Hi,
I see following page allocation failures while running IO intensive
tests on 2.6.9. My tests didn't fail, so I guess its okay. But
I never saw this before.
Its on a 4-way AMD64 machine with 7GB RAM. Tests create 10 4GB files
on 10 disks (one filesystem per disk) in parallel. (dd if=/dev/zero ...)
Thanks,
Badari
swapper: page allocation failure. order:0, mode:0x20
Call Trace:<IRQ> <ffffffff8015fd9f>{__alloc_pages+815}
<ffffffff8015fa20>{__get_free_pages+16}
<ffffffff801635c6>{kmem_getpages+38}
<ffffffff801639e9>{cache_alloc_refill+665}
<ffffffff80163b96>{kmem_cache_alloc+54}
<ffffffff8031a85d>{scsi_get_command+45}
<ffffffff8031edb0>{scsi_prep_fn+272}
<ffffffff802d62e8>{elv_next_request+72}
<ffffffff8031f0a8>{scsi_request_fn+72}
<ffffffff802d96a1>{blk_run_queue+49}
<ffffffff8031f6af>{scsi_end_request+223}
<ffffffff8031f90d>{scsi_io_completion+573}
<ffffffff80319c26>{scsi_finish_command+214}
<ffffffff8031a56a>{scsi_softirq+234}
<ffffffff8013dc11>{__do_softirq+113}
<ffffffff8013dcc5>{do_softirq+53}
<ffffffff80113f1f>{do_IRQ+335}
<ffffffff80110d27>{ret_from_intr+0}
<EOI> <ffffffff8010f5d0>{default_idle+0}
<ffffffff8010f5f0>{default_idle+32}
<ffffffff8010f9fd>{cpu_idle+29}
Badari Pulavarty <[email protected]> wrote:
>
> Hi,
>
> I see following page allocation failures while running IO intensive
> tests on 2.6.9. My tests didn't fail, so I guess its okay. But
> I never saw this before.
>
> Its on a 4-way AMD64 machine with 7GB RAM. Tests create 10 4GB files
> on 10 disks (one filesystem per disk) in parallel. (dd if=/dev/zero ...)
Tried increasing /proc/sys/vm/min_free_kbytes?
$ cat /proc/sys/vm/min_free_kbytes
886
What do you recommend it be set to?
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Andrew Morton
Sent: Thursday, October 28, 2004 5:20 AM
To: Badari Pulavarty
Cc: [email protected]
Subject: Re: 2.6.9 page allocation failures
Badari Pulavarty <[email protected]> wrote:
>
> Hi,
>
> I see following page allocation failures while running IO intensive
> tests on 2.6.9. My tests didn't fail, so I guess its okay. But
> I never saw this before.
>
> Its on a 4-way AMD64 machine with 7GB RAM. Tests create 10 4GB files
> on 10 disks (one filesystem per disk) in parallel. (dd if=/dev/zero
...)
Tried increasing /proc/sys/vm/min_free_kbytes?