Port-sparc archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

patch for smp vs. monotonic time going backwards

i'd like people to test this out please.  i think i figured.
out the problem -- the stat clock is called with some fuz
on the interval, and this was triggering the hardclock()
call on not-cpu0 to break the timings.

because we don't want to give up having the stat clock have
this fuz, i've introduced an IPI to trigger a hardclock call
on each CPU, and call this after cpu0 calls hardclock itself.

with this in place, i'm unable to observe getrusage() seeing
garbage regularly, though at some point i did notice the
[system] process lose 0.02s of total cpu time, so there is
still a much smaller problem remaining (perhaps we have to
give up the stat clock, or perhaps switch two a tick-tock
on the stat clock, that calls hardclock() regularly and
statclock() with the fuz.)

in any case, this patch seems like a big step forward to me
but i would like to see test results from others before i
commit it and consider pullups.




Home | Main Index | Thread Index | Old Index