[PATCH 7/8] accel/habanalabs: change the heartbeat scheduling point

[Ofir Bitton]

From: Farah Kassabri <fkassabri at habana.ai>

Currently we schedule the heartbeat thread at late init, only then
we set the INTS_REGISTER packet which enables events to be received
from firmware.

Init may take some time and we want to give firmware 2 full cycles of
heartbeat thread after it received INTS_REGISTER.

The patch will move the heartbeat thread scheduling to be after driver
is done with all initializations.

Signed-off-by: Farah Kassabri <fkassabri at habana.ai>
Reviewed-by: Ofir Bitton <obitton at habana.ai>
---
 drivers/accel/habanalabs/common/device.c | 54 +++++++++++++++---------
 1 file changed, 33 insertions(+), 21 deletions(-)

diff --git a/drivers/accel/habanalabs/common/device.c b/drivers/accel/habanalabs/common/device.c
index 31daa9184537..5ca7014def00 100644
--- a/drivers/accel/habanalabs/common/device.c
+++ b/drivers/accel/habanalabs/common/device.c
@@ -1150,23 +1150,6 @@ static int device_late_init(struct hl_device *hdev)
 	}
 
 	hdev->high_pll = hdev->asic_prop.high_pll;
-
-	if (hdev->heartbeat) {
-		hdev->heartbeat_debug_info.heartbeat_event_counter = 0;
-
-		/*
-		 * Before scheduling the heartbeat driver will check if eq event has received.
-		 * for the first schedule we need to set the indication as true then for the next
-		 * one this indication will be true only if eq event was sent by FW.
-		 */
-		hdev->eq_heartbeat_received = true;
-
-		INIT_DELAYED_WORK(&hdev->work_heartbeat, hl_device_heartbeat);
-
-		schedule_delayed_work(&hdev->work_heartbeat,
-				usecs_to_jiffies(HL_HEARTBEAT_PER_USEC));
-	}
-
 	hdev->late_init_done = true;
 
 	return 0;
@@ -1183,9 +1166,6 @@ static void device_late_fini(struct hl_device *hdev)
 	if (!hdev->late_init_done)
 		return;
 
-	if (hdev->heartbeat)
-		cancel_delayed_work_sync(&hdev->work_heartbeat);
-
 	if (hdev->asic_funcs->late_fini)
 		hdev->asic_funcs->late_fini(hdev);
 
@@ -1286,8 +1266,12 @@ static void hl_abort_waiting_for_completions(struct hl_device *hdev)
 static void cleanup_resources(struct hl_device *hdev, bool hard_reset, bool fw_reset,
 				bool skip_wq_flush)
 {
-	if (hard_reset)
+	if (hard_reset) {
+		if (hdev->heartbeat)
+			cancel_delayed_work_sync(&hdev->work_heartbeat);
+
 		device_late_fini(hdev);
+	}
 
 	/*
 	 * Halt the engines and disable interrupts so we won't get any more
@@ -1565,6 +1549,26 @@ static void handle_reset_trigger(struct hl_device *hdev, u32 flags)
 	}
 }
 
+static inline void device_heartbeat_schedule(struct hl_device *hdev)
+{
+	if (!hdev->heartbeat)
+		return;
+
+	hdev->heartbeat_debug_info.heartbeat_event_counter = 0;
+
+	/*
+	 * Before scheduling the heartbeat driver will check if eq event has received.
+	 * for the first schedule we need to set the indication as true then for the next
+	 * one this indication will be true only if eq event was sent by FW.
+	 */
+	hdev->eq_heartbeat_received = true;
+
+	INIT_DELAYED_WORK(&hdev->work_heartbeat, hl_device_heartbeat);
+
+	schedule_delayed_work(&hdev->work_heartbeat,
+			usecs_to_jiffies(HL_HEARTBEAT_PER_USEC));
+}
+
 /*
  * hl_device_reset - reset the device
  *
@@ -1934,6 +1938,8 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 	if (hard_reset) {
 		hdev->reset_info.hard_reset_cnt++;
 
+		device_heartbeat_schedule(hdev);
+
 		/* After reset is done, we are ready to receive events from
 		 * the F/W. We can't do it before because we will ignore events
 		 * and if those events are fatal, we won't know about it and
@@ -2368,6 +2374,12 @@ int hl_device_init(struct hl_device *hdev)
 		goto out_disabled;
 	}
 
+	/* Scheduling the EQ heartbeat thread must come after driver is done with all
+	 * initializations, as we want to make sure the FW gets enough time to be prepared
+	 * to respond to heartbeat packets.
+	 */
+	device_heartbeat_schedule(hdev);
+
 	dev_notice(hdev->dev,
 		"Successfully added device %s to habanalabs driver\n",
 		dev_name(&(hdev)->pdev->dev));
-- 
2.34.1