1 files changed, 36 insertions, 31 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
index 0c7275bca8f7..f5411b798e11 100644
--- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
+++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c
@@ -93,7 +93,7 @@ static int xgpu_ai_poll_ack(struct amdgpu_device *adev)
 		timeout -= 5;
 	} while (timeout > 1);
 
-	pr_err("Doesn't get TRN_MSG_ACK from pf in %d msec\n", AI_MAILBOX_POLL_ACK_TIMEDOUT);
+	dev_err(adev->dev, "Doesn't get TRN_MSG_ACK from pf in %d msec\n", AI_MAILBOX_POLL_ACK_TIMEDOUT);
 
 	return -ETIME;
 }
@@ -111,7 +111,7 @@ static int xgpu_ai_poll_msg(struct amdgpu_device *adev, enum idh_event event)
 		timeout -= 10;
 	} while (timeout > 1);
 
-	pr_err("Doesn't get msg:%d from pf, error=%d\n", event, r);
+	dev_err(adev->dev, "Doesn't get msg:%d from pf, error=%d\n", event, r);
 
 	return -ETIME;
 }
@@ -132,7 +132,7 @@ static void xgpu_ai_mailbox_trans_msg (struct amdgpu_device *adev,
 		xgpu_ai_mailbox_set_valid(adev, false);
 		trn = xgpu_ai_peek_ack(adev);
 		if (trn) {
-			pr_err("trn=%x ACK should not assert! wait again !\n", trn);
+			dev_err_ratelimited(adev->dev, "trn=%x ACK should not assert! wait again !\n", trn);
 			msleep(1);
 		}
 	} while(trn);
@@ -155,7 +155,7 @@ static void xgpu_ai_mailbox_trans_msg (struct amdgpu_device *adev,
 	/* start to poll ack */
 	r = xgpu_ai_poll_ack(adev);
 	if (r)
-		pr_err("Doesn't get ack from pf, continue\n");
+		dev_err(adev->dev, "Doesn't get ack from pf, continue\n");
 
 	xgpu_ai_mailbox_set_valid(adev, false);
 }
@@ -173,7 +173,7 @@ static int xgpu_ai_send_access_requests(struct amdgpu_device *adev,
 		req == IDH_REQ_GPU_RESET_ACCESS) {
 		r = xgpu_ai_poll_msg(adev, IDH_READY_TO_ACCESS_GPU);
 		if (r) {
-			pr_err("Doesn't get READY_TO_ACCESS_GPU from pf, give up\n");
+			dev_err(adev->dev, "Doesn't get READY_TO_ACCESS_GPU from pf, give up\n");
 			return r;
 		}
 		/* Retrieve checksum from mailbox2 */
@@ -231,7 +231,7 @@ static int xgpu_ai_mailbox_ack_irq(struct amdgpu_device *adev,
 					struct amdgpu_irq_src *source,
 					struct amdgpu_iv_entry *entry)
 {
-	DRM_DEBUG("get ack intr and do nothing.\n");
+	dev_dbg(adev->dev, "get ack intr and do nothing.\n");
 	return 0;
 }
 
@@ -249,38 +249,33 @@ static int xgpu_ai_set_mailbox_ack_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
-static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
+static void xgpu_ai_ready_to_reset(struct amdgpu_device *adev)
 {
-	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, flr_work);
-	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
-	int timeout = AI_MAILBOX_POLL_FLR_TIMEDOUT;
-
-	/* block amdgpu_gpu_recover till msg FLR COMPLETE received,
-	 * otherwise the mailbox msg will be ruined/reseted by
-	 * the VF FLR.
-	 */
-	if (atomic_cmpxchg(&adev->reset_domain->in_gpu_reset, 0, 1) != 0)
-		return;
-
-	down_write(&adev->reset_domain->sem);
-
-	amdgpu_virt_fini_data_exchange(adev);
-
 	xgpu_ai_mailbox_trans_msg(adev, IDH_READY_TO_RESET, 0, 0, 0);
+}
 
+static int xgpu_ai_wait_reset(struct amdgpu_device *adev)
+{
+	int timeout = AI_MAILBOX_POLL_FLR_TIMEDOUT;
 	do {
-		if (xgpu_ai_mailbox_peek_msg(adev) == IDH_FLR_NOTIFICATION_CMPL)
-			goto flr_done;
-
+		if (xgpu_ai_mailbox_peek_msg(adev) == IDH_FLR_NOTIFICATION_CMPL) {
+			dev_dbg(adev->dev, "Got AI IDH_FLR_NOTIFICATION_CMPL after %d ms\n", AI_MAILBOX_POLL_FLR_TIMEDOUT - timeout);
+			return 0;
+		}
 		msleep(10);
 		timeout -= 10;
 	} while (timeout > 1);
 
-	dev_warn(adev->dev, "waiting IDH_FLR_NOTIFICATION_CMPL timeout\n");
+	dev_dbg(adev->dev, "waiting AI IDH_FLR_NOTIFICATION_CMPL timeout\n");
+	return -ETIME;
+}
+
+static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
+{
+	struct amdgpu_virt *virt = container_of(work, struct amdgpu_virt, flr_work);
+	struct amdgpu_device *adev = container_of(virt, struct amdgpu_device, virt);
 
-flr_done:
-	atomic_set(&adev->reset_domain->in_gpu_reset, 0);
-	up_write(&adev->reset_domain->sem);
+	amdgpu_virt_fini_data_exchange(adev);
 
 	/* Trigger recovery for world switch failure if no TDR */
 	if (amdgpu_device_should_recover_gpu(adev)
@@ -292,6 +287,7 @@ flr_done:
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+		set_bit(AMDGPU_HOST_FLR, &reset_context.flags);
 
 		amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 	}
@@ -319,7 +315,7 @@ static int xgpu_ai_mailbox_rcv_irq(struct amdgpu_device *adev,
 
 	switch (event) {
 		case IDH_FLR_NOTIFICATION:
-		if (amdgpu_sriov_runtime(adev) && !amdgpu_in_reset(adev))
+		if (amdgpu_sriov_runtime(adev))
 			WARN_ONCE(!amdgpu_reset_domain_schedule(adev->reset_domain,
 								&adev->virt.flr_work),
 				  "Failed to queue work! at %s",
@@ -412,12 +408,21 @@ static void xgpu_ai_ras_poison_handler(struct amdgpu_device *adev,
 	xgpu_ai_send_access_requests(adev, IDH_RAS_POISON);
 }
 
+static bool xgpu_ai_rcvd_ras_intr(struct amdgpu_device *adev)
+{
+	enum idh_event msg = xgpu_ai_mailbox_peek_msg(adev);
+
+	return (msg == IDH_RAS_ERROR_DETECTED || msg == 0xFFFFFFFF);
+}
+
 const struct amdgpu_virt_ops xgpu_ai_virt_ops = {
 	.req_full_gpu	= xgpu_ai_request_full_gpu_access,
 	.rel_full_gpu	= xgpu_ai_release_full_gpu_access,
 	.reset_gpu = xgpu_ai_request_reset,
-	.wait_reset = NULL,
+	.ready_to_reset = xgpu_ai_ready_to_reset,
+	.wait_reset = xgpu_ai_wait_reset,
 	.trans_msg = xgpu_ai_mailbox_trans_msg,
 	.req_init_data  = xgpu_ai_request_init_data,
 	.ras_poison_handler = xgpu_ai_ras_poison_handler,
+	.rcvd_ras_intr = xgpu_ai_rcvd_ras_intr,
 };