drm/amdkfd: add sysfs counters for vm fault and migration

This is part of SVM profiling API, export sysfs counters for
per-process, per-GPU vm retry fault, pages migrated in and out of GPU vram.

counters will not be updated in parallel in GPU retry fault handler and
migration to vram/ram path, use READ_ONCE to avoid compiler
optimization.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -730,6 +730,15 @@ struct kfd_process_device {
 	 *  number of CU's a device has along with number of other competing processes
 	 */
 	struct attribute attr_cu_occupancy;
+
+	/* sysfs counters for GPU retry fault and page migration tracking */
+	struct kobject *kobj_counters;
+	struct attribute attr_faults;
+	struct attribute attr_page_in;
+	struct attribute attr_page_out;
+	uint64_t faults;
+	uint64_t page_in;
+	uint64_t page_out;
 };
 
 #define qpd_to_pdd(x) container_of(x, struct kfd_process_device, qpd)

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -416,6 +416,29 @@ static ssize_t kfd_procfs_stats_show(struct kobject *kobj,
 	return 0;
 }
 
+static ssize_t kfd_sysfs_counters_show(struct kobject *kobj,
+				       struct attribute *attr, char *buf)
+{
+	struct kfd_process_device *pdd;
+
+	if (!strcmp(attr->name, "faults")) {
+		pdd = container_of(attr, struct kfd_process_device,
+				   attr_faults);
+		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->faults));
+	}
+	if (!strcmp(attr->name, "page_in")) {
+		pdd = container_of(attr, struct kfd_process_device,
+				   attr_page_in);
+		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->page_in));
+	}
+	if (!strcmp(attr->name, "page_out")) {
+		pdd = container_of(attr, struct kfd_process_device,
+				   attr_page_out);
+		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->page_out));
+	}
+	return 0;
+}
+
 static struct attribute attr_queue_size = {
 	.name = "size",
 	.mode = KFD_SYSFS_FILE_MODE
@@ -456,6 +479,15 @@ static struct kobj_type procfs_stats_type = {
 	.release = kfd_procfs_kobj_release,
 };
 
+static const struct sysfs_ops sysfs_counters_ops = {
+	.show = kfd_sysfs_counters_show,
+};
+
+static struct kobj_type sysfs_counters_type = {
+	.sysfs_ops = &sysfs_counters_ops,
+	.release = kfd_procfs_kobj_release,
+};
+
 int kfd_procfs_add_queue(struct queue *q)
 {
 	struct kfd_process *proc;
@@ -544,6 +576,50 @@ static void kfd_procfs_add_sysfs_stats(struct kfd_process *p)
 	}
 }
 
+static void kfd_procfs_add_sysfs_counters(struct kfd_process *p)
+{
+	int ret = 0;
+	int i;
+	char counters_dir_filename[MAX_SYSFS_FILENAME_LEN];
+
+	if (!p || !p->kobj)
+		return;
+
+	/*
+	 * Create sysfs files for each GPU which supports SVM
+	 * - proc/<pid>/counters_<gpuid>/
+	 * - proc/<pid>/counters_<gpuid>/faults
+	 * - proc/<pid>/counters_<gpuid>/page_in
+	 * - proc/<pid>/counters_<gpuid>/page_out
+	 */
+	for_each_set_bit(i, p->svms.bitmap_supported, p->n_pdds) {
+		struct kfd_process_device *pdd = p->pdds[i];
+		struct kobject *kobj_counters;
+
+		snprintf(counters_dir_filename, MAX_SYSFS_FILENAME_LEN,
+			"counters_%u", pdd->dev->id);
+		kobj_counters = kfd_alloc_struct(kobj_counters);
+		if (!kobj_counters)
+			return;
+
+		ret = kobject_init_and_add(kobj_counters, &sysfs_counters_type,
+					   p->kobj, counters_dir_filename);
+		if (ret) {
+			pr_warn("Creating KFD proc/%s folder failed",
+				counters_dir_filename);
+			kobject_put(kobj_counters);
+			return;
+		}
+
+		pdd->kobj_counters = kobj_counters;
+		kfd_sysfs_create_file(kobj_counters, &pdd->attr_faults,
+				      "faults");
+		kfd_sysfs_create_file(kobj_counters, &pdd->attr_page_in,
+				      "page_in");
+		kfd_sysfs_create_file(kobj_counters, &pdd->attr_page_out,
+				      "page_out");
+	}
+}
 
 static void kfd_procfs_add_sysfs_files(struct kfd_process *p)
 {
@@ -777,6 +853,7 @@ struct kfd_process *kfd_create_process(struct file *filep)
 
 		kfd_procfs_add_sysfs_stats(process);
 		kfd_procfs_add_sysfs_files(process);
+		kfd_procfs_add_sysfs_counters(process);
 	}
 out:
 	if (!IS_ERR(process))
@@ -919,44 +996,60 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 	p->n_pdds = 0;
 }
 
-/* No process locking is needed in this function, because the process
- * is not findable any more. We must assume that no other thread is
- * using it any more, otherwise we couldn't safely free the process
- * structure in the end.
- */
-static void kfd_process_wq_release(struct work_struct *work)
+static void kfd_process_remove_sysfs(struct kfd_process *p)
 {
-	struct kfd_process *p = container_of(work, struct kfd_process,
-					     release_work);
+	struct kfd_process_device *pdd;
 	int i;
 
-	/* Remove the procfs files */
-	if (p->kobj) {
-		sysfs_remove_file(p->kobj, &p->attr_pasid);
-		kobject_del(p->kobj_queues);
-		kobject_put(p->kobj_queues);
-		p->kobj_queues = NULL;
+	if (!p->kobj)
+		return;
 
-		for (i = 0; i < p->n_pdds; i++) {
-			struct kfd_process_device *pdd = p->pdds[i];
+	sysfs_remove_file(p->kobj, &p->attr_pasid);
+	kobject_del(p->kobj_queues);
+	kobject_put(p->kobj_queues);
+	p->kobj_queues = NULL;
 
-			sysfs_remove_file(p->kobj, &pdd->attr_vram);
-			sysfs_remove_file(p->kobj, &pdd->attr_sdma);
+	for (i = 0; i < p->n_pdds; i++) {
+		pdd = p->pdds[i];
 
-			sysfs_remove_file(pdd->kobj_stats, &pdd->attr_evict);
-			if (pdd->dev->kfd2kgd->get_cu_occupancy)
-				sysfs_remove_file(pdd->kobj_stats,
-						  &pdd->attr_cu_occupancy);
-			kobject_del(pdd->kobj_stats);
-			kobject_put(pdd->kobj_stats);
-			pdd->kobj_stats = NULL;
-		}
+		sysfs_remove_file(p->kobj, &pdd->attr_vram);
+		sysfs_remove_file(p->kobj, &pdd->attr_sdma);
 
-		kobject_del(p->kobj);
-		kobject_put(p->kobj);
-		p->kobj = NULL;
+		sysfs_remove_file(pdd->kobj_stats, &pdd->attr_evict);
+		if (pdd->dev->kfd2kgd->get_cu_occupancy)
+			sysfs_remove_file(pdd->kobj_stats,
+					  &pdd->attr_cu_occupancy);
+		kobject_del(pdd->kobj_stats);
+		kobject_put(pdd->kobj_stats);
+		pdd->kobj_stats = NULL;
+	}
+
+	for_each_set_bit(i, p->svms.bitmap_supported, p->n_pdds) {
+		pdd = p->pdds[i];
+
+		sysfs_remove_file(pdd->kobj_counters, &pdd->attr_faults);
+		sysfs_remove_file(pdd->kobj_counters, &pdd->attr_page_in);
+		sysfs_remove_file(pdd->kobj_counters, &pdd->attr_page_out);
+		kobject_del(pdd->kobj_counters);
+		kobject_put(pdd->kobj_counters);
+		pdd->kobj_counters = NULL;
 	}
 
+	kobject_del(p->kobj);
+	kobject_put(p->kobj);
+	p->kobj = NULL;
+}
+
+/* No process locking is needed in this function, because the process
+ * is not findable any more. We must assume that no other thread is
+ * using it any more, otherwise we couldn't safely free the process
+ * structure in the end.
+ */
+static void kfd_process_wq_release(struct work_struct *work)
+{
+	struct kfd_process *p = container_of(work, struct kfd_process,
+					     release_work);
+	kfd_process_remove_sysfs(p);
 	kfd_iommu_unbind_process(p);
 
 	kfd_process_free_outstanding_kfd_bos(p);