Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [Critical] AB-BA deadlock between `dma_resv_lock` and page fault handling (`mmap_lock`) due to synchronous work cancellation in `xe_vma_destroy_late` - [Low] Missing `#endif` include guard in `xe_sync_types.h` causes compilation error --
commit b5c1e8acd90810b219cbad1a85281b1b796fc3ea Author: Srinivasan Shanmugam <[email protected]> drm/xe: Convert xe_user_fence to drm_user_fence This commit replaces the open-coded user fence implementation in xe_sync.c with the new common drm_user_fence helper. The xe_user_fence structure now embeds drm_user_fence, and lifecycle functions are moved to the new ops structure. > diff --git a/drivers/gpu/drm/xe/xe_sync.c b/drivers/gpu/drm/xe/xe_sync.c > index 37866768d64c2..f2eb66b8fe54d 100644 > --- a/drivers/gpu/drm/xe/xe_sync.c > +++ b/drivers/gpu/drm/xe/xe_sync.c [ ... ] > @@ -19,36 +18,60 @@ [ ... ] > +static void xe_ufence_worker(struct drm_user_fence *base, bool mm_ok) > { > - struct xe_user_fence *ufence = container_of(kref, struct xe_user_fence, > - refcount); > + struct xe_user_fence *ufence = container_of(base, struct xe_user_fence, > base); [ ... ] > + if (mm_ok) { > + if (copy_to_user(ufence->addr, &ufence->value, > sizeof(ufence->value))) [Severity: Critical] Could this copy_to_user() trigger a page fault that attempts to re-acquire the dma_resv_lock? If a userspace application mmaps a BO, creates a user fence pointing into that mapped BO address, and binds it via VM_BIND, this worker will target the mapped BO. When this occurs concurrently with VMA teardown, it appears to lead to a deadlock. See below. [ ... ] > diff --git a/drivers/gpu/drm/xe/xe_sync_types.h > b/drivers/gpu/drm/xe/xe_sync_types.h > index b88f1833e28cc..2a75f998558e6 100644 > --- a/drivers/gpu/drm/xe/xe_sync_types.h > +++ b/drivers/gpu/drm/xe/xe_sync_types.h > @@ -28,4 +27,3 @@ struct xe_sync_entry { > u32 flags; > }; > > -#endif [Severity: Low] Does this missing include guard cause a build failure? It looks like the #endif for #ifndef _XE_SYNC_TYPES_H_ was accidentally deleted, which will result in an unterminated #ifndef compiler error for any C file including this header. > diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c > index 19b3d0be79282..ef0784601f226 100644 > --- a/drivers/gpu/drm/xe/xe_vm.c > +++ b/drivers/gpu/drm/xe/xe_vm.c > @@ -1214,6 +1214,7 @@ static void xe_vma_destroy_late(struct xe_vma *vma) > struct xe_bo *bo = xe_vma_bo(vma); > > if (vma->ufence) { > + xe_sync_ufence_cancel_sync(vma->ufence); [Severity: Critical] Does this synchronous cancellation create an AB-BA deadlock? During VMA teardown, xe_vma_destroy_unlocked() acquires the BO's dma_resv_lock (and vm->lock) and then synchronously waits for the user fence worker to finish via cancel_work_sync() inside xe_sync_ufence_cancel_sync(). If the worker is concurrently executing copy_to_user() (as noted above) and triggers a page fault, the page fault handler (xe_gem_fault) will attempt to acquire the same dma_resv_lock to resolve the fault. Since the teardown thread already holds this lock while waiting for the worker, the two threads will deadlock. -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=2
