[Xenomai] Slow execution of RT task

Philippe Gerum rpm at xenomai.org
Wed Apr 16 16:23:43 CEST 2014


On 04/16/2014 11:43 AM, Jeroen Van den Keybus wrote:
>> Could you patch the code fragment in, enable CONFIG_IPIPE_TRACE, then get a
>> dump from /proc/ipipe/trace/frozen once the warning has triggered?
>> Maybe getting a bit more context by setting
>> /proc/ipipe/trace/backtrace_points to 2048 would help too. TIA,
>
> I've attached the (somewhat lengthy) trace.
>
> BTW I now loaded xeno_nucleus and xeno_native in succession. The
> modprobing of xeno_native again locked until a new ssh terminal was
> opened (typing chars in the waiting ssh client didn't help).
>

 From that trace, it looks like an initial warning is issued on the same 
CPU, before the second one (i.e. rcutree.c:480) is raised:

:    #func               -2040	  0.088  print_oops_end_marker+0x0 
(warn_slowpath_common+0x66)
:    #func               -2040	  0.090  printk+0x0 
(print_oops_end_marker+0x2e)
:|   #begin   0x80000001 -2040	  0.117  printk+0x66 
(print_oops_end_marker+0x2e)

The trace backlog is not deep enough to locate the offending spot, but 
the lines above typically close a WARN_ON() handling. Once this first 
warning is sent out, the CPU then waits for 2s before receiving a timer 
IRQ from the APIC:

:    #func               -2032! 1998.529  intel_idle+0x0 
(cpuidle_enter_state+0x40)
:|   #begin   0x000000ef   -34	  0.097  apic_timer_interrupt+0x6d 
(intel_idle+0xab)

Could you check your kernel logs for this initial warning? Maybe this 
would shed light, assuming the second bug may be a side-effect of the 
first one.

TIA,

-- 
Philippe.




More information about the Xenomai mailing list