Re: [BUG] Failed to obtain stack trace via bpf_get_stackid on ARM64 architecture

On Wed, 24 Sep 2025 17:04:16 +0900 Masami Hiramatsu (Google) [off-list ref] wrote:

quoted

After testing, it was found that the stack could not be obtained because user_mode(regs) returned 1. 
Referring to the arch_ftrace_fill_perf_regs function in your email 
(https://lore.kernel.org/all/173518997908.391279.15910334347345106424.stgit@devnote2/ (local)), 
I made the following modification: by setting the value of pstate, the stack can now be obtained successfully.

diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
index 058a99aa44bd..f2814175e958 100644
--- a/arch/arm64/include/asm/ftrace.h
+++ b/arch/arm64/include/asm/ftrace.h

@@ -159,11 +159,13 @@ ftrace_partial_regs(const struct ftrace_regs *fregs, struct pt_regs *regs)
 {
        struct __arch_ftrace_regs *afregs = arch_ftrace_regs(fregs);
 
        memcpy(regs->regs, afregs->regs, sizeof(afregs->regs));
        regs->sp = afregs->sp;
        regs->pc = afregs->pc;
        regs->regs[29] = afregs->fp;
        regs->regs[30] = afregs->lr;
+       regs->pstate = PSR_MODE_EL1h;

Good catch! 

Should I submit this patch, or will you carry out a more complete fix?

quoted

By the way, during my testing, I also noticed that when executing bpf_get_stackid via kprobes or tracepoints, 
the command bpftrace -e 'kprobe:bpf_get_stackid {printf("bpf_get_stackid\n");}' produces no output. 

I think this is because the bpf_get_stackid is a kind of recursive
event from kprobes. Kprobe handler can not be reentered.

quoted

However, it does output something when bpf_get_stackid is invoked via uprobes. 
This phenomenon also occurs on the x86 architecture, could this be a bug as well?

Maybe if bpf_get_stackid() is kicked from uprobes, it is not recursive
call from kprobes, so it works.

So it is expected behavior, not a bug. Sorry for confusion.


Thank you,

Thank you very much for your explanation.