Date: Wed, 25 Jun 2025 16:01:58 +1200
From: Tao Liu <ltao@redhat.com>
Subject: [Crash-utility] [PATCH v4 0/5] gdb multi-stack unwinding
        support
To: devel@lists.crash-utility.osci.io
Cc: Tao Liu <ltao@redhat.com>
Message-ID: <20250625040203.60334-1-ltao@redhat.com>
Content-Type: text/plain; charset="US-ASCII"; x-default=true

This patchset is based on Alexy's work [1], and is the follow-up of the
previous "gdb stack unwinding support for crash utility" patchset.

Currently gdb target analyzes only one task at a time and it backtraces
only straight stack until end of the stack. If stacks were concatenated
during exceptions or interrupts, gdb bt will show only the topmost one.

This patchset will introduce multiple stacks support for gdb stack unwinding,
which can be observed as a different threads from gdb perspective. A
short usage is as follows:

    'set <PID>'        - to switch to a specific task
    'gdb info threads' - to see list of in-kernel stacks of this task.
    'gdb thread <ID>'  - to switch to the stack.
    'gdb bt'           - to unwind it.

E.g, with the patchset:

crash> bt
PID: 17636    TASK: ffff88032e0742c0  CPU: 11   COMMAND: "kworker/11:4"
 #0 [ffff88037fca6b58] machine_kexec at ffffffff8103cef2
 #1 [ffff88037fca6ba8] crash_kexec at ffffffff810c9aa3
 #2 [ffff88037fca6c70] panic at ffffffff815f0444
 ...
 #9 [ffff88037fca6ec8] do_nmi at ffffffff815fd980
#10 [ffff88037fca6ef0] end_repeat_nmi at ffffffff815fcec1
    [exception RIP: memcpy+13]
    RIP: ffffffff812f5b1d  RSP: ffff88034f2a9728  RFLAGS: 00010046
    RAX: ffffc900139fe000  RBX: ffff880374b7a1b0  RCX: 0000000000000030
    RBP: ffff88034f2a9778   R8: 000000007fffffff   R9: 00000000ffffffff
    ...               
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
--- <NMI exception stack> ---
#11 [ffff88034f2a9728] memcpy at ffffffff812f5b1d
#12 [ffff88034f2a9728] mga_dirty_update at ffffffffa024ad2b [mgag200]
#13 [ffff88034f2a9780] mga_imageblit at ffffffffa024ae3f [mgag200]
#14 [ffff88034f2a97a0] bit_putcs at ffffffff813424ef
...

crash> info threads
  Id   Target Id                    Frame
* 1    17636 kworker/11:4 (stack 0) crash_setup_regs (oldregs=0x0, newregs=0xffff88037fca6bb0)
  2    17636 kworker/11:4 (stack 1) 0xffffffff812f5b1d in memcpy ()

crash> thread 2
crash> gdb bt
#0  0xffffffff812f5b1d in memcpy () at arch/x86/lib/memcpy_64.S:69
...

There are 2 stacks of the current task, and we can list/switch-to/unwind
each stack.

[1]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg01204.html

v2 -> v1: 1) Rebase this patchset onto gdb-16.2 [2].
          2) Improved the silent_call_bt() to catch the error FATAL.

[2]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg01354.html

v3 -> v2: 1) Rebase this patchset to crash v9.0.0.
          2) Fix v2's segfault in cmd "bt -E".
          3) Elimit repeat stacks by adding constraints before
             gdb_add_substack().

v4 -> v3: 1) Fix compiling warning of silent_call_bt()
          2) Add known issue link found in ppc arch.

Tao Liu (5):
  Add multi-threads support in crash target
  Call cmd_bt silently after "set pid"
  x86_64: Add gdb multi-stack unwind support
  arm64: Add gdb multi-stack unwind support
  ppc64: Add gdb multi-stack unwind support

 arm64.c         | 102 +++++++++++++++++++++++++++++++++--
 crash_target.c  |  49 +++++++++++++++--
 defs.h          |   3 +-
 gdb_interface.c |   6 +--
 kernel.c        |  44 +++++++++++++++
 ppc64.c         |  78 ++++++++++++++++++++++++---
 task.c          |   4 +-
 x86_64.c        | 138 +++++++++++++++++++++++++++++++++++++++++++++---
 8 files changed, 394 insertions(+), 30 deletions(-)

--
2.47.0