First, thanks for taking the time for the review!
I can do that. But how? Do I respin everything as x/5? Do I only send out
the PERL script as 5/4? Do I redo patch 1 as 1a/4 and 1b/4 (or 1.1/4 and
1.2/4)?
I like flexibility. So I allow this module to be configured with either
one or both (or none - if you just want some dead code!) of the available
interfaces. As an example, CONFIG_..._PLL_IF_SYSFS adds the sysfs
attribute to directly control the PLL from user space via writes to the
attribute file under /sys. On my system, a PowerMac 8600, this shows up in
/sys/devices/system/cpu/cpu0/ppc750gxpll. CONFIG_..._PLL_IF_CPU_FREQ adds
the public API functions that are used by the cpufreq driver.
CONFIG_..._PLL_IF_HRTIMER causes it to use the HR timer stuff. This
results in MUCH lower latencies (102 usec vs 3900 usec (HZ=250)). But I
did not want to REQUIRE HR timers (even though they are pretty cool!).
Did I mention that I like flexibility?
Seriously, should I add a check that at least one of ..._SYSFS or
..._CPU_FREQ is defined and refuse to compile otherwise? Via a pragma
or something?
I ... uh ... have no idea? Should I remove it?
This was just some goofing off in a debug message. I was playing around
trying to see what effect various code arrangements had on accuracy.
Since this is in debug code I was kinda hoping for some leniancy.
The "#if 0" is removing code that ... prevented preemption between the
processor speed switch and the changing of the loops_per_jiffy value.
The idea was that I did not want to change any sleeps. I now think
that processor speed is not a determining factor in delays. I think
the time base register (or decrementer) are used. I don't even change
the loops_per_jiffy any more (Maybe that is incorrect?). I left this
code in there to potentially jog my memory if I find myself debugging
some problem in the future? Would a comment be helpful?
At the tone the timebase will be 12499983. No trailing 0s. Very bad
for accuracy.
--
