Latest working kernel version:

Dunno.

Earliest failing kernel version:

2.6.25.15

Distribution:

Archlinux

Hardware Environment:

Asus M2400N, Pentium M 1.6 GHz, 768 MB RAM, Intel IPW2915

Software Environment:

X.org, KDE... Nothing special.

Problem Description:

ipw2200/0: page allocation failure. order:3, mode:0x4020
Pid: 7450, comm: ipw2200/0 Not tainted 2.6.25.15-AP #1
 [<c015f289>] __alloc_pages+0x2b9/0x380
 [<c017b61c>] __slab_alloc+0x2fc/0x610
 [<c017b72a>] __slab_alloc+0x40a/0x610
 [<f038d8fb>] ipw_rx_queue_replenish+0x5b/0x100 [ipw2200]
 [<c017c708>] __kmalloc_track_caller+0xb8/0x100
 [<f038d8fb>] ipw_rx_queue_replenish+0x5b/0x100 [ipw2200]
 [<c0313e85>] __alloc_skb+0x55/0x120
 [<f038d8fb>] ipw_rx_queue_replenish+0x5b/0x100 [ipw2200]
 [<f038f4d0>] ipw_bg_rx_queue_replenish+0x0/0x40 [ipw2200]
 [<f038f4f6>] ipw_bg_rx_queue_replenish+0x26/0x40 [ipw2200]
 [<c012ebbc>] run_workqueue+0x6c/0x150
 [<c012f08f>] worker_thread+0x7f/0xe0
 [<c01324e0>] autoremove_wake_function+0x0/0x50
 [<c012f010>] worker_thread+0x0/0xe0
 [<c0132097>] kthread+0x37/0x70
 [<c0132060>] kthread+0x0/0x70
 [<c0104d13>] kernel_thread_helper+0x7/0x14
 =======================
Mem-info:
DMA per-cpu:
CPU    0: hi:    0, btch:   1 usd:   0
Normal per-cpu:
CPU    0: hi:  186, btch:  31 usd:   3
Active:132552 inactive:16871 dirty:9 writeback:1 unstable:0
 free:1692 slab:7344 mapped:13890 pagetables:956 bounce:0
DMA free:2992kB min:72kB low:88kB high:108kB active:1988kB inactive:4176kB present:16256kB pages_scanned:0 all_unreclaimable? no
lowmem_reserve[]: 0 737 737
Normal free:3776kB min:3436kB low:4292kB high:5152kB active:528220kB inactive:63308kB present:755144kB pages_scanned:0 all_unreclaimable? no
lowmem_reserve[]: 0 0 0
DMA: 36*4kB 12*8kB 12*16kB 10*32kB 13*64kB 11*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 2992kB
Normal: 778*4kB 31*8kB 4*16kB 1*32kB 1*64kB 0*128kB 1*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 3776kB
54636 total pagecache pages
Swap cache: add 171402, delete 147872, find 34791/42583
Free swap  = 1630068kB
Total swap = 2000084kB
Free swap:       1630068kB
194368 pages of RAM
0 pages of HIGHMEM
2728 reserved pages
111884 pages shared
23530 pages swap cached
9 pages dirty
1 pages writeback
13890 pages mapped
7344 pages slab
956 pages pagetables

Steps to reproduce:

Attach the card to an AP and run *huge* data transfers in *both* directions. Hundreds of these long messages appear in dmesg. Some of them end like this:

ipw2200: Firmware error detected.  Restarting.
ipw2200: Unable to load firmware: -12
ipw2200: Unable to load firmware: -12
ipw2200: Failed to up device

A total network failure follows. Re-modprobing the module can fix it. However, the firmware is a binary blackbox, AFAIK... So the memory allocation failure should be more interesting than the strange firmware failure.

Some error messages (very rarely) reported allocation failures in other processes too, but I'd guess there's less than 1% of them. One of those processes was hald-addon-input.

(BTW, I had a *similar* problem on a dual CPU server (IBM xSeries 330) with an Atheros WiFi card, with lots of strange page failures in vital kernel processes. (Dmesg messages suggested a reboot, which I did.) But the server uses a tainted kernel (reiquiring ath_hal and ath_pci to serve as an AP), so I didn't post the messages here.)