Re: [RFC PATCH v2 2/4] KVM: arm64: Add PSCI SYSTEM_OFF2 function for hibernation

From: Marc Zyngier <maz@kernel.org>
To: David Woodhouse <dwmw2@infradead.org>
Cc: linux-arm-kernel@lists.infradead.org, kvm@vger.kernel.org,
	Paolo Bonzini <pbonzini@redhat.com>,
	Jonathan Corbet <corbet@lwn.net>,
	Oliver Upton <oliver.upton@linux.dev>,
	James Morse <james.morse@arm.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Zenghui Yu <yuzenghui@huawei.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Lorenzo Pieralisi <lpieralisi@kernel.org>,
	"Rafael J. Wysocki" <rafael@kernel.org>,
	Len Brown <len.brown@intel.com>, Pavel Machek <pavel@ucw.cz>,
	Mostafa Saleh <smostafa@google.com>,
	Jean-Philippe Brucker <jean-philippe@linaro.org>,
	linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org,
	kvmarm@lists.linux.dev, linux-pm@vger.kernel.org
Subject: Re: [RFC PATCH v2 2/4] KVM: arm64: Add PSCI SYSTEM_OFF2 function for hibernation
Date: Mon, 18 Mar 2024 18:07:11 +0000	[thread overview]
Message-ID: <86r0g7za5s.wl-maz@kernel.org> (raw)
In-Reply-To: <6a2107864d45bd6ac403c218d68bf97025eca971.camel@infradead.org>

On Mon, 18 Mar 2024 17:54:06 +0000,
David Woodhouse <dwmw2@infradead.org> wrote:
> 
> [1  <text/plain; UTF-8 (quoted-printable)>]
> On Mon, 2024-03-18 at 17:29 +0000, Marc Zyngier wrote:
> > 
> > Again, I really oppose this way of doing things. We already have an
> > infrastructure for selecting PSCI levels. You may not like it, but it
> > exists, and I'm not going entertain supporting yet another bike-shed
> > model. Adding an orthogonal cap for a feature that is specific to a
> > new PSCI version is just awful.
> 
> Huh? This isn't a "new bike-shed model". This is a straight copy of
> what we *already* have for SYSTEM_RESET2.

There is no KVM capability for SYSTEM_RESET2. It is directly
advertised to the guest when PSCI 1.1 is supported.

> If I were bike-shedding, I wouldn't do separate caps for them; I'd have
> done it as a *bitmask* of the optional PSCI calls that should be
> enabled.
>
> The *mandatory* ones should obviously come from the PSCI version alone,
> but I can't see how that makes sense for the optional ones...

The guest is in a position to probe for what is supported or not with
the PSCI_FEATURES call.  Why would you add anything else?

> 
> > Please make PSCI 1.3 the only version of PSCI supporting suspend in a
> > non-optional way, and be done with it.
> 
> SYSTEM_OFF2 is an *optional* feature in PSCI v1.3. As are
> CLEAR_INV_MEMREGION and CLEAR_INV_MEMREGION_ATTRIBUTES.
> 
> Are you suggesting that enabling v1.3 should automatically enable *all*
> of the optional features that were defined in that version (and
> previous versions) of the spec?

No. We have everything we need to incrementally *add* features. So you
can perfectly implement PSCI 1.3 with only SYSTEM_OFF2, and only later
on add the rest, if ever.

	M.

-- 
Without deviation from the norm, progress is not possible.