Re: [RFA/RFC] mips tracepoint: fix Bug 12013

[Pedro Alves <pedro@codesourcery.com>]

On Tuesday 28 December 2010 17:23:59, Pedro Alves wrote:
> On Tuesday 28 December 2010 17:08:52, Joel Brobecker wrote:
> > > Joel,  do you think this patch can check in to 7.2.1.
> > 
> > I think that it's borderline, and I'm tempted to say no, considering
> > that this is not a crash or a regression (just a limitation).  But
> > it looks relatively safe to me.  So, if Pedro agrees, it's OK with me.
> 
> Yeah.  I understand that without this patch, MIPS tracepoints are
> practically useless on 7.2, so on those grounds, I'd be okay.
> 
> However, I just went to check whether collecting a pseudo or
> user register in x86/x86_64 still gives a reasonable error, and
> I now get:
> 
>  >./gdb -q ./gdb
>  Reading symbols from /home/pedro/gdb/baseline/build/gdb/gdb...done.
>  (top-gdb) start
>  Temporary breakpoint 3 at 0x4565c3: file ../../src/gdb/gdb.c, line 29.
>  Starting program: /home/pedro/gdb/baseline/build/gdb/gdb 
>  [Thread debugging using libthread_db enabled]
> 
>  Temporary breakpoint 3, main (argc=1, argv=0x7fffffffe108) at ../../src/gdb/gdb.c:29
>  29        memset (&args, 0, sizeof args);
>  (top-gdb) maint agent $sp
>  ../../src/gdb/regcache.c:166: internal-error: register_type: Assertion `regnum >= 0 && regnum < descr->nr_cooked_registers' failed.
>  A problem internal to GDB has been detected,
>  further debugging may prove unreliable.
>  Quit this debugging session? (y or n) 
> 
> While before (on 7.2), we'd get:
> 
>  (top-gdb) maint agent $sp
>  'sp' is a pseudo-register; GDB cannot yet trace pseudoregister contents.
> 
> We'll need to get this fixed this before considering a backport to 7.2.
> 
> 

Fixed for now.  I've reinstated the exact same old error, but under
a stricter check: we now error out for user registers, but let
pseudo-registers pass:

 (top-gdb) maint agent $pc
 'pc' is a pseudo-register; GDB cannot yet trace pseudoregister contents.

Pseudo-registers will error out further on, on the new errors that Hui
added when the new gdbarch callbacks aren't implemented.  On x86_64:

 (top-gdb) maint agent $rsp
 Scope: 0x4565c3
 Reg mask: 80
   0  end

 (top-gdb) maint agent $esp
 'esp' is a pseudo-register; GDB cannot yet trace its contents.

-- 
Pedro Alves

2010-12-28  Pedro Alves  <pedro@codesourcery.com>

	gdb/
	* ax-gdb.c (gen_expr) <OP_REGISTER>: Error out if trying to
	collect a user register.
	
---
 gdb/ax-gdb.c |    6 ++++++
 1 file changed, 6 insertions(+)

Index: src/gdb/ax-gdb.c
===================================================================
--- src.orig/gdb/ax-gdb.c	2010-12-28 16:22:01.000000000 +0000
+++ src/gdb/ax-gdb.c	2010-12-28 17:54:45.000000000 +0000
@@ -1978,6 +1978,12 @@ gen_expr (struct expression *exp, union 
 	if (reg == -1)
 	  internal_error (__FILE__, __LINE__,
 			  _("Register $%s not available"), name);
+	/* No support for tracing user registers yet.  */
+	if (reg >= gdbarch_num_regs (exp->gdbarch)
+	    + gdbarch_num_pseudo_regs (exp->gdbarch))
+	  error (_("'%s' is a pseudo-register; "
+		   "GDB cannot yet trace pseudoregister contents."),
+		 name);
 	value->kind = axs_lvalue_register;
 	value->u.reg = reg;
 	value->type = register_type (exp->gdbarch, reg);