2.6.13-rt13 SMP crashes on boot

2.6.13-rt13 SMP crashes on boot

[John Rigg]

I've been trying to get 2.6.13 with RT-preempt patch to run on my
dual Opteron with debugging code (with CONFIG_PREEMPT_RT=y).
It crashes on boot if I enable latency tracing in .config. This happens 
with -rt13, -rt12, and -rt4 versions of the patch. Unfortunately it
occurs before any logging, so I can't provide any log data.
The crash appears to happen just after drive hda is
detected, while it's trying to mount my ext3 / filesystem (on hda7).
The mount fails then the screen fills with a segfault message in an
infinite loop:

segfault at ffffffff8010f20c rip ffffffff8010f20c rsp 00007fffffacdc88

A hard reboot is then required.

Without latency tracing it boots and runs fine with the following:

Kernel debug
	MagicSysRq
Detect soft lockups
Wakeup latency timing
Compile with debug info
Collect scheduler stats
Debug preemptible kernel

With latency tracing the crash occurs even if I disable all other
kernel hacking except for
Magic SysRq, Wakeup latency timing, and Compile w. debug info.

Hardware is 2 x Opteron 240, MSI K8T Master2-FAR (VIA K8T 800) m/board,
1GB ECC RAM. System is Debian amd64 unstable (gcc-4.0_4.0.1-7,
binutils_2.16.1cvs20050902-1).
Copy of .config and lspci output are here:
http://www.sound-man.co.uk/crash

Regards,
John

Re: 2.6.13-rt13 SMP crashes on boot

[Steven Rostedt]

Hi John,

First, always CC Ingo on all issues related to the -rt patch.  Although
I think he's on vacation now, but even so, he will probably miss this
email.

On Mon, 2005-09-19 at 19:54 +0100, John Rigg wrote:
> I've been trying to get 2.6.13 with RT-preempt patch to run on my
> dual Opteron with debugging code (with CONFIG_PREEMPT_RT=y).
> It crashes on boot if I enable latency tracing in .config. This happens 
> with -rt13, -rt12, and -rt4 versions of the patch. Unfortunately it

What was the last release that it worked on?  If you could try older
releases, that would really help to know when the problem occurred.
Lots of things have gone on recently since -rt1 and it would help if we
can pin point when this happened, so we would have a good idea of what
happened.
	
> occurs before any logging, so I can't provide any log data.

Do you have a serial output? Or compile your network driver into the
kernel and turn on netconsole.  Usually when someone says "before any
logging", I think of before there's output to the screen. But your next
statements seems to show that that is not the case.

> The crash appears to happen just after drive hda is
> detected, while it's trying to mount my ext3 / filesystem (on hda7).
> The mount fails then the screen fills with a segfault message in an
> infinite loop:

This may just be a 64bit issue as well somewhere.

-- Steve

Re: 2.6.13-rt13 SMP crashes on boot

[Thomas Gleixner]

On Mon, 2005-09-19 at 16:49 -0400, Steven Rostedt wrote:
> Hi John,
> 
> First, always CC Ingo on all issues related to the -rt patch.  Although
> I think he's on vacation now, but even so, he will probably miss this
> email.
> 
> On Mon, 2005-09-19 at 19:54 +0100, John Rigg wrote:
> > I've been trying to get 2.6.13 with RT-preempt patch to run on my
> > dual Opteron with debugging code (with CONFIG_PREEMPT_RT=y).
> > It crashes on boot if I enable latency tracing in .config. This happens 
> > with -rt13, -rt12, and -rt4 versions of the patch. Unfortunately it

Do you have high resolution timers turned on ?

If yes, there is a SMP bug in -rt13. -rt14 has fixed this

tglx