[patch] Fix a reread_symbols regression by mmap [Re: [patch] Use mmap instead of obstack_alloc for dwarf debug sections.]

[Jan Kratochvil <jan.kratochvil@redhat.com>]

Hi Paul,

while sure your patch is great it has a regression.  Assuming a regression
reproducer needs to link GDB with -lmcheck (but did verify it).

(I do not try to advocate reread_symbols but its code is there now.)


Thanks,
Jan


This patch will fix back (tested on {x86_64,i686}-fedora-linux-gnu):

-FAIL: gdb.ada/exec_changed.exp: start second
-dwarf2read.c:10943: internal-error: munmap_section_buffer: Assertion `munmap ((void *) map_begin, map_length) == 0' failed.^M
+PASS: gdb.ada/exec_changed.exp: start second

-FAIL: gdb.base/chng-syms.exp: (timeout) running with invalidated bpt condition after executable changes
-dwarf2read.c:10943: internal-error: munmap_section_buffer: Assertion `munmap ((void *) map_begin, map_length) == 0' failed.^M
+PASS: gdb.base/chng-syms.exp: running with invalidated bpt condition after executable changes

-FAIL: gdb.base/reread.exp: run to foo() second time (timeout)
-dwarf2read.c:10943: internal-error: munmap_section_buffer: Assertion `munmap ((void *) map_begin, map_length) == 0' failed.^M
-FAIL: gdb.base/reread.exp: shell mv /home/jkratoch/hammock/20090623build-gdbcvs/fedora-11-i386/build/src/gdb/testsuite.unix.-m32/gdb.base/reread /home/jkratoch/hammock/20090623build-gdbcvs/fedora-11-i386/build/src/gdb/testsuite.unix.-m32/gdb.base/reread2 (GDB internal error)
-ERROR: Could not resync from internal error (timeout)
-[more mess]
+PASS: gdb.base/reread.exp: run to foo() second time
+PASS: gdb.base/reread.exp: second pass: breakpoint foo in first file
+PASS: gdb.base/reread.exp: second pass: run to foo()
+PASS: gdb.base/reread.exp: second pass: continue to completion
+PASS: gdb.base/reread.exp: second pass: run to foo() second time
 
munmap(0x9595959595959000, 10778685752873425706) = -1 EINVAL (Invalid argument)
write(1, "dwarf2read.c:10943: internal-error: munmap_section_buffer: Assertion `munmap ((void *) map_begin, map_length) == 0' failed.\nA problem internal to GDB has been detected,\nfurther debugging may prove unre"..., 209) = 209


2009-06-23  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Fix a regression by the mmap patch from 2009-06-16.
	* symfile.c (reread_symbols): Move sym_finish and clear_objfile_data
	calls before deleting the OBJFILE obstack.  Extend the comment.

--- gdb/symfile.c	17 Jun 2009 18:34:34 -0000	1.232
+++ gdb/symfile.c	23 Jun 2009 14:56:34 -0000
@@ -2334,7 +2334,16 @@ reread_symbols (void)
 
 	      /* Nuke all the state that we will re-read.  Much of the following
 	         code which sets things to NULL really is necessary to tell
-	         other parts of GDB that there is nothing currently there.  */
+	         other parts of GDB that there is nothing currently there.
+		 
+		 Try to keep the freeing order compatible with free_objfile.  */
+
+	      if (objfile->sf != NULL)
+		{
+		  (*objfile->sf->sym_finish) (objfile);
+		}
+
+	      clear_objfile_data (objfile);
 
 	      /* FIXME: Do we have to free a whole linked list, or is this
 	         enough?  */
@@ -2371,11 +2380,6 @@ reread_symbols (void)
 		      sizeof (objfile->msymbol_hash));
 	      memset (&objfile->msymbol_demangled_hash, 0,
 		      sizeof (objfile->msymbol_demangled_hash));
-	      clear_objfile_data (objfile);
-	      if (objfile->sf != NULL)
-		{
-		  (*objfile->sf->sym_finish) (objfile);
-		}
 
 	      objfile->psymbol_cache = bcache_xmalloc ();
 	      objfile->macro_cache = bcache_xmalloc ();