On 2022-03-28 13:55, Jonathan Marek wrote:

This matches the implementation of iommu_map_sgtable() used for the non-per-process page tables path.

This works around the dma_map_sgtable() call (used to invalidate cache) overwriting sgt->nents with 1 (which is probably a separate issue).

FWIW that's expected behaviour. The sgtable APIs use nents to keep track of the number of DMA segments, while orig_nents represents the physical segments. IIUC you're not actually using the DMA mapping, just relying on its side-effects, so it's still the physical segments that you care about for your private IOMMU mapping here.

Fixes: b145c6e65eb0 ("drm/msm: Add support to create a local pagetable") Signed-off-by: Jonathan Marek jonathan@marek.ca

---

drivers/gpu/drm/msm/msm_iommu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/msm/msm_iommu.c b/drivers/gpu/drm/msm/msm_iommu.c index bcaddbba564df..22935ef26a3a1 100644 --- a/drivers/gpu/drm/msm/msm_iommu.c +++ b/drivers/gpu/drm/msm/msm_iommu.c @@ -58,7 +58,7 @@ static int msm_iommu_pagetable_map(struct msm_mmu *mmu, u64 iova, u64 addr = iova; unsigned int i;

• for_each_sg(sgt->sgl, sg, sgt->nents, i) {

• for_each_sg(sgt->sgl, sg, sgt->orig_nents, i) {

Even better would be to use for_each_sgtable_sg(), which exists primarily for the sake of avoiding this exact confusion.

Robin.

size_t size = sg->length;
phys_addr_t phys = sg_phys(sg);