Re: Getting empty callchain from perf_callchain_kernel()

From: Song Liu
Date: Fri May 17 2019 - 14:43:56 EST


+Alexei, Daniel, and bpf

> On May 17, 2019, at 2:10 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Fri, May 17, 2019 at 04:15:39PM +0800, Kairui Song wrote:
>> Hi, I think the actual problem is that bpf_get_stackid_tp (and maybe
>> some other bfp functions) is now broken, or, strating an unwind
>> directly inside a bpf program will end up strangely. It have following
>> kernel message:
>
> Urgh, what is that bpf_get_stackid_tp() doing to get the regs? I can't
> follow.

I guess we need something like the following? (we should be able to
optimize the PER_CPU stuff).

Thanks,
Song


diff --git i/kernel/trace/bpf_trace.c w/kernel/trace/bpf_trace.c
index f92d6ad5e080..c525149028a7 100644
--- i/kernel/trace/bpf_trace.c
+++ w/kernel/trace/bpf_trace.c
@@ -696,11 +696,13 @@ static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
.arg5_type = ARG_CONST_SIZE_OR_ZERO,
};

+static DEFINE_PER_CPU(struct pt_regs, bpf_stackid_tp_regs);
BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
u64, flags)
{
- struct pt_regs *regs = *(struct pt_regs **)tp_buff;
+ struct pt_regs *regs = this_cpu_ptr(&bpf_stackid_tp_regs);

+ perf_fetch_caller_regs(regs);
/*
* Same comment as in bpf_perf_event_output_tp(), only that this time
* the other helper's function body cannot be inlined due to being
@@ -719,10 +721,13 @@ static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
.arg3_type = ARG_ANYTHING,
};

+static DEFINE_PER_CPU(struct pt_regs, bpf_stack_tp_regs);
BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
u64, flags)
{
- struct pt_regs *regs = *(struct pt_regs **)tp_buff;
+ struct pt_regs *regs = this_cpu_ptr(&bpf_stack_tp_regs);
+
+ perf_fetch_caller_regs(regs);

return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
(unsigned long) size, flags, 0);