virperf.c 10.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 * virperf.c: methods for managing perf events
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library.  If not, see
 * <http://www.gnu.org/licenses/>.
 */
#include <config.h>

J
Ján Tomko 已提交
20
#include <unistd.h>
21 22 23
#ifndef WIN32
# include <sys/ioctl.h>
#endif
24 25 26 27 28 29 30 31 32 33
#if defined HAVE_SYS_SYSCALL_H
# include <sys/syscall.h>
#endif

#include "virperf.h"
#include "virerror.h"
#include "virlog.h"
#include "virfile.h"
#include "virstring.h"
#include "virtypedparam.h"
34
#include "viralloc.h"
35 36 37 38 39

VIR_LOG_INIT("util.perf");

#define VIR_FROM_THIS VIR_FROM_PERF

40 41
VIR_ENUM_IMPL(virPerfEvent,
              VIR_PERF_EVENT_LAST,
Q
Qiaowei Ren 已提交
42 43
              "cmt", "mbmt", "mbml",
              "cpu_cycles", "instructions",
44
              "cache_references", "cache_misses",
45
              "branch_instructions", "branch_misses",
46
              "bus_cycles", "stalled_cycles_frontend",
47
              "stalled_cycles_backend", "ref_cpu_cycles",
48
              "cpu_clock", "task_clock", "page_faults",
49
              "context_switches", "cpu_migrations",
50
              "page_faults_min", "page_faults_maj",
51 52
              "alignment_faults", "emulation_faults",
);
53 54 55 56 57 58 59 60 61 62 63 64 65

struct virPerfEvent {
    int fd;
    bool enabled;
    union {
        /* cmt */
        struct {
            int scale;
        } cmt;
    } efields;
};
typedef struct virPerfEvent *virPerfEventPtr;

66
struct _virPerf {
67 68 69 70 71 72 73
    struct virPerfEvent events[VIR_PERF_EVENT_LAST];
};

#if defined(__linux__) && defined(HAVE_SYS_SYSCALL_H)

# include <linux/perf_event.h>

74 75 76 77 78 79
struct virPerfEventAttr {
    unsigned int attrType;
    unsigned long long attrConfig;
};

static struct virPerfEventAttr attrs[] = {
80 81 82 83 84 85 86 87
    [VIR_PERF_EVENT_CMT] = {
        .attrType = 0,
        .attrConfig = 1
    },
    [VIR_PERF_EVENT_MBMT] = {
        .attrType = 0,
        .attrConfig = 2
    },
88
    [VIR_PERF_EVENT_MBML] = {
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
        .attrType = 0,
        .attrConfig = 3
    },
    [VIR_PERF_EVENT_CPU_CYCLES] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_CPU_CYCLES
    },
    [VIR_PERF_EVENT_INSTRUCTIONS] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_INSTRUCTIONS
    },
    [VIR_PERF_EVENT_CACHE_REFERENCES] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_CACHE_REFERENCES
    },
    [VIR_PERF_EVENT_CACHE_MISSES] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_CACHE_MISSES
    },
    [VIR_PERF_EVENT_BRANCH_INSTRUCTIONS] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_BRANCH_INSTRUCTIONS
    },
    [VIR_PERF_EVENT_BRANCH_MISSES] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_BRANCH_MISSES
    },
    [VIR_PERF_EVENT_BUS_CYCLES] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_BUS_CYCLES
    },
    [VIR_PERF_EVENT_STALLED_CYCLES_FRONTEND] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_STALLED_CYCLES_FRONTEND
    },
    [VIR_PERF_EVENT_STALLED_CYCLES_BACKEND] = {
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_STALLED_CYCLES_BACKEND
    },
    [VIR_PERF_EVENT_REF_CPU_CYCLES] = {
129
# ifdef PERF_COUNT_HW_REF_CPU_CYCLES
130 131
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_REF_CPU_CYCLES
132
# else
133 134
        .attrType = 0,
        .attrConfig = 0,
135 136
# endif
    },
137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
    [VIR_PERF_EVENT_CPU_CLOCK] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_CPU_CLOCK
    },
    [VIR_PERF_EVENT_TASK_CLOCK] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_TASK_CLOCK
    },
    [VIR_PERF_EVENT_PAGE_FAULTS] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_PAGE_FAULTS
    },
    [VIR_PERF_EVENT_CONTEXT_SWITCHES] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_CONTEXT_SWITCHES
    },
    [VIR_PERF_EVENT_CPU_MIGRATIONS] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_CPU_MIGRATIONS
    },
    [VIR_PERF_EVENT_PAGE_FAULTS_MIN] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_PAGE_FAULTS_MIN
    },
    [VIR_PERF_EVENT_PAGE_FAULTS_MAJ] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_PAGE_FAULTS_MAJ
    },
    [VIR_PERF_EVENT_ALIGNMENT_FAULTS] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_ALIGNMENT_FAULTS
    },
    [VIR_PERF_EVENT_EMULATION_FAULTS] = {
        .attrType = PERF_TYPE_SOFTWARE,
        .attrConfig = PERF_COUNT_SW_EMULATION_FAULTS
    },
173
};
174
G_STATIC_ASSERT(G_N_ELEMENTS(attrs) == VIR_PERF_EVENT_LAST);
175 176
typedef struct virPerfEventAttr *virPerfEventAttrPtr;

177

178
static int
179
virPerfRdtAttrInit(void)
180 181
{
    char *tmp = NULL;
182
    unsigned int attr_type = 0;
183
    g_autofree char *buf = NULL;
184

185
    if (virFileReadAllQuiet("/sys/devices/intel_cqm/type", 10, &buf) < 0)
186
        return -1;
187 188 189 190

    if ((tmp = strchr(buf, '\n')))
        *tmp = '\0';

191
    if (virStrToLong_ui(buf, NULL, 10, &attr_type) < 0) {
192
        virReportSystemError(errno, "%s",
193
                             _("failed to get rdt event type"));
194
        return -1;
195 196
    }

197 198 199 200 201 202 203 204 205 206 207 208 209 210
    attrs[VIR_PERF_EVENT_CMT].attrType = attr_type;
    attrs[VIR_PERF_EVENT_MBMT].attrType = attr_type;
    attrs[VIR_PERF_EVENT_MBML].attrType = attr_type;

    return 0;
}


int
virPerfEventEnable(virPerfPtr perf,
                   virPerfEventType type,
                   pid_t pid)
{
    struct perf_event_attr attr;
211
    virPerfEventPtr event = &(perf->events[type]);
212
    virPerfEventAttrPtr event_attr = &attrs[type];
213

214 215 216
    if (event->enabled)
        return 0;

217
    if (event_attr->attrType == 0 && (type == VIR_PERF_EVENT_CMT ||
218 219
                                      type == VIR_PERF_EVENT_MBMT ||
                                      type == VIR_PERF_EVENT_MBML)) {
220
        virReportError(VIR_ERR_ARGUMENT_UNSUPPORTED,
221
                       _("unable to enable host cpu perf event for %s"),
222
                       virPerfEventTypeToString(type));
223 224 225 226
        return -1;
    }

    if (type == VIR_PERF_EVENT_CMT) {
227
        g_autofree char *buf = NULL;
228

229 230 231 232
        if (virFileReadAll("/sys/devices/intel_cqm/events/llc_occupancy.scale",
                           10, &buf) < 0)
            goto error;

233
        if (virStrToLong_i(buf, NULL, 10, &event->efields.cmt.scale) < 0) {
234 235 236 237
            virReportSystemError(errno, "%s",
                                 _("failed to get cmt scaling factor"));
            goto error;
        }
238 239
    }

240 241 242 243 244 245 246 247 248
    memset(&attr, 0, sizeof(attr));
    attr.size = sizeof(attr);
    attr.inherit = 1;
    attr.disabled = 1;
    attr.enable_on_exec = 0;
    attr.type = event_attr->attrType;
    attr.config = event_attr->attrConfig;

    event->fd = syscall(__NR_perf_event_open, &attr, pid, -1, -1, 0);
249 250
    if (event->fd < 0) {
        virReportSystemError(errno,
251
                             _("unable to open host cpu perf event for %s"),
252
                             virPerfEventTypeToString(type));
253
        goto error;
254 255 256
    }

    if (ioctl(event->fd, PERF_EVENT_IOC_ENABLE) < 0) {
257
        virReportSystemError(errno,
258
                             _("unable to enable host cpu perf event for %s"),
259
                             virPerfEventTypeToString(type));
260
        goto error;
261 262 263 264 265
    }

    event->enabled = true;
    return 0;

266
 error:
267 268 269 270 271 272 273 274
    VIR_FORCE_CLOSE(event->fd);
    return -1;
}

int
virPerfEventDisable(virPerfPtr perf,
                    virPerfEventType type)
{
275
    virPerfEventPtr event = &(perf->events[type]);
276

277 278 279
    if (!event->enabled)
        return 0;

280 281
    if (ioctl(event->fd, PERF_EVENT_IOC_DISABLE) < 0) {
        virReportSystemError(errno,
282
                             _("unable to disable host cpu perf event for %s"),
283
                             virPerfEventTypeToString(type));
284 285 286 287 288 289 290 291 292 293 294
        return -1;
    }

    event->enabled = false;
    VIR_FORCE_CLOSE(event->fd);
    return 0;
}

bool virPerfEventIsEnabled(virPerfPtr perf,
                           virPerfEventType type)
{
295
    return perf && perf->events[type].enabled;
296 297 298 299 300 301 302
}

int
virPerfReadEvent(virPerfPtr perf,
                 virPerfEventType type,
                 uint64_t *value)
{
303 304
    virPerfEventPtr event = &perf->events[type];
    if (!event->enabled)
305 306
        return -1;

307
    if (saferead(event->fd, value, sizeof(uint64_t)) < 0) {
308 309 310 311 312 313 314 315 316 317 318 319
        virReportSystemError(errno, "%s",
                             _("Unable to read cache data"));
        return -1;
    }

    if (type == VIR_PERF_EVENT_CMT)
        *value *= event->efields.cmt.scale;

    return 0;
}

#else
320 321 322 323 324 325 326
static int
virPerfRdtAttrInit(void)
{
    return 0;
}


327
int
J
Ján Tomko 已提交
328 329 330
virPerfEventEnable(virPerfPtr perf G_GNUC_UNUSED,
                   virPerfEventType type G_GNUC_UNUSED,
                   pid_t pid G_GNUC_UNUSED)
331 332 333 334 335 336 337
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

int
J
Ján Tomko 已提交
338 339
virPerfEventDisable(virPerfPtr perf G_GNUC_UNUSED,
                    virPerfEventType type G_GNUC_UNUSED)
340 341 342 343 344 345 346
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

bool
J
Ján Tomko 已提交
347 348
virPerfEventIsEnabled(virPerfPtr perf G_GNUC_UNUSED,
                      virPerfEventType type G_GNUC_UNUSED)
349 350 351 352 353
{
    return false;
}

int
J
Ján Tomko 已提交
354 355 356
virPerfReadEvent(virPerfPtr perf G_GNUC_UNUSED,
                 virPerfEventType type G_GNUC_UNUSED,
                 uint64_t *value G_GNUC_UNUSED)
357 358 359 360 361 362 363
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

#endif
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378

virPerfPtr
virPerfNew(void)
{
    size_t i;
    virPerfPtr perf;

    if (VIR_ALLOC(perf) < 0)
        return NULL;

    for (i = 0; i < VIR_PERF_EVENT_LAST; i++) {
        perf->events[i].fd = -1;
        perf->events[i].enabled = false;
    }

379 380 381
    if (virPerfRdtAttrInit() < 0)
        virResetLastError();

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
    return perf;
}

void
virPerfFree(virPerfPtr perf)
{
    size_t i;

    if (perf == NULL)
        return;

    for (i = 0; i < VIR_PERF_EVENT_LAST; i++) {
        if (perf->events[i].enabled)
            virPerfEventDisable(perf, i);
    }

    VIR_FREE(perf);
}