2 Copyright 2018 Evgeny Naumov
4 Permission is hereby granted, free of charge, to any person obtaining a copy of
5 this software and associated documentation files (the "Software"), to deal in
6 the Software without restriction, including without limitation the rights to
7 use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
8 of the Software, and to permit persons to whom the Software is furnished to do
9 so, subject to the following conditions:
11 The above copyright notice and this permission notice shall be included in all
12 copies or substantial portions of the Software.
14 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
17 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
19 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 #include "daemon/collectd.h"
24 #include "daemon/common.h"
25 #include "daemon/plugin.h"
31 #define MAX_DEVNAME_LEN 256
32 #define PLUGIN_NAME "gpu_nvidia"
34 static nvmlReturn_t nv_status = NVML_SUCCESS;
35 static char *nv_errline = "";
37 #define TRY_CATCH(f, catch) \
38 if ((nv_status = f) != NVML_SUCCESS) { \
43 #define TRY_CATCH_OPTIONAL(f, catch) \
44 if ((nv_status = f) != NVML_SUCCESS && \
45 nv_status != NVML_ERROR_NOT_SUPPORTED) { \
50 #define TRY(f) TRY_CATCH(f, catch)
51 #define TRYOPT(f) TRY_CATCH_OPTIONAL(f, catch)
53 #define KEY_GPUINDEX "GPUIndex"
54 #define KEY_IGNORESELECTED "IgnoreSelected"
56 static const char *config_keys[] = {
57 KEY_GPUINDEX, KEY_IGNORESELECTED,
59 static const unsigned int n_config_keys = STATIC_ARRAY_SIZE(config_keys);
61 // This is a bitflag, necessitating the (extremely conservative) assumption
62 // that there are no more than 64 GPUs on this system.
63 static uint64_t conf_match_mask = 0;
64 static bool conf_mask_is_exclude = 0;
66 static int nvml_config(const char *key, const char *value) {
68 if (strcasecmp(key, KEY_GPUINDEX) == 0) {
70 unsigned long device_ix = strtoul(value, &eptr, 10);
72 ERROR(PLUGIN_NAME ": Failed to parse GPUIndex value \"%s\"", value);
75 if (device_ix >= 64) {
77 ": At most 64 GPUs (0 <= GPUIndex < 64) are supported!");
80 conf_match_mask |= (1 << device_ix);
81 } else if (strcasecmp(key, KEY_IGNORESELECTED)) {
82 conf_mask_is_exclude = IS_TRUE(value);
84 ERROR(PLUGIN_NAME ": Unrecognized config option %s", key);
91 static int nvml_init(void) {
95 catch : ERROR(PLUGIN_NAME ": NVML init failed with %d", nv_status);
99 static int nvml_shutdown(void) {
103 catch : ERROR(PLUGIN_NAME ": NVML shutdown failed with %d", nv_status);
107 static void nvml_submit_gauge(const char *plugin_instance, const char *type,
108 const char *type_instance, gauge_t nvml) {
110 value_list_t vl = VALUE_LIST_INIT;
112 vl.values = &(value_t){.gauge = nvml};
115 sstrncpy(vl.plugin, PLUGIN_NAME, sizeof(vl.plugin));
116 sstrncpy(vl.plugin_instance, plugin_instance, sizeof(vl.plugin_instance));
118 sstrncpy(vl.type, type, sizeof(vl.type));
120 if (type_instance != NULL) {
121 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
124 plugin_dispatch_values(&vl);
127 static int nvml_read(void) {
129 unsigned int device_count;
130 TRY_CATCH(nvmlDeviceGetCount(&device_count), catch_nocount);
132 if (device_count > 64) {
136 for (unsigned int ix = 0; ix < device_count; ix++) {
138 unsigned int is_match =
139 ((1 << ix) & conf_match_mask) || (conf_match_mask == 0);
140 if (conf_mask_is_exclude == !!is_match) {
145 TRY(nvmlDeviceGetHandleByIndex(ix, &dev));
147 char dev_name[MAX_DEVNAME_LEN + 1] = {0};
148 TRY(nvmlDeviceGetName(dev, dev_name, sizeof(dev_name) - 1));
150 // Try to be as lenient as possible with the variety of devices that are
151 // out there, ignoring any NOT_SUPPORTED errors gently.
152 nvmlMemory_t meminfo;
153 TRYOPT(nvmlDeviceGetMemoryInfo(dev, &meminfo))
154 if (nv_status == NVML_SUCCESS) {
155 nvml_submit_gauge(dev_name, "memory", "used", meminfo.used);
156 nvml_submit_gauge(dev_name, "memory", "free", meminfo.free);
159 nvmlUtilization_t utilization;
160 TRYOPT(nvmlDeviceGetUtilizationRates(dev, &utilization))
161 if (nv_status == NVML_SUCCESS)
162 nvml_submit_gauge(dev_name, "percent", "gpu_used", utilization.gpu);
164 unsigned int fan_speed;
165 TRYOPT(nvmlDeviceGetFanSpeed(dev, &fan_speed))
166 if (nv_status == NVML_SUCCESS)
167 nvml_submit_gauge(dev_name, "fanspeed", NULL, fan_speed);
169 unsigned int core_temp;
170 TRYOPT(nvmlDeviceGetTemperature(dev, NVML_TEMPERATURE_GPU, &core_temp))
171 if (nv_status == NVML_SUCCESS)
172 nvml_submit_gauge(dev_name, "temperature", "core", core_temp);
174 unsigned int sm_clk_mhz;
175 TRYOPT(nvmlDeviceGetClockInfo(dev, NVML_CLOCK_SM, &sm_clk_mhz))
176 if (nv_status == NVML_SUCCESS)
177 nvml_submit_gauge(dev_name, "frequency", "multiprocessor",
180 unsigned int mem_clk_mhz;
181 TRYOPT(nvmlDeviceGetClockInfo(dev, NVML_CLOCK_MEM, &mem_clk_mhz))
182 if (nv_status == NVML_SUCCESS)
183 nvml_submit_gauge(dev_name, "frequency", "memory", 1e6 * mem_clk_mhz);
185 unsigned int power_mW;
186 TRYOPT(nvmlDeviceGetPowerUsage(dev, &power_mW))
187 if (nv_status == NVML_SUCCESS)
188 nvml_submit_gauge(dev_name, "power", NULL, 1e-3 * power_mW);
192 // Failures here indicate transient errors or removal of GPU. In either
193 // case it will either be resolved or the GPU will no longer be enumerated
194 // the next time round.
195 catch : WARNING(PLUGIN_NAME
196 ": NVML call \"%s\" failed (%d) on dev at index %d!",
197 nv_errline, nv_status, ix);
203 // Failures here indicate serious misconfiguration; we bail out totally.
205 ERROR(PLUGIN_NAME ": Failed to enumerate NVIDIA GPUs (\"%s\" returned %d)",
206 nv_errline, nv_status);
210 void module_register(void) {
211 plugin_register_init(PLUGIN_NAME, nvml_init);
212 plugin_register_config(PLUGIN_NAME, nvml_config, config_keys, n_config_keys);
213 plugin_register_read(PLUGIN_NAME, nvml_read);
214 plugin_register_shutdown(PLUGIN_NAME, nvml_shutdown);