[Crash-utility] Re: [PATCH v7 00/15] gdb stack unwinding support for crash utility

Hi lianbo, On Wed, Sep 11, 2024 at 2:26 PM lijiang <lijiang(a)redhat.com&gt; wrote: > > Hi, Tao > > Thank you for the update. > > The following patch is a regression issue, so I tend to discuss it as a separate patch. > [PATCH v7 01/15] Fix the regression of cpumask_t for xen hyper

> In addition, I found another issue in my tests(on ppc64le), the gdb bt can display the back trace for the panic task, but when I switch to another task, the gdb bt can not display the back trace: > > crash> gdb bt > #0 0xc0000000002bde04 in crash_setup_regs (newregs=0xc00000003264b858, oldregs=0x0) at ./arch/powerpc/include/asm/kexec.h:133 > #1 0xc0000000002be4f8 in __crash_kexec (regs=0x0) at kernel/crash_core.c:122 > #2 0xc00000000016c254 in panic (fmt=0xc0000000015eef20 "sysrq triggered crash\n") at kernel/panic.c:373 > #3 0xc000000000a708b8 in sysrq_handle_crash (key=<optimized out>) at drivers/tty/sysrq.c:154 > #4 0xc000000000a713d4 in __handle_sysrq (key=key@entry=99 'c', check_mask=check_mask@entry=false) at drivers/tty/sysrq.c:612 > #5 0xc000000000a71e94 in write_sysrq_trigger (file=<optimized out>, buf=<optimized out>, count=2, ppos=<optimized out>) at drivers/tty/sysrq.c:1181 > #6 0xc00000000073260c in pde_write (pde=0xc00000000af9cc00, file=<optimized out>, buf=<optimized out>, count=<optimized out>, ppos=<optimized out>) at fs/proc/inode.c:334 > #7 proc_reg_write (file=<optimized out>, buf=<optimized out>, count=<optimized out>, ppos=<optimized out>) at fs/proc/inode.c:346 > #8 0xc00000000063c0e0 in vfs_write (file=0xc0000000092d2900, buf=0x10012536f60 <error: Cannot access memory at address 0x10012536f60>, count=2, pos=0xc00000003264bd30) at fs/read_write.c:588 > #9 vfs_write (file=0xc0000000092d2900, buf=0x10012536f60 <error: Cannot access memory at address 0x10012536f60>, count=<optimized out>, pos=0xc00000003264bd30) at fs/read_write.c:570 > #10 0xc00000000063c690 in ksys_write (fd=<optimized out>, buf=0x10012536f60 <error: Cannot access memory at address 0x10012536f60>, count=2) at fs/read_write.c:643 > #11 0xc000000000031a28 in system_call_exception (regs=0xc00000003264be80, r0=<optimized out>) at arch/powerpc/kernel/syscall.c:153 > #12 0xc00000000000d05c in system_call_vectored_common () at arch/powerpc/kernel/interrupt_64.S:198 > > crash> ps > PID PPID CPU TASK ST %MEM VSZ RSS COMM > 0 0 0 c000000002bda980 RU 0.0 0 0 [swapper/0] > > 0 0 1 c000000003864c80 RU 0.0 0 0 [swapper/1] > ... > 8017 923 0 c000000043a20000 IN 0.2 22528 16256 sshd-session > 8025 8017 6 c000000032271880 IN 0.1 22784 11840 sshd-session > > 8026 8025 0 c000000043a26600 RU 0.1 9664 6208 bash > ... > 11645 2 3 c000000032264c80 ID 0.0 0 0 [kworker/u32:2] > 11738 6188 2 c00000003811b180 IN 0.1 43520 9408 pickup > 12326 2 0 c00000003226b280 ID 0.0 0 0 [kworker/0:1] > 13112 6089 2 c00000000c809900 IN 0.0 7232 3456 sleep > > Let's take the "pickup" task as an example: > > crash> set 11738 > PID: 11738 > COMMAND: "pickup" > TASK: c00000003811b180 [THREAD_INFO: c00000003811b180] > CPU: 2 > STATE: TASK_INTERRUPTIBLE > > crash> gdb bt > #0 0xc0000000a7f876a0 in ?? () > gdb: gdb request failed: bt > crash> set gdb on > gdb: on > gdb> bt > #0 0xc0000000a7f876a0 in ?? () > gdb> There is a bug for ppc64 crash of newer version kernel. The code for determining the address of pt_regs from stack is outdated, see the following code from crash: ppc64.c:get_ppc64_frame() readmem(sp+STACK_FRAME_OVERHEAD, KVADDR, &regs, sizeof(struct ppc64_pt_regs), "PPC64 pt_regs", FAULT_ON_ERROR); The pt_regs is expected to be placed at sp+STACK_FRAME_OVERHEAD, aka sp+112. However since kernel >= v6.2, the value is no longer appropriate: linux kernel:arch/powerpc/kernel/process.c:copy_thread(): kregs = (struct pt_regs *)(sp + STACK_SWITCH_FRAME_REGS); p->thread.ksp = sp; linux kernel:arch/powerpc/include/asm/ptrace.h: #ifdef CONFIG_PPC64_ELF_ABI_V2 #define STACK_FRAME_MIN_SIZE 32 STACK_SWITCH_FRAME_REGS (STACK_FRAME_MIN_SIZE + 16)

If we apply the change to crash, i.e: readmem(sp+0x30, KVADDR, &regs, sizeof(struct ppc64_pt_regs), "PPC64 pt_regs", FAULT_ON_ERROR); The stack unwinding can work as expected, you can have a test locally to see if the above change works for you. So this bug isn't related to the gdb stack unwinding support to me, just a bug relating to a newer version of kernel.

there are plenty of places in crash which use the old STACK_FRAME_OVERHEAD value, maybe they all need to be updated.

Thanks, Tao Liu > > Anyway, I did the same test on x86 64 and aarch64, it can work well as expected. Can you help to double check on ppc64 architecture? > > X86 64: > crash> set 14599 > PID: 14599 > COMMAND: "pickup" > TASK: ffff8f57a0d7c180 [THREAD_INFO: ffff8f57a0d7c180] > CPU: 41 > STATE: TASK_INTERRUPTIBLE > crash> gdb bt > #0 0xffffffff8b3efe29 in context_switch (rq=0xffff8f6f1f835900, prev=0xffff8f57a0d7c180, next=0xffff8f5786720000, rf=0xffff9df22fea7b80) at kernel/sched/core.c:5208 > #1 __schedule (sched_mode=sched_mode@entry=0) at kernel/sched/core.c:6549 > #2 0xffffffff8b3f0217 in __schedule_loop (sched_mode=<optimized out>) at kernel/sched/core.c:6626 > #3 schedule () at kernel/sched/core.c:6641 > #4 0xffffffff8b3f6eef in schedule_hrtimeout_range_clock (expires=expires@entry=0xffff9df22fea7cb0, delta=<optimized out>, delta@entry=99999999, mode=mode@entry=HRTIMER_MODE_ABS, clock_id=clock_id@entry=1) at kernel/time/hrtimer.c:2293 > #5 0xffffffff8b3f7003 in schedule_hrtimeout_range (expires=expires@entry=0xffff9df22fea7cb0, delta=delta@entry=99999999, mode=mode@entry=HRTIMER_MODE_ABS) at kernel/time/hrtimer.c:2340 > #6 0xffffffff8aae301c in ep_poll (ep=0xffff8f5790d15d40, events=events@entry=0x7ffea91b6b90, maxevents=maxevents@entry=100, timeout=timeout@entry=0xffff9df22fea7d58) at fs/eventpoll.c:2062 > #7 0xffffffff8aae3138 in do_epoll_wait (epfd=epfd@entry=8, events=events@entry=0x7ffea91b6b90, maxevents=maxevents@entry=100, to=0xffff9df22fea7d58) at fs/eventpoll.c:2464 > #8 0xffffffff8aae44a1 in __do_sys_epoll_wait (epfd=<optimized out>, events=0x7ffea91b6b90, maxevents=<optimized out>, timeout=<optimized out>) at fs/eventpoll.c:2476 > #9 __se_sys_epoll_wait (epfd=<optimized out>, events=<optimized out>, maxevents=<optimized out>, timeout=<optimized out>) at fs/eventpoll.c:2471 > #10 __x64_sys_epoll_wait (regs=<optimized out>) at fs/eventpoll.c:2471 > #11 0xffffffff8b3e293d in do_syscall_x64 (regs=0xffff9df22fea7f48, nr=232) at arch/x86/entry/common.c:52 > #12 do_syscall_64 (regs=0xffff9df22fea7f48, nr=232) at arch/x86/entry/common.c:83 > #13 0xffffffff8b40012f in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:121 > crash> > > > aarch64: > crash> set 9338 > PID: 9338 > COMMAND: "pickup" > TASK: ffff0000c7b05400 [THREAD_INFO: ffff0000c7b05400] > CPU: 3 > STATE: TASK_INTERRUPTIBLE > crash> gdb bt > #0 __switch_to (prev=<unavailable>, prev@entry=0xffff0000c7b05400, next=next@entry=<unavailable>) at arch/arm64/kernel/process.c:555 > #1 0xffffafc5b5ebd744 in context_switch (rq=0xffff00077bbd0ec0, prev=0xffff0000c7b05400, next=<unavailable>, rf=0xffff80008ac63a60) at kernel/sched/core.c:5208 > #2 __schedule (sched_mode=sched_mode@entry=0) at kernel/sched/core.c:6549 > #3 0xffffafc5b5ebdc2c in __schedule_loop (sched_mode=<optimized out>) at kernel/sched/core.c:6626 > #4 schedule () at kernel/sched/core.c:6641 > #5 0xffffafc5b5ec6030 in schedule_hrtimeout_range_clock (expires=expires@entry=0xffff80008ac63be8, delta=delta@entry=99999999, mode=mode@entry=HRTIMER_MODE_ABS, clock_id=clock_id@entry=1) at kernel/time/hrtimer.c:2293 > #6 0xffffafc5b5ec618c in schedule_hrtimeout_range (expires=expires@entry=0xffff80008ac63be8, delta=delta@entry=99999999, mode=mode@entry=HRTIMER_MODE_ABS) at kernel/time/hrtimer.c:2340 > #7 0xffffafc5b545d33c in ep_poll (ep=<unavailable>, events=events@entry=0xffffde5c3f68, maxevents=maxevents@entry=100, timeout=timeout@entry=0xffff80008ac63ce0) at fs/eventpoll.c:2062 > #8 0xffffafc5b545d4e4 in do_epoll_wait (epfd=epfd@entry=8, events=events@entry=0xffffde5c3f68, maxevents=maxevents@entry=100, to=to@entry=0xffff80008ac63ce0) at fs/eventpoll.c:2464 > #9 0xffffafc5b545d534 in do_epoll_pwait (epfd=epfd@entry=8, events=events@entry=0xffffde5c3f68, maxevents=maxevents@entry=100, to=to@entry=0xffff80008ac63ce0, sigsetsize=<optimized out>, sigmask=<optimized out>) at fs/eventpoll.c:2498 > #10 0xffffafc5b545e7c8 in do_epoll_pwait (epfd=8, events=0xffffde5c3f68, maxevents=100, to=0xffff80008ac63ce0, sigmask=<optimized out>, sigsetsize=<optimized out>) at fs/eventpoll.c:2495 > #11 __do_sys_epoll_pwait (epfd=8, events=0xffffde5c3f68, maxevents=100, timeout=<optimized out>, sigmask=<optimized out>, sigsetsize=<optimized out>) at fs/eventpoll.c:2511 > #12 __se_sys_epoll_pwait (epfd=8, events=281474412330856, maxevents=100, timeout=<optimized out>, sigmask=<optimized out>, sigsetsize=<optimized out>) at fs/eventpoll.c:2505 > #13 __arm64_sys_epoll_pwait (regs=<optimized out>) at fs/eventpoll.c:2505 > #14 0xffffafc5b4fa99bc in __invoke_syscall (regs=0xffff80008ac63eb0, syscall_fn=<optimized out>) at arch/arm64/kernel/syscall.c:35 > #15 invoke_syscall (regs=regs@entry=0xffff80008ac63eb0, scno=<optimized out>, sc_nr=sc_nr@entry=463, syscall_table=<optimized out>) at arch/arm64/kernel/syscall.c:49 > #16 0xffffafc5b4fa9ac8 in el0_svc_common (sc_nr=463, syscall_table=<optimized out>, regs=0xffff80008ac63eb0, scno=<optimized out>) at arch/arm64/kernel/syscall.c:132 > #17 do_el0_svc (regs=regs@entry=0xffff80008ac63eb0) at arch/arm64/kernel/syscall.c:151 > #18 0xffffafc5b5eb6fa4 in el0_svc (regs=0xffff80008ac63eb0) at arch/arm64/kernel/entry-common.c:712 > #19 0xffffafc5b5eb74c0 in el0t_64_sync_handler (regs=<optimized out>) at arch/arm64/kernel/entry-common.c:730 > #20 0xffffafc5b4f91634 in el0t_64_sync () at arch/arm64/kernel/entry.S:598 > crash> > > BTW: other changes are fine to me. > > Thanks > Lianbo > > On Wed, Sep 4, 2024 at 3:54 PM < devel-request(a)lists.crash-utility.osci.io&gt; wrote: >> >> Date: Wed, 4 Sep 2024 19:49:25 +1200 >> From: Tao Liu <ltao(a)redhat.com&gt; >> Subject: [Crash-utility] [PATCH v7 00/15] gdb stack unwinding support >> for crash utility >> To: devel(a)lists.crash-utility.osci.io >> Cc: Tao Liu <ltao(a)redhat.com&gt; >> Message-ID: <20240904074940.21331-1-ltao(a)redhat.com&gt; >> Content-Type: text/plain; charset=UTF-8 >> >> This patchset is a rebase/merged version of the following 3 patchsets: >> >> 1): [PATCH v10 0/5] Improve stack unwind on ppc64 [1] >> 2): [PATCH 0/5] x86_64 gdb stack unwinding support [2] >> 3): Clean up on top of one-thread-v2 [3] >> >> A complete description of gdb stack unwinding support for crash can be >> found in [1]. >> >> This patchset can be divided into the following 3 parts: >> >> 1) part1: preparations before stack unwinding support, some >> bugs/regressions found when drafting this patchset. >> 2) part2: common part for all CPU archs, mainly dealing with >> crash_target.c/gdb_interface.c files, in order to >> support different archs. >> 3) part3: arch specific, for each ppc64/x86_64/arm64/vmware >> stack unwinding support. >> >> === part 3 >> arm64: Add gdb stack unwinding support >> vmware_guestdump: Various format versions support >> x86_64: Add gdb stack unwinding support >> ppc64: correct gdb passthroughs by implementing machdep->get_current_task_reg >> >> === part 2 >> Conditionally output gdb stack unwinding stop reasons >> Stop stack unwinding at non-kernel address >> Print task pid/command instead of CPU index >> Rename get_cpu_reg to get_current_task_reg >> Let crash change gdb context >> Leave only one gdb thread for crash >> Remove 'frame' from prohibited commands list >> >> === part 1 >> Fix gdb_interface: restore gdb's output streams at end of gdb_interface >> x86_64: Fix invalid input "=>" for bt command >> Fix cpumask_t recursive dependence issue >> Fix the regression of cpumask_t for xen hyper >> === >> >> v7 -> v6: >> 1) Reorganise the patchset, re-divided them into 3 part against the >> previous 2 parts. >> 2) Re-dealed with the cpumask_t part, which solved the comment No.4 >> pointed out by lianbo in [4]. >> 3) Add conditional output for the failing message of gdb stack unwinding. >> see [PATCH 11/15] Conditionally output gdb stack unwinding stop reasons >> 4) Redraft the commit messages, updated some outdated info. >> 5) Merged "Let crash change gdb context" and "set_context(): check if >> context is already current" into one. >> >> [4]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg01067.html >> >> v6 -> v5: >> 1) Refactor patch 4 & 9, which changed the function signature of struct >> get_cpu_reg/get_current_task_reg, and let each patch compile with no >> error when added on. >> 2) Rebased the patchset on top of latest upstream: >> ("79b93ecb2e72ec Fix a "Bus error" issue caused by 'crash --osrelease' or >> crash loading") >> >> v5 -> v4: >> 1) Plenty of code refactoring based on Lianbo's comments on v4. >> 2) Removed the magic number when dealing with regs bitmap, see [6]. >> 3) Rebased the patchset on top of latest upstream: >> ("1c6da3eaff8207 arm64: Fix bt command show wrong stacktrace on ramdump source") >> >> v4 -> v3: >> Fixed the author issue in [PATCH v3 06/16] Fix gdb_interface: restore gdb's >> output streams at end of gdb_interface. >> >> v3 -> v2: >> 1) Updated CC list as pointed out in [4] >> 2) Compiling issues as in [5] >> >> v2 -> v1: >> 1) Added the patch: x86_64: Fix invalid input "=>" for bt command, >> thanks for Kazu's testing. >> 2) Modify the patch: x86_64: Add gdb stack unwinding support, added the >> pcp_save, spp_save and sp, for restoring the value in match of the original >> code logic. >> >> [1]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00469.html >> [2]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00488.html >> [3]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00554.html >> [4]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00681.html >> [5]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00715.html >> [6]: https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg00819.html >> >> Aditya Gupta (3): >> Fix gdb_interface: restore gdb's output streams at end of >> gdb_interface >> Remove 'frame' from prohibited commands list >> ppc64: correct gdb passthroughs by implementing >> machdep->get_current_task_reg >> >> Alexey Makhalov (1): >> vmware_guestdump: Various format versions support >> >> Tao Liu (11): >> Fix the regression of cpumask_t for xen hyper >> Fix cpumask_t recursive dependence issue >> x86_64: Fix invalid input "=>" for bt command >> Leave only one gdb thread for crash >> Let crash change gdb context >> Rename get_cpu_reg to get_current_task_reg >> Print task pid/command instead of CPU index >> Stop stack unwinding at non-kernel address >> Conditionally output gdb stack unwinding stop reasons >> x86_64: Add gdb stack unwinding support >> arm64: Add gdb stack unwinding support >> >> arm64.c | 120 +++++++++++++++-- >> crash_target.c | 71 ++++++---- >> defs.h | 194 ++++++++++++++++++++++++++- >> gdb-10.2.patch | 96 ++++++++++++++ >> gdb_interface.c | 39 ++---- >> kernel.c | 63 +++++++-- >> ppc64.c | 174 +++++++++++++++++++++++- >> symbols.c | 15 +++ >> task.c | 34 +++-- >> tools.c | 16 ++- >> unwind_x86_64.h | 4 - >> vmware_guestdump.c | 321 +++++++++++++++++++++++++++++++------------- >> x86_64.c | 323 ++++++++++++++++++++++++++++++++++++++++----- >> 13 files changed, 1247 insertions(+), 223 deletions(-) >> >> -- >> 2.40.1