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