On 09/09/2020 14:03, Chris Wilson wrote:
Quoting Tvrtko Ursulin (2020-09-09 13:44:57)
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
When walking DMA mapped scatterlists sg_dma_len has to be used since it
can be different (coalesced) from the backing store entry.
This also means we have to end the walk when encountering a zero length
DMA entry and cannot rely on the normal sg list end marker.
Both issues were there in theory for some time but were hidden by the fact
Intel IOMMU driver was never coalescing entries. As there are ongoing
efforts to change this we need to start handling it.
Does this change anything if we were already feeding in coalesced
entries? I doubt we made all sg generators compact the lists though.
I don't think so. My takeaway from this was a realization that page and
dma coalescing can be completely different (more so once the iommu
patches get in). I did not think of any other problems as long as the
dma remapping fits into the number of available sg table entries we
allocated. It is just a bit weird, this duality of sg_table "views".
The change looks reasonable, it seems you have kept the dma and page
iterators distinct, which was the only worry that occurred to me.
It surprised me we have two iterators, well one and a half. The sgt_dma
one being the half of an iterator. :)
Regards,
Tvrtko
P.S. Well as I was writing this reply full test results came in and
something is still broken, so scratch that..
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
