On Mon, 7 Apr 2008, Peter Zijlstra wrote:
quoted text
> On Mon, 2008-04-07 at 20:40 +0100, Hugh Dickins wrote:
> > 
> > My supposition is that once a page has been allocated from __GFP_HIGH
> > reserves to a scsi sense_slab, swap_writepages are liable to gobble up
> > the rest of the page with bio allocations which they wouldn't have had
> > access to traditionally (i.e. under SLAB).
> > 
> > So an unexpected behaviour emerges from SLUB's slab merging.
> 
> Somewhere along the line of my swap over network patches I made
> 'robustified' SLAB to ensure these sorts of things could not happen - it
> came at a cost though.
> 
> It would basically fail[*] allocations that had a higher low watermark
> than what was used to allocate the current slab.
> 
> [*] - well, it would attempt to allocate a new slab to raise the current
> watermark, but failing that it would fail the allocation.

Thanks, Peter: that sounds just right to me; but a larger change than
we'd want to jump into for this one particular issue - it might have
its own unexpected consequences.

quoted text
> > If we had a SLAB_NOMERGE flag, would we want to apply it to the
> > bio cache or to the scsi_sense_cache or to both?  My difficulty
> > in answering that makes me wonder whether such a flag is right.
> 
> If this is critical to avoid memory deadlocks, I would suggest using
> mempools (or my reserve framework).

No, the critical part of it has been dealt with (small fix to scsi
free_list handling: which resembles a mempool, but done its own way).

What remains is about "unsightly" behaviour, the system having a
tendency to collapse briefly into far-from-efficient operation
when out of memory.

Hugh
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/