TSC instead of ACPI: powerd doesn't work anymore (to be expected?)

[M. Warner Losh]

In message: <20051030195936.GZ4115 at funkthat.com>
            John-Mark Gurney <gurney_j at resnet.uoregon.edu> writes:
: Scott Long wrote this message on Sun, Oct 30, 2005 at 07:30 -0700:
: > >Correct, but why is it felt the latency of the ACPI timer is an issue?
: > >Of course we all want things to as fast as possible, but is that just an
: > >abstract desire, or a real issue was run into?
: > >
: > 
: > ACPI-fast requires an ioport read which takes about 1us (according to 
: > Google).  Do that 1000 times a second and you have each CPU spending
: > 1% of its time doing nothing but reading the clock.  Yikes.
: 
: Math correction: .1%...  1us * 1000 = 1ms, 1ms = .001s, .001 * 100% = .1%

Actually, ioport reads can be faster than the 1us that's widely
quoted.  ioport reads can be as fast as ~125ns (2 cycles at 16MHz).
However, experience has shown that they are rarely this fast.  I've
seen 8MB/s pio over the pci bus on some custom hardware we have, which
2Mreads/sec which is about 500ns per read.  I think that the pci
hardware that I was reading had a few extra wait states...

The 1us/read is for devices on the ISA bus or for hardware that
emulates this timing.

Warner