Re: [2.6.25-rc2, 2.6.24-rc8] page allocation failure...

["Kok, Auke" <auke-jan.h.kok@intel.com>]

Andrew Morton wrote:
> On Sun, 17 Feb 2008 13:20:59 +0000 "Daniel J Blueman" <daniel.blueman@gmail.com> wrote:
> 
>> I'm still hitting this with e1000e on 2.6.25-rc2, 10 times again.

are you sure? I don't think that's the case and you're seeing e1000 dumps here...

>> It's clearly non-fatal, but then do we expect it to occur?
>>
>> Daniel
>>
>> --- [dmesg]
>>
>> [ 1250.822786] swapper: page allocation failure. order:3, mode:0x4020
>> [ 1250.822786] Pid: 0, comm: swapper Not tainted 2.6.25-rc2-119 #2
>> [ 1250.822786]
>> [ 1250.822786] Call Trace:
>> [ 1250.822786]  <IRQ>  [<ffffffff8025fe9e>] __alloc_pages+0x34e/0x3a0
>> [ 1250.822786]  [<ffffffff8048c6df>] ? __netdev_alloc_skb+0x1f/0x40
>> [ 1250.822786]  [<ffffffff8027acc2>] __slab_alloc+0x102/0x3d0
>> [ 1250.822786]  [<ffffffff8048c6df>] ? __netdev_alloc_skb+0x1f/0x40
>> [ 1250.822786]  [<ffffffff8027b8cb>] __kmalloc_track_caller+0x7b/0xc0
>> [ 1250.822786]  [<ffffffff8048b74f>] __alloc_skb+0x6f/0x160
>> [ 1250.822786]  [<ffffffff8048c6df>] __netdev_alloc_skb+0x1f/0x40
>> [ 1250.822786]  [<ffffffff8042652d>] e1000_alloc_rx_buffers+0x1ed/0x260
>> [ 1250.822786]  [<ffffffff80426b5a>] e1000_clean_rx_irq+0x22a/0x330
>> [ 1250.822786]  [<ffffffff80422981>] e1000_clean+0x1e1/0x540
>> [ 1250.822786]  [<ffffffff8024b7a5>] ? tick_program_event+0x45/0x70
>> [ 1250.822786]  [<ffffffff804930ba>] net_rx_action+0x9a/0x150
>> [ 1250.822786]  [<ffffffff802336b4>] __do_softirq+0x74/0xf0
>> [ 1250.822786]  [<ffffffff8020c5fc>] call_softirq+0x1c/0x30
>> [ 1250.822786]  [<ffffffff8020eaad>] do_softirq+0x3d/0x80
>> [ 1250.822786]  [<ffffffff80233635>] irq_exit+0x85/0x90
>> [ 1250.822786]  [<ffffffff8020eba5>] do_IRQ+0x85/0x100
>> [ 1250.822786]  [<ffffffff8020a5b0>] ? mwait_idle+0x0/0x50
>> [ 1250.822786]  [<ffffffff8020b981>] ret_from_intr+0x0/0xa
>> [ 1250.822786]  <EOI>  [<ffffffff8020a5f5>] ? mwait_idle+0x45/0x50
>> [ 1250.822786]  [<ffffffff80209a92>] ? enter_idle+0x22/0x30
>> [ 1250.822786]  [<ffffffff8020a534>] ? cpu_idle+0x74/0xa0
>> [ 1250.822786]  [<ffffffff80527825>] ? rest_init+0x55/0x60
> 
> They're regularly reported with e1000 too - I don't think aything really
> changed.
> 
> e1000 has this crazy problem where because of a cascade of follies (mainly
> borked hardware) it has to do a 32kb allocation for a 9kb(?) packet.  It
> would be sad if that was carried over into e1000e?

can't be, I personally removed that code.

for MTU > 1500 e1000e uses a plain normal sized SKB. for anything bigger e1000e
uses pages.

so I don't see how this bug could still be showing up for e1000e at all. The large
skb receive code is all gone (literally, removed).

*please* rmmod e1000; modprobe e1000e and show the dumps again so we know for sure
that we're not looking at e1000 dumps.

short fix: increase ring size for e1000 with `modprobe e1000 RxDescriptors=4096`
(or use ethtool) and `echo -n 8192 > /proc/sys/vm/min_free_kbytes` or something
like that.

what nic hardware is this on? lspci?

Auke