mm in user space

mm in user space

[Jim Potter]

I think I'm missing something here...
I'm porting Linux over to an embedded PPC project and I'm getting a DSI
crash when the kernel starts executing elf programs (specifically at
NEW_AUX_ENT in create_elf_tables() in fs/binfmt_elf.c; the stack
accesses are at 0x80000000, and the BATs are only expecting references
at 0xC0000000).
It's been a long day -- what am I missing?

--
Sincerely,

Jim Potter
45th Parallel Processing
jrp@wvi.com


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: mm in user space

[Jim Potter]

DOH!  It's *supposed* to be taking a DSI trap (slapping forehead).

> I think I'm missing something here...
> I'm porting Linux over to an embedded PPC project and I'm getting a DSI
> crash when the kernel starts executing elf programs (specifically at
> NEW_AUX_ENT in create_elf_tables() in fs/binfmt_elf.c; the stack
> accesses are at 0x80000000, and the BATs are only expecting references
> at 0xC0000000).
> It's been a long day -- what am I missing?
>
> --
> Sincerely,
>
> Jim Potter
> 45th Parallel Processing
> jrp@wvi.com
>

--
Sincerely,

Jim Potter
45th Parallel Processing
jrp@wvi.com


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: mm in user space

[Sébastien Côté]

Hi!

I got exactly the same problem, did you manage to fix it?  If so, could
you tell me how ?  I really don't see it right now.  Maybe it's time to
go home...

Thanks!

Sébastien Côté



Jim Potter wrote :

DOH!  It's *supposed* to be taking a DSI trap (slapping forehead).

> I think I'm missing something here...
> I'm porting Linux over to an embedded PPC project and I'm getting a DSI
> crash when the kernel starts executing elf programs (specifically at
> NEW_AUX_ENT in create_elf_tables() in fs/binfmt_elf.c; the stack
> accesses are at 0x80000000, and the BATs are only expecting references
> at 0xC0000000).
> It's been a long day -- what am I missing?
>
> --
> Sincerely,
>
> Jim Potter
> 45th Parallel Processing
> jrp@wvi.com
>

--
Sincerely,

Jim Potter
45th Parallel Processing
jrp@wvi.com

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: mm in user space

[Sébastien Côté]

Jim Potter wrote:
>
> > I got exactly the same problem, did you manage to fix it?
> Yes, I did (blushing).
> While doing the early boot-tasks I had installed a custom DSI trap
> handler, to alert me to all the stupid mistakes that sometimes get made
> when bringing up Linux on a new hardware platform.  When I got to the
> "user space" task of create_elf_tables(), it's *supposed* to take the
> regular DSI trap to fixup the TLB, pagetables, etc...
>
> After restoring the default DSI trap handler in head.S, it worked fine.
>
> What kind of problem are you having with your system?

I lose the system in create_elf_tables, when NEW_AUX_ENT is first called
but I don't really know what happens then because the debugger screws
up.

Today, I noticed that the values of PAGE_OFFSET (from which KERNELBASE
is set) and KERNELLOAD had been changed to 0x0 on the system..  This
seemed very bad to me so I set them back to 0xc0000000 but then the
debugger doesn't want to load the kernel image in memory (it tries to
load it at 0xc0000000 which isn't possible).. Then I just changed
KERNELLOAD back to 0x0 and kept KERNELBASE to 0xc0000000.

Now the system locks up at the very beginning (prom_init) but since
KERNELLOAD and KERNELBASE are different, I wasn't hoping for much (not
really beacuse they differ, but because the linker used 0x0).

Do you think it would be possible to get everything to work with
KERNELBASE and KERNELLOAD = 0x0 ?

Sébastien Côté

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/