Re: [PATCH] x86,perf: Unmask LVTPC only if we have APIC supported

[Cyrill Gorcunov <gorcunov@gmail.com>]

On Sat, Mar 13, 2010 at 01:40:36PM +0100, Ingo Molnar wrote:
> 
> * Cyrill Gorcunov <gorcunov@gmail.com> wrote:
> 
> > On Sat, Mar 13, 2010 at 01:24:32PM +0100, Ingo Molnar wrote:
> > > 
> > > * Cyrill Gorcunov <gorcunov@openvz.org> wrote:
> > > 
> > > > Ingo reported
> > > > |
> > > > | There's a build failure on -tip with the P4 driver, on UP 32-bit, if
> > > > | PERF_EVENTS is enabled but UP_APIC is disabled:
> > > > |
> > > > | arch/x86/built-in.o: In function `p4_pmu_handle_irq':
> > > > | perf_event.c:(.text+0xa756): undefined reference to `apic'
> > > > | perf_event.c:(.text+0xa76e): undefined reference to `apic'
> > > > |
> > > > 
> > > > So we have to unmask LVTPC only if we're configured to have one.
> > > > 
> > > > Reported-by: Ingo Molnar <mingo@elte.hu>
> > > > CC: Lin Ming <ming.m.lin@intel.com>
> > > > CC: Peter Zijlstra <peterz@infradead.org>
> > > > Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
> > > > ---
> > > >  arch/x86/kernel/cpu/perf_event_p4.c |    2 ++
> > > >  1 file changed, 2 insertions(+)
> > > > 
> > > > Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
> > > > =====================================================================
> > > > --- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event_p4.c
> > > > +++ linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
> > > > @@ -365,8 +365,10 @@ static int p4_pmu_handle_irq(struct pt_r
> > > >  	}
> > > >  
> > > >  	if (handled) {
> > > > +#ifdef CONFIG_X86_LOCAL_APIC
> > > >  		/* p4 quirk: unmask it again */
> > > >  		apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
> > > > +#endif
> > > >  		inc_irq_stat(apic_perf_irqs);
> > > 
> > > This ugly #ifdef looks like a workaround though. Why doesnt apic_write() map 
> > > to nothing in that case?
> > > 
> > > 	Ingo
> > > 
> > 
> > It is. I mean -- it maps to nothing if apic is disabled. But the scenario is 
> > that no apic configured at all. Actually I wonder how this code is supposed 
> > to work without apic support.
> > 
> > Pehpaps better to make a p4 quirk helper here, since #ifdef at this point 
> > looks ugly indeed.
> > 
> > Don't apply it then. Will back with other solution.
> 
> apic_write() is really just equivalent to a spin_lock() on UP without 
> UP_IOAPIC set - it should do nothing. So if it does something and fails the 
> build, then that should be fixed - not the P4 PMU code.
> 
> 	Ingo
> 

Looking at code a bit and config deps I think the former proposal with
#ifdef is minimal (in amount of changes) and sufficient. perf_event.c
uses #ifdef CONFIG_X86_LOCAL_APIC for the very same reason.

The former issue with config dependencies is that we may need to compile
perf_event.c without CONFIG_LOCAL_APIC support at all (and this is a case
for which you've posted the config). CONFIG_LOCAL_APIC deps on X86_UP_APIC,
the config has no X86_UP_APIC support and as result -- no CONFIG_LOCAL_APIC and
no apic.o compiled.

So, as expected, no apic_write/read and friends there. We may introduce
apic_write/read weak(s) but this would only mess the code more and would
smell unpleasant I think :) .

All-in-once: unresolved external symbol here, which could be fixed either
by introducing dummy symbol, or conditional compilation. I think the second
is preferred if the issue is just one line code.

Or you mean something different and I took a wrong mind-path?

	-- Cyrill