m68k Kconfig warning

m68k Kconfig warning

[Randy Dunlap]

Hi Geert,

Just noticed this.  I don't know what the right fix is.
Would you take care of it, please?

on Linux 5.4, m68k allmodconfig:

WARNING: unmet direct dependencies detected for NEED_MULTIPLE_NODES
  Depends on [n]: DISCONTIGMEM [=n] || NUMA
  Selected by [y]:
  - SINGLE_MEMORY_CHUNK [=y] && MMU [=y]


thanks.
-- 
~Randy
Reported-by: Randy Dunlap <rdunlap@infradead.org>

Re: m68k Kconfig warning

[Geert Uytterhoeven]

Hi Randy,

On Wed, Nov 27, 2019 at 2:27 AM Randy Dunlap <rdunlap@infradead.org> wrote:
> Just noticed this.  I don't know what the right fix is.
> Would you take care of it, please?
>
> on Linux 5.4, m68k allmodconfig:
>
> WARNING: unmet direct dependencies detected for NEED_MULTIPLE_NODES
>   Depends on [n]: DISCONTIGMEM [=n] || NUMA
>   Selected by [y]:
>   - SINGLE_MEMORY_CHUNK [=y] && MMU [=y]

This has been basically there forever, but working.

Gr{oetje,eeting}s,

                        Geert


--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: m68k Kconfig warning

[Kars de Jong]

Hi Geert & Randy,

Op wo 27 nov. 2019 om 08:12 schreef Geert Uytterhoeven <geert@linux-m68k.org>:
> On Wed, Nov 27, 2019 at 2:27 AM Randy Dunlap <rdunlap@infradead.org> wrote:
> > Just noticed this.  I don't know what the right fix is.
> > Would you take care of it, please?
> >
> > on Linux 5.4, m68k allmodconfig:
> >
> > WARNING: unmet direct dependencies detected for NEED_MULTIPLE_NODES
> >   Depends on [n]: DISCONTIGMEM [=n] || NUMA
> >   Selected by [y]:
> >   - SINGLE_MEMORY_CHUNK [=y] && MMU [=y]
>
> This has been basically there forever, but working.

The reason for SINGLE_MEMORY_CHUNK depending on NEED_MULTIPLE_NODES is
historic due to the way it is implemented.
I played with it this weekend and I got a working version of FLATMEM,
which can replace SINGLE_MEMORY_CHUNK.

I'll clean it up and send a patch later this week. A possible next
step might be to replace DISCONTIGMEM with SPARSEMEM (since
DISCONTIGMEM has been deprecated).

Kind regards,

Kars.

Re: m68k Kconfig warning

[Geert Uytterhoeven]

Hi Kars,.

On Mon, Dec 2, 2019 at 12:42 PM Kars de Jong <karsdejong@home.nl> wrote:
> Op wo 27 nov. 2019 om 08:12 schreef Geert Uytterhoeven <geert@linux-m68k.org>:
> > On Wed, Nov 27, 2019 at 2:27 AM Randy Dunlap <rdunlap@infradead.org> wrote:
> > > Just noticed this.  I don't know what the right fix is.
> > > Would you take care of it, please?
> > >
> > > on Linux 5.4, m68k allmodconfig:
> > >
> > > WARNING: unmet direct dependencies detected for NEED_MULTIPLE_NODES
> > >   Depends on [n]: DISCONTIGMEM [=n] || NUMA
> > >   Selected by [y]:
> > >   - SINGLE_MEMORY_CHUNK [=y] && MMU [=y]
> >
> > This has been basically there forever, but working.
>
> The reason for SINGLE_MEMORY_CHUNK depending on NEED_MULTIPLE_NODES is
> historic due to the way it is implemented.
> I played with it this weekend and I got a working version of FLATMEM,
> which can replace SINGLE_MEMORY_CHUNK.

Nice, thanks!

> step might be to replace DISCONTIGMEM with SPARSEMEM (since
> DISCONTIGMEM has been deprecated).

Mike Rapoport has patches for that:
"[PATCH v2 0/3] m68k/mm: switch from DISCONTIGMEM to SPARSEMEM"

Unfortunately they're not on lore, and there were some issues with them.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: m68k Kconfig warning

[Mike Rapoport]

On Mon, Dec 02, 2019 at 02:32:28PM +0100, Geert Uytterhoeven wrote:
> Hi Kars,.
> 
> On Mon, Dec 2, 2019 at 12:42 PM Kars de Jong <karsdejong@home.nl> wrote:
> > Op wo 27 nov. 2019 om 08:12 schreef Geert Uytterhoeven <geert@linux-m68k.org>:
> > > On Wed, Nov 27, 2019 at 2:27 AM Randy Dunlap <rdunlap@infradead.org> wrote:
> > > > Just noticed this.  I don't know what the right fix is.
> > > > Would you take care of it, please?
> > > >
> > > > on Linux 5.4, m68k allmodconfig:
> > > >
> > > > WARNING: unmet direct dependencies detected for NEED_MULTIPLE_NODES
> > > >   Depends on [n]: DISCONTIGMEM [=n] || NUMA
> > > >   Selected by [y]:
> > > >   - SINGLE_MEMORY_CHUNK [=y] && MMU [=y]
> > >
> > > This has been basically there forever, but working.
> >
> > The reason for SINGLE_MEMORY_CHUNK depending on NEED_MULTIPLE_NODES is
> > historic due to the way it is implemented.
> > I played with it this weekend and I got a working version of FLATMEM,
> > which can replace SINGLE_MEMORY_CHUNK.
> 
> Nice, thanks!
> 
> > step might be to replace DISCONTIGMEM with SPARSEMEM (since
> > DISCONTIGMEM has been deprecated).
> 
> Mike Rapoport has patches for that:
> "[PATCH v2 0/3] m68k/mm: switch from DISCONTIGMEM to SPARSEMEM"
> 
> Unfortunately they're not on lore, and there were some issues with them.

The patches are here:
https://www.spinics.net/lists/linux-m68k/msg13588.html

Aside from some technicalities we had troubles deciding what should be the
section size. With larger section size we might end up with wasted memory
for memory maps and with smaller section size we'll have to limit the
addressable physical memory...


> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> -- 
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds

-- 
Sincerely yours,
Mike.

Re: m68k Kconfig warning

[Kars de Jong]

Hi Mike!

Op ma 2 dec. 2019 om 17:01 schreef Mike Rapoport <rppt@linux.ibm.com>:
> The patches are here:
> https://www.spinics.net/lists/linux-m68k/msg13588.html
>
> Aside from some technicalities we had troubles deciding what should be the
> section size. With larger section size we might end up with wasted memory
> for memory maps and with smaller section size we'll have to limit the
> addressable physical memory...

I read through that thread. I believe our current page->flags needs 22
bits for the normal FLAGS and 2 bits for the ZONES, which leaves 8
bits for the SECTION. This is what Geert found out (it worked with
MAX_PHYSMEM_BITS = 30 and SECTION_SIZE_BITS = 22).
I think we can reduces ZONES to a single bit. We currently put all
memory in ZONE_DMA, so I think we might just as well put it in
ZONE_NORMAL and disable ZONE_DMA.

That would enable us to have 8 MB sections and use the full 32-bit
address space, correct? That may be a working compromise?

Kind regards,

Kars.