Re: kernel: __alloc_pages: 0-order allocation failed

Re: kernel: __alloc_pages: 0-order allocation failed

[Valerio Vanni]

On Wed, 22 Jun 2005 19:40:08 +0200, Chris Wedgwood <cw@f00f.org>
wrote:
>> as before) or the kernel is in some way locked up (in particular: is
>> it necessary/better to reboot? Is there some risk of filesystem
>> corruption?).
>
>It's memory allocation failures.  This might not work until memory is
>free but it shouldn't kill the kernel of be a huge problem if it's
>just the result of one ore more processes being memory hungry

I'm blaming myself for not having given a closer look before shutting
down the machine.
But, as I said, the shut down happened regularly. Later, to be sure, i
ran an fsck on all partitions and they were ok.

>It could also occur if there is a memory leak, in which case there is
>a bug that needs to be fixed and a reboot would be needed (I would
>only suspect that if it did it often and processes were not using much
>memory though).

It was the first time: this machine had been running on the same
kernel for six month with very long uptimes interrupted only by ups
control daemon (during blackouts).

Ah, no: the kernel had been changed about one month ago, but it was
the same version (2.4.26) and the same .config (except ext3 support
which I moved from modules to kernel itself).


-- 
Ci sono 10 tipi di persone al mondo: quelle che capiscono il sistema binario
e quelle che non lo capiscono.

Re: kernel: __alloc_pages: 0-order allocation failed

[Chris Wedgwood]

On Wed, Jun 22, 2005 at 04:32:32PM +0200, Valerio Vanni wrote:

> I mean: is it simply a situation of excessive memory requests that is
> fixed by killing one or more processes (and the kernel is still alive
> as before) or the kernel is in some way locked up (in particular: is
> it necessary/better to reboot? Is there some risk of filesystem
> corruption?).

It's memory allocation failures.  This might not work until memory is
free but it shouldn't kill the kernel of be a huge problem if it's
just the result of one ore more processes being memory hungry
(ie. when those processes go away things should be fine).

It could also occur if there is a memory leak, in which case there is
a bug that needs to be fixed and a reboot would be needed (I would
only suspect that if it did it often and processes were not using much
memory though).

kernel: __alloc_pages: 0-order allocation failed

[Valerio Vanni]

I found this error on a 2.4.26 kernel:

>Jun 19 23:00:08 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1d2/0)
>Jun 19 23:00:11 server kernel: __alloc_pages: 2-order allocation failed (gfp=0x1f0/0)
>Jun 19 23:00:11 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1d2/0)
>Jun 19 23:00:11 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1f0/0)
>Jun 19 23:00:12 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1d2/0)
>Jun 19 23:00:12 server kernel: __alloc_pages: 2-order allocation failed (gfp=0x1f0/0)
>Jun 19 23:00:13 server syslogd: /var/log/messages: Cannot allocate memory
>Jun 19 23:00:13 server kernel: __alloc_pages: 0-order allocation failed  (gfp=0x1f0/0)
>Jun 19 23:00:13 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1d2/0)
>Jun 19 23:00:13 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1d2/0)
>Jun 19 23:00:13 server kernel: VM: killing process fetchnews
>Jun 19 23:00:15 server kernel: __alloc_pages: 0-order allocation failed (gfp=0x1f0/0)

between these lines there were other saying
>fetchnews[958]: reading XOVER info from /var/spool/news/...

Then I could shut the machine correctly down. No other process than
fetchnews had been killed

Aftere many searches on the net I still don't understand a thing about
this error: how much is it critical?

I mean: is it simply a situation of excessive memory requests that is
fixed by killing one or more processes (and the kernel is still alive
as before) or the kernel is in some way locked up (in particular: is
it necessary/better to reboot? Is there some risk of filesystem
corruption?).

-- 
Ci sono 10 tipi di persone al mondo: quelle che capiscono il sistema binario
e quelle che non lo capiscono.