On Fri, Apr 15, 2016 at 8:55 PM, Shaohua Li <s...@fb.com> wrote: > Fair enough, it makes sense to ignore the DAC overhead. My point is the > cache case should respect the limit, since devices might not be able to > handle DMA64. Even without IOMMU, we use GFP_DMA32 or swiotlb to > guarantee DMA address is in range.
Absolutely, the cache should respect the limit. That'll be fixed in the next version; sorry I forgot to do it in this revision. > For the 'force DMA32' logic in > intel-iommu, it's in the code in day one, but I can't remember the > reason. My guess is it's to workaround driver/device bugs. I'd suggest > just deleting the logic and let cache only handle DMA64. Having looked at this further, DMA64 limits get cut down to the max address the IOMMU supports (in intel_alloc_iova()). In principle, DMA32 (or close to it) might end up as the limit even for DMA64. So it seems cleaner to just flush the rcache if we can't allocate; that takes care of everything. _______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
