Re: [PATCH 2/2] KVM : powerpc/booke: Allow debug interrupt injection to guest

[Alexander Graf <agraf@suse.de>]

On 01.07.14 12:30, Bharat.Bhushan@freescale.com wrote:
>
>> -----Original Message-----
>> From: Alexander Graf [mailto:agraf@suse.de]
>> Sent: Tuesday, July 01, 2014 3:42 PM
>> To: Bhushan Bharat-R65777; Wood Scott-B07421
>> Cc: kvm-ppc@vger.kernel.org; kvm@vger.kernel.org
>> Subject: Re: [PATCH 2/2] KVM : powerpc/booke: Allow debug interrupt injection to
>> guest
>>
>>
>> On 01.07.14 12:06, Bharat.Bhushan@freescale.com wrote:
>>>> -----Original Message-----
>>>> From: Alexander Graf [mailto:agraf@suse.de]
>>>> Sent: Tuesday, July 01, 2014 11:53 AM
>>>> To: Wood Scott-B07421
>>>> Cc: Bhushan Bharat-R65777; kvm-ppc@vger.kernel.org;
>>>> kvm@vger.kernel.org
>>>> Subject: Re: [PATCH 2/2] KVM : powerpc/booke: Allow debug interrupt
>>>> injection to guest
>>>>
>>>>
>>>>
>>>>> Am 30.06.2014 um 22:25 schrieb Scott Wood <scottwood@freescale.com>:
>>>>>
>>>>>> On Sun, 2014-06-29 at 23:38 -0500, Bhushan Bharat-R65777 wrote:
>>>>>>
>>>>>>> -----Original Message-----
>>>>>>> From: Wood Scott-B07421
>>>>>>> Sent: Friday, June 27, 2014 11:53 PM
>>>>>>> To: Bhushan Bharat-R65777
>>>>>>> Cc: agraf@suse.de; kvm-ppc@vger.kernel.org; kvm@vger.kernel.org
>>>>>>> Subject: Re: [PATCH 2/2] KVM : powerpc/booke: Allow debug
>>>>>>> interrupt injection to guest
>>>>>>>
>>>>>>>> On Fri, 2014-06-27 at 11:55 +0530, Bharat Bhushan wrote:
>>>>>>>> -    /* Force enable debug interrupts when user space wants to debug */
>>>>>>>> -    if (vcpu->guest_debug) {
>>>>>>>> +    /*
>>>>>>>> +     * Force enable debug interrupts when user space wants to debug
>>>>>>>> +     * and there is no debug interrupt pending for guest to handle.
>>>>>>>> +     */
>>>>>>>> +    if (vcpu->guest_debug && !kvmppc_core_pending_debug(vcpu)) {
>>>>>>> Are you trying to allow the guest to be simultaneously debugged by
>>>>>>> itself and by host userspace?  How does this work?
>>>>>> Not actually, Currently we are not partitioning debug resources
>>>>>> between host userspace and guest. In fact we do not emulate debug
>>>>>> registers for guest. But we want host userspace to pass the
>>>>>> interrupt to guest if it is not able to handle.
>>>>> I don't understand the logic here.  A debug interrupt should be
>>>>> injected when the programming model in the guest says that a debug
>>>>> interrupt should happen.  How can that occur currently?  If the
>>>>> guest didn't set up the debug registers and QEMU still can't handle
>>>>> the debug interrupt, that's a bug in QEMU (or KVM, or the hardware...).
>>>>> Injecting the interrupt into the guest just adds another bug on top of that.
>>>> I don't think QEMU should be aware of these limitations.
>>>>
>>>>>>>> #ifdef CONFIG_KVM_BOOKE_HV
>>>>>>>>          /*
>>>>>>>>           * Since there is no shadow MSR, sync MSR_DE into the
>>>>>>>> guest @@
>>>>>>>> -264,6 +272,16 @@ static void kvmppc_core_dequeue_watchdog(struct
>>>>>>>> kvm_vcpu
>>>>>>> *vcpu)
>>>>>>>>      clear_bit(BOOKE_IRQPRIO_WATCHDOG,
>>>>>>>> &vcpu->arch.pending_exceptions); }
>>>>>>>>
>>>>>>>> +static void kvmppc_core_queue_debug(struct kvm_vcpu *vcpu) {
>>>>>>>> +    kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DEBUG); }
>>>>>>>> +
>>>>>>>> +static void kvmppc_core_dequeue_debug(struct kvm_vcpu *vcpu) {
>>>>>>>> +    clear_bit(BOOKE_IRQPRIO_DEBUG,
>>>>>>>> +&vcpu->arch.pending_exceptions); }
>>>>>>> Is there currently no support for a guest debugging itself (i.e.
>>>>>>> guest_debug unset) on e500v2?
>>>>>> Yes, It is not yet supported (IACx/DACx/DBCR/DBSR/DSRRx are not yet
>>>> emulated).
>>>>> How is it useful to inject a debug exception into the guest, until
>>>>> these things are emulated?
>>>> We don't have to touch QEMU later then ;). But I agree that it would
>>>> make a lot of sense to enable guest debugging along the way - it can't be
>> that hard, no?
>>> Copy pasting my response in another email:
>>>
>>> "
>>> Ok, Till we add support for guest to used debug resource, can we say that
>> userspace will still try to inject debug interrupt (as it does not know guest
>> capability) to guest but KVM will:
>>>    - clear guest dbsr
>> Can we fake the value of DBSR that the guest sees?
> Yes; we can hack the dbsr emulation with a comment :)
> But why we want that?

I guess I don't fully grasp what you're proposing :).


Alex