Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit b588c193 authored by Eric Holmberg's avatar Eric Holmberg
Browse files

msm: smsm: add snapshot queue scheduling delay 



When debugging delayed SMSM callbacks, it is difficult to determine if
the delay is due to a scheduling issue for the callback worker or a
delayed interrupt issue.  In addition, sometimes the interrupt timestamp
needs to be determined from a crash dump.

Add timestamp when the SMSM interrupt snapshot is queued and calculate
the time delta when the snapshot is dequeued.

Change-Id: Ife9bd790ff4b1943c75ee9bbfdb0b77d2aca70f5
Signed-off-by: default avatarEric Holmberg <eholmber@codeaurora.org>
parent dc077d46

arch/arm/mach-msm/smd.c

+28 −1
Original line number Diff line number Diff line
@@ -56,7 +56,7 @@ 


#define SMD_VERSION 0x00020000

#define SMSM_SNAPSHOT_CNT 64

#define SMSM_SNAPSHOT_SIZE ((SMSM_NUM_ENTRIES + 1) * 4)

#define SMSM_SNAPSHOT_SIZE ((SMSM_NUM_ENTRIES + 1) * 4 + sizeof(uint64_t))

#define RSPIN_INIT_WAIT_MS 1000

#define SMD_FIFO_FULL_RESERVE 4
@@ -2560,7 +2560,9 @@ static void smsm_cb_snapshot(uint32_t use_wakelock) 
	uint32_t new_state;

	unsigned long flags;

	int ret;

	uint64_t timestamp;



	timestamp = sched_clock();

	ret = kfifo_avail(&smsm_snapshot_fifo);

	if (ret < SMSM_SNAPSHOT_SIZE) {

		pr_err("%s: SMSM snapshot full %d\n", __func__, ret);
@@ -2602,6 +2604,12 @@ static void smsm_cb_snapshot(uint32_t use_wakelock) 
		}

	}



	ret = kfifo_in(&smsm_snapshot_fifo, &timestamp, sizeof(timestamp));

	if (ret != sizeof(timestamp)) {

		pr_err("%s: SMSM snapshot failure %d\n", __func__, ret);

		goto restore_snapshot_count;

	}



	/* queue wakelock usage flag */

	ret = kfifo_in(&smsm_snapshot_fifo,

			&use_wakelock, sizeof(use_wakelock));
@@ -2868,8 +2876,12 @@ void notify_smsm_cb_clients_worker(struct work_struct *work) 
	uint32_t use_wakelock;

	int ret;

	unsigned long flags;

	uint64_t t_snapshot;

	uint64_t t_start;

	unsigned long nanosec_rem;



	while (kfifo_len(&smsm_snapshot_fifo) >= SMSM_SNAPSHOT_SIZE) {

		t_start = sched_clock();

		mutex_lock(&smsm_lock);

		for (n = 0; n < SMSM_NUM_ENTRIES; n++) {

			state_info = &smsm_states[n];
@@ -2900,6 +2912,15 @@ void notify_smsm_cb_clients_worker(struct work_struct *work) 
			}

		}



		ret = kfifo_out(&smsm_snapshot_fifo, &t_snapshot,

				sizeof(t_snapshot));

		if (ret != sizeof(t_snapshot)) {

			pr_err("%s: snapshot underflow %d\n",

				__func__, ret);

			mutex_unlock(&smsm_lock);

			return;

		}



		/* read wakelock flag */

		ret = kfifo_out(&smsm_snapshot_fifo, &use_wakelock,

				sizeof(use_wakelock));
@@ -2927,6 +2948,12 @@ void notify_smsm_cb_clients_worker(struct work_struct *work) 
			spin_unlock_irqrestore(&smsm_snapshot_count_lock,

					flags);

		}



		t_start = t_start - t_snapshot;

		nanosec_rem = do_div(t_start, 1000000000U);

		SMSM_POWER_INFO(

			"SMSM snapshot queue response time %6u.%09lu s\n",

			(unsigned)t_start, nanosec_rem);

	}

}