module: reduce stack usage for each_symbol()
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 5 Aug 2010 18:59:05 +0000 (12:59 -0600)
committerRusty Russell <rusty@rustcorp.com.au>
Thu, 5 Aug 2010 03:29:06 +0000 (12:29 +0930)
commit44032e631691adf1f406843d5e54deb795973ff7
treee9cb4f72974386a2b174bf6e21f41be9aece7a10
parent22e268ebecc549f1b1907f114cb44d6044bdee3c
module: reduce stack usage for each_symbol()

And now that I'm looking at that call-chain (to see if it would make sense
to use some other more specific lock - doesn't look like it: all the
readers are using RCU and this is the only writer), I also give you this
trivial one-liner. It changes each_symbol() to not put that constant array
on the stack, resulting in changing

        movq    $C.388.31095, %rsi      #, tmp85
        subq    $376, %rsp      #,
        movq    %rdi, %rbx      # fn, fn
        leaq    -208(%rbp), %rdi        #, tmp84
        movq    %rbx, %rdx      # fn,
        rep movsl
        xorl    %esi, %esi      #
        leaq    -208(%rbp), %rdi        #, tmp87
        movq    %r12, %rcx      # data,
        call    each_symbol_in_section.clone.0  #

into

        xorl    %esi, %esi      #
        subq    $216, %rsp      #,
        movq    %rdi, %rbx      # fn, fn
        movq    $arr.31078, %rdi        #,
        call    each_symbol_in_section.clone.0  #

which is not so much about being obviously shorter and simpler because we
don't unnecessarily copy that constant array around onto the stack, but
also about having a much smaller stack footprint (376 vs 216 bytes - see
the update of 'rsp').

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
kernel/module.c