Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Critical] drm/msm: bogus physical addresses in __do_map() when dma_iommu is active --
commit b6b3547f9503a1fd43535b38edfaee1a6356607a Author: Rob Clark <[email protected]> drm/msm: Convert map helper to use dma-address This commit switches __do_map() in drivers/gpu/drm/msm/msm_iommu.c from iterating over scatterlist page entries with for_each_sgtable_sg() and sg_phys() to iterating over DMA-mapped entries with for_each_sgtable_dma_sg(), sg_dma_len(), and sg_dma_address(). > diff --git a/drivers/gpu/drm/msm/msm_iommu.c b/drivers/gpu/drm/msm/msm_iommu.c > index 479e5450a4dcd..c2950d32c5075 100644 > --- a/drivers/gpu/drm/msm/msm_iommu.c > +++ b/drivers/gpu/drm/msm/msm_iommu.c > @@ -153,9 +153,9 @@ __do_map(unsigned long pgsize_bitmap, u64 iova, struct > sg_table *sgt, size_t off > u64 addr = iova; > unsigned int i; > > - for_each_sgtable_sg (sgt, sg, i) { > - size_t size = sg->length; > - phys_addr_t phys = sg_phys(sg); > + for_each_sgtable_dma_sg (sgt, sg, i) { > + size_t size = sg_dma_len(sg); > + phys_addr_t phys = sg_dma_address(sg); [Severity: Critical] Could passing sg_dma_address(sg) and sg_dma_len(sg) as physical addresses into ops->map() program virtual IOVAs from an inactive default DMA domain into the GPU and display SMMU page tables on platforms where use_dma_iommu() is true on drm_dev_dma_dev(obj->dev)? In drivers/gpu/drm/msm/msm_gem.c:get_pages() (and via dma_buf_map_attachment() for PRIME imports), the scatterlist is DMA-mapped against drm_dev_dma_dev(obj->dev): ret = dma_map_sgtable(drm_dev_dma_dev(obj->dev), msm_obj->sgt, DMA_BIDIRECTIONAL, map_attrs); On platforms where the iommus property is placed directly on the display controller node (mdp_dev, which is obj->dev->dev) rather than a parent mdss node, as handled in drivers/gpu/drm/msm/msm_kms.c:msm_kms_init_vm(): if (device_iommu_mapped(mdp_dev)) iommu_dev = mdp_dev; else if (mdss_dev && device_iommu_mapped(mdss_dev)) iommu_dev = mdss_dev; and the IOMMU driver does not return IOMMU_DOMAIN_IDENTITY from .def_domain_type, the IOMMU core allocates a default DMA domain during probe and drivers/iommu/dma-iommu.c:iommu_setup_dma_ops() sets dev->dma_iommu: dma_iommu = iommu_is_dma_domain(domain); dev_assign_dma_iommu(dev, dma_iommu); For example, drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c: qcom_smmu_client_of_match[] does not list "qcom,mdp5" (affecting MDP5 platforms such as SDM630, SDM660, and MSM8996), and qcom_iommu_ops in drivers/iommu/arm/arm-smmu/qcom_iommu.c does not implement .def_domain_type at all. When msm_iommu_new() later attaches an unmanaged domain via iommu_attach_device(), group->default_domain remains the DMA domain and use_dma_iommu(dev) stays true. As a result, kernel/dma/mapping.c: dma_go_direct() returns false even when get_dma_ops(dev) is NULL: if (use_dma_iommu(dev)) return false; if (likely(!ops)) return true; This causes dma_map_sgtable() to call iommu_dma_map_sg(), which allocates virtual IOVAs from the inactive group->default_domain, coalesces segments, and stores those IOVAs and coalesced lengths into sg_dma_address(sg) and sg_dma_len(sg). When __do_map() subsequently passes phys = sg_dma_address(sg) and size = sg_dma_len(sg) to io_pgtable_ops->map_pages() or iommu_domain->ops->map_pages() (which expect CPU physical addresses), won't this map bogus physical addresses into the active GPU and display page tables? -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=5
