Re: [PATCH] kvm: x86: correct the misleading comment in vmx_handle_external_intr

From: Radim KrÄmÃÅ
Date: Tue Oct 11 2016 - 14:23:46 EST

2016-10-10 08:23+0800, Longpeng(Mike):
> Since Paolo has removed irq-enable-operation in vmx_handle_external_intr
> (KVM: x86: use guest_exit_irqoff), the original comment about the IF bit
> in rflags is incorrect now.
> Signed-off-by: Longpeng(Mike) <longpeng2@xxxxxxxxxx>
> ---
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> @@ -8647,9 +8647,12 @@ static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
> register void *__sp asm(_ASM_SP);
> /*
> - * If external interrupt exists, IF bit is set in rflags/eflags on the
> - * interrupt stack frame, and interrupt will be enabled on a return
> - * from interrupt handler.

Good catch, thanks.
We want to change it, but I think that the new comment is an overkill.

I am generally not a fan of code comments that describe what the code
does; code speaks for itself and it is better to fix the code, e.g.
split into well named functions, instead of duplicating it.

> + * If external interrupt exists, fakes an interrupt stack and jump to
> + * idt table to let real handler to handle it.

This is the duplication I was talking about. If the corresponding part
of the code is not obvious, it would be better to rework it instead.

> Because most of bits in
> + * rflags are cleared when VM exit(Intel SDM volum 3, chapter 27.5.3),
> + * the IF bit is 0 in rflags on the interrupt stack frame, so interrupt
> + * is still disabled when return from the irq handler, but it will be
> + * enabled later by the caller.

This part is acceptable as it gives a new information code, yet the
function does not modify flags, which makes it unremarkable.
And dependencies on the caller would be better described in a header
(if we cannot express them well in the code).

The most comment-worthy thing about this function is the reason why we
execute the interrupt handler manually, i.e. the dependency on
VM_EXIT_ACK_INTR_ON_EXIT, but that is easy to tell from the commit
message and convenient access to git history is essential in a workflow,
so providing a leeway could be counter-productive.

I would go with no comment for now.