Re: [PATCH v3 6/6] drm/xe/xe_ras: Track offlined pages by firmware to avoid duplicates

["Ghimiray, Himal Prasad" <himal.prasad.ghimiray@intel.com>]

On 28-09-2026 11:48, Riana Tauro wrote:
> A memory scrubber can report multiple errors at the same address. Track
> pages already successfully offlined by the firmware so that subsequent
> reports for the same address are removed from the firmware queue instead
> of issuing redundant page offline requests.
> 
> Signed-off-by: Riana Tauro <riana.tauro@intel.com>
> ---
> v2: use xe page shift (Sashiko)
> ---
>   drivers/gpu/drm/xe/xe_device.c    |  4 ++-
>   drivers/gpu/drm/xe/xe_ras.c       | 44 ++++++++++++++++++++++++++++---
>   drivers/gpu/drm/xe/xe_ras.h       |  2 +-
>   drivers/gpu/drm/xe/xe_ras_types.h |  3 +++
>   4 files changed, 47 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index 205cb4e7f9e8..1ec7c74cbd58 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -957,7 +957,9 @@ int xe_device_probe(struct xe_device *xe)
>   	if (err)
>   		return err;
>   
> -	xe_ras_init(xe);
> +	err = xe_ras_init(xe);
> +	if (err)
> +		return err;
>   
>   	/*
>   	 * Now that GT is initialized (TTM in particular),
> diff --git a/drivers/gpu/drm/xe/xe_ras.c b/drivers/gpu/drm/xe/xe_ras.c
> index cfc7f07e9ced..7dee1c804eeb 100644
> --- a/drivers/gpu/drm/xe/xe_ras.c
> +++ b/drivers/gpu/drm/xe/xe_ras.c
> @@ -20,6 +20,9 @@
>   #include "xe_sysctrl_mailbox_types.h"
>   #include "xe_ttm_vram_mgr.h"
>   
> +/* Any non-null entry marks the page as offlined by firmware */
> +#define XE_RAS_PAGE_OFFLINED		xa_mk_value(1)
> +
>   #define CORE_COMPUTE_UNCORR_TYPE	GENMASK(26, 25)
>   /*
>    * Uncorrectable error type for core compute errors.
> @@ -210,6 +213,7 @@ static int send_page_offline_cmd(struct xe_device *xe, u64 page_address,
>   	struct xe_sysctrl_mailbox_command command = {0};
>   	struct xe_ras_page_offline_request request = {0};
>   	struct xe_ras_page_offline_response response = {0};
> +	struct xe_ras_state *state = &xe->ras.state;
>   	size_t rlen;
>   	int ret;
>   
> @@ -249,11 +253,16 @@ static int send_page_offline_cmd(struct xe_device *xe, u64 page_address,
>   		xe_log_err(xe, SYSCTRL, ret, "page offline command failed with status %u\n",
>   			   response.status);
>   
> +	if (action == XE_RAS_PAGE_ACTION_OFFLINE)
> +		xa_store(&state->offlined_pages, page_address >> XE_PTE_SHIFT,
> +			 XE_RAS_PAGE_OFFLINED, GFP_KERNEL);
> +
>   	return ret;
>   }
>   
>   static int handle_page_offline(struct xe_device *xe, u64 page_address, bool send_cmd)
>   {
> +	struct xe_ras_state *state = &xe->ras.state;
>   	enum xe_ras_page_action action;
>   	u64 addr;
>   	int ret = 0;
> @@ -297,7 +306,11 @@ static int handle_page_offline(struct xe_device *xe, u64 page_address, bool send
>   			   page_address);
>   		return ret;
>   	case -EEXIST:
> -		action = XE_RAS_PAGE_ACTION_OFFLINE;
> +		if (xa_load(&state->offlined_pages, page_address >> XE_PTE_SHIFT))
> +			action = XE_RAS_PAGE_ACTION_REMOVE;
> +		else
> +			action = XE_RAS_PAGE_ACTION_OFFLINE;
> +
>   		xe_log_err(xe, DEVICE_MEMORY, ret,
>   			   "Double-bit ECC error detected at physical address 0x%llx, page soft-offlined\n",
>   			   page_address);
> @@ -341,6 +354,8 @@ static void get_offline_pages(struct xe_device *xe, u32 cmd, void *req, size_t r
>   {
>   	struct xe_sysctrl_mailbox_command command = {0};
>   	struct xe_ras_offline_list_request *list_req;
> +	struct xe_ras_state *state = &xe->ras.state;
> +	unsigned long index;
>   	u32 total_pages = 0, count = 0;
>   	ssize_t rlen;
>   	int ret, i;
> @@ -370,8 +385,15 @@ static void get_offline_pages(struct xe_device *xe, u32 cmd, void *req, size_t r
>   			return;
>   		}
>   
> -		for (i = 0; i < common->pages_returned && i < XE_RAS_NUM_PAGES; i++)
> +		for (i = 0; i < common->pages_returned && i < XE_RAS_NUM_PAGES; i++) {
>   			handle_page_offline(xe, common->page_addresses[i], offline);
> +			/* The pages are already offlined by firmware */
> +			if (!offline) {
> +				index = common->page_addresses[i] >> XE_PTE_SHIFT;
> +				xa_store(&state->offlined_pages, index,
> +					 XE_RAS_PAGE_OFFLINED, GFP_KERNEL);
> +			}
> +		}
>   
>   		count += common->pages_returned;
>   		if (!common->pages_returned)
> @@ -1098,6 +1120,13 @@ static const struct attribute_group gpu_health_group = {
>   	.attrs = gpu_health_attrs,
>   };
>   
> +static void ras_fini(void *arg)
> +{
> +	struct xe_device *xe = arg;
> +
> +	xa_destroy(&xe->ras.state.offlined_pages);
> +}
> +
>   /**
>    * xe_ras_get_disable_page_offline - Get the page offline user policy
>    * @xe: xe device instance
> @@ -1125,15 +1154,18 @@ u32 xe_ras_get_max_pages(struct xe_device *xe)
>    * @xe: xe device instance
>    *
>    * Initialize Xe RAS
> + *
> + * Return: 0 on success, negative error code otherwise
>    */
> -void xe_ras_init(struct xe_device *xe)
> +int xe_ras_init(struct xe_device *xe)
>   {
> +	struct xe_ras_state *state = &xe->ras.state;
>   	int ret;
>   
>   	xe_drm_ras_init(xe);
>   
>   	if (!xe->info.has_sysctrl)
> -		return;
> +		return 0;
>   
>   	/*
>   	 * TODO: Replace platform check with xe->info.has_disable_vram_page_offline
> @@ -1145,10 +1177,14 @@ void xe_ras_init(struct xe_device *xe)
>   	if (IS_ENABLED(CONFIG_PCIEAER))
>   		ras_usp_aer_init(xe);
>   
> +	xa_init(&state->offlined_pages);
> +
>   	get_queued_pages(xe);
>   	get_offlined_list(xe);
>   
>   	ret = devm_device_add_group(xe->drm.dev, &gpu_health_group);
>   	if (ret)
>   		xe_err(xe, "Failed to create GPU health sysfs, err=%d\n", ret);
> +
> +	return devm_add_action_or_reset(xe->drm.dev, ras_fini, xe);
>   }
> diff --git a/drivers/gpu/drm/xe/xe_ras.h b/drivers/gpu/drm/xe/xe_ras.h
> index 561c656e2ad3..e41a52869085 100644
> --- a/drivers/gpu/drm/xe/xe_ras.h
> +++ b/drivers/gpu/drm/xe/xe_ras.h
> @@ -19,7 +19,7 @@ int xe_ras_get_counter(struct xe_device *xe, u8 severity, u8 component, u32 *val
>   int xe_ras_clear_counter(struct xe_device *xe, u8 severity, u8 component);
>   int xe_ras_get_threshold(struct xe_device *xe, u8 severity, u8 component, u32 *threshold);
>   int xe_ras_set_threshold(struct xe_device *xe, u8 severity, u8 component, u32 threshold);
> -void xe_ras_init(struct xe_device *xe);
> +int xe_ras_init(struct xe_device *xe);
>   enum xe_ras_recovery_action xe_ras_process_errors(struct xe_device *xe);
>   bool xe_ras_get_disable_page_offline(struct xe_device *xe);
>   u32 xe_ras_get_max_pages(struct xe_device *xe);
> diff --git a/drivers/gpu/drm/xe/xe_ras_types.h b/drivers/gpu/drm/xe/xe_ras_types.h
> index e09b50a6f77b..50d70e2e2fd3 100644
> --- a/drivers/gpu/drm/xe/xe_ras_types.h
> +++ b/drivers/gpu/drm/xe/xe_ras_types.h
> @@ -7,6 +7,7 @@
>   #define _XE_RAS_TYPES_H_
>   
>   #include <linux/types.h>
> +#include <linux/xarray.h>
>   
>   #define XE_RAS_NUM_COUNTERS			16
>   #define XE_RAS_NUM_ERROR_ARR			3
> @@ -417,5 +418,7 @@ struct xe_ras_state {
>   	bool disable_page_offline;
>   	/** @max_pages: Total number of pages that can be stored by firmware */
>   	u32 max_pages;
> +	/** @offlined_pages: XArray of pages offlined by firmware */
> +	struct xarray offlined_pages;

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>

>   };
>   #endif