Commit 90de6805 authored by Tal Cohen's avatar Tal Cohen Committed by Greg Kroah-Hartman
Browse files

habanalabs: use separate structure info for each error collect data 



Create separate info structure for each error type.
The structures shall be used inside the large structure that contains
the last session error.
This is more scalable for adding more errors in the future.

Signed-off-by: default avatarTal Cohen <talcohen@habana.ai>
Reviewed-by: default avatarOded Gabbay <ogabbay@kernel.org>
Signed-off-by: default avatarOded Gabbay <ogabbay@kernel.org>
Signed-off-by: default avatarGreg Kroah-Hartman <gregkh@linuxfoundation.org>
parent f873a27f

drivers/misc/habanalabs/common/command_submission.c

+3 −4
Original line number Diff line number Diff line
@@ -735,11 +735,10 @@ static void cs_timedout(struct work_struct *work) 
	hdev = cs->ctx->hdev;



	/* Save only the first CS timeout parameters */

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

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

	if (!rc) {

		hdev->last_error.open_dev_timestamp = hdev->last_successful_open_ktime;

		hdev->last_error.cs_timeout_timestamp = ktime_get();

		hdev->last_error.cs_timeout_seq = cs->sequence;

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

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

	}



	switch (cs->type) {

drivers/misc/habanalabs/common/habanalabs.h

+42 −29
Original line number Diff line number Diff line
@@ -2568,37 +2568,50 @@ struct hl_clk_throttle { 
};



/**

 * struct last_error_session_info - info about last session in which CS timeout or

 *                                    razwi error occurred.

 * @open_dev_timestamp: device open timestamp.

 * @cs_timeout_timestamp: CS timeout timestamp.

 * @razwi_timestamp: razwi timestamp.

 * @cs_write_disable: if set writing to CS parameters in the structure is disabled so the

 *                    first (root cause) CS timeout will not be overwritten.

 * @razwi_write_disable: if set writing to razwi parameters in the structure is disabled so the

 * struct cs_timeout_info - info of last CS timeout occurred.

 * @timestamp: CS timeout timestamp.

 * @write_disable: if set writing to CS parameters in the structure is disabled so,

 *                 the first (root cause) CS timeout will not be overwritten.

 * @seq: CS timeout sequence number.

 */

struct cs_timeout_info {

	ktime_t		timestamp;

	atomic_t	write_disable;

	u64		seq;

};



/**

 * struct razwi_info - info about last razwi error occurred.

 * @timestamp: razwi timestamp.

 * @write_disable: if set writing to razwi parameters in the structure is disabled so the

 *                 first (root cause) razwi will not be overwritten.

 * @cs_timeout_seq: CS timeout sequence number.

 * @razwi_addr: address that caused razwi.

 * @razwi_engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does

 * @addr: address that caused razwi.

 * @engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does

 *               not have engine id it will be set to U16_MAX.

 * @razwi_engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible

 * @engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible

 *               engines which one them caused the razwi. In that case, it will contain the

 *               second possible engine id, otherwise it will be set to U16_MAX.

 * @razwi_non_engine_initiator: in case the initiator of the razwi does not have engine id.

 * @razwi_type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.

 * @non_engine_initiator: in case the initiator of the razwi does not have engine id.

 * @type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.

 */

struct razwi_info {

	ktime_t		timestamp;

	atomic_t	write_disable;

	u64		addr;

	u16		engine_id_1;

	u16		engine_id_2;

	u8		non_engine_initiator;

	u8		type;

};



/**

 * struct last_error_session_info - info about last session errors occurred.

 * @cs_timeout: CS timeout error last information.

 * @razwi: razwi last information.

 */

struct last_error_session_info {

	ktime_t		open_dev_timestamp;

	ktime_t		cs_timeout_timestamp;

	ktime_t		razwi_timestamp;

	atomic_t	cs_write_disable;

	atomic_t	razwi_write_disable;

	u64		cs_timeout_seq;

	u64		razwi_addr;

	u16		razwi_engine_id_1;

	u16		razwi_engine_id_2;

	u8		razwi_non_engine_initiator;

	u8		razwi_type;

	struct	cs_timeout_info	cs_timeout;

	struct	razwi_info	razwi;

};



/**

drivers/misc/habanalabs/common/habanalabs_drv.c

+2 −2
Original line number Diff line number Diff line
@@ -195,8 +195,8 @@ int hl_device_open(struct inode *inode, struct file *filp) 


	hl_debugfs_add_file(hpriv);



	atomic_set(&hdev->last_error.cs_write_disable, 0);

	atomic_set(&hdev->last_error.razwi_write_disable, 0);

	atomic_set(&hdev->last_error.cs_timeout.write_disable, 0);

	atomic_set(&hdev->last_error.razwi.write_disable, 0);



	hdev->open_counter++;

	hdev->last_successful_open_jif = jiffies;

drivers/misc/habanalabs/common/habanalabs_ioctl.c

+9 −9
Original line number Diff line number Diff line
@@ -569,7 +569,7 @@ static int last_err_open_dev_info(struct hl_fpriv *hpriv, struct hl_info_args *a 
	if ((!max_size) || (!out))

		return -EINVAL;



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

	info.timestamp = ktime_to_ns(hdev->last_successful_open_ktime);



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

}
@@ -584,8 +584,8 @@ 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->last_error.cs_timeout.seq;

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



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

}
@@ -600,12 +600,12 @@ 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->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;



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

}

drivers/misc/habanalabs/gaudi/gaudi.c

+7 −8
Original line number Diff line number Diff line
@@ -7404,19 +7404,18 @@ 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_disable, 0, 1);

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

		if (!rc) {

			hdev->last_error.open_dev_timestamp = hdev->last_successful_open_ktime;

			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->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;

			/*

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

			hdev->last_error.razwi.type = razwi_type;



		}

	}