Re: [PATCH v2 2/2] qemu-kvm: x86: Add support for event states

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 2985 bytes --]

Jan Kiszka wrote:
> This patch extends the qemu-kvm state sync logic with the event substate
> from the new VCPU state interface, giving access to yet missing
> exception, interrupt and NMI states.
> 
> The patch  does not switch the rest of qemu-kvm's code to the new
> interface as it is expected to be morphed into upstream's version
> anyway. Instead, a full conversion will be submitted for upstream.
> 
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
> 
> Changes from v1:
>  - added support for has_error_code
>  - properly clear exception_index and interrupt_injected in case the
>    kernel does not support KVM_GET_VCPU_STATE
> 
>  qemu-kvm-x86.c        |   86 +++++++++++++++++++++++++++++++++++++++++++++++++
>  target-i386/cpu.h     |    5 +++
>  target-i386/machine.c |    5 +++
>  3 files changed, 96 insertions(+), 0 deletions(-)
> 
> diff --git a/qemu-kvm-x86.c b/qemu-kvm-x86.c
> index e03a4ba..dc3e6cb 100644
> --- a/qemu-kvm-x86.c
> +++ b/qemu-kvm-x86.c
> @@ -903,6 +903,82 @@ static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
>  	| (rhs->avl * DESC_AVL_MASK);
>  }
>  
> +static void kvm_get_events(CPUState *env)
> +{
> +#ifdef KVM_CAP_VCPU_STATE
> +    struct {
> +        struct kvm_vcpu_state header;
> +        struct kvm_vcpu_substate substates[1];
> +    } request;
> +    struct kvm_x86_event_state events;
> +    int r;
> +
> +    request.header.nsubstates = 1;
> +    request.header.substates[0].type = KVM_X86_VCPU_STATE_EVENTS;
> +    request.header.substates[0].offset = (size_t)&events - (size_t)&request;
> +    r = kvm_vcpu_ioctl(env, KVM_GET_VCPU_STATE, &request);
> +    if (r == 0) {
> +        env->exception_index =
> +            events.exception.injected ? events.exception.nr : -1;
> +        env->has_error_code = events.exception.has_error_code;
> +        env->error_code = events.exception.error_code;
> +
> +        env->interrupt_injected =
> +            events.interrupt.injected ? events.interrupt.nr : -1;
> +        env->soft_interrupt = events.interrupt.soft;
> +
> +        env->nmi_injected = events.nmi.injected;
> +        env->nmi_pending = events.nmi.pending;
> +        if (events.nmi.masked) {
> +            env->hflags2 |= HF2_NMI_MASK;
> +        } else {
> +            env->hflags2 &= ~HF2_NMI_MASK;
> +        }
> +
> +        env->sipi_vector = events.sipi_vector;
> +
> +        return;
> +    }
> +#endif
> +    env->exception_index = -1;
> +    env->interrupt_injected = -1;
> +    env->nmi_injected = 0;
> +    env->nmi_pending = 0;
> +    env->hflags2 &= ~HF2_NMI_MASK;

Frankly, I don't recall anymore why I added these 5 lines. This state is
reached on every guest reset and won't change afterwards unless the
kernel actually supports KVM_X86_VCPU_STATE_EVENTS.

This is temporary code, the upstream version will not contain it, but if
you want me to clean this up, just drop me a note.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]