Re: [PATCH v5 12/13] drm/xe/xe_ras: Add support to retrieve info queue data

[Raag Jadav <raag.jadav@intel.com>]

On Fri, Oct 23, 2026 at 10:04:34AM +0530, Badal Nilawar wrote:
> Retrieve the RAS info queue data, in multiple chunks, and assemble it into
> a flat buffer.
> 
> Assisted-by: LLM
> Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
> ---
> v5:
>  - Marked unused fields reserved
>  - Handle response.status
> v4:
>  - Simplify info queue handling and validation (Raag)
>  - Clean up info queue naming and definitions (Raag)
>  - Validate each chunk against the requested offset
> ---
>  drivers/gpu/drm/xe/xe_ras.c                   | 130 ++++++++++++++++++
>  drivers/gpu/drm/xe/xe_ras.h                   |   3 +
>  drivers/gpu/drm/xe/xe_ras_types.h             |  91 +++++++++++-
>  drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h |   2 +
>  4 files changed, 224 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_ras.c b/drivers/gpu/drm/xe/xe_ras.c
> index b3dc4d8ad190..12ac9a29555a 100644
> --- a/drivers/gpu/drm/xe/xe_ras.c
> +++ b/drivers/gpu/drm/xe/xe_ras.c
> @@ -1114,6 +1114,136 @@ int xe_ras_set_threshold(struct xe_device *xe, u8 severity, u8 component, u32 th
>  	return 0;
>  }
>  
> +static int get_info_queue_data(struct xe_device *xe,

Let's not add needless verbiage, get_info_queue() is sufficient.

> +			       struct xe_ras_get_info_queue_request *req,

'request'

> +			       struct xe_ras_get_info_queue_response *out)

'response'

> +{
> +	struct xe_ras_get_info_queue_response response = {0};

Why not reuse caller response?

> +	struct xe_sysctrl_mailbox_command command = {0};
> +	const struct xe_ras_info_queue_header *hdr;
> +	size_t rlen;
> +	int ret;
> +
> +	xe_sysctrl_create_command(&command, XE_SYSCTRL_GROUP_GFSP, XE_SYSCTRL_CMD_GET_INFO_QUEUE,
> +				  req, sizeof(*req), &response, sizeof(response));
> +
> +	ret = xe_sysctrl_send_command(&xe->sc, &command, &rlen);
> +	if (ret) {
> +		xe_err(xe, "sysctrl: failed to get info queue data %d\n", ret);

Redundant 'data'.

> +		return ret;
> +	}
> +
> +	if (rlen != sizeof(response)) {
> +		xe_err(xe, "sysctrl: unexpected get info queue data response length %zu (expected %zu)\n",

Ditto.

> +		       rlen, sizeof(response));
> +		return -EIO;
> +	}
> +
> +	ret = ras_status_to_errno(response.status);
> +	if (ret) {
> +		xe_err(xe, "sysctrl: get info queue command failed with status %#x\n",
> +		       response.status);
> +		return ret;
> +	}
> +
> +	hdr = &response.queue_response.header;
> +	xe_dbg(xe, "[RAS] info queue: total_size=%u chunk_offset=%u chunk_size=%u flags=0x%x num_headers=%u\n",
> +	       hdr->total_size, hdr->chunk_offset, hdr->chunk_size,
> +	       hdr->flags, hdr->num_headers);
> +
> +	*out = response;
> +	return 0;
> +}
> +
> +static bool chunk_is_valid(const struct xe_ras_info_queue_header *hdr,
> +			   u32 expected_offset, u32 dest_size)
> +{
> +	u32 end;
> +
> +	return hdr->chunk_offset == expected_offset &&
> +	       hdr->chunk_size && hdr->chunk_size <= XE_RAS_INFO_QUEUE_MAX_CHUNK_SIZE &&
> +	       !check_add_overflow(hdr->chunk_offset, hdr->chunk_size, &end) &&
> +	       end <= XE_RAS_INFO_QUEUE_MAX_TOTAL_SIZE && end <= dest_size;
> +}
> +
> +/**
> + * xe_ras_drain_info_queue_raw() - Drain the full RAS info queue into a flat buffer
> + * @xe: xe device
> + * @counter_resp: counter response carrying the first embedded chunk and the
> + *                counter identifier used as the source context for subsequent

Are we still relying on source context?

> + *                GET_INFO_QUEUE fetches
> + * @dest_buf: destination buffer supplied by the caller
> + * @dest_size: size of @dest_buf in bytes; also caps the total amount of data
> + *             assembled from the info queue
> + *
> + * Copies the first chunk already embedded in @counter_resp, then loops
> + * issuing GET_INFO_QUEUE to fetch any remaining chunks until the queue
> + * signals no more data or a transport/bounds error is encountered.  On
> + * transport or bounds errors the function stops and returns whatever has
> + * been assembled so far.
> + *
> + * Returns: Number of valid bytes written into @dest_buf.  Zero if @dest_buf is
> + *          NULL or @dest_size is 0.
> + */
> +u32 xe_ras_drain_info_queue_raw(struct xe_device *xe,

Redundant 'raw'.

> +				const struct xe_ras_get_counter_response *counter_resp,
> +				u8 *dest_buf, u32 dest_size)
> +{
> +	const struct xe_ras_info_queue_header *first_hdr;
> +	u32 flags, offset = 0;
> +	int ret;
> +
> +	xe_assert(xe, counter_resp->has_info_queue);
> +
> +	if (!dest_buf || !dest_size)

Do we actually have such a user, or is it possible due to some firmware
quirk? If no, this can be xe_assert().

> +		return 0;
> +
> +	first_hdr = &counter_resp->first.header;
> +
> +	if (!chunk_is_valid(first_hdr, offset, dest_size)) {
> +		xe_warn(xe, "[RAS]: invalid first info queue chunk (offset=%u size=%u, expected offset=%u)\n",
> +			first_hdr->chunk_offset, first_hdr->chunk_size, offset);
> +		return 0;
> +	}
> +
> +	/* Copy first chunk already embedded in the counter response */
> +	memcpy(dest_buf, counter_resp->first.data, first_hdr->chunk_size);
> +	offset = first_hdr->chunk_size;
> +	flags = first_hdr->flags;

The code upto this point can be refactored into something like
get_first_chunk().

> +	/* Fetch any remaining chunks */
> +	while ((flags & XE_RAS_INFO_QUEUE_FLAG_MORE_DATA) && offset < dest_size) {
> +		struct xe_ras_get_info_queue_request req = {};
> +		struct xe_ras_get_info_queue_response resp = {};
> +		struct xe_ras_info_queue_header *hdr;
> +
> +		req.queue_request.offset = offset;
> +		req.queue_request.size = min(XE_RAS_INFO_QUEUE_MAX_CHUNK_SIZE, dest_size - offset);
> +		req.queue_request.session_id = counter_resp->counter;
> +
> +		ret = get_info_queue_data(xe, &req, &resp);
> +		if (ret) {
> +			xe_err(xe, "[RAS]: info queue drain aborted: fetch at offset=%u, ret = %d\n",
> +			       offset, ret);
> +			return offset;
> +		}
> +
> +		hdr = &resp.queue_response.header;
> +
> +		if (!chunk_is_valid(hdr, offset, dest_size)) {
> +			xe_warn(xe, "[RAS]: invalid info queue chunk (offset=%u size=%u, expected offset=%u)\n",
> +				hdr->chunk_offset, hdr->chunk_size, offset);
> +			return offset;
> +		}
> +
> +		memcpy(dest_buf + offset, resp.queue_response.data, hdr->chunk_size);
> +		offset += hdr->chunk_size;
> +		flags = hdr->flags;

Ditto, can be refactored into get_remaining_chunks().

> +	}
> +
> +	return min(offset, first_hdr->total_size);
> +}
> +
>  static ssize_t gpu_health_show(struct device *dev, struct device_attribute *attr, char *buf)
>  {
>  	struct xe_ras_get_health_response response = {0};
> diff --git a/drivers/gpu/drm/xe/xe_ras.h b/drivers/gpu/drm/xe/xe_ras.h
> index c736d5a01c47..ccb79f2c2844 100644
> --- a/drivers/gpu/drm/xe/xe_ras.h
> +++ b/drivers/gpu/drm/xe/xe_ras.h
> @@ -25,5 +25,8 @@ bool xe_ras_get_disable_vram_page_offline(struct xe_device *xe);
>  u32 xe_ras_get_max_pages(struct xe_device *xe);
>  int xe_ras_get_counter_response(struct xe_device *xe, const struct xe_ras_error_class *counter,
>  				struct xe_ras_get_counter_response *out);
> +u32 xe_ras_drain_info_queue_raw(struct xe_device *xe,
> +				const struct xe_ras_get_counter_response *counter_resp,
> +				u8 *dest_buf, u32 dest_size);
>  
>  #endif
> diff --git a/drivers/gpu/drm/xe/xe_ras_types.h b/drivers/gpu/drm/xe/xe_ras_types.h
> index a86f9d2b8c92..67d57182aac9 100644
> --- a/drivers/gpu/drm/xe/xe_ras_types.h
> +++ b/drivers/gpu/drm/xe/xe_ras_types.h
> @@ -19,6 +19,10 @@
>  #define XE_RAS_MEMORY_POISON			BIT(2)
>  #define XE_RAS_MEMORY_DATA_PARITY		BIT(5)
>  
> +#define XE_RAS_INFO_QUEUE_MAX_CHUNK_SIZE	200
> +#define XE_RAS_INFO_QUEUE_MAX_TOTAL_SIZE	5120
> +#define XE_RAS_INFO_QUEUE_FLAG_MORE_DATA	0x02
> +
>  /**
>   * enum xe_ras_page_action - Page offline actions for page offline request
>   *
> @@ -110,6 +114,83 @@ struct xe_ras_threshold_crossed {
>  	struct xe_ras_error_class counters[XE_RAS_NUM_COUNTERS];
>  } __packed;
>  
> +/**
> + * struct xe_ras_info_queue_header - Metadata for large info queue data transfers
> + *
> + * Provides chunk metadata for commands that support extended info queue
> + * functionality. Used when the total data exceeds a single mailbox response.
> + */
> +struct xe_ras_info_queue_header {
> +	/** @total_size: Total size of the complete info queue data in bytes */
> +	u32 total_size;
> +	/** @chunk_offset: Offset of this chunk within the total data in bytes */
> +	u32 chunk_offset;
> +	/** @chunk_size: Size of the data in this chunk in bytes */
> +	u32 chunk_size;
> +	/** @reserved0: Reserved for future use */
> +	u32 reserved0;
> +	/** @flags: Info queue control flags (XE_RAS_INFO_QUEUE_FLAG_*) */
> +	u32 flags:8;
> +	/** @reserved1: Reserved for future use */
> +	u32 reserved1:4;
> +	/** @num_headers: Number of detailed counter headers at start of the data */
> +	u32 num_headers:5;
> +	/** @reserved2: Reserved for future use */
> +	u32 reserved2:15;
> +	/** @reserved3: Reserved for future use */
> +	u32 reserved3;
> +} __packed;
> +
> +/**
> + * struct xe_ras_info_queue_data_request - Request for a specific chunk of info queue data
> + *
> + * Allows the driver to request continuation of large info queue transfers
> + * by specifying an offset and size within the full data set.
> + */
> +struct xe_ras_info_queue_data_request {
> +	/** @offset: Byte offset of the requested data chunk */
> +	u32 offset;
> +	/** @size: Maximum size of the requested chunk in bytes */
> +	u32 size;
> +	/** @session_id: Session ID to correlate multi-chunk transfers */
> +	struct xe_ras_error_class session_id;
> +	/** @reserved: Reserved for future use */
> +	u32 reserved;
> +} __packed;
> +
> +/**
> + * struct xe_ras_info_queue_data_response - Generic response for commands with info queues
> + *
> + * Standard response format for any command that returns an info queue
> + * payload. May be embedded in a command-specific response structure.
> + */
> +struct xe_ras_info_queue_data_response {
> +	/** @header: Info queue metadata for this chunk */
> +	struct xe_ras_info_queue_header header;
> +	/** @data: Info queue data for this chunk */
> +	u8 data[XE_RAS_INFO_QUEUE_MAX_CHUNK_SIZE];
> +} __packed;
> +
> +/**
> + * struct xe_ras_get_info_queue_request - Request for XE_SYSCTRL_CMD_GET_INFO_QUEUE
> + */
> +struct xe_ras_get_info_queue_request {
> +	/** @queue_request: Info queue request parameters */
> +	struct xe_ras_info_queue_data_request queue_request;
> +	/** @reserved: Reserved for future use */
> +	u32 reserved[16];
> +} __packed;
> +
> +/**
> + * struct xe_ras_get_info_queue_response - Response for XE_SYSCTRL_CMD_GET_INFO_QUEUE
> + */
> +struct xe_ras_get_info_queue_response {
> +	/** @status: Status of the retrieval operation */
> +	u32 status;

Curious, why does get_info_queue() require a status but get_counter()
does not? The spec doesn't explain this.

Raag

> +	/** @queue_response: Info queue data chunk */
> +	struct xe_ras_info_queue_data_response queue_response;
> +} __packed;
> +
>  /**
>   * struct xe_ras_get_counter_request - Request structure for get counter
>   */
> @@ -132,8 +213,14 @@ struct xe_ras_get_counter_response {
>  	u64 timestamp;
>  	/** @threshold: Threshold value for the counter */
>  	u32 threshold;
> -	/** @reserved: Reserved  */
> -	u32 reserved[57];
> +	/** @reserved: Reserved for future use */
> +	u32 reserved:9;
> +	/** @has_info_queue: Set if info queue is available */
> +	u32 has_info_queue:1;
> +	/** @reserved1: Reserved for future use */
> +	u32 reserved1:22;
> +	/** @first: First info queue chunk, if @has_info_queue is set */
> +	struct xe_ras_info_queue_data_response first;
>  } __packed;
>  
>  /**
> diff --git a/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h b/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> index 1f3f782d770e..9188457ef7ae 100644
> --- a/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> +++ b/drivers/gpu/drm/xe/xe_sysctrl_mailbox_types.h
> @@ -37,6 +37,7 @@ enum xe_sysctrl_group {
>   * @XE_SYSCTRL_CMD_GET_OFFLINE_QUEUE: Retrieve list of offlined queued pages from firmware
>   * @XE_SYSCTRL_CMD_GET_HEALTH: Retrieve gpu health
>   * @XE_SYSCTRL_CMD_SET_HEALTH: Set gpu health
> + * @XE_SYSCTRL_CMD_GET_INFO_QUEUE: Retrieve a chunk of info queue data
>   */
>  enum xe_sysctrl_gfsp_cmd {
>  	XE_SYSCTRL_CMD_GET_SOC_ERROR		= 0x01,
> @@ -50,6 +51,7 @@ enum xe_sysctrl_gfsp_cmd {
>  	XE_SYSCTRL_CMD_GET_OFFLINE_QUEUE	= 0x0A,
>  	XE_SYSCTRL_CMD_GET_HEALTH		= 0x0B,
>  	XE_SYSCTRL_CMD_SET_HEALTH		= 0x0C,
> +	XE_SYSCTRL_CMD_GET_INFO_QUEUE		= 0x0D,
>  };
>  
>  /**
> -- 
> 2.54.0
>