emuxki/midi related KDLs?

[jonas.kirilla]

Two different KDL panics recently, (r24336?) happening at late bootup.

thread: media_server_addon - unexpected exception "invalid opcode exception"

thread: MidiPortProducer - unhandled page fault in kernel space

See screenshots. Look for "midi". Coincidence?

[jonas.kirilla]

/Users/kirilla/Desktop/DSC03665.jpg

[jonas.kirilla]

Sorry for all the confusing, disordered, undescribed screenshots. :( Most of them are almost identical versions of "invalid opcode exception" in the media_server_addon, with DSC03683.jpg and DSC03684.jpg being page faults instead. The one exception being the unhandled page fault of MidiPortProducer in DSC03675.jpg, which I suppose could be unrelated. These KDLs started appearing around the same time. Before there had been none.

[korli]

Is this new ? I don't understand well why the midi dev entry of emuxki is opened by the multiaudio addon ...

[stippi]

Are you already using an image which has the fixed VideoWindow media add-on already?

[jonas.kirilla]

Replying to stippi:

Are you already using an image which has the fixed VideoWindow media add-on already?

If you're referring to r24273 (5 days ago): yes, I'm way past that rev. If you're referring to r24362 (2 hours ago): no, I have not tried that revision yet.

[jonas.kirilla]

The "invalid opcode exception" still happens in r24361. Will try a later revision ASAP.

[jonas.kirilla]

It happens also with r24371.

[korli]

Could you activate TRACE_DEVFS in src/system/kernel/fs/devfs.cpp and provide a serial log ?

The emuxki driver has this peculiarity : it publishes two sets of entries in dev/audio/hmultiaudio and dev/midi.

[jonas.kirilla]

syslog - Sun, 2008 Mar 23, 19:59:07

[jonas.kirilla]

syslog.old.zip - Sun, 2008 Mar 23, 19:58:36

[jonas.kirilla]

I hope syslog and syslog.old is good enough. The serial output won't fit in SerialConnect's buffer. Syslog.old was too large for Trac, so I had to zip it. Sorry about that. Let me know if you want me to try something else or find a better log.

BTW, having enabled TRACE_DEVFS, moving the mouse results in exceptionally large CPU load, kernel side, in 2 threads, one being the USB mouse watcher thread in the input_server and the other being the multi addon thread in the media_addon_server, the one KDL:ing in screenshot: http://dev.haiku-os.org/attachment/ticket/1909/DSC03665.jpg I find this strange.

[korli]

Could you try again with r24597 or superior ?

[jonas.kirilla]

Having rebooted Haiku maybe 20 times (shutdown -r in UserBootscript) it didn't happen, so I think it's fixed. Thanks Jérôme!

[korli]

Fixed in r24597. Thanks for testing!