3 # Copyright (c) 2017 Red Hat Inc
6 # Eduardo Habkost <ehabkost@redhat.com>
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.
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.
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.
23 Run QEMU with all combinations of -machine and -device types,
24 check for crashes and unexpected errors.
35 from itertools import chain
36 from pathlib import Path
39 from qemu.machine import QEMUMachine
40 from qemu.qmp import ConnectError
41 except ModuleNotFoundError as exc:
42 path = Path(__file__).resolve()
43 print(f"Module '{exc.name}' not found.")
44 print(" Try 'make check-venv' from your build directory,")
45 print(" and then one way to run this script is like so:")
46 print(f' > $builddir/pyvenv/bin/python3 "{path}"')
49 logger = logging.getLogger('device-crash-test')
53 # Purposes of the following rule list:
54 # * Avoiding verbose log messages when we find known non-fatal
56 # * Avoiding fatal errors when we find known crashes
57 # * Skipping machines/devices that are known not to work out of
58 # the box, when running in --quick mode
60 # Keeping the rule list updated is desirable, but not required,
61 # because unexpected cases where QEMU exits with exitcode=1 will
62 # just trigger a INFO message.
64 # Valid error rule keys:
65 # * accel: regexp, full match only
66 # * machine: regexp, full match only
67 # * device: regexp, full match only
68 # * log: regexp, partial match allowed
69 # * exitcode: if not present, defaults to 1. If None, matches any exitcode
70 # * warn: if True, matching failures will be logged as warnings
71 # * expected: if True, QEMU is expected to always fail every time
72 # when testing the corresponding test case
73 # * loglevel: log level of log output when there's a match.
75 # Machines that won't work out of the box:
76 # MACHINE | ERROR MESSAGE
77 {'machine':'niagara', 'expected':True}, # Unable to load a firmware for -M niagara
78 {'machine':'boston', 'expected':True}, # Please provide either a -kernel or -bios argument
79 {'machine':'leon3_generic', 'expected':True}, # Can't read bios image (null)
81 # devices that don't work out of the box because they require extra options to "-device DEV":
82 # DEVICE | ERROR MESSAGE
83 {'device':'.*-(i386|x86_64)-cpu', 'expected':True}, # CPU socket-id is not set
84 {'device':'icp', 'expected':True}, # icp_realize: required link 'xics' not found: Property '.xics' not found
85 {'device':'ics', 'expected':True}, # ics_base_realize: required link 'xics' not found: Property '.xics' not found
86 # "-device ide-cd" does work on more recent QEMU versions, so it doesn't have expected=True
87 {'device':'ide-cd'}, # No drive specified
88 {'device':'ide-hd', 'expected':True}, # No drive specified
89 {'device':'ipmi-bmc-extern', 'expected':True}, # IPMI external bmc requires chardev attribute
90 {'device':'isa-debugcon', 'expected':True}, # Can't create serial device, empty char device
91 {'device':'isa-ipmi-bt', 'expected':True}, # IPMI device requires a bmc attribute to be set
92 {'device':'isa-ipmi-kcs', 'expected':True}, # IPMI device requires a bmc attribute to be set
93 {'device':'isa-parallel', 'expected':True}, # Can't create serial device, empty char device
94 {'device':'ivshmem-doorbell', 'expected':True}, # You must specify a 'chardev'
95 {'device':'ivshmem-plain', 'expected':True}, # You must specify a 'memdev'
96 {'device':'loader', 'expected':True}, # please include valid arguments
97 {'device':'nand', 'expected':True}, # Unsupported NAND block size 0x1
98 {'device':'nvdimm', 'expected':True}, # 'memdev' property is not set
99 {'device':'nvme', 'expected':True}, # Device initialization failed
100 {'device':'pc-dimm', 'expected':True}, # 'memdev' property is not set
101 {'device':'pci-bridge', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0.
102 {'device':'pci-bridge-seat', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0.
103 {'device':'pxb', 'expected':True}, # Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0.
104 {'device':'pxb-cxl', 'expected':True}, # pxb-cxl devices cannot reside on a PCI bus.
105 {'device':'scsi-block', 'expected':True}, # drive property not set
106 {'device':'scsi-generic', 'expected':True}, # drive property not set
107 {'device':'scsi-hd', 'expected':True}, # drive property not set
108 {'device':'spapr-pci-host-bridge', 'expected':True}, # BUID not specified for PHB
109 {'device':'spapr-rng', 'expected':True}, # spapr-rng needs an RNG backend!
110 {'device':'spapr-vty', 'expected':True}, # chardev property not set
111 {'device':'tpm-tis', 'expected':True}, # tpm_tis: backend driver with id (null) could not be found
112 {'device':'unimplemented-device', 'expected':True}, # property 'size' not specified or zero
113 {'device':'usb-braille', 'expected':True}, # Property chardev is required
114 {'device':'usb-mtp', 'expected':True}, # rootdir property must be configured
115 {'device':'usb-redir', 'expected':True}, # Parameter 'chardev' is missing
116 {'device':'usb-serial', 'expected':True}, # Property chardev is required
117 {'device':'usb-storage', 'expected':True}, # drive property not set
118 {'device':'vfio-amd-xgbe', 'expected':True}, # -device vfio-amd-xgbe: vfio error: wrong host device name
119 {'device':'vfio-calxeda-xgmac', 'expected':True}, # -device vfio-calxeda-xgmac: vfio error: wrong host device name
120 {'device':'vfio-pci', 'expected':True}, # No provided host device
121 {'device':'vfio-pci-igd-lpc-bridge', 'expected':True}, # VFIO dummy ISA/LPC bridge must have address 1f.0
122 {'device':'vhost-scsi.*', 'expected':True}, # vhost-scsi: missing wwpn
123 {'device':'vhost-vsock-device', 'expected':True}, # guest-cid property must be greater than 2
124 {'device':'vhost-vsock-pci', 'expected':True}, # guest-cid property must be greater than 2
125 {'device':'virtio-9p-ccw', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL
126 {'device':'virtio-9p-device', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL
127 {'device':'virtio-9p-pci', 'expected':True}, # 9pfs device couldn't find fsdev with the id = NULL
128 {'device':'virtio-blk-ccw', 'expected':True}, # drive property not set
129 {'device':'virtio-blk-device', 'expected':True}, # drive property not set
130 {'device':'virtio-blk-device', 'expected':True}, # drive property not set
131 {'device':'virtio-blk-pci', 'expected':True}, # drive property not set
132 {'device':'virtio-crypto-ccw', 'expected':True}, # 'cryptodev' parameter expects a valid object
133 {'device':'virtio-crypto-device', 'expected':True}, # 'cryptodev' parameter expects a valid object
134 {'device':'virtio-crypto-pci', 'expected':True}, # 'cryptodev' parameter expects a valid object
135 {'device':'virtio-input-host-device', 'expected':True}, # evdev property is required
136 {'device':'virtio-input-host-pci', 'expected':True}, # evdev property is required
137 {'device':'xen-pvdevice', 'expected':True}, # Device ID invalid, it must always be supplied
138 {'device':'vhost-vsock-ccw', 'expected':True}, # guest-cid property must be greater than 2
139 {'device':'zpci', 'expected':True}, # target must be defined
140 {'device':'pnv-(occ|icp|lpc)', 'expected':True}, # required link 'xics' not found: Property '.xics' not found
141 {'device':'powernv-cpu-.*', 'expected':True}, # pnv_core_realize: required link 'xics' not found: Property '.xics' not found
143 # ioapic devices are already created by pc and will fail:
144 {'machine':'q35|pc.*', 'device':'kvm-ioapic', 'expected':True}, # Only 1 ioapics allowed
145 {'machine':'q35|pc.*', 'device':'ioapic', 'expected':True}, # Only 1 ioapics allowed
147 # "spapr-cpu-core needs a pseries machine"
148 {'machine':'(?!pseries).*', 'device':'.*-spapr-cpu-core', 'expected':True},
150 # KVM-specific devices shouldn't be tried without accel=kvm:
151 {'accel':'(?!kvm).*', 'device':'kvmclock', 'expected':True},
153 # xen-specific machines and devices:
154 {'accel':'(?!xen).*', 'machine':'xen.*', 'expected':True},
155 {'accel':'(?!xen).*', 'device':'xen-.*', 'expected':True},
157 # this fails on some machine-types, but not all, so they don't have expected=True:
158 {'device':'vmgenid'}, # vmgenid requires DMA write support in fw_cfg, which this machine type does not provide
160 # Silence INFO messages for errors that are common on multiple
162 {'log':r"No '[\w-]+' bus found for device '[\w-]+'"},
163 {'log':r"images* must be given with the 'pflash' parameter"},
164 {'log':r"(Guest|ROM|Flash|Kernel) image must be specified"},
165 {'log':r"[cC]ould not load [\w ]+ (BIOS|bios) '[\w-]+\.bin'"},
166 {'log':r"Couldn't find rom image '[\w-]+\.bin'"},
167 {'log':r"speed mismatch trying to attach usb device"},
168 {'log':r"Can't create a second ISA bus"},
169 {'log':r"duplicate fw_cfg file name"},
170 # sysbus-related error messages: most machines reject most dynamic sysbus devices:
171 {'log':r"Option '-device [\w.,-]+' cannot be handled by this machine"},
172 {'log':r"Device [\w.,-]+ is not supported by this machine yet"},
173 {'log':r"Device [\w.,-]+ can not be dynamically instantiated"},
174 {'log':r"Platform Bus: Can not fit MMIO region of size "},
175 # other more specific errors we will ignore:
176 {'device':'.*-spapr-cpu-core', 'log':r"CPU core type should be"},
177 {'log':r"MSI(-X)? is not supported by interrupt controller"},
178 {'log':r"pxb-pcie? devices cannot reside on a PCIe? bus"},
179 {'log':r"Ignoring smp_cpus value"},
180 {'log':r"sd_init failed: Drive 'sd0' is already in use because it has been automatically connected to another device"},
181 {'log':r"This CPU requires a smaller page size than the system is using"},
182 {'log':r"MSI-X support is mandatory in the S390 architecture"},
183 {'log':r"rom check and register reset failed"},
184 {'log':r"Unable to initialize GIC, CPUState for CPU#0 not valid"},
185 {'log':r"Multiple VT220 operator consoles are not supported"},
186 {'log':r"core 0 already populated"},
187 {'log':r"could not find stage1 bootloader"},
189 # other exitcode=1 failures not listed above will just generate INFO messages:
190 {'exitcode':1, 'loglevel':logging.INFO},
192 # everything else (including SIGABRT and SIGSEGV) will be a fatal error:
193 {'exitcode':None, 'fatal':True, 'loglevel':logging.FATAL},
197 def errorRuleTestCaseMatch(rule, t):
198 """Check if a test case specification can match a error rule
200 This only checks if a error rule is a candidate match
201 for a given test case, it won't check if the test case
202 results/output match the rule. See ruleListResultMatch().
204 return (('machine' not in rule or
205 'machine' not in t or
206 re.match(rule['machine'] + '$', t['machine'])) and
207 ('accel' not in rule or
209 re.match(rule['accel'] + '$', t['accel'])) and
210 ('device' not in rule or
212 re.match(rule['device'] + '$', t['device'])))
215 def ruleListCandidates(t):
216 """Generate the list of candidates that can match a test case"""
217 for i, rule in enumerate(ERROR_RULE_LIST):
218 if errorRuleTestCaseMatch(rule, t):
222 def findExpectedResult(t):
223 """Check if there's an expected=True error rule for a test case
225 Returns (i, rule) tuple, where i is the index in
226 ERROR_RULE_LIST and rule is the error rule itself.
228 for i, rule in ruleListCandidates(t):
229 if rule.get('expected'):
233 def ruleListResultMatch(rule, r):
234 """Check if test case results/output match a error rule
236 It is valid to call this function only if
237 errorRuleTestCaseMatch() is True for the rule (e.g. on
238 rules returned by ruleListCandidates())
240 assert errorRuleTestCaseMatch(rule, r['testcase'])
241 return ((rule.get('exitcode', 1) is None or
242 r['exitcode'] == rule.get('exitcode', 1)) and
243 ('log' not in rule or
244 re.search(rule['log'], r['log'], re.MULTILINE)))
247 def checkResultRuleList(r):
248 """Look up error rule for a given test case result
250 Returns (i, rule) tuple, where i is the index in
251 ERROR_RULE_LIST and rule is the error rule itself.
253 for i, rule in ruleListCandidates(r['testcase']):
254 if ruleListResultMatch(rule, r):
257 raise Exception("this should never happen")
260 def qemuOptsEscape(s):
261 """Escape option value QemuOpts"""
262 return s.replace(",", ",,")
265 def formatTestCase(t):
266 """Format test case info as "key=value key=value" for prettier logging output"""
267 return ' '.join('%s=%s' % (k, v) for k, v in t.items())
270 def qomListTypeNames(vm, **kwargs):
271 """Run qom-list-types QMP command, return type names"""
272 types = vm.command('qom-list-types', **kwargs)
273 return [t['name'] for t in types]
277 """Parse 'info qdm' output"""
278 args = {'command-line': 'info qdm'}
279 devhelp = vm.command('human-monitor-command', **args)
280 for l in devhelp.split('\n'):
282 if l == '' or l.endswith(':'):
284 d = {'name': re.search(r'name "([^"]+)"', l).group(1),
285 'no-user': (re.search(', no-user', l) is not None)}
289 class QemuBinaryInfo(object):
290 def __init__(self, binary, devtype):
295 self._machine_info = {}
297 dbg("devtype: %r", devtype)
298 args = ['-S', '-machine', 'none,accel=kvm:tcg']
299 dbg("querying info for QEMU binary: %s", binary)
300 vm = QEMUMachine(binary=binary, args=args)
303 self.alldevs = set(qomListTypeNames(vm, implements=devtype, abstract=False))
304 # there's no way to query DeviceClass::user_creatable using QMP,
306 self.no_user_devs = set([d['name'] for d in infoQDM(vm, ) if d['no-user']])
307 self.machines = list(m['name'] for m in vm.command('query-machines'))
308 self.user_devs = self.alldevs.difference(self.no_user_devs)
309 self.kvm_available = vm.command('query-kvm')['enabled']
313 def machineInfo(self, machine):
314 """Query for information on a specific machine-type
316 Results are cached internally, in case the same machine-
317 type is queried multiple times.
319 if machine in self._machine_info:
320 return self._machine_info[machine]
323 args = ['-S', '-machine', '%s' % (machine)]
324 dbg("querying machine info for binary=%s machine=%s", self.binary, machine)
325 vm = QEMUMachine(binary=self.binary, args=args)
328 mi['runnable'] = True
330 dbg("exception trying to run binary=%s machine=%s", self.binary, machine, exc_info=sys.exc_info())
331 dbg("log: %r", vm.get_log())
332 mi['runnable'] = False
335 self._machine_info[machine] = mi
342 def getBinaryInfo(args, binary):
343 if binary not in BINARY_INFO:
344 BINARY_INFO[binary] = QemuBinaryInfo(binary, args.devtype)
345 return BINARY_INFO[binary]
348 def checkOneCase(args, testcase):
349 """Check one specific case
351 Returns a dictionary containing failure information on error,
354 binary = testcase['binary']
355 accel = testcase['accel']
356 machine = testcase['machine']
357 device = testcase['device']
359 dbg("will test: %r", testcase)
361 args = ['-S', '-machine', '%s,accel=%s' % (machine, accel),
362 '-device', qemuOptsEscape(device)]
363 cmdline = ' '.join([binary] + args)
364 dbg("will launch QEMU: %s", cmdline)
365 vm = QEMUMachine(binary=binary, args=args, qmp_timer=15)
371 except Exception as this_exc:
373 exc_traceback = traceback.format_exc()
374 dbg("Exception while running test case")
380 if exc is not None or ec != 0:
382 'exc_traceback':exc_traceback,
389 def binariesToTest(args, testcase):
393 r = [f.path for f in os.scandir('.')
394 if f.name.startswith('qemu-system-') and
395 f.is_file() and os.access(f, os.X_OK)]
399 def accelsToTest(args, testcase):
400 if getBinaryInfo(args, testcase['binary']).kvm_available and not args.tcg_only:
405 def machinesToTest(args, testcase):
406 return getBinaryInfo(args, testcase['binary']).machines
409 def devicesToTest(args, testcase):
410 return getBinaryInfo(args, testcase['binary']).user_devs
413 TESTCASE_VARIABLES = [
414 ('binary', binariesToTest),
415 ('accel', accelsToTest),
416 ('machine', machinesToTest),
417 ('device', devicesToTest),
421 def genCases1(args, testcases, var, fn):
422 """Generate new testcases for one variable
424 If an existing item already has a variable set, don't
425 generate new items and just return it directly. This
426 allows the "-t" command-line option to be used to choose
427 a specific test case.
429 for testcase in testcases:
431 yield testcase.copy()
433 for i in fn(args, testcase):
439 def genCases(args, testcase):
440 """Generate test cases for all variables
442 cases = [testcase.copy()]
443 for var, fn in TESTCASE_VARIABLES:
444 dbg("var: %r, fn: %r", var, fn)
445 cases = genCases1(args, cases, var, fn)
449 def casesToTest(args, testcase):
450 cases = genCases(args, testcase)
453 cases = random.sample(cases, min(args.random, len(cases)))
456 dbg("%d test cases to test", len(cases))
459 random.shuffle(cases)
463 def logFailure(f, level):
465 logger.log(level, "failed: %s", formatTestCase(t))
466 logger.log(level, "cmdline: %s", f['cmdline'])
467 for l in f['log'].strip().split('\n'):
468 logger.log(level, "log: %s", l)
469 logger.log(level, "exit code: %r", f['exitcode'])
471 # If the Exception is merely a QMP connect error,
472 # reduce the logging level for its traceback to
473 # improve visual clarity.
474 if isinstance(f.get('exc'), ConnectError):
475 logger.log(level, "%s.%s: %s",
476 type(f['exc']).__module__,
477 type(f['exc']).__qualname__,
479 level = logging.DEBUG
481 if f['exc_traceback']:
482 logger.log(level, "exception:")
483 for l in f['exc_traceback'].split('\n'):
484 logger.log(level, " %s", l.rstrip('\n'))
488 parser = argparse.ArgumentParser(description="QEMU -device crash test")
489 parser.add_argument('-t', metavar='KEY=VALUE', nargs='*',
490 help="Limit test cases to KEY=VALUE",
491 action='append', dest='testcases', default=[])
492 parser.add_argument('-d', '--debug', action='store_true',
494 parser.add_argument('-v', '--verbose', action='store_true', default=True,
495 help='verbose output')
496 parser.add_argument('-q', '--quiet', dest='verbose', action='store_false',
497 help='non-verbose output')
498 parser.add_argument('-r', '--random', type=int, metavar='COUNT',
499 help='run a random sample of COUNT test cases',
501 parser.add_argument('--shuffle', action='store_true',
502 help='Run test cases in random order')
503 parser.add_argument('--dry-run', action='store_true',
504 help="Don't run any tests, just generate list")
505 parser.add_argument('-D', '--devtype', metavar='TYPE',
506 help="Test only device types that implement TYPE")
507 parser.add_argument('-Q', '--quick', action='store_true', default=True,
508 help="Quick mode: skip test cases that are expected to fail")
509 parser.add_argument('-F', '--full', action='store_false', dest='quick',
510 help="Full mode: test cases that are expected to fail")
511 parser.add_argument('--strict', action='store_true', dest='strict',
512 help="Treat all warnings as fatal")
513 parser.add_argument('--tcg-only', action='store_true', dest='tcg_only',
514 help="Only test with TCG accelerator")
515 parser.add_argument('qemu', nargs='*', metavar='QEMU',
516 help='QEMU binary to run')
517 args = parser.parse_args()
525 logging.basicConfig(stream=sys.stdout, level=lvl, format='%(levelname)s: %(message)s')
528 # Async QMP, when in use, is chatty about connection failures.
529 # This script knowingly generates a ton of connection errors.
530 # Silence this logger.
531 logging.getLogger('qemu.qmp.qmp_client').setLevel(logging.CRITICAL)
539 dbg("testcases: %r", args.testcases)
541 for t in chain(*args.testcases):
543 k, v = kv.split('=', 1)
546 if len(binariesToTest(args, tc)) == 0:
547 print("No QEMU binary found", file=sys.stderr)
548 parser.print_usage(sys.stderr)
551 for t in casesToTest(args, tc):
552 logger.info("running test case: %s", formatTestCase(t))
555 expected_match = findExpectedResult(t)
558 not getBinaryInfo(args, t['binary']).machineInfo(t['machine'])['runnable'])):
559 dbg("skipped: %s", formatTestCase(t))
567 f = checkOneCase(args, t)
568 except KeyboardInterrupt:
572 i, rule = checkResultRuleList(f)
573 dbg("testcase: %r, rule list match: %r", t, rule)
574 wl_stats.setdefault(i, []).append(f)
575 level = rule.get('loglevel', logging.DEBUG)
577 if rule.get('fatal') or (args.strict and level >= logging.WARN):
578 fatal_failures.append(f)
580 dbg("success: %s", formatTestCase(t))
582 logger.warn("Didn't fail as expected: %s", formatTestCase(t))
584 logger.info("Total: %d test cases", total)
586 logger.info("Skipped %d test cases", skipped)
589 stats = sorted([(len(wl_stats.get(i, [])), rule) for i, rule in
590 enumerate(ERROR_RULE_LIST)], key=lambda x: x[0])
591 for count, rule in stats:
592 dbg("error rule stats: %d: %r", count, rule)
595 for f in fatal_failures:
597 logger.error("Fatal failure: %s", formatTestCase(t))
598 logger.error("Fatal failures on some machine/device combinations")
601 if __name__ == '__main__':