From ba767b9d01a2c552d76cf6f46b125d50ec4147a6 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Thomas=20Hellstr=C3=B6m?= Date: Tue, 4 Mar 2025 18:33:42 +0100 Subject: [PATCH] drm/xe/userptr: Unmap userptrs in the mmu notifier MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit If userptr pages are freed after a call to the xe mmu notifier, the device will not be blocked out from theoretically accessing these pages unless they are also unmapped from the iommu, and this violates some aspects of the iommu-imposed security. Ensure that userptrs are unmapped in the mmu notifier to mitigate this. A naive attempt would try to free the sg table, but the sg table itself may be accessed by a concurrent bind operation, so settle for only unmapping. v3: - Update lockdep asserts. - Fix a typo (Matthew Auld) Fixes: 81e058a3e7fd ("drm/xe: Introduce helper to populate userptr") Cc: Oak Zeng Cc: Matthew Auld Cc: # v6.10+ Signed-off-by: Thomas Hellström Reviewed-by: Matthew Auld Acked-by: Matthew Brost Link: https://patchwork.freedesktop.org/patch/msgid/20250304173342.22009-4-thomas.hellstrom@linux.intel.com --- drivers/gpu/drm/xe/xe_hmm.c | 51 ++++++++++++++++++++++++++------ drivers/gpu/drm/xe/xe_hmm.h | 2 ++ drivers/gpu/drm/xe/xe_vm.c | 4 +++ drivers/gpu/drm/xe/xe_vm_types.h | 4 +++ 4 files changed, 52 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_hmm.c b/drivers/gpu/drm/xe/xe_hmm.c index be284b852307e..392102515f3d8 100644 --- a/drivers/gpu/drm/xe/xe_hmm.c +++ b/drivers/gpu/drm/xe/xe_hmm.c @@ -150,6 +150,45 @@ static int xe_build_sg(struct xe_device *xe, struct hmm_range *range, DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_NO_KERNEL_MAPPING); } +static void xe_hmm_userptr_set_mapped(struct xe_userptr_vma *uvma) +{ + struct xe_userptr *userptr = &uvma->userptr; + struct xe_vm *vm = xe_vma_vm(&uvma->vma); + + lockdep_assert_held_write(&vm->lock); + lockdep_assert_held(&vm->userptr.notifier_lock); + + mutex_lock(&userptr->unmap_mutex); + xe_assert(vm->xe, !userptr->mapped); + userptr->mapped = true; + mutex_unlock(&userptr->unmap_mutex); +} + +void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma) +{ + struct xe_userptr *userptr = &uvma->userptr; + struct xe_vma *vma = &uvma->vma; + bool write = !xe_vma_read_only(vma); + struct xe_vm *vm = xe_vma_vm(vma); + struct xe_device *xe = vm->xe; + + if (!lockdep_is_held_type(&vm->userptr.notifier_lock, 0) && + !lockdep_is_held_type(&vm->lock, 0) && + !(vma->gpuva.flags & XE_VMA_DESTROYED)) { + /* Don't unmap in exec critical section. */ + xe_vm_assert_held(vm); + /* Don't unmap while mapping the sg. */ + lockdep_assert_held(&vm->lock); + } + + mutex_lock(&userptr->unmap_mutex); + if (userptr->sg && userptr->mapped) + dma_unmap_sgtable(xe->drm.dev, userptr->sg, + write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE, 0); + userptr->mapped = false; + mutex_unlock(&userptr->unmap_mutex); +} + /** * xe_hmm_userptr_free_sg() - Free the scatter gather table of userptr * @uvma: the userptr vma which hold the scatter gather table @@ -161,16 +200,9 @@ static int xe_build_sg(struct xe_device *xe, struct hmm_range *range, void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma) { struct xe_userptr *userptr = &uvma->userptr; - struct xe_vma *vma = &uvma->vma; - bool write = !xe_vma_read_only(vma); - struct xe_vm *vm = xe_vma_vm(vma); - struct xe_device *xe = vm->xe; - struct device *dev = xe->drm.dev; - - xe_assert(xe, userptr->sg); - dma_unmap_sgtable(dev, userptr->sg, - write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE, 0); + xe_assert(xe_vma_vm(&uvma->vma)->xe, userptr->sg); + xe_hmm_userptr_unmap(uvma); sg_free_table(userptr->sg); userptr->sg = NULL; } @@ -297,6 +329,7 @@ int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma, xe_mark_range_accessed(&hmm_range, write); userptr->sg = &userptr->sgt; + xe_hmm_userptr_set_mapped(uvma); userptr->notifier_seq = hmm_range.notifier_seq; up_read(&vm->userptr.notifier_lock); kvfree(pfns); diff --git a/drivers/gpu/drm/xe/xe_hmm.h b/drivers/gpu/drm/xe/xe_hmm.h index 9602cb7d976dd..0ea98d8e7bbc7 100644 --- a/drivers/gpu/drm/xe/xe_hmm.h +++ b/drivers/gpu/drm/xe/xe_hmm.h @@ -13,4 +13,6 @@ struct xe_userptr_vma; int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma, bool is_mm_mmap_locked); void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma); + +void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma); #endif diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index dd422ac95dc04..3dbd3d38008a5 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -621,6 +621,8 @@ static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uv err = xe_vm_invalidate_vma(vma); XE_WARN_ON(err); } + + xe_hmm_userptr_unmap(uvma); } static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni, @@ -1039,6 +1041,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm, INIT_LIST_HEAD(&userptr->invalidate_link); INIT_LIST_HEAD(&userptr->repin_link); vma->gpuva.gem.offset = bo_offset_or_userptr; + mutex_init(&userptr->unmap_mutex); err = mmu_interval_notifier_insert(&userptr->notifier, current->mm, @@ -1080,6 +1083,7 @@ static void xe_vma_destroy_late(struct xe_vma *vma) * them anymore */ mmu_interval_notifier_remove(&userptr->notifier); + mutex_destroy(&userptr->unmap_mutex); xe_vm_put(vm); } else if (xe_vma_is_null(vma)) { xe_vm_put(vm); diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h index 1fe79bf23b6b5..eca73c4197d4e 100644 --- a/drivers/gpu/drm/xe/xe_vm_types.h +++ b/drivers/gpu/drm/xe/xe_vm_types.h @@ -59,12 +59,16 @@ struct xe_userptr { struct sg_table *sg; /** @notifier_seq: notifier sequence number */ unsigned long notifier_seq; + /** @unmap_mutex: Mutex protecting dma-unmapping */ + struct mutex unmap_mutex; /** * @initial_bind: user pointer has been bound at least once. * write: vm->userptr.notifier_lock in read mode and vm->resv held. * read: vm->userptr.notifier_lock in write mode or vm->resv held. */ bool initial_bind; + /** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */ + bool mapped; #if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT) u32 divisor; #endif -- 2.39.5