On Fri, Mar 07, 2025 at 03:30:41PM +0800, Andy Yan wrote:
>
> Hi All,
> At 2025-03-07 09:08:48, "Andy Yan" <andyshrk@xxxxxxx> wrote:
> >Hi All,
> >
> >At 2025-03-06 23:41:24, "Simona Vetter" <simona.vetter@xxxxxxxx> wrote:
> >>On Thu, Mar 06, 2025 at 08:10:16AM +0100, Maxime Ripard wrote:
> >>> On Thu, Mar 06, 2025 at 09:16:24AM +0800, Andy Yan wrote:
> >>> >
> >>> > Hi Maxime and Dmitry:
> >>> >
> >>> > At 2025-03-06 04:13:53, "Dmitry Baryshkov" <dmitry.baryshkov@xxxxxxxxxx> wrote:
> >>> > >On Wed, Mar 05, 2025 at 02:19:36PM +0100, Maxime Ripard wrote:
> >>> > >> Hi Andy,
> >>> > >>
> >>> > >> On Wed, Mar 05, 2025 at 07:55:19PM +0800, Andy Yan wrote:
> >>> > >> > At 2025-03-04 19:10:47, "Maxime Ripard" <mripard@xxxxxxxxxx> wrote:
> >>> > >> > >With the bridges switching over to drm_bridge_connector, the direct
> >>> > >> > >association between a bridge driver and its connector was lost.
> >>> > >> > >
> >>> > >> > >This is mitigated for atomic bridge drivers by the fact you can access
> >>> > >> > >the encoder, and then call drm_atomic_get_old_connector_for_encoder() or
> >>> > >> > >drm_atomic_get_new_connector_for_encoder() with drm_atomic_state.
> >>> > >> > >
> >>> > >> > >This was also made easier by providing drm_atomic_state directly to all
> >>> > >> > >atomic hooks bridges can implement.
> >>> > >> > >
> >>> > >> > >However, bridge drivers don't have a way to access drm_atomic_state
> >>> > >> > >outside of the modeset path, like from the hotplug interrupt path or any
> >>> > >> > >interrupt handler.
> >>> > >> > >
> >>> > >> > >Let's introduce a function to retrieve the connector currently assigned
> >>> > >> > >to an encoder, without using drm_atomic_state, to make these drivers'
> >>> > >> > >life easier.
> >>> > >> > >
> >>> > >> > >Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
> >>> > >> > >Co-developed-by: Simona Vetter <simona.vetter@xxxxxxxx>
> >>> > >> > >Signed-off-by: Maxime Ripard <mripard@xxxxxxxxxx>
> >>> > >> > >---
> >>> > >> > > drivers/gpu/drm/drm_atomic.c | 45 ++++++++++++++++++++++++++++++++++++++++++++
> >>> > >> > > include/drm/drm_atomic.h | 3 +++
> >>> > >> > > 2 files changed, 48 insertions(+)
> >>> > >> > >
> >>> > >> > >diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
> >>> > >> > >index 9ea2611770f43ce7ccba410406d5f2c528aab022..b926b132590e78f8d41d48eb4da4bccf170ee236 100644
> >>> > >> > >--- a/drivers/gpu/drm/drm_atomic.c
> >>> > >> > >+++ b/drivers/gpu/drm/drm_atomic.c
> >>> > >> > >@@ -985,10 +985,55 @@ drm_atomic_get_new_connector_for_encoder(const struct drm_atomic_state *state,
> >>> > >> > >
> >>> > >> > > return NULL;
> >>> > >> > > }
> >>> > >> > > EXPORT_SYMBOL(drm_atomic_get_new_connector_for_encoder);
> >>> > >> > >
> >>> > >> > >+/**
> >>> > >> > >+ * drm_atomic_get_connector_for_encoder - Get connector currently assigned to an encoder
> >>> > >> > >+ * @encoder: The encoder to find the connector of
> >>> > >> > >+ * @ctx: Modeset locking context
> >>> > >> > >+ *
> >>> > >> > >+ * This function finds and returns the connector currently assigned to
> >>> > >> > >+ * an @encoder.
> >>> > >> > >+ *
> >>> > >> > >+ * Returns:
> >>> > >> > >+ * The connector connected to @encoder, or an error pointer otherwise.
> >>> > >> > >+ * When the error is EDEADLK, a deadlock has been detected and the
> >>> > >> > >+ * sequence must be restarted.
> >>> > >> > >+ */
> >>> > >> > >+struct drm_connector *
> >>> > >> > >+drm_atomic_get_connector_for_encoder(const struct drm_encoder *encoder,
> >>> > >> > >+ struct drm_modeset_acquire_ctx *ctx)
> >>> > >> > >+{
> >>> > >> > >+ struct drm_connector_list_iter conn_iter;
> >>> > >> > >+ struct drm_connector *out_connector = ERR_PTR(-EINVAL);
> >>> > >> > >+ struct drm_connector *connector;
> >>> > >> > >+ struct drm_device *dev = encoder->dev;
> >>> > >> > >+ int ret;
> >>> > >> > >+
> >>> > >> > >+ ret = drm_modeset_lock(&dev->mode_config.connection_mutex, ctx);
> >>> > >> > >+ if (ret)
> >>> > >> > >+ return ERR_PTR(ret);
> >>> > >> >
> >>> > >> > It seems that this will cause a deadlock when called from a hotplug
> >>> > >> > handling path, I have a WIP DP diver[0], which suggested by Dmitry to
> >>> > >> > use this API from a &drm_bridge_funcs.detect callback to get the
> >>> > >> > connector, as detect is called by drm_helper_probe_detect, which will
> >>> > >> > hold connection_mutex first, so the deaklock happens:
> >>> > >> >
> >>> > >> > drm_helper_probe_detect(struct drm_connector *connector,
> >>> > >> > struct drm_modeset_acquire_ctx *ctx,
> >>> > >> > bool force)
> >>> > >> > {
> >>> > >> > const struct drm_connector_helper_funcs *funcs = connector->helper_private;
> >>> > >> > struct drm_device *dev = connector->dev;
> >>> > >> > int ret;
> >>> > >> >
> >>> > >> > if (!ctx)
> >>> > >> > return drm_helper_probe_detect_ctx(connector, force);
> >>> > >> >
> >>> > >> > ret = drm_modeset_lock(&dev->mode_config.connection_mutex, ctx);
> >>> > >> > if (ret)
> >>> > >> > return ret;
> >>> > >> >
> >>> > >> > if (funcs->detect_ctx)
> >>> > >> > ret = funcs->detect_ctx(connector, ctx, force);
> >>> > >> > else if (connector->funcs->detect)
> >>> > >> > ret = connector->funcs->detect(connector, force);
> >>> > >> > else
> >>> > >> > ret = connector_status_connected;
> >>> > >> >
> >>> > >> > if (ret != connector->status)
> >>> > >> > connector->epoch_counter += 1;
> >>> > >> >
> >>> > >> > So I wonder can we let drm_bridge_funcs.detect pass a connector for
> >>> > >> > this case ?
> >>> > >>
> >>> > >> Do you actually see a deadlock occurring? AFAIK, drm_modeset_lock is
> >>> > >> fine with reentrancy from the same context, so it should work just fine.
> >>> > >
> >>> > >Andy, that probably means that you should use .detect_ctx() and pass the
> >>> > >context to drm_atomic_get_connector_for_encoder().
> >>> >
> >>> > Unfortunately, the drm_bridge_funcs does not have a .detect_ctx() version .
> >>> > The call chain is:
> >>> > drm_helper_probe_detect
> >>> > --> drm_bridge_connector_detect(struct drm_connector *connector, bool force)
> >>> > --> drm_bridge_funcs.detect(bridge)
> >>> > The ctx got dropped when drm_helper_probe_detect call drm_bridge_connector_detect
> >>> > The connector got dropped when connector call it's bridege.detect
> >>> >
> >>> > So I think the simplest solution is to have drm_bridge_funcs.detect
> >>> > directly pass the connector
> >>>
> >>> I don't disagree on principle, but I think a better first step would be
> >>> to provide a detect_ctx hook to bridges.
> >>
> >>Yup. There's other reasons you really want to get at the locking context
> >>in detect callbacks, doing this special case by passing something for
> >>everyone doesn't sound like the right approach to me.
> >
> >Ok, I will add a detect_ctx hook for bridge. Thanks for your advice.
> >
> >Just confirm that can I send this add detect_ctx hook patch alone first?
> >I think this patch will be easy to merge, so it can help my WIP DP driver stay light on dependencies。
Yeah I think sending this prep work as a standalone thing is ok,
especially since it looks like there's going to be more work involved.
> When I try to add the detect_ctx hook to bridge, I found that there is still a case that there is no ctx to
> pass to detect_ctx:
>
> int drm_helper_probe_single_connector_modes(struct drm_connector *connector,
> uint32_t maxX, uint32_t maxY)
> {
> ...............
> struct drm_modeset_acquire_ctx ctx;
>
> WARN_ON(!mutex_is_locked(&dev->mode_config.mutex));
>
> drm_modeset_acquire_init(&ctx, 0);
>
> drm_dbg_kms(dev, "[CONNECTOR:%d:%s]\n", connector->base.id,
> connector->name);
>
> retry:
> ret = drm_modeset_lock(&dev->mode_config.connection_mutex, &ctx);
> ...................
> ret = drm_helper_probe_detect(connector, &ctx, true);
> ......................................
>
> count = drm_helper_probe_get_modes(connector);
>
> Then in drm_bridge_connector_get_modes:
>
> static int drm_bridge_connector_get_modes(struct drm_connector *connector)
> {
> struct drm_bridge_connector *bridge_connector =
> to_drm_bridge_connector(connector);
> struct drm_bridge *bridge;
>
> ...........................................
> /*
> * If display exposes EDID, then we parse that in the normal way to
> * build table of supported modes.
> */
> bridge = bridge_connector->bridge_edid;
> if (bridge)
> return drm_bridge_connector_get_modes_edid(connector, bridge);
>
> static int drm_bridge_connector_get_modes_edid(struct drm_connector *connector,
> struct drm_bridge *bridge)
> {
> enum drm_connector_status status;
> const struct drm_edid *drm_edid;
> int n;
>
> status = drm_bridge_connector_detect(connector, false);
>
> ......................
>
> -drm_bridge_connector_detect(struct drm_connector *connector, bool force)
> +drm_bridge_connector_detect(struct drm_connector *connector,
> + struct drm_modeset_acquire_ctx *ctx,
> + bool force)
> {
> struct drm_bridge_connector *bridge_connector =
> to_drm_bridge_connector(connector);
> @@ -186,7 +188,7 @@ drm_bridge_connector_detect(struct drm_connector *connector, bool force)
> enum drm_connector_status status;
>
> if (detect) {
> status = detect->funcs->detect(detect, ctx);
>
> There is still no ctx in this call chain.
>
> So there will be deadlock if I use drm_atomic_get_new_connector_for_encoder to find connector in
> my bridge detect_ctx hook.
Hm yeah looks like bridge callchain isn't split up into ->detec and
->get_modes in exactly the same way as connectors. I think the clean
solution would be to pull drm_bridge_connector_detect() out from
drm_bridge_connector_get_modes(), but that might unravel a huge amount of
work. But wiring the ctx through all the get_mode functions might also not
work out great, hence why I'd try moving _detect() first.
Once you have bridge_connector_detect at the same split-up like any other
connectors, it should be possible to add the contect to
drm_bridge_connector_detect and have it everywhere. In general I think
calling drm_bridge_connector_detect with a NULL ctx should be a bug, and
we should catch that with a WARN_ON and bail out. Otherwise the locking is
going to be a complete mess.
And yeah definitely do this prep work as a separate series, I think the
need is clearly established, so we don't need a user right away to justify
this rework.
Cheers, Sima
>
>
>
> >
> >
> >>-Sima
> >>--
> >>Simona Vetter
> >>Software Engineer, Intel Corporation
> >>http://blog.ffwll.ch
--
Simona Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
