Fix a deadlock in ViEEncoder::DeliverFrame.

The deadlock can happen when using HW encoder. HW encoder calls
the encode complete callback on libjingle worker thread instead
of ViECaptureThread. The capture thread can hold VieEncoder::|data_cs_|
and wait for ModuleRtpRtcpImpl::|critical_section_module_ptrs_|.
When libjingle worker thread runs encode complete callback, it
can hold ModuleRtpRtcpImpl::|critical_section_module_ptrs_| and
wait for VieEncoder::|data_cs_|.

|default_rtp_rtcp_| is not guarded by |data_cs|. So move it out of
the critical section to avoid the deadlock.

BUG=chromium:352567
TEST=Run apprtc loopback on CrOS.
Run apprtc between CrOS and Linux.
Run vie_auto_test.

R=henrik.lundin@webrtc.org, pbos@webrtc.org, stefan@webrtc.org

Review URL: https://webrtc-codereview.appspot.com/10039004

git-svn-id: http://webrtc.googlecode.com/svn/trunk@5721 4adac7df-926f-26a2-2b94-8c16560cd09d
This commit is contained in:
wuchengli@chromium.org 2014-03-19 03:44:20 +00:00
parent 8b61011b6f
commit ac4b87c258

View File

@ -575,13 +575,13 @@ void ViEEncoder::DeliverFrame(int id,
ViEId(engine_id_, channel_id_),
"%s: %llu", __FUNCTION__,
video_frame->timestamp());
if (default_rtp_rtcp_->SendingMedia() == false) {
// We've paused or we have no channels attached, don't encode.
return;
}
{
CriticalSectionScoped cs(data_cs_.get());
time_of_last_incoming_frame_ms_ = TickTime::MillisecondTimestamp();
if (default_rtp_rtcp_->SendingMedia() == false) {
// We've paused or we have no channels attached, don't encode.
return;
}
if (EncoderPaused()) {
if (!encoder_paused_and_dropped_frame_) {
TRACE_EVENT_ASYNC_BEGIN0("webrtc", "EncoderPaused", this);