Err, weirdly I thought I had updated this commit to remove the comment; lemme do that and resend. -a On Mon, 9 Nov 2020 at 10:03, Adrian Chadd <adrian@freebsd.org> wrote:

Fix receive path hangs if another thread closes down the hackrf receive whilst this buffer receive function is waiting to be woken up.

Now:

* Sleep for up to 100ms each time waiting for the cond to be kicked; * Check whether streaming is still enabled each time rather than only when the function is entered.

This fixes hangs where consumers like gqrx via gnuradio will do a stop_rx/start_rx very quickly to change something, and the buffer receive path is waiting for a buffer. --- lib/hackrf/hackrf_source_c.cc | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-)

diff --git a/lib/hackrf/hackrf_source_c.cc b/lib/hackrf/hackrf_source_c.cc index 662d04a..21656f9 100644 --- a/lib/hackrf/hackrf_source_c.cc +++ b/lib/hackrf/hackrf_source_c.cc @@ -30,6 +30,7 @@

#include <stdexcept> #include <iostream> +#include <chrono>

#include <gnuradio/io_signature.h>

@@ -203,8 +204,22 @@ int hackrf_source_c::work( int noutput_items, { std::unique_lock<std::mutex> lock(_buf_mutex);

- while (_buf_used < 3 && running) // collect at least 3 buffers - _buf_cond.wait( lock ); + // XXX adrian - tomorrow, turn this loop into a wait_for (1 second) + // and re-check dev.get / running ! + // Otherwise this will hang reasonably reliably if we race where + // receive is stopped in one thread, and we've already passed + // the hackrf_is_streaming() check; we'll never finish this loop + // and consumers (eg gqrx) will fail. + + while (_buf_used < 3 && running) { // collect at least 3 buffers + _buf_cond.wait_for( lock , std::chrono::milliseconds(100)); + + // Re-check whether the device has closed or stopped streaming + if ( _dev.get() ) + running = (hackrf_is_streaming( _dev.get() ) == HACKRF_TRUE); + else + running = false; + } }

if ( ! running ) -- 2.28.0