drm/v3d: Create a CPU job extension for the copy performance query job

A CPU job is a type of job that performs operations that requires CPU
intervention. A copy performance query job is a job that copy the complete
or partial result of a query to a buffer. In order to copy the result of
a performance query to a buffer, we need to get the values from the
performance monitors.

So, create a user extension for the CPU job that enables the creation
of a copy performance query job. This user extension will allow the creation
of a CPU job that copy the results of a performance query to a BO with the
possibility to indicate the availability with a availability bit.

Signed-off-by: Maíra Canal <[e-mail]>
Reviewed-by: Iago Toral Quiroga <[e-mail]>
Link: https://patchwork.freedesktop.org/patch/msgid/[e-mail]

drivers/gpu/drm/v3d/v3d_drv.h

···
 	V3D_CPU_JOB_TYPE_RESET_TIMESTAMP_QUERY,
 	V3D_CPU_JOB_TYPE_COPY_TIMESTAMP_QUERY,
 	V3D_CPU_JOB_TYPE_RESET_PERFORMANCE_QUERY,
+	V3D_CPU_JOB_TYPE_COPY_PERFORMANCE_QUERY,
 };
 
 struct v3d_timestamp_query {

drivers/gpu/drm/v3d/v3d_sched.c

···
 	}
 }
 
+static void
+v3d_write_performance_query_result(struct v3d_cpu_job *job, void *data, u32 query)
+{
+	struct v3d_performance_query_info *performance_query = &job->performance_query;
+	struct v3d_copy_query_results_info *copy = &job->copy;
+	struct v3d_file_priv *v3d_priv = job->base.file->driver_priv;
+	struct v3d_dev *v3d = job->base.v3d;
+	struct v3d_perfmon *perfmon;
+	u64 counter_values[V3D_PERFCNT_NUM];
+
+	for (int i = 0; i < performance_query->nperfmons; i++) {
+		perfmon = v3d_perfmon_find(v3d_priv,
+					   performance_query->queries[query].kperfmon_ids[i]);
+		if (!perfmon) {
+			DRM_DEBUG("Failed to find perfmon.");
+			continue;
+		}
+
+		v3d_perfmon_stop(v3d, perfmon, true);
+
+		memcpy(&counter_values[i * DRM_V3D_MAX_PERF_COUNTERS], perfmon->values,
+		       perfmon->ncounters * sizeof(u64));
+
+		v3d_perfmon_put(perfmon);
+	}
+
+	for (int i = 0; i < performance_query->ncounters; i++)
+		write_to_buffer(data, i, copy->do_64bit, counter_values[i]);
+}
+
+static void
+v3d_copy_performance_query(struct v3d_cpu_job *job)
+{
+	struct v3d_performance_query_info *performance_query = &job->performance_query;
+	struct v3d_copy_query_results_info *copy = &job->copy;
+	struct v3d_bo *bo = to_v3d_bo(job->base.bo[0]);
+	struct dma_fence *fence;
+	bool available, write_result;
+	u8 *data;
+
+	v3d_get_bo_vaddr(bo);
+
+	data = ((u8 *)bo->vaddr) + copy->offset;
+
+	for (int i = 0; i < performance_query->count; i++) {
+		fence = drm_syncobj_fence_get(performance_query->queries[i].syncobj);
+		available = fence ? dma_fence_is_signaled(fence) : false;
+
+		write_result = available || copy->do_partial;
+		if (write_result)
+			v3d_write_performance_query_result(job, data, i);
+
+		if (copy->availability_bit)
+			write_to_buffer(data, performance_query->ncounters,
+					copy->do_64bit, available ? 1u : 0u);
+
+		data += copy->stride;
+
+		dma_fence_put(fence);
+	}
+
+	v3d_put_bo_vaddr(bo);
+}
+
 static const v3d_cpu_job_fn cpu_job_function[] = {
 	[V3D_CPU_JOB_TYPE_INDIRECT_CSD] = v3d_rewrite_csd_job_wg_counts_from_indirect,
 	[V3D_CPU_JOB_TYPE_TIMESTAMP_QUERY] = v3d_timestamp_query,
 	[V3D_CPU_JOB_TYPE_RESET_TIMESTAMP_QUERY] = v3d_reset_timestamp_queries,
 	[V3D_CPU_JOB_TYPE_COPY_TIMESTAMP_QUERY] = v3d_copy_query_results,
 	[V3D_CPU_JOB_TYPE_RESET_PERFORMANCE_QUERY] = v3d_reset_performance_queries,
+	[V3D_CPU_JOB_TYPE_COPY_PERFORMANCE_QUERY] = v3d_copy_performance_query,
 };
 
 static struct dma_fence *

drivers/gpu/drm/v3d/v3d_submit.c

···
 	return 0;
 }
 
+static int
+v3d_get_cpu_copy_performance_query_params(struct drm_file *file_priv,
+					  struct drm_v3d_extension __user *ext,
+					  struct v3d_cpu_job *job)
+{
+	u32 __user *syncs;
+	u64 __user *kperfmon_ids;
+	struct drm_v3d_copy_performance_query copy;
+
+	if (!job) {
+		DRM_DEBUG("CPU job extension was attached to a GPU job.\n");
+		return -EINVAL;
+	}
+
+	if (job->job_type) {
+		DRM_DEBUG("Two CPU job extensions were added to the same CPU job.\n");
+		return -EINVAL;
+	}
+
+	if (copy_from_user(&copy, ext, sizeof(copy)))
+		return -EFAULT;
+
+	if (copy.pad)
+		return -EINVAL;
+
+	job->job_type = V3D_CPU_JOB_TYPE_COPY_PERFORMANCE_QUERY;
+
+	job->performance_query.queries = kvmalloc_array(copy.count,
+							sizeof(struct v3d_performance_query),
+							GFP_KERNEL);
+	if (!job->performance_query.queries)
+		return -ENOMEM;
+
+	syncs = u64_to_user_ptr(copy.syncs);
+	kperfmon_ids = u64_to_user_ptr(copy.kperfmon_ids);
+
+	for (int i = 0; i < copy.count; i++) {
+		u32 sync;
+		u64 ids;
+		u32 __user *ids_pointer;
+		u32 id;
+
+		if (copy_from_user(&sync, syncs++, sizeof(sync))) {
+			kvfree(job->performance_query.queries);
+			return -EFAULT;
+		}
+
+		job->performance_query.queries[i].syncobj = drm_syncobj_find(file_priv, sync);
+
+		if (copy_from_user(&ids, kperfmon_ids++, sizeof(ids))) {
+			kvfree(job->performance_query.queries);
+			return -EFAULT;
+		}
+
+		ids_pointer = u64_to_user_ptr(ids);
+
+		for (int j = 0; j < copy.nperfmons; j++) {
+			if (copy_from_user(&id, ids_pointer++, sizeof(id))) {
+				kvfree(job->performance_query.queries);
+				return -EFAULT;
+			}
+
+			job->performance_query.queries[i].kperfmon_ids[j] = id;
+		}
+	}
+	job->performance_query.count = copy.count;
+	job->performance_query.nperfmons = copy.nperfmons;
+	job->performance_query.ncounters = copy.ncounters;
+
+	job->copy.do_64bit = copy.do_64bit;
+	job->copy.do_partial = copy.do_partial;
+	job->copy.availability_bit = copy.availability_bit;
+	job->copy.offset = copy.offset;
+	job->copy.stride = copy.stride;
+
+	return 0;
+}
+
 /* Whenever userspace sets ioctl extensions, v3d_get_extensions parses data
  * according to the extension id (name).
  */
···
 			break;
 		case DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY:
 			ret = v3d_get_cpu_reset_performance_params(file_priv, user_ext, job);
+			break;
+		case DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY:
+			ret = v3d_get_cpu_copy_performance_query_params(file_priv, user_ext, job);
 			break;
 		default:
 			DRM_DEBUG_DRIVER("Unknown extension id: %d\n", ext.id);
···
 	[V3D_CPU_JOB_TYPE_RESET_TIMESTAMP_QUERY] = 1,
 	[V3D_CPU_JOB_TYPE_COPY_TIMESTAMP_QUERY] = 2,
 	[V3D_CPU_JOB_TYPE_RESET_PERFORMANCE_QUERY] = 0,
+	[V3D_CPU_JOB_TYPE_COPY_PERFORMANCE_QUERY] = 1,
 };
 
 /**

include/uapi/drm/v3d_drm.h

···
 #define DRM_V3D_EXT_ID_CPU_RESET_TIMESTAMP_QUERY	0x04
 #define DRM_V3D_EXT_ID_CPU_COPY_TIMESTAMP_QUERY	0x05
 #define DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY	0x06
+#define DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY	0x07
 	__u32 flags; /* mbz */
 };
 
···
 	__u64 kperfmon_ids;
 };
 
+/**
+ * struct drm_v3d_copy_performance_query - ioctl extension for the CPU job to copy
+ * performance query results to a buffer
+ *
+ * When an extension DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY is defined, it
+ * points to this extension to define a copy performance query submission. This
+ * CPU job will copy the performance queries results to a BO with the offset
+ * and stride defined in the extension.
+ */
+struct drm_v3d_copy_performance_query {
+	struct drm_v3d_extension base;
+
+	/* Define if should write to buffer using 64 or 32 bits */
+	__u8 do_64bit;
+
+	/* Define if it can write to buffer even if the query is not available */
+	__u8 do_partial;
+
+	/* Define if it should write availability bit to buffer */
+	__u8 availability_bit;
+
+	/* mbz */
+	__u8 pad;
+
+	/* Offset of the buffer in the BO */
+	__u32 offset;
+
+	/* Stride of the buffer in the BO */
+	__u32 stride;
+
+	/* Number of performance monitors */
+	__u32 nperfmons;
+
+	/* Number of performance counters related to this query pool */
+	__u32 ncounters;
+
+	/* Number of queries */
+	__u32 count;
+
+	/* Array of performance queries's syncobjs to indicate its availability */
+	__u64 syncs;
+
+	/* Array of u64 user-pointers that point to an array of kperfmon_ids */
+	__u64 kperfmon_ids;
+};
+
 struct drm_v3d_submit_cpu {
 	/* Pointer to a u32 array of the BOs that are referenced by the job.
 	 *
···
 	 *
 	 * For DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY, it must contain no
 	 * BOs.
+	 *
+	 * For DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY, it must contain one
+	 * BO, where the performance queries will be written.
 	 */
 	__u64 bo_handles;