drm/amdgpu: flush TLB if valid PDE turns into PTE

Mapping huge page, 2MB aligned address with 2MB size, uses PDE0 as PTE.
If previously valid PDE0, PDE0.V=1 and PDE0.P=0 turns into PTE, this
requires TLB flush, otherwise page table walker will not read updated
PDE0.

Change page table update mapping to return table_freed flag to indicate
the previously valid PDE may have turned into a PTE if page table is
freed.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63035b7..57a6ad0 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1583,6 +1583,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,
                        while (cursor.pfn < frag_start) {
                                amdgpu_vm_free_pts(adev, params->vm, &cursor);
                                amdgpu_vm_pt_next(adev, &cursor);
+                               params->table_freed = true;
                        }
 
                } else if (frag >= shift) {
@@ -1610,6 +1611,7 @@ static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,
  * @res: ttm_resource to map
  * @pages_addr: DMA addresses to use for mapping
  * @fence: optional resulting fence
+ * @table_freed: return true if page table is freed
  *
  * Fill in the page table entries between @start and @last.
  *
@@ -1624,7 +1626,8 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
                                uint64_t flags, uint64_t offset,
                                struct ttm_resource *res,
                                dma_addr_t *pages_addr,
-                               struct dma_fence **fence)
+                               struct dma_fence **fence,
+                               bool *table_freed)
 {
        struct amdgpu_vm_update_params params;
        struct amdgpu_res_cursor cursor;
@@ -1719,6 +1722,9 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
        r = vm->update_funcs->commit(&params, fence);
 
+       if (table_freed)
+               *table_freed = params.table_freed;
+
 error_unlock:
        amdgpu_vm_eviction_unlock(vm);
        return r;
@@ -1815,7 +1821,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
                                                resv, mapping->start,
                                                mapping->last, update_flags,
                                                mapping->offset, mem,
-                                               pages_addr, last_update);
+                                               pages_addr, last_update, NULL);
                if (r)
                        return r;
        }
@@ -2026,7 +2032,7 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
                r = amdgpu_vm_bo_update_mapping(adev, adev, vm, false, false,
                                                resv, mapping->start,
                                                mapping->last, init_pte_value,
-                                               0, NULL, NULL, &f);
+                                               0, NULL, NULL, &f, NULL);
                amdgpu_vm_free_mapping(adev, vm, mapping, f);
                if (r) {
                        dma_fence_put(f);
@@ -3364,7 +3370,7 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
        }
 
        r = amdgpu_vm_bo_update_mapping(adev, adev, vm, true, false, NULL, addr,
-                                       addr, flags, value, NULL, NULL,
+                                       addr, flags, value, NULL, NULL, NULL,
                                        NULL);
        if (r)
                goto error_unlock;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 1ae5ea8..f8bb1e6 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -231,6 +231,11 @@ struct amdgpu_vm_update_params {
         * @num_dw_left: number of dw left for the IB
         */
        unsigned int num_dw_left;
+
+       /**
+        * @table_freed: return true if page table is freed when updating
+        */
+       bool table_freed;
 };
 
 struct amdgpu_vm_update_funcs {
@@ -404,7 +409,7 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
                                uint64_t flags, uint64_t offset,
                                struct ttm_resource *res,
                                dma_addr_t *pages_addr,
-                               struct dma_fence **fence);
+                               struct dma_fence **fence, bool *free_table);
 int amdgpu_vm_bo_update(struct amdgpu_device *adev,
                        struct amdgpu_bo_va *bo_va,
                        bool clear);

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index a4d1fb3..879c857 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1086,7 +1086,7 @@ svm_range_unmap_from_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 
        return amdgpu_vm_bo_update_mapping(adev, adev, vm, false, true, NULL,
                                           start, last, init_pte_value, 0,
-                                          NULL, NULL, fence);
+                                          NULL, NULL, fence, NULL);
 }
 
 static int
@@ -1139,6 +1139,7 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
                     struct amdgpu_device *bo_adev, struct dma_fence **fence)
 {
        struct amdgpu_bo_va bo_va;
+       bool table_freed = false;
        uint64_t pte_flags;
        int r = 0;
 
@@ -1160,7 +1161,8 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
                                        prange->mapping.last, pte_flags,
                                        prange->mapping.offset,
                                        prange->ttm_res,
-                                       dma_addr, &vm->last_update);
+                                       dma_addr, &vm->last_update,
+                                       &table_freed);
        if (r) {
                pr_debug("failed %d to map to gpu 0x%lx\n", r, prange->start);
                goto out;
@@ -1176,6 +1178,13 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm,
        if (fence)
                *fence = dma_fence_get(vm->last_update);
 
+       if (table_freed) {
+               struct kfd_process *p;
+
+               p = container_of(prange->svms, struct kfd_process, svms);
+               amdgpu_amdkfd_flush_gpu_tlb_pasid((struct kgd_dev *)adev,
+                                                 p->pasid);
+       }
 out:
        prange->mapping.bo_va = NULL;
        return r;