Messages in this thread | | | Date | Thu, 10 Sep 2009 20:24:08 +0200 | From | Frederic Weisbecker <> | Subject | Re: [RFC GIT PULL] hw-breakpoints: Rewrite on top of perf counters |
| |
On Thu, Sep 10, 2009 at 08:09:56PM +1000, Paul Mackerras wrote: > Frederic Weisbecker writes: > > > This is a rewrite of the hardware breakpoints on top of perf counters. > > On powerpc, it doesn't build. I get: > > CC kernel/perf_counter.o > kernel/perf_counter.c:31:31: error: asm/hw_breakpoint.h: No such file or directory > kernel/perf_counter.c: In function 'bp_perf_counter_init': > kernel/perf_counter.c:3964: error: implicit declaration of function 'register_perf_hw_breakpoint' > kernel/perf_counter.c:3966: error: implicit declaration of function '__register_perf_hw_breakpoint' > kernel/perf_counter.c:3968: error: 'perf_ops_bp' undeclared (first use in this function) > kernel/perf_counter.c:3968: error: (Each undeclared identifier is reported only once > kernel/perf_counter.c:3968: error: for each function it appears in.) > make[2]: *** [kernel/perf_counter.o] Error 1 > make[1]: *** [kernel] Error 2 > make: *** [sub-make] Error 2 > > Seems like every architecture now needs an asm/hw_breakpoint.h. What > is the minimum required in that file? Looks like we would require a > perf_ops_bp, at least. > > Could you please either arrange things so that architectures that > don't have hardware breakpoints hooked up to perf_counters don't need > an asm/hw_breakpoint.h, or add minimal versions of that file for every > architecture, so as not to break bisection unnecessarily? > > Paul.
I focused so much on the breakpoint <- perf dependency that I've even forgotten the opposite dependency :)
I'll fix it in the 3rd patch. We have an <asm-generic/hw-breakpoint.h> so I can just add an #ifdef HAVE_HW_BREAKPOINT on its top.
Also I'll export the various breakpoints api dependant out of kernel/perf_counter.c and provide stubs for the off-case.
Thanks.
| |