lkml.org 
[lkml]   [1999]   [Nov]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
Subject[patch] zoned-2.3.28-J5

this is zoned-2.3.28-J5, and it should fix all (most) known problems.
Please let me know if the '16MB RAM' case is still problematic. I've
tested it (without swap), and memory balance appears to be just fine.

-- mingo

Changes in zoned-2.3.28-J5:

- further page_alloc.c cleanups/speedups.

- Alan and Rogier convinced me to optimize the 'top level zone is empty'
case a bit more.

- show_free_areas() works again.

- some more include file fixes


Changes in zoned-2.3.28-H2:

- fixed NFS to work out of high memory - tested with moderate load. All
NFS caches (directory, symlink, data, etc.) are in high memory.

- modules fix ...

- page->virtual is filled out for non-highmem pages too, this is a
nice speedup in certain cases.

- fixed a bug in highmem support which might cause user-datapage
corruption in certain cases.


Changes in zoned-2.3.28-G5:

- this one should actually compile if modules support is turned on ...


Changes in zoned-2.3.28-G4:

- implemented 'zone chains' zonelist_t and gfp_mask indexed zonelists[]
speedups (Linus' idea) to handle fallback zones. This should enable
advanced NUMA-style allocations as well. [fallback to different CPUs is
possible via changing build_zonelists().]

- <=16MB RAM boxes should boot just fine now.

- added page->zone for easier deallocation and generic cleanliness. This
also helps NUMA.

- cleaned up the page-allocator namespace, there are only two 'core'
page-allocation functions left: __alloc_pages() and __free_pages_ok().

- modules should compile again.

- we are now inlining the 'put_page_testzero()' part of __free_page_ok.
This is subtle as page->count for reserved pages is now 'rotating' -
this is fine though and lets us to put the rare PageReserved() branch
into __free_page_ok().

- cleaned up pgtable.h, split into lowlevel and highlevel parts, this
fixes dependencies in mm.h & misc.c.

- serial.c didnt clear freshly allocated bootmem - as a result now all
bootmem allocations are explicitly cleared, it's not performance
critical anyway.

- fixed code,data,initmem reporting.

- fixed boot task's swapper_pg_dir clearing

-- mingo
[unhandled content-type:application/octet-stream]
\
 
 \ /
  Last update: 2005-03-22 13:55    [W:0.075 / U:0.700 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site