Hi Maarten,
Thank you for the patch.
On Wednesday, 30 August 2017 15:17:51 EEST Maarten Lankhorst wrote:
> Currently we neatly track the crtc state, but forget to look at
> plane/connector state.
>
> When doing a nonblocking modeset, immediately followed by a setprop
> before the modeset completes, the setprop will see the modesets new
> state as the old state and free it.
>
> This has to be solved by waiting for hw_done on the connector, even
> if it's not assigned to a crtc. When a connector is unbound we take
> the last crtc commit, and when it stays unbound we create a new
> fake crtc commit for that gets signaled on hw_done for all the
> planes/connectors.
>
> We wait for it the same way as we do for crtc's, which will make
> sure we never run into a use-after-free situation.
>
> Changes since v1:
> - Only create a single disable commit. (danvet)
> - Fix leak in intel_legacy_cursor_update.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx>
> Testcase: kms_atomic_transition.plane-use-after-nonblocking-unbind*
> Cc: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/drm_atomic.c | 4 +
> drivers/gpu/drm/drm_atomic_helper.c | 156 ++++++++++++++++++++++++++++++--
> drivers/gpu/drm/i915/intel_display.c | 2 +
> include/drm/drm_atomic.h | 12 +++
> include/drm/drm_connector.h | 7 ++
> include/drm/drm_plane.h | 7 ++
> 6 files changed, 182 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
> index 2cce48f203e0..75f5f74de9bf 100644
> --- a/drivers/gpu/drm/drm_atomic.c
> +++ b/drivers/gpu/drm/drm_atomic.c
> @@ -192,6 +192,10 @@ void drm_atomic_state_default_clear(struct
> drm_atomic_state *state) }
> state->num_private_objs = 0;
>
> + if (state->fake_commit) {
> + drm_crtc_commit_put(state->fake_commit);
> + state->fake_commit = NULL;
> + }
> }
> EXPORT_SYMBOL(drm_atomic_state_default_clear);
>
> diff --git a/drivers/gpu/drm/drm_atomic_helper.c
> b/drivers/gpu/drm/drm_atomic_helper.c index 8ccb8b6536c0..034f563fb130
> 100644
> --- a/drivers/gpu/drm/drm_atomic_helper.c
> +++ b/drivers/gpu/drm/drm_atomic_helper.c
> @@ -1644,6 +1644,40 @@ static void release_crtc_commit(struct completion
> *completion) drm_crtc_commit_put(commit);
> }
>
> +static void init_commit(struct drm_crtc_commit *commit, struct drm_crtc
> *crtc)
> +{
You could allocate the commit in this function too, the kzalloc() is currently
duplicated. The function should probably be called alloc_commit() then.
> + init_completion(&commit->flip_done);
> + init_completion(&commit->hw_done);
> + init_completion(&commit->cleanup_done);
> + INIT_LIST_HEAD(&commit->commit_entry);
> + kref_init(&commit->ref);
> + commit->crtc = crtc;
> +}
> +
> +static struct drm_crtc_commit *
> +fake_or_crtc_commit(struct drm_atomic_state *state, struct drm_crtc *crtc)
> +{
> + struct drm_crtc_commit *commit;
> +
> + if (crtc) {
> + struct drm_crtc_state *new_crtc_state;
> +
> + new_crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
> +
> + commit = new_crtc_state->commit;
> + } else if (!state->fake_commit) {
> + state->fake_commit = commit = kzalloc(sizeof(*commit), GFP_KERNEL);
> + if (!commit)
> + return NULL;
> +
> + init_commit(commit, NULL);
> + } else
> + commit = state->fake_commit;
> +
> + drm_crtc_commit_get(commit);
I believe the reference counting is right. The double reference in the second
case (kref_init() when initializing the commit and drm_crtc_commit_get())
should not cause a leak. The kref_init() takes a reference to store the commit
in state->fake_commit, released in drm_atomic_state_default_clear(), and the
drm_crtc_commit_get() takes a reference returned by the function, stored in
new_*_state->commit by the caller.
This being said, I think the reference counting is confusing, as proved by
Daniel thinking there was a leak here (or by me thinking there's no leak while
there's one :-)). To make the implementation clearer, I propose turning the
definition of drm_crtc_commit_get() to
static inline struct drm_crtc_commit *
drm_crtc_commit_get(struct drm_crtc_commit *commit)
{
kref_get(&commit->ref);
return commit;
}
and writing this function as
/* Return a new reference to the commit object */
static struct drm_crtc_commit *
fake_or_crtc_commit(struct drm_atomic_state *state, struct drm_crtc *crtc)
{
struct drm_crtc_commit *commit;
if (crtc) {
struct drm_crtc_state *new_crtc_state;
new_crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
commit = new_crtc_state->commit;
} else {
if (!state->fake_commit)
state->fake_commit = alloc_commit(NULL);
commit = state->fake_commit;
}
return drm_crtc_commit_get(commit);
}
> + return commit;
> +}
> +
> /**
> * drm_atomic_helper_setup_commit - setup possibly nonblocking commit
> * @state: new modeset state to be committed
> @@ -1692,6 +1726,10 @@ int drm_atomic_helper_setup_commit(struct
> drm_atomic_state *state, {
> struct drm_crtc *crtc;
> struct drm_crtc_state *old_crtc_state, *new_crtc_state;
> + struct drm_connector *conn;
> + struct drm_connector_state *old_conn_state, *new_conn_state;
> + struct drm_plane *plane;
> + struct drm_plane_state *old_plane_state, *new_plane_state;
> struct drm_crtc_commit *commit;
> int i, ret;
>
> @@ -1700,12 +1738,7 @@ int drm_atomic_helper_setup_commit(struct
> drm_atomic_state *state, if (!commit)
> return -ENOMEM;
>
> - init_completion(&commit->flip_done);
> - init_completion(&commit->hw_done);
> - init_completion(&commit->cleanup_done);
> - INIT_LIST_HEAD(&commit->commit_entry);
> - kref_init(&commit->ref);
> - commit->crtc = crtc;
> + init_commit(commit, crtc);
>
> new_crtc_state->commit = commit;
>
> @@ -1741,6 +1774,36 @@ int drm_atomic_helper_setup_commit(struct
> drm_atomic_state *state, drm_crtc_commit_get(commit);
> }
>
> + for_each_oldnew_connector_in_state(state, conn, old_conn_state,
> new_conn_state, i) {
> + if (new_conn_state->crtc)
> + continue;
> +
> + if (nonblock && old_conn_state->commit &&
> + !try_wait_for_completion(&old_conn_state->commit->flip_done))
> + return -EBUSY;
flip_done only, not hw_done ? A comment that explains why would be helpful.
> +
> + commit = fake_or_crtc_commit(state, old_conn_state->crtc);
> + if (!commit)
> + return -ENOMEM;
> +
> + new_conn_state->commit = commit;
> + }
> +
> + for_each_oldnew_plane_in_state(state, plane, old_plane_state,
> new_plane_state, i) {
> + if (new_plane_state->crtc)
> + continue;
> +
> + if (nonblock && old_plane_state->commit &&
> + !try_wait_for_completion(&old_plane_state->commit->flip_done))
> + return -EBUSY;
Same here.
> + commit = fake_or_crtc_commit(state, old_plane_state->crtc);
> + if (!commit)
> + return -ENOMEM;
> +
> + new_plane_state->commit = commit;
> + }
> +
> return 0;
> }
> EXPORT_SYMBOL(drm_atomic_helper_setup_commit);
> @@ -1761,6 +1824,10 @@ void drm_atomic_helper_wait_for_dependencies(struct
> drm_atomic_state *old_state) {
> struct drm_crtc *crtc;
> struct drm_crtc_state *old_crtc_state;
> + struct drm_plane *plane;
> + struct drm_plane_state *old_plane_state;
> + struct drm_connector *conn;
> + struct drm_connector_state *old_conn_state;
> struct drm_crtc_commit *commit;
> int i;
> long ret;
> @@ -1785,6 +1852,48 @@ void drm_atomic_helper_wait_for_dependencies(struct
> drm_atomic_state *old_state) DRM_ERROR("[CRTC:%d:%s] flip_done timed
> out\n",
> crtc->base.id, crtc->name);
> }
> +
> + for_each_old_connector_in_state(old_state, conn, old_conn_state, i) {
> + commit = old_conn_state->commit;
> +
> + if (!commit)
> + continue;
> +
> + ret = wait_for_completion_timeout(&commit->hw_done,
> + 10*HZ);
> + if (ret == 0)
> + DRM_ERROR("[CONNECTOR:%d:%s] hw_done timed out\n",
> + conn->base.id, conn->name);
> +
> + /* Currently no support for overwriting flips, hence
> + * stall for previous one to execute completely. */
> + ret = wait_for_completion_timeout(&commit->flip_done,
> + 10*HZ);
> + if (ret == 0)
> + DRM_ERROR("[CONNECTOR:%d:%s] flip_done timed out\n",
> + conn->base.id, conn->name);
There's probably something I don't get, but given that the connector state -
>commit pointer is only set for commits that have no CRTC, do we have a
guarantee that there will be a page flip to be signalled ? Can't we for
instance disable a connector without a page flip ? Or set a property on an
already disabled connector ?
> + }
> +
> + for_each_old_plane_in_state(old_state, plane, old_plane_state, i) {
> + commit = old_plane_state->commit;
> +
> + if (!commit)
> + continue;
> +
> + ret = wait_for_completion_timeout(&commit->hw_done,
> + 10*HZ);
> + if (ret == 0)
> + DRM_ERROR("[PLANE:%d:%s] hw_done timed out\n",
> + plane->base.id, plane->name);
> +
> + /* Currently no support for overwriting flips, hence
> + * stall for previous one to execute completely. */
> + ret = wait_for_completion_timeout(&commit->flip_done,
> + 10*HZ);
> + if (ret == 0)
> + DRM_ERROR("[PLANE:%d:%s] flip_done timed out\n",
> + plane->base.id, plane->name);
> + }
> }
> EXPORT_SYMBOL(drm_atomic_helper_wait_for_dependencies);
>
> @@ -1819,6 +1928,11 @@ void drm_atomic_helper_commit_hw_done(struct
> drm_atomic_state *old_state) WARN_ON(new_crtc_state->event);
> complete_all(&commit->hw_done);
> }
> +
> + if (old_state->fake_commit) {
> + complete_all(&old_state->fake_commit->hw_done);
> + complete_all(&old_state->fake_commit->flip_done);
> + }
> }
> EXPORT_SYMBOL(drm_atomic_helper_commit_hw_done);
>
> @@ -2241,6 +2355,28 @@ int drm_atomic_helper_swap_state(struct
> drm_atomic_state *state, if (ret)
> return ret;
> }
> +
A comment would be helpful here too. In particular explaining why you wait on
hw_done only isn't straightforward.
> + for_each_old_connector_in_state(state, connector, old_conn_state, i) {
> + commit = old_conn_state->commit;
> +
> + if (!commit)
> + continue;
> +
> + ret = wait_for_completion_interruptible(&commit->hw_done);
> + if (ret)
> + return ret;
> + }
> +
> + for_each_old_plane_in_state(state, plane, old_plane_state, i) {
> + commit = old_plane_state->commit;
> +
> + if (!commit)
> + continue;
> +
> + ret = wait_for_completion_interruptible(&commit->hw_done);
> + if (ret)
> + return ret;
> + }
> }
>
> for_each_oldnew_connector_in_state(state, connector, old_conn_state,
> new_conn_state, i) { @@ -3223,6 +3359,7 @@ void
> __drm_atomic_helper_plane_duplicate_state(struct drm_plane *plane,
> drm_framebuffer_get(state->fb);
>
> state->fence = NULL;
> + state->commit = NULL;
> }
> EXPORT_SYMBOL(__drm_atomic_helper_plane_duplicate_state);
>
> @@ -3264,6 +3401,9 @@ void __drm_atomic_helper_plane_destroy_state(struct
> drm_plane_state *state)
>
> if (state->fence)
> dma_fence_put(state->fence);
> +
> + if (state->commit)
> + drm_crtc_commit_put(state->commit);
> }
> EXPORT_SYMBOL(__drm_atomic_helper_plane_destroy_state);
>
> @@ -3342,6 +3482,7 @@ __drm_atomic_helper_connector_duplicate_state(struct
> drm_connector *connector, memcpy(state, connector->state, sizeof(*state));
> if (state->crtc)
> drm_connector_get(connector);
> + state->commit = NULL;
> }
> EXPORT_SYMBOL(__drm_atomic_helper_connector_duplicate_state);
>
> @@ -3468,6 +3609,9 @@ __drm_atomic_helper_connector_destroy_state(struct
> drm_connector_state *state) {
> if (state->crtc)
> drm_connector_put(state->connector);
> +
> + if (state->commit)
> + drm_crtc_commit_put(state->commit);
> }
> EXPORT_SYMBOL(__drm_atomic_helper_connector_destroy_state);
>
> diff --git a/drivers/gpu/drm/i915/intel_display.c
> b/drivers/gpu/drm/i915/intel_display.c index 3f3cb96aa11e..70ce02753177
> 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -13109,8 +13109,10 @@ intel_legacy_cursor_update(struct drm_plane *plane,
>
> /* Swap plane state */
> new_plane_state->fence = old_plane_state->fence;
> + new_plane_state->commit = old_plane_state->commit;
> *to_intel_plane_state(old_plane_state) =
> *to_intel_plane_state(new_plane_state); new_plane_state->fence = NULL;
> + new_plane_state->commit = NULL;
> new_plane_state->fb = old_fb;
> to_intel_plane_state(new_plane_state)->vma = NULL;
>
> diff --git a/include/drm/drm_atomic.h b/include/drm/drm_atomic.h
> index e76d9f95c191..ba976f4a4938 100644
> --- a/include/drm/drm_atomic.h
> +++ b/include/drm/drm_atomic.h
> @@ -236,6 +236,18 @@ struct drm_atomic_state {
> struct drm_modeset_acquire_ctx *acquire_ctx;
>
> /**
> + * @fake_commit:
> + *
> + * Used for signaling unbound planes/connectors.
> + * When a connector or plane is not bound to any CRTC, it's still
> important
> + * to preserve linearity to prevent the atomic states from being freed to
> early.
> + *
> + * This commit (if set) is not bound to any crtc, but will be completed
> when
> + * drm_atomic_helper_commit_hw_done() is called.
> + */
> + struct drm_crtc_commit *fake_commit;
> +
> + /**
> * @commit_work:
> *
> * Work item which can be used by the driver or helpers to execute the
> diff --git a/include/drm/drm_connector.h b/include/drm/drm_connector.h
> index ea8da401c93c..8837649d16e8 100644
> --- a/include/drm/drm_connector.h
> +++ b/include/drm/drm_connector.h
> @@ -347,6 +347,13 @@ struct drm_connector_state {
>
> struct drm_atomic_state *state;
>
> + /**
> + * @commit: Tracks the pending commit to prevent use-after-free
> conditions.
> + *
> + * Is only set when @crtc is NULL.
> + */
> + struct drm_crtc_commit *commit;
> +
> struct drm_tv_connector_state tv;
>
> /**
> diff --git a/include/drm/drm_plane.h b/include/drm/drm_plane.h
> index 73f90f9d057f..7d96116fd4c4 100644
> --- a/include/drm/drm_plane.h
> +++ b/include/drm/drm_plane.h
> @@ -123,6 +123,13 @@ struct drm_plane_state {
> */
> bool visible;
>
> + /**
> + * @commit: Tracks the pending commit to prevent use-after-free
> conditions.
> + *
> + * Is only set when @crtc is NULL.
> + */
> + struct drm_crtc_commit *commit;
> +
> struct drm_atomic_state *state;
> };
--
Regards,
Laurent Pinchart
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
