I think I narrowed down this problem a little bit.
When I set RCU_BOOST, it was quite easy to generate cpu stall, but the
system is not really hung. So I turned off the cpu stall checking by doing
echo "1" > /sys/module/rcutree/parameters/rcu_cpu_stall_suppress
at boot time.
Then I tried to start VMs. Sometime VM started fine, sometime VM hung, I
need to destroied. When the starting of VM hung, the WCHAN of qemu-kvm
was wait_rcu_gp. And when I did "virsh destroy id", the VM also hung,
qemu-kvm was stuck at rcu_barrier. It really has something to do with KVM.
Dong
On 7/6/12 2:20 AM, Dong Liu wrote:
I can quite reliably trigger this cpu stall error now. Just try to start
several KVM guests.
It seems the kvm progress uses too much CPU?
Since ps -efl showed that the priority of the kvm kernel thread was -20.
So I tried,
CONFIG_RCU_BOOST=y
CONFIG_RCU_BOOST_PRIO=50
CONFIG_RCU_BOOST_DELAY=500
But I still got the same error :(
Are there any way I can use to narrow down this error?
Is it possible to non-preemptible RCU in PREEMPT_RT kernel? I know it
will reduce the system's RT performance, but we can make sure the reason
of this cpu stall error is the preemptible RCU
Thanks,
Dong
On 6/27/12 11:22 PM, Dong Liu wrote:
Hi,
I'm running 3.4.3-rt11 and 3.2.18-rt29 on a system with Intel I7 860, I
noticed that with hyperthread enabled, I'm getting kernel error
rcu_preempt detected stalls on CPUs/tasks: { 3} (detected by 5, t=60002
jiffies)
But with hyperthread disabled, no such error.
Is preempt_rt not supposed to be used with hyperthread?
Thanks,
Dong
P.S.
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
