lkml.org 
[lkml]   [2014]   [Jul]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH 09/16] rcu: Remove redundant check for online cpu
    On Wed, Jul 23, 2014 at 04:16:11PM -0400, Pranith Kumar wrote:
    > On Wed, Jul 23, 2014 at 3:15 PM, Paul E. McKenney
    > <paulmck@linux.vnet.ibm.com> wrote:
    > >> > If you change the "awake" to something like "am_online", I could get
    > >> > behind this one.
    > >>
    > >> OK! I will submit that in the next series(with the zalloc check).
    > >
    > > You caught me at a weak moment... This change just adds an extra
    > > line of code and doesn't really help anything.
    > >
    > > So please leave this one out.
    > >
    >
    > <resending as the assembly was garbled>
    >
    > It adds an extra line of code and generates better assembly code. Last
    > try to convince you before I give up :-)

    If you got this kind of savings in __rcu_read_lock() or
    __rcu_read_unlock(), I might be interested. Hard to get excited about
    __call_rcu_core(), especially given that a smarter compiler might be
    able to make this transformation on its own.

    Thanx, Paul

    > Size:
    > text data bss dec hex filename
    > before 30664 7844 32 38540 968c kernel/rcu/tree.o
    > after 30648 7844 32 38524 967c kernel/rcu/tree.o
    >
    > Assembly:
    >
    > Before:
    >
    > if (!rcu_is_watching() && cpu_online(smp_processor_id()))
    > 26d3: 83 e2 01 and $0x1,%edx
    > 26d6: 75 1f jne 26f7 <__call_rcu+0x1c7>
    > 26d8: 65 8b 14 25 00 00 00 mov %gs:0x0,%edx
    > 26df: 00
    > 26dc: R_X86_64_32S cpu_number
    > 26e0: 48 8b 0d 00 00 00 00 mov 0x0(%rip),%rcx # 26e7 <__call_rcu+0x1b7>
    > 26e3: R_X86_64_PC32 cpu_online_mask-0x4
    > 26e7: 89 d2 mov %edx,%edx
    > 26e9: 48 0f a3 11 bt %rdx,(%rcx)
    > 26ed: 19 d2 sbb %edx,%edx
    > 26ef: 85 d2 test %edx,%edx
    > 26f1: 0f 85 29 02 00 00 jne 2920 <__call_rcu+0x3f0>
    > invoke_rcu_core();
    >
    > /* If interrupts were disabled or CPU offline, don't invoke RCU core. */
    > if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
    > 26f7: 48 f7 45 d0 00 02 00 testq $0x200,-0x30(%rbp)
    > 26fe: 00
    > 26ff: 0f 84 e6 fe ff ff je 25eb <__call_rcu+0xbb>
    > 2705: 65 8b 14 25 00 00 00 mov %gs:0x0,%edx
    > 270c: 00
    > 2709: R_X86_64_32S cpu_number
    > 270d: 48 8b 0d 00 00 00 00 mov 0x0(%rip),%rcx # 2714 <__call_rcu+0x1e4>
    > 2710: R_X86_64_PC32 cpu_online_mask-0x4
    > 2714: 89 d2 mov %edx,%edx
    > 2716: 48 0f a3 11 bt %rdx,(%rcx)
    > 271a: 19 d2 sbb %edx,%edx
    > 271c: 85 d2 test %edx,%edx
    > 271e: 0f 84 c7 fe ff ff je 25eb <__call_rcu+0xbb>
    >
    > After:
    >
    > bool cpu_up = cpu_online(smp_processor_id());
    > 26c1: 65 8b 14 25 00 00 00 mov %gs:0x0,%edx
    > 26c8: 00
    > 26c5: R_X86_64_32S cpu_number
    > 26c9: 48 8b 0d 00 00 00 00 mov 0x0(%rip),%rcx # 26d0 <__call_rcu+0x1a0>
    > 26cc: R_X86_64_PC32 cpu_online_mask-0x4
    > 26d0: 89 d2 mov %edx,%edx
    > 26d2: 48 0f a3 11 bt %rdx,(%rcx)
    > 26d6: 19 d2 sbb %edx,%edx
    > 26d8: 85 d2 test %edx,%edx
    > 26da: 41 0f 95 c4 setne %r12b
    >
    > if (!rcu_is_watching() && cpu_up)
    > 26f0: 83 e2 01 and $0x1,%edx
    > 26f3: 75 09 jne 26fe <__call_rcu+0x1ce>
    > 26f5: 45 84 e4 test %r12b,%r12b
    > 26f8: 0f 85 12 02 00 00 jne 2910 <__call_rcu+0x3e0>
    > invoke_rcu_core();
    >
    > /* If interrupts were disabled or CPU offline, don't invoke RCU core. */
    > if (irqs_disabled_flags(flags) || !cpu_up)
    > 26fe: 48 f7 45 d0 00 02 00 testq $0x200,-0x30(%rbp)
    > 2705: 00
    > 2706: 0f 84 df fe ff ff je 25eb <__call_rcu+0xbb>
    > 270c: 45 84 e4 test %r12b,%r12b
    > 270f: 0f 84 d6 fe ff ff je 25eb <__call_rcu+0xbb>
    >
    > --
    > Pranith
    >



    \
     
     \ /
      Last update: 2014-07-23 23:41    [W:3.602 / U:1.144 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site