perf, bpf: allow bpf programs attach to tracepoints
authorAlexei Starovoitov <ast@fb.com>
Thu, 7 Apr 2016 01:43:25 +0000 (18:43 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 8 Apr 2016 01:04:26 +0000 (21:04 -0400)
commit98b5c2c65c2951772a8fc661f50d675e450e8bce
treea72f04b6056a18b50d6244e8f4b6f41e95f9e0c5
parent1e1dcd93b468901e114f279c94a0b356adc5e7cd
perf, bpf: allow bpf programs attach to tracepoints

introduce BPF_PROG_TYPE_TRACEPOINT program type and allow it to be attached
to the perf tracepoint handler, which will copy the arguments into
the per-cpu buffer and pass it to the bpf program as its first argument.
The layout of the fields can be discovered by doing
'cat /sys/kernel/debug/tracing/events/sched/sched_switch/format'
prior to the compilation of the program with exception that first 8 bytes
are reserved and not accessible to the program. This area is used to store
the pointer to 'struct pt_regs' which some of the bpf helpers will use:
+---------+
| 8 bytes | hidden 'struct pt_regs *' (inaccessible to bpf program)
+---------+
| N bytes | static tracepoint fields defined in tracepoint/format (bpf readonly)
+---------+
| dynamic | __dynamic_array bytes of tracepoint (inaccessible to bpf yet)
+---------+

Not that all of the fields are already dumped to user space via perf ring buffer
and broken application access it directly without consulting tracepoint/format.
Same rule applies here: static tracepoint fields should only be accessed
in a format defined in tracepoint/format. The order of fields and
field sizes are not an ABI.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/trace/perf.h
include/uapi/linux/bpf.h
kernel/events/core.c