Re: [PATCH 1/5] kvm: add exit_to_guest_mode() and enter_from_guest_mode()

[Mark Rutland]

On Thu, Jan 13, 2022 at 08:32:20PM +0000, Sean Christopherson wrote:
> On Tue, Jan 11, 2022, Mark Rutland wrote:
> > Atop this, new exit_to_guest_mode() and enter_from_guest_mode() helpers
> > are added to handle the ordering of lockdep, tracing, and RCU manageent.
> > These are named to align with exit_to_user_mode() and
> > enter_from_user_mode().
> > 
> > Subsequent patches will migrate architectures over to the new helpers,
> > following a sequence:
> > 
> > 	guest_timing_enter_irqoff();
> > 
> > 	exit_to_guest_mode();
> 
> I'm not a fan of this nomenclature.  First and foremost, virtualization refers to
> transfers to guest mode as VM-Enter, and transfers from guest mode as VM-Exit.
> It's really, really confusing to read this code from a virtualization perspective.
> The functions themselves are contradictory as the "enter" helper calls functions
> with "exit" in their name, and vice versa.

Sure; FWIW I wasn't happy with the naming either, but I couldn't find anything
that was entirely clear, because it depends on whether you consider this an
entry..exit of guest context or an exit..entry of regular kernel context. I
went with exit_to_guest_mode() and enter_from_guest_mode() because that clearly
corresponded to exit_to_user_mode() and enter_from_user_mode(), and the
convention in the common entry code is to talk in terms of the regular kernel
context.

While I was working on this, I had guest_context_enter_irqoff() for
kernel->guest and guest_context_exit_irqoff() for guest->kernel, which also
matched the style of guest_timing_{enter,exit}_irqoff().

I'm happy to change to that, if that works for you?

> We settled on xfer_to_guest_mode_work() for a similar conundrum in the past, though
> I don't love using xfer_to/from_guest_mode() as that makes it sound like those
> helpers handle the actual transition into guest mode, i.e. runs the vCPU.
> 
> To avoid too much bikeshedding, what about reusing the names we all compromised
> on when we did this for x86 and call them kvm_guest_enter/exit_irqoff()?  If x86
> is converted in the first patch then we could even avoid temporary #ifdefs.

I'd like to keep this somewhat orthogonal to the x86 changes (e.g. as other
architectures will need backports to stable at least for the RCU bug fix), so
I'd rather use a name that isn't immediately coupled with x86 changes.

Does the guest_context_{enter,exit}_irqoff() naming above work for you?

Thanks,
Mark.