On 09/20/16 at 09:45am, Alexei Starovoitov wrote:
that's a great idea. Instead of adding aborted, ring_full, blahblah counters
everywhere that cost performance, let's add tracepoints that are nops
when not in use.
And if all drivers call the same tracepoints it will be even better.
Monitoring trace_xdp_aborted tracepoint would be generic way to debug
'divide by zero'.
Not opposed but I still need an indication to start tracing ;-) A
single counter would be enough though.