Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

From: Segher Boessenkool <segher@kernel.crashing.org>
To: Nick Desaulniers <ndesaulniers@google.com>
Cc: apinski@marvell.com, Josh Poimboeuf <jpoimboe@redhat.com>,
	Borislav Petkov <bp@alien8.de>, Vasily Gorbik <gor@linux.ibm.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Ingo Molnar <mingo@kernel.org>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Peter Zijlstra <peterz@infradead.org>,
	Luc Van Oostenryck <luc.vanoostenryck@gmail.com>,
	x86@kernel.org, llvm@lists.linux.dev,
	linux-sparse@vger.kernel.org, linux-kernel@vger.kernel.org,
	kernel test robot <lkp@intel.com>,
	Nathan Chancellor <nathan@kernel.org>,
	linux-toolchains@vger.kernel.org
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__
Date: Mon, 31 Jan 2022 16:13:33 -0600	[thread overview]
Message-ID: <20220131221333.GR614@gate.crashing.org> (raw)
In-Reply-To: <CAKwvOdnB=OfCc_31b=PMjxKewvb3CV2WDwhMWrigviWaHrGMhw@mail.gmail.com>

On Mon, Jan 31, 2022 at 12:45:20PM -0800, Nick Desaulniers wrote:
> On Tue, Jan 25, 2022 at 3:34 PM Segher Boessenkool
> <segher@kernel.crashing.org> wrote:
> >
> > Hi!
> >
> > On Mon, Jan 24, 2022 at 03:26:36PM -0800, Nick Desaulniers wrote:
> > > I'm more confident that we can remove the `volatile` keyword (I was
> > > thinking about adding a new diagnostic to clang to warn that volatile
> > > is redundate+implied for asm goto or inline asm that doesn't have
> > > outputs) though that's not the problem here and will probably generate
> > > some kernel wide cleanup before we could enable such a flag.
> >
> > Its main value is that it would discourage users from thinking volatile
> > is magic.  Seriously worth some pain!
> 
> https://reviews.llvm.org/D118297
> PTAL

""
  Really the volatile asm-qualifier exists only to signal that an asm
  statement should not be DCE'd (when it has outputs but they are unused),
  CSE'd, or LICM'd. It is not a general compiler barrier.

It means that the asm has a side effect (one unknown to the compiler),
so it must be executed in the real machine just where it would be in the
abstract machine.  It *can* be CSEd, it *can* be DCEd, it can even be
optimised by LICM in certain cases: but it has to be executed as often
(and in the same order etc.) in the resulting machine code as it would
be if you single-stepped through the source code by hand.

Those are fine examples if you add "in most cases" (and that they are
just examples, it's not an exhaustive list).

Thanks,

Segher