notwithstanding, some Linux-kernel architectures
+can have nested NMIs, which RCU must handle correctly.
+Andy Lutomirski
+<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
+with this requirement;
+he also kindly surprised me with
+<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
+that meets this requirement.
+
+<h3><a name="Loadable Modules">Loadable Modules</a></h3>
+
+<p>
+The Linux kernel has loadable modules, and these modules can
+also be unloaded.
+After a given module has been unloaded, any attempt to call
+one of its functions results in a segmentation fault.
+The module-unload functions must therefore cancel any
+delayed calls to loadable-module functions, for example,
+any outstanding <tt>mod_timer()</tt> must be dealt with
+via <tt>del_timer_sync()</tt> or similar.
+
+<p>
+Unfortunately, there is no way to cancel an RCU callback;
+once you invoke <tt>call_rcu()</tt>, the callback function is
+going to eventually be invoked, unless the system goes down first.
+Because it is normally considered socially irresponsible to crash the system
+in response to a module unload request, we need some other way
+to deal with in-flight RCU callbacks.
+
+<p>
+RCU therefore provides
+<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
+which waits until all in-flight RCU callbacks have been invoked.
+If a module uses <tt>call_rcu()</tt>, its exit function should therefore
+prevent any future invocation of <tt>call_rcu()</tt>, then invoke
+<tt>rcu_barrier()</tt>.
+In theory, the underlying module-unload code could invoke
+<tt>rcu_barrier()</tt> unconditionally, but in practice this would
+incur unacceptable latencies.
+
+<p>
+Nikita Danilov noted this requirement for an analogous filesystem-unmount
+situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
+The need for <tt>rcu_barrier()</tt> for module unloading became
+apparent later.
+
+<h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
+
+<p>
+The Linux kernel supports CPU hotplug, which means that CPUs
+can come and go.
+It is of course illegal to use any RCU API member from an offline CPU.
+This requirement was present from day one in DYNIX/ptx, but
+on the other hand, the Linux kernel's CPU-hotplug implementation
+is &ldquo;interesting.&rdquo;
+
+<p>
+The Linux-kernel CPU-hotplug implementation has notifiers that
+are used to allow the various kernel subsystems (including RCU)
+to respond appropriately to a given CPU-hotplug operation.
+Most RCU operations may be invoked from CPU-hotplug notifiers,
+including even normal synchronous grace-period operations
+such as <tt>synchronize_rcu()</tt>.
+However, expedited grace-period operations such as
+<tt>synchronize_rcu_expedited()</tt> are not supported,
+due to the fact that current implementations block CPU-hotplug
+operations, which could result in deadlock.
+
+<p>
+In addition, all-callback-wait operations such as
+<tt>rcu_barrier()</tt> are also not supported, due to the
+fact that there are phases of CPU-hotplug operations where
+the outgoing CPU's callbacks will not be invoked until after
+the CPU-hotplug operation ends, which could also result in deadlock.
+
+<h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
+
+<p>
+RCU depends on the scheduler, and the scheduler uses RCU to
+protect some of its data structures.
+This means the scheduler is forbidden from acquiring
+the runqueue locks and the priority-inheritance locks
+in the middle of an outermost RCU read-side critical section unless
+it also releases them before exiting that same
+RCU read-side critical section.
+This same prohibition also applies to any lock that is acquired
+while holding any lock to which this prohibition applies.
+Violating this rule results in deadlock.
+
+<p>
+For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
+implementation must be written carefully to avoid similar deadlocks.
+In particular, <tt>rcu_read_unlock()</tt> must tolerate an
+interrupt where the interrupt handler invokes both
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+This possibility requires <tt>rcu_read_unlock()</tt> to use
+negative nesting levels to avoid destructive recursion via
+interrupt handler's use of RCU.
+
+<p>
+This pair of mutual scheduler-RCU requirements came as a
+<a href="http://lwn.net/Articles/453002/">complete surprise</a>.
+
+<p>
+As noted above, RCU makes use of kthreads, and it is necessary to
+avoid excessive CPU-time accumulation by these kthreads.
+This requirement was no surprise, but RCU's violation of it
+when running context-switch-heavy workloads when built with
+<tt>CONFIG_NO_HZ_FULL=y</tt>
+<a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
+RCU has made good progress towards meeting this requirement, even
+for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
+but there is room for further improvement.
+
+<h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
+
+<p>
+It is possible to use tracing on RCU code, but tracing itself
+uses RCU.
+For this reason, <tt>rcu_dereference_raw_notrace()</tt>
+is provided for use by tracing, which avoids the destructive
+recursion that could otherwise ensue.
+This API is also used by virtualization in some architectures,
+where RCU readers execute in environments in which tracing
+cannot be used.
+The tracing folks both located the requirement and provided the
+needed fix, so this surprise requirement was relatively painless.
+
+<h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
+
+<p>
+Interrupting idle CPUs is considered socially unacceptable,
+especially by people with battery-powered embedded systems.
+RCU therefore conserves energy by detecting which CPUs are
+idle, including tracking CPUs that have been interrupted from idle.
+This is a large part of the energy-efficiency requirement,
+so I learned of this via an irate phone call.
+
+<p>
+Because RCU avoids interrupting idle CPUs, it is illegal to
+execute an RCU read-side critical section on an idle CPU.
+(Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
+if you try it.)
+The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
+event tracing is provided to work around this restriction.
+In addition, <tt>rcu_is_watching()</tt> may be used to
+test whether or not it is currently legal to run RCU read-side
+critical sections on this CPU.
+I learned of the need for diagnostics on the one hand
+and <tt>RCU_NONIDLE()</tt> on the other while inspecting
+idle-loop code.
+Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
+which is used quite heavily in the idle loop.
+
+<p>
+It is similarly socially unacceptable to interrupt an
+<tt>nohz_full</tt> CPU running in userspace.
+RCU must therefore track <tt>nohz_full</tt> userspace
+execution.
+And in
+<a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
+kernels, RCU must separately track idle CPUs on the one hand and
+CPUs that are either idle or executing in userspace on the other.
+In both cases, RCU must be able to sample state at two points in
+time, and be able to determine whether or not some other CPU spent
+any time idle and/or executing in userspace.
+
+<p>
+These energy-efficiency requirements have proven quite difficult to
+understand and to meet, for example, there have been more than five
+clean-sheet rewrites of RCU's energy-efficiency code, the last of
+which was finally able to demonstrate
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
+As noted earlier,
+I learned of many of these requirements via angry phone calls:
+Flaming me on the Linux-kernel mailing list was apparently not
+sufficient to fully vent their ire at RCU's energy-efficiency bugs!
+
+<h3><a name="Performance, Scalability, Response Time, and Reliability">
+Performance, Scalability, Response Time, and Reliability</a></h3>
+
+<p>
+Expanding on the
+<a href="#Performance and Scalability">earlier discussion</a>,
+RCU is used heavily by hot code paths in performance-critical
+portions of the Linux kernel's networking, security, virtualization,
+and scheduling code paths.
+RCU must therefore use efficient implementations, especially in its
+read-side primitives.
+To that end, it would be good if preemptible RCU's implementation
+of <tt>rcu_read_lock()</tt> could be inlined, however, doing
+this requires resolving <tt>#include</tt> issues with the
+<tt>task_struct</tt> structure.
+
+<p>
+The Linux kernel supports hardware configurations with up to
+4096 CPUs, which means that RCU must be extremely scalable.
+Algorithms that involve frequent acquisitions of global locks or
+frequent atomic operations on global variables simply cannot be
+tolerated within the RCU implementation.
+RCU therefore makes heavy use of a combining tree based on the
+<tt>rcu_node</tt> structure.
+RCU is required to tolerate all CPUs continuously invoking any
+combination of RCU's runtime primitives with minimal per-operation
+overhead.
+In fact, in many cases, increasing load must <i>decrease</i> the
+per-operation overhead, witness the batching optimizations for
+<tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
+<tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
+As a general rule, RCU must cheerfully accept whatever the
+rest of the Linux kernel decides to throw at it.
+
+<p>
+The Linux kernel is used for real-time workloads, especially
+in conjunction with the
+<a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
+The real-time-latency response requirements are such that the
+traditional approach of disabling preemption across RCU
+read-side critical sections is inappropriate.
+Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
+use an RCU implementation that allows RCU read-side critical
+sections to be preempted.
+This requirement made its presence known after users made it
+clear that an earlier
+<a href="http://lwn.net/Articles/107930/">real-time patch</a>
+did not meet their needs, in conjunction with some
+<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>
+encountered by a very early version of the -rt patchset.
+
+<p>
+In addition, RCU must make do with a sub-100-microsecond real-time latency
+budget.
+In fact, on smaller systems with the -rt patchset, the Linux kernel
+provides sub-20-microsecond real-time latencies for the whole kernel,
+including RCU.
+RCU's scalability and latency must therefore be sufficient for
+these sorts of configurations.
+To my surprise, the sub-100-microsecond real-time latency budget
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
+applies to even the largest systems [PDF]</a>,
+up to and including systems with 4096 CPUs.
+This real-time requirement motivated the grace-period kthread, which
+also simplified handling of a number of race conditions.
+
+<p>
+Finally, RCU's status as a synchronization primitive means that
+any RCU failure can result in arbitrary memory corruption that can be
+extremely difficult to debug.
+This means that RCU must be extremely reliable, which in
+practice also means that RCU must have an aggressive stress-test
+suite.
+This stress-test suite is called <tt>rcutorture</tt>.
+
+<p>
+Although the need for <tt>rcutorture</tt> was no surprise,
+the current immense popularity of the Linux kernel is posing
+interesting&mdash;and perhaps unprecedented&mdash;validation
+challenges.
+To see this, keep in mind that there are well over one billion
+instances of the Linux kernel running today, given Android
+smartphones, Linux-powered televisions, and servers.
+This number can be expected to increase sharply with the advent of
+the celebrated Internet of Things.
+
+<p>
+Suppose that RCU contains a race condition that manifests on average
+once per million years of runtime.
+This bug will be occurring about three times per <i>day</i> across
+the installed base.
+RCU could simply hide behind hardware error rates, given that no one
+should really expect their smartphone to last for a million years.
+However, anyone taking too much comfort from this thought should
+consider the fact that in most jurisdictions, a successful multi-year
+test of a given mechanism, which might include a Linux kernel,
+suffices for a number of types of safety-critical certifications.
+In fact, rumor has it that the Linux kernel is already being used
+in production for safety-critical applications.
+I don't know about you, but I would feel quite bad if a bug in RCU
+killed someone.
+Which might explain my recent focus on validation and verification.
+
+<h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
+
+<p>
+One of the more surprising things about RCU is that there are now
+no fewer than five <i>flavors</i>, or API families.
+In addition, the primary flavor that has been the sole focus up to
+this point has two different implementations, non-preemptible and
+preemptible.
+The other four flavors are listed below, with requirements for each
+described in a separate section.
+
+<ol>
+<li> <a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
+<li> <a href="#Sched Flavor">Sched Flavor</a>
+<li> <a href="#Sleepable RCU">Sleepable RCU</a>
+<li> <a href="#Tasks RCU">Tasks RCU</a>
+</ol>
+
+<h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>
+
+<p>
+The softirq-disable (AKA &ldquo;bottom-half&rdquo;,
+hence the &ldquo;_bh&rdquo; abbreviations)
+flavor of RCU, or <i>RCU-bh</i>, was developed by
+Dipankar Sarma to provide a flavor of RCU that could withstand the
+network-based denial-of-service attacks researched by Robert
+Olsson.
+These attacks placed so much networking load on the system
+that some of the CPUs never exited softirq execution,
+which in turn prevented those CPUs from ever executing a context switch,
+which, in the RCU implementation of that time, prevented grace periods
+from ever ending.
+The result was an out-of-memory condition and a system hang.
+
+<p>
+The solution was the creation of RCU-bh, which does
+<tt>local_bh_disable()</tt>
+across its read-side critical sections, and which uses the transition
+from one type of softirq processing to another as a quiescent state
+in addition to context switch, idle, user mode, and offline.
+This means that RCU-bh grace periods can complete even when some of
+the CPUs execute in softirq indefinitely, thus allowing algorithms
+based on RCU-bh to withstand network-based denial-of-service attacks.
+
+<p>
+Because
+<tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
+disable and re-enable softirq handlers, any attempt to start a softirq
+handlers during the
+RCU-bh read-side critical section will be deferred.
+In this case, <tt>rcu_read_unlock_bh()</tt>
+will invoke softirq processing, which can take considerable time.
+One can of course argue that this softirq overhead should be associated
+with the code following the RCU-bh read-side critical section rather
+than <tt>rcu_read_unlock_bh()</tt>, but the fact
+is that most profiling tools cannot be expected to make this sort
+of fine distinction.
+For example, suppose that a three-millisecond-long RCU-bh read-side
+critical section executes during a time of heavy networking load.
+There will very likely be an attempt to invoke at least one softirq
+handler during that three milliseconds, but any such invocation will
+be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
+This can of course make it appear at first glance as if
+<tt>rcu_read_unlock_bh()</tt> was executing very slowly.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
+includes
+<tt>rcu_read_lock_bh()</tt>,
+<tt>rcu_read_unlock_bh()</tt>,
+<tt>rcu_dereference_bh()</tt>,
+<tt>rcu_dereference_bh_check()</tt>,
+<tt>synchronize_rcu_bh()</tt>,
+<tt>synchronize_rcu_bh_expedited()</tt>,
+<tt>call_rcu_bh()</tt>,
+<tt>rcu_barrier_bh()</tt>, and
+<tt>rcu_read_lock_bh_held()</tt>.
+
+<h3><a name="Sched Flavor">Sched Flavor</a></h3>
+
+<p>
+Before preemptible RCU, waiting for an RCU grace period had the
+side effect of also waiting for all pre-existing interrupt
+and NMI handlers.
+However, there are legitimate preemptible-RCU implementations that
+do not have this property, given that any point in the code outside
+of an RCU read-side critical section can be a quiescent state.
+Therefore, <i>RCU-sched</i> was created, which follows &ldquo;classic&rdquo;
+RCU in that an RCU-sched grace period waits for for pre-existing
+interrupt and NMI handlers.
+In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
+APIs have identical implementations, while kernels built with
+<tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
+
+<p>
+Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
+<tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
+disable and re-enable preemption, respectively.
+This means that if there was a preemption attempt during the
+RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
+will enter the scheduler, with all the latency and overhead entailed.
+Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
+as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
+However, the highest-priority task won't be preempted, so that task
+will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
+includes
+<tt>rcu_read_lock_sched()</tt>,
+<tt>rcu_read_unlock_sched()</tt>,
+<tt>rcu_read_lock_sched_notrace()</tt>,
+<tt>rcu_read_unlock_sched_notrace()</tt>,
+<tt>rcu_dereference_sched()</tt>,
+<tt>rcu_dereference_sched_check()</tt>,
+<tt>synchronize_sched()</tt>,
+<tt>synchronize_rcu_sched_expedited()</tt>,
+<tt>call_rcu_sched()</tt>,
+<tt>rcu_barrier_sched()</tt>, and
+<tt>rcu_read_lock_sched_held()</tt>.
+However, anything that disables preemption also marks an RCU-sched
+read-side critical section, including
+<tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
+<tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
+and so on.
+
+<h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
+
+<p>
+For well over a decade, someone saying &ldquo;I need to block within
+an RCU read-side critical section&rdquo; was a reliable indication
+that this someone did not understand RCU.
+After all, if you are always blocking in an RCU read-side critical
+section, you can probably afford to use a higher-overhead synchronization
+mechanism.
+However, that changed with the advent of the Linux kernel's notifiers,
+whose RCU read-side critical
+sections almost never sleep, but sometimes need to.
+This resulted in the introduction of
+<a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
+or <i>SRCU</i>.
+
+<p>
+SRCU allows different domains to be defined, with each such domain
+defined by an instance of an <tt>srcu_struct</tt> structure.
+A pointer to this structure must be passed in to each SRCU function,
+for example, <tt>synchronize_srcu(&amp;ss)</tt>, where
+<tt>ss</tt> is the <tt>srcu_struct</tt> structure.
+The key benefit of these domains is that a slow SRCU reader in one
+domain does not delay an SRCU grace period in some other domain.
+That said, one consequence of these domains is that read-side code
+must pass a &ldquo;cookie&rdquo; from <tt>srcu_read_lock()</tt>
+to <tt>srcu_read_unlock()</tt>, for example, as follows:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+As noted above, it is legal to block within SRCU read-side critical sections,
+however, with great power comes great responsibility.
+If you block forever in one of a given domain's SRCU read-side critical
+sections, then that domain's grace periods will also be blocked forever.
+Of course, one good way to block forever is to deadlock, which can
+happen if any operation in a given domain's SRCU read-side critical
+section can block waiting, either directly or indirectly, for that domain's
+grace period to elapse.
+For example, this results in a self-deadlock:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 synchronize_srcu(&amp;ss);
+ 6 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+However, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
+deadlock would still be possible.
+Furthermore, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
+and if an <tt>ss1</tt>-domain SRCU read-side critical section
+acquired another mutex that was held across as <tt>ss</tt>-domain
+<tt>synchronize_srcu()</tt>,
+deadlock would again be possible.
+Such a deadlock cycle could extend across an arbitrarily large number
+of different SRCU domains.
+Again, with great power comes great responsibility.
+
+<p>
+Unlike the other RCU flavors, SRCU read-side critical sections can
+run on idle and even offline CPUs.
+This ability requires that <tt>srcu_read_lock()</tt> and
+<tt>srcu_read_unlock()</tt> contain memory barriers, which means
+that SRCU readers will run a bit slower than would RCU readers.
+It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
+API, which, in combination with <tt>srcu_read_unlock()</tt>,
+guarantees a full memory barrier.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
+includes
+<tt>srcu_read_lock()</tt>,
+<tt>srcu_read_unlock()</tt>,
+<tt>srcu_dereference()</tt>,
+<tt>srcu_dereference_check()</tt>,
+<tt>synchronize_srcu()</tt>,
+<tt>synchronize_srcu_expedited()</tt>,
+<tt>call_srcu()</tt>,
+<tt>srcu_barrier()</tt>, and
+<tt>srcu_read_lock_held()</tt>.
+It also includes
+<tt>DEFINE_SRCU()</tt>,
+<tt>DEFINE_STATIC_SRCU()</tt>, and
+<tt>init_srcu_struct()</tt>
+APIs for defining and initializing <tt>srcu_struct</tt> structures.
+
+<h3><a name="Tasks RCU">Tasks RCU</a></h3>
+
+<p>
+Some forms of tracing use &ldquo;tramopolines&rdquo; to handle the
+binary rewriting required to install different types of probes.
+It would be good to be able to free old trampolines, which sounds
+like a job for some form of RCU.
+However, because it is necessary to be able to install a trace
+anywhere in the code, it is not possible to use read-side markers
+such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+In addition, it does not work to have these markers in the trampoline
+itself, because there would need to be instructions following
+<tt>rcu_read_unlock()</tt>.
+Although <tt>synchronize_rcu()</tt> would guarantee that execution
+reached the <tt>rcu_read_unlock()</tt>, it would not be able to
+guarantee that execution had completely left the trampoline.
+
+<p>
+The solution, in the form of
+<a href="http://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
+is to have implicit
+read-side critical sections that are delimited by voluntary context
+switches, that is, calls to <tt>schedule()</tt>,
+<tt>cond_resched_rcu_qs()</tt>, and
+<tt>synchronize_rcu_tasks()</tt>.
+In addition, transitions to and from userspace execution also delimit
+tasks-RCU read-side critical sections.
+
+<p>
+The tasks-RCU API is quite compact, consisting only of
+<tt>call_rcu_tasks()</tt>,
+<tt>synchronize_rcu_tasks()</tt>, and
+<tt>rcu_barrier_tasks()</tt>.
+
+<h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
+
+<p>
+One of the tricks that RCU uses to attain update-side scalability is
+to increase grace-period latency with increasing numbers of CPUs.
+If this becomes a serious problem, it will be necessary to rework the
+grace-period state machine so as to avoid the need for the additional
+latency.
+
+<p>
+Expedited grace periods scan the CPUs, so their latency and overhead
+increases with increasing numbers of CPUs.
+If this becomes a serious problem on large systems, it will be necessary
+to do some redesign to avoid this scalability problem.
+
+<p>
+RCU disables CPU hotplug in a few places, perhaps most notably in the
+expedited grace-period and <tt>rcu_barrier()</tt> operations.
+If there is a strong reason to use expedited grace periods in CPU-hotplug
+notifiers, it will be necessary to avoid disabling CPU hotplug.
+This would introduce some complexity, so there had better be a <i>very</i>
+good reason.
+
+<p>
+The tradeoff between grace-period latency on the one hand and interruptions
+of other CPUs on the other hand may need to be re-examined.
+The desire is of course for zero grace-period latency as well as zero
+interprocessor interrupts undertaken during an expedited grace period
+operation.
+While this ideal is unlikely to be achievable, it is quite possible that
+further improvements can be made.
+
+<p>
+The multiprocessor implementations of RCU use a combining tree that
+groups CPUs so as to reduce lock contention and increase cache locality.
+However, this combining tree does not spread its memory across NUMA
+nodes nor does it align the CPU groups with hardware features such
+as sockets or cores.
+Such spreading and alignment is currently believed to be unnecessary
+because the hotpath read-side primitives do not access the combining
+tree, nor does <tt>call_rcu()</tt> in the common case.
+If you believe that your architecture needs such spreading and alignment,
+then your architecture should also benefit from the
+<tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
+to the number of CPUs in a socket, NUMA node, or whatever.
+If the number of CPUs is too large, use a fraction of the number of
+CPUs.
+If the number of CPUs is a large prime number, well, that certainly
+is an &ldquo;interesting&rdquo; architectural choice!
+More flexible arrangements might be considered, but only if
+<tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
+if the inadequacy has been demonstrated by a carefully run and
+realistic system-level workload.
+
+<p>
+Please note that arrangements that require RCU to remap CPU numbers will
+require extremely good demonstration of need and full exploration of
+alternatives.
+
+<p>
+There is an embarrassingly large number of flavors of RCU, and this
+number has been increasing over time.
+Perhaps it will be possible to combine some at some future date.
+
+<p>
+RCU's various kthreads are reasonably recent additions.
+It is quite likely that adjustments will be required to more gracefully
+handle extreme loads.
+It might also be necessary to be able to relate CPU utilization by
+RCU's kthreads and softirq handlers to the code that instigated this
+CPU utilization.
+For example, RCU callback overhead might be charged back to the
+originating <tt>call_rcu()</tt> instance, though probably not
+in production kernels.
+
+<h2><a name="Summary">Summary</a></h2>
+
+<p>
+This document has presented more than two decade's worth of RCU
+requirements.
+Given that the requirements keep changing, this will not be the last
+word on this subject, but at least it serves to get an important
+subset of the requirements set forth.
+
+<h2><a name="Acknowledgments">Acknowledgments</a></h2>
+
+I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
+Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
+Andy Lutomirski for their help in rendering
+this article human readable, and to Michelle Rankin for her support
+of this effort.
+Other contributions are acknowledged in the Linux kernel's git archive.
+The cartoon is copyright (c) 2013 by Melissa Broussard,
+and is provided
+under the terms of the Creative Commons Attribution-Share Alike 3.0
+United States license.
+
+<p>@@QQAL@@
+
+</body></html>
diff --git a/Documentation/RCU/Design/htmlqqz.sh b/Documentation/RCU/Design/htmlqqz.sh
new file mode 100755
index 000000000000..d354f069559b
--- /dev/null
+++ b/Documentation/RCU/Design/htmlqqz.sh
@@ -0,0 +1,108 @@
+#!/bin/sh
+#
+# Usage: sh htmlqqz.sh file
+#
+# Extracts and converts quick quizzes in a proto-HTML document file.htmlx.
+# Commands, all of which must be on a line by themselves:
+#
+# "<p>@@QQ@@": Start of a quick quiz.
+# "<p>@@QQA@@": Start of a quick-quiz answer.
+# "<p>@@QQE@@": End of a quick-quiz answer, and thus of the quick quiz.
+# "<p>@@QQAL@@": Place to put quick-quiz answer list.
+#
+# Places the result in file.html.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, you can access it online at
+# http://www.gnu.org/licenses/gpl-2.0.html.
+#
+# Copyright (c) 2013 Paul E. McKenney, IBM Corporation.
+
+fn=$1
+if test ! -r $fn.htmlx
+then
+ echo "Error: $fn.htmlx unreadable."
+ exit 1
+fi
+
+echo "<!-- DO NOT HAND EDIT. -->" > $fn.html
+echo "<!-- Instead, edit $fn.htmlx and run 'sh htmlqqz.sh $fn' -->" >> $fn.html
+awk < $fn.htmlx >> $fn.html '
+
+state == "" && $1 != "<p>@@QQ@@" && $1 != "<p>@@QQAL@@" {
+ print $0;
+ if ($0 ~ /^<p>@@QQ/)
+ print "Bad Quick Quiz command: " NR " (expected <p>@@QQ@@ or <p>@@QQAL@@)." > "/dev/stderr"
+ next;
+}
+
+state == "" && $1 == "<p>@@QQ@@" {
+ qqn++;
+ qqlineno = NR;
+ haveqq = 1;
+ state = "qq";
+ print "<p><a name=\"Quick Quiz " qqn "\"><b>Quick Quiz " qqn "</b>:</a>"
+ next;
+}
+
+state == "qq" && $1 != "<p>@@QQA@@" {
+ qq[qqn] = qq[qqn] $0 "\n";
+ print $0
+ if ($0 ~ /^<p>@@QQ/)
+ print "Bad Quick Quiz command: " NR ". (expected <p>@@QQA@@)" > "/dev/stderr"
+ next;
+}
+
+state == "qq" && $1 == "<p>@@QQA@@" {
+ state = "qqa";
+ print "<br><a href=\"#qq" qqn "answer\">Answer</a>"
+ next;
+}
+
+state == "qqa" && $1 != "<p>@@QQE@@" {
+ qqa[qqn] = qqa[qqn] $0 "\n";
+ if ($0 ~ /^<p>@@QQ/)
+ print "Bad Quick Quiz command: " NR " (expected <p>@@QQE@@)." > "/dev/stderr"
+ next;
+}
+
+state == "qqa" && $1 == "<p>@@QQE@@" {
+ state = "";
+ next;
+}
+
+state == "" && $1 == "<p>@@QQAL@@" {
+ haveqq = "";
+ print "<h3><a name=\"Answers to Quick Quizzes\">"
+ print "Answers to Quick Quizzes</a></h3>"
+ print "";
+ for (i = 1; i <= qqn; i++) {
+ print "<a name=\"qq" i "answer\"></a>"
+ print "<p><b>Quick Quiz " i "</b>:"
+ print qq[i];
+ print "";
+ print "</p><p><b>Answer</b>:"
+ print qqa[i];
+ print "";
+ print "</p><p><a href=\"#Quick%20Quiz%20" i "\"><b>Back to Quick Quiz " i "</b>.</a>"
+ print "";
+ }
+ next;
+}
+
+END {
+ if (state != "")
+ print "Unterminated Quick Quiz: " qqlineno "." > "/dev/stderr"
+ else if (haveqq)
+ print "Missing \"<p>@@QQAL@@\", no Quick Quiz." > "/dev/stderr"
+}'
--
2.5.2


\
 
 \ /
  Last update: 2015-12-05 01:01    [W:0.587 / U:1.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site