summaryrefslogtreecommitdiffstats
path: root/src/stateless/rx
diff options
context:
space:
mode:
authorIdo Barnea <ibarnea@cisco.com>2016-05-17 12:16:36 +0300
committerIdo Barnea <ibarnea@cisco.com>2016-05-18 19:23:48 +0300
commitf8d4267c09469019a275e522e0d2493720b5f9a1 (patch)
tree04bf92d3bfc03a02ae68c9d921b77e516302ccdf /src/stateless/rx
parent3e024d8bdcebdd3100851f28b5724a2ecbfc923a (diff)
more error counter
Diffstat (limited to 'src/stateless/rx')
-rw-r--r--src/stateless/rx/trex_stateless_rx_core.cpp82
-rw-r--r--src/stateless/rx/trex_stateless_rx_core.h20
2 files changed, 59 insertions, 43 deletions
diff --git a/src/stateless/rx/trex_stateless_rx_core.cpp b/src/stateless/rx/trex_stateless_rx_core.cpp
index 4a7c0b29..463fa7d3 100644
--- a/src/stateless/rx/trex_stateless_rx_core.cpp
+++ b/src/stateless/rx/trex_stateless_rx_core.cpp
@@ -32,12 +32,14 @@ void CRxCoreStateless::create(const CRxSlCfg &cfg) {
for (int i = 0; i < MAX_FLOW_STATS_PAYLOAD; i++) {
// This is the seq num value we expect next packet to have.
// Init value should match m_seq_num in CVirtualIFPerSideStats
- m_per_flow_seq[i] = UINT32_MAX - 1; // catch wrap around issues early
- m_per_flow_hist[i].Create();
- m_per_flow_jitter[i].reset();
- m_per_flow_seq_error[i] = 0;
- m_per_flow_out_of_order[i] = 0;
- m_per_flow_dup[i] = 0;
+ m_rfc2544[i].seq = UINT32_MAX - 1; // catch wrap around issues early
+ m_rfc2544[i].latency.Create();
+ m_rfc2544[i].jitter.reset();
+ m_rfc2544[i].seq_err = 0;
+ m_rfc2544[i].seq_err_events_too_big = 0;
+ m_rfc2544[i].seq_err_events_too_low = 0;
+ m_rfc2544[i].out_of_order = 0;
+ m_rfc2544[i].dup = 0;
}
}
@@ -138,47 +140,51 @@ void CRxCoreStateless::handle_rx_pkt(CLatencyManagerPerPortStl *lp, rte_mbuf_t *
if (fsp_head->magic == FLOW_STAT_PAYLOAD_MAGIC) {
hw_id = fsp_head->hw_id;
seq = fsp_head->seq;
- if (unlikely(seq != m_per_flow_seq[hw_id])) {
- if (seq < m_per_flow_seq[hw_id]) {
- if (m_per_flow_seq[hw_id] - seq > 100000) {
+ if (unlikely(seq != m_rfc2544[hw_id].seq)) {
+ if (seq < m_rfc2544[hw_id].seq) {
+ if (m_rfc2544[hw_id].seq - seq > 100000) {
// packet loss while we had wrap around
- m_per_flow_seq_error[hw_id] += seq - m_per_flow_seq[hw_id];
- m_per_flow_seq[hw_id] = seq + 1;
+ m_rfc2544[hw_id].seq_err += seq - m_rfc2544[hw_id].seq;
+ m_rfc2544[hw_id].seq_err_events_too_big++;
+ m_rfc2544[hw_id].seq = seq + 1;
} else {
- if (seq == (m_per_flow_seq[hw_id] - 1)) {
- m_per_flow_dup[hw_id] += 1;
+ if (seq == (m_rfc2544[hw_id].seq - 1)) {
+ m_rfc2544[hw_id].dup += 1;
} else {
- m_per_flow_out_of_order[hw_id] += 1;
+ m_rfc2544[hw_id].out_of_order += 1;
// We thought it was lost, but it was just out of order
- m_per_flow_seq_error[hw_id] -= 1;
+ m_rfc2544[hw_id].seq_err -= 1;
+ m_rfc2544[hw_id].seq_err_events_too_low++;
}
}
} else {
- if (unlikely (m_per_flow_seq[hw_id] - seq > 100000)) {
+ if (unlikely (m_rfc2544[hw_id].seq - seq > 100000)) {
// packet reorder while we had wrap around
- if (seq == (m_per_flow_seq[hw_id] - 1)) {
- m_per_flow_dup[hw_id] += 1;
+ if (seq == (m_rfc2544[hw_id].seq - 1)) {
+ m_rfc2544[hw_id].dup += 1;
} else {
- m_per_flow_out_of_order[hw_id] += 1;
+ m_rfc2544[hw_id].out_of_order += 1;
// We thought it was lost, but it was just out of order
- m_per_flow_seq_error[hw_id] -= 1;
+ m_rfc2544[hw_id].seq_err -= 1;
+ m_rfc2544[hw_id].seq_err_events_too_low++;
}
} else {
- // seq > m_per_flow_seq[hw_id]. Assuming lost packets
- m_per_flow_seq_error[hw_id] += seq - m_per_flow_seq[hw_id];
- m_per_flow_seq[hw_id] = seq + 1;
+ // seq > m_rfc2544[hw_id].seq. Assuming lost packets
+ m_rfc2544[hw_id].seq_err += seq - m_rfc2544[hw_id].seq;
+ m_rfc2544[hw_id].seq_err_events_too_big++;
+ m_rfc2544[hw_id].seq = seq + 1;
}
}
} else {
- m_per_flow_seq[hw_id] = seq + 1;
+ m_rfc2544[hw_id].seq = seq + 1;
}
lp->m_port.m_rx_pg_stat_payload[hw_id].add_pkts(1);
lp->m_port.m_rx_pg_stat_payload[hw_id].add_bytes(m->pkt_len);
uint64_t d = (os_get_hr_tick_64() - fsp_head->time_stamp );
dsec_t ctime = ptime_convert_hr_dsec(d);
- m_per_flow_hist[hw_id].Add(ctime);
- m_per_flow_last_max[hw_id].update(ctime);
- m_per_flow_jitter[hw_id].calc(ctime);
+ m_rfc2544[hw_id].latency.Add(ctime);
+ m_rfc2544[hw_id].last_max.update(ctime);
+ m_rfc2544[hw_id].jitter.calc(ctime);
}
} else {
hw_id = get_hw_id(ip_id);
@@ -331,20 +337,24 @@ int CRxCoreStateless::get_rfc2544_info(rfc2544_info_t *rfc2544_info, int min, in
Json::Reader reader;
for (int hw_id = min; hw_id <= max; hw_id++) {
- rfc2544_info[hw_id - min].set_err_cntrs(m_per_flow_seq_error[hw_id], m_per_flow_out_of_order[hw_id]);
- rfc2544_info[hw_id - min].set_jitter(m_per_flow_jitter[hw_id].get_jitter());
- m_per_flow_hist[hw_id].update();
- m_per_flow_hist[hw_id].dump_json("", json_str);
+ rfc2544_info[hw_id - min].set_err_cntrs(m_rfc2544[hw_id].seq_err, m_rfc2544[hw_id].out_of_order
+ , m_rfc2544[hw_id].dup, m_rfc2544[hw_id].seq_err_events_too_big
+ , m_rfc2544[hw_id].seq_err_events_too_low);
+ rfc2544_info[hw_id - min].set_jitter(m_rfc2544[hw_id].jitter.get_jitter());
+ m_rfc2544[hw_id].latency.update();
+ m_rfc2544[hw_id].latency.dump_json("", json_str);
// This is a hack. We need to make the dump_json return json object.
reader.parse( json_str.c_str(), json);
rfc2544_info[hw_id - min].set_latency_json(json);
- rfc2544_info[hw_id - min].set_last_max(m_per_flow_last_max[hw_id].switchMax());
+ rfc2544_info[hw_id - min].set_last_max(m_rfc2544[hw_id].last_max.switchMax());
if (reset) {
- m_per_flow_seq_error[hw_id] = 0;
- m_per_flow_out_of_order[hw_id] = 0;
- m_per_flow_hist[hw_id].Reset();
- m_per_flow_jitter[hw_id].reset();
+ m_rfc2544[hw_id].seq_err = 0;
+ m_rfc2544[hw_id].seq_err_events_too_big = 0;
+ m_rfc2544[hw_id].seq_err_events_too_low = 0;
+ m_rfc2544[hw_id].out_of_order = 0;
+ m_rfc2544[hw_id].latency.Reset();
+ m_rfc2544[hw_id].jitter.reset();
}
}
return 0;
diff --git a/src/stateless/rx/trex_stateless_rx_core.h b/src/stateless/rx/trex_stateless_rx_core.h
index 604ad260..abe8e754 100644
--- a/src/stateless/rx/trex_stateless_rx_core.h
+++ b/src/stateless/rx/trex_stateless_rx_core.h
@@ -109,6 +109,18 @@ class CRxSlCfg {
CPortLatencyHWBase * m_ports[TREX_MAX_PORTS];
};
+struct per_flow_rfc2544_info {
+ uint32_t seq; // expected next seq num
+ CTimeHistogram latency; // latency info
+ CJitter jitter;
+ uint64_t seq_err; // How many packet seq num gaps we saw (packets lost or out of order)
+ uint64_t seq_err_events_too_big; // How many packet seq num greater than expected events we had
+ uint64_t seq_err_events_too_low; // How many packet seq num lower than expected events we had
+ uint64_t out_of_order; // Packets we got with seq num lower than expected
+ uint64_t dup; // Packets we got with same seq num
+ CLastMax last_max; // maximum for last measurement period (reset whenever we read it).
+};
+
class CRxCoreStateless {
enum state_e {
STATE_IDLE,
@@ -156,12 +168,6 @@ class CRxCoreStateless {
CCpuUtlCp m_cpu_cp_u;
// Used for acking "work" (go out of idle) messages from cp
volatile bool m_ack_start_work_msg __rte_cache_aligned;
- uint32_t m_per_flow_seq[MAX_FLOW_STATS_PAYLOAD]; // expected next seq num
- CTimeHistogram m_per_flow_hist[MAX_FLOW_STATS_PAYLOAD]; /* latency info */
- CJitter m_per_flow_jitter[MAX_FLOW_STATS_PAYLOAD];
- uint64_t m_per_flow_seq_error[MAX_FLOW_STATS_PAYLOAD]; // How many packet seq num gaps we saw (packets lost or out of order)
- uint64_t m_per_flow_out_of_order[MAX_FLOW_STATS_PAYLOAD]; // Packets we got with seq num lower than expected
- uint64_t m_per_flow_dup[MAX_FLOW_STATS_PAYLOAD]; // Packets we got with same seq num
- CLastMax m_per_flow_last_max[MAX_FLOW_STATS_PAYLOAD];
+ struct per_flow_rfc2544_info m_rfc2544[MAX_FLOW_STATS_PAYLOAD];
};
#endif