Updated GPU firmware

[mba]

Would the picture from the pandora crash help at all? It also printed what looks like a trace dump of some kind

I think that's okay. I know what's happening with that, but I need the VCHIQ guy to suggest the correct fix. I believe it only occurs when killing an application that is playing audio.

Hi Dom,

thanks for the update. I had the same crash as thogue when starting SqueezePlay. First time it crashed immediately. Second time it couldn't initialise alsa and the third time it played for 2minutes before crashing.
Undefined instruction vchiq_core.c:3183 the PC is pointing to an instrcution in vchiq_bulk_transmit.
(I also have a poor-quality photo of the dump taken with my phone if you need it )

I'm hoping the changes will solve dropouts in audio I have when using SqueezePlay. You never now

[dom]

@firmware updaters: Why not add a changed file and place it in the boot directory? Even if it's just a tweet.

If you want a changelist then:
https://github.com/raspberrypi/firmware/commits/master

[dom]

Using 240/16
On another note: The error message stopped after 18 mins of uptime. It was repeated 9 times.

Okay, the 240M build was broken. I've just pushed out a fix. (No change to other splits).

[lupine]

Using 240/16
On another note: The error message stopped after 18 mins of uptime. It was repeated 9 times.

Okay, the 240M build was broken. I've just pushed out a fix. (No change to other splits).

I am still getting the same timeout messages after applying the fix.

[kulve]

If you want a changelist then:
https://github.com/raspberrypi/firmware/commits/master

I also realized that I can enable watch in github to get email notifications from changes and I hope that resolves my notification need.

[joan]

...
If you want a changelist then:
https://github.com/raspberrypi/firmware/commits/master

I think it is easier to remember that a change list will be in /boot of your Pi.

[Frank B]

I am still getting the same timeout messages after applying the fix.

me too.

[dom]

I just locked up twice in a row in pianobar running inside lxterminal (have not changed any audio settings/configuration)

went headless still locked up

I took a picture but the jist :
kernel BUG at drivers/misc/vc04_services/interface/vchiq_arm/vchiq_core.c:3183
internal error: Oops - undefined instruction 0 [#1] PREEMPT

I've removed the BUG() call in latest update. It seems to allow killing ALSA apps without killing kernel.

However I really need to talk to the VCHIQ guy on Monday to try and get some proper fixes.

[milhouse]

I am still getting the same timeout messages after applying the fix.

me too.

Yep, also here with 240/16 split (testing sdhci-bcm2708.cycle_delay=1000 with 750/255/+3 overclock settings)

Code: Select all

pi@raspberrypi ~ $ uname -a
Linux raspberrypi 3.2.27+ #238 PREEMPT Fri Oct 5 23:19:10 BST 2012 armv6l GNU/Linux

pi@raspberrypi ~ $ vcgencmd version
Oct  6 2012 15:41:08
Copyright (c) 2012 Broadcom
version 342154 (release)

pi@raspberrypi ~ $ tail -f /var/log/syslog
Oct  6 18:16:51 raspberrypi dbus[1929]: [system] Successfully activated service 'org.freedesktop.PolicyKit1'
Oct  6 18:16:51 raspberrypi dbus[1929]: [system] Successfully activated service 'org.freedesktop.ConsoleKit'
Oct  6 18:17:11 raspberrypi /USR/SBIN/CRON[2107]: (root) CMD (   cd / && run-parts --report /etc/cron.hourly)
Oct  6 18:20:10 raspberrypi kernel: [  241.172631] INFO: task VCHIQka-0:27 blocked for more than 120 seconds.
Oct  6 18:20:10 raspberrypi kernel: [  241.172655] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
Oct  6 18:20:10 raspberrypi kernel: [  241.172672] VCHIQka-0       D c037aa9c     0    27      2 0x00000000
Oct  6 18:20:10 raspberrypi kernel: [  241.172754] [<c037aa9c>] (__schedule+0x2bc/0x568) from [<c0042e24>] (kthread+0x68/0x8c)
Oct  6 18:20:10 raspberrypi kernel: [  241.172805] [<c0042e24>] (kthread+0x68/0x8c) from [<c000e930>] (kernel_thread_exit+0x0/0x8)
^C

Timeout message has appeared just the once so far.

[mba]

I just locked up twice in a row in pianobar running inside lxterminal (have not changed any audio settings/configuration)

went headless still locked up

I took a picture but the jist :
kernel BUG at drivers/misc/vc04_services/interface/vchiq_arm/vchiq_core.c:3183
internal error: Oops - undefined instruction 0 [#1] PREEMPT

I've removed the BUG() call in latest update. It seems to allow killing ALSA apps without killing kernel.

However I really need to talk to the VCHIQ guy on Monday to try and get some proper fixes.

The new version gave me a new oops when using SqueezePlay. It started fine and played for 3sec and the I got:
Unable to handle kernel NULL ptr dereference at virtual address 000001b8

If you need any screendumps of the oops's just let me know.

[zardoz99]

Yep, this firmware is seriously broken. 240/16 split, headless, ssh drbd.

I can no longer keep either of my Raspis up for more than a couple of minutes before they crash with the hung task scenario.

I'm powering them off to avoid the crashes corrupting the file systems.

Role on Monday...

[milhouse]

Yep, this firmware is seriously broken. 240/16 split, headless, ssh drbd.

I can no longer keep either of my Raspis up for more than a couple of minutes before they crash with the hung task scenario.

I'm powering them off to avoid the crashes corrupting the file systems.

Role on Monday...

240/16 split seems to have been fixed - have you picked up the latest version (see uname/version details above). I've been testing Raspbian using 240/16, headless with ssh, without problems for the last hour or so. I get the timeout messages, but the system is otherwise stable.

[zardoz99]

Yes, I'm running the latest version....

Z.

[dom]

The new version gave me a new oops when using SqueezePlay. It started fine and played for 3sec and the I got:
Unable to handle kernel NULL ptr dereference at virtual address 000001b8

If you need any screendumps of the oops's just let me know.

Yes, that sounds like a new bug - screenshots of the stacktrace could be useful.

[fury]

Bit the bullet, updated the firmware, recompiled Quake 3

Maybe it's just my imagination, but I think I got a boost from 40 fps to 45 fps on high graphical quality.

Is there any chance of a firmware that allows control of the FPS cap or AA?

[mba]

The new version gave me a new oops when using SqueezePlay. It started fine and played for 3sec and the I got:
Unable to handle kernel NULL ptr dereference at virtual address 000001b8

If you need any screendumps of the oops's just let me know.

Yes, that sounds like a new bug - screenshots of the stacktrace could be useful.

Here you go, sorry for the quality but had to keep it under 64kb.

[dom]

Here you go, sorry for the quality but had to keep it under 64kb.

The "good bit" is off the bottom of screen. Can you press space and let it scroll to bottom of stack trace?

[Pitey]

Hey dom,

i'm using Raspbian Wheezy with the RetroArch emulator. I updated the firmware yesterday and since then the games freeze after ~30 seconds. I tried different overclocks (Medium, None) and memory splits (192, 128).
After the freeze i can't even log on via ssh.

Is there a workaround or do i have to wait till there's a new update?

[schicki]

I also got serveral freezes since the firmware update yesterday.

When playing music with mpd (Music Player Daemon) the system just froze after a few seconds.
Then I tried sudo /etc/init.d/mpd stop after reboot to stop it from freezing, but it also resulted in a freeze.
Finally the pi didn't boot anymore. It looked liked it booted but then freezed at some point. I couldn't ssh into it anymore. After replacing the kernel.img with an older one it came up again. Maybe the kernel.img was corrupted with all the system freezes.

[kulve]

In general you may be able to get some kernel backtraces by logging in over ssh and then:

Code: Select all

cat /proc/kmsg

The kernel may lock up or freeze faster than it can write the backtrace over the network but at least it could provide an easier way for pasting those kernel dumps to this forum.

[Pitey]

Okay, now i could log in via ssh after the picture froze:

Code: Select all

pi@raspberrypi ~ $ sudo uname -a
Linux raspberrypi 3.2.27+ #244 PREEMPT Sat Oct 6 17:26:38 BST 2012 armv6l GNU/Linux

pi@raspberrypi ~ $ vcgencmd version
Oct  6 2012 15:42:12
Copyright (c) 2012 Broadcom
version 342154 (release)

pi@raspberrypi ~ $ sudo cat /proc/kmsg
<6>[  437.516095] vchiq: 0: prs BULK_TX_DONE@ffd9fe40 (31->21) unexpected

pi@raspberrypi ~ $ sudo tail -f /var/log/syslog
Oct  7 16:36:17 raspberrypi dbus[1778]: [system] Activating service name='org.freedesktop.ConsoleKit' (using servicehelper)
Oct  7 16:36:17 raspberrypi dbus[1778]: [system] Activating service name='org.freedesktop.PolicyKit1' (using servicehelper)
Oct  7 16:36:18 raspberrypi polkitd[1977]: started daemon version 0.105 using authority implementation `local' version `0.105'
Oct  7 16:36:18 raspberrypi dbus[1778]: [system] Successfully activated service'org.freedesktop.PolicyKit1'
Oct  7 16:36:18 raspberrypi dbus[1778]: [system] Successfully activated service'org.freedesktop.ConsoleKit'

[kulve]

I haven't tried this on earlier kernels but at least the latest one is able to download at full 11MB/s speed.

With scp it drops to ~3.7MB/s and with "scp -c blowfish" it's about 4.4MB/s. Both to memory instead of SD card, so it's completely CPU bound.

I tried those only once, so nothing too accurate or scientific here.

[kulve]

"cat /proc/kmsg" blocks and prints only messages from that point on, so keep it in a separate window. You can also run "dmesg" to see past kernel messages if the system is not frozen.

[mba]

Here you go, sorry for the quality but had to keep it under 64kb.

The "good bit" is off the bottom of screen. Can you press space and let it scroll to bottom of stack trace?

Its completely frozen and doesn't react to the keyboard. I tried it twice. I have no other idea of getting the dump, cat /var/kmesg doesn't work either over ssh.

[cult_hero13]

If my information is of any help, I updated to #244 PREEMPT yesterday. I'm running in headless mode with some network services running and see load consistently above 1.0 where it used to run near .5. top shows there is nothing in particular causing the high load, and nothing out of the ordinary running, though I can't say I've ever seen these running:

VCHIQka-0
VCHIQ-0
VCHIQr-0
VCHIQs-0

/pro/kmsg shows what others have posted:

<3>[ 1321.227931] INFO: task VCHIQka-0:27 blocked for more than 120 seconds.
<3>[ 1321.227955] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
<6>[ 1321.227971] VCHIQka-0 D c037aa8c 0 27 2 0x00000000
<4>[ 1321.228049] [<c037aa8c>] (__schedule+0x2bc/0x568) from [<c0042e24>] (kthread+0x68/0x8c)
<4>[ 1321.228096] [<c0042e24>] (kthread+0x68/0x8c) from [<c000e930>]
(kernel_thread_exit+0x0/0x8)

Other than the higher than the higher than normal load, everything seems to run fine. I've had no shutdowns other than what I've initiated.

3.2.27+ #244 PREEMPT Sat Oct 6 17:26:38 BST 2012 armv6l GNU/Linux

Oct 5 2012 23:28:10
Copyright (c) 2012 Broadcom
version 342141 (release)

240/16 split.