scripts/kvm/kvm_stat

   1 #!/usr/bin/python
   2 #
   3 # top-like utility for displaying kvm statistics
   4 #
   5 # Copyright 2006-2008 Qumranet Technologies
   6 # Copyright 2008-2011 Red Hat, Inc.
   7 #
   8 # Authors:
   9 #  Avi Kivity <avi@redhat.com>
  10 #
  11 # This work is licensed under the terms of the GNU GPL, version 2.  See
  12 # the COPYING file in the top-level directory.
  13
  14 import curses
  15 import sys, os, time, optparse, ctypes
  16 from ctypes import *
  17
  18 class DebugfsProvider(object):
  19     def __init__(self):
  20         self.base = '/sys/kernel/debug/kvm'
  21         self._fields = os.listdir(self.base)
  22     def fields(self):
  23         return self._fields
  24     def select(self, fields):
  25         self._fields = fields
  26     def read(self):
  27         def val(key):
  28             return int(file(self.base + '/' + key).read())
  29         return dict([(key, val(key)) for key in self._fields])
  30
  31 vmx_exit_reasons = {
  32     0: 'EXCEPTION_NMI',
  33     1: 'EXTERNAL_INTERRUPT',
  34     2: 'TRIPLE_FAULT',
  35     7: 'PENDING_INTERRUPT',
  36     8: 'NMI_WINDOW',
  37     9: 'TASK_SWITCH',
  38     10: 'CPUID',
  39     12: 'HLT',
  40     14: 'INVLPG',
  41     15: 'RDPMC',
  42     16: 'RDTSC',
  43     18: 'VMCALL',
  44     19: 'VMCLEAR',
  45     20: 'VMLAUNCH',
  46     21: 'VMPTRLD',
  47     22: 'VMPTRST',
  48     23: 'VMREAD',
  49     24: 'VMRESUME',
  50     25: 'VMWRITE',
  51     26: 'VMOFF',
  52     27: 'VMON',
  53     28: 'CR_ACCESS',
  54     29: 'DR_ACCESS',
  55     30: 'IO_INSTRUCTION',
  56     31: 'MSR_READ',
  57     32: 'MSR_WRITE',
  58     33: 'INVALID_STATE',
  59     36: 'MWAIT_INSTRUCTION',
  60     39: 'MONITOR_INSTRUCTION',
  61     40: 'PAUSE_INSTRUCTION',
  62     41: 'MCE_DURING_VMENTRY',
  63     43: 'TPR_BELOW_THRESHOLD',
  64     44: 'APIC_ACCESS',
  65     48: 'EPT_VIOLATION',
  66     49: 'EPT_MISCONFIG',
  67     54: 'WBINVD',
  68     55: 'XSETBV',
  69     56: 'APIC_WRITE',
  70     58: 'INVPCID',
  71 }
  72
  73 svm_exit_reasons = {
  74     0x000: 'READ_CR0',
  75     0x003: 'READ_CR3',
  76     0x004: 'READ_CR4',
  77     0x008: 'READ_CR8',
  78     0x010: 'WRITE_CR0',
  79     0x013: 'WRITE_CR3',
  80     0x014: 'WRITE_CR4',
  81     0x018: 'WRITE_CR8',
  82     0x020: 'READ_DR0',
  83     0x021: 'READ_DR1',
  84     0x022: 'READ_DR2',
  85     0x023: 'READ_DR3',
  86     0x024: 'READ_DR4',
  87     0x025: 'READ_DR5',
  88     0x026: 'READ_DR6',
  89     0x027: 'READ_DR7',
  90     0x030: 'WRITE_DR0',
  91     0x031: 'WRITE_DR1',
  92     0x032: 'WRITE_DR2',
  93     0x033: 'WRITE_DR3',
  94     0x034: 'WRITE_DR4',
  95     0x035: 'WRITE_DR5',
  96     0x036: 'WRITE_DR6',
  97     0x037: 'WRITE_DR7',
  98     0x040: 'EXCP_BASE',
  99     0x060: 'INTR',
 100     0x061: 'NMI',
 101     0x062: 'SMI',
 102     0x063: 'INIT',
 103     0x064: 'VINTR',
 104     0x065: 'CR0_SEL_WRITE',
 105     0x066: 'IDTR_READ',
 106     0x067: 'GDTR_READ',
 107     0x068: 'LDTR_READ',
 108     0x069: 'TR_READ',
 109     0x06a: 'IDTR_WRITE',
 110     0x06b: 'GDTR_WRITE',
 111     0x06c: 'LDTR_WRITE',
 112     0x06d: 'TR_WRITE',
 113     0x06e: 'RDTSC',
 114     0x06f: 'RDPMC',
 115     0x070: 'PUSHF',
 116     0x071: 'POPF',
 117     0x072: 'CPUID',
 118     0x073: 'RSM',
 119     0x074: 'IRET',
 120     0x075: 'SWINT',
 121     0x076: 'INVD',
 122     0x077: 'PAUSE',
 123     0x078: 'HLT',
 124     0x079: 'INVLPG',
 125     0x07a: 'INVLPGA',
 126     0x07b: 'IOIO',
 127     0x07c: 'MSR',
 128     0x07d: 'TASK_SWITCH',
 129     0x07e: 'FERR_FREEZE',
 130     0x07f: 'SHUTDOWN',
 131     0x080: 'VMRUN',
 132     0x081: 'VMMCALL',
 133     0x082: 'VMLOAD',
 134     0x083: 'VMSAVE',
 135     0x084: 'STGI',
 136     0x085: 'CLGI',
 137     0x086: 'SKINIT',
 138     0x087: 'RDTSCP',
 139     0x088: 'ICEBP',
 140     0x089: 'WBINVD',
 141     0x08a: 'MONITOR',
 142     0x08b: 'MWAIT',
 143     0x08c: 'MWAIT_COND',
 144     0x08d: 'XSETBV',
 145     0x400: 'NPF',
 146 }
 147
 148 # From include/uapi/linux/kvm.h, KVM_EXIT_xxx
 149 userspace_exit_reasons = {
 150      0: 'UNKNOWN',
 151      1: 'EXCEPTION',
 152      2: 'IO',
 153      3: 'HYPERCALL',
 154      4: 'DEBUG',
 155      5: 'HLT',
 156      6: 'MMIO',
 157      7: 'IRQ_WINDOW_OPEN',
 158      8: 'SHUTDOWN',
 159      9: 'FAIL_ENTRY',
 160     10: 'INTR',
 161     11: 'SET_TPR',
 162     12: 'TPR_ACCESS',
 163     13: 'S390_SIEIC',
 164     14: 'S390_RESET',
 165     15: 'DCR',
 166     16: 'NMI',
 167     17: 'INTERNAL_ERROR',
 168     18: 'OSI',
 169     19: 'PAPR_HCALL',
 170     20: 'S390_UCONTROL',
 171     21: 'WATCHDOG',
 172     22: 'S390_TSCH',
 173     23: 'EPR',
 174     24: 'SYSTEM_EVENT',
 175 }
 176
 177 x86_exit_reasons = {
 178     'vmx': vmx_exit_reasons,
 179     'svm': svm_exit_reasons,
 180 }
 181
 182 sc_perf_evt_open = None
 183 exit_reasons = None
 184
 185 ioctl_numbers = {
 186     'SET_FILTER' : 0x40082406,
 187     'ENABLE'     : 0x00002400,
 188     'DISABLE'    : 0x00002401,
 189     'RESET'      : 0x00002403,
 190 }
 191
 192 def x86_init(flag):
 193     globals().update({
 194         'sc_perf_evt_open' : 298,
 195         'exit_reasons' : x86_exit_reasons[flag],
 196     })
 197
 198 def s390_init():
 199     globals().update({
 200         'sc_perf_evt_open' : 331
 201     })
 202
 203 def ppc_init():
 204     globals().update({
 205         'sc_perf_evt_open' : 319,
 206         'ioctl_numbers' : {
 207             'SET_FILTER' : 0x80002406 | (ctypes.sizeof(ctypes.c_char_p) << 16),
 208             'ENABLE'     : 0x20002400,
 209             'DISABLE'    : 0x20002401,
 210         }
 211     })
 212
 213 def aarch64_init():
 214     globals().update({
 215         'sc_perf_evt_open' : 241
 216     })
 217
 218 def detect_platform():
 219     if os.uname()[4].startswith('ppc'):
 220         ppc_init()
 221         return
 222     elif os.uname()[4].startswith('aarch64'):
 223         aarch64_init()
 224         return
 225
 226     for line in file('/proc/cpuinfo').readlines():
 227         if line.startswith('flags'):
 228             for flag in line.split():
 229                 if flag in x86_exit_reasons:
 230                     x86_init(flag)
 231                     return
 232         elif line.startswith('vendor_id'):
 233             for flag in line.split():
 234                 if flag == 'IBM/S390':
 235                     s390_init()
 236                     return
 237
 238 detect_platform()
 239
 240 def invert(d):
 241     return dict((x[1], x[0]) for x in d.iteritems())
 242
 243 filters = {}
 244 filters['kvm_userspace_exit'] = ('reason', invert(userspace_exit_reasons))
 245 if exit_reasons:
 246     filters['kvm_exit'] = ('exit_reason', invert(exit_reasons))
 247
 248 import struct, array
 249
 250 libc = ctypes.CDLL('libc.so.6')
 251 syscall = libc.syscall
 252 get_errno = libc.__errno_location
 253 get_errno.restype = POINTER(c_int)
 254
 255 class perf_event_attr(ctypes.Structure):
 256     _fields_ = [('type', ctypes.c_uint32),
 257                 ('size', ctypes.c_uint32),
 258                 ('config', ctypes.c_uint64),
 259                 ('sample_freq', ctypes.c_uint64),
 260                 ('sample_type', ctypes.c_uint64),
 261                 ('read_format', ctypes.c_uint64),
 262                 ('flags', ctypes.c_uint64),
 263                 ('wakeup_events', ctypes.c_uint32),
 264                 ('bp_type', ctypes.c_uint32),
 265                 ('bp_addr', ctypes.c_uint64),
 266                 ('bp_len', ctypes.c_uint64),
 267                 ]
 268 def _perf_event_open(attr, pid, cpu, group_fd, flags):
 269     return syscall(sc_perf_evt_open, ctypes.pointer(attr), ctypes.c_int(pid),
 270                    ctypes.c_int(cpu), ctypes.c_int(group_fd),
 271                    ctypes.c_long(flags))
 272
 273 PERF_TYPE_HARDWARE              = 0
 274 PERF_TYPE_SOFTWARE              = 1
 275 PERF_TYPE_TRACEPOINT            = 2
 276 PERF_TYPE_HW_CACHE              = 3
 277 PERF_TYPE_RAW                   = 4
 278 PERF_TYPE_BREAKPOINT            = 5
 279
 280 PERF_SAMPLE_IP                  = 1 << 0
 281 PERF_SAMPLE_TID                 = 1 << 1
 282 PERF_SAMPLE_TIME                = 1 << 2
 283 PERF_SAMPLE_ADDR                = 1 << 3
 284 PERF_SAMPLE_READ                = 1 << 4
 285 PERF_SAMPLE_CALLCHAIN           = 1 << 5
 286 PERF_SAMPLE_ID                  = 1 << 6
 287 PERF_SAMPLE_CPU                 = 1 << 7
 288 PERF_SAMPLE_PERIOD              = 1 << 8
 289 PERF_SAMPLE_STREAM_ID           = 1 << 9
 290 PERF_SAMPLE_RAW                 = 1 << 10
 291
 292 PERF_FORMAT_TOTAL_TIME_ENABLED  = 1 << 0
 293 PERF_FORMAT_TOTAL_TIME_RUNNING  = 1 << 1
 294 PERF_FORMAT_ID                  = 1 << 2
 295 PERF_FORMAT_GROUP               = 1 << 3
 296
 297 import re
 298
 299 sys_tracing = '/sys/kernel/debug/tracing'
 300
 301 class Group(object):
 302     def __init__(self, cpu):
 303         self.events = []
 304         self.group_leader = None
 305         self.cpu = cpu
 306     def add_event(self, name, event_set, tracepoint, filter = None):
 307         self.events.append(Event(group = self,
 308                                  name = name, event_set = event_set,
 309                                  tracepoint = tracepoint, filter = filter))
 310         if len(self.events) == 1:
 311             self.file = os.fdopen(self.events[0].fd)
 312     def read(self):
 313         bytes = 8 * (1 + len(self.events))
 314         fmt = 'xxxxxxxx' + 'q' * len(self.events)
 315         return dict(zip([event.name for event in self.events],
 316                         struct.unpack(fmt, self.file.read(bytes))))
 317
 318 class Event(object):
 319     def __init__(self, group, name, event_set, tracepoint, filter = None):
 320         self.name = name
 321         attr = perf_event_attr()
 322         attr.type = PERF_TYPE_TRACEPOINT
 323         attr.size = ctypes.sizeof(attr)
 324         id_path = os.path.join(sys_tracing, 'events', event_set,
 325                                tracepoint, 'id')
 326         id = int(file(id_path).read())
 327         attr.config = id
 328         attr.sample_type = (PERF_SAMPLE_RAW
 329                             | PERF_SAMPLE_TIME
 330                             | PERF_SAMPLE_CPU)
 331         attr.sample_period = 1
 332         attr.read_format = PERF_FORMAT_GROUP
 333         group_leader = -1
 334         if group.events:
 335             group_leader = group.events[0].fd
 336         fd = _perf_event_open(attr, -1, group.cpu, group_leader, 0)
 337         if fd == -1:
 338             err = get_errno()[0]
 339             raise Exception('perf_event_open failed, errno = ' + err.__str__())
 340         if filter:
 341             import fcntl
 342             fcntl.ioctl(fd, ioctl_numbers['SET_FILTER'], filter)
 343         self.fd = fd
 344     def enable(self):
 345         import fcntl
 346         fcntl.ioctl(self.fd, ioctl_numbers['ENABLE'], 0)
 347     def disable(self):
 348         import fcntl
 349         fcntl.ioctl(self.fd, ioctl_numbers['DISABLE'], 0)
 350     def reset(self):
 351         import fcntl
 352         fcntl.ioctl(self.fd, ioctl_numbers['RESET'], 0)
 353
 354 class TracepointProvider(object):
 355     def __init__(self):
 356         path = os.path.join(sys_tracing, 'events', 'kvm')
 357         fields = [f
 358                   for f in os.listdir(path)
 359                   if os.path.isdir(os.path.join(path, f))]
 360         extra = []
 361         for f in fields:
 362             if f in filters:
 363                 subfield, values = filters[f]
 364                 for name, number in values.iteritems():
 365                     extra.append(f + '(' + name + ')')
 366         fields += extra
 367         self._setup(fields)
 368         self.select(fields)
 369     def fields(self):
 370         return self._fields
 371
 372     def _online_cpus(self):
 373         l = []
 374         pattern = r'cpu([0-9]+)'
 375         basedir = '/sys/devices/system/cpu'
 376         for entry in os.listdir(basedir):
 377             match = re.match(pattern, entry)
 378             if not match:
 379                 continue
 380             path = os.path.join(basedir, entry, 'online')
 381             if os.path.exists(path) and open(path).read().strip() != '1':
 382                 continue
 383             l.append(int(match.group(1)))
 384         return l
 385
 386     def _setup(self, _fields):
 387         self._fields = _fields
 388         cpus = self._online_cpus()
 389         import resource
 390         nfiles = len(cpus) * 1000
 391         resource.setrlimit(resource.RLIMIT_NOFILE, (nfiles, nfiles))
 392         events = []
 393         self.group_leaders = []
 394         for cpu in cpus:
 395             group = Group(cpu)
 396             for name in _fields:
 397                 tracepoint = name
 398                 filter = None
 399                 m = re.match(r'(.*)\((.*)\)', name)
 400                 if m:
 401                     tracepoint, sub = m.groups()
 402                     filter = '%s==%d\0' % (filters[tracepoint][0],
 403                                            filters[tracepoint][1][sub])
 404                 event = group.add_event(name, event_set = 'kvm',
 405                                         tracepoint = tracepoint,
 406                                         filter = filter)
 407             self.group_leaders.append(group)
 408     def select(self, fields):
 409         for group in self.group_leaders:
 410             for event in group.events:
 411                 if event.name in fields:
 412                     event.reset()
 413                     event.enable()
 414                 else:
 415                     event.disable()
 416     def read(self):
 417         from collections import defaultdict
 418         ret = defaultdict(int)
 419         for group in self.group_leaders:
 420             for name, val in group.read().iteritems():
 421                 ret[name] += val
 422         return ret
 423
 424 class Stats:
 425     def __init__(self, providers, fields = None):
 426         self.providers = providers
 427         self.fields_filter = fields
 428         self._update()
 429     def _update(self):
 430         def wanted(key):
 431             import re
 432             if not self.fields_filter:
 433                 return True
 434             return re.match(self.fields_filter, key) is not None
 435         self.values = dict()
 436         for d in providers:
 437             provider_fields = [key for key in d.fields() if wanted(key)]
 438             for key in provider_fields:
 439                 self.values[key] = None
 440             d.select(provider_fields)
 441     def set_fields_filter(self, fields_filter):
 442         self.fields_filter = fields_filter
 443         self._update()
 444     def get(self):
 445         for d in providers:
 446             new = d.read()
 447             for key in d.fields():
 448                 oldval = self.values.get(key, (0, 0))
 449                 newval = new[key]
 450                 newdelta = None
 451                 if oldval is not None:
 452                     newdelta = newval - oldval[0]
 453                 self.values[key] = (newval, newdelta)
 454         return self.values
 455
 456 if not os.access('/sys/kernel/debug', os.F_OK):
 457     print 'Please enable CONFIG_DEBUG_FS in your kernel'
 458     sys.exit(1)
 459 if not os.access('/sys/kernel/debug/kvm', os.F_OK):
 460     print "Please mount debugfs ('mount -t debugfs debugfs /sys/kernel/debug')"
 461     print "and ensure the kvm modules are loaded"
 462     sys.exit(1)
 463
 464 label_width = 40
 465 number_width = 10
 466
 467 def tui(screen, stats):
 468     curses.use_default_colors()
 469     curses.noecho()
 470     drilldown = False
 471     fields_filter = stats.fields_filter
 472     def update_drilldown():
 473         if not fields_filter:
 474             if drilldown:
 475                 stats.set_fields_filter(None)
 476             else:
 477                 stats.set_fields_filter(r'^[^\(]*$')
 478     update_drilldown()
 479     def refresh(sleeptime):
 480         screen.erase()
 481         screen.addstr(0, 0, 'kvm statistics')
 482         row = 2
 483         s = stats.get()
 484         def sortkey(x):
 485             if s[x][1]:
 486                 return (-s[x][1], -s[x][0])
 487             else:
 488                 return (0, -s[x][0])
 489         for key in sorted(s.keys(), key = sortkey):
 490             if row >= screen.getmaxyx()[0]:
 491                 break
 492             values = s[key]
 493             if not values[0] and not values[1]:
 494                 break
 495             col = 1
 496             screen.addstr(row, col, key)
 497             col += label_width
 498             screen.addstr(row, col, '%10d' % (values[0],))
 499             col += number_width
 500             if values[1] is not None:
 501                 screen.addstr(row, col, '%8d' % (values[1] / sleeptime,))
 502             row += 1
 503         screen.refresh()
 504
 505     sleeptime = 0.25
 506     while True:
 507         refresh(sleeptime)
 508         curses.halfdelay(int(sleeptime * 10))
 509         sleeptime = 3
 510         try:
 511             c = screen.getkey()
 512             if c == 'x':
 513                 drilldown = not drilldown
 514                 update_drilldown()
 515             if c == 'q':
 516                 break
 517         except KeyboardInterrupt:
 518             break
 519         except curses.error:
 520             continue
 521
 522 def batch(stats):
 523     s = stats.get()
 524     time.sleep(1)
 525     s = stats.get()
 526     for key in sorted(s.keys()):
 527         values = s[key]
 528         print '%-22s%10d%10d' % (key, values[0], values[1])
 529
 530 def log(stats):
 531     keys = sorted(stats.get().iterkeys())
 532     def banner():
 533         for k in keys:
 534             print '%10s' % k[0:9],
 535         print
 536     def statline():
 537         s = stats.get()
 538         for k in keys:
 539             print ' %9d' % s[k][1],
 540         print
 541     line = 0
 542     banner_repeat = 20
 543     while True:
 544         time.sleep(1)
 545         if line % banner_repeat == 0:
 546             banner()
 547         statline()
 548         line += 1
 549
 550 options = optparse.OptionParser()
 551 options.add_option('-1', '--once', '--batch',
 552                    action = 'store_true',
 553                    default = False,
 554                    dest = 'once',
 555                    help = 'run in batch mode for one second',
 556                    )
 557 options.add_option('-l', '--log',
 558                    action = 'store_true',
 559                    default = False,
 560                    dest = 'log',
 561                    help = 'run in logging mode (like vmstat)',
 562                    )
 563 options.add_option('-t', '--tracepoints',
 564                    action = 'store_true',
 565                    default = False,
 566                    dest = 'tracepoints',
 567                    help = 'retrieve statistics from tracepoints',
 568                    )
 569 options.add_option('-d', '--debugfs',
 570                    action = 'store_true',
 571                    default = False,
 572                    dest = 'debugfs',
 573                    help = 'retrieve statistics from debugfs',
 574                    )
 575 options.add_option('-f', '--fields',
 576                    action = 'store',
 577                    default = None,
 578                    dest = 'fields',
 579                    help = 'fields to display (regex)',
 580                    )
 581 (options, args) = options.parse_args(sys.argv)
 582
 583 providers = []
 584 if options.tracepoints:
 585     providers.append(TracepointProvider())
 586 if options.debugfs:
 587     providers.append(DebugfsProvider())
 588
 589 if len(providers) == 0:
 590     try:
 591         providers = [TracepointProvider()]
 592     except:
 593         providers = [DebugfsProvider()]
 594
 595 stats = Stats(providers, fields = options.fields)
 596
 597 if options.log:
 598     log(stats)
 599 elif not options.once:
 600     import curses.wrapper
 601     curses.wrapper(tui, stats)
 602 else:
 603     batch(stats)