diff mbox

[patch+7.12.1,2/2] Fix TLS (such as 'errno') regression

Message ID 20161009185647.GB13645@host1.jankratochvil.net
State New
Headers show

Commit Message

Jan Kratochvil Oct. 9, 2016, 6:56 p.m. UTC
Hi,

2273f0ac95a79ce29ef42025c63f90e82cf907d7 is the first bad commit
commit 2273f0ac95a79ce29ef42025c63f90e82cf907d7
Author: Tom Tromey <tromey@redhat.com>
Date:   Tue Oct 15 13:28:57 2013 -0600
    change minsyms not to be relocated at read-time
[FYI v3 06/10] change minsyms not to be relocated at read-time
Message-Id: <1393441273-32268-7-git-send-email-tromey@redhat.com>
https://sourceware.org/ml/gdb-patches/2014-02/msg00798.html

mv /usr/lib/debug /usr/lib/debug-x
echo 'int main(){}'|gcc -pthread -x c -
./gdb -q -ex start -ex 'set debug expr 1' -ex 'p errno' ./a.out 
	    0  UNOP_MEMVAL_TLS       TLS type @0x35df7e0 (__thread /* "/lib64/libc.so.6" */ <thread local variable, no debug info>)
	    4    OP_LONG               Type @0x35df850 (__CORE_ADDR), value 140737345728528 (0x7ffff77fb010)
Cannot access memory at address 0xffffef7c9698
->
	    0  UNOP_MEMVAL_TLS       TLS type @0x3ad9520 (__thread /* "/lib64/libc.so.6" */ <thread local variable, no debug info>)
	    4    OP_LONG               Type @0x3ad9590 (__CORE_ADDR), value 16 (0x10)
$1 = 0

With glibc debuginfo, that is without: mv /usr/lib/debug /usr/lib/debug-x
	    0  OP_VAR_VALUE          Block @0x3b30e70, symbol @0x3b30d10 (errno)
$1 = 0
So such case is unrelated to this patch and the regression is not visible with
glibc debuginfo installed.

I guess all these issues will be solved by Gary Benson's Infinity.
But at least for older non-Infinity glibcs GDB should not regress.

For the testcase it is important the variable is in objfile with non-zero base
address.  glibc is a shared library for 'errno' but I found easier for the
testcase to use PIE instead of a shlib.  For TLS variables in PT_EXEC the
regression obviously does not happen.

It has been found by a more complete testcase present in Fedora, the fix there
also solves more cases where FSF GDB currently cannot resolve 'errno':
	http://pkgs.fedoraproject.org/cgit/rpms/gdb.git/tree/gdb-6.5-bz185337-resolve-tls-without-debuginfo-v2.patch
	FAIL: gdb.dwarf2/dw2-errno2.exp: macros=N threads=Y: print errno for core

No regressions on {x86_64,x86_64-m32,i686}-fedora26pre-linux-gnu.

OK for check-in?


Thanks,
Jan
gdb/ChangeLog
2016-10-09  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* parse.c (write_exp_msymbol): Fix ADDR computation.

gdb/testsuite/ChangeLog
2016-10-09  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* gdb.threads/tls-nodebug-pie.c: New file.
	* gdb.threads/tls-nodebug-pie.exp: New file.

Comments

Mark Wielaard Oct. 10, 2016, 10:45 a.m. UTC | #1
On Sun, 2016-10-09 at 20:56 +0200, Jan Kratochvil wrote:
> 2273f0ac95a79ce29ef42025c63f90e82cf907d7 is the first bad commit
> commit 2273f0ac95a79ce29ef42025c63f90e82cf907d7
> Author: Tom Tromey <tromey@redhat.com>
> Date:   Tue Oct 15 13:28:57 2013 -0600
>     change minsyms not to be relocated at read-time
> [FYI v3 06/10] change minsyms not to be relocated at read-time
> Message-Id: <1393441273-32268-7-git-send-email-tromey@redhat.com>
> https://sourceware.org/ml/gdb-patches/2014-02/msg00798.html
> 
> mv /usr/lib/debug /usr/lib/debug-x
> echo 'int main(){}'|gcc -pthread -x c -
> ./gdb -q -ex start -ex 'set debug expr 1' -ex 'p errno' ./a.out 
> 	    0  UNOP_MEMVAL_TLS       TLS type @0x35df7e0 (__thread /* "/lib64/libc.so.6" */ <thread local variable, no debug info>)
> 	    4    OP_LONG               Type @0x35df850 (__CORE_ADDR), value 140737345728528 (0x7ffff77fb010)
> Cannot access memory at address 0xffffef7c9698
> ->
> 	    0  UNOP_MEMVAL_TLS       TLS type @0x3ad9520 (__thread /* "/lib64/libc.so.6" */ <thread local variable, no debug info>)
> 	    4    OP_LONG               Type @0x3ad9590 (__CORE_ADDR), value 16 (0x10)
> $1 = 0
> 
> With glibc debuginfo, that is without: mv /usr/lib/debug /usr/lib/debug-x
> 	    0  OP_VAR_VALUE          Block @0x3b30e70, symbol @0x3b30d10 (errno)
> $1 = 0
> So such case is unrelated to this patch and the regression is not visible with
> glibc debuginfo installed.

FYI. This issue, cause and fix look very similar to PR varobj/18564
"regression in showing __thread so extern variable".

Might there be other places that need similar adjustments for getting at
the values of tls variables?
Tom Tromey Oct. 10, 2016, 2:50 p.m. UTC | #2
>>>>> "Jan" == Jan Kratochvil <jan.kratochvil@redhat.com> writes:

Jan> For the testcase it is important the variable is in objfile with non-zero base
Jan> address.

My only question about this patch is why the symbol's section has a
non-zero section offset.  Is that the reason why?

Maybe it would make sense to make a SEC_THREAD_LOCAL section always have
a zero offset.  Then touching all the users wouldn't be necessary.

On the third hand it seems strange to even try to get the "address" of a
TLS symbol in this way.

Tom
Jan Kratochvil Oct. 10, 2016, 3:13 p.m. UTC | #3
On Mon, 10 Oct 2016 16:50:38 +0200, Tom Tromey wrote:
> My only question about this patch is why the symbol's section has a
> non-zero section offset.  Is that the reason why?
> 
> Maybe it would make sense to make a SEC_THREAD_LOCAL section always have
> a zero offset.  Then touching all the users wouldn't be necessary.

IIUC you mean this subexpression could be zero (which it is not):
	ANOFFSET ((objfile)->section_offsets, ((symbol)->mginfo.section)

Section Headers:
  [Nr] Name              Type            Address          Off    Size   ES Flg Lk Inf Al
  [19] .tdata            PROGBITS        0000000000200db4 000db4 000004 00 WAT  0   0  4
Symbol table '.symtab' contains 71 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
    54: 0000000000000000     4 TLS     GLOBAL DEFAULT   19 thread_local

All the sections have the same offset for PIE - the PIE load base address.

And I do not think it is correct to set that offset to zero for SHF_TLS as the
thread initialization data is really located at the section address relocated
by that PIE-load-base address.


> On the third hand it seems strange to even try to get the "address" of a
> TLS symbol in this way.

There could be a symbol's getter which asserts/throws on reading address of
a TLS symbol?  Maybe, not a part of this patch.


Jan
Tom Tromey Oct. 10, 2016, 10:43 p.m. UTC | #4
>>>>> "Jan" == Jan Kratochvil <jan.kratochvil@redhat.com> writes:

Jan> And I do not think it is correct to set that offset to zero for
Jan> SHF_TLS as the thread initialization data is really located at the
Jan> section address relocated by that PIE-load-base address.

If it's useful then that makes sense.

>> On the third hand it seems strange to even try to get the "address" of a
>> TLS symbol in this way.

Jan> There could be a symbol's getter which asserts/throws on reading
Jan> address of a TLS symbol?

Yes.

Tom
diff mbox

Patch

--- a/gdb/parse.c
+++ b/gdb/parse.c
@@ -480,13 +480,17 @@  write_exp_msymbol (struct parser_state *ps,
   struct objfile *objfile = bound_msym.objfile;
   struct gdbarch *gdbarch = get_objfile_arch (objfile);
 
-  CORE_ADDR addr = BMSYMBOL_VALUE_ADDRESS (bound_msym);
   struct obj_section *section = MSYMBOL_OBJ_SECTION (objfile, msymbol);
   enum minimal_symbol_type type = MSYMBOL_TYPE (msymbol);
-  CORE_ADDR pc;
+  CORE_ADDR addr, pc;
   const int is_tls = (section != NULL
 		      && section->the_bfd_section->flags & SEC_THREAD_LOCAL);
 
+  if (is_tls)
+    addr = MSYMBOL_VALUE_RAW_ADDRESS (bound_msym.minsym);
+  else
+    addr = BMSYMBOL_VALUE_ADDRESS (bound_msym);
+
   /* The minimal symbol might point to a function descriptor;
      resolve it to the actual code address instead.  */
   pc = gdbarch_convert_from_func_ptr_addr (gdbarch, addr, &current_target);
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/tls-nodebug-pie.c
@@ -0,0 +1,27 @@ 
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2016 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/>.  */
+
+#include <pthread.h>
+
+__thread int thread_local = 42;
+
+int main(void)
+{
+  /* Ensure we link against pthreads even with --as-needed.  */
+  pthread_testcancel();
+  return 0;
+}
--- /dev/null
+++ b/gdb/testsuite/gdb.threads/tls-nodebug-pie.exp
@@ -0,0 +1,29 @@ 
+# Copyright 2016 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/>.
+
+standard_testfile
+
+if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable \
+			  [list "additional_flags=-fPIE -pie"]] != "" } {
+    return -1
+}
+
+clean_restart ${binfile}
+if ![runto_main] then {
+   return 0
+}
+
+# Formerly: Cannot access memory at address 0xd5554d5216fc
+gdb_test "p thread_local" " = 42" "thread local storage"