Received-SPF: pass (google.com: domain of linux-kernel+bounces-52885-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Mon, 5 Feb 2024 15:19:47 +0000
From: Dave Martin <Dave.Martin@arm.com>
To: Ard Biesheuvel <ardb@kernel.org>
Cc: Fangrui Song <maskray@google.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will@kernel.org>, linux-arm-kernel@lists.infradead.org,
	Jisheng Zhang <jszhang@kernel.org>,
	Peter Smith <peter.smith@arm.com>, llvm@lists.linux.dev,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] arm64: jump_label: use constraints "Si" instead of "i"
Message-ID: <ZcD8k2UTs3wyk9Dx@e133380.arm.com>
References: <20240201223545.728028-1-maskray@google.com>
 <Zb0Qu5lR0iZUqImb@e133380.arm.com>
 <CAMj1kXGSQ4Xq68Cmq6BH37FPJ+R+N5cOfrdC+aVML4sGaNDopg@mail.gmail.com>
 <Zb0eRogn3rjkeDAs@e133380.arm.com>
 <20240202225104.f4dsagfwf6gcnddy@google.com>
 <CAMj1kXGN-nKSzxJoyM9peBTDevuPkH-+P2UzH746P-F913Dg-g@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAMj1kXGN-nKSzxJoyM9peBTDevuPkH-+P2UzH746P-F913Dg-g@mail.gmail.com>

On Sat, Feb 03, 2024 at 10:50:44AM +0100, Ard Biesheuvel wrote:
> On Fri, 2 Feb 2024 at 23:51, Fangrui Song <maskray@google.com> wrote:

[...]

> > "Si" is fine for GCC and Clang.
> > "i" is fine for Clang but not for GCC PIC.
> >
> >      https://maskray.me/blog/2024-01-30-raw-symbol-names-in-inline-assembly#aarch64
> >
> >      In gcc/config/aarch64, LEGITIMATE_PIC_OPERAND_P(X) disallows any symbol
> >      reference, which means that "i" and "s" cannot be used for PIC. Instead,
> >      the constraint "S" has been supported since the initial port (2012) to
> >      reference a symbol or label.
> >
> > I am also not familiar with
> > https://gcc.gnu.org/onlinedocs/gcc/Multi-Alternative.html (comma in a
> > constraint string). Thankfully we don't need this powerful construct:)

Ack, I had thought that this was relevant, but it is not,
and "Si" seems right.

[...]

> > I am convinced by Ard' argument that two inputs (key, branch) deserve
> > two operands.
> > The existing "i"(&((char *)key)[branch]) is kinda ugly and also longer:)
> 
> If it helps clarify things, we might do something like
> 
> ".quad  (%[key]  - .) + %[bit0]"
> 
> : : [key]"Si"(key), [bit0]"i"(branch) :  : l_yes);

I don't have a strong opinion on the naming, but something like this
seems fine.

Cheers
---Dave