On Fri, Jan 18, 2019 at 02:45:44PM -0700, Jason Gunthorpe wrote: > On Wed, Dec 26, 2018 at 08:55:25PM -0500, ira.weiny@xxxxxxxxx wrote: > > From: Ira Weiny <ira.weiny@xxxxxxxxx> > > > > A while ago I wrote these patches for MAD stack tracing. At the time I was > > proposing to remove the snoop interface.[1] > > > > For this submission I'd like to propose adding the tracing and leave the snoop > > interface in. While I still don't see a need for the snoop interface, I'm no > > longer advocating getting rid of it as the functionality of these patches is > > different. > > > > In addition I wrote a sample eBPF which shows how one can further filter at the > > tracepoints to distill the information being traced. > > > > Changes for V3: > > Rebased on current RDMA for next > > Change license text to SPDX tag > > Change dev_name to string from array > > Reorder fields for more compact ring buffer utilization > > Use a defined roce address type for safer memcpy/memset > > Adjust BPF to new umad format > > > > Changes for v2: > > Update MAINTAINERS as indicated from Doug > > Now CC'ing the Tracing maintainers so they are aware of the additions > > > > [1] https://www.spinics.net/lists/linux-rdma/msg29109.html > > > > > > Ira Weiny (6): > > IB/MAD: Add send path trace points > > IB/MAD: Add recv path trace point > > IB/MAD: Add agent trace points > > IB/UMAD: Add umad trace points > > IB/MAD: Add SMP details to MAD tracing > > BPF: Add sample code for new ib_umad tracepoint > > > > MAINTAINERS | 4 + > > drivers/infiniband/core/mad.c | 95 +++++++++- > > drivers/infiniband/core/user_mad.c | 7 + > > include/trace/events/ib_mad.h | 362 +++++++++++++++++++++++++++++++++++++ > > include/trace/events/ib_umad.h | 112 ++++++++++++ > > samples/bpf/Makefile | 3 + > > samples/bpf/ibumad_kern.c | 123 +++++++++++++ > > samples/bpf/ibumad_user.c | 120 ++++++++++++ > > Can someone from the BPF side Ack that this is OK? > > https://patchwork.kernel.org/project/linux-rdma/list/?series=60629 Looks ok-ish. TP_STRUCT__entry() is huge, but as long as TP_ARGS has only two arguments the build will not complain.
