gpu: nvgpu: move local golden context memory allocation to poweorn

- Separate out local golden context memory allocation from
  nvgpu_gr_global_ctx_init_local_golden_image() into a new function
  nvgpu_gr_global_ctx_alloc_local_golden_image().
- Add a new member local_golden_image_copy to struct
  nvgpu_gr_obj_ctx_golden_image to store copy used for context
  verification.
- Allocate local golden context memory from nvgpu_gr_obj_ctx_init()
  which is called during poweron path.
- Remove memory allocation from nvgpu_gr_obj_ctx_save_golden_ctx().
- Disable test test_gr_obj_ctx_error_injection since it needs rework
  to accomodate the new changes.
- Fix below tests to allocate local golden context memory :
  test_gr_global_ctx_local_ctx_error_injection
  test_gr_setup_alloc_obj_ctx

Bug 3307637

Change-Id: I2f760d524881fd328346838ea9ce0234358f8e51
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2633713
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/gr/global_ctx.c

@@ -363,36 +363,35 @@ bool nvgpu_gr_global_ctx_buffer_ready(
 	return false;
 }
 
-struct nvgpu_gr_global_ctx_local_golden_image *
+int nvgpu_gr_global_ctx_alloc_local_golden_image(struct gk20a *g,
-nvgpu_gr_global_ctx_init_local_golden_image(struct gk20a *g,
+		struct nvgpu_gr_global_ctx_local_golden_image **img,
-	struct nvgpu_mem *source_mem, size_t size)
+		size_t size)
 {
 	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image;
 
-#ifdef NVGPU_UNITTEST_FAULT_INJECTION_ENABLEMENT
-	if (nvgpu_posix_fault_injection_handle_call(
-			nvgpu_local_golden_image_get_fault_injection())) {
-		return NULL;
-	}
-#endif
-
 	local_golden_image = nvgpu_kzalloc(g, sizeof(*local_golden_image));
 	if (local_golden_image == NULL) {
-		return NULL;
+		return -ENOMEM;
 	}
 
 	local_golden_image->context = nvgpu_vzalloc(g, size);
 	if (local_golden_image->context == NULL) {
 		nvgpu_kfree(g, local_golden_image);
-		return NULL;
+		return -ENOMEM;
 	}
 
 	local_golden_image->size = size;
 
-	nvgpu_mem_rd_n(g, source_mem, 0, local_golden_image->context,
+	*img = local_golden_image;
-		nvgpu_safe_cast_u64_to_u32(size));
+	return 0;
+}
 
-	return local_golden_image;
+void nvgpu_gr_global_ctx_init_local_golden_image(struct gk20a *g,
+		struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image,
+		struct nvgpu_mem *source_mem, size_t size)
+{
+	nvgpu_mem_rd_n(g, source_mem, 0, local_golden_image->context,
+		nvgpu_safe_cast_u64_to_u32(local_golden_image->size));
 }
 
 #ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION

drivers/gpu/nvgpu/common/gr/obj_ctx.c

@@ -629,14 +629,11 @@ static int nvgpu_gr_obj_ctx_save_golden_ctx(struct gk20a *g,
 	struct nvgpu_mem *gr_mem;
 	u64 size;
 	u32 data;
-#ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
-	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image_temp =
-									NULL;
-#endif
 
 	nvgpu_log(g, gpu_dbg_gr, " ");
 
 	gr_mem = nvgpu_gr_ctx_get_ctx_mem(gr_ctx);
+	size = nvgpu_gr_obj_ctx_get_golden_image_size(golden_image);
 
 #ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
 	/*
@@ -644,15 +641,8 @@ static int nvgpu_gr_obj_ctx_save_golden_ctx(struct gk20a *g,
 	 * before second golden context save. This temporary copy is
 	 * saved in local_golden_image_temp.
 	 */
-
+	nvgpu_gr_global_ctx_init_local_golden_image(g,
-	size = nvgpu_gr_obj_ctx_get_golden_image_size(golden_image);
+			golden_image->local_golden_image_copy, gr_mem, size);
-
-	local_golden_image_temp =
-		nvgpu_gr_global_ctx_init_local_golden_image(g, gr_mem, size);
-	if (local_golden_image_temp == NULL) {
-		err = -ENOMEM;
-		goto clean_up;
-	}
 #endif
 
 	data = g->ops.gr.falcon.get_fecs_current_ctx_data(g, inst_block);
@@ -662,23 +652,13 @@ static int nvgpu_gr_obj_ctx_save_golden_ctx(struct gk20a *g,
 		goto clean_up;
 	}
 
-	size = nvgpu_gr_obj_ctx_get_golden_image_size(golden_image);
+	nvgpu_gr_global_ctx_init_local_golden_image(g,
-
+			golden_image->local_golden_image, gr_mem, size);
-	golden_image->local_golden_image =
-		nvgpu_gr_global_ctx_init_local_golden_image(g, gr_mem, size);
-	if (golden_image->local_golden_image == NULL) {
-		err = -ENOMEM;
-		goto clean_up;
-	}
 
 #ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
 	/* Before second golden context save restore to before known state */
 	nvgpu_gr_global_ctx_load_local_golden_image(g,
-					local_golden_image_temp, gr_mem);
+			golden_image->local_golden_image_copy, gr_mem);
-	/* free local copy now */
-	nvgpu_gr_global_ctx_deinit_local_golden_image(g,
-						local_golden_image_temp);
-	local_golden_image_temp = NULL;
 
 	/* Initiate second golden context save */
 	data = g->ops.gr.falcon.get_fecs_current_ctx_data(g, inst_block);
@@ -689,32 +669,26 @@ static int nvgpu_gr_obj_ctx_save_golden_ctx(struct gk20a *g,
 	}
 
 	/* Copy the data to local buffer */
-	local_golden_image_temp =
+	nvgpu_gr_global_ctx_init_local_golden_image(g,
-		nvgpu_gr_global_ctx_init_local_golden_image(g, gr_mem, size);
+			golden_image->local_golden_image_copy, gr_mem, size);
-	if (local_golden_image_temp == NULL) {
-		err = -ENOMEM;
-		goto clean_up;
-	}
 
 	/* Compare two golden context images */
 	if (!nvgpu_gr_global_ctx_compare_golden_images(g,
 		nvgpu_mem_is_sysmem(gr_mem),
 		golden_image->local_golden_image,
-		local_golden_image_temp,
+		golden_image->local_golden_image_copy,
 		size)) {
 		nvgpu_err(g, "golden context mismatch");
 		err = -ENOMEM;
 	}
+
+	/* free temporary copy now */
+	nvgpu_gr_global_ctx_deinit_local_golden_image(g,
+			golden_image->local_golden_image_copy);
+	golden_image->local_golden_image_copy = NULL;
 #endif
 
 clean_up:
-#ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
-	if (local_golden_image_temp != NULL) {
-		nvgpu_gr_global_ctx_deinit_local_golden_image(g,
-						local_golden_image_temp);
-	}
-#endif
-
 	if (err == 0) {
 		nvgpu_log(g, gpu_dbg_gr, "golden image saved with size = %llu", size);
 	}
@@ -961,6 +935,7 @@ int nvgpu_gr_obj_ctx_init(struct gk20a *g,
 	struct nvgpu_gr_obj_ctx_golden_image **gr_golden_image, u32 size)
 {
 	struct nvgpu_gr_obj_ctx_golden_image *golden_image;
+	int err;
 
 	nvgpu_log(g, gpu_dbg_gr, "size = %u", size);
 
@@ -973,6 +948,24 @@ int nvgpu_gr_obj_ctx_init(struct gk20a *g,
 
 	nvgpu_mutex_init(&golden_image->ctx_mutex);
 
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g,
+			&golden_image->local_golden_image, size);
+	if (err != 0) {
+		nvgpu_kfree(g, golden_image);
+		return err;
+	}
+
+#ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g,
+			&golden_image->local_golden_image_copy, size);
+	if (err != 0) {
+		nvgpu_gr_global_ctx_deinit_local_golden_image(g,
+			golden_image->local_golden_image);
+		nvgpu_kfree(g, golden_image);
+		return err;
+	}
+#endif
+
 	*gr_golden_image = golden_image;
 
 	return 0;

drivers/gpu/nvgpu/common/gr/obj_ctx_priv.h

@@ -53,6 +53,14 @@ struct nvgpu_gr_obj_ctx_golden_image {
 	 * Pointer to local Golden context image struct.
 	 */
 	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image;
+
+#ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
+	/**
+	 * Pointer to local Golden context image struct used for Golden
+	 * context verification.
+	 */
+	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image_copy;
+#endif
 };
 
 #endif /* NVGPU_GR_OBJ_CTX_PRIV_H */

drivers/gpu/nvgpu/include/nvgpu/gr/global_ctx.h

@@ -240,15 +240,30 @@ bool nvgpu_gr_global_ctx_buffer_ready(
 	u32 index);
 
 /**
- * @brief Initialize local golden context image.
+ * @brief Allocate memory for local golden context image.
  *
- * @param g [in]		Pointer to GPU driver struct.
+ * @param g [in]			Pointer to GPU driver struct.
- * @param source_mem [in]	Pointer to source memory.
+ * @param local_golden_image [in]	Pointer to local golden context image struct.
- * @param size [in]		Size of local golden context image.
+ * @param size [in]			Size of local golden context image.
  *
  * This function allocates memory to store local golden context image
  * and also for #nvgpu_gr_global_ctx_local_golden_image structure.
  *
+ * @return 0 in case of success, < 0 in case of failure.
+ * @retval -ENOMEM if local golden image memory allocation fails.
+ */
+int nvgpu_gr_global_ctx_alloc_local_golden_image(struct gk20a *g,
+		struct nvgpu_gr_global_ctx_local_golden_image **img,
+		size_t size);
+
+/**
+ * @brief Initialize local golden context image.
+ *
+ * @param g [in]			Pointer to GPU driver struct.
+ * @param local_golden_image [in]	Pointer to local golden image to be initialized.
+ * @param source_mem [in]		Pointer to source memory.
+ * @param size [in]			Size of local golden context image.
+ *
  * This function will then initialize local golden context image by
  * copying contents of #source_mem into newly created image.
  *
@@ -256,11 +271,9 @@ bool nvgpu_gr_global_ctx_buffer_ready(
  * ever graphics context image for any channel. Subsequent graphics
  * context allocations will re-use this local golden image to
  * initialize. See #nvgpu_gr_global_ctx_load_local_golden_image.
- *
- * @return Pointer to local golden context image struct.
  */
-struct nvgpu_gr_global_ctx_local_golden_image *
+void nvgpu_gr_global_ctx_init_local_golden_image(struct gk20a *g,
-nvgpu_gr_global_ctx_init_local_golden_image(struct gk20a *g,
+	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image,
 	struct nvgpu_mem *source_mem, size_t size);
 
 /**

libs/dgpu/libnvgpu-drv-dgpu_safe.export

@@ -470,6 +470,7 @@ nvgpu_gr_falcon_remove_support
 nvgpu_gr_free
 nvgpu_gr_fs_state_init
 nvgpu_gr_get_config_ptr
+nvgpu_gr_global_ctx_alloc_local_golden_image
 nvgpu_gr_global_ctx_buffer_alloc
 nvgpu_gr_global_ctx_buffer_free
 nvgpu_gr_global_ctx_buffer_get_mem

libs/igpu/libnvgpu-drv-igpu_safe.export

@@ -486,6 +486,7 @@ nvgpu_gr_falcon_remove_support
 nvgpu_gr_free
 nvgpu_gr_fs_state_init
 nvgpu_gr_get_config_ptr
+nvgpu_gr_global_ctx_alloc_local_golden_image
 nvgpu_gr_global_ctx_buffer_alloc
 nvgpu_gr_global_ctx_buffer_free
 nvgpu_gr_global_ctx_buffer_get_mem

userspace/required_tests.ini

@@ -658,7 +658,7 @@ test_gr_intr_without_channel.gr_intr_channel_free=0
 [nvgpu_gr_obj_ctx]
 test_gr_init_setup_cleanup.gr_obj_ctx_cleanup=0
 test_gr_init_setup_ready.gr_obj_ctx_setup=0
-test_gr_obj_ctx_error_injection.gr_obj_ctx_alloc_errors=0
+test_gr_obj_ctx_error_injection.gr_obj_ctx_alloc_errors=2
 
 [nvgpu_gr_setup]
 test_gr_init_setup_cleanup.gr_setup_cleanup=0

userspace/units/gr/global_ctx/nvgpu-gr-global-ctx.c

@@ -204,8 +204,6 @@ int test_gr_global_ctx_local_ctx_error_injection(struct unit_module *m,
 	struct nvgpu_mem mem;
 	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image;
 	struct nvgpu_gr_global_ctx_local_golden_image *local_golden_image_bk;
-	struct nvgpu_posix_fault_inj *kmem_fi =
-		nvgpu_kmem_get_fault_injection();
 
 	/* Allocate dummy memory */
 	err = nvgpu_dma_alloc(g, DUMMY_SIZE, &mem);
@@ -213,41 +211,27 @@ int test_gr_global_ctx_local_ctx_error_injection(struct unit_module *m,
 		unit_return_fail(m, "failed to allocate dummy memory");
 	}
 
-	/* Fail allocation of nvgpu_gr_global_ctx_local_golden_image struct */
-	nvgpu_posix_enable_fault_injection(kmem_fi, true, 0);
-	local_golden_image = nvgpu_gr_global_ctx_init_local_golden_image(g,
-				&mem, DUMMY_SIZE);
-	if (local_golden_image != NULL) {
-		unit_return_fail(m, "unexpected success");
-	}
-
-	/* Fail allocation of local_golden_image->context */
-	nvgpu_posix_enable_fault_injection(kmem_fi, true, 1);
-	local_golden_image = nvgpu_gr_global_ctx_init_local_golden_image(g,
-				&mem, DUMMY_SIZE);
-	if (local_golden_image != NULL) {
-		unit_return_fail(m, "unexpected success");
-	}
-
 	/* Successful allocation of local golden context */
-	nvgpu_posix_enable_fault_injection(kmem_fi, false, 0);
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g, &local_golden_image, DUMMY_SIZE);
-	local_golden_image = nvgpu_gr_global_ctx_init_local_golden_image(g,
+	if (err != 0) {
-				&mem, DUMMY_SIZE);
-	if (local_golden_image == NULL) {
 		unit_return_fail(m, "failed to initialize local golden image");
 	}
 
+	nvgpu_gr_global_ctx_init_local_golden_image(g, local_golden_image, &mem, DUMMY_SIZE);
+
 	/* Trigger flush error during context load */
 	g->ops.mm.cache.l2_flush = dummy_l2_flush;
 	nvgpu_gr_global_ctx_load_local_golden_image(g,
 		local_golden_image, &mem);
 
 	/* Allocate dummy local golden context image */
-	local_golden_image_bk = nvgpu_gr_global_ctx_init_local_golden_image(g,
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g, &local_golden_image_bk, DUMMY_SIZE);
-					&mem, DUMMY_SIZE);
+	if (err != 0) {
-	if (local_golden_image_bk == NULL) {
 		unit_return_fail(m, "failed to initialize local golden image");
 	}
+
+	nvgpu_gr_global_ctx_init_local_golden_image(g, local_golden_image_bk, &mem, DUMMY_SIZE);
+
 #ifdef CONFIG_NVGPU_GR_GOLDEN_CTX_VERIFICATION
 	bool valid;
 

userspace/units/gr/global_ctx/nvgpu-gr-global-ctx.h

@@ -81,7 +81,8 @@ int test_gr_global_ctx_alloc_error_injection(struct unit_module *m,
  *
  * Test Type: Feature, Error guessing
  *
- * Targets: nvgpu_gr_global_ctx_init_local_golden_image,
+ * Targets: nvgpu_gr_global_ctx_alloc_local_golden_image,
+ *          nvgpu_gr_global_ctx_init_local_golden_image,
  *          nvgpu_gr_global_ctx_load_local_golden_image,
  *          nvgpu_gr_global_ctx_compare_golden_images,
  *          nvgpu_gr_global_ctx_deinit_local_golden_image

userspace/units/gr/obj_ctx/nvgpu-gr-obj-ctx.c

@@ -314,42 +314,6 @@ int test_gr_obj_ctx_error_injection(struct unit_module *m,
 		unit_return_fail(m, "unexpected success");
 	}
 
-	/*
-	 * Fail first local golden image allocation in
-	 * nvgpu_gr_global_ctx_init_local_golden_image()
-	 */
-	nvgpu_posix_enable_fault_injection(local_golden_image_fi, true, 0);
-	err = nvgpu_gr_obj_ctx_alloc(g, golden_image, global_desc, desc,
-			config, gr_ctx, subctx, vm, &inst_block,
-			VOLTA_COMPUTE_A, 0, false, false);
-	if (err == 0) {
-		unit_return_fail(m, "unexpected success");
-	}
-
-	/*
-	 * Fail second local golden image allocation in
-	 * nvgpu_gr_global_ctx_init_local_golden_image()
-	 */
-	nvgpu_posix_enable_fault_injection(local_golden_image_fi, true, 1);
-	err = nvgpu_gr_obj_ctx_alloc(g, golden_image, global_desc, desc,
-			config, gr_ctx, subctx, vm, &inst_block,
-			VOLTA_COMPUTE_A, 0, false, false);
-	if (err == 0) {
-		unit_return_fail(m, "unexpected success");
-	}
-
-	/*
-	 * Fail third local golden image allocation in
-	 * nvgpu_gr_global_ctx_init_local_golden_image()
-	 */
-	nvgpu_posix_enable_fault_injection(local_golden_image_fi, true, 2);
-	err = nvgpu_gr_obj_ctx_alloc(g, golden_image, global_desc, desc,
-			config, gr_ctx, subctx, vm, &inst_block,
-			VOLTA_COMPUTE_A, 0, false, false);
-	if (err == 0) {
-		unit_return_fail(m, "unexpected success");
-	}
-
 	/* Disable error injection */
 	nvgpu_posix_enable_fault_injection(local_golden_image_fi, false, 0);
 
@@ -434,7 +398,7 @@ int test_gr_obj_ctx_error_injection(struct unit_module *m,
 
 struct unit_module_test nvgpu_gr_obj_ctx_tests[] = {
 	UNIT_TEST(gr_obj_ctx_setup, test_gr_init_setup_ready, NULL, 0),
-	UNIT_TEST(gr_obj_ctx_alloc_errors, test_gr_obj_ctx_error_injection, NULL, 0),
+	UNIT_TEST(gr_obj_ctx_alloc_errors, test_gr_obj_ctx_error_injection, NULL, 2),
 	UNIT_TEST(gr_obj_ctx_cleanup, test_gr_init_setup_cleanup, NULL, 0),
 };
 

userspace/units/gr/setup/nvgpu-gr-setup.c

@@ -720,6 +720,17 @@ int test_gr_setup_alloc_obj_ctx(struct unit_module *m,
 	/* Set a default size for golden image */
 	g->gr->golden_image->size = 0x800;
 
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g,
+			&g->gr->golden_image->local_golden_image, 0x800);
+	if (err != 0) {
+		unit_return_fail(m, "local golden image alloc failed\n");
+	}
+	err = nvgpu_gr_global_ctx_alloc_local_golden_image(g,
+			&g->gr->golden_image->local_golden_image_copy, 0x800);
+	if (err != 0) {
+		unit_return_fail(m, "local golden image copy alloc failed\n");
+	}
+
 	/* Test with channel and tsg */
 	err = gr_test_setup_allocate_ch_tsg(m, g);
 	if (err != 0) {