Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 7431105b authored by Dave Airlie's avatar Dave Airlie
Browse files

Merge branch 'msm-next' of git://people.freedesktop.org/~robclark/linux into drm-next 

Compared to original fixes pull req that I sent yesterday, this adds
one more fix that I found for a synchronization issue which starts to
crop up when we use XA in DDX for 2d accel on 3d core.  In particular,
accelerating presentation blit triggers this problem.

* 'msm-next' of git://people.freedesktop.org/~robclark/linux:
  drm/msm: bigger synchronization hammer
  drm/msm: fix deadlock in bo create fail path
  drm/msm/mdp4: cursor fixes
  drm/msm/mdp4: pageflip fixes
  drm/msm/mdp5: fix ref leaks in error paths
  drm/msm: fix inconsequential typo
parents 9b5a3c0a c2703b13

drivers/gpu/drm/msm/mdp/mdp4/mdp4_crtc.c

+116 −63
Original line number Diff line number Diff line
@@ -39,6 +39,7 @@ struct mdp4_crtc { 
		spinlock_t lock;

		bool stale;

		uint32_t width, height;

		uint32_t x, y;



		/* next cursor to scan-out: */

		uint32_t next_iova;
@@ -57,9 +58,16 @@ struct mdp4_crtc { 
#define PENDING_FLIP   0x2

	atomic_t pending;



	/* the fb that we currently hold a scanout ref to: */

	/* the fb that we logically (from PoV of KMS API) hold a ref

	 * to.  Which we may not yet be scanning out (we may still

	 * be scanning out previous in case of page_flip while waiting

	 * for gpu rendering to complete:

	 */

	struct drm_framebuffer *fb;



	/* the fb that we currently hold a scanout ref to: */

	struct drm_framebuffer *scanout_fb;



	/* for unref'ing framebuffers after scanout completes: */

	struct drm_flip_work unref_fb_work;
@@ -77,24 +85,73 @@ static struct mdp4_kms *get_kms(struct drm_crtc *crtc) 
	return to_mdp4_kms(to_mdp_kms(priv->kms));

}



static void update_fb(struct drm_crtc *crtc, bool async,

		struct drm_framebuffer *new_fb)

static void request_pending(struct drm_crtc *crtc, uint32_t pending)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct drm_framebuffer *old_fb = mdp4_crtc->fb;



	if (old_fb)

		drm_flip_work_queue(&mdp4_crtc->unref_fb_work, old_fb);

	atomic_or(pending, &mdp4_crtc->pending);

	mdp_irq_register(&get_kms(crtc)->base, &mdp4_crtc->vblank);

}



static void crtc_flush(struct drm_crtc *crtc)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct mdp4_kms *mdp4_kms = get_kms(crtc);

	uint32_t i, flush = 0;



	for (i = 0; i < ARRAY_SIZE(mdp4_crtc->planes); i++) {

		struct drm_plane *plane = mdp4_crtc->planes[i];

		if (plane) {

			enum mdp4_pipe pipe_id = mdp4_plane_pipe(plane);

			flush |= pipe2flush(pipe_id);

		}

	}

	flush |= ovlp2flush(mdp4_crtc->ovlp);



	DBG("%s: flush=%08x", mdp4_crtc->name, flush);



	mdp4_write(mdp4_kms, REG_MDP4_OVERLAY_FLUSH, flush);

}



static void update_fb(struct drm_crtc *crtc, struct drm_framebuffer *new_fb)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct drm_framebuffer *old_fb = mdp4_crtc->fb;



	/* grab reference to incoming scanout fb: */

	drm_framebuffer_reference(new_fb);

	mdp4_crtc->base.fb = new_fb;

	mdp4_crtc->fb = new_fb;



	if (!async) {

		/* enable vblank to pick up the old_fb */

		mdp_irq_register(&get_kms(crtc)->base, &mdp4_crtc->vblank);

	if (old_fb)

		drm_flip_work_queue(&mdp4_crtc->unref_fb_work, old_fb);

}



/* unlike update_fb(), take a ref to the new scanout fb *before* updating

 * plane, then call this.  Needed to ensure we don't unref the buffer that

 * is actually still being scanned out.

 *

 * Note that this whole thing goes away with atomic.. since we can defer

 * calling into driver until rendering is done.

 */

static void update_scanout(struct drm_crtc *crtc, struct drm_framebuffer *fb)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);



	/* flush updates, to make sure hw is updated to new scanout fb,

	 * so that we can safely queue unref to current fb (ie. next

	 * vblank we know hw is done w/ previous scanout_fb).

	 */

	crtc_flush(crtc);



	if (mdp4_crtc->scanout_fb)

		drm_flip_work_queue(&mdp4_crtc->unref_fb_work,

				mdp4_crtc->scanout_fb);



	mdp4_crtc->scanout_fb = fb;



	/* enable vblank to complete flip: */

	request_pending(crtc, PENDING_FLIP);

}



/* if file!=NULL, this is preclose potential cancel-flip path */
@@ -120,34 +177,6 @@ static void complete_flip(struct drm_crtc *crtc, struct drm_file *file) 
	spin_unlock_irqrestore(&dev->event_lock, flags);

}



static void crtc_flush(struct drm_crtc *crtc)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct mdp4_kms *mdp4_kms = get_kms(crtc);

	uint32_t i, flush = 0;



	for (i = 0; i < ARRAY_SIZE(mdp4_crtc->planes); i++) {

		struct drm_plane *plane = mdp4_crtc->planes[i];

		if (plane) {

			enum mdp4_pipe pipe_id = mdp4_plane_pipe(plane);

			flush |= pipe2flush(pipe_id);

		}

	}

	flush |= ovlp2flush(mdp4_crtc->ovlp);



	DBG("%s: flush=%08x", mdp4_crtc->name, flush);



	mdp4_write(mdp4_kms, REG_MDP4_OVERLAY_FLUSH, flush);

}



static void request_pending(struct drm_crtc *crtc, uint32_t pending)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);



	atomic_or(pending, &mdp4_crtc->pending);

	mdp_irq_register(&get_kms(crtc)->base, &mdp4_crtc->vblank);

}



static void pageflip_cb(struct msm_fence_cb *cb)

{

	struct mdp4_crtc *mdp4_crtc =
@@ -158,11 +187,9 @@ static void pageflip_cb(struct msm_fence_cb *cb) 
	if (!fb)

		return;



	drm_framebuffer_reference(fb);

	mdp4_plane_set_scanout(mdp4_crtc->plane, fb);

	crtc_flush(crtc);



	/* enable vblank to complete flip: */

	request_pending(crtc, PENDING_FLIP);

	update_scanout(crtc, fb);

}



static void unref_fb_worker(struct drm_flip_work *work, void *val)
@@ -320,6 +347,20 @@ static int mdp4_crtc_mode_set(struct drm_crtc *crtc, 
			mode->vsync_end, mode->vtotal,

			mode->type, mode->flags);



	/* grab extra ref for update_scanout() */

	drm_framebuffer_reference(crtc->fb);



	ret = mdp4_plane_mode_set(mdp4_crtc->plane, crtc, crtc->fb,

			0, 0, mode->hdisplay, mode->vdisplay,

			x << 16, y << 16,

			mode->hdisplay << 16, mode->vdisplay << 16);

	if (ret) {

		drm_framebuffer_unreference(crtc->fb);

		dev_err(crtc->dev->dev, "%s: failed to set mode on plane: %d\n",

				mdp4_crtc->name, ret);

		return ret;

	}



	mdp4_write(mdp4_kms, REG_MDP4_DMA_SRC_SIZE(dma),

			MDP4_DMA_SRC_SIZE_WIDTH(mode->hdisplay) |

			MDP4_DMA_SRC_SIZE_HEIGHT(mode->vdisplay));
@@ -341,24 +382,15 @@ static int mdp4_crtc_mode_set(struct drm_crtc *crtc, 


	mdp4_write(mdp4_kms, REG_MDP4_OVLP_CFG(ovlp), 1);



	update_fb(crtc, false, crtc->fb);



	ret = mdp4_plane_mode_set(mdp4_crtc->plane, crtc, crtc->fb,

			0, 0, mode->hdisplay, mode->vdisplay,

			x << 16, y << 16,

			mode->hdisplay << 16, mode->vdisplay << 16);

	if (ret) {

		dev_err(crtc->dev->dev, "%s: failed to set mode on plane: %d\n",

				mdp4_crtc->name, ret);

		return ret;

	}



	if (dma == DMA_E) {

		mdp4_write(mdp4_kms, REG_MDP4_DMA_E_QUANT(0), 0x00ff0000);

		mdp4_write(mdp4_kms, REG_MDP4_DMA_E_QUANT(1), 0x00ff0000);

		mdp4_write(mdp4_kms, REG_MDP4_DMA_E_QUANT(2), 0x00ff0000);

	}



	update_fb(crtc, crtc->fb);

	update_scanout(crtc, crtc->fb);



	return 0;

}
@@ -385,13 +417,24 @@ static int mdp4_crtc_mode_set_base(struct drm_crtc *crtc, int x, int y, 
	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct drm_plane *plane = mdp4_crtc->plane;

	struct drm_display_mode *mode = &crtc->mode;

	int ret;



	update_fb(crtc, false, crtc->fb);

	/* grab extra ref for update_scanout() */

	drm_framebuffer_reference(crtc->fb);



	return mdp4_plane_mode_set(plane, crtc, crtc->fb,

	ret = mdp4_plane_mode_set(plane, crtc, crtc->fb,

			0, 0, mode->hdisplay, mode->vdisplay,

			x << 16, y << 16,

			mode->hdisplay << 16, mode->vdisplay << 16);

	if (ret) {

		drm_framebuffer_unreference(crtc->fb);

		return ret;

	}



	update_fb(crtc, crtc->fb);

	update_scanout(crtc, crtc->fb);



	return 0;

}



static void mdp4_crtc_load_lut(struct drm_crtc *crtc)
@@ -419,7 +462,7 @@ static int mdp4_crtc_page_flip(struct drm_crtc *crtc, 
	mdp4_crtc->event = event;

	spin_unlock_irqrestore(&dev->event_lock, flags);



	update_fb(crtc, true, new_fb);

	update_fb(crtc, new_fb);



	return msm_gem_queue_inactive_cb(obj, &mdp4_crtc->pageflip_cb);

}
@@ -442,12 +485,12 @@ static int mdp4_crtc_set_property(struct drm_crtc *crtc, 
static void update_cursor(struct drm_crtc *crtc)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct mdp4_kms *mdp4_kms = get_kms(crtc);

	enum mdp4_dma dma = mdp4_crtc->dma;

	unsigned long flags;



	spin_lock_irqsave(&mdp4_crtc->cursor.lock, flags);

	if (mdp4_crtc->cursor.stale) {

		struct mdp4_kms *mdp4_kms = get_kms(crtc);

		struct drm_gem_object *next_bo = mdp4_crtc->cursor.next_bo;

		struct drm_gem_object *prev_bo = mdp4_crtc->cursor.scanout_bo;

		uint32_t iova = mdp4_crtc->cursor.next_iova;
@@ -479,6 +522,11 @@ static void update_cursor(struct drm_crtc *crtc) 
		mdp4_crtc->cursor.scanout_bo = next_bo;

		mdp4_crtc->cursor.stale = false;

	}



	mdp4_write(mdp4_kms, REG_MDP4_DMA_CURSOR_POS(dma),

			MDP4_DMA_CURSOR_POS_X(mdp4_crtc->cursor.x) |

			MDP4_DMA_CURSOR_POS_Y(mdp4_crtc->cursor.y));



	spin_unlock_irqrestore(&mdp4_crtc->cursor.lock, flags);

}
@@ -530,6 +578,7 @@ static int mdp4_crtc_cursor_set(struct drm_crtc *crtc, 
		drm_gem_object_unreference_unlocked(old_bo);

	}



	crtc_flush(crtc);

	request_pending(crtc, PENDING_CURSOR);



	return 0;
@@ -542,12 +591,15 @@ static int mdp4_crtc_cursor_set(struct drm_crtc *crtc, 
static int mdp4_crtc_cursor_move(struct drm_crtc *crtc, int x, int y)

{

	struct mdp4_crtc *mdp4_crtc = to_mdp4_crtc(crtc);

	struct mdp4_kms *mdp4_kms = get_kms(crtc);

	enum mdp4_dma dma = mdp4_crtc->dma;

	unsigned long flags;



	mdp4_write(mdp4_kms, REG_MDP4_DMA_CURSOR_POS(dma),

			MDP4_DMA_CURSOR_POS_X(x) |

			MDP4_DMA_CURSOR_POS_Y(y));

	spin_lock_irqsave(&mdp4_crtc->cursor.lock, flags);

	mdp4_crtc->cursor.x = x;

	mdp4_crtc->cursor.y = y;

	spin_unlock_irqrestore(&mdp4_crtc->cursor.lock, flags);



	crtc_flush(crtc);

	request_pending(crtc, PENDING_CURSOR);



	return 0;

}
@@ -713,6 +765,7 @@ struct drm_crtc *mdp4_crtc_init(struct drm_device *dev, 
	crtc = &mdp4_crtc->base;



	mdp4_crtc->plane = plane;

	mdp4_crtc->id = id;



	mdp4_crtc->ovlp = ovlp_id;

	mdp4_crtc->dma = dma_id;

drivers/gpu/drm/msm/mdp/mdp4/mdp4_plane.c

+2 −2
Original line number Diff line number Diff line
@@ -170,8 +170,8 @@ int mdp4_plane_mode_set(struct drm_plane *plane, 
			MDP4_PIPE_DST_SIZE_HEIGHT(crtc_h));



	mdp4_write(mdp4_kms, REG_MDP4_PIPE_DST_XY(pipe),

			MDP4_PIPE_SRC_XY_X(crtc_x) |

			MDP4_PIPE_SRC_XY_Y(crtc_y));

			MDP4_PIPE_DST_XY_X(crtc_x) |

			MDP4_PIPE_DST_XY_Y(crtc_y));



	mdp4_plane_set_scanout(plane, fb);

drivers/gpu/drm/msm/mdp/mdp5/mdp5_crtc.c

+6 −1
Original line number Diff line number Diff line
@@ -296,6 +296,7 @@ static int mdp5_crtc_mode_set(struct drm_crtc *crtc, 
			x << 16, y << 16,

			mode->hdisplay << 16, mode->vdisplay << 16);

	if (ret) {

		drm_framebuffer_unreference(crtc->fb);

		dev_err(crtc->dev->dev, "%s: failed to set mode on plane: %d\n",

				mdp5_crtc->name, ret);

		return ret;
@@ -343,11 +344,15 @@ static int mdp5_crtc_mode_set_base(struct drm_crtc *crtc, int x, int y, 
			0, 0, mode->hdisplay, mode->vdisplay,

			x << 16, y << 16,

			mode->hdisplay << 16, mode->vdisplay << 16);

	if (ret) {

		drm_framebuffer_unreference(crtc->fb);

		return ret;

	}



	update_fb(crtc, crtc->fb);

	update_scanout(crtc, crtc->fb);



	return ret;

	return 0;

}



static void mdp5_crtc_load_lut(struct drm_crtc *crtc)

drivers/gpu/drm/msm/msm_gem.c

+1 −1
Original line number Diff line number Diff line
@@ -644,7 +644,7 @@ struct drm_gem_object *msm_gem_new(struct drm_device *dev, 


fail:

	if (obj)

		drm_gem_object_unreference_unlocked(obj);

		drm_gem_object_unreference(obj);



	return ERR_PTR(ret);

}

drivers/gpu/drm/msm/msm_gem_submit.c

+5 −4
Original line number Diff line number Diff line
@@ -163,7 +163,7 @@ static int submit_validate_objects(struct msm_gem_submit *submit) 




		/* if locking succeeded, pin bo: */

		ret = msm_gem_get_iova(&msm_obj->base,

		ret = msm_gem_get_iova_locked(&msm_obj->base,

				submit->gpu->id, &iova);



		/* this would break the logic in the fail path.. there is no
@@ -247,7 +247,7 @@ static int submit_reloc(struct msm_gem_submit *submit, struct msm_gem_object *ob 
	/* For now, just map the entire thing.  Eventually we probably

	 * to do it page-by-page, w/ kmap() if not vmap()d..

	 */

	ptr = msm_gem_vaddr(&obj->base);

	ptr = msm_gem_vaddr_locked(&obj->base);



	if (IS_ERR(ptr)) {

		ret = PTR_ERR(ptr);
@@ -307,14 +307,12 @@ static void submit_cleanup(struct msm_gem_submit *submit, bool fail) 
{

	unsigned i;



	mutex_lock(&submit->dev->struct_mutex);

	for (i = 0; i < submit->nr_bos; i++) {

		struct msm_gem_object *msm_obj = submit->bos[i].obj;

		submit_unlock_unpin_bo(submit, i);

		list_del_init(&msm_obj->submit_entry);

		drm_gem_object_unreference(&msm_obj->base);

	}

	mutex_unlock(&submit->dev->struct_mutex);



	ww_acquire_fini(&submit->ticket);

	kfree(submit);
@@ -342,6 +340,8 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, 
	if (args->nr_cmds > MAX_CMDS)

		return -EINVAL;



	mutex_lock(&dev->struct_mutex);



	submit = submit_create(dev, gpu, args->nr_bos);

	if (!submit) {

		ret = -ENOMEM;
@@ -410,5 +410,6 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, 
out:

	if (submit)

		submit_cleanup(submit, !!ret);

	mutex_unlock(&dev->struct_mutex);

	return ret;

}