[rfc] Do not read from the executable if ptrace fails

[rfc] Do not read from the executable if ptrace fails

[Daniel Jacobowitz]

The old memory read methods used to have a target_has_all_memory
check.  It was broken at some point, probably around the introduction
of target_xfer_partial, and when I redid all the memory interfaces
last year I didn't try to fix it.  But there's a bug report about
this, which I've been meaning to come back to.

The attached fixes it.  If we try to read from a ptrace target, and
it fails, we do not continue down the target stack.  This won't affect
core debugging.

Any comments, or shall I apply this?

-- 
Daniel Jacobowitz
CodeSourcery

2007-06-24  Daniel Jacobowitz  <dan@codesourcery.com>

	PR symtab/2161
	* target.c (memory_xfer_partial): Do not continue past targets with
	all memory.

Index: target.c
===================================================================
RCS file: /cvs/src/src/gdb/target.c,v
retrieving revision 1.143
diff -u -p -r1.143 target.c
--- target.c	9 Jun 2007 13:42:16 -0000	1.143
+++ target.c	25 Jun 2007 00:29:17 -0000
@@ -1084,6 +1084,11 @@ memory_xfer_partial (struct target_ops *
       if (res > 0)
 	return res;
 
+      /* We want to continue past core files to executables, but not
+	 past a running target's memory.  */
+      if (ops->to_has_all_memory)
+	return res;
+
       ops = ops->beneath;
     }
   while (ops != NULL);

Re: [rfc] Do not read from the executable if ptrace fails

[Joel Brobecker]

> Any comments, or shall I apply this?
> 
> -- 
> Daniel Jacobowitz
> CodeSourcery
> 
> 2007-06-24  Daniel Jacobowitz  <dan@codesourcery.com>
> 
> 	PR symtab/2161
> 	* target.c (memory_xfer_partial): Do not continue past targets with
> 	all memory.

FWIW, it looks pretty good to me, and I think it should be part of 6.7
too.

-- 
Joel

Re: [rfc] Do not read from the executable if ptrace fails

[Daniel Jacobowitz]

On Mon, Jun 25, 2007 at 07:54:03PM -0400, Joel Brobecker wrote:
> > Any comments, or shall I apply this?
> > 
> > -- 
> > Daniel Jacobowitz
> > CodeSourcery
> > 
> > 2007-06-24  Daniel Jacobowitz  <dan@codesourcery.com>
> > 
> > 	PR symtab/2161
> > 	* target.c (memory_xfer_partial): Do not continue past targets with
> > 	all memory.
> 
> FWIW, it looks pretty good to me, and I think it should be part of 6.7
> too.

OK, I checked it in.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [rfc] Do not read from the executable if ptrace fails

[Ulrich Weigand]

Dan Jacobowitz wrote:
> On Mon, Jun 25, 2007 at 07:54:03PM -0400, Joel Brobecker wrote:
> > > Any comments, or shall I apply this?
> > > 
> > > -- 
> > > Daniel Jacobowitz
> > > CodeSourcery
> > > 
> > > 2007-06-24  Daniel Jacobowitz  <dan@codesourcery.com>
> > > 
> > > 	PR symtab/2161
> > > 	* target.c (memory_xfer_partial): Do not continue past targets with
> > > 	all memory.
> > 
> > FWIW, it looks pretty good to me, and I think it should be part of 6.7
> > too.
> 
> OK, I checked it in.

I only noticed now, but this breaks SPU overlay support.

Overlay support uses "LMA" addresses to refer to overlay sections
not currently loaded, and performs xfer_partial requests to read
from those (e.g. when invoking skip_prologue on a function not
currently present).

This works only if accesses to those LMA addresses go through to
the xfer_memory routine in exec.c, which implements the required
logic to retrieve those section contents from the executable file.

This used to work (probably accidentally?), but is broken by
this patch.  Any suggestions how to fix it?  Should the overlay
logic from xfer_memory be moved to memory_xfer_partial, maybe?

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [rfc] Do not read from the executable if ptrace fails

[Daniel Jacobowitz]

On Mon, Jul 23, 2007 at 09:02:33PM +0200, Ulrich Weigand wrote:
> I only noticed now, but this breaks SPU overlay support.

:-(

> Overlay support uses "LMA" addresses to refer to overlay sections
> not currently loaded, and performs xfer_partial requests to read
> from those (e.g. when invoking skip_prologue on a function not
> currently present).
> 
> This works only if accesses to those LMA addresses go through to
> the xfer_memory routine in exec.c, which implements the required
> logic to retrieve those section contents from the executable file.
> 
> This used to work (probably accidentally?), but is broken by
> this patch.  Any suggestions how to fix it?  Should the overlay
> logic from xfer_memory be moved to memory_xfer_partial, maybe?

That sounds reasonable.  What about if we check for an unmapped
overlay address at the same point we handle trust_readonly?

-- 
Daniel Jacobowitz
CodeSourcery

Re: [rfc] Do not read from the executable if ptrace fails

[Ulrich Weigand]

Daniel Jacobowitz wrote:

> > This used to work (probably accidentally?), but is broken by
> > this patch.  Any suggestions how to fix it?  Should the overlay
> > logic from xfer_memory be moved to memory_xfer_partial, maybe?
> 
> That sounds reasonable.  What about if we check for an unmapped
> overlay address at the same point we handle trust_readonly?

Like so?  Tested on spu-elf, fixes the overlay test case.

Bye,
Ulrich


ChangeLog:

	* target.c (memory_xfer_partial): Accesses to unmapped overlay
	sections should always go to the executable file.

Index: gdb/target.c
===================================================================
RCS file: /cvs/src/src/gdb/target.c,v
retrieving revision 1.144
diff -c -p -r1.144 target.c
*** gdb/target.c	1 Jul 2007 22:35:55 -0000	1.144
--- gdb/target.c	23 Jul 2007 20:39:33 -0000
*************** memory_xfer_partial (struct target_ops *
*** 1017,1022 ****
--- 1017,1030 ----
  	return xfer_memory (memaddr, readbuf, len, 0, NULL, ops);
      }
  
+   /* Likewise for accesses to unmapped overlay sections.  */
+   if (readbuf != NULL && overlay_debugging)
+     {
+       asection *section = find_pc_overlay (memaddr);
+       if (pc_in_unmapped_range (memaddr, section))
+ 	return xfer_memory (memaddr, readbuf, len, 0, NULL, ops);
+     }
+ 
    /* Try GDB's internal data cache.  */
    region = lookup_mem_region (memaddr);
    /* region->hi == 0 means there's no upper bound.  */


-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com

Re: [rfc] Do not read from the executable if ptrace fails

[Daniel Jacobowitz]

On Mon, Jul 23, 2007 at 11:24:56PM +0200, Ulrich Weigand wrote:
> 	* target.c (memory_xfer_partial): Accesses to unmapped overlay
> 	sections should always go to the executable file.

Looks fine to me.  I can't see any cases where this would be wrong.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [rfc] Do not read from the executable if ptrace fails

[Ulrich Weigand]

Daniel Jacobowitz wrote:
> On Mon, Jul 23, 2007 at 11:24:56PM +0200, Ulrich Weigand wrote:
> > 	* target.c (memory_xfer_partial): Accesses to unmapped overlay
> > 	sections should always go to the executable file.
> 
> Looks fine to me.  I can't see any cases where this would be wrong.

I've checked this in now, thanks.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com