linux-kernel - Re: [PATCHv2 11/11] arm64: use instrumented atomics

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20180704174133.GD9668@arm.com>
Date:   Wed, 4 Jul 2018 18:41:34 +0100
From:   Will Deacon <will.deacon@....com>
To:     Mark Rutland <mark.rutland@....com>
Cc:     linux-kernel@...r.kernel.org, peterz@...radead.org,
        boqun.feng@...il.com, Catalin Marinas <catalin.marinas@....com>
Subject: Re: [PATCHv2 11/11] arm64: use instrumented atomics

On Wed, Jul 04, 2018 at 05:37:23PM +0100, Mark Rutland wrote:
> On Wed, Jul 04, 2018 at 04:24:22PM +0100, Will Deacon wrote:
> > This is mostly straightforward, but the cmpxchg/cmpxchg_double cases grow
> > an 'arch' prefix which suggests that they're part of the API with the core
> > atomics and also makes them horribly ugly.
> 
> I agree this isn't nice.
> 
> > This just looks to be an artifact of __LL_SC_CALL pasting that in. Can
> > you drop that auto pasting of 'arch', and instead change the
> > non-cmpxchg-case callers of the macro to include the arch prefix
> > instead, please?
> 
> That leads to having an arch___llsc_ prefix in some cases, which is
> equally hideous.

Yuck, how does that come about?

> How about I remove the prefix mangling entirely, and always give
> functions an __llsc_ or __lse_ prefix. Then, unify the two in our
> atomic.h with:
> 
> #ifdef CONFIG_ARM64_LSE
> #define ATOMIC_PFX	__lse_
> #else
> #define ATOMIC_PFX	__ll_sc_
> #endif
> 
> #define arch_atomic_foo		ATOMIC_PFX##atomic_foo
> #define arch_atomic_bar		ATOMIC_PFX##atomic_bar
> 
> ... which clearly delineates the implementation from core API.
> 
> Does that sound ok to you?

Why do we need two prefixes? The only reason we throw out __ll_sc_ at the
moment is so the out-of-line atomics have a different name from the inlined
ones. What I'd like is:

	atomic_foo
	-> arch_atomic_foo
	   -> optionally calls __ll_sc_arch_atomic_foo

which I think is very similar to what we already do (i.e. the inlined macro
is always called arch_atomic_foo, regardless of lse or ll/sc).

Will