On Wed, Aug 03, 2005 at 09:21:49AM +0200, Brice GIBOUDEAU wrote:
> Hi,
>
> I have a Linux (Debian Sarge) server with more than 4Gb of memory, I'm
> running XFS on other servers for a long time without any problem. (Less or
> equal to 4Gb of memory)
> Actually i'm using 2.4.31 kernel, when i add PEA (64Gb of memory) in my
> kernel i have a memory leak when i'm working on XFS File system.
>
> Exemple :
> ---------
> I have 2 volumes (/xfs1 & /xfs2) of 1TB with million of small files on
> /xfs1, i trie to make a backup of file from /xfs1 to /xfs2 (tar -cvf
> /xfs2/backup.tar /xfs1), after several minutes the memory is full, and i
> get this messages :
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0xf0/0)
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0xf0/0)
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0x1f0/0)
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0x1f0/0)
> -> kernel: __alloc_pages: 1-order allocation failed (gfp=0x1f0/0)
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0xf0/0)
> -> kernel: __alloc_pages: 0-order allocation failed (gfp=0x1f0/0)
Sounds more like a lowmem/highmem balance/reclaim VM issue than a
memory leak. You could be running out of low memory which, IIRC,
holds most of the kernel data and if you're not using highmem I/O,
all the bounce buffers to do I/O out of pages in highmem.
Where are these failures happening (did you get a stack trace)?
Are they in XFS, or somewhere else?
What does /proc/meminfo and /proc/slabinfo tell you? That is, what
memory is exhausted and what is using it all?
Cheers,
Dave.
--
Dave Chinner
R&D Software Enginner
SGI Australian Software Group
|