Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
In-Reply-To: <998a8c2c39546115f93e4c6e2ad7f918626f6145.1509006199.git.luto@kernel.org>
References: <cover.1509006199.git.luto@kernel.org> <998a8c2c39546115f93e4c6e2ad7f918626f6145.1509006199.git.luto@kernel.org>
From:   Brian Gerst <brgerst@gmail.com>
Date:   Wed, 1 Nov 2017 13:25:22 -0400
Message-ID: <CAMzpN2hw6q5G7A-2vb7mHqECf4-Vd8D7XEVGXg7U-Mi06qY_4w@mail.gmail.com>
Subject: Re: [PATCH 07/18] x86/asm/64: Merge the fast and slow SYSRET paths
To:     Andy Lutomirski <luto@kernel.org>
Cc:     X86 ML <x86@kernel.org>, Borislav Petkov <bpetkov@suse.de>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Dave Hansen <dave.hansen@intel.com>,
        Linus Torvalds <torvalds@linux-foundation.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Oct 26, 2017 at 4:26 AM, Andy Lutomirski <luto@kernel.org> wrote:
> They did almost the same thing.  Remove a bunch of pointless
> instructions (mostly hidden in macros) and reduce cognitive load by
> merging them.
>
> Signed-off-by: Andy Lutomirski <luto@kernel.org>
> ---
>  arch/x86/entry/entry_64.S | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
> index 9dafafa3e0ec..c855ee91a3a5 100644
> --- a/arch/x86/entry/entry_64.S
> +++ b/arch/x86/entry/entry_64.S
> @@ -220,10 +220,9 @@ entry_SYSCALL_64_fastpath:
>         TRACE_IRQS_ON           /* user mode is traced as IRQs on */
>         movq    RIP(%rsp), %rcx
>         movq    EFLAGS(%rsp), %r11
> -       RESTORE_C_REGS_EXCEPT_RCX_R11
> -       movq    RSP(%rsp), %rsp
> +       addq    $6*8, %rsp      /* skip extra regs -- they were preserved */
>         UNWIND_HINT_EMPTY
> -       USERGS_SYSRET64
> +       jmp     .Lpop_c_regs_except_rcx_r11_and_sysret
>
>  1:
>         /*
> @@ -315,6 +314,7 @@ syscall_return_via_sysret:
>         /* rcx and r11 are already restored (see code above) */
>         UNWIND_HINT_EMPTY
>         POP_EXTRA_REGS
> +.Lpop_c_regs_except_rcx_r11_and_sysret:
>         popq    %rsi    /* skip r11 */
>         popq    %r10
>         popq    %r9

Wouldn't it be more logical to keep the SYSRET path at the end of the
fast path (reverse of what you are doing here)?  That way the fast
path falls through without jumping.

--
Brian Gerst

From 1582867162831366043@xxx Wed Nov 01 12:35:59 +0000 2017
X-GM-THRID: 1582308295795346120
X-Gmail-Labels: Inbox,Category Forums