On Mon, Feb 02, 2015 at 01:39:33PM +0000, Will Deacon wrote:

On Fri, Jan 30, 2015 at 08:44:48AM +0000, Yong Zhang wrote:

quoted

On Fri, Jan 30, 2015 at 2:56 AM, Nathan Sullivan <nathan.sullivan at ni.com<mailto:nathan.sullivan@ni.com>> wrote:

While investigating a performance issue on RT 3.14, I noticed that
__HAVE_ARCH_CMPXCHG is not defined for ARM.  This causes -rt locks to
always use the slow path and impacts performance.  ARMv6 and above
have a cmpxchg implementation already that will work just fine.

In 2012, Yong Zhang submitted a patch to fix this[1], but it mysteriously
disappeared afterwards.  I did notice that the s-o-b line has a different
email address than the windriver one.  Yong, do you mind re-submitting the

It actually doesn't matter. Anyway feel free to add another s-o-b to
the patch:

Signed-off-by: Yong Zhang <yong.zhang at windriver.com<mailto:yong.zhang@windriver.com>>

I did a quick scan of architectures and looked at the various cmpxchg
implementations.  Assuming that __HAVE_ARCH_CMPXCHG really means "this
architecture implements a native cmpxchg which is faster than the
generic implementation", I put what I found in the following table; note
that I'm not 100% confident these are correct, as in some instances I
had to make a judgement call as to whether the implementation is "fast",
hopefully it's still useful :).

                 native         native
                cmpxchg (UP)   cmpxchg (SMP)    __HAVE_ARCH_CMPXCHG
          alpha     X               X                      X
            arc     X[1]            X[1]
arm (pre ARMv6)                    [2]
   arm (ARMv6+)     X               X
          arm64     X               X
          avr32     X               X                      X
       blackfin                     X
            c6x                    [2]
           cris                    [3]
            frv     X               X
        hexagon     X               X                      X
           ia64     X               X                      X
           m32r     X               X                      X
           m68k     X[4]            X[4]                   X[4]
          metag     X               X                      X
     microblaze                    [2]
           mips     X               X
        mn10300                     X[5]
          nios2                    [2]
       openrisc                    [6]
         parisc     X               X                      X
        powerpc     X               X                      X
           s390     X               X                      X
          score    [7]             [7]                     X[7]
             sh     X               X                      X
          sparc     X               X                      X
           tile     X               X
      unicore32                    [2]
            x86     X               X                      X
         xtensa     X               X

1: Conditional on CONFIG_ARC_HAS_LLSC
2: SMP not a supported configuration
3: No cmpxchg() implementation at all?
4: Conditional on CONFIG_RMW_INSNS
5: Conditional on CONFIG_MN10300_HAS_ATOMIC_OPS_UNIT
6: SMP not supported?  Uses generic cmpxchg.
7: Should use generic cmpxchg/cmpxchg_local?  Setting __HAVE_ARCH_CMPXCHG a bug?

quoted

+#define __HAVE_ARCH_CMPXCHG 1

Is it even possible to have CONFIG_SMP=y on an architecture that doesn't
support native cmpxchg? Certainly, the asm-generic cmpxchg.h barfs if
SMP, so couldn't we just change/extend the __HAVE_ARCH_CMPXCHG check in
rtmutex.c to work automatically with SMP architectures? That would help
arm64 too.

Looks like there are a few arch-specific conditionals that might need to
be worked out for arc, m68k, and mn10300; but this could potentially
work.

If we don't go down that route, we should probably at least add
__HAVE_ARCH_CMPXCHG for arm64, blackfin (on SMP), frv, mips, tile, and
xtensa as well.

  Josh