Re: [PATCH] arm64/sve: Lower the maximum allocation for the SVE ptrace regset

From: Dave Martin <Dave.Martin@arm.com>
To: Mark Brown <broonie@kernel.org>
Cc: Will Deacon <will@kernel.org>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Oleg Nesterov <oleg@redhat.com>,
	Al Viro <viro@zeniv.linux.org.uk>,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org,
	Doug Anderson <dianders@chromium.org>
Subject: Re: [PATCH] arm64/sve: Lower the maximum allocation for the SVE ptrace regset
Date: Wed, 7 Feb 2024 13:51:42 +0000	[thread overview]
Message-ID: <ZcOK7oIe2f/BFlDj@e133380.arm.com> (raw)
In-Reply-To: <ZcOBH7ip/KMhLYGO@finisterre.sirena.org.uk>

On Wed, Feb 07, 2024 at 01:09:51PM +0000, Mark Brown wrote:
> On Wed, Feb 07, 2024 at 12:23:56PM +0000, Dave Martin wrote:
> > On Mon, Feb 05, 2024 at 05:41:47PM +0000, Mark Brown wrote:
> > > On Mon, Feb 05, 2024 at 05:11:59PM +0000, Dave Martin wrote:
> 
> > > > If the kernel is now juggling two #defines for the maximum vector size,
> > > > this feels like it may seed bitrot...
> 
> > > Ideally we'd just not have the existing define externally but it's there
> > > and it's been used.
> 
> > To clarify, is this intended as a temporary band-aid against silly
> > behaviour while a cleaner solution is found, or a permanent limitation?
> 
> Ideally we'd just make everything dynamic, other than the regset issue
> and the bitmasks used for VL enumeration we're there already.  Making
> the bitmasks dynamically sized is more painful but are also doing
> enumeration that userspace doesn't need to do.

For the bitmasks, we'd be saving (512 - 16) / 8 = 62 bytes for each of
SVE and SME (I think).

The tradeoff really didn't seem worth it...

> 
> > We'd need to change various things if the architectural max VL actually
> > grew, so no forward-portability is lost immediately if the kernel
> > adopts 16 internally, but I'm still a little concerned that people may
> > poke about in the kernel code as a reference and this will muddy the
> > waters regarding how to do the right thing in userspace (I know people
> > shouldn't, but...)
> 
> I think if we fix the ptrace regset issue we're doing a good enough job
> of just using fully dynamic sizing with no limits other than what's been
> enumerated there.  We could possibly deal with the enumberation code by
> changing it to use ZCR/SMCR_ELx_LEN_ based defines so that it's
> obviously coming from what we can possibly write to the register but
> it's a bit less clear how to do that neatly.

OK, but we still seem to have two competing approaches: clamp SVE_VQ_MAX
for kernel internal purposes, or restore the dynamic sizing of
NT_ARM_SVE based on the new regset core behaviour.

Are you saying we should or both, or otherwise which one?

Cheers
---Dave