[Skiboot] [PATCH 0/4] simulator SMP boot speedups

Nicholas Piggin npiggin at gmail.com
Tue Jun 6 21:46:34 AEST 2017


I'll cancel the ambulance...

Thanks for catching and fixing the bug!

Thanks,
Nick

On Tue, 06 Jun 2017 21:34:17 +1000
Stewart Smith <stewart at linux.vnet.ibm.com> wrote:

> Did you know there's an emacs shortcut for ROT13 that's suspiciously
> close to the "Send email" command for a tired Stewart? No. Neither did
> I, but now I do.
> 
> Fgrjneg Fzvgu <fgrjneg at yvahk.iarg.voz.pbz> jevgrf:
> > Nicholas Piggin <npiggin at gmail.com> writes:  
> >> This is a first pass at splitting out the boot speedups patch
> >> I sent earlier. It boots Linux to userspace with 2c16t POWER8
> >> simulated machine in about 12 seconds versus 45 seconds without
> >> the patches.
> >>
> >> The simulator doesn't like spinning for the timebase or polling
> >> for a condition at a high SMT priority, which causes cycles to
> >> be spent on useless work which hurts a lot more than on real
> >> hardware.
> >>
> >> This is probably not for merge yet (needs more testing on different
> >> machines first), but comments would be appreciated.
> >>
> >> Thanks,
> >> Nick
> >>
> >>
> >> Nicholas Piggin (4):
> >>   cpu: Introduce smt_lowest()
> >>   Improve cpu_idle when PM is disabled
> >>   Convert important polling loops to spin at lowest SMT priority
> >>   Run change_all_hile in parallel on all CPUs  
> >
> > I took patches 1-3 as of 05b8834b5a4ff9bd14548594700d354bc4486c5c, with
> > a fixup to the polling loops patch so that we worked with the lpc-uart
> > when the tx fifo was full.
> >
> > I wanted to double check something in the hile patch before taking it,
> > but it got late and I had a giant pile of things to merge already, so
> > will look at it later :)  
> 



More information about the Skiboot mailing list