答复: [外部邮件] Re: [PATCH] iommu/dma-iommu: Fix wrong scatterlist length assignment in P2PDMA path

["Li,Rongqing" <lirongqing@baidu.com>]

> On 2026-05-30 05:28, lirongqing wrote:
> > From: Li RongQing <lirongqing@baidu.com>
> >
> > In iommu_dma_map_sg(), when handling PCI P2PDMA cases, the DMA
> length
> > of the current scatterlist segment `s` is incorrectly assigned from
> > the head entry `sg->length` instead of the current entry `s->length`.
> >
> > This typo causes all P2PDMA segments in the scatterlist to inherit the
> > length of the first segment, leading to corrupted DMA lengths for
> > multi- segment scatterlists.
> >
> > Fix this by using `s->length` instead of `sg->length`.
> >
> > Fixes: a25e7962db ("PCI/P2PDMA: Refactor the p2pdma mapping helpers")
> > Signed-off-by: Li RongQing <lirongqing@baidu.com>
> 
> Hmm, I thought I had tested this stuff pretty thoroughly so I'm surprised I
> missed that. But I guess seeing the users of these specifically don't allow
> mixing with anonymous memory the vast majority of tests would only have
> one segment.
> 
> Can you confirm that you've reproduced and tested this and it actually fixes a
> bug? Not just seeing something that looks strange. I have a vague feeling it
> was intentional but looking at the code it looks wrong to me too.
> 
> Thanks,
> 
Cc: Christoph Hellwig

Thanks for your feedback!

This was found via code inspection. You are right that if nents == 1, or if nents > 1 but all segments happen to have the exact same length, this typo remains benign by pure coincidence.

However, for a multi-segment scatterlist with varying lengths (e.g., a short 2KB head segment followed by standard 4KB segments), sg_dma_len(s) will mistakenly assign the 2KB head length to all subsequent 4KB segments. This will inevitably lead to silent data truncation or IOMMU page faults.

Since sg_phys(s) correctly takes the current segment's address, sg_dma_len(s) should definitely use s->length to match it.

I don't have a specific P2PDMA test rig to force multi-segment lists on hand right now, but the typo and its impact on varied segment lengths seem deterministic.

Thanks

[Li,Rongqing] 

> Logan