• Denys Vlasenko's avatar
    include/linux/unaligned: force inlining of byteswap operations · e3bde956
    Denys Vlasenko authored
    Sometimes gcc mysteriously doesn't inline
    very small functions we expect to be inlined. See
    
        https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122
    
    With this .config:
    http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os,
    the following functions get deinlined many times.
    Examples of disassembly:
    
    <get_unaligned_be16> (24 copies, 108 calls):
           66 8b 07                mov    (%rdi),%ax
           55                      push   %rbp
           48 89 e5                mov    %rsp,%rbp
           86 e0                   xchg   %ah,%al
           5d                      pop    %rbp
           c3                      retq
    
    <get_unaligned_be32> (25 copies, 181 calls):
           8b 07                   mov    (%rdi),%eax
           55                      push   %rbp
           48 89 e5                mov    %rsp,%rbp
           0f c8                   bswap  %eax
           5d                      pop    %rbp
           c3                      retq
    
    <get_unaligned_be64> (23 copies, 94 calls):
           48 8b 07                mov    (%rdi),%rax
           55                      push   %rbp
           48 89 e5                mov    %rsp,%rbp
           48 0f c8                bswap  %rax
           5d                      pop    %rbp
           c3                      retq
    
    <put_unaligned_be16> (2 copies, 11 calls):
           89 f8                   mov    %edi,%eax
           55                      push   %rbp
           c1 ef 08                shr    $0x8,%edi
           c1 e0 08                shl    $0x8,%eax
           09 c7                   or     %eax,%edi
           48 89 e5                mov    %rsp,%rbp
           66 89 3e                mov    %di,(%rsi)
    
    <put_unaligned_be32> (8 copies, 43 calls):
           55                      push   %rbp
           0f cf                   bswap  %edi
           89 3e                   mov    %edi,(%rsi)
           48 89 e5                mov    %rsp,%rbp
           5d                      pop    %rbp
           c3                      retq
    
    <put_unaligned_be64> (26 copies, 157 calls):
           55                      push   %rbp
           48 0f cf                bswap  %rdi
           48 89 3e                mov    %rdi,(%rsi)
           48 89 e5                mov    %rsp,%rbp
           5d                      pop    %rbp
           c3                      retq
    
    This patch fixes this via s/inline/__always_inline/.
    
    It only affects arches with efficient unaligned access insns, such as x86.
    (arched which lack such ops do not include linux/unaligned/access_ok.h)
    
    Code size decrease after the patch is ~8.5k:
    
        text     data      bss       dec     hex filename
    92197848 20826112 36417536 149441496 8e84bd8 vmlinux
    92189231 20826144 36417536 149432911 8e82a4f vmlinux6_unaligned_be_after
    Signed-off-by: 's avatarDenys Vlasenko <dvlasenk@redhat.com>
    Acked-by: 's avatarIngo Molnar <mingo@kernel.org>
    Cc: Thomas Graf <tgraf@suug.ch>
    Cc: Peter Zijlstra <peterz@infradead.org>
    Cc: David Rientjes <rientjes@google.com>
    Cc: Arnd Bergmann <arnd@arndb.de>
    Signed-off-by: 's avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: 's avatarLinus Torvalds <torvalds@linux-foundation.org>
    e3bde956
Name
Last commit
Last update
..
access_ok.h Loading commit data...
be_byteshift.h Loading commit data...
be_memmove.h Loading commit data...
be_struct.h Loading commit data...
generic.h Loading commit data...
le_byteshift.h Loading commit data...
le_memmove.h Loading commit data...
le_struct.h Loading commit data...
memmove.h Loading commit data...
packed_struct.h Loading commit data...