In English

(http://www.netbsd.org/docs/kernel/profiling/)

In Japanese

(http://www.netbsd.org/ja/docs/kernel/profiling/)

名前は何でもいいけれど、例えば sys/arch/amd64/conf/GENERIC.PROF という名前で次の内容を用意して

include "arch/amd64/conf/GENERIC"
makeoptions     PROF="-pg"
options         GPROF

time sudo ./build.sh -m amd64 -j 4 -T /export/src/tooldir.NetBSD-4.99.49-x86_64 kernel=GENERIC.PROF

で、以前に見た気がするが:

/var/tmp//cci6nNif.s: Assembler messages:
/var/tmp//cci6nNif.s:189: Error: suffix or operands invalid for `pushf'
/var/tmp//cci6nNif.s:189: Error: suffix or operands invalid for `pop'
/var/tmp//cci6nNif.s:237: Error: suffix or operands invalid for `push'
/var/tmp//cci6nNif.s:237: Error: suffix or operands invalid for `popf'
/var/tmp//cci6nNif.s:283: Error: suffix or operands invalid for `push'
/var/tmp//cci6nNif.s:283: Error: suffix or operands invalid for `popf'

もう一度 nbmake

cd /export/src/sys/arch/amd64/compile/obj/GENERIC.PROF
sudo /export/src/tooldir.NetBSD-4.99.49-x86_64/bin/nbmake-amd64

もう一度

cd /export/src/sys/arch/amd64/compile/obj/GENERIC.PROF/lib/kern;
pts/0:makoto@berona 9:38:22/080115(...lib/kern)> sudo /export/src/tooldir.NetBSD-4.99.49-x86_64/bin/nbmake -f /export/src/sys/lib/libkern/Makefile KERNDIR=/export/src/sys/lib/libkern CC=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-gcc CFLAGS=\ -mcmodel=kernel\ -mno-red-zone\ -ffreestanding\ -fno-zero-initialized-in-bss\ \ \ -O2\ -fno-omit-frame-pointer\ -Werror\ -Wall\ -Wno-main\ -Wno-format-zero-length\ -Wpointer-arith\ -Wmissing-prototypes\ -Wstrict-prototypes\ -Wswitch\ -Wshadow\ -Wcast-qual\ -Wwrite-strings\ -Wno-sign-compare\ -Wno-pointer-sign\ -Wno-attributes\ \ -fno-strict-aliasing\ \ AS=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-as AFLAGS=\ -x\ assembler-with-cpp\ -traditional-cpp\ \ -D_LOCORE\ -Wa,-fatal-warnings\ LORDER=NM=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-nm\ MKTEMP=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/nbmktemp\ /export/src/tooldir.NetBSD-4.99.49-x86_64/bin/nblorder TSORT=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/nbtsort\ -q LD=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-ld STRIP=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-strip AR=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-ar NM=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-nm RANLIB=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-ranlib SIZE=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-size MACHINE=amd64 MACHINE_ARCH=x86_64 KERNCPPFLAGS=-Damd64\ -Dx86_64\ -I../../.\ -I/export/src/sys/contrib/dev/ath/netbsd\ -I/export/src/sys/../common/include\ -I/export/src/sys/arch\ -I/export/src/sys\ -nostdinc\ -DLKM\ -DGPROF\ -DMAXUSERS=32\ -D_KERNEL\ -D_KERNEL_OPT\ -I/export/src/sys/lib/libkern/../../../common/lib/libc/quad\ -I/export/src/sys/lib/libkern/../../../common/lib/libc/string\ -I/export/src/sys/lib/libkern/../../../common/lib/libc/arch/x86_64/string\ -I/export/src/sys/dist/ipf KERNMISCCPPFLAGS= LINTFLAGS=-bcehnxzFS libkern.po

更に cc だけ起動

/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-gcc -mcmodel=kernel -mno-red-zone -ffreestanding -fno-zero-initialized-in-bss -O2 -fno-omit-frame-pointer -Werror -Wall -Wno-main -Wno-format-zero-length -Wpointer-arith -Wmissing-prototypes -Wstrict-prototypes -Wswitch -Wshadow -Wcast-qual -Wwrite-strings -Wno-sign-compare -Wno-pointer-sign -Wno-attributes -fno-strict-aliasing AS=/export/src/tooldir.NetBSD-4.99.49-x86_64/bin/x86_64--netbsd-as -I/export/src/sys/lib/libkern/arch/x86_64 -Damd64 -Dx86_64 -I../../. -I/export/src/sys/contrib/dev/ath/netbsd -I/export/src/sys/../common/include -I/export/src/sys/arch -I/export/src/sys -nostdinc -DLKM -DGPROF -DMAXUSERS=32 -D_KERNEL -D_KERNEL_OPT -I/export/src/sys/lib/libkern/../../../common/lib/libc/quad p-I/export/src/sys/lib/libkern/../../../common/lib/libc/string -I/export/src/sys/lib/libkern/../../../common/lib/libc/arch/x86_64/string -I/export/src/sys/dist/ipf -I/export/src/sys/lib/libkern/../../../common/lib/libc/quad -I/export/src/sys/lib/libkern/../../../common/lib/libc/string -I/export/src/sys/lib/libkern/../../../common/lib/libc/arch/x86_64/string -I/export/src/sys/lib/libkern/../../../common/include -c -DGPROF -DPROF -pg /export/src/sys/lib/libkern/../../../common/lib/libc/gmon/mcount.c -o mcount.po

Following location are flagged:

    188 /APP    
    189         pushfl; popl %rax
    190 /NO_APP  

    236 /APP
    237         pushl %rax; popfl
    238 /NO_APP

    282 /APP
    283         pushl %rax; popfl
    284 /NO_APP

Those lines are derived from: src/sys/arch/amd64/include/profile.h

    135 static inline u_long
    136 mcount_read_psl(void)
    137 {
    138         u_long  ef;
    139
    140         __asm volatile("pushfl; popl %0" : "=r" (ef));
    141         return (ef);
    142 }

    144 static inline void
    145 mcount_write_psl(u_long ef)
    146 {
    147         __asm volatile("pushl %0; popfl" : : "r" (ef)); 
    148 }

oshimaya さんの 2006/04/19 の日記 にも同じ字がある。あれ、自分で メール を書いている ? でも ef が何故 ax に変わるのかな ? CPUID instruction という種類のもの ? もしかして patch

Profiling kernel, textsize=6140264 [ffffffff80100000..ffffffff806db168]

   118  11:33   sudo kgmon -b
   119  11:33   sudo ./build.sh -m amd64 -j 4 -T /export/src/tooldir.NetBSD-4.99.49-x86_64 kernel=GENERIC.PROF
   120  11:42   sudo kgmon -h
   121  11:42   sudo kgmon -p
   126  11:44   gprof /netbsd gmon.out > gmon.txt
   127  11:46   less gmon.txt

Flat profile:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  ms/call  ms/call  name    
 80.37    390.90   390.90                             x86_stihlt
  1.68    399.05     8.15                             mutex_enter
  1.46    406.16     7.11                             Xspllower
  1.34    412.69     6.53  5817777     0.00     0.00  sse2_zero_page
  0.87    416.93     4.24 11307330     0.00     0.00  pmap_enter
  0.71    420.37     3.44 30737349     0.00     0.00  pvtree_SPLAY
  0.61    423.36     2.99  7294905     0.00     0.01  uvm_fault_internal
  0.60    426.27     2.91                             mutex_spin_enter
  0.59    429.13     2.86  2555182     0.00     0.00  copyout
  0.50    431.58     2.45   420077     0.01     0.01  _kernel_lock
  0.43    433.66     2.08 51671343     0.00     0.00  lockmgr
  0.41    435.64     1.98                             calltrap
  0.38    437.50     1.86 25561689     0.00     0.00  pool_cache_get_paddr
  0.37    439.32     1.82                             x86_pause
  0.36    441.07     1.75  6560728     0.00     0.00  uvm_pagealloc_pgfl
  0.33    442.69     1.62 10687457     0.00     0.00  cache_lookup
  0.31    444.20     1.51  5833980     0.00     0.00  pmap_zero_page

sys/arch/amd64/amd64/cpufunc.S:

   310  NENTRY(x86_stihlt)
   311          pushq   %rbp
   312          movq    %rsp, %rbp
   313          sti
   314          hlt
   315          leave
   316          ret

But the calls (call count) is not shown, so, is this just a wait ? No, the blank on 'calls' means not-profiled.

sti	Set Interrupt flag
hlt	Halt
leave	High Level Procedure Exit

x86 Assembly Language Reference Manual