Files
osmo-trx/CommonLibs/trx_rate_ctr.cpp
Pau Espin Pedrol 1d165a043e Transceiver: Add several rate_ctr for rx error conditions
Since there's now a rate counter, we can drop log level for those events
which can be bursty and hence print lots of output in short periods of
time, which may affect performance. This way setting them to INFO it's
enough to avoid getting them in stderr unless explicitly configured by
the user (for instance to debug stuff), while still allowing a good
enough level to be enabled for other targets such as gsmtap.

Related: OS#4679
Change-Id: I000f7112e35ac68d3d922444f78468b1ea74cbba
2020-07-27 11:52:46 +02:00

408 lines
16 KiB
C++

/*
* Copyright (C) 2019 sysmocom - s.f.m.c. GmbH
* All Rights Reserved
*
* SPDX-License-Identifier: AGPL-3.0+
*
* Author: Pau Espin Pedrol <pespin@sysmocom.de>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
* See the COPYING file in the main directory for details.
*/
/*
* rate_ctr API uses several osmocom select loop features, and as a result,
* calls to it must be done through the main thread (the one running the osmocom
* loop in osmo-trx).
* Since read/write from/to SDR is done in separate threads (even read and write
* each use a different thread), we must use some sort of message passing system
* between main thread feeding rate_ctr structures and the Rx/Tx threads
* generating the events.
* The idea is that upon read/write issues, lower layers (SDR APIs) provide us with
* underrun/overrun/droppedPackets information, and in that case we pass that up
* the stack through signal <SS_DEVICE,S_DEVICE_COUNTER_CHANGE> with signal_cb
* being a pointer to a "struct device_counters" structure, which contains
* device (implementation agnostic) statful counters for different kind of
* statistics.
* That signal is processed here in device_sig_cb, where a copy of the "struct
* device_counters" structure is held and the main thread is instructed through
* a timerfd to update rate_ctr APIs against this copy. All this is done inside
* a mutex to avoid different race conditions (between Rx andTx threads, and
* between Rx/Tx and main thread). For the same reason, callers of signal
* <SS_DEVICE,S_DEVICE_COUNTER_CHANGE> (device_sig_cb), that is Rx/Tx threads,
* must do so with PTHREAD_CANCEL_DISABLE, in order to avoid possible deadlocks
* in case the main thread decides to cancel other threads due to a shutdown
* operation (fi SIGKILL received)
*/
#include <string.h>
#include <stdint.h>
#include <inttypes.h>
#include <netinet/in.h>
#include <arpa/inet.h>
extern "C" {
#include <osmocom/core/talloc.h>
#include <osmocom/core/utils.h>
#include <osmocom/core/rate_ctr.h>
#include <osmocom/core/select.h>
#include <osmocom/core/stats.h>
#include <osmocom/core/timer.h>
#include "osmo_signal.h"
#include "trx_vty.h"
#include "trx_rate_ctr.h"
}
#include "Threads.h"
#include "Logger.h"
/* Used in dev_ctrs_pending, when set it means that channel slot contains unused
(non-pending) counter data */
#define PENDING_CHAN_NONE SIZE_MAX
static void *trx_rate_ctr_ctx;
static struct rate_ctr_group** rate_ctrs;
static struct device_counters* dev_ctrs_pending;
static struct trx_counters* trx_ctrs_pending;
static size_t chan_len;
static struct osmo_fd dev_rate_ctr_timerfd;
static struct osmo_fd trx_rate_ctr_timerfd;
static Mutex dev_rate_ctr_mutex;
static Mutex trx_rate_ctr_mutex;
struct osmo_timer_list threshold_timer;
static LLIST_HEAD(threshold_list);
static unsigned int threshold_timer_sched_secs;
static bool threshold_initied;
const struct value_string rate_ctr_intv[] = {
{ RATE_CTR_INTV_SEC, "per-second" },
{ RATE_CTR_INTV_MIN, "per-minute" },
{ RATE_CTR_INTV_HOUR, "per-hour" },
{ RATE_CTR_INTV_DAY, "per-day" },
{ 0, NULL }
};
const struct value_string trx_chan_ctr_names[] = {
{ TRX_CTR_DEV_RX_OVERRUNS, "rx_overruns" },
{ TRX_CTR_DEV_TX_UNDERRUNS, "tx_underruns" },
{ TRX_CTR_DEV_RX_DROP_EV, "rx_drop_events" },
{ TRX_CTR_DEV_RX_DROP_SMPL, "rx_drop_samples" },
{ TRX_CTR_DEV_TX_DROP_EV, "tx_drop_events" },
{ TRX_CTR_DEV_TX_DROP_SMPL, "tx_drop_samples" },
{ TRX_CTR_TRX_TX_STALE_BURSTS, "tx_stale_bursts" },
{ TRX_CTR_TRX_TX_UNAVAILABLE_BURSTS, "tx_unavailable_bursts" },
{ TRX_CTR_TRX_TRXD_FN_REPEATED, "tx_trxd_fn_repeated" },
{ TRX_CTR_TRX_TRXD_FN_OUTOFORDER, "tx_trxd_fn_outoforder" },
{ TRX_CTR_TRX_TRXD_FN_SKIPPED, "tx_trxd_fn_skipped" },
{ TRX_CTR_TRX_RX_EMPTY_BURST, "rx_empty_burst" },
{ TRX_CTR_TRX_RX_CLIPPING, "rx_clipping" },
{ TRX_CTR_TRX_RX_NO_BURST_DETECTED, "rx_no_burst_detected" },
{ 0, NULL }
};
static const struct rate_ctr_desc trx_chan_ctr_desc[] = {
[TRX_CTR_DEV_RX_OVERRUNS] = { "device:rx_overruns", "Number of Rx overruns in FIFO queue" },
[TRX_CTR_DEV_TX_UNDERRUNS] = { "device:tx_underruns", "Number of Tx underruns in FIFO queue" },
[TRX_CTR_DEV_RX_DROP_EV] = { "device:rx_drop_events", "Number of times Rx samples were dropped by HW" },
[TRX_CTR_DEV_RX_DROP_SMPL] = { "device:rx_drop_samples", "Number of Rx samples dropped by HW" },
[TRX_CTR_DEV_TX_DROP_EV] = { "device:tx_drop_events", "Number of times Tx samples were dropped by HW" },
[TRX_CTR_DEV_TX_DROP_SMPL] = { "device:tx_drop_samples", "Number of Tx samples dropped by HW" },
[TRX_CTR_TRX_TX_STALE_BURSTS] = { "trx:tx_stale_bursts", "Number of Tx burts dropped by TRX due to arriving too late" },
[TRX_CTR_TRX_TX_UNAVAILABLE_BURSTS] = { "trx:tx_unavailable_bursts","Number of Tx burts unavailable (not enqueued) at the time they should be transmitted" },
[TRX_CTR_TRX_TRXD_FN_REPEATED] = { "trx:tx_trxd_fn_repeated", "Number of Tx burts received from TRXD with repeated FN" },
[TRX_CTR_TRX_TRXD_FN_OUTOFORDER] = { "trx:tx_trxd_fn_outoforder","Number of Tx burts received from TRXD with a past FN" },
[TRX_CTR_TRX_TRXD_FN_SKIPPED] = { "trx:tx_trxd_fn_skipped", "Number of Tx burts potentially skipped due to FN jumps" },
[TRX_CTR_TRX_RX_EMPTY_BURST] = { "trx:rx_empty_burst", "Number of Rx bursts empty" },
[TRX_CTR_TRX_RX_CLIPPING] = { "trx:rx_clipping", "Number of Rx bursts discarded due to clipping" },
[TRX_CTR_TRX_RX_NO_BURST_DETECTED] = { "trx:rx_no_burst_detected", "Number of Rx burts discarded due to burst detection error" },
};
static const struct rate_ctr_group_desc trx_chan_ctr_group_desc = {
.group_name_prefix = "trx:chan",
.group_description = "osmo-trx statistics",
.class_id = OSMO_STATS_CLASS_GLOBAL,
.num_ctr = ARRAY_SIZE(trx_chan_ctr_desc),
.ctr_desc = trx_chan_ctr_desc,
};
static int dev_rate_ctr_timerfd_cb(struct osmo_fd *ofd, unsigned int what) {
size_t chan;
struct rate_ctr *ctr;
LOGC(DCTR, INFO) << "Main thread is updating Device counters";
dev_rate_ctr_mutex.lock();
for (chan = 0; chan < chan_len; chan++) {
if (dev_ctrs_pending[chan].chan == PENDING_CHAN_NONE)
continue;
LOGCHAN(chan, DCTR, DEBUG) << "rate_ctr update";
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_RX_OVERRUNS];
rate_ctr_add(ctr, dev_ctrs_pending[chan].rx_overruns - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_TX_UNDERRUNS];
rate_ctr_add(ctr, dev_ctrs_pending[chan].tx_underruns - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_RX_DROP_EV];
rate_ctr_add(ctr, dev_ctrs_pending[chan].rx_dropped_events - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_RX_DROP_SMPL];
rate_ctr_add(ctr, dev_ctrs_pending[chan].rx_dropped_samples - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_TX_DROP_EV];
rate_ctr_add(ctr, dev_ctrs_pending[chan].tx_dropped_events - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_DEV_TX_DROP_SMPL];
rate_ctr_add(ctr, dev_ctrs_pending[chan].tx_dropped_samples - ctr->current);
/* Mark as done */
dev_ctrs_pending[chan].chan = PENDING_CHAN_NONE;
}
if (osmo_timerfd_disable(&dev_rate_ctr_timerfd) < 0)
LOGC(DCTR, ERROR) << "Failed to disable timerfd";
dev_rate_ctr_mutex.unlock();
return 0;
}
static int trx_rate_ctr_timerfd_cb(struct osmo_fd *ofd, unsigned int what) {
size_t chan;
struct rate_ctr *ctr;
LOGC(DCTR, INFO) << "Main thread is updating Transceiver counters";
trx_rate_ctr_mutex.lock();
for (chan = 0; chan < chan_len; chan++) {
if (trx_ctrs_pending[chan].chan == PENDING_CHAN_NONE)
continue;
LOGCHAN(chan, DCTR, DEBUG) << "rate_ctr update";
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_TX_STALE_BURSTS];
rate_ctr_add(ctr, trx_ctrs_pending[chan].tx_stale_bursts - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_TX_UNAVAILABLE_BURSTS];
rate_ctr_add(ctr, trx_ctrs_pending[chan].tx_unavailable_bursts - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_TRXD_FN_REPEATED];
rate_ctr_add(ctr, trx_ctrs_pending[chan].tx_trxd_fn_repeated - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_TRXD_FN_OUTOFORDER];
rate_ctr_add(ctr, trx_ctrs_pending[chan].tx_trxd_fn_outoforder - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_TRXD_FN_SKIPPED];
rate_ctr_add(ctr, trx_ctrs_pending[chan].tx_trxd_fn_skipped - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_RX_EMPTY_BURST];
rate_ctr_add(ctr, trx_ctrs_pending[chan].rx_empty_burst - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_RX_CLIPPING];
rate_ctr_add(ctr, trx_ctrs_pending[chan].rx_clipping - ctr->current);
ctr = &rate_ctrs[chan]->ctr[TRX_CTR_TRX_RX_NO_BURST_DETECTED];
rate_ctr_add(ctr, trx_ctrs_pending[chan].rx_no_burst_detected - ctr->current);
/* Mark as done */
trx_ctrs_pending[chan].chan = PENDING_CHAN_NONE;
}
if (osmo_timerfd_disable(&trx_rate_ctr_timerfd) < 0)
LOGC(DCTR, ERROR) << "Failed to disable timerfd";
trx_rate_ctr_mutex.unlock();
return 0;
}
/* Callback function to be called every time we receive a signal from DEVICE */
static int device_sig_cb(unsigned int subsys, unsigned int signal,
void *handler_data, void *signal_data)
{
struct device_counters *dev_ctr;
struct trx_counters *trx_ctr;
/* Delay sched around 20 ms, in case we receive several calls from several
* channels batched */
struct timespec next_sched = {.tv_sec = 0, .tv_nsec = 20*1000*1000};
/* no automatic re-trigger */
struct timespec intv_sched = {.tv_sec = 0, .tv_nsec = 0};
switch (signal) {
case S_DEVICE_COUNTER_CHANGE:
dev_ctr = (struct device_counters *)signal_data;
LOGCHAN(dev_ctr->chan, DCTR, INFO) << "Received counter change from radioDevice";
dev_rate_ctr_mutex.lock();
dev_ctrs_pending[dev_ctr->chan] = *dev_ctr;
if (osmo_timerfd_schedule(&dev_rate_ctr_timerfd, &next_sched, &intv_sched) < 0) {
LOGC(DCTR, ERROR) << "Failed to schedule timerfd: " << errno << " = "<< strerror(errno);
}
dev_rate_ctr_mutex.unlock();
break;
case S_TRX_COUNTER_CHANGE:
trx_ctr = (struct trx_counters *)signal_data;
LOGCHAN(trx_ctr->chan, DCTR, INFO) << "Received counter change from Transceiver";
trx_rate_ctr_mutex.lock();
trx_ctrs_pending[trx_ctr->chan] = *trx_ctr;
if (osmo_timerfd_schedule(&trx_rate_ctr_timerfd, &next_sched, &intv_sched) < 0) {
LOGC(DCTR, ERROR) << "Failed to schedule timerfd: " << errno << " = "<< strerror(errno);
}
trx_rate_ctr_mutex.unlock();
break;
default:
break;
}
return 0;
}
/************************************
* ctr_threshold APIs
************************************/
static const char* ctr_threshold_2_vty_str(struct ctr_threshold *ctr)
{
static char buf[256];
int rc = 0;
rc += snprintf(buf, sizeof(buf), "ctr-error-threshold %s", get_value_string(trx_chan_ctr_names, ctr->ctr_id));
rc += snprintf(buf + rc, sizeof(buf) - rc, " %d %s", ctr->val, get_value_string(rate_ctr_intv, ctr->intv));
return buf;
}
static void threshold_timer_cb(void *data)
{
struct ctr_threshold *ctr_thr;
struct rate_ctr *rate_ctr;
size_t chan;
LOGC(DCTR, DEBUG) << "threshold_timer_cb fired!";
llist_for_each_entry(ctr_thr, &threshold_list, list) {
for (chan = 0; chan < chan_len; chan++) {
rate_ctr = &rate_ctrs[chan]->ctr[ctr_thr->ctr_id];
LOGCHAN(chan, DCTR, INFO) << "checking threshold: " << ctr_threshold_2_vty_str(ctr_thr)
<< " ("<< rate_ctr->intv[ctr_thr->intv].rate << " vs " << ctr_thr->val << ")";
if (rate_ctr->intv[ctr_thr->intv].rate >= ctr_thr->val) {
LOGCHAN(chan, DCTR, FATAL) << "threshold reached, stopping! " << ctr_threshold_2_vty_str(ctr_thr)
<< " ("<< rate_ctr->intv[ctr_thr->intv].rate << " vs " << ctr_thr->val << ")";
osmo_signal_dispatch(SS_MAIN, S_MAIN_STOP_REQUIRED, NULL);
return;
}
}
}
osmo_timer_schedule(&threshold_timer, threshold_timer_sched_secs, 0);
}
static size_t ctr_threshold_2_seconds(struct ctr_threshold *ctr)
{
size_t mult = 0;
switch (ctr->intv) {
case RATE_CTR_INTV_SEC:
mult = 1;
break;
case RATE_CTR_INTV_MIN:
mult = 60;
break;
case RATE_CTR_INTV_HOUR:
mult = 60*60;
break;
case RATE_CTR_INTV_DAY:
mult = 60*60*24;
break;
default:
OSMO_ASSERT(false);
}
return mult;
}
static void threshold_timer_update_intv() {
struct ctr_threshold *ctr, *min_ctr;
size_t secs, min_secs;
/* Avoid scheduling timer until itself and other structures are prepared
by trx_rate_ctr_init */
if (!threshold_initied)
return;
if (llist_empty(&threshold_list)) {
if (osmo_timer_pending(&threshold_timer))
osmo_timer_del(&threshold_timer);
return;
}
min_ctr = llist_first_entry(&threshold_list, struct ctr_threshold, list);
min_secs = ctr_threshold_2_seconds(min_ctr);
llist_for_each_entry(ctr, &threshold_list, list) {
secs = ctr_threshold_2_seconds(ctr);
if (min_secs > secs)
min_secs = secs;
}
threshold_timer_sched_secs = OSMO_MAX((int)(min_secs / 2 - 1), 1);
LOGC(DCTR, INFO) << "New ctr-error-threshold check interval: "
<< threshold_timer_sched_secs << " seconds";
osmo_timer_schedule(&threshold_timer, threshold_timer_sched_secs, 0);
}
/* Init rate_ctr subsystem. Expected to be called during process start by main thread before VTY is ready */
void trx_rate_ctr_init(void *ctx, struct trx_ctx* trx_ctx)
{
size_t i;
trx_rate_ctr_ctx = ctx;
chan_len = trx_ctx->cfg.num_chans;
dev_ctrs_pending = (struct device_counters*) talloc_zero_size(ctx, chan_len * sizeof(struct device_counters));
trx_ctrs_pending = (struct trx_counters*) talloc_zero_size(ctx, chan_len * sizeof(struct trx_counters));
rate_ctrs = (struct rate_ctr_group**) talloc_zero_size(ctx, chan_len * sizeof(struct rate_ctr_group*));
for (i = 0; i < chan_len; i++) {
dev_ctrs_pending[i].chan = PENDING_CHAN_NONE;
trx_ctrs_pending[i].chan = PENDING_CHAN_NONE;
rate_ctrs[i] = rate_ctr_group_alloc(ctx, &trx_chan_ctr_group_desc, i);
if (!rate_ctrs[i]) {
LOGCHAN(i, DCTR, ERROR) << "Failed to allocate rate ctr";
exit(1);
}
}
dev_rate_ctr_timerfd.fd = -1;
if (osmo_timerfd_setup(&dev_rate_ctr_timerfd, dev_rate_ctr_timerfd_cb, NULL) < 0) {
LOGC(DCTR, ERROR) << "Failed to setup timerfd";
exit(1);
}
trx_rate_ctr_timerfd.fd = -1;
if (osmo_timerfd_setup(&trx_rate_ctr_timerfd, trx_rate_ctr_timerfd_cb, NULL) < 0) {
LOGC(DCTR, ERROR) << "Failed to setup timerfd";
exit(1);
}
osmo_signal_register_handler(SS_DEVICE, device_sig_cb, NULL);
/* Now set up threshold checks */
threshold_initied = true;
osmo_timer_setup(&threshold_timer, threshold_timer_cb, NULL);
threshold_timer_update_intv();
}
void trx_rate_ctr_threshold_add(struct ctr_threshold *ctr)
{
struct ctr_threshold *new_ctr;
new_ctr = talloc_zero(trx_rate_ctr_ctx, struct ctr_threshold);
*new_ctr = *ctr;
LOGC(DCTR, NOTICE) << "Adding new threshold check: " << ctr_threshold_2_vty_str(new_ctr);
llist_add(&new_ctr->list, &threshold_list);
threshold_timer_update_intv();
}
int trx_rate_ctr_threshold_del(struct ctr_threshold *del_ctr)
{
struct ctr_threshold *ctr;
llist_for_each_entry(ctr, &threshold_list, list) {
if (ctr->intv != del_ctr->intv ||
ctr->ctr_id != del_ctr->ctr_id ||
ctr->val != del_ctr->val)
continue;
LOGC(DCTR, NOTICE) << "Deleting threshold check: " << ctr_threshold_2_vty_str(del_ctr);
llist_del(&ctr->list);
talloc_free(ctr);
threshold_timer_update_intv();
return 0;
}
return -1;
}
void trx_rate_ctr_threshold_write_config(struct vty *vty, char *indent_prefix)
{
struct ctr_threshold *ctr;
llist_for_each_entry(ctr, &threshold_list, list) {
vty_out(vty, "%s%s%s", indent_prefix, ctr_threshold_2_vty_str(ctr), VTY_NEWLINE);
}
}