Re: adding swap workarounds oom - was: Re: Out of Memory: Killed process 16498 (java).

[Jens Axboe <axboe@suse.de>]

On Wed, Mar 01 2006, Andi Kleen wrote:
> On Wednesday 01 March 2006 14:41, Jens Axboe wrote:
> > On Wed, Mar 01 2006, Andy Chittenden wrote:
> > > with revised patch that does:
> > > 
> > >                 printk("sg%d: dma=%llx, dma_len=%u/%u, pfn=%lu\n", i,
> > > (unsigned long long) sg->dma_address, sg->dma_length, sg->offset,
> > > page_to_pfn(sg->page));
> > 
> > That is correct, thanks!
> > 
> > > hda: DMA table too small
> > > ide dma table, 255 entries, bounce pfn 1310720
> > > sg0: dma=81c8800, dma_len=4096/0, pfn=1296369
> > 
> > Still the same badness here, it's 2kb into a page so straddles two pages
> > for one entry.
> 
> That's normal if it was in the IOMMU and a merged entry. 
> 
> You can try iommu=nomerge.
> 
> Or maybe the higher layers are passing in physically continuous pages
> that get merged? Not too unlikely at boot.

But that would have to be 1kb or 512b io going in, I would expect 4kb to
be the base entries for any normal type of setup (with a 4kb fs). The
8kb could be one such merged entry, I agree.

> > Andi, any idea what is going on here? Why is this throwing up all of a
> > sudden??
> 
> What is throwing up exactly?
>
> There was a change recently in the merging algorithm, but it shouldn't
> cause any bad side effects for correct users of *_map_sg()

That the request we end up passing to blk_rq_map_sg() and then to
pci_map_sg() ends up with more entries than the driver advertised. So
far I think only Andy reported this, and then only with your
blk_queue_bounce_limit() patch applied.

-- 
Jens Axboe

-
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/