KVM guest interrupts, traps, and, exceptions

KVM guest interrupts, traps, and, exceptions

[Pekka Enberg]

Hi Avi,

I'm continuing my journey through the KVM API documentation and I'm now 
scratching my head on how interrupts, traps, and exceptions are 
supported by the API. I see KVM_CREATE_IRQCHIP and KVM_IRQ_LINE but I'm 
unable to work out how the host is notified of such events. For example, 
if the guest does

   int $0x10

is there some KVM_EXIT "exit reason" like we have for IO? Is there some 
other mechanism for that? As I haven't yet set up anything related to 
IRQs I can see the host just deadlock on user interrupts.

			Pekka

Re: KVM guest interrupts, traps, and, exceptions

[Alexander Graf]

On 27.03.2010, at 16:43, Pekka Enberg wrote:

> Hi Avi,
> 
> I'm continuing my journey through the KVM API documentation and I'm now scratching my head on how interrupts, traps, and exceptions are supported by the API. I see KVM_CREATE_IRQCHIP and KVM_IRQ_LINE but I'm unable to work out how the host is notified of such events. For example, if the guest does
> 
>  int $0x10

IRQs here mean hardware interrupts. They usually are pushed from host -> guest. So all you need is a way to tell the guest "you got an interrupt". Your userspace is the one giving interrupts, so there's no need to an exit.

The "int $0x10" instruction describes a software interrupt. These are handled 100% in guest space.

> is there some KVM_EXIT "exit reason" like we have for IO? Is there some other mechanism for that? As I haven't yet set up anything related to IRQs I can see the host just deadlock on user interrupts.

I don't think I quite understand? If there's anything running inside guest context the host shouldn't be affected at all. Host interrupts always intercept guest execution.


Alex

Re: KVM guest interrupts, traps, and, exceptions

[Pekka Enberg]

Hi Alexander,

Alexander Graf wrote:
>> I'm continuing my journey through the KVM API documentation and I'm now scratching my head on how interrupts, traps, and exceptions are supported by the API. I see KVM_CREATE_IRQCHIP and KVM_IRQ_LINE but I'm unable to work out how the host is notified of such events. For example, if the guest does
>>
>>  int $0x10
> 
> IRQs here mean hardware interrupts. They usually are pushed from host -> guest. So all you need is a way to tell the guest "you got an interrupt". Your userspace is the one giving interrupts, so there's no need to an exit.
> 
> The "int $0x10" instruction describes a software interrupt. These are handled 100% in guest space.
> 
>> is there some KVM_EXIT "exit reason" like we have for IO? Is there some other mechanism for that? As I haven't yet set up anything related to IRQs I can see the host just deadlock on user interrupts.
> 
> I don't think I quite understand? If there's anything running inside guest context the host shouldn't be affected at all. Host interrupts always intercept guest execution.

OK, so Avi asked me not to look at KVM code or qemu code if I don't 
understand something in Documentation/kvm/api.txt and instead, complain 
about it. That might also partly explain why I am so confused.

I'm not talking in host -> guest interrupts but interrupts happening in 
the _guest_. Now if I understand your explanation correctly, host is not 
involved in interrupt processing but instead I just need to set up IDTs 
properly in the guest? That makes sense and explains why guest 
interrupts are not mentioned in the KVM API documentation at all. ;-)

			Pekka

Re: KVM guest interrupts, traps, and, exceptions

[Alexander Graf]

On 27.03.2010, at 17:42, Pekka Enberg wrote:

> Hi Alexander,
> 
> Alexander Graf wrote:
>>> I'm continuing my journey through the KVM API documentation and I'm now scratching my head on how interrupts, traps, and exceptions are supported by the API. I see KVM_CREATE_IRQCHIP and KVM_IRQ_LINE but I'm unable to work out how the host is notified of such events. For example, if the guest does
>>> 
>>> int $0x10
>> IRQs here mean hardware interrupts. They usually are pushed from host -> guest. So all you need is a way to tell the guest "you got an interrupt". Your userspace is the one giving interrupts, so there's no need to an exit.
>> The "int $0x10" instruction describes a software interrupt. These are handled 100% in guest space.
>>> is there some KVM_EXIT "exit reason" like we have for IO? Is there some other mechanism for that? As I haven't yet set up anything related to IRQs I can see the host just deadlock on user interrupts.
>> I don't think I quite understand? If there's anything running inside guest context the host shouldn't be affected at all. Host interrupts always intercept guest execution.
> 
> OK, so Avi asked me not to look at KVM code or qemu code if I don't understand something in Documentation/kvm/api.txt and instead, complain about it. That might also partly explain why I am so confused.

Oh I see :-). That's an eager goal.

> I'm not talking in host -> guest interrupts but interrupts happening in the _guest_. Now if I understand your explanation correctly, host is not involved in interrupt processing but instead I just need to set up IDTs properly in the guest? That makes sense and explains why guest interrupts are not mentioned in the KVM API documentation at all. ;-)

Heh, yep. SW interrupts run 100% in guest context. Interrupt handling (IDT interpretation etc.) also happens in guest context.


Alex