Re: [PATCH v3 05/15] arm64: kvm: Build hyp-entry.S separately for VHE/nVHE

From: Marc Zyngier <maz@kernel.org>
To: David Brazdil <dbrazdil@google.com>
Cc: kernel-team@android.com, android-kvm@google.com,
	Catalin Marinas <catalin.marinas@arm.com>,
	linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org,
	Will Deacon <will@kernel.org>,
	kvmarm@lists.cs.columbia.edu
Subject: Re: [PATCH v3 05/15] arm64: kvm: Build hyp-entry.S separately for VHE/nVHE
Date: Thu, 25 Jun 2020 09:12:50 +0100	[thread overview]
Message-ID: <491f3c8877897a4ac69d69fb7354c1cb@kernel.org> (raw)
In-Reply-To: <20200622102041.myve2otyoj5q7j5s@google.com>

Hi David,

On 2020-06-22 11:20, David Brazdil wrote:
> Hi Marc,
> 
>> > -	void *dst = lm_alias(__bp_harden_hyp_vecs + slot * SZ_2K);
>> > +	char *vec = has_vhe() ? __bp_harden_hyp_vecs
>> > +			      : kvm_nvhe_sym(__bp_harden_hyp_vecs);
>> 
>> If we get this construct often, then something that abstracts
>> the uggliness of the symbol duality would be nice...
> 
> Agreed, I do hope that this will end up being limited to finding the 
> address of
> the hyp-init vector once EL2 becomes self-contained. Even this vector 
> selection
> can be done in EL2 where the symbol duality does not exist.
> If we were to hide it, there is a trade off between code "elegance" and 
> clarity
> of what's happening under the hood. I was thinking we could extract 
> this
> `has_vhe() ? foo : __kvm_nvhe_foo` as a `#define foo` but I do worry 
> that
> anybody debugging this code would be cursing my name.

I would say that whoever is debugging this code better have an 
understanding
of how things are supposed to work. Given that this is only a handful of
people so far, I think your name is safe! ;-)

> It would also not work
> with other macros that take symbol names, notably kvm_ksym_ref. But 
> that can be
> rewritten to accept a pointer instead. The more verbose but less magic 
> approach
> is to have a bunch of different helpers for various situations, eg.
> __pa_symbol_nvhe. What would be your preference?

I'd be happy with the (maybe temporary) magic approach. It helps 
reasoning
about things, and makes the transition smoother. Yes, bugs could crop up
there, but given the static nature of obtaining a symbol's address, I'm
fairly confident we'll get it right. The same cannot be said about 
pointers
though.

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm