On 16/02/2015 16:09, Gerhard Wiesinger wrote:
> On 16.02.2015 15:18, Paolo Bonzini wrote:
>>
>> On 15/02/2015 09:18, Gerhard Wiesinger wrote:
>>>
>>> Can you grab some random backtraces ("thread apply all bt full") with
>>> gdb?
>
> Very low load on the machine, so I guess most will be sleeping and hard
> to catch non sleeping ones. See below
This likely means that the 100% threads are not the I/O (event loop)
threads, but the VCPU threads.
>>> Kernel (host/guest): 3.18.6-200.fc21.x86_64 #1 SMP
>>> qemu-kvm-2.2.0-5.fc21.x86_64
>>>
>>> Bug 1178975 - endless loop in clock_gettime() on a kvm-based VM
>>> https://bugzilla.redhat.com/show_bug.cgi?id=1178975
>>> is fixed (didn't occour with the test program posted at
>>> https://bugzilla.redhat.com/show_bug.cgi?id=1178975#c28 in 30min,
>>> happened before reproduceable in 2min, still running)
>>>
>>> So I guess there is another problem in the kernel with volatile and gcc
>>> optimizations (or maybe in qemu-KVM)
>> No, this doesn't look like volatile.
>
> But why 100% on 2 cores?
Because you have 2 VCPUs, I guess.
Paolo
