Reasons for page allocation failure?

[smaxer]

Hi,

I was wondering what could be the reasons for the huge amount of page allocation failures I get on both my raspberry pis. (Both Rev. B one with 512MB Ram)
I've also tried to use another and new SD card but it didn't help.

Here is an example of one logged page allocation failure.... but as said...they happen often.
$this->bbcode_second_pass_quote('', '
')[ 123.156248] mount: page allocation failure: order:5, mode:0xd0
[ 123.156338] [<c0012af4>] (unwind_backtrace+0x0/0xf0) from [<c00ba190>] (warn_alloc_failed+0xd0/0x118)
[ 123.156377] [<c00ba190>] (warn_alloc_failed+0xd0/0x118) from [<c00bc854>] (__alloc_pages_nodemask+0x534/0x71c)
[ 123.156420] [<c00bc854>] (__alloc_pages_nodemask+0x534/0x71c) from [<c00e8eac>] (cache_alloc_refill+0x338/0x690)
[ 123.156450] [<c00e8eac>] (cache_alloc_refill+0x338/0x690) from [<c00e9370>] (__kmalloc+0x16c/0x1a4)
[ 123.156499] [<c00e9370>] (__kmalloc+0x16c/0x1a4) from [<c01b46a8>] (ext4_kvzalloc+0x14/0x3c)
[ 123.156531] [<c01b46a8>] (ext4_kvzalloc+0x14/0x3c) from [<c01b9e3c>] (ext4_fill_super+0x2390/0x2748)
[ 123.156573] [<c01b9e3c>] (ext4_fill_super+0x2390/0x2748) from [<c00f6f88>] (mount_bdev+0x19c/0x1d8)
[ 123.156622] [<c00f6f88>] (mount_bdev+0x19c/0x1d8) from [<c01a92a8>] (ext4_mount+0x14/0x20)
[ 123.156655] [<c01a92a8>] (ext4_mount+0x14/0x20) from [<c00f79e4>] (mount_fs+0x44/0x184)
[ 123.156694] [<c00f79e4>] (mount_fs+0x44/0x184) from [<c0110238>] (vfs_kern_mount+0x4c/0xc0)
[ 123.156724] [<c0110238>] (vfs_kern_mount+0x4c/0xc0) from [<c0110304>] (do_kern_mount+0x34/0xd0)
[ 123.156755] [<c0110304>] (do_kern_mount+0x34/0xd0) from [<c0111f58>] (do_mount+0x2d8/0x740)
[ 123.156783] [<c0111f58>] (do_mount+0x2d8/0x740) from [<c0112444>] (sys_mount+0x84/0xb8)
[ 123.156817] [<c0112444>] (sys_mount+0x84/0xb8) from [<c000db20>] (ret_fast_syscall+0x0/0x30)
[ 123.156829] Mem-info:
[ 123.156838] Normal per-cpu:
[ 123.156850] CPU 0: hi: 186, btch: 31 usd: 0
[ 123.156879] active_anon:5840 inactive_anon:28 isolated_anon:0
active_file:3256 inactive_file:35672 isolated_file:32
unevictable:0 dirty:60 writeback:0 unstable:0
free:67124 slab_reclaimable:2206 slab_unreclaimable:1204
mapped:8468 shmem:76 pagetables:393 bounce:0
[ 123.156937] Normal free:268496kB min:8192kB low:10240kB high:12288kB active_anon:23360kB inactive_anon:112kB active_file:13024kB inactive_file:142688kB unevictable:0kB isolated(anon):0kB isolated(file):128kB present:483488kB mlocked:0kB dirty:240kB writeback:0kB mapped:33872kB shmem:304kB slab_reclaimable:8824kB slab_unreclaimable:4816kB kernel_stack:976kB pagetables:1572kB unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:0 all_unreclaimable? no
[ 123.156953] lowmem_reserve[]: 0 0
[ 123.156973] Normal: 9644*4kB 9538*8kB 9415*16kB 73*32kB 8*64kB 1*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 268496kB
[ 123.157026] 39023 total pagecache pages
[ 123.157035] 0 pages in swap cache
[ 123.157045] Swap cache stats: add 0, delete 0, find 0/0
[ 123.157052] Free swap = 0kB
[ 123.157059] Total swap = 0kB
[ 123.223892] 121856 pages of RAM
[ 123.223916] 67349 free pages
[ 123.223925] 3663 reserved pages
[ 123.223932] 3409 slab pages
[ 123.223939] 285596 pages shared
[ 123.223947] 0 pages swap cached
[ 123.223963] SLAB: Unable to allocate memory on node 0 (gfp=0xd0)
[ 123.223979] cache: size-131072, object size: 131072, order: 5
[ 123.223994] node 0: slabs: 1/1, objs: 1/1, free: 0

[sdjf]

One possible explanation is you might have run out of RAM. I was getting the same Page Allocation Failure messages before things crashed on my Pi, see the following answer/discussion and possible solution:

http://www.raspberrypi.org/phpBB3/viewt ... 05#p369705

In my case, I was doing too many things at once, and finally gave up trying to run Midori browser using VNC over ethernet while dialup was running. Are there any memory or byte-intense applications running when this happens for you?

[moonman]

The easy solution for me was to change the memory split to give the VPU 128 MB of RAM, while by default it has 2/3 ram dedicated. On 256 MB model if you are not using XBMC or desktop you can use the minimum (32?).

[smaxer]

Hi Guys,

thanks for reply, but I don't think it has something to do with low memory because I get those page allocation failures very quick right after startup. And it also says:

$this->bbcode_second_pass_quote('', '
')[ 123.223892] 121856 pages of RAM
[ 123.223916] 67349 free pages

Hi,

Got the same problem and I found why and how to get rid of these allocation failures.
I also had many page allocation failure while still having plenty of free RAM.
The cause was a heavy memory fragmentation that prevent allocation of a contiguous memory zone.

Your log show you have free RAM (268496kB)
$this->bbcode_second_pass_quote('', '[') 123.156937] Normal free:268496kB min:8192kB low:10240kB high:12288kB active_anon:23360kB inactive_anon:112kB active_file:13024kB inactive_file:142688kB unevictable:0kB isolated(anon):0kB isolated(file):128kB present:483488kB mlocked:0kB dirty:240kB writeback:0kB mapped:33872kB shmem:304kB slab_reclaimable:8824kB slab_unreclaimable:4816kB kernel_stack:976kB pagetables:1572kB unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:0 all_unreclaimable? no

[WarheadsSE]

Ludo, you could always have informed the devs via an issue @ github, or a pull request with a better option set.

Note to any readers: using a kernel other than the packaged, or self-compiled from the package will likely end with "not our problem"

[zeludo]

I plan to inform devs about that, just wanted to investigate a bit more on kernel config to confirm my thoughts about allocator and memory compaction by compiling a kernel with only these modifications (I even think that only activating compaction would do the trick).
Then i'll make a pull request.

[pepedog]

Lido,
Thank you

[zeludo]

Made a pull request https://github.com/archlinuxarm/PKGBUILDs/pull/630