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
  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 ioctl_numbers = {
 181     'SET_FILTER' : 0x40082406,
 182     'ENABLE'     : 0x00002400,
 183     'DISABLE'    : 0x00002401,
 184 }
 185
 186 def x86_init(flag):
 187     globals().update({
 188         'sc_perf_evt_open' : 298,
 189         'exit_reasons' : x86_exit_reasons[flag],
 190     })
 191
 192 def s390_init():
 193     globals().update({
 194         'sc_perf_evt_open' : 331
 195     })
 196
 197 def ppc_init():
 198     globals().update({
 199         'sc_perf_evt_open' : 319,
 200         'ioctl_numbers' : {
 201             'SET_FILTER' : 0x80002406 | (ctypes.sizeof(ctypes.c_char_p) << 16),
 202             'ENABLE'     : 0x20002400,
 203             'DISABLE'    : 0x20002401,
 204         }
 205     })
 206
 207 def detect_platform():
 208     if os.uname()[4].startswith('ppc'):
 209         ppc_init()
 210         return
 211
 212     for line in file('/proc/cpuinfo').readlines():
 213         if line.startswith('flags'):
 214             for flag in line.split():
 215                 if flag in x86_exit_reasons:
 216                     x86_init(flag)
 217                     return
 218         elif line.startswith('vendor_id'):
 219             for flag in line.split():
 220                 if flag == 'IBM/S390':
 221                     s390_init()
 222                     return
 223
 224 detect_platform()
 225
 226 def invert(d):
 227     return dict((x[1], x[0]) for x in d.iteritems())
 228
 229 filters = {}
 230 filters['kvm_userspace_exit'] = ('reason', invert(userspace_exit_reasons))
 231 if exit_reasons:
 232     filters['kvm_exit'] = ('exit_reason', invert(exit_reasons))
 233
 234 import struct, array
 235
 236 libc = ctypes.CDLL('libc.so.6')
 237 syscall = libc.syscall
 238 class perf_event_attr(ctypes.Structure):
 239     _fields_ = [('type', ctypes.c_uint32),
 240                 ('size', ctypes.c_uint32),
 241                 ('config', ctypes.c_uint64),
 242                 ('sample_freq', ctypes.c_uint64),
 243                 ('sample_type', ctypes.c_uint64),
 244                 ('read_format', ctypes.c_uint64),
 245                 ('flags', ctypes.c_uint64),
 246                 ('wakeup_events', ctypes.c_uint32),
 247                 ('bp_type', ctypes.c_uint32),
 248                 ('bp_addr', ctypes.c_uint64),
 249                 ('bp_len', ctypes.c_uint64),
 250                 ]
 251 def _perf_event_open(attr, pid, cpu, group_fd, flags):
 252     return syscall(sc_perf_evt_open, ctypes.pointer(attr), ctypes.c_int(pid),
 253                    ctypes.c_int(cpu), ctypes.c_int(group_fd),
 254                    ctypes.c_long(flags))
 255
 256 PERF_TYPE_HARDWARE              = 0
 257 PERF_TYPE_SOFTWARE              = 1
 258 PERF_TYPE_TRACEPOINT            = 2
 259 PERF_TYPE_HW_CACHE              = 3
 260 PERF_TYPE_RAW                   = 4
 261 PERF_TYPE_BREAKPOINT            = 5
 262
 263 PERF_SAMPLE_IP                  = 1 << 0
 264 PERF_SAMPLE_TID                 = 1 << 1
 265 PERF_SAMPLE_TIME                = 1 << 2
 266 PERF_SAMPLE_ADDR                = 1 << 3
 267 PERF_SAMPLE_READ                = 1 << 4
 268 PERF_SAMPLE_CALLCHAIN           = 1 << 5
 269 PERF_SAMPLE_ID                  = 1 << 6
 270 PERF_SAMPLE_CPU                 = 1 << 7
 271 PERF_SAMPLE_PERIOD              = 1 << 8
 272 PERF_SAMPLE_STREAM_ID           = 1 << 9
 273 PERF_SAMPLE_RAW                 = 1 << 10
 274
 275 PERF_FORMAT_TOTAL_TIME_ENABLED  = 1 << 0
 276 PERF_FORMAT_TOTAL_TIME_RUNNING  = 1 << 1
 277 PERF_FORMAT_ID                  = 1 << 2
 278 PERF_FORMAT_GROUP               = 1 << 3
 279
 280 import re
 281
 282 sys_tracing = '/sys/kernel/debug/tracing'
 283
 284 class Group(object):
 285     def __init__(self, cpu):
 286         self.events = []
 287         self.group_leader = None
 288         self.cpu = cpu
 289     def add_event(self, name, event_set, tracepoint, filter = None):
 290         self.events.append(Event(group = self,
 291                                  name = name, event_set = event_set,
 292                                  tracepoint = tracepoint, filter = filter))
 293         if len(self.events) == 1:
 294             self.file = os.fdopen(self.events[0].fd)
 295     def read(self):
 296         bytes = 8 * (1 + len(self.events))
 297         fmt = 'xxxxxxxx' + 'q' * len(self.events)
 298         return dict(zip([event.name for event in self.events],
 299                         struct.unpack(fmt, self.file.read(bytes))))
 300
 301 class Event(object):
 302     def __init__(self, group, name, event_set, tracepoint, filter = None):
 303         self.name = name
 304         attr = perf_event_attr()
 305         attr.type = PERF_TYPE_TRACEPOINT
 306         attr.size = ctypes.sizeof(attr)
 307         id_path = os.path.join(sys_tracing, 'events', event_set,
 308                                tracepoint, 'id')
 309         id = int(file(id_path).read())
 310         attr.config = id
 311         attr.sample_type = (PERF_SAMPLE_RAW
 312                             | PERF_SAMPLE_TIME
 313                             | PERF_SAMPLE_CPU)
 314         attr.sample_period = 1
 315         attr.read_format = PERF_FORMAT_GROUP
 316         group_leader = -1
 317         if group.events:
 318             group_leader = group.events[0].fd
 319         fd = _perf_event_open(attr, -1, group.cpu, group_leader, 0)
 320         if fd == -1:
 321             raise Exception('perf_event_open failed')
 322         if filter:
 323             import fcntl
 324             fcntl.ioctl(fd, ioctl_numbers['SET_FILTER'], filter)
 325         self.fd = fd
 326     def enable(self):
 327         import fcntl
 328         fcntl.ioctl(self.fd, ioctl_numbers['ENABLE'], 0)
 329     def disable(self):
 330         import fcntl
 331         fcntl.ioctl(self.fd, ioctl_numbers['DISABLE'], 0)
 332
 333 class TracepointProvider(object):
 334     def __init__(self):
 335         path = os.path.join(sys_tracing, 'events', 'kvm')
 336         fields = [f
 337                   for f in os.listdir(path)
 338                   if os.path.isdir(os.path.join(path, f))]
 339         extra = []
 340         for f in fields:
 341             if f in filters:
 342                 subfield, values = filters[f]
 343                 for name, number in values.iteritems():
 344                     extra.append(f + '(' + name + ')')
 345         fields += extra
 346         self._setup(fields)
 347         self.select(fields)
 348     def fields(self):
 349         return self._fields
 350
 351     def _online_cpus(self):
 352         l = []
 353         pattern = r'cpu([0-9]+)'
 354         basedir = '/sys/devices/system/cpu'
 355         for entry in os.listdir(basedir):
 356             match = re.match(pattern, entry)
 357             if not match:
 358                 continue
 359             path = os.path.join(basedir, entry, 'online')
 360             if os.path.exists(path) and open(path).read().strip() != '1':
 361                 continue
 362             l.append(int(match.group(1)))
 363         return l
 364
 365     def _setup(self, _fields):
 366         self._fields = _fields
 367         cpus = self._online_cpus()
 368         import resource
 369         nfiles = len(cpus) * 1000
 370         resource.setrlimit(resource.RLIMIT_NOFILE, (nfiles, nfiles))
 371         events = []
 372         self.group_leaders = []
 373         for cpu in cpus:
 374             group = Group(cpu)
 375             for name in _fields:
 376                 tracepoint = name
 377                 filter = None
 378                 m = re.match(r'(.*)\((.*)\)', name)
 379                 if m:
 380                     tracepoint, sub = m.groups()
 381                     filter = '%s==%d\0' % (filters[tracepoint][0],
 382                                            filters[tracepoint][1][sub])
 383                 event = group.add_event(name, event_set = 'kvm',
 384                                         tracepoint = tracepoint,
 385                                         filter = filter)
 386             self.group_leaders.append(group)
 387     def select(self, fields):
 388         for group in self.group_leaders:
 389             for event in group.events:
 390                 if event.name in fields:
 391                     event.enable()
 392                 else:
 393                     event.disable()
 394     def read(self):
 395         from collections import defaultdict
 396         ret = defaultdict(int)
 397         for group in self.group_leaders:
 398             for name, val in group.read().iteritems():
 399                 ret[name] += val
 400         return ret
 401
 402 class Stats:
 403     def __init__(self, providers, fields = None):
 404         self.providers = providers
 405         self.fields_filter = fields
 406         self._update()
 407     def _update(self):
 408         def wanted(key):
 409             import re
 410             if not self.fields_filter:
 411                 return True
 412             return re.match(self.fields_filter, key) is not None
 413         self.values = dict()
 414         for d in providers:
 415             provider_fields = [key for key in d.fields() if wanted(key)]
 416             for key in provider_fields:
 417                 self.values[key] = None
 418             d.select(provider_fields)
 419     def set_fields_filter(self, fields_filter):
 420         self.fields_filter = fields_filter
 421         self._update()
 422     def get(self):
 423         for d in providers:
 424             new = d.read()
 425             for key in d.fields():
 426                 oldval = self.values.get(key, (0, 0))
 427                 newval = new[key]
 428                 newdelta = None
 429                 if oldval is not None:
 430                     newdelta = newval - oldval[0]
 431                 self.values[key] = (newval, newdelta)
 432         return self.values
 433
 434 if not os.access('/sys/kernel/debug', os.F_OK):
 435     print 'Please enable CONFIG_DEBUG_FS in your kernel'
 436     sys.exit(1)
 437 if not os.access('/sys/kernel/debug/kvm', os.F_OK):
 438     print "Please mount debugfs ('mount -t debugfs debugfs /sys/kernel/debug')"
 439     print "and ensure the kvm modules are loaded"
 440     sys.exit(1)
 441
 442 label_width = 40
 443 number_width = 10
 444
 445 def tui(screen, stats):
 446     curses.use_default_colors()
 447     curses.noecho()
 448     drilldown = False
 449     fields_filter = stats.fields_filter
 450     def update_drilldown():
 451         if not fields_filter:
 452             if drilldown:
 453                 stats.set_fields_filter(None)
 454             else:
 455                 stats.set_fields_filter(r'^[^\(]*$')
 456     update_drilldown()
 457     def refresh(sleeptime):
 458         screen.erase()
 459         screen.addstr(0, 0, 'kvm statistics')
 460         row = 2
 461         s = stats.get()
 462         def sortkey(x):
 463             if s[x][1]:
 464                 return (-s[x][1], -s[x][0])
 465             else:
 466                 return (0, -s[x][0])
 467         for key in sorted(s.keys(), key = sortkey):
 468             if row >= screen.getmaxyx()[0]:
 469                 break
 470             values = s[key]
 471             if not values[0] and not values[1]:
 472                 break
 473             col = 1
 474             screen.addstr(row, col, key)
 475             col += label_width
 476             screen.addstr(row, col, '%10d' % (values[0],))
 477             col += number_width
 478             if values[1] is not None:
 479                 screen.addstr(row, col, '%8d' % (values[1] / sleeptime,))
 480             row += 1
 481         screen.refresh()
 482
 483     sleeptime = 0.25
 484     while True:
 485         refresh(sleeptime)
 486         curses.halfdelay(int(sleeptime * 10))
 487         sleeptime = 3
 488         try:
 489             c = screen.getkey()
 490             if c == 'x':
 491                 drilldown = not drilldown
 492                 update_drilldown()
 493             if c == 'q':
 494                 break
 495         except KeyboardInterrupt:
 496             break
 497         except curses.error:
 498             continue
 499
 500 def batch(stats):
 501     s = stats.get()
 502     time.sleep(1)
 503     s = stats.get()
 504     for key in sorted(s.keys()):
 505         values = s[key]
 506         print '%-22s%10d%10d' % (key, values[0], values[1])
 507
 508 def log(stats):
 509     keys = sorted(stats.get().iterkeys())
 510     def banner():
 511         for k in keys:
 512             print '%10s' % k[0:9],
 513         print
 514     def statline():
 515         s = stats.get()
 516         for k in keys:
 517             print ' %9d' % s[k][1],
 518         print
 519     line = 0
 520     banner_repeat = 20
 521     while True:
 522         time.sleep(1)
 523         if line % banner_repeat == 0:
 524             banner()
 525         statline()
 526         line += 1
 527
 528 options = optparse.OptionParser()
 529 options.add_option('-1', '--once', '--batch',
 530                    action = 'store_true',
 531                    default = False,
 532                    dest = 'once',
 533                    help = 'run in batch mode for one second',
 534                    )
 535 options.add_option('-l', '--log',
 536                    action = 'store_true',
 537                    default = False,
 538                    dest = 'log',
 539                    help = 'run in logging mode (like vmstat)',
 540                    )
 541 options.add_option('-t', '--tracepoints',
 542                    action = 'store_true',
 543                    default = False,
 544                    dest = 'tracepoints',
 545                    help = 'retrieve statistics from tracepoints',
 546                    )
 547 options.add_option('-d', '--debugfs',
 548                    action = 'store_true',
 549                    default = False,
 550                    dest = 'debugfs',
 551                    help = 'retrieve statistics from debugfs',
 552                    )
 553 options.add_option('-f', '--fields',
 554                    action = 'store',
 555                    default = None,
 556                    dest = 'fields',
 557                    help = 'fields to display (regex)',
 558                    )
 559 (options, args) = options.parse_args(sys.argv)
 560
 561 providers = []
 562 if options.tracepoints:
 563     providers.append(TracepointProvider())
 564 if options.debugfs:
 565     providers.append(DebugfsProvider())
 566
 567 if len(providers) == 0:
 568     try:
 569         providers = [TracepointProvider()]
 570     except:
 571         providers = [DebugfsProvider()]
 572
 573 stats = Stats(providers, fields = options.fields)
 574
 575 if options.log:
 576     log(stats)
 577 elif not options.once:
 578     import curses.wrapper
 579     curses.wrapper(tui, stats)
 580 else:
 581     batch(stats)