Definition of PT_TEXT_ADDR and friends.

Hello!

When compiling gdb for uClinux with armnommu, I noticed that
(in
linux-low.c) PT_TEXT_ADDR, PT_DATA_ADDR and PT_TEXT_END_ADDR
are
defined only when __mcoldfire__ is defined.  I think this is
wrong.
Instead, IMHO, it should always be defined when we have
linux without
a mmu.

Comments?

On Tue, Mar 20, 2007 at 07:44:56AM +0100, Josef Wolf wrote:
> Hello!
> 
> When compiling gdb for uClinux with armnommu, I noticed
that (in
> linux-low.c) PT_TEXT_ADDR, PT_DATA_ADDR and
PT_TEXT_END_ADDR are
> defined only when __mcoldfire__ is defined.  I think
this is wrong.
> Instead, IMHO, it should always be defined when we have
linux without
> a mmu.
> 
> Comments?

/* These should really be defined in the kernel's ptrace.h
header.  */

We don't know what to define them _to_ on other platforms. 
If you
have another platform which fails to define them in the
kernel
headers, we can add another case.

(I know the __UCLIBC_HAS_MMU__ is out of date - I'm going to
fix that
today)

-- 
Daniel Jacobowitz
CodeSourcery

On Tue, Mar 20, 2007 at 06:56:18AM -0400, Daniel Jacobowitz
wrote:
> On Tue, Mar 20, 2007 at 07:44:56AM +0100, Josef Wolf
wrote:

> > When compiling gdb for uClinux with armnommu, I
noticed that (in
> > linux-low.c) PT_TEXT_ADDR, PT_DATA_ADDR and
PT_TEXT_END_ADDR are
> > defined only when __mcoldfire__ is defined.  I
think this is wrong.
> > Instead, IMHO, it should always be defined when we
have linux without
> > a mmu.
> > 
> > Comments?
> 
> /* These should really be defined in the kernel's
ptrace.h header.  */
> 
> We don't know what to define them _to_ on other
platforms.  If you
> have another platform which fails to define them in the
kernel
> headers, we can add another case.

At least for uClinux, ptrace() is hardwired to the same
values as in
linux-low.c.  So probably the comment is right and the
defined should
be moved into uClinux's uc0-patch.  But I don't see why the
dependency
on uclibc define.

> (I know the __UCLIBC_HAS_MMU__ is out of date - I'm
going to fix that
> today)

Hmmm, what's the correct way to find out whether the cpu has
mmu without
uclibc?  How does uclibc find out?

On Wed, Mar 21, 2007 at 08:14:12AM +0100, Josef Wolf wrote:
> At least for uClinux, ptrace() is hardwired to the same
values as in
> linux-low.c.

For all architectures, really?  I'm surprised.

> So probably the comment is right and the defined
should
> be moved into uClinux's uc0-patch.  But I don't see why
the dependency
> on uclibc define.

uClibc or the kernel.  It's not GDB's business to keep track
of that.

> > (I know the __UCLIBC_HAS_MMU__ is out of date -
I'm going to fix that
> > today)
> 
> Hmmm, what's the correct way to find out whether the
cpu has mmu without
> uclibc?  How does uclibc find out?

You configure it   But
they've renamed the configuration macro
recently.

-- 
Daniel Jacobowitz
CodeSourcery

On Wed, Mar 21, 2007 at 07:16:57AM -0400, Daniel Jacobowitz
wrote:
> On Wed, Mar 21, 2007 at 08:14:12AM +0100, Josef Wolf
wrote:

> > At least for uClinux, ptrace() is hardwired to the
same values as in
> > linux-low.c.
> 
> For all architectures, really?  I'm surprised.

I was confused, so don't be surprised   Every
architecture has its
own implementation.

> > So probably the comment is right and the defined
should
> > be moved into uClinux's uc0-patch.  But I don't
see why the dependency
> > on uclibc define.
> 
> uClibc or the kernel.  It's not GDB's business to keep
track of that.

I don't see how uclibc can provide them without kernel
support.

> > > (I know the __UCLIBC_HAS_MMU__ is out of date
- I'm going to fix that
> > > today)
> > 
> > Hmmm, what's the correct way to find out whether
the cpu has mmu without
> > uclibc?  How does uclibc find out?
> 
> You configure it   But
they've renamed the configuration macro
> recently.

Shouldn't the kernel headers provide some appropriate macro?
 There seem
to exist a CONFIG_MMU macro, but this is only checked and
never set.

On Wed, Mar 21, 2007 at 06:45:09PM +0100, Josef Wolf wrote:
> Shouldn't the kernel headers provide some appropriate
macro?  There seem
> to exist a CONFIG_MMU macro, but this is only checked
and never set.

I don't know, but none of the kernel's config macros should
generally
be used outside the kernel - uClibc seems as good a place
for it as any.

-- 
Daniel Jacobowitz
CodeSourcery