[libcamera-devel] [PATCH v1] v4l2_videodevice: Disable the watchdog timer when no buffers are queued

[Kieran Bingham]

Quoting Naushir Patuck (2022-05-05 11:15:44)
> Hi Kieran,
> 
> Thank you for the feedback!
> 
> On Thu, 5 May 2022 at 10:36, Kieran Bingham <kieran.bingham at ideasonboard.com>
> wrote:
> 
> > Hi Naush,
> >
> > Quoting Naushir Patuck via libcamera-devel (2022-05-05 09:48:24)
> > > Only enable/reset the watchdog timer when there are buffers queued in
> > the V4L2
> > > device. Otherwise, we may trigger spurious warnings when the watchdog
> > times out
> > > even if there are no buffers queued in the device.
> >
> > aha yes - I can see how that was tripping up on python interactive
> > sessions or otherwise underflowing when it's not at all a fault of the
> > V4L2VideoDevice.
> >
> >
> > >
> > > Signed-off-by: Naushir Patuck <naush at raspberrypi.com>
> > > ---
> > >  src/libcamera/v4l2_videodevice.cpp | 22 +++++++++++++++-------
> > >  1 file changed, 15 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/src/libcamera/v4l2_videodevice.cpp
> > b/src/libcamera/v4l2_videodevice.cpp
> > > index 5b4637b1a39e..430715afd554 100644
> > > --- a/src/libcamera/v4l2_videodevice.cpp
> > > +++ b/src/libcamera/v4l2_videodevice.cpp
> > > @@ -1662,8 +1662,11 @@ int V4L2VideoDevice::queueBuffer(FrameBuffer
> > *buffer)
> > >                 return ret;
> > >         }
> > >
> > > -       if (queuedBuffers_.empty())
> > > +       if (queuedBuffers_.empty()) {
> > >                 fdBufferNotifier_->setEnabled(true);
> > > +               if (watchdogDuration_)
> > > +
> >  watchdog_.start(std::chrono::duration_cast<std::chrono::milliseconds>(watchdogDuration_));
> > > +       }
> >
> > I guess this could also be set outside of the if (queuedBuffer_.empty())
> > - but it would artificially delay the watchdog every time a buffer was
> > queued. In the event that more than one buffer is required to be queued
> > (to satisfy internal requirements perhaps?) that might actually be
> > beneficial ... But I think either way is fine.
> >
> 
> It could... I thought it might be ever so slightly more logically correct
> if this
> was inside the if() block.  I'll keep it in there unless there are further
> objections.
> 
> 
> >
> > >
> > >         queuedBuffers_[buf.index] = buffer;
> > >
> > > @@ -1742,16 +1745,21 @@ FrameBuffer *V4L2VideoDevice::dequeueBuffer()
> > >                 return nullptr;
> > >         }
> > >
> > > -       if (watchdogDuration_.count())
> > > -
> >  watchdog_.start(std::chrono::duration_cast<std::chrono::milliseconds>(watchdogDuration_));
> > > -
> > >         cache_->put(buf.index);
> > >
> > >         FrameBuffer *buffer = it->second;
> > >         queuedBuffers_.erase(it);
> > >
> > > -       if (queuedBuffers_.empty())
> > > +       if (queuedBuffers_.empty()) {
> > >                 fdBufferNotifier_->setEnabled(false);
> > > +               watchdog_.stop();
> > > +       } else if (watchdogDuration_) {
> > > +               /*
> > > +                * Restart the watchdog timer if there are buffers still
> > queued
> > > +                * in the device.
> > > +                */
> > > +
> >  watchdog_.start(std::chrono::duration_cast<std::chrono::milliseconds>(watchdogDuration_));
> >
> > Why do we need/have all these casts? Is either watchdogDuration_ not a
> > duration? or is watchdog_.start() not accepting a duration? Either of
> > those would remove a lot of casting surely?
> >
> 
> This is a bit of a minor annoyance because of how I defined the underlying
> type
> of utils::Duration to a double. The std::chrono library forbids implicit
> casts
> from integer to float types.
> 
> 
> > watchdogDuration_ is a utils::Duration which stores std::nano, and I see
> > Timer start takes a std::chrono::milliseconds. I think it would make
> > sense to add a 'void start(std::chrono::duration)' to the timer class
> > and simplify these casts. But that doesn't have to be fixed in this
> > patch.
> >
> > Would you like to do that as a patch on top? If you don't want to let me
> > know and I'll handle it after this patch is merged.
> >
> 
> What we've talked about doing is to switch utils::Duration to use an integer
> underlying type, which would allow implicit conversions and remove the ugly
> casts like the above. I believe Laurent was still thinking this through as
> there
> were also plans to make utils::Duration part of the public API...

Ayeee - I see - not as simple and obvious as I thought it should be
(which explains why it isn't so ...)

Well - I don't think it should affect this patch anyway.

--
Kieran


> 
> Regards,
> Naush
> 
> 
> >
> >
> > Reviewed-by: Kieran Bingham <kieran.bingham at ideasonboard.com>
> >
> >
> > > +       }
> > >
> > >         buffer->metadata_.status = buf.flags & V4L2_BUF_FLAG_ERROR
> > >                                  ? FrameMetadata::FrameError
> > > @@ -1847,7 +1855,7 @@ int V4L2VideoDevice::streamOn()
> > >         }
> > >
> > >         state_ = State::Streaming;
> > > -       if (watchdogDuration_.count())
> > > +       if (watchdogDuration_ && !queuedBuffers_.empty())
> > >
> >  watchdog_.start(std::chrono::duration_cast<std::chrono::milliseconds>(watchdogDuration_));
> > >
> > >         return 0;
> > > @@ -1924,7 +1932,7 @@ void
> > V4L2VideoDevice::setDequeueTimeout(utils::Duration timeout)
> > >         watchdogDuration_ = timeout;
> > >
> > >         watchdog_.stop();
> > > -       if (watchdogDuration_.count() && state_ == State::Streaming)
> > > +       if (watchdogDuration_ && state_ == State::Streaming &&
> > !queuedBuffers_.empty())
> > >
> >  watchdog_.start(std::chrono::duration_cast<std::chrono::milliseconds>(timeout));
> > >  }
> > >
> > > --
> > > 2.25.1
> > >
> >