Re: CVS commit: src/sys/arch

[David Brownlee <abs%absd.org@localhost>]

On Mon, 15 Jun 2026 at 16:56, Nick Hudson <nick.hudson%gmx.co.uk@localhost> wrote:
>
> On 15/06/2026 16:18, Nick Hudson wrote:
> > Module Name:  src
> > Committed By: skrll
> > Date:         Mon Jun 15 15:18:04 UTC 2026
> >
> > Modified Files:
> >       src/sys/arch/aarch64/include: pmap_machdep.h types.h
> >       src/sys/arch/mips/include: types.h
> >       src/sys/arch/powerpc/include: types.h
> >       src/sys/arch/riscv/include: types.h
> >
> > Log Message:
> > mi pmap: remove __HAVE_UNLOCKED_PMAP now the locking is fixed.
> >
> > The MI PMAP now allows the ability to process concurrent page faults on
> > individual uvm_objects / vm_amaps in parallel, where the relevant pages
> > are already in-core. As enabled in
> >
> > https://mail-index.netbsd.org/source-changes/2020/03/22/msg115362.html
>
> When performing 6 timed kernel -j64 builds on thunderx (32 cpus) this
> change sees a 30% improvement in system time.
>
> +------------------+--------------------+--------------------+
> | Metric           | Before Fix (Avg)   | After Fix (Avg)    |
> +------------------+--------------------+--------------------+
> | System Time      | 1176.03s           | 812.22s            |
> |   Abs Change     |  -363.82s          |                    |
> |   Rel Change     |   -30.94%          |                    |
> +------------------+--------------------+--------------------+
> | User Time        | 7186.63s           | 7211.47s           |
> |   Abs Change     |   +24.84s          |                    |
> |   Rel Change     |    +0.35%          |                    |
> +------------------+--------------------+--------------------+
> | Real Time        | 393.49s            | 384.69s            |
> |                  | (6m 33.5s)         | (6m 24.7s)         |
> |   Abs Change     |    -8.80s          |                    |
> |   Rel Change     |    -2.24%          |                    |
> +------------------+--------------------+--------------------+
>

a >2% wall clock speedup is not to be sneezed at either... Nice  :)

David