[PATCH 12/15] habanalabs: soft-reset device if context-switch fails

[Oded Gabbay <oded.gabbay@gmail.com>]

This patch fix a bug in the driver, where if the TPC or MME remains in
non-IDLE even after all the command submissions are done (due to user bug
or malicious user), then future command submissions will fail in the
context-switch stage and the driver will remain in "stuck" mode.

The fix is to do a soft-reset of the device in case the context-switch
fails, because the device should be IDLE during context-switch. If it is
not IDLE, then something is wrong and we should reset the compute engines.

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
---
 drivers/misc/habanalabs/command_submission.c | 16 +++++++++-------
 drivers/misc/habanalabs/goya/goya.c          |  2 +-
 2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/misc/habanalabs/command_submission.c b/drivers/misc/habanalabs/command_submission.c
index 25ad9d805cfa..3525236ed8d9 100644
--- a/drivers/misc/habanalabs/command_submission.c
+++ b/drivers/misc/habanalabs/command_submission.c
@@ -622,13 +622,15 @@ int hl_cs_ioctl(struct hl_fpriv *hpriv, void *data)
 					"Failed to switch to context %d, rejecting CS! %d\n",
 					ctx->asid, rc);
 				/*
-				 * If we timedout, we need to soft-reset because
-				 * QMAN is probably stuck. However, we can't
-				 * call to reset here directly because of
-				 * deadlock, so need to do it at the very end
-				 * of this function
+				 * If we timedout, or if the device is not IDLE
+				 * while we want to do context-switch (-EBUSY),
+				 * we need to soft-reset because QMAN is
+				 * probably stuck. However, we can't call to
+				 * reset here directly because of deadlock, so
+				 * need to do it at the very end of this
+				 * function
 				 */
-				if (rc == -ETIMEDOUT)
+				if ((rc == -ETIMEDOUT) || (rc == -EBUSY))
 					need_soft_reset = true;
 				mutex_unlock(&hpriv->restore_phase_mutex);
 				goto out;
@@ -706,7 +708,7 @@ int hl_cs_ioctl(struct hl_fpriv *hpriv, void *data)
 		args->out.seq = cs_seq;
 	}
 
-	if ((rc == -ETIMEDOUT) && (need_soft_reset))
+	if (((rc == -ETIMEDOUT) || (rc == -EBUSY)) && (need_soft_reset))
 		hl_device_reset(hdev, false, false);
 
 	return rc;
diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c
index 39824214ce61..11597432f519 100644
--- a/drivers/misc/habanalabs/goya/goya.c
+++ b/drivers/misc/habanalabs/goya/goya.c
@@ -3138,7 +3138,7 @@ static int goya_send_job_on_qman0(struct hl_device *hdev, struct hl_cs_job *job)
 	if (!hdev->asic_funcs->is_device_idle(hdev)) {
 		dev_err_ratelimited(hdev->dev,
 			"Can't send KMD job on QMAN0 if device is not idle\n");
-		return -EFAULT;
+		return -EBUSY;
 	}
 
 	fence_ptr = hdev->asic_funcs->dma_pool_zalloc(hdev, 4, GFP_KERNEL,
-- 
2.17.1