Re: [PATCH v3 2/2] KVM: arm64: Don't acquire RCU read lock for exclusive table walks

From: Will Deacon <will@kernel.org>
To: Oliver Upton <oliver.upton@linux.dev>
Cc: kvm@vger.kernel.org, Marc Zyngier <maz@kernel.org>,
	linux-kernel@vger.kernel.org,
	Catalin Marinas <catalin.marinas@arm.com>,
	kvmarm@lists.linux.dev, kvmarm@lists.cs.columbia.edu,
	linux-arm-kernel@lists.infradead.org,
	Marek Szyprowski <m.szyprowski@samsung.com>
Subject: Re: [PATCH v3 2/2] KVM: arm64: Don't acquire RCU read lock for exclusive table walks
Date: Fri, 18 Nov 2022 12:19:50 +0000	[thread overview]
Message-ID: <20221118121949.GA3697@willie-the-truck> (raw)
In-Reply-To: <Y3Z8G3aCuRzzoq5e@google.com>

On Thu, Nov 17, 2022 at 06:23:23PM +0000, Oliver Upton wrote:
> On Thu, Nov 17, 2022 at 05:49:52PM +0000, Will Deacon wrote:
> > On Wed, Nov 16, 2022 at 04:56:55PM +0000, Oliver Upton wrote:
> 
> [...]
> 
> > > -static inline void kvm_pgtable_walk_begin(void) {}
> > > -static inline void kvm_pgtable_walk_end(void) {}
> > > +static inline void kvm_pgtable_walk_begin(struct kvm_pgtable_walker *walker)
> > > +{
> > > +	/*
> > > +	 * Due to the lack of RCU (or a similar protection scheme), only
> > > +	 * non-shared table walkers are allowed in the hypervisor.
> > > +	 */
> > > +	WARN_ON(walker->flags & KVM_PGTABLE_WALK_SHARED);
> > > +}
> > 
> > I think it would be better to propagate the error to the caller rather
> > than WARN here.
> 
> I'd really like to warn somewhere though since we're rather fscked at
> this point. Keeping that WARN close to the exceptional condition would
> help w/ debugging.
> 
> Were you envisioning bubbling the error all the way back up (i.e. early
> return from kvm_pgtable_walk())?

Yes, that's what I had in mind. WARN is fatal at EL2, so I think it's
better to fail the pgtable operation rather than bring down the entire
machine by default. Now, it _might_ be fatal anyway (e.g. if we were
handling a host stage-2 fault w/ pKVM), but the caller is in a better
position to decide the severity.

> I had really only intended these to indirect lock acquisition/release,
> so the error handling on the caller side feels weird:
> 
>   static inline int kvm_pgtable_walk_begin(struct kvm_pgtable_walker *walker)
>   {
>   	if (WARN_ON(walker->flags & KVM_PGTABLE_WALK_SHARED))
> 		return -EPERM;
> 
> 	return 0;
>   }
> 
>   r = kvm_pgtable_walk_begin()
>   if (r)
>   	return r;
> 
>   r = _kvm_pgtable_walk();
>   kvm_pgtable_walk_end();

This doesn't look particularly weird to me (modulo dropping the WARN, or
moving it to _end()), but maybe I've lost my sense of taste.

> > Since you're rejigging things anyway, can you have this
> > function return int?
> 
> If having this is a strong motivator I can do a v4.

It's a really minor point, so I'll leave it up to you guys.

WIll
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm