On 18.02.2012, at 00:00, Scott Wood wrote:
> On 02/17/2012 11:13 AM, Alexander Graf wrote:
>> Instead of checking whether we should reschedule only when we exited
>> due to an interrupt, let's always check before entering the guest back
>> again. This gets the target more in line with the other archs.
>>
>> Signed-off-by: Alexander Graf <agraf@xxxxxxx>
>> ---
>> arch/powerpc/kvm/booke.c | 15 ++++++++++-----
>> 1 files changed, 10 insertions(+), 5 deletions(-)
>>
>> diff --git a/arch/powerpc/kvm/booke.c b/arch/powerpc/kvm/booke.c
>> index bfb2092..de30b6d 100644
>> --- a/arch/powerpc/kvm/booke.c
>> +++ b/arch/powerpc/kvm/booke.c
>> @@ -572,6 +572,7 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
>> unsigned int exit_nr)
>> {
>> int r = RESUME_HOST;
>> + int resched_needed = 1;
>>
>> /* update before a new last_exit_type is rewritten */
>> kvmppc_update_timing_stats(vcpu);
>> @@ -602,25 +603,21 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
>>
>> switch (exit_nr) {
>> case BOOKE_INTERRUPT_MACHINE_CHECK:
>> - kvm_resched(vcpu);
>> r = RESUME_GUEST;
>> break;
>>
>> case BOOKE_INTERRUPT_EXTERNAL:
>> kvmppc_account_exit(vcpu, EXT_INTR_EXITS);
>> - kvm_resched(vcpu);
>> r = RESUME_GUEST;
>> break;
>>
>> case BOOKE_INTERRUPT_DECREMENTER:
>> kvmppc_account_exit(vcpu, DEC_EXITS);
>> - kvm_resched(vcpu);
>> r = RESUME_GUEST;
>> break;
>>
>> case BOOKE_INTERRUPT_DOORBELL:
>> kvmppc_account_exit(vcpu, DBELL_EXITS);
>> - kvm_resched(vcpu);
>> r = RESUME_GUEST;
>> break;
>>
>> @@ -869,8 +866,16 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
>> BUG();
>> }
>>
>> - local_irq_disable();
>> + /* make sure we reschedule if we need to */
>> + while (resched_needed) {
>> + local_irq_disable();
>>
>> + resched_needed = need_resched();
>> + if (resched_needed) {
>> + local_irq_enable();
>> + cond_resched();
>> + }
>> + }
>> kvmppc_core_prepare_to_enter(vcpu);
>>
>> if (!(r & RESUME_HOST)) {
>
> kvmppc_core_prepare_to_enter can enable interrupts (and block) if guest
> MSR_WE is set. We may take an interrupt that wants a resched after
> waking but before interrupts are disabled again.
>
> We also want to check for a resched in kvmppc_vcpu_run. So, the resched
> check belongs in kvmppc_core_prepare_to_enter, something like:
>
> /* Check pending exceptions and deliver one, if possible. */
> void kvmppc_core_prepare_to_enter(struct kvm_vcpu *vcpu)
> {
> WARN_ON_ONCE(!irqs_disabled());
>
> while (true) {
> if (signal_pending(current))
> break;
>
> if (need_resched()) {
> local_irq_enable();
> cond_resched();
> local_irq_disable();
> continue;
> }
>
> kvmppc_core_check_exceptions(vcpu);
>
> if (vcpu->arch.shared->msr & MSR_WE) {
> local_irq_enable();
> kvm_vcpu_block(vcpu);
> local_irq_disable();
>
> kvmppc_set_exit_type(vcpu,
> EMULATED_MTMSRWE_EXITS);
> continue;
> }
>
> break;
> }
> }
>
> It would be simpler (both here and in the idle hcall) if we could just
> drop support for CONFIG_PREEMPT=n. :-P
When running with CONFIG_PREEMPT=n we don't have to worry about interrupts being enabled though, since we only preempt on known good checkpoints, right? While for CONFIG_PREEMPT=y we can always get preempted, rendering most of these checks void.
So essentially we could just be lazy and do a "best effort" resched check, but not worry about races wrt guest entry/exit, right? And for real preemption modes we don't have to worry about any of the resched stuff IIUC, correct?
Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
