Re: [PATCH][2.6.7-rc3-mm1] perfctr cpumask cleanup

[Mikael Pettersson]

Paul Jackson writes:
 > > Clean up perfctr/virtual by using the new cpus_andnot() operation
 > 
 > Neat.
 > 
 > Do you still need "tmp" ?  Perhaps you could further add the
 > following patch (untested, unbuilt, ...).
 > 
 > This saves copies and stack space for one cpumask (that's
 > 512 bits on my SN2 systems).
 > 
 > Signed-off-by: Paul Jackson <pj@xxxxxxx>
 > 
 > Index: 2.6.7-rc3-mm1/drivers/perfctr/virtual.c
 > ===================================================================
 > --- 2.6.7-rc3-mm1.orig/drivers/perfctr/virtual.c	2004-06-09 15:34:34.000000000 -0700
 > +++ 2.6.7-rc3-mm1/drivers/perfctr/virtual.c	2004-06-09 15:38:32.000000000 -0700
 > @@ -403,11 +403,10 @@
 >  		return -EFAULT;
 >  
 >  	if (control.cpu_control.nractrs || control.cpu_control.nrictrs) {
 > -		cpumask_t tmp, old_mask, new_mask;
 > +		cpumask_t old_mask, new_mask;
 >  
 > -		tmp = perfctr_cpus_forbidden_mask;
 >  		old_mask = tsk->cpus_allowed;
 > -		cpus_andnot(new_mask, old_mask, tmp);
 > +		cpus_andnot(new_mask, old_mask, perfctr_cpus_forbidden_mask);

Doesn't work because cpus_andnot() requires all three parameters
to be lvalues. In UP and PowerPC builds, perfctr_cpus_forbidden_mask
is #define:d to CPU_MASK_NONE to allow client-side optimisations.

Making it always be a variable will slow down UP and PowerPC with
unoptimisable cpumask tests; alternatively I'll have to wrap my
cpumask uses with optimisation macros and/or #ifdefs.

/Mikael
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/