Thank you for your contribution! Sashiko AI review found 3 potential issues to consider: - [High] drm/amdgpu: UAF on bo_va in amdgpu_vm_update_ctx_fini() after amdgpu_vm_bo_del() - [High] drm/amdgpu: skipped delete in amdgpu_vm_update_ctx_add_freed_mapping() - [Low] drm/amdgpu: update_link placed under wrong lock comment in amdgpu_bo_va --
commit 3f1753904c640653510160bf642992a7a55fbc3d Author: Natalie Vock <[email protected]> drm/amdgpu: Track dirty bo_va structures in update ctx This commit adds an update_link list head to struct amdgpu_bo_va and tracks modified bo_va structures in a dirty_bo_vas list on struct amdgpu_vm_update_ctx. > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h > b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h > index e0090dc10682f..068c57ed537fd 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h > @@ -80,6 +80,8 @@ struct amdgpu_bo_va { > /* protected by bo being reserved */ > unsigned ref_count; > > + struct list_head update_link; > + > /* all other members protected by the VM PD being reserved */ > struct dma_fence *last_pt_update; > [Severity: Low] Should update_link be placed below the "all other members protected by the VM PD being reserved" comment? It is modified by the update_ctx helpers while only the VM PD reservation lock is held, including when the individual BO is NULL (such as fpriv->prt_va) or not reserved (when clearing overlapping mappings of other BOs in the VM). > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c > b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c > index 1e2d45979efdb..330c288053352 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c [ ... ] > @@ -1733,8 +1735,10 @@ static void > amdgpu_vm_update_ctx_add_freed_mapping(struct amdgpu_vm_update_ctx *ctx, > struct amdgpu_bo_va_mapping *mapping) > { > - WARN_ON(!list_empty(&mapping->pending_list)); > - list_add(&mapping->pending_list, &ctx->deleted); > + if (list_empty(&mapping->bo_va->update_link)) > + list_add(&mapping->bo_va->update_link, &ctx->dirty_bo_vas); > + if (list_empty(&mapping->pending_list)) > + list_add(&mapping->pending_list, &ctx->deleted); > } [Severity: High] Can skipping list_add(&mapping->pending_list, &ctx->deleted) when mapping->pending_list is non-empty leave freed mappings in vm->va? Within one update_ctx, amdgpu_vm_bo_clear_mappings() and amdgpu_vm_bo_replace_map() put overlapping mappings onto ctx->cleared via mapping->pending_list; if amdgpu_vm_bo_unmap() is then called on that mapping in the same context, amdgpu_vm_update_ctx_add_freed_mapping() leaves it on ctx->cleared instead of moving it to ctx->deleted, so amdgpu_vm_update_ctx_execute_replace() re-inserts any non-cleared remainder. Furthermore, if an update_ctx is aborted before amdgpu_vm_update_ctx_execute_replace() finishes processing ctx->cleared (for example, if amdgpu_vm_bo_replace_map() fails in amdgpu_vm_update_ctx_add_new_mapping()), neither amdgpu_vm_update_ctx_fini() here nor amdgpu_vm_update_ctx_rollback() in commit "drm/amdgpu: Add function to roll back a VM update ctx" unlinks entries from ctx->cleared. When the BO is later removed in amdgpu_vm_bo_del(), amdgpu_vm_update_ctx_add_freed_mapping() silently skips adding the mapping to update_ctx.deleted and kfree(bo_va) frees bo_va while the mapping remains in vm->va with a dangling mapping->bo_va pointer, causing a use-after-free on subsequent vm->va traversals. > int amdgpu_vm_update_ctx_execute_replace(struct amdgpu_vm_update_ctx *ctx) [ ... ] > @@ -1878,10 +1883,16 @@ void amdgpu_vm_update_ctx_init(struct > amdgpu_vm_update_ctx *ctx, > void amdgpu_vm_update_ctx_fini(struct amdgpu_vm_update_ctx *ctx) > { > struct interval_tree_node *node, *next_node; > + struct amdgpu_bo_va *bo_va, *next_bo_va; > > list_splice(&ctx->freed, &ctx->vm->delayed_freed); > amdgpu_sync_free(&ctx->sync); > > + list_for_each_entry_safe(bo_va, next_bo_va, &ctx->dirty_bo_vas, > + update_link) { > + list_del_init(&bo_va->update_link); > + } [Severity: High] Does this cause a use-after-free on bo_va when a caller unmaps and deletes a bo_va before finalizing update_ctx? In amdgpu_unmap_static_csa(), amdgpu_vm_bo_unmap() adds bo_va->update_link to update_ctx.dirty_bo_vas, and amdgpu_vm_bo_del() then frees bo_va before amdgpu_vm_update_ctx_fini() runs: drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c:amdgpu_unmap_static_csa() { ... r = amdgpu_vm_bo_unmap(&update_ctx, bo_va, csa_addr); if (r) { drm_err(adev_to_drm(adev), "failed to do bo_unmap on static CSA, err=%d\n", r); goto error_ctx; } amdgpu_vm_update_ctx_commit(&update_ctx); amdgpu_vm_bo_del(adev, bo_va); error_ctx: amdgpu_vm_update_ctx_fini(&update_ctx); ... } Because amdgpu_vm_bo_del() calls kfree(bo_va) without unlinking bo_va->update_link from the outer update_ctx.dirty_bo_vas, amdgpu_vm_update_ctx_fini() executes list_del_init(&bo_va->update_link) on the freed bo_va. Unlinking bo_va->update_link in amdgpu_vm_bo_del() before kfree(bo_va) would avoid this. -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=5
