Hi Gustavo,
On Thu, Oct 20, 2016 at 06:30:17PM -0200, Gustavo Padovan wrote:
Hi Brian,
2016-10-20 Brian Starkey <brian.star...@arm.com>:
Hi Gustavo,
I notice your branch has the sync_file refcount change in, but this
doesn't seem to take account for that. Will you be dropping that
change to match the semantics of fence_array?
I will drop the fence_get() in the out-fence patch because we already
hold the ref when we create the fence. The sync_file refcount patch will
remain.
OK, I'll work on that basis, thanks.
Couple more comments below.
On Thu, Oct 20, 2016 at 12:50:05PM -0200, Gustavo Padovan wrote:
> From: Gustavo Padovan <gustavo.pado...@collabora.co.uk>
>
> Support DRM out-fences by creating a sync_file with a fence for each CRTC
> that sets the OUT_FENCE_PTR property.
>
> We use the out_fence pointer received in the OUT_FENCE_PTR prop to send
> the sync_file fd back to userspace.
>
> The sync_file and fd are allocated/created before commit, but the
> fd_install operation only happens after we know that commit succeed.
>
> In case of error userspace should receive a fd number of -1.
>
> v2: Comment by Rob Clark:
> - Squash commit that adds DRM_MODE_ATOMIC_OUT_FENCE flag here.
>
> Comment by Daniel Vetter:
> - Add clean up code for out_fences
>
> v3: Comments by Daniel Vetter:
> - create DRM_MODE_ATOMIC_EVENT_MASK
> - userspace should fill out_fences_ptr with the crtc_ids for which
> it wants fences back.
>
> v4: Create OUT_FENCE_PTR properties and remove old approach.
>
> Signed-off-by: Gustavo Padovan <gustavo.pado...@collabora.co.uk>
> ---
> drivers/gpu/drm/drm_atomic.c | 116 ++++++++++++++++++++++++++++++++++---------
> drivers/gpu/drm/drm_crtc.c | 8 +++
> include/drm/drm_crtc.h | 19 +++++++
> 3 files changed, 119 insertions(+), 24 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
> index b3284b2..07775fc 100644
> --- a/drivers/gpu/drm/drm_atomic.c
> +++ b/drivers/gpu/drm/drm_atomic.c
> @@ -490,6 +490,8 @@ int drm_atomic_crtc_set_property(struct drm_crtc *crtc,
> &replaced);
> state->color_mgmt_changed |= replaced;
> return ret;
> + } else if (property == config->prop_out_fence_ptr) {
> + state->out_fence_ptr = u64_to_user_ptr(val);
> } else if (crtc->funcs->atomic_set_property)
> return crtc->funcs->atomic_set_property(crtc, state, property,
val);
> else
> @@ -532,6 +534,8 @@ drm_atomic_crtc_get_property(struct drm_crtc *crtc,
> *val = (state->ctm) ? state->ctm->base.id : 0;
> else if (property == config->gamma_lut_property)
> *val = (state->gamma_lut) ? state->gamma_lut->base.id : 0;
> + else if (property == config->prop_out_fence_ptr)
> + *val = 0;
> else if (crtc->funcs->atomic_get_property)
> return crtc->funcs->atomic_get_property(crtc, state, property,
val);
> else
> @@ -1474,11 +1478,9 @@ EXPORT_SYMBOL(drm_atomic_nonblocking_commit);
> */
>
> static struct drm_pending_vblank_event *create_vblank_event(
> - struct drm_device *dev, struct drm_file *file_priv,
> - struct fence *fence, uint64_t user_data)
> + struct drm_device *dev, uint64_t user_data)
> {
> struct drm_pending_vblank_event *e = NULL;
> - int ret;
>
> e = kzalloc(sizeof *e, GFP_KERNEL);
> if (!e)
> @@ -1488,17 +1490,6 @@ static struct drm_pending_vblank_event
*create_vblank_event(
> e->event.base.length = sizeof(e->event);
> e->event.user_data = user_data;
>
> - if (file_priv) {
> - ret = drm_event_reserve_init(dev, file_priv, &e->base,
> - &e->event.base);
> - if (ret) {
> - kfree(e);
> - return NULL;
> - }
> - }
> -
> - e->base.fence = fence;
> -
> return e;
> }
>
> @@ -1603,6 +1594,40 @@ void drm_atomic_clean_old_fb(struct drm_device *dev,
> }
> EXPORT_SYMBOL(drm_atomic_clean_old_fb);
>
> +static int crtc_setup_out_fence(struct drm_crtc *crtc,
> + struct drm_crtc_state *crtc_state,
> + struct drm_out_fence_state *fence_state)
> +{
> + struct fence *fence;
> +
> + fence_state->fd = get_unused_fd_flags(O_CLOEXEC);
> + if (fence_state->fd < 0) {
> + return fence_state->fd;
> + }
> +
> + fence_state->out_fence_ptr = crtc_state->out_fence_ptr;
> + crtc_state->out_fence_ptr = NULL;
> +
> + if (put_user(fence_state->fd, fence_state->out_fence_ptr))
> + return -EFAULT;
> +
> + fence = kzalloc(sizeof(*fence), GFP_KERNEL);
> + if (!fence)
> + return -ENOMEM;
> +
> + fence_init(fence, &drm_crtc_fence_ops, &crtc->fence_lock,
> + crtc->fence_context, ++crtc->fence_seqno);
> +
> + fence_state->sync_file = sync_file_create(fence);
> + if(!fence_state->sync_file) {
> + fence_put(fence);
> + return -ENOMEM;
> + }
> +
> + crtc_state->event->base.fence = fence_get(fence);
> + return 0;
> +}
> +
> int drm_mode_atomic_ioctl(struct drm_device *dev,
> void *data, struct drm_file *file_priv)
> {
> @@ -1617,9 +1642,10 @@ int drm_mode_atomic_ioctl(struct drm_device *dev,
> struct drm_plane *plane;
> struct drm_crtc *crtc;
> struct drm_crtc_state *crtc_state;
> + struct drm_out_fence_state *fence_state;
I just hit another crash - looks like fence_state needs to be
initialised to NULL here, otherwise if property lookup/set fails we
kfree() a bad pointer.
> unsigned plane_mask;
> int ret = 0;
> - unsigned int i, j;
> + unsigned int i, j, fence_idx = 0;
>
> /* disallow for drivers not supporting atomic: */
> if (!drm_core_check_feature(dev, DRIVER_ATOMIC))
> @@ -1734,12 +1760,19 @@ retry:
> drm_mode_object_unreference(obj);
> }
>
> - if (arg->flags & DRM_MODE_PAGE_FLIP_EVENT) {
> - for_each_crtc_in_state(state, crtc, crtc_state, i) {
> + fence_state = kcalloc(dev->mode_config.num_crtc, sizeof(*fence_state),
> + GFP_KERNEL);
> + if (!fence_state) {
> + ret = -ENOMEM;
> + goto out;
> + }
> +
> + for_each_crtc_in_state(state, crtc, crtc_state, i) {
> + if (arg->flags & DRM_MODE_PAGE_FLIP_EVENT ||
> + crtc_state->out_fence_ptr) {
> struct drm_pending_vblank_event *e;
>
> - e = create_vblank_event(dev, file_priv, NULL,
> - arg->user_data);
> + e = create_vblank_event(dev, arg->user_data);
> if (!e) {
> ret = -ENOMEM;
> goto out;
> @@ -1747,6 +1780,28 @@ retry:
>
> crtc_state->event = e;
> }
> +
> + if (arg->flags & DRM_MODE_PAGE_FLIP_EVENT) {
> + struct drm_pending_vblank_event *e = crtc_state->event;
> +
> + if (!file_priv)
> + continue;
> +
> + ret = drm_event_reserve_init(dev, file_priv, &e->base,
> + &e->event.base);
> + if (ret) {
> + kfree(e);
> + crtc_state->event = NULL;
> + goto out;
> + }
> + }
> +
> + if (crtc_state->out_fence_ptr) {
> + ret = crtc_setup_out_fence(crtc, crtc_state,
> + &fence_state[fence_idx++]);
> + if (ret)
> + goto out;
> + }
> }
>
> if (arg->flags & DRM_MODE_ATOMIC_TEST_ONLY) {
> @@ -1761,24 +1816,37 @@ retry:
> ret = drm_atomic_commit(state);
> }
>
> + for (i = 0; i < fence_idx; i++)
> + fd_install(fence_state[i].fd, fence_state[i].sync_file->file);
> +
> out:
> drm_atomic_clean_old_fb(dev, plane_mask, ret);
>
> - if (ret && arg->flags & DRM_MODE_PAGE_FLIP_EVENT) {
> + for_each_crtc_in_state(state, crtc, crtc_state, i) {
I think a check on ret is still needed before you iterate the CRTCs.
If the commit is successful then state has already been freed by this
point, and I get a splat.
Yes, I believe I only tested with non-blocking requests.
Right, I suppose you should get here before the free in that case.
> /*
> * TEST_ONLY and PAGE_FLIP_EVENT are mutually exclusive,
> * if they weren't, this code should be called on success
> * for TEST_ONLY too.
> */
> + if (ret && crtc_state->event)
> + drm_event_cancel_free(dev, &crtc_state->event->base);
> + }
>
> - for_each_crtc_in_state(state, crtc, crtc_state, i) {
> - if (!crtc_state->event)
> - continue;
> + if (ret && fence_state) {
> + for (i = 0; i < fence_idx; i++) {
> + if (fence_state[i].fd >= 0)
> + put_unused_fd(fence_state[i].fd);
> + if (fence_state[i].sync_file)
> + fput(fence_state[i].sync_file->file);
>
> - drm_event_cancel_free(dev, &crtc_state->event->base);
> + /* If this fails, there's not much we can do about it */
> + if (put_user(-1, fence_state->out_fence_ptr))
> + DRM_ERROR("Couldn't clear out_fence_ptr\n");
> }
> }
>
> + kfree(fence_state);
> +
> if (ret == -EDEADLK) {
> drm_atomic_state_clear(state);
> drm_modeset_backoff(&ctx);
> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c
> index b99090f..40bce97 100644
> --- a/drivers/gpu/drm/drm_crtc.c
> +++ b/drivers/gpu/drm/drm_crtc.c
> @@ -274,6 +274,8 @@ int drm_crtc_init_with_planes(struct drm_device *dev,
struct drm_crtc *crtc,
> if (drm_core_check_feature(dev, DRIVER_ATOMIC)) {
> drm_object_attach_property(&crtc->base, config->prop_active, 0);
> drm_object_attach_property(&crtc->base, config->prop_mode_id, 0);
> + drm_object_attach_property(&crtc->base,
> + config->prop_out_fence_ptr, 0);
> }
>
> return 0;
> @@ -434,6 +436,12 @@ static int drm_mode_create_standard_properties(struct
drm_device *dev)
> return -ENOMEM;
> dev->mode_config.prop_in_fence_fd = prop;
>
> + prop = drm_property_create_range(dev, DRM_MODE_PROP_ATOMIC,
> + "OUT_FENCE_PTR", 0, U64_MAX);
> + if (!prop)
> + return -ENOMEM;
> + dev->mode_config.prop_out_fence_ptr = prop;
> +
> prop = drm_property_create_object(dev, DRM_MODE_PROP_ATOMIC,
> "CRTC_ID", DRM_MODE_OBJECT_CRTC);
> if (!prop)
> diff --git a/include/drm/drm_crtc.h b/include/drm/drm_crtc.h
> index 657a33a..b898604 100644
> --- a/include/drm/drm_crtc.h
> +++ b/include/drm/drm_crtc.h
> @@ -165,6 +165,8 @@ struct drm_crtc_state {
> struct drm_property_blob *ctm;
> struct drm_property_blob *gamma_lut;
>
> + u64 __user *out_fence_ptr;
> +
I'm somewhat not convinced about stashing a __user pointer in the
CRTC atomic state. I know it gets cleared before the driver sees it,
but if anything I think that hints that this isn't the right place to
store it.
I've been trying to think of other ways to get from a property to
something drm_mode_atomic_ioctl can use - maybe we can store some
stuff in drm_atomic_state which only lives for the length of the ioctl
call and put this pointer in there.
The drm_atomic_state is still visible by the drivers. Between there and
crtc_state, I would keep in the crtc_state for now.
Mm, yeah I suppose they could get to it if they went looking for it
in ->atomic_set_property or something, but I was thinking of freeing
it before the drm_atomic_commit.
Anyway, this way is certainly simpler, and setting it to NULL should
be enough to limit the damage a driver can do :-)
Thanks,
Brian
Gustavo
