x86, perf, kprobes: Make kprobes's twobyte_is_boostable volatile

When compiling an i386_defconfig kernel with
gcc-4.6.1-9.fc15.i686, I noticed a warning about the asm operand
for test_bit in kprobes' can_boost. I discovered that this
caused only the first long of twobyte_is_boostable[] to be
output.

Jakub filed and fixed gcc PR50571 to correct the warning and
this output issue.  But to solve it for less current gcc, we can
make kprobes' twobyte_is_boostable[] volatile, and it won't be
optimized out.

Before:

    CC      arch/x86/kernel/kprobes.o
  In file included from include/linux/bitops.h:22:0,
                   from include/linux/kernel.h:17,
                   from [...]/arch/x86/include/asm/percpu.h:44,
                   from [...]/arch/x86/include/asm/current.h:5,
                   from [...]/arch/x86/include/asm/processor.h:15,
                   from [...]/arch/x86/include/asm/atomic.h:6,
                   from include/linux/atomic.h:4,
                   from include/linux/mutex.h:18,
                   from include/linux/notifier.h:13,
                   from include/linux/kprobes.h:34,
                   from arch/x86/kernel/kprobes.c:43:
  [...]/arch/x86/include/asm/bitops.h: In function ‘can_boost.part.1’:
  [...]/arch/x86/include/asm/bitops.h:319:2: warning: use of memory input without lvalue in asm operand 1 is deprecated [enabled by default]

  $ objdump -rd arch/x86/kernel/kprobes.o | grep -A1 -w bt
       551:	0f a3 05 00 00 00 00 	bt     %eax,0x0
                          554: R_386_32	.rodata.cst4

  $ objdump -s -j .rodata.cst4 -j .data arch/x86/kernel/kprobes.o

  arch/x86/kernel/kprobes.o:     file format elf32-i386

  Contents of section .data:
   0000 48000000 00000000 00000000 00000000  H...............
  Contents of section .rodata.cst4:
   0000 4c030000                             L...

Only a single long of twobyte_is_boostable[] is in the object
file.

After, with volatile:

  $ objdump -rd arch/x86/kernel/kprobes.o | grep -A1 -w bt
       551:	0f a3 05 20 00 00 00 	bt     %eax,0x20
                          554: R_386_32	.data

  $ objdump -s -j .rodata.cst4 -j .data arch/x86/kernel/kprobes.o

  arch/x86/kernel/kprobes.o:     file format elf32-i386

  Contents of section .data:
   0000 48000000 00000000 00000000 00000000  H...............
   0010 00000000 00000000 00000000 00000000  ................
   0020 4c030000 0f000200 ffff0000 ffcff0c0  L...............
   0030 0000ffff 3bbbfff8 03ff2ebb 26bb2e77  ....;.......&..w

Now all 32 bytes are output into .data instead.

Signed-off-by: Josh Stone <jistone@redhat.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Link: http://lkml.kernel.org/r/1318899645-4068-1-git-send-email-jistone@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
index f1a6244d7d93769c35d219b79de6358d8a4096e3..c0ed3d9c3b225dd4fa7266801ec917e0469ddbee 100644 (file)
--- a/arch/x86/kernel/kprobes.c
+++ b/arch/x86/kernel/kprobes.c
@@ -75,8 +75,10 @@ DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
        /*
         * Undefined/reserved opcodes, conditional jump, Opcode Extension
         * Groups, and some special opcodes can not boost.
+        * This is volatile to keep gcc from statically optimizing it out, as
+        * variable_test_bit makes gcc think only *(unsigned long*) is used.
         */
-static const u32 twobyte_is_boostable[256 / 32] = {
+static volatile const u32 twobyte_is_boostable[256 / 32] = {
        /*      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f          */
        /*      ----------------------------------------------          */
        W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */