[PING][PATCH v8 0/5] Modernize frame unwinders and add disable feature
Guinevere Larsen
guinevere@redhat.com
Tue Jan 7 12:11:05 GMT 2025
Ping :)
All of the documentation, and patches 2 and 3, have already been approved.
On 12/10/24 4:51 PM, Guinevere Larsen wrote:
> This patch series started with me trying to make it easier to test GDB's
> ability to unwind using CFI data, to improve a previous patch I sent to
> the list.
>
> The first patch is just a minor change, storing frame unwinders in a
> vector instead of through an unwinder table accessible, still accessed
> throught the registry subsystem. This isn't required (like I originally
> thought it was), but it does make the whole system more readable in my
> opinion.
>
> Patch 2 adds a new field for frame unwinders, their "class", which
> roughly correlates to which area of GDB adds that unwinder. They were
> chosen rather arbitrarily based on my understanding, where the unwinder
> is added and its name. This class will be used to simplify bulk
> disabling of unwinders introduced by patch 4.
>
> Patch 3 has the real meat of the modernization, making GDB use
> polymorphism to handle frame unwinders. This is slightly slower than
> using function pointers, but much more readable in my opinion. Also, as
> noted by Thiago Bauermann, no frame unwinders that use the legacy class
> ever pass a frame_data pointer, so that field was removed from those
> unwinders.
>
> Patch 4 adds the possibility to disable unwinders based on their name or
> class, and disable all unwinders at once if desired. It also reverts
> commit 3c3bb0580be0 since that patch pointed out that if a more generic
> system was implemented it could be reverted.
>
> Patch 5 adds the test I was trying to upstream all along.
>
> All the documentation has been approved by Eli already. Patches 2 and 3
> were already approved by Simon.
>
> Changes for v8:
> * Simplified the loop printing the table for "maint info
> frame-unwinders".
> * Used gdb::option to parse the options for "maint frame-unwinder
> disable"
> * Changed error message when no unwinder is found but some unwinders
> were disabled
> * Cosmetic fixes.
>
> Changes for v7:
> * Moved the std::vector back to being accessed through the registry system
> Removed the approvals from patch 1 because I think this change is
> reasonably big, and needs another review.
> * Several cosmetic changes to the code, following feedback from Simon
> and Thiago.
> * Fixed the test from patch 4, following Thiago's suggestion
> * Users can no longer specify the unwinder class with the
> FRAME_UNWINDER_ prefix
>
> Changes for v6:
> * Patches 1 and 2 were approved by Simon
> * Several cosmetic changes to patch 3 and 4
> * New dealloc_cache default implementation (noop), and frame_unwind_legacy
> defaults to it if no function pointer is given.
> * New prev_arch default implementation (getting the current arch), and
> frame_unwind_legacy defaults to it if no pointer is given
> * Renamed "sniffer" to "sniff", to make it more clear what the method
> does
> * Fixed some problems with the testcase.
>
> Changes for v5:
> * Reverted commit 3c3bb0580be0, with patch 4 following the advice on
> that commit
> * on patch 3, some of the original functions were converted into the
> classes' methods instead of just passing the call along.
> * some methods of frame_unwind_legacy were switched to be purely
> virtual
>
> Changes for v4:
> * Rebase on current master (No changes needed)
> * improved documentation of unwinder classes follow Eli feedback
> * added KFAIL to new test on patch 5 if running on 32 bit arm
>
> Changes for v3:
> * Fixed Linaro CI issue in 32 bit arm
> * Fixed the few comments from Eli.
> * Documented addition of unwinder class in the patch that adds it.
>
> Changes for v2:
> * Added back the test that checks if GDB can handle a mix of CUs with
> and without debuginfo in a single backtrace.
> * Fixed all the nitpicks for style.
> * Removed FRAME_UNWIND_ prefix when talking about classes
> * Thoroughly changed the documentation, to be more readable
>
> Guinevere Larsen (5):
> gdb: make gdbarch store a vector of frame unwinders
> gdb: add "unwinder class" to frame unwinders
> gdb: Migrate frame unwinders to use C++ classes
> gdb: introduce ability to disable frame unwinders
> gdb/testsuite: Test for a backtrace through object without debuginfo
>
> gdb/NEWS | 21 +
> gdb/aarch64-tdep.c | 12 +-
> gdb/alpha-mdebug-tdep.c | 6 +-
> gdb/alpha-tdep.c | 12 +-
> gdb/amd64-obsd-tdep.c | 6 +-
> gdb/amd64-tdep.c | 24 +-
> gdb/amd64-windows-tdep.c | 6 +-
> gdb/amdgpu-tdep.c | 7 +-
> gdb/arc-tdep.c | 10 +-
> gdb/arm-tdep.c | 29 +-
> gdb/avr-tdep.c | 5 +-
> gdb/bfin-tdep.c | 6 +-
> gdb/bpf-tdep.c | 6 +-
> gdb/cris-tdep.c | 12 +-
> gdb/csky-tdep.c | 10 +-
> gdb/doc/gdb.texinfo | 64 +--
> gdb/dummy-frame.c | 8 +-
> gdb/dummy-frame.h | 2 +-
> gdb/dwarf2/frame-tailcall.c | 9 +-
> gdb/dwarf2/frame-tailcall.h | 2 +-
> gdb/dwarf2/frame.c | 46 +-
> gdb/dwarf2/frame.h | 6 -
> gdb/frame-unwind.c | 420 ++++++++++++++----
> gdb/frame-unwind.h | 173 +++++++-
> gdb/frame.c | 21 +-
> gdb/frv-linux-tdep.c | 6 +-
> gdb/frv-tdep.c | 5 +-
> gdb/ft32-tdep.c | 6 +-
> gdb/h8300-tdep.c | 5 +-
> gdb/hppa-linux-tdep.c | 5 +-
> gdb/hppa-tdep.c | 17 +-
> gdb/i386-obsd-tdep.c | 5 +-
> gdb/i386-tdep.c | 30 +-
> gdb/ia64-tdep.c | 24 +-
> gdb/inline-frame.c | 5 +-
> gdb/inline-frame.h | 2 +-
> gdb/iq2000-tdep.c | 5 +-
> gdb/jit.c | 6 +-
> gdb/lm32-tdep.c | 5 +-
> gdb/loongarch-tdep.c | 7 +-
> gdb/m32c-tdep.c | 5 +-
> gdb/m32r-linux-tdep.c | 5 +-
> gdb/m32r-tdep.c | 5 +-
> gdb/m68hc11-tdep.c | 5 +-
> gdb/m68k-linux-tdep.c | 6 +-
> gdb/m68k-tdep.c | 6 +-
> gdb/mep-tdep.c | 5 +-
> gdb/microblaze-tdep.c | 6 +-
> gdb/mips-sde-tdep.c | 6 +-
> gdb/mips-tdep.c | 24 +-
> gdb/mn10300-tdep.c | 5 +-
> gdb/moxie-tdep.c | 5 +-
> gdb/msp430-tdep.c | 5 +-
> gdb/nds32-tdep.c | 14 +-
> gdb/or1k-tdep.c | 7 +-
> gdb/ppc-fbsd-tdep.c | 5 +-
> gdb/ppc-obsd-tdep.c | 5 +-
> gdb/python/py-unwind.c | 61 ++-
> gdb/record-btrace.c | 12 +-
> gdb/record.h | 4 +-
> gdb/riscv-tdep.c | 8 +-
> gdb/rl78-tdep.c | 6 +-
> gdb/rs6000-aix-tdep.c | 5 +-
> gdb/rs6000-tdep.c | 12 +-
> gdb/rx-tdep.c | 10 +-
> gdb/s12z-tdep.c | 7 +-
> gdb/s390-linux-tdep.c | 5 +-
> gdb/s390-tdep.c | 10 +-
> gdb/sentinel-frame.c | 8 +-
> gdb/sentinel-frame.h | 2 +-
> gdb/sh-tdep.c | 11 +-
> gdb/sparc-netbsd-tdep.c | 6 +-
> gdb/sparc-obsd-tdep.c | 6 +-
> gdb/sparc-sol2-tdep.c | 6 +-
> gdb/sparc-tdep.c | 6 +-
> gdb/sparc64-fbsd-tdep.c | 6 +-
> gdb/sparc64-netbsd-tdep.c | 6 +-
> gdb/sparc64-obsd-tdep.c | 12 +-
> gdb/sparc64-sol2-tdep.c | 6 +-
> gdb/sparc64-tdep.c | 6 +-
> .../backtrace-through-cu-nodebug-caller.c | 28 ++
> .../backtrace-through-cu-nodebug-main.c | 32 ++
> .../gdb.base/backtrace-through-cu-nodebug.exp | 95 ++++
> .../gdb.base/frame-info-consistent.exp | 8 +-
> gdb/testsuite/gdb.base/frame-unwind-disable.c | 22 +
> .../gdb.base/frame-unwind-disable.exp | 137 ++++++
> gdb/testsuite/gdb.base/maint.exp | 4 -
> gdb/tic6x-tdep.c | 12 +-
> gdb/tilegx-tdep.c | 5 +-
> gdb/tramp-frame.c | 70 ++-
> gdb/v850-tdep.c | 5 +-
> gdb/vax-tdep.c | 6 +-
> gdb/windows-tdep.c | 35 +-
> gdb/windows-tdep.h | 16 +-
> gdb/xstormy16-tdep.c | 5 +-
> gdb/xtensa-tdep.c | 7 +-
> gdb/z80-tdep.c | 7 +-
> 97 files changed, 1339 insertions(+), 551 deletions(-)
> create mode 100644 gdb/testsuite/gdb.base/backtrace-through-cu-nodebug-caller.c
> create mode 100644 gdb/testsuite/gdb.base/backtrace-through-cu-nodebug-main.c
> create mode 100644 gdb/testsuite/gdb.base/backtrace-through-cu-nodebug.exp
> create mode 100644 gdb/testsuite/gdb.base/frame-unwind-disable.c
> create mode 100644 gdb/testsuite/gdb.base/frame-unwind-disable.exp
>
--
Cheers,
Guinevere Larsen
She/Her/Hers
More information about the Gdb-patches
mailing list