Re: [Qemu-devel] [RFC PATCH] target-arm/psci.c: wake up sleeping CPUs (MTTCG)

[Alex Bennée]

Paolo Bonzini <address@hidden> writes:

> On 24/06/2015 19:18, Alex Bennée wrote:
>>>> >> @@ -196,6 +196,8 @@ void arm_handle_psci_call(ARMCPU *cpu)
>>>> >>          }
>>>> >>          target_cpu_class->set_pc(target_cpu_state, entry);
>>>> >>  
>>>> >> +        qemu_cond_signal(target_cpu_state->halt_cond);
>>> >
>>> > That's called qemu_cpu_kick(target_cpu_state). :)  The patch should be
>>> > acceptable now upstream, I think.
>> Oh so this might well fail in KVM too?
>> 
>> The qemu_cpu_kick does a qemu_cond_broadcast(cpu->halt_cond) which seems
>> a little excessive? Won't all sleeping CPUs wake up (and return to sleep)?
>
> On KVM (and I assume on MT-TCG), each CPU has a different halt_cond.

You are right of course, I got my sense the wrong way around. Given it
is per-cpu I wonder if you will ever have multiple threads waiting on
it?

Anyway I'll fix that up and re-submit after I've tested to see of these
test cases break current KVM.

>
> Paolo

-- 
Alex Bennée