[Crash-utility] Re: [PATCH 2/2] RISCV64: Fix 'bt' output when no ra on the stack top
Resending to the new crash list address (without kexec list)
On 2023/12/04 19:50, Song Shuai wrote:
Same as the Linux commit f766f77a74f5 ("riscv/stacktrace: Fix
stack output without ra on the stack top").
When a function doesn't have a callee, then it will not
push ra into the stack, such as lkdtm functions, so
correct the FP of the second frame and use pt_regs to get
the right PC of the second frame.
TEST:
Before this patch, the `bt -f` outputs only the first frame with
the wrong PC and FP of next frame:
```
crash> bt -f
PID: 1 TASK: ff600000000e0000 CPU: 1 COMMAND: "sh"
#0 [ff20000000013cf0] lkdtm_EXCEPTION at ffffffff805303c0
[PC: ffffffff805303c0 RA: ff20000000013d10 SP: ff20000000013cf0 SIZE: 16] <-
wrong next PC
ff20000000013cf0: 0000000000000001 ff20000000013d10 <- next FP
ff20000000013d00: ff20000000013d40
crash>
```
After this patch, the `bt` outputs the full frames:
```
crash> bt
PID: 1 TASK: ff600000000e0000 CPU: 1 COMMAND: "sh"
#0 [ff20000000013cf0] lkdtm_EXCEPTION at ffffffff805303c0
#1 [ff20000000013d00] lkdtm_do_action at ffffffff8052fe36
#2 [ff20000000013d10] direct_entry at ffffffff80530018
#3 [ff20000000013d40] full_proxy_write at ffffffff80305044
#4 [ff20000000013d80] vfs_write at ffffffff801b68b4
#5 [ff20000000013e30] ksys_write at ffffffff801b6c4a
#6 [ff20000000013e80] __riscv_sys_write at ffffffff801b6cc4
#7 [ff20000000013e90] do_trap_ecall_u at ffffffff80836798
crash>
```
Signed-off-by: Song Shuai <songshuaishuai(a)tinylab.org>
Looks good.
Acked-by: Kazuhito Hagio <k-hagio-ab(a)nec.com>
Thanks,
Kazu
> ---
> riscv64.c | 15 +++++++++++++--
> 1 file changed, 13 insertions(+), 2 deletions(-)
>
> diff --git a/riscv64.c b/riscv64.c
> index 0aaa14b..872be59 100644
> --- a/riscv64.c
> +++ b/riscv64.c
> @@ -747,11 +747,14 @@ riscv64_back_trace_cmd(struct bt_info *bt)
> {
> struct riscv64_unwind_frame current, previous;
> struct stackframe curr_frame;
> + struct riscv64_register * regs;
> int level = 0;
>
> if (bt->flags & BT_REGS_NOT_FOUND)
> return;
>
> + regs = (struct riscv64_register *) bt->machdep;
> +
> current.pc = bt->instptr;
> current.sp = bt->stkptr;
> current.fp = bt->frameptr;
> @@ -788,8 +791,16 @@ riscv64_back_trace_cmd(struct bt_info *bt)
> sizeof(curr_frame), "get stack frame", RETURN_ON_ERROR))
> return;
>
> - previous.pc = curr_frame.ra;
> - previous.fp = curr_frame.fp;
> + /* correct PC and FP of the second frame when the first frame has no callee */
> +
> + if (regs && (regs->regs[RISCV64_REGS_EPC] == current.pc) &&
curr_frame.fp & 0x7){
> + previous.pc = regs->regs[RISCV64_REGS_RA];
> + previous.fp = curr_frame.ra;
> + } else {
> + previous.pc = curr_frame.ra;
> + previous.fp = curr_frame.fp;
> + }
> +
> previous.sp = current.fp;
>
> riscv64_dump_backtrace_entry(bt, symbol, ¤t, &previous, level++);