Re: [PATCH v2 2/2] powercap/rapl: reduce ipi calls

[Jacob Pan <jacob.jun.pan@linux.intel.com>]

On Wed, 13 Jan 2016 23:02:33 +0100 (CET)
Thomas Gleixner <tglx@linutronix.de> wrote:

> On Wed, 13 Jan 2016, Srinivas Pandruvada wrote:
> > On Wed, 2016-01-13 at 22:26 +0100, Borislav Petkov wrote:
> > > 	rmwmsrl_safe_on_cpu(policy->cpu, MSR_IA32_PERF_CTL,
> > > 			    INTEL_PERF_CTL_MASK,
> > > 			    (u32)sfi_cpufreq_array[next_perf_state].ctr
> > > l_val & INTEL_PERF_CTL_MASK);
> > > 
> > > Yikes!
> > > 
> > > So yes, it can work but it is ugly, hard to parse and use, not
> > > generic
> > > enough, etc, etc.
> > > 
> > > So thanks, but no thanks.
> > > 
> > I agree, in some cases it will not make much sense to use read-
> > modify_write calls, the user may decide whether it makes sense or
> > not. But such interface is not new to Linux kernel:
> > 
> > regmap_update_bits(), which is referenced for 346 times.
> > 
> > Are you saying that any such calls are not useful?
> 
> There are certainly cases when such calls are useful. And those cases
> are when we have a sufficiently big occurence of similar code which
> is sufficiently complex to justify the library code and the export.
> 
The balance of pros and cons depends on the number of occurrence. The
lib call overhead is constant where saving from the callers are
multiplied. Anyway, I will go back to my original code until we have
enough callers to tip the balance.