Commit 8e8125f1 authored by Yuri Nudelman's avatar Yuri Nudelman Committed by Oded Gabbay
Browse files

habanalabs: add debug flag to prevent failure on timeout



Sometimes it is useful to allow the command to continue running despite
the timeout occurred, to differentiate between really stuck or just very
time consuming commands. This can be achieved by passing a new debug
flag alongside the cs, HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT.

Anyway, if the timeout occurred, a warning print shall be issued,
however this shall not fail the submission.

Signed-off-by: default avatarYuri Nudelman <ynudelman@habana.ai>
Reviewed-by: default avatarOded Gabbay <ogabbay@kernel.org>
Signed-off-by: default avatarOded Gabbay <ogabbay@kernel.org>
parent 254fac6d
Loading
Loading
Loading
Loading
+20 −5
Original line number Diff line number Diff line
@@ -556,6 +556,13 @@ static void cs_do_release(struct kref *ref)
	else if (!cs->submitted)
		cs->fence->error = -EBUSY;

	if (unlikely(cs->skip_reset_on_timeout)) {
		dev_err(hdev->dev,
			"Command submission %llu completed after %llu (s)\n",
			cs->sequence,
			div_u64(jiffies - cs->submission_time_jiffies, HZ));
	}

	if (cs->timestamp)
		cs->fence->timestamp = ktime_get();
	complete_all(&cs->fence->completion);
@@ -571,6 +578,8 @@ static void cs_timedout(struct work_struct *work)
	int rc;
	struct hl_cs *cs = container_of(work, struct hl_cs,
						 work_tdr.work);
	bool skip_reset_on_timeout = cs->skip_reset_on_timeout;

	rc = cs_get_unless_zero(cs);
	if (!rc)
		return;
@@ -581,6 +590,7 @@ static void cs_timedout(struct work_struct *work)
	}

	/* Mark the CS is timed out so we won't try to cancel its TDR */
	if (likely(!skip_reset_on_timeout))
		cs->timedout = true;

	hdev = cs->ctx->hdev;
@@ -613,11 +623,13 @@ static void cs_timedout(struct work_struct *work)

	cs_put(cs);

	if (likely(!skip_reset_on_timeout)) {
		if (hdev->reset_on_lockup)
			hl_device_reset(hdev, HL_RESET_TDR);
		else
			hdev->needs_reset = true;
	}
}

static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
			enum hl_cs_type cs_type, u64 user_sequence,
@@ -650,6 +662,9 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
	cs->type = cs_type;
	cs->timestamp = !!(flags & HL_CS_FLAGS_TIMESTAMP);
	cs->timeout_jiffies = timeout;
	cs->skip_reset_on_timeout =
		!!(flags & HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT);
	cs->submission_time_jiffies = jiffies;
	INIT_LIST_HEAD(&cs->job_list);
	INIT_DELAYED_WORK(&cs->work_tdr, cs_timedout);
	kref_init(&cs->refcount);
+5 −0
Original line number Diff line number Diff line
@@ -1421,6 +1421,7 @@ struct hl_userptr {
 * @staged_sequence: the sequence of the staged submission this CS is part of,
 *                   relevant only if staged_cs is set.
 * @timeout_jiffies: cs timeout in jiffies.
 * @submission_time_jiffies: submission time of the cs
 * @type: CS_TYPE_*.
 * @submitted: true if CS was submitted to H/W.
 * @completed: true if CS was completed by device.
@@ -1433,6 +1434,8 @@ struct hl_userptr {
 * @staged_first: true if this is the first staged CS and we need to receive
 *                timeout for this CS.
 * @staged_cs: true if this CS is part of a staged submission.
 * @skip_reset_on_timeout: true if we shall not reset the device in case
 *                         timeout occurs (debug scenario).
 */
struct hl_cs {
	u16			*jobs_in_queue_cnt;
@@ -1450,6 +1453,7 @@ struct hl_cs {
	u64			sequence;
	u64			staged_sequence;
	u64			timeout_jiffies;
	u64			submission_time_jiffies;
	enum hl_cs_type		type;
	u8			submitted;
	u8			completed;
@@ -1460,6 +1464,7 @@ struct hl_cs {
	u8			staged_last;
	u8			staged_first;
	u8			staged_cs;
	u8			skip_reset_on_timeout;
};

/**
+1 −0
Original line number Diff line number Diff line
@@ -664,6 +664,7 @@ struct hl_cs_chunk {
#define HL_CS_FLAGS_STAGED_SUBMISSION_FIRST	0x80
#define HL_CS_FLAGS_STAGED_SUBMISSION_LAST	0x100
#define HL_CS_FLAGS_CUSTOM_TIMEOUT		0x200
#define HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT	0x400

#define HL_CS_STATUS_SUCCESS		0