| Message ID | 20220908214104.3851807-2-namhyung@kernel.org (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | BPF |
| Headers | show |
| Series | [1/3] perf: Use sample_flags for callchain | expand |
| Context | Check | Description |
|---|---|---|
| bpf/vmtest-bpf-next-PR | fail | PR summary |
| netdev/tree_selection | success | Not a local patch |
| bpf/vmtest-bpf-next-VM_Test-6 | success | Logs for set-matrix |
| bpf/vmtest-bpf-next-VM_Test-1 | pending | Logs for build for s390x with gcc |
| bpf/vmtest-bpf-next-VM_Test-4 | success | Logs for llvm-toolchain |
| bpf/vmtest-bpf-next-VM_Test-5 | success | Logs for set-matrix |
| bpf/vmtest-bpf-next-VM_Test-2 | fail | Logs for build for x86_64 with gcc |
| bpf/vmtest-bpf-next-VM_Test-3 | fail | Logs for build for x86_64 with llvm-16 |
On 09/08, Namhyung Kim wrote: > If the perf_event has PERF_SAMPLE_CALLCHAIN, BPF can use it for stack > trace. > The problematic cases like PEBS and IBS already handled in the PMU driver > and > they filled the callchain info in the sample data. For others, we can > call > perf_callchain() before the BPF handler. > Signed-off-by: Namhyung Kim <namhyung@kernel.org> Reviewed-by: Stanislav Fomichev <sdf@google.com> At least from the description it make sense. We're filling a callchain when it's been requested by the event, but it's missing on the sample data (aka, software fallback?). perf_callchain also seems to always fallback to &__empty_callchain in case of an error, so seems safe. > --- > kernel/bpf/stackmap.c | 4 ++-- > kernel/events/core.c | 12 ++++++++++-- > 2 files changed, 12 insertions(+), 4 deletions(-) > diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c > index 1adbe67cdb95..aecea7451b61 100644 > --- a/kernel/bpf/stackmap.c > +++ b/kernel/bpf/stackmap.c > @@ -338,7 +338,7 @@ BPF_CALL_3(bpf_get_stackid_pe, struct > bpf_perf_event_data_kern *, ctx, > int ret; > /* perf_sample_data doesn't have callchain, use bpf_get_stackid */ > - if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY)) > + if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) > return bpf_get_stackid((unsigned long)(ctx->regs), > (unsigned long) map, flags, 0, 0); > @@ -506,7 +506,7 @@ BPF_CALL_4(bpf_get_stack_pe, struct > bpf_perf_event_data_kern *, ctx, > int err = -EINVAL; > __u64 nr_kernel; > - if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY)) > + if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) > return __bpf_get_stack(regs, NULL, NULL, buf, size, flags); > if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | > diff --git a/kernel/events/core.c b/kernel/events/core.c > index b8af9fdbf26f..2ea93ce75ad4 100644 > --- a/kernel/events/core.c > +++ b/kernel/events/core.c > @@ -10003,8 +10003,16 @@ static void bpf_overflow_handler(struct > perf_event *event, > goto out; > rcu_read_lock(); > prog = READ_ONCE(event->prog); > - if (prog) > + if (prog) { > + if (prog->call_get_stack && > + (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) && > + !(data->sample_flags & PERF_SAMPLE_CALLCHAIN)) { > + data->callchain = perf_callchain(event, regs); > + data->sample_flags |= PERF_SAMPLE_CALLCHAIN; > + } > + > ret = bpf_prog_run(prog, &ctx); > + } > rcu_read_unlock(); > out: > __this_cpu_dec(bpf_prog_active); > @@ -10030,7 +10038,7 @@ static int perf_event_set_bpf_handler(struct > perf_event *event, > if (event->attr.precise_ip && > prog->call_get_stack && > - (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY) || > + (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) || > event->attr.exclude_callchain_kernel || > event->attr.exclude_callchain_user)) { > /* > -- > 2.37.2.789.g6183377224-goog
diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c index 1adbe67cdb95..aecea7451b61 100644 --- a/kernel/bpf/stackmap.c +++ b/kernel/bpf/stackmap.c @@ -338,7 +338,7 @@ BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx, int ret; /* perf_sample_data doesn't have callchain, use bpf_get_stackid */ - if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY)) + if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) return bpf_get_stackid((unsigned long)(ctx->regs), (unsigned long) map, flags, 0, 0); @@ -506,7 +506,7 @@ BPF_CALL_4(bpf_get_stack_pe, struct bpf_perf_event_data_kern *, ctx, int err = -EINVAL; __u64 nr_kernel; - if (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY)) + if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) return __bpf_get_stack(regs, NULL, NULL, buf, size, flags); if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK | diff --git a/kernel/events/core.c b/kernel/events/core.c index b8af9fdbf26f..2ea93ce75ad4 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -10003,8 +10003,16 @@ static void bpf_overflow_handler(struct perf_event *event, goto out; rcu_read_lock(); prog = READ_ONCE(event->prog); - if (prog) + if (prog) { + if (prog->call_get_stack && + (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) && + !(data->sample_flags & PERF_SAMPLE_CALLCHAIN)) { + data->callchain = perf_callchain(event, regs); + data->sample_flags |= PERF_SAMPLE_CALLCHAIN; + } + ret = bpf_prog_run(prog, &ctx); + } rcu_read_unlock(); out: __this_cpu_dec(bpf_prog_active); @@ -10030,7 +10038,7 @@ static int perf_event_set_bpf_handler(struct perf_event *event, if (event->attr.precise_ip && prog->call_get_stack && - (!(event->attr.sample_type & __PERF_SAMPLE_CALLCHAIN_EARLY) || + (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) || event->attr.exclude_callchain_kernel || event->attr.exclude_callchain_user)) { /*
If the perf_event has PERF_SAMPLE_CALLCHAIN, BPF can use it for stack trace. The problematic cases like PEBS and IBS already handled in the PMU driver and they filled the callchain info in the sample data. For others, we can call perf_callchain() before the BPF handler. Signed-off-by: Namhyung Kim <namhyung@kernel.org> --- kernel/bpf/stackmap.c | 4 ++-- kernel/events/core.c | 12 ++++++++++-- 2 files changed, 12 insertions(+), 4 deletions(-)