Re: Regression: graceful unwind termination when we'd need unavailable/uncollect memory or registers to unwind further

[Pedro Alves <pedro@codesourcery.com>]

On Sunday 20 March 2011 18:48:45, Jan Kratochvil wrote:
> Hi Pedro,
> 
> on a testfile from the Fedora patchset there is a regression, IMO a valid one:

Agreed.

> http://pkgs.fedoraproject.org/gitweb/?p=gdb.git;a=blob_plain;f=gdb-6.8-bz457187-largefile-test.patch;hb=f15/master
>  = This file is included in this mail.

Thanks!

> The core file does not have accessible registers.  It was not a goal of the
> testcase, it tests something else.  But it has found this unrelated
> regression.  

Yeah, nice test.

> (gdb) x/i 0x400078
>    0x400078:    hlt    
> PASS: gdb.arch/i386-biarch-core.exp: .text is readable
> ->
> (gdb) x/i 0x400078
> PC not available
> FAIL: gdb.arch/i386-biarch-core.exp: .text is readable

> I believe one should be able to evaluate PC-indepenent
> expressions even if PC is not available.

Yep, and you can.  The quirk is x/i / disassemble specific.  The
error is thrown while trying to decide whether to print the "==>" that
indicates the current PC.

> Even the first part is regressing (although it stays PASS->PASS):
> (gdb) core-file gdb.arch/i386-biarch-core.core
> [New LWP 6901]
> warning: Couldn't recognize general-purpose registers in core file.
> Core was generated by `./bad'.
> Program terminated with signal 11, Segmentation fault.
> warning: Couldn't recognize general-purpose registers in core file.
> #0  0x00000000 in ?? ()
> ->
> (gdb) core-file gdb.arch/i386-biarch-core.core
> [New LWP 6901]
> warning: Couldn't recognize general-purpose registers in core file.
> PC register is not available

Fixed as well.  After the patch:

(...)
warning: Couldn't recognize general-purpose registers in core file.
Core was generated by `./bad'.
Program terminated with signal 11, Segmentation fault.

warning: Couldn't recognize general-purpose registers in core file.
#0  <unavailable> in ?? ()
(gdb) 

and:

(gdb)  x/i 0x400078
   0x400078:    hlt  

Below's the patch.  Doesn't cause regressions for me.
Did you see any other regression?

Thanks,

Pedro Alves

2011-03-27  Pedro Alves  <pedro@codesourcery.com>

	* infcmd.c (post_create_inferior): Ignore NOT_AVAILABLE_ERROR
	errors when reading the `stop_pc'.
	* printcmd.c (pc_prefix): Use get_frame_pc_if_available instead of
	get_frame_pc.

---
 gdb/infcmd.c   |   14 ++++++++++++--
 gdb/printcmd.c |    4 +---
 2 files changed, 13 insertions(+), 5 deletions(-)

Index: src/gdb/infcmd.c
===================================================================
--- src.orig/gdb/infcmd.c	2011-03-15 20:15:05.000000000 +0000
+++ src/gdb/infcmd.c	2011-03-22 10:22:17.729088002 +0000
@@ -395,6 +395,8 @@ strip_bg_char (char **args)
 void
 post_create_inferior (struct target_ops *target, int from_tty)
 {
+  volatile struct gdb_exception ex;
+
   /* Be sure we own the terminal in case write operations are performed.  */ 
   target_terminal_ours ();
 
@@ -404,8 +406,16 @@ post_create_inferior (struct target_ops
      don't need to.  */
   target_find_description ();
 
-  /* Now that we know the register layout, retrieve current PC.  */
-  stop_pc = regcache_read_pc (get_current_regcache ());
+  /* Now that we know the register layout, retrieve current PC.  But
+     if the PC is unavailable (e.g., we're opening a core file with
+     missing registers info), ignore it.  */
+  stop_pc = 0;
+  TRY_CATCH (ex, RETURN_MASK_ERROR)
+    {
+      stop_pc = regcache_read_pc (get_current_regcache ());
+    }
+  if (ex.reason < 0 && ex.error != NOT_AVAILABLE_ERROR)
+    throw_exception (ex);
 
   if (exec_bfd)
     {
Index: src/gdb/printcmd.c
===================================================================
--- src.orig/gdb/printcmd.c	2011-03-15 20:15:34.000000000 +0000
+++ src/gdb/printcmd.c	2011-03-22 10:21:41.829088002 +0000
@@ -759,9 +759,7 @@ pc_prefix (CORE_ADDR addr)
       CORE_ADDR pc;
 
       frame = get_selected_frame (NULL);
-      pc = get_frame_pc (frame);
-
-      if (pc == addr)
+      if (get_frame_pc_if_available (frame, &pc) && pc == addr)
 	return "=> ";
     }
   return "   ";