Re: [OpenBIOS] [Qemu-devel] Re: What is the current support state for Sparc emulation

[Blue Swirl]

On 7/19/07, [e-mail]  wrote:
> Unhandled Exception 0x00000007
>   PC = 0xffd0a55c NPC = 0xffd09374
>     Stopping execution
>
> So this is my report, I hope it can be of some use. I'll try to boot sparc version of OpenSolaris soon and I'l let you know how it went.

Thanks for the report. Exception 7 is unaligned access, support for
detection of unaligned accesses in QEMU was improved recently.

The exception happens inside OpenBIOS:
0xffd0a55c is in lstore (kernel/forth.c:649).
644
645     static void lstore(void)
646     {
647             const u32 *aaddr = (u32 *)cell2pointer(POP());
648             const u32 longval = POP();
649             write_long(aaddr, longval);
650     }

Maybe write_long should handle unaligned addresses?

[Qemu-devel] Re: What is the current support state for Sparc emulation

[Stefan Reinauer]

* Blue Swirl  [070719 20:46]:
> On 7/19/07, [e-mail]  wrote:
> > Unhandled Exception 0x00000007
> >   PC = 0xffd0a55c NPC = 0xffd09374
> >     Stopping execution
> >
> > So this is my report, I hope it can be of some use. I'll try to boot sparc version of OpenSolaris soon and I'l let you know how it went.
>
> Thanks for the report. Exception 7 is unaligned access, support for
> detection of unaligned accesses in QEMU was improved recently.
>
> The exception happens inside OpenBIOS:
> 0xffd0a55c is in lstore (kernel/forth.c:649).
> 644
> 645     static void lstore(void)
> 646     {
> 647             const u32 *aaddr = (u32 *)cell2pointer(POP());
> 648             const u32 longval = POP();
> 649             write_long(aaddr, longval);
> 650     }
>
> Maybe write_long should handle unaligned addresses?

Possibly. But something is wrong.

Unaligned accesses should be done with unaligned-l@  ( addr -- quad )
instead of l@.

So either the pointer was calculated wrong or Solaris does not care
about alignment in l@

If you change it, please make the change specific to sparc.

Stefan

[Qemu-devel] Re: What is the current support state for Sparc emulation

[Blue Swirl]

> > Maybe write_long should handle unaligned addresses?
>
> Possibly. But something is wrong.
>
> Unaligned accesses should be done with unaligned-l@  ( addr -- quad )
> instead of l@.
>
> So either the pointer was calculated wrong or Solaris does not care
> about alignment in l@
>
> If you change it, please make the change specific to sparc.

I noticed that the memory allocators either don't enforce alignment or
align just to 4. That could be one reason for the problem.

[Qemu-devel] Re: What is the current support state for Sparc emulation

[Stefan Reinauer]

* Blue Swirl  [070720 15:11]:
> > Unaligned accesses should be done with unaligned-l@  ( addr -- quad )
> > instead of l@.
> >
> > So either the pointer was calculated wrong or Solaris does not care
> > about alignment in l@
> >
> > If you change it, please make the change specific to sparc.
>
> I noticed that the memory allocators either don't enforce alignment or
> align just to 4. That could be one reason for the problem.

In OpenBIOS? That should definitely be fixed!

Alignment to 4 should be fine for l@ though..

Stefan