Skip to content

Commit

Permalink
Merge pull request #4296 from aleksinx/intel_rdt_llc_miss
Browse files Browse the repository at this point in the history 
[intel_rdt] Added support for LLC misses monitoring event
  • Loading branch information
@eero-t
eero-t committed May 14, 2024
2 parents 4d29f1d + 211de8d commit e817577
Show file tree
Hide file tree
Showing 3 changed files with 114 additions and 47 deletions.
1 change: 1 addition & 0 deletions src/collectd.conf.in
View file
Original file line number Diff line number Diff line change
Expand Up @@ -849,6 +849,7 @@
#<Plugin "intel_rdt">
# MonIPCEnabled true
# MonLLCRefEnabled false
# MonLLCMissEnabled false
# Cores "0-2"
# Processes "sshd"
#</Plugin>
Expand Down
7 changes: 7 additions & 0 deletions src/collectd.conf.pod
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4033,6 +4033,7 @@ B<Synopsis:>
<Plugin "intel_rdt">
MonIPCEnabled true
MonLLCRefEnabled false
MonLLCMissEnabled false
Cores "0-2" "3,4,6" "8-10,15"
Processes "sshd,qemu-system-x86" "bash"
</Plugin>
Expand All @@ -4059,6 +4060,12 @@ Determines whether or not to enable LLC references monitoring. If set to
B<false> (the default), LLC references monitoring statistics will not be
collected by intel_rdt plugin.

=item B<MonLLCMissEnabled> B<true>|B<false>

Determines whether or not to enable LLC misses monitoring. If set to B<false>
(the default), LLC misses monitoring statistics will not be collected by
intel_rdt plugin.

=item B<Cores> I<cores groups>

Monitoring of the events can be configured for group of cores
Expand Down
153 changes: 106 additions & 47 deletions src/intel_rdt.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -44,14 +44,14 @@

#define RDT_PLUGIN "intel_rdt"

#if PQOS_VERSION >= 40400
#define RDT_EVENTS \
#define RDT_EVENTS_BASE \
(PQOS_MON_EVENT_L3_OCCUP | PQOS_PERF_EVENT_IPC | PQOS_MON_EVENT_LMEM_BW | \
PQOS_MON_EVENT_TMEM_BW | PQOS_MON_EVENT_RMEM_BW | PQOS_PERF_EVENT_LLC_REF)
PQOS_MON_EVENT_TMEM_BW | PQOS_MON_EVENT_RMEM_BW | PQOS_PERF_EVENT_LLC_MISS)

#if PQOS_VERSION >= 40400
#define RDT_EVENTS (RDT_EVENTS_BASE | PQOS_PERF_EVENT_LLC_REF)
#else
#define RDT_EVENTS \
(PQOS_MON_EVENT_L3_OCCUP | PQOS_PERF_EVENT_IPC | PQOS_MON_EVENT_LMEM_BW | \
PQOS_MON_EVENT_TMEM_BW | PQOS_MON_EVENT_RMEM_BW)
#define RDT_EVENTS RDT_EVENTS_BASE
#endif

#define RDT_MAX_SOCKETS 8
Expand Down Expand Up @@ -89,6 +89,7 @@ struct rdt_ctx_s {
#if PQOS_VERSION >= 40400
bool mon_llc_ref_enabled;
#endif
bool mon_llc_miss_enabled;
core_groups_list_t cores;
enum pqos_mon_event events[RDT_MAX_CORES];
struct pqos_mon_data *pcgroups[RDT_MAX_CORES];
Expand Down Expand Up @@ -143,75 +144,127 @@ static void rdt_submit_gauge(const char *cgroup, const char *type,
plugin_dispatch_values(&vl);
}

static void rdt_submit(const struct pqos_mon_data *group) {
static int rdt_get_value(const struct pqos_mon_data *group,
const enum pqos_mon_event event_id, uint64_t *value) {
int ret;

#if PQOS_VERSION >= 40400
ret = pqos_mon_get_value(group, PQOS_PERF_EVENT_LLC_REF, value, NULL);

#else
const struct pqos_event_values *values = &group->values;

ret = PQOS_RETVAL_OK;

switch (event_id) {
case PQOS_MON_EVENT_L3_OCCUP:
*value = values->llc;
break;
case PQOS_PERF_EVENT_LLC_MISS:
*value = values->llc_misses;
break;
case PQOS_MON_EVENT_LMEM_BW:
*value = values->mbm_local;
break;
case PQOS_MON_EVENT_TMEM_BW:
*value = values->mbm_total;
break;
case PQOS_MON_EVENT_RMEM_BW:
#if PQOS_VERSION < 40000
if (group->event & (PQOS_MON_EVENT_TMEM_BW | PQOS_MON_EVENT_LMEM_BW))
*value = values->mbm_total - values->mbm_local;
#else
*value = values->mbm_remote;
#endif
default:
ret = PQOS_RETVAL_ERROR;
break;
}
#endif

if (ret != PQOS_RETVAL_OK)
return ret;

// Apply scale factor
if (event_id == PQOS_MON_EVENT_LMEM_BW ||
event_id == PQOS_MON_EVENT_TMEM_BW ||
event_id == PQOS_MON_EVENT_RMEM_BW) {
const struct pqos_monitor *mon = NULL;

ret = pqos_cap_get_event(g_rdt->pqos_cap, event_id, &mon);
if (ret == PQOS_RETVAL_OK && mon->scale_factor != 0)
*value = *value * mon->scale_factor;
}

return ret;
}

static void rdt_submit(const struct pqos_mon_data *group) {
const char *desc = (const char *)group->context;
const enum pqos_mon_event events = group->event;

if (events & PQOS_MON_EVENT_L3_OCCUP)
rdt_submit_gauge(desc, "bytes", "llc", values->llc);
if (events & PQOS_MON_EVENT_L3_OCCUP) {
uint64_t value;

int ret = rdt_get_value(group, PQOS_MON_EVENT_L3_OCCUP, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_gauge(desc, "bytes", "llc", value);
}

if (events & PQOS_PERF_EVENT_IPC) {
double value;
int ret = PQOS_RETVAL_OK;

#if PQOS_VERSION >= 40400
ret = pqos_mon_get_ipc(group, &value);
#else
value = group->values.ipc;
#endif

if (events & PQOS_PERF_EVENT_IPC)
rdt_submit_gauge(desc, "ipc", NULL, values->ipc);
if (ret == PQOS_RETVAL_OK)
rdt_submit_gauge(desc, "ipc", NULL, value);
}

#if PQOS_VERSION >= 40400
if (events & PQOS_PERF_EVENT_LLC_REF) {
uint64_t value;

int ret = pqos_mon_get_value(group, PQOS_PERF_EVENT_LLC_REF, &value, NULL);
int ret = rdt_get_value(group, PQOS_PERF_EVENT_LLC_REF, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_gauge(desc, "bytes", "llc_ref", value);
rdt_submit_derive(desc, "bytes", "llc_ref", value);
}
#endif

if (events & PQOS_MON_EVENT_LMEM_BW) {
const struct pqos_monitor *mon = NULL;
if (events & PQOS_PERF_EVENT_LLC_MISS) {
uint64_t value;

int retval =
pqos_cap_get_event(g_rdt->pqos_cap, PQOS_MON_EVENT_LMEM_BW, &mon);
if (retval == PQOS_RETVAL_OK) {
uint64_t value = values->mbm_local;
int ret = rdt_get_value(group, PQOS_PERF_EVENT_LLC_MISS, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_derive(desc, "bytes", "llc_miss", value);
}

if (mon->scale_factor != 0)
value = value * mon->scale_factor;
if (events & PQOS_MON_EVENT_LMEM_BW) {
uint64_t value;

int ret = rdt_get_value(group, PQOS_MON_EVENT_LMEM_BW, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_derive(desc, "memory_bandwidth", "local", value);
}
}

if (events & PQOS_MON_EVENT_TMEM_BW) {
const struct pqos_monitor *mon = NULL;

int retval =
pqos_cap_get_event(g_rdt->pqos_cap, PQOS_MON_EVENT_TMEM_BW, &mon);
if (retval == PQOS_RETVAL_OK) {
uint64_t value = values->mbm_total;

if (mon->scale_factor != 0)
value = value * mon->scale_factor;
uint64_t value;

int ret = rdt_get_value(group, PQOS_MON_EVENT_TMEM_BW, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_derive(desc, "memory_bandwidth", "total", value);
}
}

if (events & PQOS_MON_EVENT_RMEM_BW) {
const struct pqos_monitor *mon = NULL;

int retval =
pqos_cap_get_event(g_rdt->pqos_cap, PQOS_MON_EVENT_RMEM_BW, &mon);
if (retval == PQOS_RETVAL_OK) {
uint64_t value = values->mbm_remote;

#if PQOS_VERSION < 40000
if (events & (PQOS_MON_EVENT_TMEM_BW | PQOS_MON_EVENT_LMEM_BW))
value = values->mbm_total - values->mbm_local;
#endif

if (mon->scale_factor != 0)
value = value * mon->scale_factor;
uint64_t value;

int ret = rdt_get_value(group, PQOS_MON_EVENT_RMEM_BW, &value);
if (ret == PQOS_RETVAL_OK)
rdt_submit_derive(desc, "memory_bandwidth", "remote", value);
}
}
}

Expand Down Expand Up @@ -586,6 +639,10 @@ static int rdt_config_events(rdt_ctx_t *rdt) {
events &= ~(PQOS_PERF_EVENT_LLC_REF);
#endif

/* LLC misses monitoring is disabled */
if (!rdt->mon_llc_miss_enabled)
events &= ~(PQOS_PERF_EVENT_LLC_MISS);

DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);

for (size_t i = 0; i < rdt->cores.num_cgroups; i++) {
Expand Down Expand Up @@ -1269,6 +1326,8 @@ static int rdt_config(oconfig_item_t *ci) {
} else if (strcasecmp("MonLLCRefEnabled", child->key) == 0) {
cf_util_get_boolean(child, &g_rdt->mon_llc_ref_enabled);
#endif
} else if (strcasecmp("MonLLCMissEnabled", child->key) == 0) {
cf_util_get_boolean(child, &g_rdt->mon_llc_miss_enabled);
} else {
ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
}
Expand Down

0 comments on commit e817577

Please sign in to comment.