| Message ID | 20180919223935.999270-2-ast@kernel.org |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | BPF Maintainers |
| Headers | show |
| Series | perf, bpf: reveal invisible bpf programs | expand |
> On Sep 19, 2018, at 3:39 PM, Alexei Starovoitov <ast@kernel.org> wrote: > > introduce perf_event_mmap_bpf_prog() helper to emit RECORD_MMAP events > into perf ring buffer. > It's used by bpf load/unload logic to notify user space of addresses > and names of JITed bpf programs. > > Note that event->mmap.pid == -1 is an existing indicator of kernel event. > In addition use event->mmap.tid == BPF_FS_MAGIC to indicate bpf related > RECORD_MMAP event. > > Alternatively it's possible to introduce new 'enum perf_event_type' command > specificially for bpf prog load/unload, but existing RECORD_MMAP > is very close, so the choice made by this patch is to extend it. > > Signed-off-by: Alexei Starovoitov <ast@kernel.org> Acked-by: Song Liu <songliubraving@fb.com> I guess we should also use this for kernel modules load/unload? > --- > include/linux/perf_event.h | 1 + > kernel/events/core.c | 44 +++++++++++++++++++++++++++++++++----- > 2 files changed, 40 insertions(+), 5 deletions(-) > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h > index 53c500f0ca79..0e79af83138f 100644 > --- a/include/linux/perf_event.h > +++ b/include/linux/perf_event.h > @@ -1113,6 +1113,7 @@ static inline void perf_event_task_sched_out(struct task_struct *prev, > } > > extern void perf_event_mmap(struct vm_area_struct *vma); > +void perf_event_mmap_bpf_prog(u64 start, u64 len, char *name, int size); > extern struct perf_guest_info_callbacks *perf_guest_cbs; > extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); > extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); > diff --git a/kernel/events/core.c b/kernel/events/core.c > index 2a62b96600ad..c48244ddf993 100644 > --- a/kernel/events/core.c > +++ b/kernel/events/core.c > @@ -7152,7 +7152,7 @@ static int perf_event_mmap_match(struct perf_event *event, > { > struct perf_mmap_event *mmap_event = data; > struct vm_area_struct *vma = mmap_event->vma; > - int executable = vma->vm_flags & VM_EXEC; > + int executable = !vma || vma->vm_flags & VM_EXEC; > > return (!executable && event->attr.mmap_data) || > (executable && (event->attr.mmap || event->attr.mmap2)); > @@ -7165,12 +7165,13 @@ static void perf_event_mmap_output(struct perf_event *event, > struct perf_output_handle handle; > struct perf_sample_data sample; > int size = mmap_event->event_id.header.size; > + bool bpf_event = !mmap_event->vma; > int ret; > > if (!perf_event_mmap_match(event, data)) > return; > > - if (event->attr.mmap2) { > + if (event->attr.mmap2 && !bpf_event) { > mmap_event->event_id.header.type = PERF_RECORD_MMAP2; > mmap_event->event_id.header.size += sizeof(mmap_event->maj); > mmap_event->event_id.header.size += sizeof(mmap_event->min); > @@ -7186,12 +7187,14 @@ static void perf_event_mmap_output(struct perf_event *event, > if (ret) > goto out; > > - mmap_event->event_id.pid = perf_event_pid(event, current); > - mmap_event->event_id.tid = perf_event_tid(event, current); > + if (!bpf_event) { > + mmap_event->event_id.pid = perf_event_pid(event, current); > + mmap_event->event_id.tid = perf_event_tid(event, current); > + } > > perf_output_put(&handle, mmap_event->event_id); > > - if (event->attr.mmap2) { > + if (event->attr.mmap2 && !bpf_event) { > perf_output_put(&handle, mmap_event->maj); > perf_output_put(&handle, mmap_event->min); > perf_output_put(&handle, mmap_event->ino); > @@ -7448,6 +7451,37 @@ void perf_event_mmap(struct vm_area_struct *vma) > perf_event_mmap_event(&mmap_event); > } > > +void perf_event_mmap_bpf_prog(u64 start, u64 len, char *name, int size) > +{ > + struct perf_mmap_event mmap_event; > + > + if (!atomic_read(&nr_mmap_events)) > + return; > + > + if (!IS_ALIGNED(size, sizeof(u64))) { > + WARN_ONCE(1, "size is not aligned\n"); > + return; > + } > + > + mmap_event = (struct perf_mmap_event){ > + .file_name = name, > + .file_size = size, > + .event_id = { > + .header = { > + .type = PERF_RECORD_MMAP, > + .misc = PERF_RECORD_MISC_KERNEL, > + .size = sizeof(mmap_event.event_id) + size, > + }, > + .pid = -1, /* indicates kernel */ > + .tid = BPF_FS_MAGIC, /* bpf mmap event */ > + .start = start, > + .len = len, > + .pgoff = start, > + }, > + }; > + perf_iterate_sb(perf_event_mmap_output, &mmap_event, NULL); > +} > + > void perf_event_aux_event(struct perf_event *event, unsigned long head, > unsigned long size, u64 flags) > { > -- > 2.17.1 >
On 9/19/18 4:30 PM, Song Liu wrote: > > >> On Sep 19, 2018, at 3:39 PM, Alexei Starovoitov <ast@kernel.org> wrote: >> >> introduce perf_event_mmap_bpf_prog() helper to emit RECORD_MMAP events >> into perf ring buffer. >> It's used by bpf load/unload logic to notify user space of addresses >> and names of JITed bpf programs. >> >> Note that event->mmap.pid == -1 is an existing indicator of kernel event. >> In addition use event->mmap.tid == BPF_FS_MAGIC to indicate bpf related >> RECORD_MMAP event. >> >> Alternatively it's possible to introduce new 'enum perf_event_type' command >> specificially for bpf prog load/unload, but existing RECORD_MMAP >> is very close, so the choice made by this patch is to extend it. >> >> Signed-off-by: Alexei Starovoitov <ast@kernel.org> > > Acked-by: Song Liu <songliubraving@fb.com> > > I guess we should also use this for kernel modules load/unload? yes. that's possible. There is similar issue with modules today that get unloaded before 'perf report'. Synthetic record_mmap for modules that perf emits into perf.data has filename==module_name. It solves typical use case though. We can extend record_mmap further and let kernel emit them for module load/unload. Some new magic for 'tid' would be necessary.
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 53c500f0ca79..0e79af83138f 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -1113,6 +1113,7 @@ static inline void perf_event_task_sched_out(struct task_struct *prev, } extern void perf_event_mmap(struct vm_area_struct *vma); +void perf_event_mmap_bpf_prog(u64 start, u64 len, char *name, int size); extern struct perf_guest_info_callbacks *perf_guest_cbs; extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); diff --git a/kernel/events/core.c b/kernel/events/core.c index 2a62b96600ad..c48244ddf993 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -7152,7 +7152,7 @@ static int perf_event_mmap_match(struct perf_event *event, { struct perf_mmap_event *mmap_event = data; struct vm_area_struct *vma = mmap_event->vma; - int executable = vma->vm_flags & VM_EXEC; + int executable = !vma || vma->vm_flags & VM_EXEC; return (!executable && event->attr.mmap_data) || (executable && (event->attr.mmap || event->attr.mmap2)); @@ -7165,12 +7165,13 @@ static void perf_event_mmap_output(struct perf_event *event, struct perf_output_handle handle; struct perf_sample_data sample; int size = mmap_event->event_id.header.size; + bool bpf_event = !mmap_event->vma; int ret; if (!perf_event_mmap_match(event, data)) return; - if (event->attr.mmap2) { + if (event->attr.mmap2 && !bpf_event) { mmap_event->event_id.header.type = PERF_RECORD_MMAP2; mmap_event->event_id.header.size += sizeof(mmap_event->maj); mmap_event->event_id.header.size += sizeof(mmap_event->min); @@ -7186,12 +7187,14 @@ static void perf_event_mmap_output(struct perf_event *event, if (ret) goto out; - mmap_event->event_id.pid = perf_event_pid(event, current); - mmap_event->event_id.tid = perf_event_tid(event, current); + if (!bpf_event) { + mmap_event->event_id.pid = perf_event_pid(event, current); + mmap_event->event_id.tid = perf_event_tid(event, current); + } perf_output_put(&handle, mmap_event->event_id); - if (event->attr.mmap2) { + if (event->attr.mmap2 && !bpf_event) { perf_output_put(&handle, mmap_event->maj); perf_output_put(&handle, mmap_event->min); perf_output_put(&handle, mmap_event->ino); @@ -7448,6 +7451,37 @@ void perf_event_mmap(struct vm_area_struct *vma) perf_event_mmap_event(&mmap_event); } +void perf_event_mmap_bpf_prog(u64 start, u64 len, char *name, int size) +{ + struct perf_mmap_event mmap_event; + + if (!atomic_read(&nr_mmap_events)) + return; + + if (!IS_ALIGNED(size, sizeof(u64))) { + WARN_ONCE(1, "size is not aligned\n"); + return; + } + + mmap_event = (struct perf_mmap_event){ + .file_name = name, + .file_size = size, + .event_id = { + .header = { + .type = PERF_RECORD_MMAP, + .misc = PERF_RECORD_MISC_KERNEL, + .size = sizeof(mmap_event.event_id) + size, + }, + .pid = -1, /* indicates kernel */ + .tid = BPF_FS_MAGIC, /* bpf mmap event */ + .start = start, + .len = len, + .pgoff = start, + }, + }; + perf_iterate_sb(perf_event_mmap_output, &mmap_event, NULL); +} + void perf_event_aux_event(struct perf_event *event, unsigned long head, unsigned long size, u64 flags) {
introduce perf_event_mmap_bpf_prog() helper to emit RECORD_MMAP events into perf ring buffer. It's used by bpf load/unload logic to notify user space of addresses and names of JITed bpf programs. Note that event->mmap.pid == -1 is an existing indicator of kernel event. In addition use event->mmap.tid == BPF_FS_MAGIC to indicate bpf related RECORD_MMAP event. Alternatively it's possible to introduce new 'enum perf_event_type' command specificially for bpf prog load/unload, but existing RECORD_MMAP is very close, so the choice made by this patch is to extend it. Signed-off-by: Alexei Starovoitov <ast@kernel.org> --- include/linux/perf_event.h | 1 + kernel/events/core.c | 44 +++++++++++++++++++++++++++++++++----- 2 files changed, 40 insertions(+), 5 deletions(-)