More than 4G of memory on Sparc32?

More than 4G of memory on Sparc32?

[Blue Swirl]

Hi,

I've tried to add support to Qemu (and OpenBIOS) for more than 4G of
memory for the Sparc32 target. But now it seems that Linux can't use
this memory after all. The first show stopper is this code in
arch/sparc/prom/memory.c:

111                 /* Grrr, have to traverse the prom device tree ;( */
112                 node = prom_getchild(prom_root_node);
113                 node = prom_searchsiblings(node, "memory");
114                 num_regs = prom_getproperty(node, "available",
115                                             (char *) prom_reg_memlist,
116                                             sizeof(prom_reg_memlist));
117                 num_regs = (num_regs/sizeof(struct linux_prom_registers));
118                 for(iter=0; iter<num_regs; iter++) {
119                         prom_phys_avail[iter].start_adr 120                                 (char *) prom_reg_memlist[iter].phys_addr;
121                         prom_phys_avail[iter].num_bytes 122                                 (unsigned long)
prom_reg_memlist[iter].reg_size;
123                         prom_phys_avail[iter].theres_more 124                                 &prom_phys_avail[iter+1];
125                 }

Here prom_reg_memlist[iter].which_io is not used at all. It would be
needed to access the memory above the first 4G.

Some real Sparc32 machines could have >4G RAM (SC2000), maybe nobody
tried to run Linux on those. The BSDs aren't any better.

Will this ever be fixed? Anybody out there with a real machine and
lots of memory?

Re: More than 4G of memory on Sparc32?

[Chris Newport]

Blue Swirl wrote:

>Some real Sparc32 machines could have >4G RAM (SC2000), maybe nobody
>tried to run Linux on those. The BSDs aren't any better.
>
>Will this ever be fixed? Anybody out there with a real machine and
>lots of memory?
>  
>
AFAIK the only Sparc32 machines which could support more than
4G was the SS2000 and a Cray box which were both sun4d and therefore not 
currently supported.

These machines are rather scarce and there seems to be little
prospect of the sun4d issues being resolved, even for the smaller
SS1000.

 

Re: More than 4G of memory on Sparc32?

[David Miller]

From: "Blue Swirl" <blauwirbel@gmail.com>
Date: Wed, 26 Sep 2007 22:21:42 +0300

> Will this ever be fixed? Anybody out there with a real machine and
> lots of memory?

Since you have a qemu with >4GB support you are in the best
position the fix this :-)

Re: More than 4G of memory on Sparc32?

[Blue Swirl]

On 9/26/07, David Miller <davem@davemloft.net> wrote:
> From: "Blue Swirl" <blauwirbel@gmail.com>
> Date: Wed, 26 Sep 2007 22:21:42 +0300
>
> > Will this ever be fixed? Anybody out there with a real machine and
> > lots of memory?
>
> Since you have a qemu with >4GB support you are in the best
> position the fix this :-)

Maybe. There are other similar limits that could be explored and
fixed, for example Qemu  supports 16 CPUs on a plain Sun4m and there
are patches allowing lots of disks. But I don't see what's the point
fixing the issues if there are no real machines where removing these
limits would matter.

If I implemented in Qemu a completely new system (Sun4q?) with a
Sparc-like CPU of my own design, would you accept any patches to
support that in Linux?