When write to MMIO happens and there is an ioeventfd for that and
is handled successfully, ioeventfd_write() returns 0 (success) and
kvmppc_handle_store() returns EMULATE_DONE. Then kvmppc_emulate_mmio()
converts EMULATE_DONE to RESUME_GUEST_NV and this broke from the loop.
This adds handling of RESUME_GUEST_NV in kvmppc_vcpu_run_hv().
Cc: Michael S. Tsirkin <[email protected]>
Suggested-by: Paul Mackerras <[email protected]>
Signed-off-by: Alexey Kardashevskiy <[email protected]>
---
This definitely needs a better commit message. Please, help.
ps. it seems like ioeventfd never worked on ppc64. hm.
---
arch/powerpc/kvm/book3s_hv.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 072287f..24f363f 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -1569,7 +1569,7 @@ static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
}
- } while (r == RESUME_GUEST);
+ } while ((r == RESUME_GUEST_NV) || (r == RESUME_GUEST));
out:
vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
--
1.8.4.rc4
On 10.01.2014, at 08:21, Alexey Kardashevskiy <[email protected]> wrote:
> When write to MMIO happens and there is an ioeventfd for that and
> is handled successfully, ioeventfd_write() returns 0 (success) and
> kvmppc_handle_store() returns EMULATE_DONE. Then kvmppc_emulate_mmio()
> converts EMULATE_DONE to RESUME_GUEST_NV and this broke from the loop.
>
> This adds handling of RESUME_GUEST_NV in kvmppc_vcpu_run_hv().
>
> Cc: Michael S. Tsirkin <[email protected]>
> Suggested-by: Paul Mackerras <[email protected]>
> Signed-off-by: Alexey Kardashevskiy <[email protected]>
> ---
>
> This definitely needs a better commit message. Please, help.
> ps. it seems like ioeventfd never worked on ppc64. hm.
>
> ---
> arch/powerpc/kvm/book3s_hv.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
> index 072287f..24f363f 100644
> --- a/arch/powerpc/kvm/book3s_hv.c
> +++ b/arch/powerpc/kvm/book3s_hv.c
> @@ -1569,7 +1569,7 @@ static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
> vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
> srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
> }
> - } while (r == RESUME_GUEST);
> + } while ((r == RESUME_GUEST_NV) || (r == RESUME_GUEST));
How about
while(!(r & RESUME_FLAG_HOST));
That should cover all RESUME_GUEST_XXX cases just fine. Apart from that I agree that we should check for ! FLAG_HOST bit rather than the actual RESUME_GUEST value in all case where we check for it (read: please update all places).
Alex
On 01/13/2014 02:44 AM, Alexander Graf wrote:
>
> On 10.01.2014, at 08:21, Alexey Kardashevskiy <[email protected]> wrote:
>
>> When write to MMIO happens and there is an ioeventfd for that and
>> is handled successfully, ioeventfd_write() returns 0 (success) and
>> kvmppc_handle_store() returns EMULATE_DONE. Then kvmppc_emulate_mmio()
>> converts EMULATE_DONE to RESUME_GUEST_NV and this broke from the loop.
>>
>> This adds handling of RESUME_GUEST_NV in kvmppc_vcpu_run_hv().
>>
>> Cc: Michael S. Tsirkin <[email protected]>
>> Suggested-by: Paul Mackerras <[email protected]>
>> Signed-off-by: Alexey Kardashevskiy <[email protected]>
>> ---
>>
>> This definitely needs a better commit message. Please, help.
>> ps. it seems like ioeventfd never worked on ppc64. hm.
>>
>> ---
>> arch/powerpc/kvm/book3s_hv.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
>> index 072287f..24f363f 100644
>> --- a/arch/powerpc/kvm/book3s_hv.c
>> +++ b/arch/powerpc/kvm/book3s_hv.c
>> @@ -1569,7 +1569,7 @@ static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
>> vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
>> srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
>> }
>> - } while (r == RESUME_GUEST);
>> + } while ((r == RESUME_GUEST_NV) || (r == RESUME_GUEST));
>
> How about
>
> while(!(r & RESUME_FLAG_HOST));
Rather "while(!(r & RESUME_FLAG_HOST) && (r > 0));" and still not obvious
that this is really better.
Paul agrees with the original patch (and made a better commit message for
our internal tree) but I just cannot make him reply in this thread, keep
constantly asking him but to no avail :)
> That should cover all RESUME_GUEST_XXX cases just fine. Apart from that
> I agree that we should check for ! FLAG_HOST bit rather than the actual
> RESUME_GUEST value in all case where we check for it (read: please
> update all places).
There are 3 places remotely similar to this and none of them requires a fix
like above.
--
Alexey
On 04.02.2014, at 09:03, Alexey Kardashevskiy <[email protected]> wrote:
> On 01/13/2014 02:44 AM, Alexander Graf wrote:
>>
>> On 10.01.2014, at 08:21, Alexey Kardashevskiy <[email protected]> wrote:
>>
>>> When write to MMIO happens and there is an ioeventfd for that and
>>> is handled successfully, ioeventfd_write() returns 0 (success) and
>>> kvmppc_handle_store() returns EMULATE_DONE. Then kvmppc_emulate_mmio()
>>> converts EMULATE_DONE to RESUME_GUEST_NV and this broke from the loop.
>>>
>>> This adds handling of RESUME_GUEST_NV in kvmppc_vcpu_run_hv().
>>>
>>> Cc: Michael S. Tsirkin <[email protected]>
>>> Suggested-by: Paul Mackerras <[email protected]>
>>> Signed-off-by: Alexey Kardashevskiy <[email protected]>
>>> ---
>>>
>>> This definitely needs a better commit message. Please, help.
>>> ps. it seems like ioeventfd never worked on ppc64. hm.
>>>
>>> ---
>>> arch/powerpc/kvm/book3s_hv.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
>>> index 072287f..24f363f 100644
>>> --- a/arch/powerpc/kvm/book3s_hv.c
>>> +++ b/arch/powerpc/kvm/book3s_hv.c
>>> @@ -1569,7 +1569,7 @@ static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
>>> vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
>>> srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
>>> }
>>> - } while (r == RESUME_GUEST);
>>> + } while ((r == RESUME_GUEST_NV) || (r == RESUME_GUEST));
>>
>> How about
>>
>> while(!(r & RESUME_FLAG_HOST));
>
>
> Rather "while(!(r & RESUME_FLAG_HOST) && (r > 0));" and still not obvious
> that this is really better.
>
> Paul agrees with the original patch (and made a better commit message for
> our internal tree) but I just cannot make him reply in this thread, keep
> constantly asking him but to no avail :)
>
>
>> That should cover all RESUME_GUEST_XXX cases just fine. Apart from that
>> I agree that we should check for ! FLAG_HOST bit rather than the actual
>> RESUME_GUEST value in all case where we check for it (read: please
>> update all places).
>
> There are 3 places remotely similar to this and none of them requires a fix
> like above.
Not today. Then someone goes in and realizes that CEDE advanced version 8 should clear r15 at which point you have to set RESUME_GUEST_NV and are in the same mess again. Let's just always treat RESUME_GUEST_NV and RESUME_GUEST as identical when checking for it.
If you like, add a small helper like
static inline bool is_resume_guest(int r) {
return (r == RESUME_GUEST || r == RESUME_GUEST_NV);
}
in a header and use that one instead. That way we're guaranteed to be consistent.
Alex