Linux prom mmu mapping expectations?

Linux prom mmu mapping expectations?

[Blue Swirl]

Hi,

What kind of memory mappings Linux expects from Prom side?

Background: I've been trying to get Qemu run Linux 2.6.8.1 for sparc32, but 
currently there is a MMU fault at srmmu_early_allocate_ptable_skeleton.  The 
code in question is as follows:
0xf00fdec4:  ldd  [ %o1 + -64 ], %g2
0xf00fdec8:  ldd  [ %o1 + -56 ], %g4
0xf00fdecc:  st  %g2, [ %o0 + -64 ]
0xf00fded0:  st  %g3, [ %o0 + -60 ]
0xf00fded4:  st  %g4, [ %o0 + -56 ]
0xf00fded8:  st  %g5, [ %o0 + -52 ]
0xf00fdedc:  ldd  [ %o1 + -48 ], %g2
0xf00fdee0:  ldd  [ %o1 + -40 ], %g4

This code is in memcpy, but I can't find where it is called from in the 
function (srmmu_early_allocate_ptable_skeleton, srmmu.c).

The faulting address is 0xf00fdec4, %o1 is 0xf0000000. There is no MMU 
mapping at the address. I tried adding some code into Proll to map the area, 
but that didn't help.

Any clues, please?

_________________________________________________________________
Add photos to your messages with MSN 8. Get 2 months FREE*. 
http://join.msn.com/?pageþatures/featuredemail

Re: Linux prom mmu mapping expectations?

[David S. Miller]

On Thu, 07 Oct 2004 20:22:38 +0200
"Blue Swirl" <blueswir1@hotmail.com> wrote:

> Background: I've been trying to get Qemu run Linux 2.6.8.1 for sparc32, but 
> currently there is a MMU fault at srmmu_early_allocate_ptable_skeleton.  The 
> code in question is as follows:
> 0xf00fdec4:  ldd  [ %o1 + -64 ], %g2
> 0xf00fdec8:  ldd  [ %o1 + -56 ], %g4
> 0xf00fdecc:  st  %g2, [ %o0 + -64 ]
> 0xf00fded0:  st  %g3, [ %o0 + -60 ]
> 0xf00fded4:  st  %g4, [ %o0 + -56 ]
> 0xf00fded8:  st  %g5, [ %o0 + -52 ]
> 0xf00fdedc:  ldd  [ %o1 + -48 ], %g2
> 0xf00fdee0:  ldd  [ %o1 + -40 ], %g4
> 
> This code is in memcpy, but I can't find where it is called from in the 
> function (srmmu_early_allocate_ptable_skeleton, srmmu.c).
> 
> The faulting address is 0xf00fdec4, %o1 is 0xf0000000. There is no MMU 
> mapping at the address. I tried adding some code into Proll to map the area, 
> but that didn't help.

0xf0000000 is KERNBASE, and arch/sparc/kernel/head.S load mappings for the
entire kernel image, starting at KERNBASE, into the MMU.

Re: Linux prom mmu mapping expectations?

[Blue Swirl]

>0xf0000000 is KERNBASE, and arch/sparc/kernel/head.S load mappings for the
>entire kernel image, starting at KERNBASE, into the MMU.

Thanks. I found the bug in Proll:
--- proll_18.orig/src/system.c  2002-07-23 07:52:49.000000000 +0200
+++ proll_18.min/src/system.c   2004-10-08 18:08:50.000000000 +0200
@@ -298,8 +298,8 @@
        }

        /* We need to start from LOADBASE, but kernel wants PAGE_SIZE. */
-       pa = PAGE_SIZE;
-       for (va = PAGE_SIZE; va < LOWMEMSZ; va += PAGE_SIZE) {
+       pa = 0;
+       for (va = 0; va < LOWMEMSZ; va += PAGE_SIZE) {
                map_page(l1, va, pa, 0, highbase);
                pa += PAGE_SIZE;
        }

Now we are stuck at poke_srmmu call faulting mysteriously, even though 
poke_srmmu is correctly initialized earlier at init_cypress_604. Kill one 
bug, advance 500 instructions, next bug halts progress.

_________________________________________________________________
Tired of spam? Get advanced junk mail protection with MSN 8. 
http://join.msn.com/?pageþatures/junkmail