1 files changed, 53 insertions, 138 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 4e36551ab50b..c136bd449744 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -80,6 +80,8 @@ enum amdgpu_ras_retire_page_reservation {
 
 atomic_t amdgpu_ras_in_intr = ATOMIC_INIT(0);
 
+static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con,
+				uint64_t addr);
 static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev,
 				uint64_t addr);
 
@@ -516,9 +518,9 @@ struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
 /* obj end */
 
 static void amdgpu_ras_parse_status_code(struct amdgpu_device *adev,
-				  const char* 		invoke_type,
-				  const char* 		block_name,
-				  enum ta_ras_status 	ret)
+					 const char* invoke_type,
+					 const char* block_name,
+					 enum ta_ras_status ret)
 {
 	switch (ret) {
 	case TA_RAS_STATUS__SUCCESS:
@@ -607,7 +609,7 @@ int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
 	if (!con)
 		return -EINVAL;
 
-        info = kzalloc(sizeof(union ta_ras_cmd_input), GFP_KERNEL);
+	info = kzalloc(sizeof(union ta_ras_cmd_input), GFP_KERNEL);
 	if (!info)
 		return -ENOMEM;
 
@@ -903,13 +905,6 @@ int amdgpu_ras_error_inject(struct amdgpu_device *adev,
 	return ret;
 }
 
-int amdgpu_ras_error_cure(struct amdgpu_device *adev,
-		struct ras_cure_if *info)
-{
-	/* psp fw has no cure interface for now. */
-	return 0;
-}
-
 /* get the total error counts on all IPs */
 unsigned long amdgpu_ras_query_error_count(struct amdgpu_device *adev,
 		bool is_ce)
@@ -953,7 +948,7 @@ static char *amdgpu_ras_badpage_flags_str(unsigned int flags)
 	case AMDGPU_RAS_RETIRE_PAGE_FAULT:
 	default:
 		return "F";
-	};
+	}
 }
 
 /**
@@ -1172,7 +1167,7 @@ static void amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev)
 			con->dir, &con->disable_ras_err_cnt_harvest);
 }
 
-void amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
+static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
 		struct ras_fs_if *head)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
@@ -1194,7 +1189,6 @@ void amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
 
 void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev)
 {
-#if defined(CONFIG_DEBUG_FS)
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_manager *obj;
 	struct ras_fs_if fs_info;
@@ -1203,7 +1197,7 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev)
 	 * it won't be called in resume path, no need to check
 	 * suspend and gpu reset status
 	 */
-	if (!con)
+	if (!IS_ENABLED(CONFIG_DEBUG_FS) || !con)
 		return;
 
 	amdgpu_ras_debugfs_create_ctrl_node(adev);
@@ -1217,10 +1211,9 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev)
 			amdgpu_ras_debugfs_create(adev, &fs_info);
 		}
 	}
-#endif
 }
 
-void amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
+static void amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
 		struct ras_common_if *head)
 {
 	struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
@@ -1234,7 +1227,6 @@ void amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
 
 static void amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
 {
-#if defined(CONFIG_DEBUG_FS)
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_manager *obj, *tmp;
 
@@ -1243,7 +1235,6 @@ static void amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
 	}
 
 	con->dir = NULL;
-#endif
 }
 /* debugfs end */
 
@@ -1291,7 +1282,8 @@ static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
 
 static int amdgpu_ras_fs_fini(struct amdgpu_device *adev)
 {
-	amdgpu_ras_debugfs_remove_all(adev);
+	if (IS_ENABLED(CONFIG_DEBUG_FS))
+		amdgpu_ras_debugfs_remove_all(adev);
 	amdgpu_ras_sysfs_remove_all(adev);
 	return 0;
 }
@@ -1477,8 +1469,8 @@ static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev)
 }
 
 /* Parse RdRspStatus and WrRspStatus */
-void amdgpu_ras_error_status_query(struct amdgpu_device *adev,
-		struct ras_query_if *info)
+static void amdgpu_ras_error_status_query(struct amdgpu_device *adev,
+					  struct ras_query_if *info)
 {
 	/*
 	 * Only two block need to query read/write
@@ -1551,10 +1543,12 @@ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev,
 			.size = AMDGPU_GPU_PAGE_SIZE,
 			.flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED,
 		};
-
-		if (data->last_reserved <= i)
+		ret = amdgpu_vram_mgr_query_page_status(
+				ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM),
+				data->bps[i].retired_page);
+		if (ret == -EBUSY)
 			(*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING;
-		else if (data->bps_bo[i] == NULL)
+		else if (ret == -ENOENT)
 			(*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_FAULT;
 	}
 
@@ -1606,12 +1600,9 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev,
 	unsigned int new_space = old_space + pages;
 	unsigned int align_space = ALIGN(new_space, 512);
 	void *bps = kmalloc(align_space * sizeof(*data->bps), GFP_KERNEL);
-	struct amdgpu_bo **bps_bo =
-			kmalloc(align_space * sizeof(*data->bps_bo), GFP_KERNEL);
 
-	if (!bps || !bps_bo) {
+	if (!bps) {
 		kfree(bps);
-		kfree(bps_bo);
 		return -ENOMEM;
 	}
 
@@ -1620,14 +1611,8 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev,
 				data->count * sizeof(*data->bps));
 		kfree(data->bps);
 	}
-	if (data->bps_bo) {
-		memcpy(bps_bo, data->bps_bo,
-				data->count * sizeof(*data->bps_bo));
-		kfree(data->bps_bo);
-	}
 
 	data->bps = bps;
-	data->bps_bo = bps_bo;
 	data->space_left += align_space - old_space;
 	return 0;
 }
@@ -1639,6 +1624,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_err_handler_data *data;
 	int ret = 0;
+	uint32_t i;
 
 	if (!con || !con->eh_data || !bps || pages <= 0)
 		return 0;
@@ -1648,16 +1634,26 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 	if (!data)
 		goto out;
 
-	if (data->space_left <= pages)
-		if (amdgpu_ras_realloc_eh_data_space(adev, data, pages)) {
+	for (i = 0; i < pages; i++) {
+		if (amdgpu_ras_check_bad_page_unlock(con,
+			bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT))
+			continue;
+
+		if (!data->space_left &&
+			amdgpu_ras_realloc_eh_data_space(adev, data, 256)) {
 			ret = -ENOMEM;
 			goto out;
 		}
 
-	memcpy(&data->bps[data->count], bps, pages * sizeof(*data->bps));
-	data->count += pages;
-	data->space_left -= pages;
+		amdgpu_vram_mgr_reserve_range(
+			ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM),
+			bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT,
+			AMDGPU_GPU_PAGE_SIZE);
 
+		memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps));
+		data->count++;
+		data->space_left--;
+	}
 out:
 	mutex_unlock(&con->recovery_lock);
 
@@ -1668,7 +1664,7 @@ out:
  * write error record array to eeprom, the function should be
  * protected by recovery_lock
  */
-static int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev)
+int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_err_handler_data *data;
@@ -1730,6 +1726,20 @@ out:
 	return ret;
 }
 
+static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con,
+				uint64_t addr)
+{
+	struct ras_err_handler_data *data = con->eh_data;
+	int i;
+
+	addr >>= AMDGPU_GPU_PAGE_SHIFT;
+	for (i = 0; i < data->count; i++)
+		if (addr == data->bps[i].retired_page)
+			return true;
+
+	return false;
+}
+
 /*
  * check if an address belongs to bad page
  *
@@ -1739,26 +1749,13 @@ static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev,
 				uint64_t addr)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-	struct ras_err_handler_data *data;
-	int i;
 	bool ret = false;
 
 	if (!con || !con->eh_data)
 		return ret;
 
 	mutex_lock(&con->recovery_lock);
-	data = con->eh_data;
-	if (!data)
-		goto out;
-
-	addr >>= AMDGPU_GPU_PAGE_SHIFT;
-	for (i = 0; i < data->count; i++)
-		if (addr == data->bps[i].retired_page) {
-			ret = true;
-			goto out;
-		}
-
-out:
+	ret = amdgpu_ras_check_bad_page_unlock(con, addr);
 	mutex_unlock(&con->recovery_lock);
 	return ret;
 }
@@ -1804,80 +1801,6 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev,
 	}
 }
 
-/* called in gpu recovery/init */
-int amdgpu_ras_reserve_bad_pages(struct amdgpu_device *adev)
-{
-	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-	struct ras_err_handler_data *data;
-	uint64_t bp;
-	struct amdgpu_bo *bo = NULL;
-	int i, ret = 0;
-
-	/* Not reserve bad page when amdgpu_bad_page_threshold == 0. */
-	if (!con || !con->eh_data || (amdgpu_bad_page_threshold == 0))
-		return 0;
-
-	mutex_lock(&con->recovery_lock);
-	data = con->eh_data;
-	if (!data)
-		goto out;
-	/* reserve vram at driver post stage. */
-	for (i = data->last_reserved; i < data->count; i++) {
-		bp = data->bps[i].retired_page;
-
-		/* There are two cases of reserve error should be ignored:
-		 * 1) a ras bad page has been allocated (used by someone);
-		 * 2) a ras bad page has been reserved (duplicate error injection
-		 *    for one page);
-		 */
-		if (amdgpu_bo_create_kernel_at(adev, bp << AMDGPU_GPU_PAGE_SHIFT,
-					       AMDGPU_GPU_PAGE_SIZE,
-					       AMDGPU_GEM_DOMAIN_VRAM,
-					       &bo, NULL))
-			dev_warn(adev->dev, "RAS WARN: reserve vram for "
-					"retired page %llx fail\n", bp);
-
-		data->bps_bo[i] = bo;
-		data->last_reserved = i + 1;
-		bo = NULL;
-	}
-
-	/* continue to save bad pages to eeprom even reesrve_vram fails */
-	ret = amdgpu_ras_save_bad_pages(adev);
-out:
-	mutex_unlock(&con->recovery_lock);
-	return ret;
-}
-
-/* called when driver unload */
-static int amdgpu_ras_release_bad_pages(struct amdgpu_device *adev)
-{
-	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-	struct ras_err_handler_data *data;
-	struct amdgpu_bo *bo;
-	int i;
-
-	if (!con || !con->eh_data)
-		return 0;
-
-	mutex_lock(&con->recovery_lock);
-	data = con->eh_data;
-	if (!data)
-		goto out;
-
-	for (i = data->last_reserved - 1; i >= 0; i--) {
-		bo = data->bps_bo[i];
-
-		amdgpu_bo_free_kernel(&bo, NULL, NULL);
-
-		data->bps_bo[i] = bo;
-		data->last_reserved = i;
-	}
-out:
-	mutex_unlock(&con->recovery_lock);
-	return 0;
-}
-
 int amdgpu_ras_recovery_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
@@ -1917,18 +1840,12 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev)
 		ret = amdgpu_ras_load_bad_pages(adev);
 		if (ret)
 			goto free;
-		ret = amdgpu_ras_reserve_bad_pages(adev);
-		if (ret)
-			goto release;
 	}
 
 	return 0;
 
-release:
-	amdgpu_ras_release_bad_pages(adev);
 free:
 	kfree((*data)->bps);
-	kfree((*data)->bps_bo);
 	kfree(*data);
 	con->eh_data = NULL;
 out:
@@ -1956,12 +1873,10 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev)
 		return 0;
 
 	cancel_work_sync(&con->recovery_work);
-	amdgpu_ras_release_bad_pages(adev);
 
 	mutex_lock(&con->recovery_lock);
 	con->eh_data = NULL;
 	kfree(data->bps);
-	kfree(data->bps_bo);
 	kfree(data);
 	mutex_unlock(&con->recovery_lock);
 
@@ -2156,7 +2071,7 @@ void amdgpu_ras_late_fini(struct amdgpu_device *adev,
 
 	amdgpu_ras_sysfs_remove(adev, ras_block);
 	if (ih_info->cb)
-                amdgpu_ras_interrupt_remove_handler(adev, ih_info);
+		amdgpu_ras_interrupt_remove_handler(adev, ih_info);
 	amdgpu_ras_feature_enable(adev, ras_block, 0);
 }