drm/msm: remove fence_cbs

This was only used for atomic commit these days.  So instead just give
atomic it's own work-queue where we can do a block on each bo in turn.
Simplifies things a whole bunch and makes the 'struct fence' conversion
easier.
Signed-off-by: Rob Clark <robdclark@gmail.com>

drivers/gpu/drm/msm/msm_atomic.c

@@ -18,18 +18,16 @@
 #include "msm_drv.h"
 #include "msm_kms.h"
 #include "msm_gem.h"
-#include "msm_gpu.h"   /* temporary */
 #include "msm_fence.h"
 
 struct msm_commit {
 	struct drm_device *dev;
 	struct drm_atomic_state *state;
-	uint32_t fence;
-	struct msm_fence_cb fence_cb;
+	struct work_struct work;
 	uint32_t crtc_mask;
 };
 
-static void fence_cb(struct msm_fence_cb *cb);
+static void commit_worker(struct work_struct *work);
 
 /* block until specified crtcs are no longer pending update, and
  * atomically mark them as pending update
@@ -71,11 +69,7 @@ static struct msm_commit *commit_init(struct drm_atomic_state *state)
 	c->dev = state->dev;
 	c->state = state;
 
-	/* TODO we might need a way to indicate to run the cb on a
-	 * different wq so wait_for_vblanks() doesn't block retiring
-	 * bo's..
-	 */
-	INIT_FENCE_CB(&c->fence_cb, fence_cb);
+	INIT_WORK(&c->work, commit_worker);
 
 	return c;
 }
@@ -113,16 +107,39 @@ static void msm_atomic_wait_for_commit_done(struct drm_device *dev,
 	}
 }
 
+static void wait_fences(struct msm_commit *c, bool async)
+{
+	int nplanes = c->dev->mode_config.num_total_plane;
+	ktime_t timeout = ktime_add_ms(ktime_get(), 1000);
+	int i;
+
+	for (i = 0; i < nplanes; i++) {
+		struct drm_plane *plane = c->state->planes[i];
+		struct drm_plane_state *new_state = c->state->plane_states[i];
+
+		if (!plane)
+			continue;
+
+		if ((plane->state->fb != new_state->fb) && new_state->fb) {
+			struct drm_gem_object *obj =
+				msm_framebuffer_bo(new_state->fb, 0);
+			msm_gem_cpu_sync(obj, MSM_PREP_READ, &timeout);
+		}
+	}
+}
+
 /* The (potentially) asynchronous part of the commit.  At this point
  * nothing can fail short of armageddon.
  */
-static void complete_commit(struct msm_commit *c)
+static void complete_commit(struct msm_commit *c, bool async)
 {
 	struct drm_atomic_state *state = c->state;
 	struct drm_device *dev = state->dev;
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_kms *kms = priv->kms;
 
+	wait_fences(c, async);
+
 	kms->funcs->prepare_commit(kms, state);
 
 	drm_atomic_helper_commit_modeset_disables(dev, state);
@@ -155,17 +172,9 @@ static void complete_commit(struct msm_commit *c)
 	commit_destroy(c);
 }
 
-static void fence_cb(struct msm_fence_cb *cb)
+static void commit_worker(struct work_struct *work)
 {
-	struct msm_commit *c =
-			container_of(cb, struct msm_commit, fence_cb);
-	complete_commit(c);
-}
-
-static void add_fb(struct msm_commit *c, struct drm_framebuffer *fb)
-{
-	struct drm_gem_object *obj = msm_framebuffer_bo(fb, 0);
-	c->fence = max(c->fence, msm_gem_fence(to_msm_bo(obj), MSM_PREP_READ));
+	complete_commit(container_of(work, struct msm_commit, work), true);
 }
 
 int msm_atomic_check(struct drm_device *dev,
@@ -204,9 +213,7 @@ int msm_atomic_commit(struct drm_device *dev,
 		struct drm_atomic_state *state, bool nonblock)
 {
 	struct msm_drm_private *priv = dev->dev_private;
-	int nplanes = dev->mode_config.num_total_plane;
 	int ncrtcs = dev->mode_config.num_crtc;
-	ktime_t timeout;
 	struct msm_commit *c;
 	int i, ret;
 
@@ -230,20 +237,6 @@ int msm_atomic_commit(struct drm_device *dev,
 		c->crtc_mask |= (1 << drm_crtc_index(crtc));
 	}
 
-	/*
-	 * Figure out what fence to wait for:
-	 */
-	for (i = 0; i < nplanes; i++) {
-		struct drm_plane *plane = state->planes[i];
-		struct drm_plane_state *new_state = state->plane_states[i];
-
-		if (!plane)
-			continue;
-
-		if ((plane->state->fb != new_state->fb) && new_state->fb)
-			add_fb(c, new_state->fb);
-	}
-
 	/*
 	 * Wait for pending updates on any of the same crtc's and then
 	 * mark our set of crtc's as busy:
@@ -278,18 +271,12 @@ int msm_atomic_commit(struct drm_device *dev,
 	 * current layout.
 	 */
 
-	if (nonblock && priv->gpu) {
-		msm_queue_fence_cb(priv->gpu->fctx, &c->fence_cb, c->fence);
+	if (nonblock) {
+		queue_work(priv->atomic_wq, &c->work);
 		return 0;
 	}
 
-	timeout = ktime_add_ms(ktime_get(), 1000);
-
-	/* uninterruptible wait */
-	if (priv->gpu)
-		msm_wait_fence(priv->gpu->fctx, c->fence, &timeout, false);
-
-	complete_commit(c);
+	complete_commit(c, false);
 
 	return 0;
 

drivers/gpu/drm/msm/msm_drv.c

@@ -213,6 +213,9 @@ static int msm_unload(struct drm_device *dev)
 	flush_workqueue(priv->wq);
 	destroy_workqueue(priv->wq);
 
+	flush_workqueue(priv->atomic_wq);
+	destroy_workqueue(priv->atomic_wq);
+
 	if (kms) {
 		pm_runtime_disable(dev->dev);
 		kms->funcs->destroy(kms);
@@ -339,6 +342,7 @@ static int msm_load(struct drm_device *dev, unsigned long flags)
 	dev->dev_private = priv;
 
 	priv->wq = alloc_ordered_workqueue("msm", 0);
+	priv->atomic_wq = alloc_ordered_workqueue("msm:atomic", 0);
 	init_waitqueue_head(&priv->pending_crtcs_event);
 
 	INIT_LIST_HEAD(&priv->inactive_list);

drivers/gpu/drm/msm/msm_drv.h

@@ -109,6 +109,7 @@ struct msm_drm_private {
 	struct list_head inactive_list;
 
 	struct workqueue_struct *wq;
+	struct workqueue_struct *atomic_wq;
 
 	/* crtcs pending async atomic updates: */
 	uint32_t pending_crtcs;
@@ -192,8 +193,8 @@ void *msm_gem_vaddr(struct drm_gem_object *obj);
 void msm_gem_move_to_active(struct drm_gem_object *obj,
 		struct msm_gpu *gpu, bool write, uint32_t fence);
 void msm_gem_move_to_inactive(struct drm_gem_object *obj);
-int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op,
-		ktime_t *timeout);
+int msm_gem_cpu_sync(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout);
+int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout);
 int msm_gem_cpu_fini(struct drm_gem_object *obj);
 void msm_gem_free_object(struct drm_gem_object *obj);
 int msm_gem_new_handle(struct drm_device *dev, struct drm_file *file,

drivers/gpu/drm/msm/msm_fence.c

@@ -33,7 +33,6 @@ msm_fence_context_alloc(struct drm_device *dev, const char *name)
 	fctx->dev = dev;
 	fctx->name = name;
 	init_waitqueue_head(&fctx->event);
-	INIT_LIST_HEAD(&fctx->fence_cbs);
 
 	return fctx;
 }
@@ -86,54 +85,12 @@ int msm_wait_fence(struct msm_fence_context *fctx, uint32_t fence,
 	return ret;
 }
 
-int msm_queue_fence_cb(struct msm_fence_context *fctx,
-		struct msm_fence_cb *cb, uint32_t fence)
-{
-	struct msm_drm_private *priv = fctx->dev->dev_private;
-	int ret = 0;
-
-	mutex_lock(&fctx->dev->struct_mutex);
-	if (!list_empty(&cb->work.entry)) {
-		ret = -EINVAL;
-	} else if (fence > fctx->completed_fence) {
-		cb->fence = fence;
-		list_add_tail(&cb->work.entry, &fctx->fence_cbs);
-	} else {
-		queue_work(priv->wq, &cb->work);
-	}
-	mutex_unlock(&fctx->dev->struct_mutex);
-
-	return ret;
-}
-
 /* called from workqueue */
 void msm_update_fence(struct msm_fence_context *fctx, uint32_t fence)
 {
-	struct msm_drm_private *priv = fctx->dev->dev_private;
-
 	mutex_lock(&fctx->dev->struct_mutex);
 	fctx->completed_fence = max(fence, fctx->completed_fence);
-
-	while (!list_empty(&fctx->fence_cbs)) {
-		struct msm_fence_cb *cb;
-
-		cb = list_first_entry(&fctx->fence_cbs,
-				struct msm_fence_cb, work.entry);
-
-		if (cb->fence > fctx->completed_fence)
-			break;
-
-		list_del_init(&cb->work.entry);
-		queue_work(priv->wq, &cb->work);
-	}
-
 	mutex_unlock(&fctx->dev->struct_mutex);
 
 	wake_up_all(&fctx->event);
 }
-
-void __msm_fence_worker(struct work_struct *work)
-{
-	struct msm_fence_cb *cb = container_of(work, struct msm_fence_cb, work);
-	cb->func(cb);
-}

drivers/gpu/drm/msm/msm_fence.h

@@ -27,28 +27,12 @@ struct msm_fence_context {
 	uint32_t last_fence;          /* last assigned fence */
 	uint32_t completed_fence;     /* last completed fence */
 	wait_queue_head_t event;
-	/* callbacks deferred until bo is inactive: */
-	struct list_head fence_cbs;
 };
 
 struct msm_fence_context * msm_fence_context_alloc(struct drm_device *dev,
 		const char *name);
 void msm_fence_context_free(struct msm_fence_context *fctx);
 
-/* callback from wq once fence has passed: */
-struct msm_fence_cb {
-	struct work_struct work;
-	uint32_t fence;
-	void (*func)(struct msm_fence_cb *cb);
-};
-
-void __msm_fence_worker(struct work_struct *work);
-
-#define INIT_FENCE_CB(_cb, _func)  do {                     \
-		INIT_WORK(&(_cb)->work, __msm_fence_worker); \
-		(_cb)->func = _func;                         \
-	} while (0)
-
 int msm_wait_fence(struct msm_fence_context *fctx, uint32_t fence,
 		ktime_t *timeout, bool interruptible);
 int msm_queue_fence_cb(struct msm_fence_context *fctx,

drivers/gpu/drm/msm/msm_gem.c

@@ -439,7 +439,7 @@ void msm_gem_move_to_inactive(struct drm_gem_object *obj)
 	list_add_tail(&msm_obj->mm_list, &priv->inactive_list);
 }
 
-int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout)
+int msm_gem_cpu_sync(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout)
 {
 	struct drm_device *dev = obj->dev;
 	struct msm_drm_private *priv = dev->dev_private;
@@ -456,6 +456,13 @@ int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout)
 			ret = msm_wait_fence(priv->gpu->fctx, fence, timeout, true);
 	}
 
+	return ret;
+}
+
+int msm_gem_cpu_prep(struct drm_gem_object *obj, uint32_t op, ktime_t *timeout)
+{
+	int ret = msm_gem_cpu_sync(obj, op, timeout);
+
 	/* TODO cache maintenance */
 
 	return ret;