Real-Time Preemption, -RT-2.6.10-rc2-mm3-V0.7.31-15

Hi Ingo,

System hangs on my notebook with RT-0.7.31-15 when running "/sbin/hwclock --hctosys --localtime".
Cursor stops blinking, no response to SysRq-b.
But if I set "hardirq-preempt=0" in kernel command line, the problem is gone.

config is attached. Is there anything I can do to provide you more useful informantion?

Best Regards,
Wen-chien Jesse Sung

hm, how did you generate this .config file? Could you send me your /proc/config.gz or the .config that you have in your kernel tree?

Ingo

also, is this a vanilla -RT-2.6.10-rc2-mm3-V0.7.31-15 tree?

Ingo

Ah... Sorry, my fault, it's not a vanilla one.

I redo the test with a vanilla 2.6.10-rc2-mm3-V0.7.31-16 again, and it still hangs at hwclock. Here's the complete config file.

Best Regards,
Wen-chien Jesse Sung

ok, could you try the -17 kernel i've just uploaded to the usual place:

http://redhat.com/~mingo/realtime-preempt/

does this fix the lockup?

i believe the lockup is an interesting side-effect of threading IRQ#0:
any code within the kernel that loops on jiffies will produce a lockup, because it starves the timer IRQ thread.

The RTC driver had two such places, but i also found one in the IRQ-autodetect code. We want to eliminate such code anyway, and a lockup is certainly an effective way to detect it ;)

to debug such lockups in the future you can do:

	echo 1 > /proc/sys/kernel/debug_direct_keyboard
	/sbin/hwclock ...

and use the sysrq keys to get a stack dump of the lockup. NOTE: dont use the keyboard in this mode for too long, it can lock up.

Ingo

Yes, hwclock works fine with -17. Thanks! :)

Best Regards,
Wen-chien Jesse Sung