Overlays and symbol lookup query

Overlays and symbol lookup query

[Andrew Burgess]

[-- Attachment #1: Type: text/plain, Size: 9302 bytes --]

Hi,

I ran into an issue relating to overlays and would be grateful for any 
advice on the best way to fix the problem.
There's a demo attached, untar, then make && make test, but the 
interesting bit is the linker script:

SECTIONS
{
   .text : { test.o(.text) }

   OVERLAY :
   {
     .overlay_a { file_a.o(.text) }
     .overlay_b { file_b.o(.text) }
   }
}

file_a contains func_a, and file_b contains func_b.

I then start up gdb on the executable and do:

(gdb) info line func_a
Line 2 of "file_a.c" starts at address 0xc <func_a> and ends at 0xf 
<func_a+3>.
(gdb) info line func_b
Line 2 of "file_a.c" starts at address 0xc <func_a> and ends at 0xf 
<func_a+3>.

So gdb is clearly confused by my overlays as it thinks both symbols are 
in file_a.
I know that debugging with overlays is special, but it feels like gdb 
should give me a sensible answer in this case - don't know if you'll 
agree with this though?

The problem is caused in find_function_start_sal(symtab.c), up to that 
point we have a symbol for which we're getting the information, from the 
symbol we have a symtab. We then discard the symbol/symtab and call 
find_pc_sect_line which is forced to lookup the symtab based on the 
address, it seems reasonable that this gives me the wrong answer as it 
doesn't know which overlay is mapped in at that point.

The patch below is a _rough_ cut at solving this issue by extending 
find_pc_sect_line to take a symtab* and only look one up by address if 
the passed in symtab* is NULL, the rest of the changes are to fix call 
sites.

This doesn't completely fix the "info line" case, the printing of the 
start/end addresses is still wrong, but right now I'm just after any 
feedback on the current set of changes, or if anyone thinks there's a 
better way to fix this problem.


Thanks,

Andrew


diff -ur clean/src/gdb/block.c working/src/gdb/block.c
--- clean/src/gdb/block.c	2010-01-01 07:31:30.000000000 +0000
+++ working/src/gdb/block.c	2010-12-07 13:53:10.799284000 +0000
@@ -174,12 +174,12 @@
     in the specified section, or 0 if there is none.  */

  struct block *
-block_for_pc_sect (CORE_ADDR pc, struct obj_section *section)
+block_for_pc_sect (CORE_ADDR pc, struct symtab *symtab, struct 
obj_section *section)
  {
    struct blockvector *bl;
    struct block *b;

-  bl = blockvector_for_pc_sect (pc, section, &b, NULL);
+  bl = blockvector_for_pc_sect (pc, section, &b, symtab);
    if (bl)
      return b;
    return 0;
@@ -191,7 +191,7 @@
  struct block *
  block_for_pc (CORE_ADDR pc)
  {
-  return block_for_pc_sect (pc, find_pc_mapped_section (pc));
+  return block_for_pc_sect (pc, NULL, find_pc_mapped_section (pc));
  }

  /* Now come some functions designed to deal with C++ namespace issues.
diff -ur clean/src/gdb/block.h working/src/gdb/block.h
--- clean/src/gdb/block.h	2010-01-01 07:31:30.000000000 +0000
+++ working/src/gdb/block.h	2010-12-07 13:53:10.811286000 +0000
@@ -147,7 +147,7 @@

  extern struct block *block_for_pc (CORE_ADDR);

-extern struct block *block_for_pc_sect (CORE_ADDR, struct obj_section *);
+extern struct block * block_for_pc_sect (CORE_ADDR, struct symtab *, 
struct obj_section *);

  extern const char *block_scope (const struct block *block);

diff -ur clean/src/gdb/blockframe.c working/src/gdb/blockframe.c
--- clean/src/gdb/blockframe.c	2010-05-13 23:44:02.000000000 +0100
+++ working/src/gdb/blockframe.c	2010-12-07 13:53:51.042763000 +0000
@@ -136,7 +136,7 @@
  struct symbol *
  find_pc_sect_function (CORE_ADDR pc, struct obj_section *section)
  {
-  struct block *b = block_for_pc_sect (pc, section);
+  struct block *b = block_for_pc_sect (pc, NULL, section);

    if (b == 0)
      return 0;
diff -ur clean/src/gdb/linespec.c working/src/gdb/linespec.c
--- clean/src/gdb/linespec.c	2010-07-13 21:07:44.000000000 +0100
+++ working/src/gdb/linespec.c	2010-12-07 14:47:45.783167000 +0000
@@ -1990,6 +1990,7 @@
    values.sals = (struct symtab_and_line *)
      xmalloc (sizeof (struct symtab_and_line));
    values.sals[0] = find_pc_sect_line (SYMBOL_VALUE_ADDRESS (msymbol),
+                                      (struct symtab*) 0,
  				      (struct obj_section *) 0, 0);
    values.sals[0].section = SYMBOL_OBJ_SECTION (msymbol);

@@ -1999,7 +2000,7 @@
                                             values.sals[0].pc,
                                             &current_target);
    if (pc != values.sals[0].pc)
-    values.sals[0] = find_pc_sect_line (pc, NULL, 0);
+    values.sals[0] = find_pc_sect_line (pc, NULL, NULL, 0);

    if (funfirstline)
      skip_prologue_sal (&values.sals[0]);
diff -ur clean/src/gdb/printcmd.c working/src/gdb/printcmd.c
--- clean/src/gdb/printcmd.c	2010-10-15 19:54:12.000000000 +0100
+++ working/src/gdb/printcmd.c	2010-12-07 13:53:10.969287000 +0000
@@ -717,7 +717,7 @@
      {
        struct symtab_and_line sal;

-      sal = find_pc_sect_line (addr, section, 0);
+      sal = find_pc_sect_line (addr, (struct symtab*)NULL, section, 0);

        if (sal.symtab)
  	{
diff -ur clean/src/gdb/symtab.c working/src/gdb/symtab.c
--- clean/src/gdb/symtab.c	2010-10-17 19:49:46.000000000 +0100
+++ working/src/gdb/symtab.c	2010-12-07 14:02:52.237975000 +0000
@@ -1890,7 +1890,7 @@
  /* If it's worth the effort, we could be using a binary search.  */

  struct symtab_and_line
-find_pc_sect_line (CORE_ADDR pc, struct obj_section *section, int 
notcurrent)
+find_pc_sect_line (CORE_ADDR pc, struct symtab *symtab, struct 
obj_section *section, int notcurrent)
  {
    struct symtab *s;
    struct linetable *l;
@@ -2011,8 +2011,11 @@
  	  return find_pc_line (SYMBOL_VALUE_ADDRESS (mfunsym), 0);
        }

+  if ( symtab == NULL )
+    s = find_pc_sect_symtab (pc, section);
+  else
+    s = symtab;

-  s = find_pc_sect_symtab (pc, section);
    if (!s)
      {
        /* if no symbol information, return previous pc */
@@ -2134,7 +2137,7 @@
    section = find_pc_overlay (pc);
    if (pc_in_unmapped_range (pc, section))
      pc = overlay_mapped_address (pc, section);
-  return find_pc_sect_line (pc, section, notcurrent);
+  return find_pc_sect_line (pc, NULL, section, notcurrent);
  }
  \f
  /* Find line number LINE in any symtab whose name is the same as
@@ -2289,7 +2292,7 @@
       This also insures that we never give a range like "starts at 0x134
       and ends at 0x12c".  */

-  found_sal = find_pc_sect_line (startaddr, sal.section, 0);
+  found_sal = find_pc_sect_line (startaddr, sal.symtab, sal.section, 0);
    if (found_sal.line != sal.line)
      {
        /* The specified line (sal) has zero bytes.  */
@@ -2416,6 +2419,7 @@

    fixup_symbol_section (sym, NULL);
    sal = find_pc_sect_line (BLOCK_START (SYMBOL_BLOCK_VALUE (sym)),
+                           SYMBOL_SYMTAB (sym),
  			   SYMBOL_OBJ_SECTION (sym), 0);

    /* We always should have a line for the function start address.
@@ -2502,7 +2506,7 @@
    pc = overlay_mapped_address (pc, section);

    /* Calculate line number.  */
-  start_sal = find_pc_sect_line (pc, section, 0);
+  start_sal = find_pc_sect_line (pc, NULL, section, 0);

    /* Check if gdbarch_skip_prologue left us in mid-line, and the next
       line is still part of the same function.  */
@@ -2515,7 +2519,7 @@
        /* First pc of next line */
        pc = start_sal.end;
        /* Recalculate the line number (might not be N+1).  */
-      start_sal = find_pc_sect_line (pc, section, 0);
+      start_sal = find_pc_sect_line (pc, NULL, section, 0);
      }

    /* On targets with executable formats that don't have a concept of
@@ -2527,7 +2531,7 @@
      {
        pc = gdbarch_skip_main_prologue (gdbarch, pc);
        /* Recalculate the line number (might not be N+1).  */
-      start_sal = find_pc_sect_line (pc, section, 0);
+      start_sal = find_pc_sect_line (pc, NULL, section, 0);
      }

    /* If we still don't have a valid source line, try to find the first
@@ -2542,7 +2546,7 @@
      {
        pc = skip_prologue_using_lineinfo (pc, SYMBOL_SYMTAB (sym));
        /* Recalculate the line number.  */
-      start_sal = find_pc_sect_line (pc, section, 0);
+      start_sal = find_pc_sect_line (pc, NULL, section, 0);
      }

    do_cleanups (old_chain);
@@ -2566,7 +2570,7 @@

    /* Check if we are now inside an inlined function.  If we can,
       use the call site of the function instead.  */
-  b = block_for_pc_sect (sal->pc, sal->section);
+  b = block_for_pc_sect (sal->pc, sal->symtab, sal->section);
    function_block = NULL;
    while (b != NULL)
      {
@@ -4643,7 +4647,7 @@
        set_current_program_space (ret.sals[i].pspace);

        filter[i] = 1;
-      blocks[i] = block_for_pc_sect (ret.sals[i].pc, ret.sals[i].section);
+      blocks[i] = block_for_pc_sect (ret.sals[i].pc, 
ret.sals[i].symtab, ret.sals[i].section);

      }
    do_cleanups (old_chain);
diff -ur clean/src/gdb/symtab.h working/src/gdb/symtab.h
--- clean/src/gdb/symtab.h	2010-10-17 19:49:47.000000000 +0100
+++ working/src/gdb/symtab.h	2010-12-07 13:53:11.099286000 +0000
@@ -1106,6 +1106,7 @@
  /* Same function, but specify a section as well as an address */

  extern struct symtab_and_line find_pc_sect_line (CORE_ADDR,
+                                                 struct symtab *,
  						 struct obj_section *, int);

  /* Given a symtab and line number, return the pc there.  */




[-- Attachment #2: overlay-test.tar.bz2 --]
[-- Type: application/octet-stream, Size: 1239 bytes --]

Re: Overlays and symbol lookup query

[Petr Hluzín]

Hello

I am probably the wrong guy to answer that (I do not know how GDB
internals work here, nor have used overlays before).

You did not mention if manual overlay commands [1] were of any use.
I understand that GDB requires user to set overlays manually or to
have special support in the target [2]. The special support in the
target can be done by an overlaying library.
What does "overlay list" print?

I wonder if GDB is able to detect that a target uses overlays and
automatic overlay support is not implemented in the target.

[1] http://sourceware.org/gdb/current/onlinedocs/gdb/Overlay-Commands.html
[2] http://sourceware.org/gdb/current/onlinedocs/gdb/Automatic-Overlay-Debugging.html

-- 
Petr Hluzin