MythTV gets stuck tuning going from IVTV -> DVB

[Mark Buechler <Mark.Buechler@…>]

MythTV gets stuck when changing from an IVTV channel to a DVB channel. The frontend times out on socket communications and will not work until the backend is kicked.

Backtrace, frontend and backend logs attached. The backtrace log has a couple backtraces in it as well as some stepping.

[Mark Buechler <Mark.Buechler@…>]

Backend log

[Mark Buechler <Mark.Buechler@…>]

Backend log

[Mark Buechler <Mark.Buechler@…>]

Frontend log

[Mark Buechler <Mark.Buechler@…>]

I believe thread 8 is the culprit here. It looks to be spinning on ThreadedFileWriter::Write() on "while (count > BufFree?())"

Could tearing down the RingBuffer? cause a scenario where count will always be > BufFree?()?

• Mark.

[Shane Shrybman <gnome42@…>]

Hi Mark,

I can't reproduce the problem and I haven't analyzed the actual cause of this too closely but it doesn't seem inconceivable that you could get stuck in that loop.

Patch attached to safeguard against that.

[Mark Buechler <Mark.Buechler@…>]

Several backtraces..

[Mark Buechler <Mark.Buechler@…>]

It seems there's a few things going on here. Attached a log of several backtraces. I'll analyze it tomorrow when I'm more awake.

• Mark.

[laga]

Mark,

did you ever get around to looking into this?

[Mark Buechler <Mark.Buechler@…>]

Yes, I believe it's due to reclist_lock getting locked in the scheduler and never getting released. In trying to narrow it down, I added a bunch of debug printf's, but they changing the timing of events ever so slightly as to nearly prevent the issue altogether.

• Mark.

[gnome42]

From the bt of thread 8

RingBuffer::Write (this=0xa715b058, buf=0xa205e008, count=2211632) at RingBuffer?.cpp:1182

ThreadedFileWriter::TFW_DEF_BUF_SIZE = 2*1024*1024 => 2097152

2211632 > 2097152 so the (count > BufFree?()) condition will never clear.

This is a duplicate of #5591 fixed in trunk by [18105]