device-crash-test (24374B)
1#!/usr/bin/env python3 2# 3# Copyright (c) 2017 Red Hat Inc 4# 5# Author: 6# Eduardo Habkost <ehabkost@redhat.com> 7# 8# This program is free software; you can redistribute it and/or modify 9# it under the terms of the GNU General Public License as published by 10# the Free Software Foundation; either version 2 of the License, or 11# (at your option) any later version. 12# 13# This program is distributed in the hope that it will be useful, 14# but WITHOUT ANY WARRANTY; without even the implied warranty of 15# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16# GNU General Public License for more details. 17# 18# You should have received a copy of the GNU General Public License along 19# with this program; if not, write to the Free Software Foundation, Inc., 20# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. 21 22""" 23Run QEMU with all combinations of -machine and -device types, 24check for crashes and unexpected errors. 25""" 26 27import os 28import sys 29import glob 30import logging 31import traceback 32import re 33import random 34import argparse 35from itertools import chain 36 37sys.path.append(os.path.join(os.path.dirname(__file__), '..', 'python')) 38from qemu.machine import QEMUMachine 39 40logger = logging.getLogger('device-crash-test') 41dbg = logger.debug 42 43 44# Purposes of the following rule list: 45# * Avoiding verbose log messages when we find known non-fatal 46# (exitcode=1) errors 47# * Avoiding fatal errors when we find known crashes 48# * Skipping machines/devices that are known not to work out of 49# the box, when running in --quick mode 50# 51# Keeping the rule list updated is desirable, but not required, 52# because unexpected cases where QEMU exits with exitcode=1 will 53# just trigger a INFO message. 54 55# Valid error rule keys: 56# * accel: regexp, full match only 57# * machine: regexp, full match only 58# * device: regexp, full match only 59# * log: regexp, partial match allowed 60# * exitcode: if not present, defaults to 1. If None, matches any exitcode 61# * warn: if True, matching failures will be logged as warnings 62# * expected: if True, QEMU is expected to always fail every time 63# when testing the corresponding test case 64# * loglevel: log level of log output when there's a match. 65ERROR_RULE_LIST = [ 66 # Machines that won't work out of the box: 67 # MACHINE | ERROR MESSAGE 68 {'machine':'niagara', 'expected':True}, # Unable to load a firmware for -M niagara 69 {'machine':'boston', 'expected':True}, # Please provide either a -kernel or -bios argument 70 {'machine':'leon3_generic', 'expected':True}, # Can't read bios image (null) 71 72 # devices that don't work out of the box because they require extra options to "-device DEV": 73 # DEVICE | ERROR MESSAGE 74 {'device':'.*-(i386|x86_64)-cpu', 'expected':True}, # CPU socket-id is not set 75 {'device':'icp', 'expected':True}, # icp_realize: required link 'xics' not found: Property '.xics' not found 76 {'device':'ics', 'expected':True}, # ics_base_realize: required link 'xics' not found: Property '.xics' not found 77 # "-device ide-cd" does work on more recent QEMU versions, so it doesn't have expected=True 78 {'device':'ide-cd'}, # No drive specified 79 {'device':'ide-hd', 'expected':True}, # No drive specified 80 {'device':'ipmi-bmc-extern', 'expected':True}, # IPMI external bmc requires chardev attribute 81 {'device':'isa-debugcon', 'expected':True}, # Can't create serial device, empty char device 82 {'device':'isa-ipmi-bt', 'expected':True}, # IPMI device requires a bmc attribute to be set 83 {'device':'isa-ipmi-kcs', 'expected':True}, # IPMI device requires a bmc attribute to be set 84 {'device':'isa-parallel', 'expected':True}, # Can't create serial device, empty char device 85 {'device':'ivshmem-doorbell', 'expected':True}, # You must specify a 'chardev' 86 {'device':'ivshmem-plain', 'expected':True}, # You must specify a 'memdev' 87 {'device':'loader', 'expected':True}, # please include valid arguments 88 {'device':'nand', 'expected':True}, # Unsupported NAND block size 0x1 89 {'device':'nvdimm', 'expected':True}, # 'memdev' property is not set 90 {'device':'nvme', 'expected':True}, # Device initialization failed 91 {'device':'pc-dimm', 'expected':True}, # 'memdev' property is not set 92 {'device':'pci-bridge', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0. 93 {'device':'pci-bridge-seat', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0. 94 {'device':'pxb', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0. 95 {'device':'scsi-block', 'expected':True}, # drive property not set 96 {'device':'scsi-generic', 'expected':True}, # drive property not set 97 {'device':'scsi-hd', 'expected':True}, # drive property not set 98 {'device':'spapr-pci-host-bridge', 'expected':True}, # BUID not specified for PHB 99 {'device':'spapr-rng', 'expected':True}, # spapr-rng needs an RNG backend! 100 {'device':'spapr-vty', 'expected':True}, # chardev property not set 101 {'device':'tpm-tis', 'expected':True}, # tpm_tis: backend driver with id (null) could not be found 102 {'device':'unimplemented-device', 'expected':True}, # property 'size' not specified or zero 103 {'device':'usb-braille', 'expected':True}, # Property chardev is required 104 {'device':'usb-mtp', 'expected':True}, # rootdir property must be configured 105 {'device':'usb-redir', 'expected':True}, # Parameter 'chardev' is missing 106 {'device':'usb-serial', 'expected':True}, # Property chardev is required 107 {'device':'usb-storage', 'expected':True}, # drive property not set 108 {'device':'vfio-amd-xgbe', 'expected':True}, # -device vfio-amd-xgbe: vfio error: wrong host device name 109 {'device':'vfio-calxeda-xgmac', 'expected':True}, # -device vfio-calxeda-xgmac: vfio error: wrong host device name 110 {'device':'vfio-pci', 'expected':True}, # No provided host device 111 {'device':'vfio-pci-igd-lpc-bridge', 'expected':True}, # VFIO dummy ISA/LPC bridge must have address 1f.0 112 {'device':'vhost-scsi.*', 'expected':True}, # vhost-scsi: missing wwpn 113 {'device':'vhost-vsock-device', 'expected':True}, # guest-cid property must be greater than 2 114 {'device':'vhost-vsock-pci', 'expected':True}, # guest-cid property must be greater than 2 115 {'device':'virtio-9p-ccw', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL 116 {'device':'virtio-9p-device', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL 117 {'device':'virtio-9p-pci', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL 118 {'device':'virtio-blk-ccw', 'expected':True}, # drive property not set 119 {'device':'virtio-blk-device', 'expected':True}, # drive property not set 120 {'device':'virtio-blk-device', 'expected':True}, # drive property not set 121 {'device':'virtio-blk-pci', 'expected':True}, # drive property not set 122 {'device':'virtio-crypto-ccw', 'expected':True}, # 'cryptodev' parameter expects a valid object 123 {'device':'virtio-crypto-device', 'expected':True}, # 'cryptodev' parameter expects a valid object 124 {'device':'virtio-crypto-pci', 'expected':True}, # 'cryptodev' parameter expects a valid object 125 {'device':'virtio-input-host-device', 'expected':True}, # evdev property is required 126 {'device':'virtio-input-host-pci', 'expected':True}, # evdev property is required 127 {'device':'xen-pvdevice', 'expected':True}, # Device ID invalid, it must always be supplied 128 {'device':'vhost-vsock-ccw', 'expected':True}, # guest-cid property must be greater than 2 129 {'device':'zpci', 'expected':True}, # target must be defined 130 {'device':'pnv-(occ|icp|lpc)', 'expected':True}, # required link 'xics' not found: Property '.xics' not found 131 {'device':'powernv-cpu-.*', 'expected':True}, # pnv_core_realize: required link 'xics' not found: Property '.xics' not found 132 133 # ioapic devices are already created by pc and will fail: 134 {'machine':'q35|pc.*', 'device':'kvm-ioapic', 'expected':True}, # Only 1 ioapics allowed 135 {'machine':'q35|pc.*', 'device':'ioapic', 'expected':True}, # Only 1 ioapics allowed 136 137 # "spapr-cpu-core needs a pseries machine" 138 {'machine':'(?!pseries).*', 'device':'.*-spapr-cpu-core', 'expected':True}, 139 140 # KVM-specific devices shouldn't be tried without accel=kvm: 141 {'accel':'(?!kvm).*', 'device':'kvmclock', 'expected':True}, 142 143 # xen-specific machines and devices: 144 {'accel':'(?!xen).*', 'machine':'xen.*', 'expected':True}, 145 {'accel':'(?!xen).*', 'device':'xen-.*', 'expected':True}, 146 147 # this fails on some machine-types, but not all, so they don't have expected=True: 148 {'device':'vmgenid'}, # vmgenid requires DMA write support in fw_cfg, which this machine type does not provide 149 150 # Silence INFO messages for errors that are common on multiple 151 # devices/machines: 152 {'log':r"No '[\w-]+' bus found for device '[\w-]+'"}, 153 {'log':r"images* must be given with the 'pflash' parameter"}, 154 {'log':r"(Guest|ROM|Flash|Kernel) image must be specified"}, 155 {'log':r"[cC]ould not load [\w ]+ (BIOS|bios) '[\w-]+\.bin'"}, 156 {'log':r"Couldn't find rom image '[\w-]+\.bin'"}, 157 {'log':r"speed mismatch trying to attach usb device"}, 158 {'log':r"Can't create a second ISA bus"}, 159 {'log':r"duplicate fw_cfg file name"}, 160 # sysbus-related error messages: most machines reject most dynamic sysbus devices: 161 {'log':r"Option '-device [\w.,-]+' cannot be handled by this machine"}, 162 {'log':r"Device [\w.,-]+ is not supported by this machine yet"}, 163 {'log':r"Device [\w.,-]+ can not be dynamically instantiated"}, 164 {'log':r"Platform Bus: Can not fit MMIO region of size "}, 165 # other more specific errors we will ignore: 166 {'device':'.*-spapr-cpu-core', 'log':r"CPU core type should be"}, 167 {'log':r"MSI(-X)? is not supported by interrupt controller"}, 168 {'log':r"pxb-pcie? devices cannot reside on a PCIe? bus"}, 169 {'log':r"Ignoring smp_cpus value"}, 170 {'log':r"sd_init failed: Drive 'sd0' is already in use because it has been automatically connected to another device"}, 171 {'log':r"This CPU requires a smaller page size than the system is using"}, 172 {'log':r"MSI-X support is mandatory in the S390 architecture"}, 173 {'log':r"rom check and register reset failed"}, 174 {'log':r"Unable to initialize GIC, CPUState for CPU#0 not valid"}, 175 {'log':r"Multiple VT220 operator consoles are not supported"}, 176 {'log':r"core 0 already populated"}, 177 {'log':r"could not find stage1 bootloader"}, 178 {'log':r"No '.*' bus found for device"}, 179 180 # other exitcode=1 failures not listed above will just generate INFO messages: 181 {'exitcode':1, 'loglevel':logging.INFO}, 182 183 # everything else (including SIGABRT and SIGSEGV) will be a fatal error: 184 {'exitcode':None, 'fatal':True, 'loglevel':logging.FATAL}, 185] 186 187 188def errorRuleTestCaseMatch(rule, t): 189 """Check if a test case specification can match a error rule 190 191 This only checks if a error rule is a candidate match 192 for a given test case, it won't check if the test case 193 results/output match the rule. See ruleListResultMatch(). 194 """ 195 return (('machine' not in rule or 196 'machine' not in t or 197 re.match(rule['machine'] + '$', t['machine'])) and 198 ('accel' not in rule or 199 'accel' not in t or 200 re.match(rule['accel'] + '$', t['accel'])) and 201 ('device' not in rule or 202 'device' not in t or 203 re.match(rule['device'] + '$', t['device']))) 204 205 206def ruleListCandidates(t): 207 """Generate the list of candidates that can match a test case""" 208 for i, rule in enumerate(ERROR_RULE_LIST): 209 if errorRuleTestCaseMatch(rule, t): 210 yield (i, rule) 211 212 213def findExpectedResult(t): 214 """Check if there's an expected=True error rule for a test case 215 216 Returns (i, rule) tuple, where i is the index in 217 ERROR_RULE_LIST and rule is the error rule itself. 218 """ 219 for i, rule in ruleListCandidates(t): 220 if rule.get('expected'): 221 return (i, rule) 222 223 224def ruleListResultMatch(rule, r): 225 """Check if test case results/output match a error rule 226 227 It is valid to call this function only if 228 errorRuleTestCaseMatch() is True for the rule (e.g. on 229 rules returned by ruleListCandidates()) 230 """ 231 assert errorRuleTestCaseMatch(rule, r['testcase']) 232 return ((rule.get('exitcode', 1) is None or 233 r['exitcode'] == rule.get('exitcode', 1)) and 234 ('log' not in rule or 235 re.search(rule['log'], r['log'], re.MULTILINE))) 236 237 238def checkResultRuleList(r): 239 """Look up error rule for a given test case result 240 241 Returns (i, rule) tuple, where i is the index in 242 ERROR_RULE_LIST and rule is the error rule itself. 243 """ 244 for i, rule in ruleListCandidates(r['testcase']): 245 if ruleListResultMatch(rule, r): 246 return i, rule 247 248 raise Exception("this should never happen") 249 250 251def qemuOptsEscape(s): 252 """Escape option value QemuOpts""" 253 return s.replace(",", ",,") 254 255 256def formatTestCase(t): 257 """Format test case info as "key=value key=value" for prettier logging output""" 258 return ' '.join('%s=%s' % (k, v) for k, v in t.items()) 259 260 261def qomListTypeNames(vm, **kwargs): 262 """Run qom-list-types QMP command, return type names""" 263 types = vm.command('qom-list-types', **kwargs) 264 return [t['name'] for t in types] 265 266 267def infoQDM(vm): 268 """Parse 'info qdm' output""" 269 args = {'command-line': 'info qdm'} 270 devhelp = vm.command('human-monitor-command', **args) 271 for l in devhelp.split('\n'): 272 l = l.strip() 273 if l == '' or l.endswith(':'): 274 continue 275 d = {'name': re.search(r'name "([^"]+)"', l).group(1), 276 'no-user': (re.search(', no-user', l) is not None)} 277 yield d 278 279 280class QemuBinaryInfo(object): 281 def __init__(self, binary, devtype): 282 if devtype is None: 283 devtype = 'device' 284 285 self.binary = binary 286 self._machine_info = {} 287 288 dbg("devtype: %r", devtype) 289 args = ['-S', '-machine', 'none,accel=kvm:tcg'] 290 dbg("querying info for QEMU binary: %s", binary) 291 vm = QEMUMachine(binary=binary, args=args) 292 vm.launch() 293 try: 294 self.alldevs = set(qomListTypeNames(vm, implements=devtype, abstract=False)) 295 # there's no way to query DeviceClass::user_creatable using QMP, 296 # so use 'info qdm': 297 self.no_user_devs = set([d['name'] for d in infoQDM(vm, ) if d['no-user']]) 298 self.machines = list(m['name'] for m in vm.command('query-machines')) 299 self.user_devs = self.alldevs.difference(self.no_user_devs) 300 self.kvm_available = vm.command('query-kvm')['enabled'] 301 finally: 302 vm.shutdown() 303 304 def machineInfo(self, machine): 305 """Query for information on a specific machine-type 306 307 Results are cached internally, in case the same machine- 308 type is queried multiple times. 309 """ 310 if machine in self._machine_info: 311 return self._machine_info[machine] 312 313 mi = {} 314 args = ['-S', '-machine', '%s' % (machine)] 315 dbg("querying machine info for binary=%s machine=%s", self.binary, machine) 316 vm = QEMUMachine(binary=self.binary, args=args) 317 try: 318 vm.launch() 319 mi['runnable'] = True 320 except KeyboardInterrupt: 321 raise 322 except: 323 dbg("exception trying to run binary=%s machine=%s", self.binary, machine, exc_info=sys.exc_info()) 324 dbg("log: %r", vm.get_log()) 325 mi['runnable'] = False 326 327 vm.shutdown() 328 self._machine_info[machine] = mi 329 return mi 330 331 332BINARY_INFO = {} 333 334 335def getBinaryInfo(args, binary): 336 if binary not in BINARY_INFO: 337 BINARY_INFO[binary] = QemuBinaryInfo(binary, args.devtype) 338 return BINARY_INFO[binary] 339 340 341def checkOneCase(args, testcase): 342 """Check one specific case 343 344 Returns a dictionary containing failure information on error, 345 or None on success 346 """ 347 binary = testcase['binary'] 348 accel = testcase['accel'] 349 machine = testcase['machine'] 350 device = testcase['device'] 351 352 dbg("will test: %r", testcase) 353 354 args = ['-S', '-machine', '%s,accel=%s' % (machine, accel), 355 '-device', qemuOptsEscape(device)] 356 cmdline = ' '.join([binary] + args) 357 dbg("will launch QEMU: %s", cmdline) 358 vm = QEMUMachine(binary=binary, args=args) 359 360 exc_traceback = None 361 try: 362 vm.launch() 363 except KeyboardInterrupt: 364 raise 365 except: 366 exc_traceback = traceback.format_exc() 367 dbg("Exception while running test case") 368 finally: 369 vm.shutdown() 370 ec = vm.exitcode() 371 log = vm.get_log() 372 373 if exc_traceback is not None or ec != 0: 374 return {'exc_traceback':exc_traceback, 375 'exitcode':ec, 376 'log':log, 377 'testcase':testcase, 378 'cmdline':cmdline} 379 380 381def binariesToTest(args, testcase): 382 if args.qemu: 383 r = args.qemu 384 else: 385 r = [f.path for f in os.scandir('.') 386 if f.name.startswith('qemu-system-') and 387 f.is_file() and os.access(f, os.X_OK)] 388 return r 389 390 391def accelsToTest(args, testcase): 392 if getBinaryInfo(args, testcase['binary']).kvm_available: 393 yield 'kvm' 394 yield 'tcg' 395 396 397def machinesToTest(args, testcase): 398 return getBinaryInfo(args, testcase['binary']).machines 399 400 401def devicesToTest(args, testcase): 402 return getBinaryInfo(args, testcase['binary']).user_devs 403 404 405TESTCASE_VARIABLES = [ 406 ('binary', binariesToTest), 407 ('accel', accelsToTest), 408 ('machine', machinesToTest), 409 ('device', devicesToTest), 410] 411 412 413def genCases1(args, testcases, var, fn): 414 """Generate new testcases for one variable 415 416 If an existing item already has a variable set, don't 417 generate new items and just return it directly. This 418 allows the "-t" command-line option to be used to choose 419 a specific test case. 420 """ 421 for testcase in testcases: 422 if var in testcase: 423 yield testcase.copy() 424 else: 425 for i in fn(args, testcase): 426 t = testcase.copy() 427 t[var] = i 428 yield t 429 430 431def genCases(args, testcase): 432 """Generate test cases for all variables 433 """ 434 cases = [testcase.copy()] 435 for var, fn in TESTCASE_VARIABLES: 436 dbg("var: %r, fn: %r", var, fn) 437 cases = genCases1(args, cases, var, fn) 438 return cases 439 440 441def casesToTest(args, testcase): 442 cases = genCases(args, testcase) 443 if args.random: 444 cases = list(cases) 445 cases = random.sample(cases, min(args.random, len(cases))) 446 if args.debug: 447 cases = list(cases) 448 dbg("%d test cases to test", len(cases)) 449 if args.shuffle: 450 cases = list(cases) 451 random.shuffle(cases) 452 return cases 453 454 455def logFailure(f, level): 456 t = f['testcase'] 457 logger.log(level, "failed: %s", formatTestCase(t)) 458 logger.log(level, "cmdline: %s", f['cmdline']) 459 for l in f['log'].strip().split('\n'): 460 logger.log(level, "log: %s", l) 461 logger.log(level, "exit code: %r", f['exitcode']) 462 if f['exc_traceback']: 463 logger.log(level, "exception:") 464 for l in f['exc_traceback'].split('\n'): 465 logger.log(level, " %s", l.rstrip('\n')) 466 467 468def main(): 469 parser = argparse.ArgumentParser(description="QEMU -device crash test") 470 parser.add_argument('-t', metavar='KEY=VALUE', nargs='*', 471 help="Limit test cases to KEY=VALUE", 472 action='append', dest='testcases', default=[]) 473 parser.add_argument('-d', '--debug', action='store_true', 474 help='debug output') 475 parser.add_argument('-v', '--verbose', action='store_true', default=True, 476 help='verbose output') 477 parser.add_argument('-q', '--quiet', dest='verbose', action='store_false', 478 help='non-verbose output') 479 parser.add_argument('-r', '--random', type=int, metavar='COUNT', 480 help='run a random sample of COUNT test cases', 481 default=0) 482 parser.add_argument('--shuffle', action='store_true', 483 help='Run test cases in random order') 484 parser.add_argument('--dry-run', action='store_true', 485 help="Don't run any tests, just generate list") 486 parser.add_argument('-D', '--devtype', metavar='TYPE', 487 help="Test only device types that implement TYPE") 488 parser.add_argument('-Q', '--quick', action='store_true', default=True, 489 help="Quick mode: skip test cases that are expected to fail") 490 parser.add_argument('-F', '--full', action='store_false', dest='quick', 491 help="Full mode: test cases that are expected to fail") 492 parser.add_argument('--strict', action='store_true', dest='strict', 493 help="Treat all warnings as fatal") 494 parser.add_argument('qemu', nargs='*', metavar='QEMU', 495 help='QEMU binary to run') 496 args = parser.parse_args() 497 498 if args.debug: 499 lvl = logging.DEBUG 500 elif args.verbose: 501 lvl = logging.INFO 502 else: 503 lvl = logging.WARN 504 logging.basicConfig(stream=sys.stdout, level=lvl, format='%(levelname)s: %(message)s') 505 506 fatal_failures = [] 507 wl_stats = {} 508 skipped = 0 509 total = 0 510 511 tc = {} 512 dbg("testcases: %r", args.testcases) 513 if args.testcases: 514 for t in chain(*args.testcases): 515 for kv in t.split(): 516 k, v = kv.split('=', 1) 517 tc[k] = v 518 519 if len(binariesToTest(args, tc)) == 0: 520 print("No QEMU binary found", file=sys.stderr) 521 parser.print_usage(sys.stderr) 522 return 1 523 524 for t in casesToTest(args, tc): 525 logger.info("running test case: %s", formatTestCase(t)) 526 total += 1 527 528 expected_match = findExpectedResult(t) 529 if (args.quick and 530 (expected_match or 531 not getBinaryInfo(args, t['binary']).machineInfo(t['machine'])['runnable'])): 532 dbg("skipped: %s", formatTestCase(t)) 533 skipped += 1 534 continue 535 536 if args.dry_run: 537 continue 538 539 try: 540 f = checkOneCase(args, t) 541 except KeyboardInterrupt: 542 break 543 544 if f: 545 i, rule = checkResultRuleList(f) 546 dbg("testcase: %r, rule list match: %r", t, rule) 547 wl_stats.setdefault(i, []).append(f) 548 level = rule.get('loglevel', logging.DEBUG) 549 logFailure(f, level) 550 if rule.get('fatal') or (args.strict and level >= logging.WARN): 551 fatal_failures.append(f) 552 else: 553 dbg("success: %s", formatTestCase(t)) 554 if expected_match: 555 logger.warn("Didn't fail as expected: %s", formatTestCase(t)) 556 557 logger.info("Total: %d test cases", total) 558 if skipped: 559 logger.info("Skipped %d test cases", skipped) 560 561 if args.debug: 562 stats = sorted([(len(wl_stats.get(i, [])), rule) for i, rule in 563 enumerate(ERROR_RULE_LIST)], key=lambda x: x[0]) 564 for count, rule in stats: 565 dbg("error rule stats: %d: %r", count, rule) 566 567 if fatal_failures: 568 for f in fatal_failures: 569 t = f['testcase'] 570 logger.error("Fatal failure: %s", formatTestCase(t)) 571 logger.error("Fatal failures on some machine/device combinations") 572 return 1 573 574if __name__ == '__main__': 575 sys.exit(main())