[PATCH v2 00/24] Multi-target support

Pedro Alves pedro@palves.net
Thu Aug 6 15:16:10 GMT 2020


On 8/4/20 4:30 AM, Kevin Buettner via Gdb-patches wrote:
> On Fri, 10 Jan 2020 20:13:22 +0000
> Pedro Alves <palves@redhat.com> wrote:
> 
>> I've now merged the multi-target work to master, including
>> the couple follow up patches developed and discussed on this
>> thread.
> 
> I've run into a regression stemming from this commit:
> 
> 5b6d1e4fa4 (HEAD, refs/bisect/bad) Multi-target support
> 
> More info in Bug 26336:
> 
> https://sourceware.org/bugzilla/show_bug.cgi?id=26336
> 
> (I thought at first that I introduced this problem with my recent
> core file work, but realized that was not the case after I disabled
> it.  I then did a bisection starting from late last year.)

Thanks for the bisect and the initial analysis.  This patch fixes
it for me.  No regressions for me on either -m32 nor -m64, and
the corefile.exp regression is fixed.  Let me know what you think.

>From a82de99c678920ba3f7a296419a8aaa376c92d98 Mon Sep 17 00:00:00 2001
From: Pedro Alves <pedro@palves.net>
Date: Thu, 6 Aug 2020 13:05:02 +0100
Subject: [PATCH] gdb.base/corefile.exp regression for unix/-m32 on x86_64 (PR
 26336)

gdb.base/corefile.exp is showing an unexpected failure and an
unresolved testcase when testing against unix/-m32:

 (gdb) PASS: gdb.base/corefile.exp: attach: sanity check we see the core file
 attach 15741
 gdb/dwarf2-frame.c:1009: internal-error: dwarf2_frame_cache* dwarf2_frame_cache(frame_info*, void**): Assertion `fde != NULL' failed.
 A problem internal to GDB has been detected,
 further debugging may prove unreliable.
 Quit this debugging session? (y or n) FAIL: gdb.base/corefile.exp: attach: with core (GDB internal error)
 Resyncing due to internal error.

This regressed with:

 From 5b6d1e4fa4fc6827c7b3f0e99ff120dfa14d65d2 Mon Sep 17 00:00:00 2001
 From: Pedro Alves <palves@redhat.com>
 Date: Fri, 10 Jan 2020 20:06:08 +0000
 Subject: [PATCH] Multi-target support

The assertion is here:

 #0  internal_error (file=0xbffffccb0 <error: Cannot access memory at address 0xbffffccb0>, line=0, fmt=0x555556327320 "en_US.UTF-8") at sr
 c/gdbsupport/errors.cc:51
 #1  0x00005555557d4e45 in dwarf2_frame_cache (this_frame=0x55555672f950, this_cache=0x55555672f968) at src/gdb/dwarf2/frame.c:1013
 #2  0x00005555557d5886 in dwarf2_frame_this_id (this_frame=0x55555672f950, this_cache=0x55555672f968, this_id=0x55555672f9b0) at src/gdb/d
 warf2/frame.c:1226
 #3  0x00005555558b184e in compute_frame_id (fi=0x55555672f950) at src/gdb/frame.c:558
 #4  0x00005555558b19b2 in get_frame_id (fi=0x55555672f950) at src/gdb/frame.c:588
 #5  0x0000555555bda338 in scoped_restore_current_thread::scoped_restore_current_thread (this=0x7fffffffd0d8) at src/gdb/thread.c:1458
 #6  0x00005555556ce41f in scoped_restore_current_pspace_and_thread::scoped_restore_current_pspace_and_thread (During symbol reading: .debug_line address at offset 0x1db2d3
 is 0 [in module /home/pedro/gdb/cascais-builds/binutils-gdb/gdb/gdb]
 this=0x7fffffffd0d0) at src/gdb/progspace-and-thread.h:29
 #7  0x0000555555898ea6 in remove_target_sections (owner=0x555556935550) at src/gdb/exec.c:798
 #8  0x0000555555b700b6 in symfile_free_objfile (objfile=0x555556935550) at src/gdb/symfile.c:3742
 #9  0x000055555565050e in std::_Function_handler<void (objfile*), void (*)(objfile*)>::_M_invoke(std::_Any_data const&, objfile*&&) (__functor=..., __args#0=@0x7fffffffd190
 : 0x555556935550) at /usr/include/c++/9/bits/std_function.h:300
 #10 0x0000555555a3053d in std::function<void (objfile*)>::operator()(objfile*) const (this=0x555556752a20, __args#0=0x555556935550) at /usr/include/c++/9/bits/std_function.
 h:688
 #11 0x0000555555a2ff01 in gdb::observers::observable<objfile*>::notify (this=0x5555562eaa80 <gdb::observers::free_objfile>, args#0=0x555556935550) at /net/cascais.nfs/gdb/b
 inutils-gdb/src/gdb/../gdbsupport/observable.h:106
 #12 0x0000555555a2c56a in objfile::~objfile (this=0x555556935550, __in_chrg=<optimized out>) at src/gdb/objfiles.c:521
 #13 0x0000555555a31d46 in std::_Sp_counted_ptr<objfile*, (__gnu_cxx::_Lock_policy)2>::_M_dispose (this=0x555556c1f6f0) at /usr/include/c++/9/bits/shared_ptr_base.h:377
 #14 0x00005555556d3444 in std::_Sp_counted_base<(__gnu_cxx::_Lock_policy)2>::_M_release (this=0x555556c1f6f0) at /usr/include/c++/9/bits/shared_ptr_base.h:155
 #15 0x00005555556cec77 in std::__shared_count<(__gnu_cxx::_Lock_policy)2>::~__shared_count (this=0x555556b99ee8, __in_chrg=<optimized out>) at /usr/include/c++/9/bits/shared_ptr_base.h:730
 #16 0x0000555555a2f8da in std::__shared_ptr<objfile, (__gnu_cxx::_Lock_policy)2>::~__shared_ptr (this=0x555556b99ee0, __in_chrg=<optimized out>) at /usr/include/c++/9/bits/shared_ptr_base.h:1169
 #17 0x0000555555a2f8fa in std::shared_ptr<objfile>::~shared_ptr (this=0x555556b99ee0, __in_chrg=<optimized out>) at /usr/include/c++/9/bits/shared_ptr.h:103
 #18 0x0000555555a63fba in __gnu_cxx::new_allocator<std::_List_node<std::shared_ptr<objfile> > >::destroy<std::shared_ptr<objfile> > (this=0x55555679f0c0, __p=0x555556b99ee0) at /usr/include/c++/9/ext/new_allocator.h:153
 #19 0x0000555555a638fb in std::allocator_traits<std::allocator<std::_List_node<std::shared_ptr<objfile> > > >::destroy<std::shared_ptr<objfile> > (__a=..., __p=0x555556b99ee0) at /usr/include/c++/9/bits/alloc_traits.h:497
 #20 0x0000555555a6351c in std::__cxx11::list<std::shared_ptr<objfile>, std::allocator<std::shared_ptr<objfile> > >::_M_erase (this=0x55555679f0c0, __position=std::shared_ptr<objfile> (expired, weak count 1) = {get() = 0x555556935550}) at /usr/include/c++/9/bits/stl_list.h:1921
 #21 0x0000555555a62dab in std::__cxx11::list<std::shared_ptr<objfile>, std::allocator<std::shared_ptr<objfile> > >::erase (this=0x55555679f0c0, __position=std::shared_ptr<objfile> (expired, weak count 1) = {get() = 0x555556935550}) at /usr/include/c++/9/bits/list.tcc:158
 #22 0x0000555555a614dd in program_space::remove_objfile (this=0x55555679f080, objfile=0x555556935550) at src/gdb/progspace.c:207
 #23 0x0000555555a2c4dc in objfile::unlink (this=0x555556935550) at src/gdb/objfiles.c:497
 #24 0x0000555555a2da65 in objfile_purge_solibs () at src/gdb/objfiles.c:904
 #25 0x0000555555b3af74 in no_shared_libraries (ignored=0x0, from_tty=1) at src/gdb/solib.c:1236
 #26 0x0000555555bbafc7 in target_pre_inferior (from_tty=1) at src/gdb/target.c:1900
 #27 0x0000555555940afb in attach_command (args=0x5555563277c7 "15741", from_tty=1) at src/gdb/infcmd.c:2582
 ...


The problem is that the multi-target commit added a
scoped_restore_current_thread to remove_target_sections (frame #7
above).  scoped_restore_current_thread's ctor fetches the selected
frame's frame id.  If the frame had not had its frame id computed yet,
it is computed then (frame #4 above).  Because it has been determined
earlier that the frame's unwinder is the DWARF unwinder, we end up
here:

 static struct dwarf2_frame_cache *
 dwarf2_frame_cache (struct frame_info *this_frame, void **this_cache)
 {
 ...
   /* Find the correct FDE.  */
   fde = dwarf2_frame_find_fde (&pc1, &cache->per_objfile);
   gdb_assert (fde != NULL);

And, that assertion fails.  The assertion is reasonable, because the
DWARF unwinder only claims the frame if it managed to find the FDE
earlier (in dwarf2_frame_sniffer).

(unix/-m32 is thus really a red herring here -- it's just that on
x86_64 -m64, the frame is not claimed by the DWARF unwinder.)

The reason the assertion is failing, is because the objfile that
contains the FDE has been removed from the objfiles list already when
we get here (frame #22 above).  This suggests that the fix should be
to invalidate DWARF frames when their objfile is removed.  Or to keep
it simple and safe, invalidate the frame cache when an objfile is
removed.  That is what this commit does.

OOC, I checked why is it that when you unload a file with plain "(gdb)
file", we don't hit the assertion.  It must be because we're already
flushing the frame cache somewhere else in that case.  And indeed, we
flush the frame cache here:

 (gdb) bt
 #0  reinit_frame_cache () at src/gdb/frame.c:1857
 #1  0x0000555555ad1ad6 in registers_changed_ptid (target=0x0, ptid=...) at src/gdb/regcache.c:470
 #2  0x0000555555ad1b58 in registers_changed () at src/gdb/regcache.c:485
 #3  0x00005555558d095e in set_target_gdbarch (new_gdbarch=0x555556d5f5b0) at src/gdb/gdbarch.c:5528
 #4  0x0000555555677175 in set_gdbarch_from_file (abfd=0x0) at src/gdb/arch-utils.c:601
 #5  0x0000555555897c6b in exec_file_attach (filename=0x0, from_tty=1) at src/gdb/exec.c:409
 #6  0x000055555589852d in exec_file_command (args=0x0, from_tty=1) at src/gdb/exec.c:571
 #7  0x00005555558985a1 in file_command (arg=0x0, from_tty=1) at src/gdb/exec.c:583
 #8  0x000055555572b55f in do_const_cfunc (c=0x55555672e200, args=0x0, from_tty=1) at src/gdb/cli/cli-decode.c:95
 #9  0x000055555572f3d3 in cmd_func (cmd=0x55555672e200, args=0x0, from_tty=1) at src/gdb/cli/cli-decode.c:2181
 #10 0x0000555555be1ecc in execute_command (p=0x555556327804 "", from_tty=1) at src/gdb/top.c:668
 #11 0x0000555555895427 in command_handler (command=0x555556327800 "file") at src/gdb/event-top.c:588
 #12 0x00005555558958af in command_line_handler (rl=...) at src/gdb/event-top.c:773
 #13 0x0000555555894b3e in gdb_rl_callback_handler (rl=0x55555a09e240 "file") at src/gdb/event-top.c:219
 #14 0x0000555555ccfeec in rl_callback_read_char () at src/readline/readline/callback.c:281
 #15 0x000055555589495a in gdb_rl_callback_read_char_wrapper_noexcept () at src/gdb/event-top.c:177
 #16 0x0000555555894a08 in gdb_rl_callback_read_char_wrapper (client_data=0x555556327520) at src/gdb/event-top.c:194
 #17 0x00005555558952a5 in stdin_event_handler (error=0, client_data=0x555556327520) at src/gdb/event-top.c:516
 #18 0x0000555555e027d6 in handle_file_event (file_ptr=0x555558d20840, ready_mask=1) at src/gdbsupport/event-loop.cc:548
 #19 0x0000555555e02d88 in gdb_wait_for_event (block=1) at src/gdbsupport/event-loop.cc:673
 #20 0x0000555555e01c42 in gdb_do_one_event () at src/gdbsupport/event-loop.cc:215
 #21 0x00005555559c47c2 in start_event_loop () at src/gdb/main.c:356
 #22 0x00005555559c490d in captured_command_loop () at src/gdb/main.c:416
 #23 0x00005555559c6217 in captured_main (data=0x7fffffffdc00) at src/gdb/main.c:1253
 #24 0x00005555559c6289 in gdb_main (args=0x7fffffffdc00) at src/gdb/main.c:1268
 #25 0x0000555555621756 in main (argc=3, argv=0x7fffffffdd18) at src/gdb/gdb.c:32

gdb/ChangeLog:

	* progspace.c (program_space::remove_objfile): Invalidate the
	frame cache.
---
 gdb/progspace.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/gdb/progspace.c b/gdb/progspace.c
index a0b14a6d2eb..462083ce1f6 100644
--- a/gdb/progspace.c
+++ b/gdb/progspace.c
@@ -198,6 +198,12 @@ program_space::add_objfile (std::shared_ptr<objfile> &&objfile,
 void
 program_space::remove_objfile (struct objfile *objfile)
 {
+  /* Removing an objfile from the objfile list invalidates any frame
+     that was built using frame info found in the objfile.  Reinit the
+     frame cache to get rid of any frame that might otherwise
+     reference stale info.  */
+  reinit_frame_cache ();
+
   auto iter = std::find_if (objfiles_list.begin (), objfiles_list.end (),
 			    [=] (const std::shared_ptr<::objfile> &objf)
 			    {

base-commit: 1a9f72a7a8f445b8d665eb36b053a18e758e63e6
-- 
2.14.5



More information about the Gdb-patches mailing list