Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20211229021258.176670-1-morbo@google.com> <20220204005742.1222997-1-morbo@google.com>
 <CAKwvOdk=VdDo1fhWJVa4eO0UjuQwtV9kC-cJd0J9-6guU2vafA@mail.gmail.com> <0d93ea4d847b42ca9c5603cb97cbda8a@AcuMS.aculab.com>
In-Reply-To: <0d93ea4d847b42ca9c5603cb97cbda8a@AcuMS.aculab.com>
From:   Bill Wendling <morbo@google.com>
Date:   Tue, 8 Feb 2022 15:18:50 -0800
Message-ID: <CAGG=3QW09mBaLNgu4+3oBw-NfYx+0xn86Cd5APYxGdSNgXzg0Q@mail.gmail.com>
Subject: Re: [PATCH v3] x86: use builtins to read eflags
To:     David Laight <David.Laight@aculab.com>
Cc:     Nick Desaulniers <ndesaulniers@google.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        "x86@kernel.org" <x86@kernel.org>,
        "H . Peter Anvin" <hpa@zytor.com>,
        Nathan Chancellor <nathan@kernel.org>,
        Juergen Gross <jgross@suse.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Andy Lutomirski <luto@kernel.org>,
        "llvm@lists.linux.dev" <llvm@lists.linux.dev>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Feb 8, 2022 at 1:14 AM David Laight <David.Laight@aculab.com> wrote:
>
> From: Nick Desaulniers
> > Sent: 07 February 2022 22:12
> >
> > On Thu, Feb 3, 2022 at 4:57 PM Bill Wendling <morbo@google.com> wrote:
> > >
> > > GCC and Clang both have builtins to read and write the EFLAGS register.
> > > This allows the compiler to determine the best way to generate this
> > > code, which can improve code generation.
> > >
> > > This issue arose due to Clang's issue with the "=rm" constraint.  Clang
> > > chooses to be conservative in these situations, and so uses memory
> > > instead of registers. This is a known issue, which is currently being
> > > addressed.
>
> How much performance would be lost by just using "=r"?
>
> You get two instructions if the actual target is memory.
> This might be a marginal code size increase - but not much,
> It might also slow things down if the execution is limited
> by the instruction decoder.
>
> But on Intel cpu 'pop memory' is 2 uops, exactly the same
> as 'pop register' 'store register' (and I think amd is similar).
> So the actual execution time is exactly the same for both.
>
> Also it looks like clang's builtin is effectively "=r".
> Compiling:
> long fl;
> void native_save_fl(void) {
>        fl = __builtin_ia32_readeflags_u64();
> }
> Not only generates a stack frame, it also generates:
> pushf; pop %rax; mov mem, %rax.
>
It used to be "=r" (see f1f029c7bfbf4e), but was changed back to "=rm"
in ab94fcf528d127. This pinging back and forth is another reason to
use the builtins and be done with it.

-bw