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 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
/*
 * 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/>.
 *
 * Authors:
 *  Ren Qiaowei <qiaowei.ren@intel.com>
 */
#include <config.h>

#include <sys/ioctl.h>
#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"

VIR_LOG_INIT("util.perf");

#define VIR_FROM_THIS VIR_FROM_PERF

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

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

63
struct _virPerf {
64 65 66 67 68 69 70
    struct virPerfEvent events[VIR_PERF_EVENT_LAST];
};

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

# include <linux/perf_event.h>

71 72 73 74 75 76
struct virPerfEventAttr {
    unsigned int attrType;
    unsigned long long attrConfig;
};

static struct virPerfEventAttr attrs[] = {
77 78 79 80 81 82 83 84
    [VIR_PERF_EVENT_CMT] = {
        .attrType = 0,
        .attrConfig = 1
    },
    [VIR_PERF_EVENT_MBMT] = {
        .attrType = 0,
        .attrConfig = 2
    },
85
    [VIR_PERF_EVENT_MBML] = {
86 87 88 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
        .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] = {
126
# ifdef PERF_COUNT_HW_REF_CPU_CYCLES
127 128
        .attrType = PERF_TYPE_HARDWARE,
        .attrConfig = PERF_COUNT_HW_REF_CPU_CYCLES
129
# else
130 131
        .attrType = 0,
        .attrConfig = 0,
132 133
# endif
    },
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
    [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
    },
170
};
171
verify(ARRAY_CARDINALITY(attrs) == VIR_PERF_EVENT_LAST);
172 173
typedef struct virPerfEventAttr *virPerfEventAttrPtr;

174

175
static int
176
virPerfRdtAttrInit(void)
177 178
{
    char *tmp = NULL;
179
    unsigned int attr_type = 0;
180
    VIR_AUTOFREE(char *) buf = NULL;
181

182
    if (virFileReadAllQuiet("/sys/devices/intel_cqm/type", 10, &buf) < 0)
183
        return -1;
184 185 186 187

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

188
    if (virStrToLong_ui(buf, NULL, 10, &attr_type) < 0) {
189
        virReportSystemError(errno, "%s",
190
                             _("failed to get rdt event type"));
191
        return -1;
192 193
    }

194 195 196 197 198 199 200 201 202 203 204 205 206 207
    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;
208
    virPerfEventPtr event = &(perf->events[type]);
209
    virPerfEventAttrPtr event_attr = &attrs[type];
210

211 212 213
    if (event->enabled)
        return 0;

214 215 216
    if (event_attr->attrType == 0 && (type == VIR_PERF_EVENT_CMT ||
                                       type == VIR_PERF_EVENT_MBMT ||
                                       type == VIR_PERF_EVENT_MBML)) {
217
        virReportError(VIR_ERR_ARGUMENT_UNSUPPORTED,
218
                       _("unable to enable host cpu perf event for %s"),
219
                       virPerfEventTypeToString(type));
220 221 222 223
        return -1;
    }

    if (type == VIR_PERF_EVENT_CMT) {
224 225
        VIR_AUTOFREE(char *) buf = NULL;

226 227 228 229
        if (virFileReadAll("/sys/devices/intel_cqm/events/llc_occupancy.scale",
                           10, &buf) < 0)
            goto error;

230
        if (virStrToLong_i(buf, NULL, 10, &event->efields.cmt.scale) < 0) {
231 232 233 234
            virReportSystemError(errno, "%s",
                                 _("failed to get cmt scaling factor"));
            goto error;
        }
235 236
    }

237 238 239 240 241 242 243 244 245
    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);
246 247
    if (event->fd < 0) {
        virReportSystemError(errno,
248
                             _("unable to open host cpu perf event for %s"),
249
                             virPerfEventTypeToString(type));
250
        goto error;
251 252 253
    }

    if (ioctl(event->fd, PERF_EVENT_IOC_ENABLE) < 0) {
254
        virReportSystemError(errno,
255
                             _("unable to enable host cpu perf event for %s"),
256
                             virPerfEventTypeToString(type));
257
        goto error;
258 259 260 261 262
    }

    event->enabled = true;
    return 0;

263
 error:
264 265 266 267 268 269 270 271
    VIR_FORCE_CLOSE(event->fd);
    return -1;
}

int
virPerfEventDisable(virPerfPtr perf,
                    virPerfEventType type)
{
272
    virPerfEventPtr event = &(perf->events[type]);
273

274 275 276
    if (!event->enabled)
        return 0;

277 278
    if (ioctl(event->fd, PERF_EVENT_IOC_DISABLE) < 0) {
        virReportSystemError(errno,
279
                             _("unable to disable host cpu perf event for %s"),
280
                             virPerfEventTypeToString(type));
281 282 283 284 285 286 287 288 289 290 291
        return -1;
    }

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

bool virPerfEventIsEnabled(virPerfPtr perf,
                           virPerfEventType type)
{
292
    return perf && perf->events[type].enabled;
293 294 295 296 297 298 299
}

int
virPerfReadEvent(virPerfPtr perf,
                 virPerfEventType type,
                 uint64_t *value)
{
300 301
    virPerfEventPtr event = &perf->events[type];
    if (!event->enabled)
302 303
        return -1;

304
    if (saferead(event->fd, value, sizeof(uint64_t)) < 0) {
305 306 307 308 309 310 311 312 313 314 315 316
        virReportSystemError(errno, "%s",
                             _("Unable to read cache data"));
        return -1;
    }

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

    return 0;
}

#else
317 318 319 320 321 322 323
static int
virPerfRdtAttrInit(void)
{
    return 0;
}


324 325
int
virPerfEventEnable(virPerfPtr perf ATTRIBUTE_UNUSED,
R
Roman Bogorodskiy 已提交
326
                   virPerfEventType type ATTRIBUTE_UNUSED,
327 328 329 330 331 332 333 334 335
                   pid_t pid ATTRIBUTE_UNUSED)
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

int
virPerfEventDisable(virPerfPtr perf ATTRIBUTE_UNUSED,
R
Roman Bogorodskiy 已提交
336
                    virPerfEventType type ATTRIBUTE_UNUSED)
337 338 339 340 341 342 343
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

bool
R
Roman Bogorodskiy 已提交
344 345
virPerfEventIsEnabled(virPerfPtr perf ATTRIBUTE_UNUSED,
                      virPerfEventType type ATTRIBUTE_UNUSED)
346 347 348 349 350
{
    return false;
}

int
R
Roman Bogorodskiy 已提交
351 352 353
virPerfReadEvent(virPerfPtr perf ATTRIBUTE_UNUSED,
                 virPerfEventType type ATTRIBUTE_UNUSED,
                 uint64_t *value ATTRIBUTE_UNUSED)
354 355 356 357 358 359 360
{
    virReportSystemError(ENXIO, "%s",
                         _("Perf not supported on this platform"));
    return -1;
}

#endif
361 362 363 364 365 366 367 368 369 370 371 372 373 374 375

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;
    }

376 377 378
    if (virPerfRdtAttrInit() < 0)
        virResetLastError();

379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
    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);
}