Messages in this thread |  | | Subject | Reproducible SMP/SCSI Hangs under 2.0.xx | Date | Mon, 16 Dec 1996 14:46:28 -0700 | From | Bill Reynolds <> |
| |
I have enclosed some routines which will reproducibly hang a 2.0.xx/SMP box running SCSI. It has been tested with NCR and Buslogic controllers. Here is the README:
This is code that tickles a bug under SMP/SCSI linux 2.0.27 (and earlier 2.0.xx revs) which causes the machine to hang. There are two programs, db_create, which creates a large file (of size `cat size` Mb) of zeroes, and load, which mmaps the file and repeatedly does both sequential and random seeks on it (consequently beating the heck out of both the memory and disk systems). We find that the hang is easiest to reproduce when 'size' is 2-3X physical memory. For an 8M system, 30M is adequate. A trick we use is to put 'append mem=8M' in the lilo.conf. For our full configuration (64M), we needed 'size' around 200M.
Build the programs by typing `make'. To reproduce the hang, first fire up the RUN_ON_VC1 script, which creates a file, foo, (default size: 30M) and then runs load on it. On another virtual console, run the RUN_ON_VC2 script, which first sleeps 20 seconds and then repeatedly creates another file, foo1, (default size: 30M). After a few minutes, all of the disk lights go off and the machine hangs, requiring a hard reset. Note that we do not see the hang as often if we omit the sleep.
We have observed this problem on two of our systems, a dual processor AMI Goliath PPro200/512k system, and a dual processor SuperMicro P6DOF PPro200/512k. We were able to reproduce the hang with kernel versions back to 2.0.14. We have observed it with both a BusLogic BT958 SCSI controller and a Symbios (NCR) 53C825 SCSI controller using the BSD driver. We did not observe a hang with the Eckhardt NCR driver. We do not observe the problem if we have compiled the kernel single processor. We can reduce the frequency of the hangs (i.e. you have to run the db_create script many more times on a loaded machine to get the hang) by: 1) Installing Alan Cox's irq forwarding patches or 2) Either reducing the tagged queueing depth with the BusLogic, or disabling tagged queueing altogether. We have not observed the hang on machines which have disabled tagged queueing and use the forwarding patches, although we have not run long enough tests to see if the problem has gone away entirely.
Any help with this problem would be greatly appreciated.
Bill Reynolds bill@santafe.edu
================= Bill Reynolds bill@santafe.edu
------------Begin----------------- begin 664 kill_kernel.tar.gz M'XL("!*_M3(``VMI;&Q?:V5R;F5L+G1A<@#M&FE3VT@V7]&O>`/#1B;"EB_( MA'AK"3BS5#!0P&RV=I+RM*66W;&L]K2D$,_QW_>]UFESSA;)UDST*L%2Z_6[ M+QU3X?O#*5<!]QM//A-`I[6[VX4G`+"[TUGZ3<$&V&EUN^WV;L?&XV:S8[>? M0/=S"52&.(R8`G@R0D/<A7??]3\I3$O^/^_O'P[ZC\^C:=L[*WXO^[_3:G9R M_V,$H/];[=WF$[`?7Y3K\)7[_W(B0L!_CG0Y1!,6022<J<]#8#"*QQ`'+E=P M,3AK7!Q<'($O@O@3M.IVO;4+)@M<@S/E"T2AM4^?0/&/80VN)L*9@,/B$`E% M$PXSYDQ$@!PD3%@PKL/EA"L.#/]'5]*8*SE6;!9:X(Z&CN(LXE9&1)^1.#Y3 M8PZ>\#F8TH-0_,+A)P<EIJ.?C,&H!KC\"U>2(R&4#7S)W(S.;,;FB2R:`EU6 M?$ZT77\!+NZ!D8PF1LA_CGD0">8G./A'SB#D?!J"#$!$8#HR2+%PYPA)B&"L M*4^X,P491P;*0<02U?E,JH4FYHIP"N$BC/@LK-7A+<F"RXG9:3^:AIS!62AX M&!EH+91123=V.*K!`WA*NCXEG-9V^]\PGRQ"X:"H"9,ZO)8*.<'S0<K&,MKV M@-"9BQ*CLG78ATBAB^&*`[J'KB&;>1S!4S:?<Q0':?6>#Y!)0$(9OO!E'57V M$NHR5N#%O@^T),:Q0O71+N9.9U"SB&C`N<O=3%*F)(:0T;+M0=TP7L7"=[6J MF<=AM(!H,2<+_C1C4_X40Z.L=686"TVEPHC^<B.>Z_7S'TZ&IR?#?QTT(724 MF$?78X9\C5NEM,!TN<=B/PF7%V27FG9*1'95<1#J<$E\7(?3`"^B"S&R/PH5 MQ4PK'$HBA\C:,CG_U@K_1-30YQQ#KF5C].!>-RRQRR//R&5-N>42-U=%!A(9 M_>=%B(:J\2N883KB9LM@Z!`,.K**#C)?C"=1"&.)JU[&-\]",BBFB,*0$(I, MSW!)N8;B(4?=3V24U@)TIRL!):,$*$*482J@%&@ICU#D3.CX-;3&=3",MX3X M$:^,0JX^<N*.<88^'?E\1B;&I"=Y*9C2?,",!9?,C%@.#T.IC/W!$7PO?<$P MD\[.E,0@:G2;K2R%DAQ?W047\9RK@7"4A+.=P]/71GFK3KHK77I0%(AN"C6X M$L@PZ8GPD:L0XQOCE#E32DBJ<\V.IK.L(MI`;]2)S^!5'![+L7#@U>5WW>>@ MJR=&0:2D[W-E)*)?+&8C(4,P3P[.:]!M'SQO=5=1,4NS^O+JXA!<)5"F.I`` MKG`-<DXJ@W9C)C_A]YTI^34"))]O?)O[--M&KLM\DWA4:^;(V1R#,0F=U!PD MBL\+:VMR#J.(SFQH>$H71V>11:0.-S!%':O/0L8)=3)]DD=%S4_3".,T6!A8 MT!!+S+BNNTQG)PI3ZB1C7I3-&A:2%]"LP5&`#=WWR63[/@N,`_GI*78X]3/F ME+I":]"5.8N<"1'&OE6#OM"9IU7(3!VQ\1BYH2(QQS7#Y7,T:F[9S+L6D<", MPVC2.Y=W`?,CB5).4KN3XF6'N47,H8JI9F%:1+25$M*$N"(/Q0^5;]W0KBEF M$>>)C,>3W)O$E@SN2T3C`5TSL'1$NOQ3=@LO*\PZ$":,R@<:FEVQ!5!'5-Q? M8`W?#Q;8Z/QY9HQ28E_)&,O[B,.8W(G-$3N*XHZ@4D?5'T<H..>+0/IN:-!` M]8^0!1'S>!U-;QC_[UGH:X2E^3_O9X_+`^?_7;RMN_7^K[.;S/^==J?=W&GB M_-]NVKO5_/\E8..;QD@$C7!B;.B!&H>^P:L%%@;CXN@__5YIOC8VC`T]`89R MQK%8LE`&%G:]I^F\JE9+<MI+_'0)=^L!*YW\:#;Q:%K%626*L*>,<.Q"'GJ& MP*%)L]-W)\O37#;CTPB2W`-\2Y+"@(_9B`0G(MR92%C7ZV:N40W6$UP#"RP2 M>@%[V`E3W'JCZ$(T?"68Z\;-ZX:+E?$O4:_*^3_`^9M,_-@\Z/Z_<_O]/W3: M=IK_[6[;WM7YW^Q4^?\EX."@-W8<,`Y>'^]_?]';GHLYA^VQ07<4+Y*$S1/` M,.B\+I/U.NY:^]8\.!V<'1WWZTXM7Z7?%$=JG..CDS<Y@H1MJ8^,G#"1+$ZN MT5VZE)^4MRRS*:\CKT)^Q^<L>&&LJ1EL>XER#DV918IOU:DJR:V_1&X_!,KY M7[+SH_*X)_];[69W-?^[W2K_OPALB,#Q8Y?#RS!RA:Q/_FZ4EA9A`\T3+:]Z M3A#YUQ&CQ1R;[[7E&=[,/0@Y#@2*L((:N;X8T9JQX6)OPMN1PZ.+-\-7QZ<' M;\!\WORNM87153,<O,^%K;/Q[(3-^)Z!467,F`A,IL:.A7\^UFA-!!'0BD9( MMM"U']_O&;_2DN)CE`$G&4(4>]D6SRTVC&+OQT($W$A74KX]3<Q^GVST-'/X MI@?M&FCR"-Y<(47/1,6X4A:L_Q!B(=L,J03IZ09_!WJ,>1>L6QGAVEZZG7\2 MD=G,3W^GWQF?A3PR43#+M@K1$J0XP/O2J:D%:[ZO99)]8WIN3\YYD%VQ3H?G MAV_/?SL='ISW]R\M>P>M6KM5[G5S$\<IO._7-[-(!U4@@5-YK25^-\E-!XTM MM*Z@)ZTTQ;DL8B.&=[1;#;KX:[X3_1'@G!;3\QH62<_,R7A8P2,XF?%Q3U_0 M7%OO:Q02!!D>^?!DY$_#'N$V"AOMW:4=-D#I),]V-]-'MR-<F8:PZ8'V$6EL MK*UI$IGFFH]E:M%J!:<MO=XP4\GJM4()J4S1L_=`O$R$Q*-GS\CR&\+#F(?+ M_L6EL79-1E<FDEF;[KMHW1++KM_@@2L\W(;.OE(BXNAPBT*DA`6]'FPWB=/: M?2[6)&B^US%ZFZ_7EER]EGCZ;BW>1>\BFJ7K[]1Z6>S?RQ%^777:$*PG-M0, M;3S^_:MIVH\(Y?Z?S&^/S^.>_M_>;357^G^KN].M^O^7@+O[OQ@'[(96_^"> M?L\`0<4AH.HPV#\;OM['<?\P[_/%$A:IK#`8JXT^/:>7>\-P,C-#/.6ZDUKS M:$B/!VI%KR\NYJU]*\5*!X#5;K\5\"M])CUOF#P(V<MZ5]H?.(Q].6*^?IA! M'*@L)ST,:V_>9TN"):TV;;%Y#?Z?^FQ!]<X1@00AV7L^O<>D5F!;%_W^FV'_ MY+!6Z@-WR*`+OW[<XO*(._J=G\?H]4!)GON$T#+T[`>SLA]*.W-CK_`0>JYG M)DZLZ9>_IFW151RJSD\OA_2MPV_ZZ.WYT67?HG"[^.?^>?_0`K)//BHA'3)0 M3JJ(RWO4()Z`IQ@6GGYO&LWCZ":%YAK)//GA^/A.)1U?AM3'DU7%HU@%))YN M?3>-NRL#[P-&7NN#%>2Y@5G:LXM,88I9B7E+"35;3B>:CIMVJW/;7+PR&+=N M-V$V%V<L'AH)&7XO'4OT112]9VHE:J5*D2!:?]-5(I=-!]Y+_8,C)6ZB<;+5 MN6\6UIGH9B_[TS$I>;DN(9QAL4BF15B:%C,A[E0IT%'=*$FT=]-45.0.O8^; M2-_%Z7!*;N%CKI;*1M#0*B4ZTP"ZMW>[>A@&-`#?"3B]68B'<^M-8VV0C[1Z M'C7%9CH&Z]*#=V\WSI\I7Q60'H.4"06E)1K9?IHRT;D_BO>]Y$.-XN:@F"%O M%0498%W/=VZ:P79JER4Y4RGA3C$=.5_@K_VFL$8FIZ:'@>',%_HN#05^AGRM M:R%&>"3L!Q+VPTN@13P@>4E)W//L0UG1\IBLC[`__M$9N#S_D4"?8\:XY_LO MA';Q_F?7UM]_V=7SGR\";;NZ:_J:X<;WO\W'Y7'O^]_VM>>_37NGRO\O`7_P M_>]G>B%[XSO86U[!IEMR:9(/.^F"_L22/N3,/J_+OMZD"^7O.AWZ>*DL6%T_ M^R`BQ"X[KBIC!1544$$%%51000455%!!!1544$$%%51000455%!!!1544$$% /%53P9X#_`L98T8(`4``` ` end
------------End-------------------
|  |