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);
case AMDGPU_RAS_RETIRE_PAGE_FAULT:
default:
return "F";
- };
+ }
}
/**
return scnprintf(buf, PAGE_SIZE, "feature mask: 0x%x\n", con->features);
}
-static void amdgpu_ras_sysfs_add_bad_page_node(struct amdgpu_device *adev)
-{
- struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
- struct attribute_group group;
- struct bin_attribute *bin_attrs[] = {
- &con->badpages_attr,
- NULL,
- };
-
- con->badpages_attr = (struct bin_attribute) {
- .attr = {
- .name = "gpu_vram_bad_pages",
- .mode = S_IRUGO,
- },
- .size = 0,
- .private = NULL,
- .read = amdgpu_ras_sysfs_badpages_read,
- };
-
- group.name = RAS_FS_NAME;
- group.bin_attrs = bin_attrs;
-
- sysfs_bin_attr_init(bin_attrs[0]);
-
- sysfs_update_group(&adev->dev->kobj, &group);
-}
-
-static int amdgpu_ras_sysfs_create_feature_node(struct amdgpu_device *adev)
-{
- struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
- struct attribute *attrs[] = {
- &con->features_attr.attr,
- NULL
- };
- struct attribute_group group = {
- .name = RAS_FS_NAME,
- .attrs = attrs,
- };
-
- con->features_attr = (struct device_attribute) {
- .attr = {
- .name = "features",
- .mode = S_IRUGO,
- },
- .show = amdgpu_ras_sysfs_features_read,
- };
-
- sysfs_attr_init(attrs[0]);
-
- return sysfs_create_group(&adev->dev->kobj, &group);
-}
-
static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device *adev)
{
struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
/* debugfs end */
/* ras fs */
-
+static BIN_ATTR(gpu_vram_bad_pages, S_IRUGO,
+ amdgpu_ras_sysfs_badpages_read, NULL, 0);
+static DEVICE_ATTR(features, S_IRUGO,
+ amdgpu_ras_sysfs_features_read, NULL);
static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
{
- amdgpu_ras_sysfs_create_feature_node(adev);
+ struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+ struct attribute_group group = {
+ .name = RAS_FS_NAME,
+ };
+ struct attribute *attrs[] = {
+ &con->features_attr.attr,
+ NULL
+ };
+ struct bin_attribute *bin_attrs[] = {
+ NULL,
+ NULL,
+ };
+ int r;
- if (amdgpu_bad_page_threshold != 0)
- amdgpu_ras_sysfs_add_bad_page_node(adev);
+ /* add features entry */
+ con->features_attr = dev_attr_features;
+ group.attrs = attrs;
+ sysfs_attr_init(attrs[0]);
+
+ if (amdgpu_bad_page_threshold != 0) {
+ /* add bad_page_features entry */
+ bin_attr_gpu_vram_bad_pages.private = NULL;
+ con->badpages_attr = bin_attr_gpu_vram_bad_pages;
+ bin_attrs[0] = &con->badpages_attr;
+ group.bin_attrs = bin_attrs;
+ sysfs_bin_attr_init(bin_attrs[0]);
+ }
+
+ r = sysfs_create_group(&adev->dev->kobj, &group);
+ if (r)
+ dev_err(adev->dev, "Failed to create RAS sysfs group!");
return 0;
}
}
}
+/* Parse RdRspStatus and WrRspStatus */
+void amdgpu_ras_error_status_query(struct amdgpu_device *adev,
+ struct ras_query_if *info)
+{
+ /*
+ * Only two block need to query read/write
+ * RspStatus at current state
+ */
+ switch (info->head.block) {
+ case AMDGPU_RAS_BLOCK__GFX:
+ if (adev->gfx.funcs->query_ras_error_status)
+ adev->gfx.funcs->query_ras_error_status(adev);
+ break;
+ case AMDGPU_RAS_BLOCK__MMHUB:
+ if (adev->mmhub.funcs->query_ras_error_status)
+ adev->mmhub.funcs->query_ras_error_status(adev);
+ break;
+ default:
+ break;
+ }
+}
+
+static void amdgpu_ras_query_err_status(struct amdgpu_device *adev)
+{
+ struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+ struct ras_manager *obj;
+
+ if (!con)
+ return;
+
+ list_for_each_entry(obj, &con->head, node) {
+ struct ras_query_if info = {
+ .head = obj->head,
+ };
+
+ amdgpu_ras_error_status_query(adev, &info);
+ }
+}
+
/* recovery begin */
/* return 0 on success.
.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;
}
}
list_for_each_entry(remote_adev,
- device_list_handle, gmc.xgmi.head)
+ device_list_handle, gmc.xgmi.head) {
+ amdgpu_ras_query_err_status(remote_adev);
amdgpu_ras_log_on_err_counter(remote_adev);
+ }
amdgpu_put_xgmi_hive(hive);
}
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;
}
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;
}
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;
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);
* 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;
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
*
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;
}
}
}
-/* 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);
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:
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);
projects / linux.git / blobdiff