Re: [PATCH 0/5] Rewrite Motorola MMU page-table layout

[Michael Schmitz]

Peter,

Am 30.01.2020 um 08:31 schrieb Peter Zijlstra:
> On Thu, Jan 30, 2020 at 07:52:11AM +1300, Michael Schmitz wrote:
> > Peter,
> >
> > On Thu, Jan 30, 2020 at 12:54 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > > On Wed, Jan 29, 2020 at 11:49:13AM +0100, John Paul Adrian Glaubitz wrote:
> > >
> > > > > [1] https://wiki.debian.org/M68k/QemuSystemM68k
> > >
> > > Now, if only debian would actually ship that :/
> > >
> > > AFAICT that emulates a q800 which is another 68040 and should thus not
> > > differ from ARAnyM.
> > >
> > > I'm fairly confident in the 040 bits, it's the 020/030 things that need
> > > coverage.
> >
> > I'll take a look - unless this eats up way more kernel memory for page
> > tables, it should still boot on my Falcon.
>
> It should actually be better in most cases I think, since we no longer
> require all 16 pte-tables to map consecutive (virtual) memory.

Not much difference:

             total       used       free     shared    buffers     cached
Mem:         10712      10120        592          0       1860       2276
-/+ buffers/cache:       5984       4728
Swap:      2097144       1552    2095592


vs. vanilla 5.5rc5:
             total       used       free     shared    buffers     cached
Mem:         10716      10104        612          0       1588       2544
-/+ buffers/cache:       5972       4744
Swap:      2097144       1296    2095848

By sheer coincidence, the boot with your patch series happened to run a 
full filesystem check on the root filesystem, so I'd say it got a good 
workout re: paging and swapping (even though it's just a paltry 4 GB).

Haven't tried any VM stress testing yet (not sure what to do for that; 
it's been years since I tried that sort of stuff).

Cheers,

	Michael