This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Accessing tls variables across files causes a bug


On Mon, 01 Dec 2008 19:29:55 +0100, Ulrich Weigand wrote:

Followed your advices, updated the patch.


Thanks,
Jan

------------------------------------------------------------------------------

> Jan Kratochvil wrote:
> 
> > +	/* SYMBOL_VALUE_ADDRESS should return the translated address.  */
> 
> I'm not sure I understand this comment -- the translated address of
> a thread-local variable obviously depends on the thread in which it
> is evaluated, so how should a single location like SYMBOL_VALUE_ADDRESS
> be able to hold that value for all threads?

SYMBOL_VALUE_ADDRESS is used many times across GDB code and I assumed in most
of such cases it would be more appropriate to return the absolute address
valid for current inferior_ptid.

But I now briefly checked all these points and found out in fact all of them
work with SYMBOL_VALUE_ADDRESS referring to function symbols - PC values (just
except value_static_field() but that is also not in use with TLS symbols).
Sure PC values are never TLS-translated.


> > +	    /* SYMBOL_VALUE_ADDRESS should return the translated address.  */
> > +	    if (section
> > +		&& (section->the_bfd_section->flags & SEC_THREAD_LOCAL) != 0)
> > +	      {
> > +		printf_filtered (_("a thread-local variable at offset %s "
> > +				   "at final address "), paddr_nz (load_addr));
> > +		load_addr = target_translate_tls_address (section->objfile,
> > +							  load_addr);
> > +	      }
> 
> Again, I think "at final address" may be misleading; if we give an absolute
> address, it should explicitly mention it is relative to the current thread.

`final' was put there as it printed before:

Symbol "X" is a thread-local variable at offset 0x1234 at address 0xabcd0000.

and I found it ambiguous whether the bytes of the variable X are in fact
placed at (a) address 0xabcd0000 or (b) address 0xabcd1234.


> In the LOC_COMPUTED case, we'd output something like "at offset ...
> in the thread-local storage for ...".  For consistency reasons, I'd
> prefer to have the same output in the LOC_UNDEFINED case as well.

While I copied the message from the LOC_COMPUTED case I agree it should be
consistent and if I wish the absolute address there it should be printed even
in the LOC_COMPUTED case which would be more a scope of a different patch.


> On the other hand, I'm wondering if we should perform the resolution here
> at all; isn't "info address" also allowed when the target is not actually
> running, so we don't even have a current thread?

While I forgot about this case I find the current behavior appropriate:

(gdb) info address a_thread_local
Symbol "a_thread_local" is a thread-local variable at offset 0 at final address Cannot find thread-local variables on this target

But it is no longer printed due to the consistency advice above.


On Mon, 01 Dec 2008 19:29:31 +0100, Gary Funck wrote:
> --- gdb/printcmd.c      20 Nov 2008 16:13:11 -0000      1.138
> +++ gdb/printcmd.c      30 Nov 2008 21:10:06 -0000
> @@ -1234,6 +1234,7 @@ address_info (char *exp, int from_tty)
> +       struct obj_section *obj_section;
> Is obj_section referenced?

Removed.
2008-12-01  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Fix resolving external references to TLS variables.
	* findvar.c: Include `objfiles.h'.
	(read_var_value <LOC_UNRESOLVED>): New variable `obj_section'.  Handle
	SEC_THREAD_LOCAL variables.
	* printcmd.c (address_info <LOC_UNRESOLVED>): Handle SEC_THREAD_LOCAL
	variables.

2008-12-01  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Test resolving external references to TLS variables.
	* gdb.threads/tls.exp: New tests to examine A_THREAD_LOCAL and
	FILE2_THREAD_LOCAL.
	(testfile2, srcfile2): New variables.
	* gdb.threads/tls.c (file2_thread_local)
	(function_referencing_file2_thread_local): New.
	* gdb.threads/tls2.c: New file.

--- gdb/findvar.c	5 Sep 2008 11:37:17 -0000	1.118
+++ gdb/findvar.c	1 Dec 2008 21:50:25 -0000
@@ -34,6 +34,7 @@
 #include "regcache.h"
 #include "user-regs.h"
 #include "block.h"
+#include "objfiles.h"
 
 /* Basic byte-swapping routines.  GDB has needed these for a long time...
    All extract a target-format integer at ADDR which is LEN bytes long.  */
@@ -536,6 +537,7 @@ read_var_value (struct symbol *var, stru
     case LOC_UNRESOLVED:
       {
 	struct minimal_symbol *msym;
+	struct obj_section *obj_section;
 
 	msym = lookup_minimal_symbol (SYMBOL_LINKAGE_NAME (var), NULL, NULL);
 	if (msym == NULL)
@@ -545,6 +547,11 @@ read_var_value (struct symbol *var, stru
 					   SYMBOL_OBJ_SECTION (msym));
 	else
 	  addr = SYMBOL_VALUE_ADDRESS (msym);
+
+	obj_section = SYMBOL_OBJ_SECTION (msym);
+	if (obj_section
+	    && (obj_section->the_bfd_section->flags & SEC_THREAD_LOCAL) != 0)
+	  addr = target_translate_tls_address (obj_section->objfile, addr);
       }
       break;
 
--- gdb/printcmd.c	20 Nov 2008 16:13:11 -0000	1.138
+++ gdb/printcmd.c	1 Dec 2008 21:50:28 -0000
@@ -1241,16 +1241,25 @@ address_info (char *exp, int from_tty)
 	else
 	  {
 	    section = SYMBOL_OBJ_SECTION (msym);
-	    printf_filtered (_("static storage at address "));
 	    load_addr = SYMBOL_VALUE_ADDRESS (msym);
-	    fputs_filtered (paddress (load_addr), gdb_stdout);
-	    if (section_is_overlay (section))
+
+	    if (section
+		&& (section->the_bfd_section->flags & SEC_THREAD_LOCAL) != 0)
+	      printf_filtered (_("a thread-local variable at offset %s "
+				 "in the thread-local storage for `%s'"),
+			       paddr_nz (load_addr), section->objfile->name);
+	    else
 	      {
-		load_addr = overlay_unmapped_address (load_addr, section);
-		printf_filtered (_(",\n -- loaded at "));
+		printf_filtered (_("static storage at address "));
 		fputs_filtered (paddress (load_addr), gdb_stdout);
-		printf_filtered (_(" in overlay section %s"),
-				 section->the_bfd_section->name);
+		if (section_is_overlay (section))
+		  {
+		    load_addr = overlay_unmapped_address (load_addr, section);
+		    printf_filtered (_(",\n -- loaded at "));
+		    fputs_filtered (paddress (load_addr), gdb_stdout);
+		    printf_filtered (_(" in overlay section %s"),
+				     section->the_bfd_section->name);
+		  }
 	      }
 	  }
       }
--- gdb/testsuite/gdb.threads/tls.c	29 Jul 2003 21:51:25 -0000	1.2
+++ gdb/testsuite/gdb.threads/tls.c	1 Dec 2008 21:50:30 -0000
@@ -20,6 +20,9 @@
 __thread int a_thread_local;
 __thread int another_thread_local;
 
+/* psymtabs->symtabs resolving check.  */
+extern __thread int file2_thread_local;
+
 /* Global variable just for info addr in gdb.  */
 int a_global;
 
@@ -119,6 +122,12 @@ void *spin( vp )
 }
 
 void
+function_referencing_file2_thread_local (void)
+{
+  file2_thread_local = file2_thread_local;
+}
+
+void
 do_pass()
 {
     int i;
--- gdb/testsuite/gdb.threads/tls.exp	6 Aug 2008 12:52:08 -0000	1.9
+++ gdb/testsuite/gdb.threads/tls.exp	1 Dec 2008 21:50:30 -0000
@@ -15,7 +15,9 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
 
 set testfile tls
+set testfile2 tls2
 set srcfile ${testfile}.c
+set srcfile2 ${testfile2}.c
 set binfile ${objdir}/${subdir}/${testfile}
 
 if [istarget "*-*-linux"] then {
@@ -24,7 +26,7 @@ if [istarget "*-*-linux"] then {
     set target_cflags ""
 }
 
-if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable [list debug "incdir=${objdir}"]] != "" } {
+if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile} ${srcdir}/${subdir}/${srcfile2}" "${binfile}" executable [list debug "incdir=${objdir}"]] != "" } {
     return -1
 }
 
@@ -284,6 +286,20 @@ gdb_test "info address a_global" \
 setup_kfail "gdb/1294" "*-*-*"
 gdb_test "info address me" ".*me.*is a variable at offset.*" "info address me"
 
+
+# Test LOC_UNRESOLVED references resolving for `extern' TLS variables.
+
+gdb_test "p a_thread_local" " = \[0-9\]+"
+# Here it could crash with: Cannot access memory at address 0x0
+gdb_test "p file2_thread_local" " = \[0-9\]+"
+# Depending on the current lookup scope we get LOC_UNRESOLVED or LOC_COMPUTED
+# both printing:
+# Symbol "file2_thread_local" is a thread-local variable at offset 8 in the thread-local storage for `.../gdb.threads/tls'.
+gdb_test "info address file2_thread_local" "Symbol \"file2_thread_local\" is a thread-local variable.*"
+# Here it could also crash with: Cannot access memory at address 0x0
+gdb_test "p a_thread_local" " = \[0-9\]+" "p a_thread_local second time"
+gdb_test "info address a_thread_local" "Symbol \"a_thread_local\" is a thread-local variable.*"
+
 # Done!
 #
 gdb_exit
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ gdb/testsuite/gdb.threads/tls2.c	1 Dec 2008 21:50:30 -0000
@@ -0,0 +1,28 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2008 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+   Please email any bugs, comments, and/or additions to this file to:
+   bug-gdb@prep.ai.mit.edu  */
+
+extern __thread int a_thread_local;
+__thread int file2_thread_local;
+
+void
+function_referencing_a_thread_local (void)
+{
+  a_thread_local = a_thread_local;
+}

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]