Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.34 as permitted sender) client-ip=23.128.96.34;
MIME-Version: 1.0
References: <20230721161018.50214-1-brgerst@gmail.com> <ZR5yTecBhCFsVOtQ@gmail.com>
In-Reply-To: <ZR5yTecBhCFsVOtQ@gmail.com>
From:   Brian Gerst <brgerst@gmail.com>
Date:   Thu, 5 Oct 2023 11:13:16 -0400
Message-ID: <CAMzpN2j7qddPEUdD+ZX3dtyQkPq6e4gzwcu5szkZ2esh_8zm9g@mail.gmail.com>
Subject: Re: [PATCH v2 0/6] x86: Clean up fast syscall return validation
To:     Ingo Molnar <mingo@kernel.org>
Cc:     linux-kernel@vger.kernel.org, x86@kernel.org,
        Thomas Gleixner <tglx@linutronix.de>,
        Borislav Petkov <bp@alien8.de>,
        "H . Peter Anvin" <hpa@zytor.com>,
        Andy Lutomirski <luto@kernel.org>,
        =?UTF-8?Q?Mika_Penttil=C3=A4?= <mpenttil@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Uros Bizjak <ubizjak@gmail.com>,
        Denys Vlasenko <dvlasenk@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Josh Poimboeuf <jpoimboe@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, Oct 5, 2023 at 4:22=E2=80=AFAM Ingo Molnar <mingo@kernel.org> wrote=
:
>
>
> * Brian Gerst <brgerst@gmail.com> wrote:
>
> > This patch set cleans up the tests done to determine if a fast syscall
> > return instruction can be used to return to userspace.  It converts the
> > code to C, and refactors existing code to be more readable.
> >
> > v2:
> >  - Fix shift value for canonical RIP test and use
> >    __is_canonical_address()
> >
> > Brian Gerst (6):
> >   x86/entry/64: Remove obsolete comment on tracing vs. SYSRET
> >   x86/entry/64: Convert SYSRET validation tests to C
> >   x86/entry/compat: Combine return value test from syscall handler
> >   x86/entry/32: Convert do_fast_syscall_32() to bool return type
> >   x86/entry/32: Remove SEP test for SYSEXIT
> >   x86/entry/32: Clean up syscall fast exit tests
> >
> >  arch/x86/entry/common.c          | 99 +++++++++++++++++++++-----------
> >  arch/x86/entry/entry_32.S        |  2 +-
> >  arch/x86/entry/entry_64.S        | 68 +---------------------
> >  arch/x86/entry/entry_64_compat.S | 12 ++--
> >  arch/x86/include/asm/syscall.h   |  6 +-
> >  5 files changed, 77 insertions(+), 110 deletions(-)
>
> Ok, so I've applied patches #1, #3, #4 and #5 to tip:x86/entry,
> (ie. skipped #2 & #6 for now), as they look correct and are good
> improvements. None of these four patches depend on the skipped
> patches in some way I missed, correct?
>
> As for #2, I looked at the before/after disassembly, and the new
> C code in do_syscall_64() looked suboptimal on x86-64 defconfig,
> if I was reading it right.
>
> Mind re-evaluating that, and if you still think the C conversion
> is a good idea, mind putting a before/after analysis of the
> generated instructions into the changelog? This is our primary
> system call return path after all.

Looking at the compiled output, the only suboptimal code appears to be
the canonical address test, where the C code uses the CL register for
the shifts instead of immediates.

 180:   e9 00 00 00 00          jmp    185 <do_syscall_64+0x85>
                        181: R_X86_64_PC32      .altinstr_aux-0x4
 185:   b9 07 00 00 00          mov    $0x7,%ecx
 18a:   eb 05                   jmp    191 <do_syscall_64+0x91>
 18c:   b9 10 00 00 00          mov    $0x10,%ecx
 191:   48 89 c2                mov    %rax,%rdx
 194:   48 d3 e2                shl    %cl,%rdx
 197:   48 d3 fa                sar    %cl,%rdx
 19a:   48 39 d0                cmp    %rdx,%rax
 19d:   75 39                   jne    1d8 <do_syscall_64+0xd8>

Was there anything else specifically that you can point out?

Brian Gerst