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
This commit is contained in:
Dave Airlie 2014-02-11 12:56:17 +10:00
commit 7431105b14
6 changed files with 130 additions and 74 deletions

View File

@ -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);
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;
if (old_fb)
drm_flip_work_queue(&mdp4_crtc->unref_fb_work, old_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 @@ fail:
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;

View File

@ -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);

View File

@ -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)

View File

@ -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);
}

View File

@ -163,7 +163,7 @@ retry:
/* 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;
}

View File

@ -298,8 +298,6 @@ int msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
struct msm_drm_private *priv = dev->dev_private;
int i, ret;
mutex_lock(&dev->struct_mutex);
submit->fence = ++priv->next_fence;
gpu->submitted_fence = submit->fence;
@ -331,7 +329,6 @@ int msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
msm_gem_move_to_active(&msm_obj->base, gpu, true, submit->fence);
}
hangcheck_timer_reset(gpu);
mutex_unlock(&dev->struct_mutex);
return ret;
}