drm/amdgpu: enable separate timeout setting for every ring type V4

Every ring type can have its own timeout setting.

 - V2: update lockup_timeout parameter format and cosmetic fixes
 - V3: invalidate 0 and negative values
 - V4: update lockup_timeout parameter format
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -118,7 +118,6 @@ extern int amdgpu_disp_priority;
 extern int amdgpu_hw_i2c;
 extern int amdgpu_pcie_gen2;
 extern int amdgpu_msi;
-extern int amdgpu_lockup_timeout;
 extern int amdgpu_dpm;
 extern int amdgpu_fw_load_type;
 extern int amdgpu_aspm;
@@ -415,6 +414,7 @@ struct amdgpu_fpriv {
 };
 
 int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv);
+int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev);
 
 int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		  unsigned size, struct amdgpu_ib *ib);
@@ -943,6 +943,11 @@ struct amdgpu_device {
 	struct work_struct		xgmi_reset_work;
 
 	bool                            in_baco_reset;
+
+	long				gfx_timeout;
+	long				sdma_timeout;
+	long				video_timeout;
+	long				compute_timeout;
 };
 
 static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_bo_device *bdev)

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -910,8 +910,10 @@ static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
  * Validates certain module parameters and updates
  * the associated values used by the driver (all asics).
  */
-static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
+static int amdgpu_device_check_arguments(struct amdgpu_device *adev)
 {
+	int ret = 0;
+
 	if (amdgpu_sched_jobs < 4) {
 		dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n",
 			 amdgpu_sched_jobs);
@@ -956,12 +958,15 @@ static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
 		amdgpu_vram_page_split = 1024;
 	}
 
-	if (amdgpu_lockup_timeout == 0) {
-		dev_warn(adev->dev, "lockup_timeout msut be > 0, adjusting to 10000\n");
-		amdgpu_lockup_timeout = 10000;
+	ret = amdgpu_device_get_job_timeout_settings(adev);
+	if (ret) {
+		dev_err(adev->dev, "invalid lockup_timeout parameter syntax\n");
+		return ret;
 	}
 
 	adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type);
+
+	return ret;
 }
 
 /**
@@ -2473,7 +2478,9 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	mutex_init(&adev->lock_reset);
 	mutex_init(&adev->virt.dpm_mutex);
 
-	amdgpu_device_check_arguments(adev);
+	r = amdgpu_device_check_arguments(adev);
+	if (r)
+		return r;
 
 	spin_lock_init(&adev->mmio_idx_lock);
 	spin_lock_init(&adev->smc_idx_lock);

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -81,6 +81,8 @@
 #define KMS_DRIVER_MINOR	32
 #define KMS_DRIVER_PATCHLEVEL	0
 
+#define AMDGPU_MAX_TIMEOUT_PARAM_LENTH	256
+
 int amdgpu_vram_limit = 0;
 int amdgpu_vis_vram_limit = 0;
 int amdgpu_gart_size = -1; /* auto */
@@ -93,7 +95,7 @@ int amdgpu_disp_priority = 0;
 int amdgpu_hw_i2c = 0;
 int amdgpu_pcie_gen2 = -1;
 int amdgpu_msi = -1;
-int amdgpu_lockup_timeout = 10000;
+char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENTH];
 int amdgpu_dpm = -1;
 int amdgpu_fw_load_type = -1;
 int amdgpu_aspm = -1;
@@ -227,12 +229,21 @@ MODULE_PARM_DESC(msi, "MSI support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(msi, amdgpu_msi, int, 0444);
 
 /**
- * DOC: lockup_timeout (int)
- * Set GPU scheduler timeout value in ms. Value 0 is invalidated, will be adjusted to 10000.
- * Negative values mean 'infinite timeout' (MAX_JIFFY_OFFSET). The default is 10000.
- */
-MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms > 0 (default 10000)");
-module_param_named(lockup_timeout, amdgpu_lockup_timeout, int, 0444);
+ * DOC: lockup_timeout (string)
+ * Set GPU scheduler timeout value in ms.
+ *
+ * The format can be [Non-Compute] or [GFX,Compute,SDMA,Video]. That is there can be one or
+ * multiple values specified. 0 and negative values are invalidated. They will be adjusted
+ * to default timeout.
+ *  - With one value specified, the setting will apply to all non-compute jobs.
+ *  - With multiple values specified, the first one will be for GFX. The second one is for Compute.
+ *    And the third and fourth ones are for SDMA and Video.
+ * By default(with no lockup_timeout settings), the timeout for all non-compute(GFX, SDMA and Video)
+ * jobs is 10000. And there is no timeout enforced on compute jobs.
+ */
+MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (default: 10000 for non-compute jobs and no timeout for compute jobs), "
+		"format is [Non-Compute] or [GFX,Compute,SDMA,Video]");
+module_param_string(lockup_timeout, amdgpu_lockup_timeout, sizeof(amdgpu_lockup_timeout), 0444);
 
 /**
  * DOC: dpm (int)
@@ -1216,6 +1227,62 @@ int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv)
 	return 0;
 }
 
+int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
+{
+	char *input = amdgpu_lockup_timeout;
+	char *timeout_setting = NULL;
+	int index = 0;
+	long timeout;
+	int ret = 0;
+
+	/*
+	 * By default timeout for non compute jobs is 10000.
+	 * And there is no timeout enforced on compute jobs.
+	 */
+	adev->gfx_timeout = adev->sdma_timeout = adev->video_timeout = 10000;
+	adev->compute_timeout = MAX_SCHEDULE_TIMEOUT;
+
+	if (strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENTH)) {
+		while ((timeout_setting = strsep(&input, ",")) &&
+				strnlen(timeout_setting, AMDGPU_MAX_TIMEOUT_PARAM_LENTH)) {
+			ret = kstrtol(timeout_setting, 0, &timeout);
+			if (ret)
+				return ret;
+
+			/* Invalidate 0 and negative values */
+			if (timeout <= 0) {
+				index++;
+				continue;
+			}
+
+			switch (index++) {
+			case 0:
+				adev->gfx_timeout = timeout;
+				break;
+			case 1:
+				adev->compute_timeout = timeout;
+				break;
+			case 2:
+				adev->sdma_timeout = timeout;
+				break;
+			case 3:
+				adev->video_timeout = timeout;
+				break;
+			default:
+				break;
+			}
+		}
+		/*
+		 * There is only one value specified and
+		 * it should apply to all non-compute jobs.
+		 */
+		if (index == 1)
+			adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout;
+	}
+
+	return ret;
+}
+
 static bool
 amdgpu_get_crtc_scanout_position(struct drm_device *dev, unsigned int pipe,
 				 bool in_vblank_irq, int *vpos, int *hpos,

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

@@ -427,9 +427,13 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 				  unsigned num_hw_submission)
 {
+	struct amdgpu_device *adev = ring->adev;
 	long timeout;
 	int r;
 
+	if (!adev)
+		return -EINVAL;
+
 	/* Check that num_hw_submission is a power of two */
 	if ((num_hw_submission & (num_hw_submission - 1)) != 0)
 		return -EINVAL;
@@ -451,12 +455,31 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 
 	/* No need to setup the GPU scheduler for KIQ ring */
 	if (ring->funcs->type != AMDGPU_RING_TYPE_KIQ) {
-		/* for non-sriov case, no timeout enforce on compute ring */
-		if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE)
-				&& !amdgpu_sriov_vf(ring->adev))
-			timeout = MAX_SCHEDULE_TIMEOUT;
-		else
-			timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
+		switch (ring->funcs->type) {
+		case AMDGPU_RING_TYPE_GFX:
+			timeout = adev->gfx_timeout;
+			break;
+		case AMDGPU_RING_TYPE_COMPUTE:
+			/*
+			 * For non-sriov case, no timeout enforce
+			 * on compute ring by default. Unless user
+			 * specifies a timeout for compute ring.
+			 *
+			 * For sriov case, always use the timeout
+			 * as gfx ring
+			 */
+			if (!amdgpu_sriov_vf(ring->adev))
+				timeout = adev->compute_timeout;
+			else
+				timeout = adev->gfx_timeout;
+			break;
+		case AMDGPU_RING_TYPE_SDMA:
+			timeout = adev->sdma_timeout;
+			break;
+		default:
+			timeout = adev->video_timeout;
+			break;
+		}
 
 		r = drm_sched_init(&ring->sched, &amdgpu_sched_ops,
 				   num_hw_submission, amdgpu_job_hang_limit,

drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c

@@ -343,7 +343,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 
 	/* Trigger recovery for world switch failure if no TDR */
 	if (amdgpu_device_should_recover_gpu(adev)
-		&& amdgpu_lockup_timeout == MAX_SCHEDULE_TIMEOUT)
+		&& adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT)
 		amdgpu_device_gpu_recover(adev, NULL);
 }