drm/amdgpu: fix vf error handling

The error handling for virtual functions assumed a single
vf per VM and didn't properly account for bare metal.  Make
the error arrays per device and add locking.
Reviewed-by: NGavin Wan <gavin.wan@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

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

@@ -2040,6 +2040,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	mutex_init(&adev->srbm_mutex);
 	mutex_init(&adev->grbm_idx_mutex);
 	mutex_init(&adev->mn_lock);
+	mutex_init(&adev->virt.vf_errors.lock);
 	hash_init(adev->mn_hash);
 
 	amdgpu_check_arguments(adev);
@@ -2125,7 +2126,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	r = amdgpu_atombios_init(adev);
 	if (r) {
 		dev_err(adev->dev, "amdgpu_atombios_init failed\n");
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
 		goto failed;
 	}
 
@@ -2136,7 +2137,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	if (amdgpu_vpost_needed(adev)) {
 		if (!adev->bios) {
 			dev_err(adev->dev, "no vBIOS found\n");
-			amdgpu_vf_error_put(AMDGIM_ERROR_VF_NO_VBIOS, 0, 0);
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0);
 			r = -EINVAL;
 			goto failed;
 		}
@@ -2144,7 +2145,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
 		if (r) {
 			dev_err(adev->dev, "gpu post error!\n");
-			amdgpu_vf_error_put(AMDGIM_ERROR_VF_GPU_POST_ERROR, 0, 0);
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_POST_ERROR, 0, 0);
 			goto failed;
 		}
 	} else {
@@ -2156,7 +2157,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		r = amdgpu_atomfirmware_get_clock_info(adev);
 		if (r) {
 			dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
-			amdgpu_vf_error_put(AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
 			goto failed;
 		}
 	} else {
@@ -2164,7 +2165,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		r = amdgpu_atombios_get_clock_info(adev);
 		if (r) {
 			dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
-			amdgpu_vf_error_put(AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
 			goto failed;
 		}
 		/* init i2c buses */
@@ -2175,7 +2176,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	r = amdgpu_fence_driver_init(adev);
 	if (r) {
 		dev_err(adev->dev, "amdgpu_fence_driver_init failed\n");
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
 		goto failed;
 	}
 
@@ -2185,7 +2186,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	r = amdgpu_init(adev);
 	if (r) {
 		dev_err(adev->dev, "amdgpu_init failed\n");
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
 		amdgpu_fini(adev);
 		goto failed;
 	}
@@ -2205,7 +2206,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	r = amdgpu_ib_pool_init(adev);
 	if (r) {
 		dev_err(adev->dev, "IB initialization failed (%d).\n", r);
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
 		goto failed;
 	}
 
@@ -2254,7 +2255,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	r = amdgpu_late_init(adev);
 	if (r) {
 		dev_err(adev->dev, "amdgpu_late_init failed\n");
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
 		goto failed;
 	}
 
@@ -2936,7 +2937,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 		}
 	} else {
 		dev_err(adev->dev, "asic resume failed (%d).\n", r);
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_ASIC_RESUME_FAIL, 0, r);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ASIC_RESUME_FAIL, 0, r);
 		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
 			if (adev->rings[i] && adev->rings[i]->sched.thread) {
 				kthread_unpark(adev->rings[i]->sched.thread);
@@ -2950,7 +2951,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 	if (r) {
 		/* bad news, how to tell it to userspace ? */
 		dev_info(adev->dev, "GPU reset failed\n");
-		amdgpu_vf_error_put(AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r);
 	}
 	else {
 		dev_info(adev->dev, "GPU reset successed!\n");

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

@@ -25,30 +25,21 @@
 #include "amdgpu_vf_error.h"
 #include "mxgpu_ai.h"
 
-#define AMDGPU_VF_ERROR_ENTRY_SIZE    16 
-
-/* struct error_entry - amdgpu VF error information. */
-struct amdgpu_vf_error_buffer {
-	int read_count;
-	int write_count;
-	uint16_t code[AMDGPU_VF_ERROR_ENTRY_SIZE];
-	uint16_t flags[AMDGPU_VF_ERROR_ENTRY_SIZE];
-	uint64_t data[AMDGPU_VF_ERROR_ENTRY_SIZE];
-};
-
-struct amdgpu_vf_error_buffer admgpu_vf_errors;
-
-
-void amdgpu_vf_error_put(uint16_t sub_error_code, uint16_t error_flags, uint64_t error_data)
+void amdgpu_vf_error_put(struct amdgpu_device *adev,
+			 uint16_t sub_error_code,
+			 uint16_t error_flags,
+			 uint64_t error_data)
 {
 	int index;
 	uint16_t error_code = AMDGIM_ERROR_CODE(AMDGIM_ERROR_CATEGORY_VF, sub_error_code);
 
-	index = admgpu_vf_errors.write_count % AMDGPU_VF_ERROR_ENTRY_SIZE;
-	admgpu_vf_errors.code [index] = error_code;
-	admgpu_vf_errors.flags [index] = error_flags;
-	admgpu_vf_errors.data [index] = error_data;
-	admgpu_vf_errors.write_count ++;
+	mutex_lock(&adev->virt.vf_errors.lock);
+	index = adev->virt.vf_errors.write_count % AMDGPU_VF_ERROR_ENTRY_SIZE;
+	adev->virt.vf_errors.code [index] = error_code;
+	adev->virt.vf_errors.flags [index] = error_flags;
+	adev->virt.vf_errors.data [index] = error_data;
+	adev->virt.vf_errors.write_count ++;
+	mutex_unlock(&adev->virt.vf_errors.lock);
 }
 
 
@@ -58,7 +49,8 @@ void amdgpu_vf_error_trans_all(struct amdgpu_device *adev)
 	u32 data1, data2, data3;
 	int index;
 
-	if ((NULL == adev) || (!amdgpu_sriov_vf(adev)) || (!adev->virt.ops) || (!adev->virt.ops->trans_msg)) {
+	if ((NULL == adev) || (!amdgpu_sriov_vf(adev)) ||
+	    (!adev->virt.ops) || (!adev->virt.ops->trans_msg)) {
 		return;
 	}
 /*
@@ -68,18 +60,22 @@ void amdgpu_vf_error_trans_all(struct amdgpu_device *adev)
 		return;
 	}
 */
+
+	mutex_lock(&adev->virt.vf_errors.lock);
 	/* The errors are overlay of array, correct read_count as full. */
-	if (admgpu_vf_errors.write_count - admgpu_vf_errors.read_count > AMDGPU_VF_ERROR_ENTRY_SIZE) {
-		admgpu_vf_errors.read_count = admgpu_vf_errors.write_count - AMDGPU_VF_ERROR_ENTRY_SIZE;
+	if (adev->virt.vf_errors.write_count - adev->virt.vf_errors.read_count > AMDGPU_VF_ERROR_ENTRY_SIZE) {
+		adev->virt.vf_errors.read_count = adev->virt.vf_errors.write_count - AMDGPU_VF_ERROR_ENTRY_SIZE;
 	}
 
-	while (admgpu_vf_errors.read_count < admgpu_vf_errors.write_count) {
-		index =admgpu_vf_errors.read_count % AMDGPU_VF_ERROR_ENTRY_SIZE;
-		data1 = AMDGIM_ERROR_CODE_FLAGS_TO_MAILBOX (admgpu_vf_errors.code[index], admgpu_vf_errors.flags[index]);
-		data2 = admgpu_vf_errors.data[index] & 0xFFFFFFFF;
-		data3 = (admgpu_vf_errors.data[index] >> 32) & 0xFFFFFFFF;
+	while (adev->virt.vf_errors.read_count < adev->virt.vf_errors.write_count) {
+		index =adev->virt.vf_errors.read_count % AMDGPU_VF_ERROR_ENTRY_SIZE;
+		data1 = AMDGIM_ERROR_CODE_FLAGS_TO_MAILBOX(adev->virt.vf_errors.code[index],
+							   adev->virt.vf_errors.flags[index]);
+		data2 = adev->virt.vf_errors.data[index] & 0xFFFFFFFF;
+		data3 = (adev->virt.vf_errors.data[index] >> 32) & 0xFFFFFFFF;
 
 		adev->virt.ops->trans_msg(adev, IDH_LOG_VF_ERROR, data1, data2, data3);
-		admgpu_vf_errors.read_count ++;
+		adev->virt.vf_errors.read_count ++;
 	}
+	mutex_unlock(&adev->virt.vf_errors.lock);
 }

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

@@ -56,7 +56,10 @@ enum AMDGIM_ERROR_CATEGORY {
 	AMDGIM_ERROR_CATEGORY_MAX
 };
 
-void amdgpu_vf_error_put(uint16_t sub_error_code, uint16_t error_flags, uint64_t error_data);
+void amdgpu_vf_error_put(struct amdgpu_device *adev,
+			 uint16_t sub_error_code,
+			 uint16_t error_flags,
+			 uint64_t error_data);
 void amdgpu_vf_error_trans_all (struct amdgpu_device *adev);
 
 #endif /* __VF_ERROR_H__ */

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

@@ -36,6 +36,18 @@ struct amdgpu_mm_table {
 	uint64_t		gpu_addr;
 };
 
+#define AMDGPU_VF_ERROR_ENTRY_SIZE    16
+
+/* struct error_entry - amdgpu VF error information. */
+struct amdgpu_vf_error_buffer {
+	struct mutex lock;
+	int read_count;
+	int write_count;
+	uint16_t code[AMDGPU_VF_ERROR_ENTRY_SIZE];
+	uint16_t flags[AMDGPU_VF_ERROR_ENTRY_SIZE];
+	uint64_t data[AMDGPU_VF_ERROR_ENTRY_SIZE];
+};
+
 /**
  * struct amdgpu_virt_ops - amdgpu device virt operations
  */
@@ -59,6 +71,7 @@ struct amdgpu_virt {
 	struct work_struct		flr_work;
 	struct amdgpu_mm_table		mm_table;
 	const struct amdgpu_virt_ops	*ops;
+	struct amdgpu_vf_error_buffer   vf_errors;
 };
 
 #define AMDGPU_CSA_SIZE    (8 * 1024)