drm/nouveau/nvkm: add locking to subdev/engine init paths

This wasn't really needed before; the main place this could race is with
channel recovery, but (through potentially fragile means) shouldn't have
been possible.

However, a number of upcoming patches benefit from having better control
over subdev init, necessitating some improvements here.

- allows subdev/engine oneinit() without init() (host/fifo patches)
- merges engine use locking/tracking into subdev, and extends it to fix
  some issues that will arise with future usage patterns (acr patches)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

drivers/gpu/drm/nouveau/include/nvkm/core/engine.h

@@ -12,12 +12,6 @@ struct nvkm_engine {
 	const struct nvkm_engine_func *func;
 	struct nvkm_subdev subdev;
 	spinlock_t lock;
-
-	struct {
-		refcount_t refcount;
-		struct mutex mutex;
-		bool enabled;
-	} use;
 };
 
 struct nvkm_engine_func {

drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h

@@ -17,12 +17,19 @@ struct nvkm_subdev {
 	struct nvkm_device *device;
 	enum nvkm_subdev_type type;
 	int inst;
+
 	char name[16];
 	u32 debug;
-	struct list_head head;
+
+	struct {
+		refcount_t refcount;
+		struct mutex mutex;
+		bool enabled;
+	} use;
 
 	struct nvkm_inth inth;
 
+	struct list_head head;
 	void **pself;
 	bool oneinit;
 };
@@ -40,11 +47,23 @@ struct nvkm_subdev_func {
 extern const char *nvkm_subdev_type[NVKM_SUBDEV_NR];
 int nvkm_subdev_new_(const struct nvkm_subdev_func *, struct nvkm_device *, enum nvkm_subdev_type,
 		     int inst, struct nvkm_subdev **);
-void nvkm_subdev_ctor(const struct nvkm_subdev_func *, struct nvkm_device *,
+void __nvkm_subdev_ctor(const struct nvkm_subdev_func *, struct nvkm_device *,
 			enum nvkm_subdev_type, int inst, struct nvkm_subdev *);
+
+static inline void
+nvkm_subdev_ctor(const struct nvkm_subdev_func *func, struct nvkm_device *device,
+		 enum nvkm_subdev_type type, int inst, struct nvkm_subdev *subdev)
+{
+	__nvkm_subdev_ctor(func, device, type, inst, subdev);
+	mutex_init(&subdev->use.mutex);
+}
+
 void nvkm_subdev_disable(struct nvkm_device *, enum nvkm_subdev_type, int inst);
 void nvkm_subdev_del(struct nvkm_subdev **);
+int  nvkm_subdev_ref(struct nvkm_subdev *);
+void nvkm_subdev_unref(struct nvkm_subdev *);
 int  nvkm_subdev_preinit(struct nvkm_subdev *);
+int  nvkm_subdev_oneinit(struct nvkm_subdev *);
 int  nvkm_subdev_init(struct nvkm_subdev *);
 int  nvkm_subdev_fini(struct nvkm_subdev *, bool suspend);
 int  nvkm_subdev_info(struct nvkm_subdev *, u64, u64 *);

drivers/gpu/drm/nouveau/nvkm/core/engine.c

@@ -39,12 +39,9 @@ void
 nvkm_engine_unref(struct nvkm_engine **pengine)
 {
 	struct nvkm_engine *engine = *pengine;
+
 	if (engine) {
-		if (refcount_dec_and_mutex_lock(&engine->use.refcount, &engine->use.mutex)) {
-			nvkm_subdev_fini(&engine->subdev, false);
-			engine->use.enabled = false;
-			mutex_unlock(&engine->use.mutex);
-		}
+		nvkm_subdev_unref(&engine->subdev);
 		*pengine = NULL;
 	}
 }
@@ -53,21 +50,13 @@ struct nvkm_engine *
 nvkm_engine_ref(struct nvkm_engine *engine)
 {
 	int ret;
+
 	if (engine) {
-		if (!refcount_inc_not_zero(&engine->use.refcount)) {
-			mutex_lock(&engine->use.mutex);
-			if (!refcount_inc_not_zero(&engine->use.refcount)) {
-				engine->use.enabled = true;
-				if ((ret = nvkm_subdev_init(&engine->subdev))) {
-					engine->use.enabled = false;
-					mutex_unlock(&engine->use.mutex);
+		ret = nvkm_subdev_ref(&engine->subdev);
+		if (ret)
 			return ERR_PTR(ret);
 	}
-				refcount_set(&engine->use.refcount, 1);
-			}
-			mutex_unlock(&engine->use.mutex);
-		}
-	}
+
 	return engine;
 }
 
@@ -117,26 +106,6 @@ nvkm_engine_init(struct nvkm_subdev *subdev)
 	struct nvkm_engine *engine = nvkm_engine(subdev);
 	struct nvkm_fb *fb = subdev->device->fb;
 	int ret = 0, i;
-	s64 time;
-
-	if (!engine->use.enabled) {
-		nvkm_trace(subdev, "init skipped, engine has no users\n");
-		return ret;
-	}
-
-	if (engine->func->oneinit && !engine->subdev.oneinit) {
-		nvkm_trace(subdev, "one-time init running...\n");
-		time = ktime_to_us(ktime_get());
-		ret = engine->func->oneinit(engine);
-		if (ret) {
-			nvkm_trace(subdev, "one-time init failed, %d\n", ret);
-			return ret;
-		}
-
-		engine->subdev.oneinit = true;
-		time = ktime_to_us(ktime_get()) - time;
-		nvkm_trace(subdev, "one-time init completed in %lldus\n", time);
-	}
 
 	if (engine->func->init)
 		ret = engine->func->init(engine);
@@ -146,6 +115,17 @@ nvkm_engine_init(struct nvkm_subdev *subdev)
 	return ret;
 }
 
+static int
+nvkm_engine_oneinit(struct nvkm_subdev *subdev)
+{
+	struct nvkm_engine *engine = nvkm_engine(subdev);
+
+	if (engine->func->oneinit)
+		return engine->func->oneinit(engine);
+
+	return 0;
+}
+
 static int
 nvkm_engine_preinit(struct nvkm_subdev *subdev)
 {
@@ -161,7 +141,6 @@ nvkm_engine_dtor(struct nvkm_subdev *subdev)
 	struct nvkm_engine *engine = nvkm_engine(subdev);
 	if (engine->func->dtor)
 		return engine->func->dtor(engine);
-	mutex_destroy(&engine->use.mutex);
 	return engine;
 }
 
@@ -169,6 +148,7 @@ const struct nvkm_subdev_func
 nvkm_engine = {
 	.dtor = nvkm_engine_dtor,
 	.preinit = nvkm_engine_preinit,
+	.oneinit = nvkm_engine_oneinit,
 	.init = nvkm_engine_init,
 	.fini = nvkm_engine_fini,
 	.info = nvkm_engine_info,
@@ -179,10 +159,9 @@ int
 nvkm_engine_ctor(const struct nvkm_engine_func *func, struct nvkm_device *device,
 		 enum nvkm_subdev_type type, int inst, bool enable, struct nvkm_engine *engine)
 {
-	nvkm_subdev_ctor(&nvkm_engine, device, type, inst, &engine->subdev);
 	engine->func = func;
-	refcount_set(&engine->use.refcount, 0);
-	mutex_init(&engine->use.mutex);
+	nvkm_subdev_ctor(&nvkm_engine, device, type, inst, &engine->subdev);
+	refcount_set(&engine->subdev.use.refcount, 0);
 
 	if (!nvkm_boolopt(device->cfgopt, engine->subdev.name, enable)) {
 		nvkm_debug(&engine->subdev, "disabled\n");

drivers/gpu/drm/nouveau/nvkm/core/subdev.c

@@ -54,7 +54,7 @@ int
 nvkm_subdev_fini(struct nvkm_subdev *subdev, bool suspend)
 {
 	struct nvkm_device *device = subdev->device;
-	const char *action = suspend ? "suspend" : "fini";
+	const char *action = suspend ? "suspend" : subdev->use.enabled ? "fini" : "reset";
 	s64 time;
 
 	nvkm_trace(subdev, "%s running...\n", action);
@@ -68,6 +68,7 @@ nvkm_subdev_fini(struct nvkm_subdev *subdev, bool suspend)
 				return ret;
 		}
 	}
+	subdev->use.enabled = false;
 
 	nvkm_mc_reset(device, subdev->type, subdev->inst);
 
@@ -97,17 +98,15 @@ nvkm_subdev_preinit(struct nvkm_subdev *subdev)
 	return 0;
 }
 
-int
-nvkm_subdev_init(struct nvkm_subdev *subdev)
+static int
+nvkm_subdev_oneinit_(struct nvkm_subdev *subdev)
 {
 	s64 time;
 	int ret;
 
-	nvkm_trace(subdev, "init running...\n");
-	time = ktime_to_us(ktime_get());
+	if (!subdev->func->oneinit || subdev->oneinit)
+		return 0;
 
-	if (subdev->func->oneinit && !subdev->oneinit) {
-		s64 time;
 	nvkm_trace(subdev, "one-time init running...\n");
 	time = ktime_to_us(ktime_get());
 	ret = subdev->func->oneinit(subdev);
@@ -119,8 +118,29 @@ nvkm_subdev_init(struct nvkm_subdev *subdev)
 	subdev->oneinit = true;
 	time = ktime_to_us(ktime_get()) - time;
 	nvkm_trace(subdev, "one-time init completed in %lldus\n", time);
+	return 0;
+}
+
+static int
+nvkm_subdev_init_(struct nvkm_subdev *subdev)
+{
+	s64 time;
+	int ret;
+
+	if (subdev->use.enabled) {
+		nvkm_trace(subdev, "init skipped, already running\n");
+		return 0;
 	}
 
+	nvkm_trace(subdev, "init running...\n");
+	time = ktime_to_us(ktime_get());
+
+	ret = nvkm_subdev_oneinit_(subdev);
+	if (ret)
+		return ret;
+
+	subdev->use.enabled = true;
+
 	if (subdev->func->init) {
 		ret = subdev->func->init(subdev);
 		if (ret) {
@@ -134,6 +154,64 @@ nvkm_subdev_init(struct nvkm_subdev *subdev)
 	return 0;
 }
 
+int
+nvkm_subdev_init(struct nvkm_subdev *subdev)
+{
+	int ret;
+
+	mutex_lock(&subdev->use.mutex);
+	if (refcount_read(&subdev->use.refcount) == 0) {
+		nvkm_trace(subdev, "init skipped, no users\n");
+		mutex_unlock(&subdev->use.mutex);
+		return 0;
+	}
+
+	ret = nvkm_subdev_init_(subdev);
+	mutex_unlock(&subdev->use.mutex);
+	return ret;
+}
+
+int
+nvkm_subdev_oneinit(struct nvkm_subdev *subdev)
+{
+	int ret;
+
+	mutex_lock(&subdev->use.mutex);
+	ret = nvkm_subdev_oneinit_(subdev);
+	mutex_unlock(&subdev->use.mutex);
+	return ret;
+}
+
+void
+nvkm_subdev_unref(struct nvkm_subdev *subdev)
+{
+	if (refcount_dec_and_mutex_lock(&subdev->use.refcount, &subdev->use.mutex)) {
+		nvkm_subdev_fini(subdev, false);
+		mutex_unlock(&subdev->use.mutex);
+	}
+}
+
+int
+nvkm_subdev_ref(struct nvkm_subdev *subdev)
+{
+	int ret;
+
+	if (subdev && !refcount_inc_not_zero(&subdev->use.refcount)) {
+		mutex_lock(&subdev->use.mutex);
+		if (!refcount_inc_not_zero(&subdev->use.refcount)) {
+			if ((ret = nvkm_subdev_init_(subdev))) {
+				mutex_unlock(&subdev->use.mutex);
+				return ret;
+			}
+
+			refcount_set(&subdev->use.refcount, 1);
+		}
+		mutex_unlock(&subdev->use.mutex);
+	}
+
+	return 0;
+}
+
 void
 nvkm_subdev_del(struct nvkm_subdev **psubdev)
 {
@@ -146,6 +224,7 @@ nvkm_subdev_del(struct nvkm_subdev **psubdev)
 		list_del(&subdev->head);
 		if (subdev->func->dtor)
 			*psubdev = subdev->func->dtor(subdev);
+		mutex_destroy(&subdev->use.mutex);
 		time = ktime_to_us(ktime_get()) - time;
 		nvkm_trace(subdev, "destroy completed in %lldus\n", time);
 		kfree(*psubdev);
@@ -167,7 +246,7 @@ nvkm_subdev_disable(struct nvkm_device *device, enum nvkm_subdev_type type, int
 }
 
 void
-nvkm_subdev_ctor(const struct nvkm_subdev_func *func, struct nvkm_device *device,
+__nvkm_subdev_ctor(const struct nvkm_subdev_func *func, struct nvkm_device *device,
 		   enum nvkm_subdev_type type, int inst, struct nvkm_subdev *subdev)
 {
 	subdev->func = func;
@@ -180,6 +259,8 @@ nvkm_subdev_ctor(const struct nvkm_subdev_func *func, struct nvkm_device *device
 	else
 		strscpy(subdev->name, nvkm_subdev_type[type], sizeof(subdev->name));
 	subdev->debug = nvkm_dbgopt(device->dbgopt, subdev->name);
+
+	refcount_set(&subdev->use.refcount, 1);
 	list_add_tail(&subdev->head, &device->subdev);
 }