Received-SPF: pass (google.com: domain of linux-kernel+bounces-44810-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Tue, 30 Jan 2024 15:10:28 +0000
From: Dave Martin <Dave.Martin@arm.com>
To: Mark Brown <broonie@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will@kernel.org>,
	Jackson Cooper-Driver <Jackson.Cooper-Driver@arm.com>,
	linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] arm64/sme: Restore SMCR_EL1.EZT0 on exit from suspend
Message-ID: <ZbkRZNNS5WrfRpjU@e133380.arm.com>
References: <20240130-arm64-sme-resume-v1-0-0e60ebba18df@kernel.org>
 <20240130-arm64-sme-resume-v1-2-0e60ebba18df@kernel.org>
 <ZbjVTigk0YlGd3mA@e133380.arm.com>
 <ca380b64-3420-4817-b3b4-584b8640c0ac@sirena.org.uk>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ca380b64-3420-4817-b3b4-584b8640c0ac@sirena.org.uk>

On Tue, Jan 30, 2024 at 02:34:23PM +0000, Mark Brown wrote:
> On Tue, Jan 30, 2024 at 10:54:06AM +0000, Dave Martin wrote:
> > On Tue, Jan 30, 2024 at 12:02:49AM +0000, Mark Brown wrote:
> 
> > > +	if (system_supports_sme2())
> > > +		smcr |= SMCR_ELx_EZT0;
> 
> > Side question: since ZT0 is likely to be sporadically used, maybe it
> > is worth having separate lazy restore for it versus the main SME state?
> > (Not relevant for this series though, and probably best deferred until
> > there is hardware to benchmark on.  Also, ZT0 is small compared with
> > the SME state proper...)
> 
> One of the advantages SME has here is that we've got a clear indication
> if userspace is actively using the registers through SMSTART and SMSTOP.
> We only restore ZT0 at all whenever PSTATE.ZA is set and the strong

Good point.  I was still thinking in SVE mode there.

> recommendation is that should only be set when either ZA or ZT0 are in
> active use for power and performance reasons.  While it is likely that
> there will be code that uses ZA but doesn't touch ZT0 I would expect
> that the overhead of entering the kernel to do a lazy restore will be
> sufficiently high for it to be an unreasonable penalty on code that does
> touch it, as you say it's not *that* big compared to likely ZA sizes.

Agreed.

Cheers
---Dave