Commit 97a78e3d authored by Dani Liberman's avatar Dani Liberman Committed by Oded Gabbay
Browse files

habanalabs: rename error info structure 



As a preparation for adding more errors to it,
change to more suitable name.

Signed-off-by: default avatarDani Liberman <dliberman@habana.ai>
Reviewed-by: default avatarOded Gabbay <ogabbay@kernel.org>
Signed-off-by: default avatarOded Gabbay <ogabbay@kernel.org>
parent 04d53cd2

drivers/misc/habanalabs/common/command_submission.c

+3 −3
Original line number Diff line number Diff line
@@ -826,10 +826,10 @@ static void cs_timedout(struct work_struct *work) 
	}



	/* Save only the first CS timeout parameters */

	rc = atomic_cmpxchg(&hdev->last_error.cs_timeout.write_enable, 1, 0);

	rc = atomic_cmpxchg(&hdev->captured_err_info.cs_timeout.write_enable, 1, 0);

	if (rc) {

		hdev->last_error.cs_timeout.timestamp = ktime_get();

		hdev->last_error.cs_timeout.seq = cs->sequence;

		hdev->captured_err_info.cs_timeout.timestamp = ktime_get();

		hdev->captured_err_info.cs_timeout.seq = cs->sequence;



		event_mask = device_reset ? (HL_NOTIFIER_EVENT_CS_TIMEOUT |

				HL_NOTIFIER_EVENT_DEVICE_RESET) : HL_NOTIFIER_EVENT_CS_TIMEOUT;

drivers/misc/habanalabs/common/habanalabs.h

+6 −6
Original line number Diff line number Diff line
@@ -2981,12 +2981,12 @@ struct undefined_opcode_info { 
};



/**

 * struct last_error_session_info - info about last session errors occurred.

 * @cs_timeout: CS timeout error last information.

 * @razwi: razwi last information.

 * struct hl_error_info - holds information collected during an error.

 * @cs_timeout: CS timeout error information.

 * @razwi: razwi information.

 * @undef_opcode: undefined opcode information

 */

struct last_error_session_info {

struct hl_error_info {

	struct cs_timeout_info		cs_timeout;

	struct razwi_info		razwi;

	struct undefined_opcode_info	undef_opcode;
@@ -3111,7 +3111,7 @@ struct hl_reset_info { 
 * @state_dump_specs: constants and dictionaries needed to dump system state.

 * @multi_cs_completion: array of multi-CS completion.

 * @clk_throttling: holds information about current/previous clock throttling events

 * @last_error: holds information about last session in which CS timeout or razwi error occurred.

 * @captured_err_info: holds information about errors.

 * @reset_info: holds current device reset information.

 * @stream_master_qid_arr: pointer to array with QIDs of master streams.

 * @fw_major_version: major version of current loaded preboot.
@@ -3286,7 +3286,7 @@ struct hl_device { 
	struct multi_cs_completion	multi_cs_completion[

							MULTI_CS_MAX_USER_CTX];

	struct hl_clk_throttle		clk_throttling;

	struct last_error_session_info	last_error;

	struct hl_error_info		captured_err_info;



	struct hl_reset_info		reset_info;

drivers/misc/habanalabs/common/habanalabs_drv.c

+3 −3
Original line number Diff line number Diff line
@@ -211,9 +211,9 @@ int hl_device_open(struct inode *inode, struct file *filp) 


	hl_debugfs_add_file(hpriv);



	atomic_set(&hdev->last_error.cs_timeout.write_enable, 1);

	atomic_set(&hdev->last_error.razwi.write_enable, 1);

	hdev->last_error.undef_opcode.write_enable = true;

	atomic_set(&hdev->captured_err_info.cs_timeout.write_enable, 1);

	atomic_set(&hdev->captured_err_info.razwi.write_enable, 1);

	hdev->captured_err_info.undef_opcode.write_enable = true;



	hdev->open_counter++;

	hdev->last_successful_open_jif = jiffies;

drivers/misc/habanalabs/common/habanalabs_ioctl.c

+15 −15
Original line number Diff line number Diff line
@@ -593,8 +593,8 @@ static int cs_timeout_info(struct hl_fpriv *hpriv, struct hl_info_args *args) 
	if ((!max_size) || (!out))

		return -EINVAL;



	info.seq = hdev->last_error.cs_timeout.seq;

	info.timestamp = ktime_to_ns(hdev->last_error.cs_timeout.timestamp);

	info.seq = hdev->captured_err_info.cs_timeout.seq;

	info.timestamp = ktime_to_ns(hdev->captured_err_info.cs_timeout.timestamp);



	return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;

}
@@ -609,12 +609,12 @@ static int razwi_info(struct hl_fpriv *hpriv, struct hl_info_args *args) 
	if ((!max_size) || (!out))

		return -EINVAL;



	info.timestamp = ktime_to_ns(hdev->last_error.razwi.timestamp);

	info.addr = hdev->last_error.razwi.addr;

	info.engine_id_1 = hdev->last_error.razwi.engine_id_1;

	info.engine_id_2 = hdev->last_error.razwi.engine_id_2;

	info.no_engine_id = hdev->last_error.razwi.non_engine_initiator;

	info.error_type = hdev->last_error.razwi.type;

	info.timestamp = ktime_to_ns(hdev->captured_err_info.razwi.timestamp);

	info.addr = hdev->captured_err_info.razwi.addr;

	info.engine_id_1 = hdev->captured_err_info.razwi.engine_id_1;

	info.engine_id_2 = hdev->captured_err_info.razwi.engine_id_2;

	info.no_engine_id = hdev->captured_err_info.razwi.non_engine_initiator;

	info.error_type = hdev->captured_err_info.razwi.type;



	return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;

}
@@ -629,13 +629,13 @@ static int undefined_opcode_info(struct hl_fpriv *hpriv, struct hl_info_args *ar 
	if ((!max_size) || (!out))

		return -EINVAL;



	info.timestamp = ktime_to_ns(hdev->last_error.undef_opcode.timestamp);

	info.engine_id = hdev->last_error.undef_opcode.engine_id;

	info.cq_addr = hdev->last_error.undef_opcode.cq_addr;

	info.cq_size = hdev->last_error.undef_opcode.cq_size;

	info.stream_id = hdev->last_error.undef_opcode.stream_id;

	info.cb_addr_streams_len = hdev->last_error.undef_opcode.cb_addr_streams_len;

	memcpy(info.cb_addr_streams, hdev->last_error.undef_opcode.cb_addr_streams,

	info.timestamp = ktime_to_ns(hdev->captured_err_info.undef_opcode.timestamp);

	info.engine_id = hdev->captured_err_info.undef_opcode.engine_id;

	info.cq_addr = hdev->captured_err_info.undef_opcode.cq_addr;

	info.cq_size = hdev->captured_err_info.undef_opcode.cq_size;

	info.stream_id = hdev->captured_err_info.undef_opcode.stream_id;

	info.cb_addr_streams_len = hdev->captured_err_info.undef_opcode.cb_addr_streams_len;

	memcpy(info.cb_addr_streams, hdev->captured_err_info.undef_opcode.cb_addr_streams,

			sizeof(info.cb_addr_streams));



	return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;

drivers/misc/habanalabs/gaudi/gaudi.c

+16 −15
Original line number Diff line number Diff line
@@ -6894,9 +6894,9 @@ static void gaudi_handle_sw_config_stream_data(struct hl_device *hdev, u32 strea 
							stream, cq_ptr, size);



	if (event_mask & HL_NOTIFIER_EVENT_UNDEFINED_OPCODE) {

		hdev->last_error.undef_opcode.cq_addr = cq_ptr;

		hdev->last_error.undef_opcode.cq_size = size;

		hdev->last_error.undef_opcode.stream_id = stream;

		hdev->captured_err_info.undef_opcode.cq_addr = cq_ptr;

		hdev->captured_err_info.undef_opcode.cq_size = size;

		hdev->captured_err_info.undef_opcode.stream_id = stream;

	}

}
@@ -6962,7 +6962,7 @@ static void gaudi_handle_last_pqes_on_err(struct hl_device *hdev, u32 qid_base, 
	}



	if (event_mask & HL_NOTIFIER_EVENT_UNDEFINED_OPCODE) {

		struct undefined_opcode_info *undef_opcode = &hdev->last_error.undef_opcode;

		struct undefined_opcode_info *undef_opcode = &hdev->captured_err_info.undef_opcode;

		u32 arr_idx = undef_opcode->cb_addr_streams_len;



		if (arr_idx == 0) {
@@ -7046,11 +7046,11 @@ static void gaudi_handle_qman_err_generic(struct hl_device *hdev, 
		}

		/* check for undefined opcode */

		if (glbl_sts_val & TPC0_QM_GLBL_STS1_CP_UNDEF_CMD_ERR_MASK &&

				hdev->last_error.undef_opcode.write_enable) {

			memset(&hdev->last_error.undef_opcode, 0,

						sizeof(hdev->last_error.undef_opcode));

				hdev->captured_err_info.undef_opcode.write_enable) {

			memset(&hdev->captured_err_info.undef_opcode, 0,

						sizeof(hdev->captured_err_info.undef_opcode));



			hdev->last_error.undef_opcode.write_enable = false;

			hdev->captured_err_info.undef_opcode.write_enable = false;

			*event_mask |= HL_NOTIFIER_EVENT_UNDEFINED_OPCODE;

		}
@@ -7332,18 +7332,19 @@ static void gaudi_print_irq_info(struct hl_device *hdev, u16 event_type, 
		gaudi_print_and_get_mmu_error_info(hdev, &razwi_addr, &razwi_type);



		/* In case it's the first razwi, save its parameters*/

		rc = atomic_cmpxchg(&hdev->last_error.razwi.write_enable, 1, 0);

		rc = atomic_cmpxchg(&hdev->captured_err_info.razwi.write_enable, 1, 0);

		if (rc) {

			hdev->last_error.razwi.timestamp = ktime_get();

			hdev->last_error.razwi.addr = razwi_addr;

			hdev->last_error.razwi.engine_id_1 = engine_id_1;

			hdev->last_error.razwi.engine_id_2 = engine_id_2;

			hdev->captured_err_info.razwi.timestamp = ktime_get();

			hdev->captured_err_info.razwi.addr = razwi_addr;

			hdev->captured_err_info.razwi.engine_id_1 = engine_id_1;

			hdev->captured_err_info.razwi.engine_id_2 = engine_id_2;

			/*

			 * If first engine id holds non valid value the razwi initiator

			 * does not have engine id

			 */

			hdev->last_error.razwi.non_engine_initiator = (engine_id_1 == U16_MAX);

			hdev->last_error.razwi.type = razwi_type;

			hdev->captured_err_info.razwi.non_engine_initiator =

									(engine_id_1 == U16_MAX);

			hdev->captured_err_info.razwi.type = razwi_type;



		}

	}