On Wed, Apr 24, 2024 at 04:05:31PM +0300, Tomi Valkeinen wrote: > On 19/04/2024 18:36, Laurent Pinchart wrote: > > On Tue, Apr 16, 2024 at 04:55:12PM +0300, Tomi Valkeinen wrote: > >> At the moment the v4l2_subdev_enable/disable_streams() functions call > >> fallback helpers to handle the case where the subdev only implements > >> .s_stream(), and the main function handles the case where the subdev > >> implements streams (V4L2_SUBDEV_FL_STREAMS, which implies > >> .enable/disable_streams()). > >> > >> What is missing is support for subdevs which do not implement streams > >> support, but do implement .enable/disable_streams(). Example cases of > >> these subdevices are single-stream cameras, where using > >> .enable/disable_streams() is not required but helps us remove the users > >> of the legacy .s_stream(), and subdevices with multiple source pads (but > >> single stream per pad), where .enable/disable_streams() allows the > >> subdevice to control the enable/disable state per pad. > >> > >> The two single-streams cases (.s_stream() and .enable/disable_streams()) > >> are very similar, and with small changes we can change the > >> v4l2_subdev_enable/disable_streams() functions to support all three > >> cases, without needing separate fallback functions. > >> > >> A few potentially problematic details, though: > >> > >> - For the single-streams cases we use sd->enabled_pads field, which > >> limits the number of pads for the subdevice to 64. For simplicity I > >> added the check for this limitation to the beginning of the function, > >> and it also applies to the streams case. > >> > >> - The fallback functions only allowed the target pad to be a source pad. > >> It is not very clear to me why this check was needed, but it was not > >> needed in the streams case. However, I doubt the > >> v4l2_subdev_enable/disable_streams() code has ever been tested with > >> sink pads, so to be on the safe side, I added the same check > >> to the v4l2_subdev_enable/disable_streams() functions. > >> > > > > Now that v4l2_subdev_enable_streams() is usable by (almost) every > > driver, should we update documentation to indicate that manual calls to > > .s_stream() should be avoided ? > > How about: > > diff --git a/include/media/v4l2-subdev.h b/include/media/v4l2-subdev.h > index dabe1b5dfe4a..f52bb790773c 100644 > --- a/include/media/v4l2-subdev.h > +++ b/include/media/v4l2-subdev.h > @@ -450,6 +450,11 @@ enum v4l2_subdev_pre_streamon_flags { > * already started or stopped subdev. Also see call_s_stream wrapper in > * v4l2-subdev.c. > * > + * New drivers should instead implement &v4l2_subdev_pad_ops.enable_streams > + * and &v4l2_subdev_pad_ops.disable_streams operations, and use > + * v4l2_subdev_s_stream_helper for the &v4l2_subdev_video_ops.s_stream > + * operation to support legacy users.A > + * Add * Drivers should also not call the .s_stream() subdev operation directly, * but use the v4l2_subdev_enable_streams() and * v4l2_subdev_disable_streams() helpers. > * @g_pixelaspect: callback to return the pixelaspect ratio. > * > * @s_dv_timings: Set custom dv timings in the sub device. This is used > > > Although now that I think about it, can we "ever" get rid of s_stream? > enable_streams is only usable if the subdev has pads. Do we have pad-less subdevs that can stream ? > If that's the > case, the text above should probably clarify when one should continue > using s_stream. > > >> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@xxxxxxxxxxxxxxxx> > >> --- > >> drivers/media/v4l2-core/v4l2-subdev.c | 187 ++++++++++++++-------------------- > >> 1 file changed, 79 insertions(+), 108 deletions(-) > >> > >> diff --git a/drivers/media/v4l2-core/v4l2-subdev.c b/drivers/media/v4l2-core/v4l2-subdev.c > >> index e45fd42da1e3..10406acfb9d0 100644 > >> --- a/drivers/media/v4l2-core/v4l2-subdev.c > >> +++ b/drivers/media/v4l2-core/v4l2-subdev.c > >> @@ -2106,6 +2106,13 @@ static void v4l2_subdev_collect_streams(struct v4l2_subdev *sd, > >> u64 *found_streams, > >> u64 *enabled_streams) > >> { > >> + if (!(sd->flags & V4L2_SUBDEV_FL_STREAMS)) { > >> + *found_streams = BIT_ULL(0); > >> + *enabled_streams = > >> + (sd->enabled_pads & BIT_ULL(pad)) ? BIT_ULL(0) : 0; > >> + return; > >> + } > >> + > >> *found_streams = 0; > >> *enabled_streams = 0; > >> > >> @@ -2127,6 +2134,14 @@ static void v4l2_subdev_set_streams_enabled(struct v4l2_subdev *sd, > >> u32 pad, u64 streams_mask, > >> bool enabled) > >> { > >> + if (!(sd->flags & V4L2_SUBDEV_FL_STREAMS)) { > >> + if (enabled) > >> + sd->enabled_pads |= BIT_ULL(pad); > >> + else > >> + sd->enabled_pads &= ~BIT_ULL(pad); > >> + return; > >> + } > >> + > >> for (unsigned int i = 0; i < state->stream_configs.num_configs; ++i) { > >> struct v4l2_subdev_stream_config *cfg = > >> &state->stream_configs.configs[i]; > >> @@ -2136,51 +2151,6 @@ static void v4l2_subdev_set_streams_enabled(struct v4l2_subdev *sd, > >> } > >> } > >> > >> -static int v4l2_subdev_enable_streams_fallback(struct v4l2_subdev *sd, u32 pad, > >> - u64 streams_mask) > >> -{ > >> - struct device *dev = sd->entity.graph_obj.mdev->dev; > >> - int ret; > >> - > >> - /* > >> - * The subdev doesn't implement pad-based stream enable, fall back > >> - * to the .s_stream() operation. > >> - */ > >> - if (!(sd->entity.pads[pad].flags & MEDIA_PAD_FL_SOURCE)) > >> - return -EOPNOTSUPP; > >> - > >> - /* > >> - * .s_stream() means there is no streams support, so the only allowed > >> - * stream is the implicit stream 0. > >> - */ > >> - if (streams_mask != BIT_ULL(0)) > >> - return -EOPNOTSUPP; > >> - > >> - /* > >> - * We use a 64-bit bitmask for tracking enabled pads, so only subdevices > >> - * with 64 pads or less can be supported. > >> - */ > >> - if (pad >= sizeof(sd->enabled_pads) * BITS_PER_BYTE) > >> - return -EOPNOTSUPP; > >> - > >> - if (sd->enabled_pads & BIT_ULL(pad)) { > >> - dev_dbg(dev, "pad %u already enabled on %s\n", > >> - pad, sd->entity.name); > >> - return -EALREADY; > >> - } > >> - > >> - /* Start streaming when the first pad is enabled. */ > >> - if (!sd->enabled_pads) { > >> - ret = v4l2_subdev_call(sd, video, s_stream, 1); > >> - if (ret) > >> - return ret; > >> - } > >> - > >> - sd->enabled_pads |= BIT_ULL(pad); > >> - > >> - return 0; > >> -} > >> - > >> int v4l2_subdev_enable_streams(struct v4l2_subdev *sd, u32 pad, > >> u64 streams_mask) > >> { > >> @@ -2189,21 +2159,33 @@ int v4l2_subdev_enable_streams(struct v4l2_subdev *sd, u32 pad, > >> bool already_streaming; > >> u64 enabled_streams; > >> u64 found_streams; > >> + bool use_s_stream; > >> int ret; > >> > >> /* A few basic sanity checks first. */ > >> if (pad >= sd->entity.num_pads) > >> return -EINVAL; > >> > >> + if (!(sd->entity.pads[pad].flags & MEDIA_PAD_FL_SOURCE)) > >> + return -EOPNOTSUPP; > >> + > >> + /* > >> + * We use a 64-bit bitmask for tracking enabled pads, so only subdevices > >> + * with 64 pads or less can be supported. > >> + */ > >> + if (pad >= sizeof(sd->enabled_pads) * BITS_PER_BYTE) > >> + return -EOPNOTSUPP; > >> + > >> if (!streams_mask) > >> return 0; > >> > >> /* Fallback on .s_stream() if .enable_streams() isn't available. */ > >> - if (!v4l2_subdev_has_op(sd, pad, enable_streams)) > >> - return v4l2_subdev_enable_streams_fallback(sd, pad, > >> - streams_mask); > >> + use_s_stream = !v4l2_subdev_has_op(sd, pad, enable_streams); > >> > >> - state = v4l2_subdev_lock_and_get_active_state(sd); > >> + if (!use_s_stream) > >> + state = v4l2_subdev_lock_and_get_active_state(sd); > >> + else > >> + state = NULL; > >> > >> /* > >> * Verify that the requested streams exist and that they are not > >> @@ -2231,9 +2213,18 @@ int v4l2_subdev_enable_streams(struct v4l2_subdev *sd, u32 pad, > >> > >> already_streaming = v4l2_subdev_is_streaming(sd); > >> > >> - /* Call the .enable_streams() operation. */ > >> - ret = v4l2_subdev_call(sd, pad, enable_streams, state, pad, > >> - streams_mask); > >> + if (!use_s_stream) { > >> + /* Call the .enable_streams() operation. */ > >> + ret = v4l2_subdev_call(sd, pad, enable_streams, state, pad, > >> + streams_mask); > >> + } else { > >> + /* Start streaming when the first pad is enabled. */ > >> + if (!already_streaming) > >> + ret = v4l2_subdev_call(sd, video, s_stream, 1); > >> + else > >> + ret = 0; > >> + } > >> + > >> if (ret) { > >> dev_dbg(dev, "enable streams %u:%#llx failed: %d\n", pad, > >> streams_mask, ret); > >> @@ -2243,34 +2234,32 @@ int v4l2_subdev_enable_streams(struct v4l2_subdev *sd, u32 pad, > >> /* Mark the streams as enabled. */ > >> v4l2_subdev_set_streams_enabled(sd, state, pad, streams_mask, true); > > > > state gets set to a non-null value above if the subdev has an > > enable_streams operation, while in v4l2_subdev_set_streams_enabled(), > > That's not quite true, v4l2_subdev_lock_and_get_active_state() returns > NULL if there's no state. If the driver has .enable_streams but no > subdev-state, we'll have NULL state. But is that a driver bug? Probably, > although I'm not quite sure. Whereas V4L2_SUBDEV_FL_STREAMS without > state is a driver bug. > > > you use it if the V4L2_SUBDEV_FL_STREAMS flag is set. The second > > condition is a subset of the first one, so it should be safe now, but I > > wonder if we're at risk of crashes in the future when reworking the > > code. I'm probably worrying needlessly. > > I agree that this is a bit messy. We have multiple different scenarios > which we handle in the same functions, and it's somewhat unclear what > op/state/flag combinations the drivers are allowed to have and what not. > > I could add a !state check here, and then use sd->enabled_pads. But that > doesn't feel correct, and I think it's just better to crash and then fix > it if we end up there. > > >> > >> - if (!already_streaming) > >> + if (!use_s_stream && !already_streaming) > >> v4l2_subdev_enable_privacy_led(sd); > > > > Once everybody switches to v4l2_subdev_enable_streams() (am I dreaming?) > > we should move LED handling from the .s_stream() wrapper to here for the > > fallback case too. How about a TODO comment ? > > Let's get back to this after we figure out if s_stream can ever be removed. > > >> > >> done: > >> - v4l2_subdev_unlock_state(state); > >> + if (!use_s_stream) > > > > if (state) > > > > would be better I think. > > We get the state and state lock if (!use_s_stream), so I think it makes > sense to use the same condition when unlocking to be symmetric. I'm sure > I can be convinced to use if (state) too, though =). > > Tomi > > >> + v4l2_subdev_unlock_state(state); > >> > >> return ret; > >> } > >> EXPORT_SYMBOL_GPL(v4l2_subdev_enable_streams); > >> > >> -static int v4l2_subdev_disable_streams_fallback(struct v4l2_subdev *sd, u32 pad, > >> - u64 streams_mask) > >> +int v4l2_subdev_disable_streams(struct v4l2_subdev *sd, u32 pad, > >> + u64 streams_mask) > >> { > >> struct device *dev = sd->entity.graph_obj.mdev->dev; > >> + struct v4l2_subdev_state *state; > >> + u64 enabled_streams; > >> + u64 found_streams; > >> + bool use_s_stream; > >> int ret; > >> > >> - /* > >> - * If the subdev doesn't implement pad-based stream enable, fall back > >> - * to the .s_stream() operation. > >> - */ > >> - if (!(sd->entity.pads[pad].flags & MEDIA_PAD_FL_SOURCE)) > >> - return -EOPNOTSUPP; > >> + /* A few basic sanity checks first. */ > >> + if (pad >= sd->entity.num_pads) > >> + return -EINVAL; > >> > >> - /* > >> - * .s_stream() means there is no streams support, so the only allowed > >> - * stream is the implicit stream 0. > >> - */ > >> - if (streams_mask != BIT_ULL(0)) > >> + if (!(sd->entity.pads[pad].flags & MEDIA_PAD_FL_SOURCE)) > >> return -EOPNOTSUPP; > >> > >> /* > >> @@ -2280,46 +2269,16 @@ static int v4l2_subdev_disable_streams_fallback(struct v4l2_subdev *sd, u32 pad, > >> if (pad >= sizeof(sd->enabled_pads) * BITS_PER_BYTE) > >> return -EOPNOTSUPP; > >> > >> - if (!(sd->enabled_pads & BIT_ULL(pad))) { > >> - dev_dbg(dev, "pad %u already disabled on %s\n", > >> - pad, sd->entity.name); > >> - return -EALREADY; > >> - } > >> - > >> - /* Stop streaming when the last streams are disabled. */ > >> - if (!(sd->enabled_pads & ~BIT_ULL(pad))) { > >> - ret = v4l2_subdev_call(sd, video, s_stream, 0); > >> - if (ret) > >> - return ret; > >> - } > >> - > >> - sd->enabled_pads &= ~BIT_ULL(pad); > >> - > >> - return 0; > >> -} > >> - > >> -int v4l2_subdev_disable_streams(struct v4l2_subdev *sd, u32 pad, > >> - u64 streams_mask) > >> -{ > >> - struct device *dev = sd->entity.graph_obj.mdev->dev; > >> - struct v4l2_subdev_state *state; > >> - u64 enabled_streams; > >> - u64 found_streams; > >> - int ret; > >> - > >> - /* A few basic sanity checks first. */ > >> - if (pad >= sd->entity.num_pads) > >> - return -EINVAL; > >> - > >> if (!streams_mask) > >> return 0; > >> > >> /* Fallback on .s_stream() if .disable_streams() isn't available. */ > >> - if (!v4l2_subdev_has_op(sd, pad, disable_streams)) > >> - return v4l2_subdev_disable_streams_fallback(sd, pad, > >> - streams_mask); > >> + use_s_stream = !v4l2_subdev_has_op(sd, pad, disable_streams); > >> > >> - state = v4l2_subdev_lock_and_get_active_state(sd); > >> + if (!use_s_stream) > >> + state = v4l2_subdev_lock_and_get_active_state(sd); > >> + else > >> + state = NULL; > >> > >> /* > >> * Verify that the requested streams exist and that they are not > >> @@ -2345,9 +2304,19 @@ int v4l2_subdev_disable_streams(struct v4l2_subdev *sd, u32 pad, > >> > >> dev_dbg(dev, "disable streams %u:%#llx\n", pad, streams_mask); > >> > >> - /* Call the .disable_streams() operation. */ > >> - ret = v4l2_subdev_call(sd, pad, disable_streams, state, pad, > >> - streams_mask); > >> + if (!use_s_stream) { > >> + /* Call the .disable_streams() operation. */ > >> + ret = v4l2_subdev_call(sd, pad, disable_streams, state, pad, > >> + streams_mask); > >> + } else { > >> + /* Stop streaming when the last streams are disabled. */ > >> + > >> + if (!(sd->enabled_pads & ~BIT_ULL(pad))) > >> + ret = v4l2_subdev_call(sd, video, s_stream, 0); > >> + else > >> + ret = 0; > >> + } > >> + > >> if (ret) { > >> dev_dbg(dev, "disable streams %u:%#llx failed: %d\n", pad, > >> streams_mask, ret); > >> @@ -2357,10 +2326,12 @@ int v4l2_subdev_disable_streams(struct v4l2_subdev *sd, u32 pad, > >> v4l2_subdev_set_streams_enabled(sd, state, pad, streams_mask, false); > >> > >> done: > >> - if (!v4l2_subdev_is_streaming(sd)) > >> - v4l2_subdev_disable_privacy_led(sd); > >> + if (!use_s_stream) { > >> + if (!v4l2_subdev_is_streaming(sd)) > >> + v4l2_subdev_disable_privacy_led(sd); > > > > Do we want to disable the privacy LED on failure in all cases, in > > particular when the .s_stream() or .disable_streams() operations are not > > even called ? > > > > And now that I wrote that, I realize the !v4l2_subdev_is_streaming() > > condition will not be true if the operations failed, so it should be > > fine. > > > > Reviewed-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx> > > > >> > >> - v4l2_subdev_unlock_state(state); > >> + v4l2_subdev_unlock_state(state); > >> + } > >> > >> return ret; > >> } > >> > > > -- Regards, Laurent Pinchart