[PATCH] ARM: fix perf build with uclibc toolchains

[Will Deacon]

On Wed, Aug 10, 2011 at 11:56:00AM +0100, Florian Fainelli wrote:
> On Wednesday 10 August 2011 12:21:55 Will Deacon wrote:
> > On Tue, Aug 09, 2011 at 09:08:53PM +0100, Will Deacon wrote:
> > > On Tue, Aug 09, 2011 at 12:24:17PM +0100, Florian Fainelli wrote:
> > > > libio.h is not provided by uClibc, in order to be able to test the
> > > > definition of __UCLIBC__ we need to include stdlib.h, which also
> > > > includes stddef.h, providing the definition of 'NULL'.
> > > > 
> > > > Signed-off-by: Florian Fainelli <florian at openwrt.org>
> > > > ---
> > > > diff --git a/tools/perf/arch/arm/util/dwarf-regs.c
> > > > b/tools/perf/arch/arm/util/dwarf-regs.c index fff6450..e8d5c55 100644
> > > > --- a/tools/perf/arch/arm/util/dwarf-regs.c
> > > > +++ b/tools/perf/arch/arm/util/dwarf-regs.c
> > > > @@ -8,7 +8,10 @@
> > > > 
> > > >   * published by the Free Software Foundation.
> > > >   */
> > > > 
> > > > +#include <stdlib.h>
> > > > +#ifndef __UCLIBC__
> > > > 
> > > >  #include <libio.h>
> > > > 
> > > > +#endif
> > > > 
> > > >  #include <dwarf-regs.h>
> > > >  
> > > >  struct pt_regs_dwarfnum {
> > > 
> > > Nice to see somebody trying to get this going with uclinux! I'll give
> > > this a quick compile-check on my boards and then I'll include it in my
> > > perf-fixes branch for 3.1.
> > 
> > Seems happy enough on my MMU boards. Since you're using this on noMMU
> > platforms, does that mean that you can build a static perf binary out of
> > the box or do you need further patches for this?
> 
> uClibc is not only for noMMU boards, I just built an uclibc toolchain for arm 
> targetting realview and run-tested this with qemu. This was the only patch 
> required to cross-compile perf for me. I also tested this with eglibc.

Ok, great. Cross-compiling perf usually just causes endless pain for me.

> By the way, perf top reports no even being generated for realview under qemu, 
> is that normal?

My guess is that qemu isn't emulating the cycle counter, which is required
to generate interrupts for use by perf. You could try disabling
HW_PERF_EVENTS which will force perf to fall back to a CPU tick instead.

Will