Re: [PATCH 3/6 v4] cpufreq: tolerate inexact values when collecting stats

On 11/11/2012 10:38 AM, Borislav Petkov wrote:

On Wed, Nov 07, 2012 at 12:32:43PM -0600, Mark Langsdorf wrote:

quoted

When collecting stats, if a frequency doesn't match the table, go through
the table again with both the search frequency and table values shifted
left by 10 bits.

Why would that second pass succeed?

It's effectively a divide by 1024 and minimizes any jitter in the
measured frequency value.

And why is this in generic code (I'm assuming this is a Calxeda-specific
case)?

The function is buried pretty deep in the cpufreq_stat code. It didn't
seem appropriate to make it a function pointer as part of struct
cpufreq_driver.

--Mark Langsdorf
Calxeda, Inc.