https://bugs.kde.org/show_bug.cgi?id=368907
--- Comment #38 from Mauro Carvalho Chehab <mchehab+...@kernel.org> --- Created attachment 106506 --> https://bugs.kde.org/attachment.cgi?id=106506&action=edit Disable notifier if no data (In reply to Jan Palus from comment #36) > Looks like it's still the same, but I'm not really sure whether we are on > the same page -- writeToPipe() is not looping. It is rather called a lot > once the bug is triggered. I've made following small adjustment to its code: > > void DvbLiveViewInternal::writeToPipe() > { > + static int notEmptyCnt = 0; > + static int emptyCnt = 0; > + if (buffers.isEmpty()) { > + ++emptyCnt; > + } else { > + ++notEmptyCnt; > + } > + printf("empty: %d, not empty: %d\n", emptyCnt, notEmptyCnt); > while (!buffers.isEmpty()) { > const QByteArray ¤tBuffer = buffers.at(0); > int bytesWritten = int(write(writeFd, currentBuffer.constData(), > currentBuffer.size())); > > That's output after 10 seconds "without bug": > > empty: 0, not empty: 614 > > Now after bug is triggered empty counter skyrockets to some crazy value > after 10 secs (and continues which is most probably the cause of high CPU > usage): > > empty: 5053133, not empty: 534 > > as if notifications about pipe being ready for writing were suddenly coming > very very frequently Ah, OK. Well, I guess all we need to do is to disable the notifier to solve this issue. Still, I think that the first patch will make it more reliable. Please test with both the previous patch and this one. -- You are receiving this mail because: You are watching all bug changes.
