public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
* [PATCH v2 0/5] [gdb/symtab] Assume epilogue unwind info is valid unless gcc < 4.5.0
@ 2023-02-13 14:23 Tom de Vries
  2023-02-13 14:23 ` [PATCH v2 1/5] [gdb/symtab] Factor out compunit_epilogue_unwind_valid Tom de Vries
                   ` (4 more replies)
  0 siblings, 5 replies; 9+ messages in thread
From: Tom de Vries @ 2023-02-13 14:23 UTC (permalink / raw)
  To: gdb-patches; +Cc: Tom Tromey

[ This patch series is a resubmission of this patch (
https://sourceware.org/pipermail/gdb-patches/2023-January/196325.html ).

The problem with the earlier version was that I tested it with a broken
version of test-case gdb.base/unwind-on-each-insn.exp, something that was
fixed by commit d9195131530 ("[gdb/testsuite] Simplify
gdb.base/unwind-on-each-insn.exp.tcl").

Retesting with a fixed test-case revealed the need for the patch
"[gdb/tdep] Add amd64/i386 epilogue override unwinders".

Furthermore, I've added two introductory refactoring patches to make following
changes more compact.

Now that this turned into a patch series, I've also split off
the dwarf2/read.c change into a separate patch. ]

The gcc 4.4.x (and earlier) compilers had the problem that the unwind info in
the epilogue was inaccurate.

In order to work around this in gdb, epilogue unwinders were added with a
higher priority than the dwarf unwinders in the amd64 and i386 targets:
- amd64_epilogue_frame_unwind, and
- i386_epilogue_frame_unwind
see:
- submission emails:
  https://sourceware.org/pipermail/gdb-patches/2009-July/066779.html
  https://sourceware.org/pipermail/gdb-patches/2009-August/067684.html
- gdb commits 872761f485e and 06da04c6105

Subsequently, the epilogue unwind info problem got fixed in gcc 4.5.0, see:
- submission email
  https://gcc.gnu.org/pipermail/gcc-patches/2009-May/261377.html
- gcc commit cd9c1ca866b
- release notes gcc 4.5.0 ( https://gcc.gnu.org/gcc-4.5/changes.html ):
  GCC now generates unwind info also for epilogues.

However, the epilogue unwinders prevented gdb from taking advantage of the
fixed epilogue unwind info, so the scope of the epilogue unwinders was
limited, bailing out for gcc >= 4.5.0, see:
- submisssion email
  https://sourceware.org/pipermail/gdb-patches/2011-June/083429.html
- gdb commit e0d00bc749e "Disable epilogue unwinders on recent GCCs"

This scope limitation mechanism works well for gcc -g: the producer is
available in the debug info, and we can determine whether we're dealing
with reliable epilogue unwind info or not.

For gcc -g0 though, epilogue unwind information is available in .eh_frame, but
the producer is not availabe to determine whether that information is reliable
or not, and consequently the info is ignored:
- in the case of using gcc <= 4.4.x, that is the ok decision and we're working
  around the gcc problem, but
- in the case of gcc >= 4.5.0, that means we fail to take advantage of fixed
  epilogue unwind info.

Furthermore, let's review the history of what epilogue unwind information is
trusted by gdb:
- initial position: trust all information
- after the epilogue unwinders were added: trust none
- after the scope limitation: only trust gcc >= 4.5.0.

So, while we started out with trusting info from all compilers, we end up
trusting only gcc >= 4.5.0, which seems a bit of an overreach for a workaround
for a problem in the gcc compiler.

Fix these two issues by reversing the burden of proof:
- currently we assume epilogue unwind info is invalid unless we can proof that
  gcc >= 4.5.0.
- instead, assume epilogue unwind info is valid unless we can proof that
  gcc < 4.5.0.

An added benefit of this is that it makes the amd64 and i386 targets more
similar to other targets, which makes comparing behaviour easier.  Note that
some other targets also have an epilogue unwinder, but none of those have a
higher priority than the dwarf unwinders.

Tested on x86_64-linux with gcc 7.5.0, with target boards unix/-m64 and
unix/-m32.

A regression test-case has been added to test the -g0
-fasynchronous-unwind-tables scenario, on amd64 (but not on i386).

PR tdep/30028
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30028
---
 gdb/amd64-tdep.c                              |  4 +-
 gdb/dwarf2/read.c                             | 12 ++++-
 gdb/i386-tdep.c                               |  4 +-
 .../gdb.base/unwind-on-each-insn-amd64-2.exp  | 52 ++++++++++++++++++
 .../gdb.base/unwind-on-each-insn-amd64-2.s    | 54 +++++++++++++++++++
 5 files changed, 123 insertions(+), 3 deletions(-)
 create mode 100644 gdb/testsuite/gdb.base/unwind-on-each-insn-amd64-2.exp
 create mode 100644 gdb/testsuite/gdb.base/unwind-on-each-insn-amd64-2.s

Tom de Vries (5):
  [gdb/symtab] Factor out compunit_epilogue_unwind_valid
  [gdb/tdep] Fix amd64/i386_stack_frame_destroyed_p
  [gdb/tdep] Add amd64/i386 epilogue override unwinders
  [gdb/symtab] Trust epilogue unwind info for unknown producer (-g0
    case)
  [gdb/symtab] Trust epilogue unwind info for unknown or non-gcc
    producer

 gdb/amd64-tdep.c                              | 69 +++++++++++++++---
 gdb/dwarf2/read.c                             |  8 ++-
 gdb/i386-tdep.c                               | 71 +++++++++++++++----
 gdb/symtab.h                                  | 13 ++++
 .../gdb.base/unwind-on-each-insn-amd64-2.exp  | 52 ++++++++++++++
 .../gdb.base/unwind-on-each-insn-amd64-2.s    | 54 ++++++++++++++
 6 files changed, 243 insertions(+), 24 deletions(-)
 create mode 100644 gdb/testsuite/gdb.base/unwind-on-each-insn-amd64-2.exp
 create mode 100644 gdb/testsuite/gdb.base/unwind-on-each-insn-amd64-2.s


base-commit: 97c195191578e9a68bfbb810eea373f5f3efcb7d
-- 
2.35.3


^ permalink raw reply	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2023-02-20 10:27 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-02-13 14:23 [PATCH v2 0/5] [gdb/symtab] Assume epilogue unwind info is valid unless gcc < 4.5.0 Tom de Vries
2023-02-13 14:23 ` [PATCH v2 1/5] [gdb/symtab] Factor out compunit_epilogue_unwind_valid Tom de Vries
2023-02-14 15:56   ` Tom Tromey
2023-02-14 22:36     ` Tom de Vries
2023-02-20 10:27       ` Tom de Vries
2023-02-13 14:23 ` [PATCH v2 2/5] [gdb/tdep] Fix amd64/i386_stack_frame_destroyed_p Tom de Vries
2023-02-13 14:23 ` [PATCH v2 3/5] [gdb/tdep] Add amd64/i386 epilogue override unwinders Tom de Vries
2023-02-13 14:23 ` [PATCH v2 4/5] [gdb/symtab] Trust epilogue unwind info for unknown producer (-g0 case) Tom de Vries
2023-02-13 14:23 ` [PATCH v2 5/5] [gdb/symtab] Trust epilogue unwind info for unknown or non-gcc producer Tom de Vries

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).