Comment bug in locore.S?

[Olivier Houchard]

On Thu, Sep 27, 2007 at 11:46:33PM +0200, Marius Nuennerich wrote:
> On Fri, 28 Sep 2007 00:10:27 +0200
> Olivier Houchard <mlfbsd at ci0.org> wrote:
> 
> > On Thu, Sep 27, 2007 at 11:13:31PM +0200, Marius Nuennerich wrote:
> > > On Thu, 27 Sep 2007 23:41:41 +0200
> > > Olivier Houchard <mlfbsd at ci0.org> wrote:
> > > 
> > > > On Thu, Sep 27, 2007 at 10:41:33PM +0200, Marius Nuennerich wrote:
> > > > > On Thu, 27 Sep 2007 19:03:39 +0200
> > > > > Olivier Houchard <mlfbsd at ci0.org> wrote:
> > > > > 
> > > > > > On Thu, Sep 27, 2007 at 05:18:22PM +0200, Marius Nuennerich wrote:
> > > > > > > Hi all,
> > > > > > > 
> > > > > > 
> > > > > > Hi Marius,
> > > > > > 
> > > > > > > I'm still trying to boot FreeBSD on the neo1973. So far I managed to
> > > > > > > put bytes over the serial console so I can see where I am. At the
> > > > > > > moment the (virtual) machine crashes in locore.S (1.16) line 170:
> > > > > > > ldr pc, .Lvirt_done
> > > > > > > 
> > > > > > > qemu: fatal: Trying to execute code outside RAM or ROM at 0xc0008???
> > > > > > > 
> > > > > > > Which I don't understand because the section should be mapped by the
> > > > > > > MMU to 0x30008??? at that time.
> > > > > > > 
> > > > > > > In line 137 the sequence in the comment is described as
> > > > > > > /* # of sections, PA|attr, VA */
> > > > > > > but in line 185 the order is
> > > > > > > n_sec, VA, PA|attr
> > > > > > > 
> > > > > > > Is the comment wrong?
> > > > > > > 
> > > > > > 
> > > > > > No the comment is right, 0xc0000000 should be mapped.
> > > > > > Did you define STARTUP_PAGETABLE_ADDR ? It's not marked as such, but it is
> > > > > > mandatory now.
> > > > > 
> > > > > Yes, in the kernel config there are these lines:
> > > > > makeoptions     KERNPHYSADDR=0x30008000
> > > > > makeoptions     KERNVIRTADDR=0xc0000000
> > > > > options         KERNPHYSADDR=0x30008000
> > > > > options         KERNVIRTADDR=0xc0000000
> > > > > options         PHYSADDR=0x30000000
> > > > > options         STARTUP_PAGETABLE_ADDR=0x30800000
> > > > > 
> > > > > It's from Andrew Turners config file which he posted on
> > > > > http://wiki.openmoko.org/wiki/FreeBSD
> > > > > in the tarball.
> > > > > 
> > > > 
> > > > Hmmm, I really see no reason why the PA <=> PA mapping would work, but not
> > > > the VA <=> PA one.
> > > > I'll try to setup qemu. Can it provide useful informations about its 
> > > > internal state, ie CPU registers, TLB, etc ?
> > > 
> > > Yes, it looks like this:
> > > qemu: fatal: Trying to execute code outside RAM or ROM at 0xc0008030
> > > 
> > > R00=30800000 R01=c01afdb8 R02=c0000070 R03=00000000
> > > R04=3000824c R05=3000808c R06=00000002 R07=30008050
> > > R08=33f17fdc R09=00000000 R10=00000041 R11=50000020
> > > R12=33f14934 R13=c019b140 R14=33f8b9e8 R15=c0008030
> > > PSR=60000153 -ZC- A svc32
> > 
> > There's something I seriously miss. At this point, r2 should be 0 :
> > 	ldmia	r1, {r1, r2, sp}	/* Set initial stack and */
> > 	sub	r2, r2, r1		/* get zero init data */
> > 	mov	r3, #0
> > .L1:
> > 	str	r3, [r1], #0x0004	/* get zero init data */
> > 	subs	r2, r2, #4
> > 	bgt	.L1
> > 	ldr	pc, .Lvirt_done
> > 
> > And not 0xc0000070
> > What are the values of _edata and _end ?
> > If you've been able to do all those stores, the mapping should work anyway,
> > I'm really confused.
> 
> I'm sorry, I just realized that it doesn't crash right there. My debug
> ouput on serial line is ACDFFFGabcdeHIfA which indicates that it
> does jump but to an address before line 110. I attach my modified
> locore.S hopefully it won't get scrubbed.

Ah no wonder it explodes then, it goes back from the start, disables the MMU,
and can't figure out where 0xc0000000 is.
Now the big question is why it jumps there.
I'd guess qemu can do single-stepping.
Could you try to see which value it loads into pc ?

Regards,

Olivier