mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Peter Zijlstra <peterz@infradead.org>
To: Eric Dumazet <edumazet@google.com>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Eric Dumazet <eric.dumazet@gmail.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, "H. Peter Anvin" <hpa@zytor.com>,
	Borislav Petkov <bp@alien8.de>
Subject: Re: [PATCH] x86/tsc: use real seqcount_latch in cyc2ns_read_begin()
Date: Thu, 11 Oct 2018 17:24:38 +0200	[thread overview]
Message-ID: <20181011152438.GM9848@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <CANn89iJdztFEMzeCoXu9J4X6roKKx+fuPkwzi9N9G7aiO=_=FQ@mail.gmail.com>

On Thu, Oct 11, 2018 at 08:00:42AM -0700, Eric Dumazet wrote:
> Yes, but the code size is bigger (I have looked at the disassembly)
> 
> All these %gs plus offset add up

> Total length : 0xA7 bytes

effective length: 0x78 bytes

> 00000000000002a0 <native_sched_clock>:
>      2a0: 4c 8d 54 24 08        lea    0x8(%rsp),%r10
>      2a5: 48 83 e4 f0          and    $0xfffffffffffffff0,%rsp
>      2a9: 41 ff 72 f8          pushq  -0x8(%r10)
>      2ad: 55                    push   %rbp
>      2ae: 48 89 e5              mov    %rsp,%rbp
>      2b1: 41 52                push   %r10

       2b3: 66 66 66 66 90	k8_nop5_atomic

>      2b8: 0f 31                rdtsc
>      2ba: 48 c1 e2 20          shl    $0x20,%rdx
>      2be: 48 09 c2              or     %rax,%rdx
>      2c1: 49 89 d2              mov    %rdx,%r10
>      2c4: 49 c7 c1 00 00 00 00 mov    $0x0,%r9
> 		2c7: R_X86_64_32S .data..percpu..shared_aligned
>      2cb: 65 8b 05 00 00 00 00 mov    %gs:0x0(%rip),%eax        # 2d2
> <native_sched_clock+0x32>
> 		2ce: R_X86_64_PC32 .data..percpu..shared_aligned+0x1c
>      2d2: 89 c1                mov    %eax,%ecx
>      2d4: 83 e1 01              and    $0x1,%ecx
>      2d7: 48 c1 e1 04          shl    $0x4,%rcx
>      2db: 4c 01 c9              add    %r9,%rcx
>      2de: 65 48 8b 79 08        mov    %gs:0x8(%rcx),%rdi
>      2e3: 65 8b 31              mov    %gs:(%rcx),%esi
>      2e6: 65 8b 49 04          mov    %gs:0x4(%rcx),%ecx
>      2ea: 65 44 8b 05 00 00 00 mov    %gs:0x0(%rip),%r8d        # 2f2
> <native_sched_clock+0x52>
>      2f1: 00
> 		2ee: R_X86_64_PC32 .data..percpu..shared_aligned+0x1c
>      2f2: 44 39 c0              cmp    %r8d,%eax
>      2f5: 75 d4                jne    2cb <native_sched_clock+0x2b>
>      2f7: 89 f6                mov    %esi,%esi
>      2f9: 48 89 f0              mov    %rsi,%rax
>      2fc: 49 f7 e2              mul    %r10
>      2ff: 48 0f ad d0          shrd   %cl,%rdx,%rax
>      303: 48 d3 ea              shr    %cl,%rdx
>      306: f6 c1 40              test   $0x40,%cl
>      309: 48 0f 45 c2          cmovne %rdx,%rax
>      30d: 48 01 f8              add    %rdi,%rax
>      310: 41 5a                pop    %r10
>      312: 5d                    pop    %rbp
>      313: 49 8d 62 f8          lea    -0x8(%r10),%rsp
>      317: c3                    retq
> 
> New version :
> 
> Total length = 0x91 bytes

effective: 0x71

> 
> 00000000000002a0 <native_sched_clock>:
>      2a0: 4c 8d 54 24 08        lea    0x8(%rsp),%r10
>      2a5: 48 83 e4 f0          and    $0xfffffffffffffff0,%rsp
>      2a9: 41 ff 72 f8          pushq  -0x8(%r10)
>      2ad: 55                    push   %rbp
>      2ae: 48 89 e5              mov    %rsp,%rbp
>      2b1: 41 52                push   %r10

       2b3: 66 66 66 66 90	k8_nop5_atomic

>      2b8: 0f 31                rdtsc
>      2ba: 48 c1 e2 20          shl    $0x20,%rdx
>      2be: 48 09 c2              or     %rax,%rdx
>      2c1: 49 89 d1              mov    %rdx,%r9
>      2c4: 49 c7 c0 00 00 00 00 mov    $0x0,%r8
>		2c7: R_X86_64_32S .data..percpu..shared_aligned
>      2cb: 65 4c 03 05 00 00 00 add    %gs:0x0(%rip),%r8        # 2d3
> <native_sched_clock+0x33>
>      2d2: 00
>		2cf: R_X86_64_PC32 this_cpu_off-0x4
>      2d3: 41 8b 40 20          mov    0x20(%r8),%eax
>      2d7: 89 c6                mov    %eax,%esi
>      2d9: 83 e6 01              and    $0x1,%esi
>      2dc: 48 c1 e6 04          shl    $0x4,%rsi
>      2e0: 4c 01 c6              add    %r8,%rsi
>      2e3: 8b 3e                mov    (%rsi),%edi
>      2e5: 8b 4e 04              mov    0x4(%rsi),%ecx
>      2e8: 48 8b 76 08          mov    0x8(%rsi),%rsi
>      2ec: 41 3b 40 20          cmp    0x20(%r8),%eax
>      2f0: 75 e1                jne    2d3 <native_sched_clock+0x33>
>      2f2: 48 89 f8              mov    %rdi,%rax
>      2f5: 49 f7 e1              mul    %r9
>      2f8: 48 0f ad d0          shrd   %cl,%rdx,%rax
>      2fc: 48 d3 ea              shr    %cl,%rdx
>      2ff: f6 c1 40              test   $0x40,%cl
>      302: 48 0f 45 c2          cmovne %rdx,%rax
>      306: 48 01 f0              add    %rsi,%rax
>      309: 41 5a                pop    %r10
>      30b: 5d                    pop    %rbp
>      30c: 49 8d 62 f8          lea    -0x8(%r10),%rsp
>      310: c3                    retq

Ah, right you are. But my version only touches the one cacheline,
whereas yours will do that extra cpu offset load, which might or might
not be hot.

Difficult..

You have some weird stack setup though.. mine doesn't have that:

$ objdump -dr defconfig-build/arch/x86/kernel/tsc.o | awk '/<native_sched_clock>:$/ {p=1} /^$/ {p=0} {if (p) print $0}'
0000000000000b00 <native_sched_clock>:
b00:   55                      push   %rbp
b01:   48 89 e5                mov    %rsp,%rbp
b04:   66 66 66 66 90          k8_nop5_atomic
b09:   0f 31                   rdtsc
b0b:   48 c1 e2 20             shl    $0x20,%rdx
b0f:   48 09 c2                or     %rax,%rdx
b12:   49 89 d2                mov    %rdx,%r10
b15:   49 c7 c1 00 00 00 00    mov    $0x0,%r9
		b18: R_X86_64_32S       .data..percpu..shared_aligned
b1c:   65 8b 05 00 00 00 00    mov    %gs:0x0(%rip),%eax        # b23 <native_sched_clock+0x23>
		b1f: R_X86_64_PC32      .data..percpu..shared_aligned+0x1c
b23:   89 c1                   mov    %eax,%ecx
b25:   83 e1 01                and    $0x1,%ecx
b28:   48 c1 e1 04             shl    $0x4,%rcx
b2c:   4c 01 c9                add    %r9,%rcx
b2f:   65 48 8b 79 08          mov    %gs:0x8(%rcx),%rdi
b34:   65 8b 31                mov    %gs:(%rcx),%esi
b37:   65 8b 49 04             mov    %gs:0x4(%rcx),%ecx
b3b:   65 44 8b 05 00 00 00    mov    %gs:0x0(%rip),%r8d        # b43 <native_sched_clock+0x43>
b42:   00
		b3f: R_X86_64_PC32      .data..percpu..shared_aligned+0x1c
b43:   44 39 c0                cmp    %r8d,%eax
b46:   75 d4                   jne    b1c <native_sched_clock+0x1c>
b48:   89 f6                   mov    %esi,%esi
b4a:   48 89 f0                mov    %rsi,%rax
b4d:   49 f7 e2                mul    %r10
b50:   48 0f ad d0             shrd   %cl,%rdx,%rax
b54:   48 d3 ea                shr    %cl,%rdx
b57:   f6 c1 40                test   $0x40,%cl
b5a:   48 0f 45 c2             cmovne %rdx,%rax
b5e:   48 01 f8                add    %rdi,%rax
b61:   5d                      pop    %rbp
b62:   c3                      retq

Which gets me to 0x62 effective bytes.




  parent reply	other threads:[~2018-10-11 15:24 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-10-11  0:33 Eric Dumazet
2018-10-11  7:31 ` Peter Zijlstra
2018-10-11  8:40   ` Peter Zijlstra
     [not found]     ` <CANn89iJdztFEMzeCoXu9J4X6roKKx+fuPkwzi9N9G7aiO=_=FQ@mail.gmail.com>
2018-10-11 15:24       ` Peter Zijlstra [this message]
2018-10-11 15:17   ` Eric Dumazet
2018-10-11 23:25 ` kbuild test robot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20181011152438.GM9848@hirez.programming.kicks-ass.net \
    --to=peterz@infradead.org \
    --cc=bp@alien8.de \
    --cc=edumazet@google.com \
    --cc=eric.dumazet@gmail.com \
    --cc=hpa@zytor.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=tglx@linutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®