Re: [Xen-devel] [PATCH V3] x86/HVM: Introduce struct hvm_pi_ops

[Suravee Suthikulpanit <Suravee.Suthikulpanit@xxxxxxx>]

Jan,

I would like to updated the following to be more correct.

On 1/12/17 19:37, Jan Beulich wrote:
> > > > On 12.01.17 at 05:47, <suravee.suthikulpanit@xxxxxxx> wrote:
> > --- a/xen/include/asm-x86/hvm/domain.h
> > +++ b/xen/include/asm-x86/hvm/domain.h
> > @@ -72,6 +72,38 @@ struct hvm_ioreq_server {
> >      bool_t                 bufioreq_atomic;
> >  };
> >
> > +/*
> > + * This structure defines function hooks to support hardware-assisted
> > + * virtual interrupt delivery to guest. (e.g. VMX PI and SVM AVIC).
> > + *
> > + * The way we implement this now is by looking at what needs to happen on
> > + * the following runstate transitions:
> > + *
> > + * A: runnable -> running  : switch_to
> > + * B: running  -> runnable : switch_from
> I don't think this is accurate - switch_from may be used when
> transitioning to states other than runnable.
> Both hooks really are context switch rather than run state related afaict.
>
> > + * C: running  -> blocked  : vcpu_block
> > + * D: blocked  -> runnable : do_resume
> This one is, I think, similarly not so much run state related, but
> instead used when a vCPU is about to enter guest context again.
So, what if I generalize and simplify as following:

+/*
+ * This structure defines function hooks to support hardware-assisted
+ * virtual interrupt delivery to guest. (e.g. VMX PI and SVM AVIC).
+ *
+ * These hooks are defined by the underlying arch-specific code
+ * as needed. For example:
+ *   - When the domain is enabled with virtual IPI delivery
+ *   - When the domain is enabled with virtual I/O int delivery
+ *     and actually has a physical device assigned .
+ */
+struct hvm_pi_ops {
+    /* We add hooks into ctx_switch_from. */
+    void (*switch_from) (struct vcpu *v);
+
+    /* We add hooks into ctx_switch_to */
+    void (*switch_to) (struct vcpu *v);
+
+    /*
+     * We add a new arch hook, arch_vcpu_block(),
+     * which is called from vcpu_block() and vcpu_do_poll().
+     */
+    void (*vcpu_block) (struct vcpu *);
+
+    /* We add a hook on the vmentry path. */
+    void (*do_resume) (struct vcpu *v);
+};

Thanks,
Suravee

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel