[lkml]   [2009]   [Nov]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 0/5] Reduce GFP_ATOMIC allocation failures, candidate fix V3
On Thu, Nov 12, 2009 at 07:30:30PM +0000, Mel Gorman wrote:

> [Bug #14265] ifconfig: page allocation failure. order:5, mode:0x8020 w/ e100

> Patches 1-3 should be tested first. The testing I've done shows that the
> page allocator and behaviour of congestion_wait() is more in line with
> 2.6.30 than the vanilla kernels.
> It'd be nice to have 2 more tests, applying each patch on top noting any
> behaviour change. i.e. ideally there would be results for
> o patches 1+2+3
> o patches 1+2+3+4
> o patches 1+2+3+4+5
> Of course, any tests results are welcome. The rest of the mail is the
> results of my own tests.

I've tried testing 3+4+5 against 2.6.32-rc7 (1+2 seem to be in
mainline) and got failure. I've noticed something strange (I think).
I was unable to trigger failures when system was under heavy memory
pressure (i.e. my testing - gitk, firefoxes, etc.). When I killed
almost all memory hogs, put system into sleep and resumed -- it
failed. free(1) showed:

total used free shared buffers cached
Mem: 255240 194052 61188 0 4040 49364
-/+ buffers/cache: 140648 114592
Swap: 514040 72712 441328

Is that ok? Wild guess -- maybe kswapd doesn't take fragmentation (or
other factors) into account as hard as it used to in 2.6.30?


 \ /
  Last update: 2009-11-15 13:09    [W:0.425 / U:0.348 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site