Attention is currently required from: Hoernchen.
9 comments:
File include/osmocom/mgcp/mgcp_trunk.h:
Patch Set #13, Line 38: struct mgcp_ratectr_trunk ratectr; /* NULL for the threads! */
And this kind of stuff is what makes me speak again about having different structs for the global state and one for each worker thread.
File src/libosmo-mgcp/mgcp_endp.c:
Patch Set #13, Line 128: if (endp->callid) {
So these stats are being dropped? why? I don't recall reading about it in the commit description.
File src/libosmo-mgcp/mgcp_stat.c:
Patch Set #13, Line 84: atomic_uint_least64_t packets_rx = conn->atomic_counters[RTP_PACKETS_RX_CTR];
AFAIU these don't need to be atomic, since they are basically copied locally and used here, no synchonization needs to happen back.
File src/libosmo-mgcp/mgcp_threads.c:
Patch Set #13, Line 428: for (int i = 0; i < _MGCP_GENERAL_NUM_ENUMS; i++) {
This kind of loops make me thing whether we'd be better using a mutex rather than accessing crazy amounts of atomic variables (barriers, etc.)
Patch Set #13, Line 433: atomic_uint_least64_t endpoints_used = 0;
why does this need to be atomic? It's just a local variable not being accessed by other threads?
Patch Set #13, Line 490: /* wait for the threads to be done with init, so main thread can safely read atomics*/
space: atomics */
File src/libosmo-mgcp/mgcp_threads_vty.c:
Patch Set #13, Line 34: atomic_uint_least64_t tx_packets, tx_bytes;
same ehre, why are these atomic?
File src/libosmo-mgcp/mgcp_trunk.c:
Patch Set #13, Line 143: trunk->number_endpoints = number_endpoints;
another stat being removed here?
File tests/mgcp/mgcp_test.c:
Patch Set #13, Line 1315: atomic_uint_least64_t test_ctr_in;
no need to be atomic?
To view, visit change 26190. To unsubscribe, or for help writing mail filters, visit settings.