[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAFULd4YxvMtTEfQL-RiLisTxDwoJZZxXXB+3CWqCpzZkUf85JA@mail.gmail.com>
Date: Thu, 19 Oct 2023 20:16:24 +0200
From: Uros Bizjak <ubizjak@...il.com>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: peterz@...radead.org, Nadav Amit <namit@...are.com>,
"the arch/x86 maintainers" <x86@...nel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Andy Lutomirski <luto@...nel.org>,
Brian Gerst <brgerst@...il.com>,
Denys Vlasenko <dvlasenk@...hat.com>,
"H . Peter Anvin" <hpa@...or.com>,
Thomas Gleixner <tglx@...utronix.de>,
Josh Poimboeuf <jpoimboe@...hat.com>,
Nick Desaulniers <ndesaulniers@...gle.com>
Subject: Re: [PATCH v2 -tip] x86/percpu: Use C for arch_raw_cpu_ptr()
On Thu, Oct 19, 2023 at 8:06 PM Linus Torvalds
<torvalds@...ux-foundation.org> wrote:
>
> On Thu, 19 Oct 2023 at 10:21, Uros Bizjak <ubizjak@...il.com> wrote:
> >
> > > A compiler that were to rematerializes an inline asm (instead of
> > > spilling) would be a bad joke. That's not an optimization, that's just
> > > a crazy bad compiler with a code generation bug.
> >
> > But that is what the compiler does without volatile.
>
> Do you actually have a real case of that, or are basing it purely off
> insane documentation?
>
> Because remat of inline asm really _is_ insane.
The following testcase pushes the compiler to the limit:
--cut here--
extern void ex (int);
static int read (void)
{
int ret;
asm ("# -> %0" : "=r"(ret));
return ret;
}
int foo (void)
{
int ret = read ();
ex (ret);
asm volatile ("clobber" : : : "ax", "cx", "dx", "bx", "bp", "si", "di");
return ret;
}
extern int m;
int bar (void)
{
int ret = m;
ex (ret);
asm volatile ("clobber" : : : "ax", "cx", "dx", "bx", "bp", "si", "di");
return ret;
}
--cut here--
Please compile the above with -S -O2 -m32 (so we don't have to list
all 16 x86_64 registers).
And NO (whee...), there is no rematerialization of asm (foo() ). OTOH,
there is also no rematerialization from non-volatile memory (bar() ),
although it would be more optimal than spill to/fill from a frame pair
of moves. I wonder what are "certain circumstances" that the
documentation is referring to.
Uros.
Uros.
Powered by blists - more mailing lists