[24/34] drm/amdkfd: SVM API call to restore page tables

Commit Message

Felix Kuehling April 15, 2021, 1:23 a.m. UTC

From: Alex Sierra <[e-mail]>

Use SVM API to restore page tables when retry fault and
compute context are enabled.

Signed-off-by: Alex Sierra <[e-mail]>
Reviewed-by: Felix Kuehling <[e-mail]>
Signed-off-by: Felix Kuehling <[e-mail]>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

Patch

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index f06de6dd7f16..f95bcda8463f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -37,6 +37,7 @@ 
 #include "amdgpu_gmc.h"
 #include "amdgpu_xgmi.h"
 #include "amdgpu_dma_buf.h"
+#include "kfd_svm.h"
 
 /**
  * DOC: GPUVM
@@ -3298,6 +3299,7 @@  void amdgpu_vm_set_task_info(struct amdgpu_vm *vm)
 bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 			    uint64_t addr)
 {
+	bool is_compute_context = false;
 	struct amdgpu_bo *root;
 	uint64_t value, flags;
 	struct amdgpu_vm *vm;
@@ -3305,15 +3307,25 @@  bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 
 	spin_lock(&adev->vm_manager.pasid_lock);
 	vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
-	if (vm)
+	if (vm) {
 		root = amdgpu_bo_ref(vm->root.base.bo);
-	else
+		is_compute_context = vm->is_compute_context;
+	} else {
 		root = NULL;
+	}
 	spin_unlock(&adev->vm_manager.pasid_lock);
 
 	if (!root)
 		return false;
 
+	addr /= AMDGPU_GPU_PAGE_SIZE;
+
+	if (is_compute_context &&
+	    !svm_range_restore_pages(adev, pasid, addr)) {
+		amdgpu_bo_unref(&root);
+		return true;
+	}
+
 	r = amdgpu_bo_reserve(root, true);
 	if (r)
 		goto error_unref;
@@ -3327,18 +3339,16 @@  bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 	if (!vm)
 		goto error_unlock;
 
-	addr /= AMDGPU_GPU_PAGE_SIZE;
 	flags = AMDGPU_PTE_VALID | AMDGPU_PTE_SNOOPED |
 		AMDGPU_PTE_SYSTEM;
 
-	if (vm->is_compute_context) {
+	if (is_compute_context) {
 		/* Intentionally setting invalid PTE flag
 		 * combination to force a no-retry-fault
 		 */
 		flags = AMDGPU_PTE_EXECUTABLE | AMDGPU_PDE_PTE |
 			AMDGPU_PTE_TF;
 		value = 0;
-
 	} else if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_NEVER) {
 		/* Redirect the access to the dummy page */
 		value = adev->dummy_page_addr;