3 # Jailhouse, a Linux-based partitioning hypervisor
5 # Copyright (c) Siemens AG, 2014
7 # This work is licensed under the terms of the GNU GPL, version 2. See
8 # the COPYING file in the top-level directory.
10 # This script should help to create a basic jailhouse configuration file.
11 # It needs to be executed on the target machine, where it will gather
12 # information about the system. For more advanced scenarios you will have
13 # to change the generated C-code.
15 from __future__ import print_function
22 from mako.template import Template
27 template_default_dir = datadir + "/jailhouse"
29 template_default_dir = os.path.abspath(os.path.dirname(sys.argv[0]))
33 # pretend to be part of the jailhouse tool
34 sys.argv[0] = sys.argv[0].replace('-', ' ')
36 parser = argparse.ArgumentParser()
37 parser.add_argument('-g', '--generate-collector',
38 help='generate a script to collect input files on '
41 parser.add_argument('-r', '--root',
42 help='gather information in ROOT/, the default is "/" '
43 'which means creating a config for localhost',
47 parser.add_argument('-t', '--template-dir',
48 help='the directory where the templates are located,'
49 'the default is "' + template_default_dir + '"',
50 default=template_default_dir,
54 memargs = [['--mem-inmates', '2M', 'inmate'],
55 ['--mem-hv', '64M', 'hypervisor']]
58 parser.add_argument(entry[0],
59 help='the amount of ' + entry[2] +
60 ' memory, default is "' + entry[1] +
61 '", format "xxx[K|M|G]"',
66 parser.add_argument('file', metavar='FILE',
67 help='name of file to write out',
70 options = parser.parse_args()
72 inputs = {'files': set(), 'files_opt': set(), 'files_intel': set()}
75 inputs['files'].add('/proc/iomem')
76 inputs['files'].add('/proc/cpuinfo')
77 inputs['files'].add('/proc/cmdline')
78 inputs['files'].add('/proc/ioports')
79 inputs['files'].add('/sys/bus/pci/devices/*/config')
80 inputs['files'].add('/sys/devices/system/cpu/cpu*/uevent')
81 inputs['files'].add('/sys/firmware/acpi/tables/APIC')
82 inputs['files'].add('/sys/firmware/acpi/tables/MCFG')
84 inputs['files_opt'].add('/sys/class/dmi/id/product_name')
85 inputs['files_opt'].add('/sys/class/dmi/id/sys_vendor')
86 inputs['files_opt'].add('/sys/devices/jailhouse/enabled')
87 # platform specific files
88 inputs['files_intel'].add('/sys/firmware/acpi/tables/DMAR')
91 def kmg_multiply(value, kmg):
92 if (kmg == 'K' or kmg == 'k'):
94 if (kmg == 'M' or kmg == 'm'):
95 return 1024**2 * value
96 if (kmg == 'G' or kmg == 'g'):
97 return 1024**3 * value
101 def kmg_multiply_str(str):
102 m = re.match(r'([0-9a-fA-FxX]+)([KMG]?)', str)
104 return kmg_multiply(int(m.group(1)), m.group(2))
105 raise RuntimeError('kmg_multiply_str can not parse input "' + str + '"')
108 def check_input_listed(name, optional=False):
109 set = inputs['files_opt']
110 if optional is False:
111 set = inputs['files']
113 if cpuvendor == 'GenuineIntel':
114 set = set.union(inputs['files_intel'])
117 if fnmatch.fnmatch(name, file):
119 raise RuntimeError('"' + name + '" is not a listed input file')
122 def input_open(name, mode='r', optional=False):
123 check_input_listed(name, optional)
125 f = open(options.root + name, mode)
126 except Exception as e:
128 return open("/dev/null", mode)
133 def input_readline(name, optional=False):
134 f = input_open(name, optional=optional)
140 def input_listdir(dir, wildcards):
142 check_input_listed(os.path.join(dir, w))
143 dirs = os.listdir(options.root + dir)
149 def __init__(self, id, start, len, flags, content, msix_address):
154 self.content = content
155 self.msix_address = msix_address
158 def __eq__(self, other):
159 return self.id == other.id and self.start == other.start and \
160 self.len == other.len and self.flags == other.flags
163 RW = 'JAILHOUSE_PCICAPS_WRITE'
166 def parse_pcicaps(dir):
168 f = input_open(os.path.join(dir, 'config'), 'rb')
170 (status,) = struct.unpack('<H', f.read(2))
171 # capability list supported?
172 if (status & (1 << 4)) == 0:
175 # walk capability list
177 (next,) = struct.unpack('B', f.read(1))
182 (id, next) = struct.unpack('<BB', f.read(2))
183 if id == 0x01: # Power Management
184 # this cap can be handed out completely
186 flags = PCICapability.RW
187 elif id == 0x05: # MSI
188 # access will be moderated by hypervisor
190 (msgctl,) = struct.unpack('<H', f.read(2))
191 if (msgctl & (1 << 7)) != 0: # 64-bit support
193 if (msgctl & (1 << 8)) != 0: # per-vector masking support
195 flags = PCICapability.RW
196 elif id == 0x11: # MSI-X
197 # access will be moderated by hypervisor
199 (table,) = struct.unpack('<xxI', f.read(6))
200 f.seek(0x10 + (table & 7) * 4)
201 (bar,) = struct.unpack('<I', f.read(4))
203 raise RuntimeError('Invalid MSI-X BAR found')
205 bar |= struct.unpack('<I', f.read(4))[0] << 32
206 msix_address = (bar & 0xfffffffffffffff0) + table & 0xfffffff8
207 flags = PCICapability.RW
209 # unknown/unhandled cap, mark its existence
211 flags = PCICapability.RD
213 content = f.read(len - 2)
214 caps.append(PCICapability(id, cap, len, flags, content,
221 def __init__(self, type, domain, bus, dev, fn, caps, path):
231 self.num_caps = len(caps)
232 self.num_msi_vectors = 0
234 self.num_msix_vectors = 0
235 self.msix_region_size = 0
236 self.msix_address = 0
238 if c.id in (0x05, 0x11):
239 msg_ctrl = struct.unpack('<H', c.content[:2])[0]
240 if c.id == 0x05: # MSI
241 self.num_msi_vectors = 1 << ((msg_ctrl >> 1) & 0x7)
242 self.msi_64bits = (msg_ctrl >> 7) & 1
244 vectors = (msg_ctrl & 0x7ff) + 1
245 self.num_msix_vectors = vectors
246 self.msix_region_size = (vectors * 16 + 0xfff) & 0xf000
247 self.msix_address = c.msix_address
250 return 'PCIDevice: %02x:%02x.%x' % (self.bus, self.dev, self.fn)
253 return self.bus << 8 | self.dev << 3 | self.fn
256 def parse_pcidevice_sysfsdir(basedir, dir):
257 dpath = os.path.join(basedir, dir)
258 f = input_open(os.path.join(dpath, 'config'), 'rb')
260 (classcode,) = struct.unpack('<H', f.read(2))
262 if classcode == 0x0604:
263 type = 'JAILHOUSE_PCI_TYPE_BRIDGE'
265 type = 'JAILHOUSE_PCI_TYPE_DEVICE'
267 domain = int(a[0], 16)
270 caps = PCICapability.parse_pcicaps(dpath)
271 return PCIDevice(type, domain, bus, int(df[0], 16), int(df[1], 16),
275 class PCIPCIBridge(PCIDevice):
277 def get_2nd_busses(dev):
278 assert dev.type == 'JAILHOUSE_PCI_TYPE_BRIDGE'
279 f = input_open(os.path.join(dev.path, 'config'), 'rb')
281 (secondbus, subordinate) = struct.unpack('<BB', f.read(2))
283 return (secondbus, subordinate)
287 def __init__(self, start, stop, typestr, comments=None):
290 self.typestr = typestr
294 self.comments = comments
297 return 'MemRegion: %08x-%08x : %s' % \
298 (self.start, self.stop, self.typestr)
301 # round up to full PAGE_SIZE
302 return int((self.stop - self.start + 0xfff) / 0x1000) * 0x1000
304 def flagstr(self, p=''):
306 self.typestr == 'System RAM' or
307 self.typestr == 'Kernel' or
308 self.typestr == 'RAM buffer' or
309 self.typestr == 'ACPI DMAR RMRR'
311 s = 'JAILHOUSE_MEM_READ | JAILHOUSE_MEM_WRITE |\n'
312 s += p + '\t\tJAILHOUSE_MEM_EXECUTE | JAILHOUSE_MEM_DMA'
314 return 'JAILHOUSE_MEM_READ | JAILHOUSE_MEM_WRITE'
318 def __init__(self, id, address, gsi_base, iommu=0, bdf=0):
320 self.address = address
321 self.gsi_base = gsi_base
326 return 'IOAPIC %d, GSI base %d' % (self.id, self.gsi_base)
328 def irqchip_id(self):
329 # encode the IOMMU number into the irqchip ID
330 return (self.iommu << 16) | self.bdf
333 class IOMemRegionTree:
334 def __init__(self, region, level):
343 s = (' ' * (self.level - 1)) + str(self.region)
344 if self.parent and self.parent.region:
345 s += ' --> ' + self.parent.region.typestr
347 for c in self.children:
351 def regions_split_by_kernel(self):
352 kernel = [x for x in self.children if
353 x.region.typestr.startswith('Kernel ')]
355 if (len(kernel) == 0):
361 kernel_start = kernel[0].region.start
362 kernel_stop = kernel[len(kernel) - 1].region.stop
364 # align this for 16M, but only if we have enough space
365 kernel_stop = (kernel_stop & ~0xFFFFFF) + 0xFFFFFF
366 if (kernel_stop > r.stop):
372 # before Kernel if any
373 if (r.start < kernel_start):
374 before_kernel = MemRegion(r.start, kernel_start - 1, s)
376 kernel_region = MemRegion(kernel_start, kernel_stop, "Kernel")
378 # after Kernel if any
379 if (r.stop > kernel_stop):
380 after_kernel = MemRegion(kernel_stop + 1, r.stop, s)
382 return [before_kernel, kernel_region, after_kernel]
385 def parse_iomem_line(line):
386 a = line.split(':', 1)
387 level = int(a[0].count(' ') / 2) + 1
388 region = a[0].split('-', 1)
390 return level, MemRegion(int(region[0], 16), int(region[1], 16), a[1])
393 def parse_iomem_file():
394 root = IOMemRegionTree(None, 0)
395 f = input_open('/proc/iomem')
399 (level, r) = IOMemRegionTree.parse_iomem_line(line)
400 t = IOMemRegionTree(r, level)
401 if (t.level > lastlevel):
403 if (t.level == lastlevel):
404 t.parent = lastnode.parent
405 if (t.level < lastlevel):
407 while(t.level < p.level):
411 t.parent.children.append(t)
418 # find HPET regions in tree
420 def find_hpet_regions(tree):
423 for tree in tree.children:
427 if (s.find('HPET') >= 0):
430 # if the tree continues recurse further down ...
431 if (len(tree.children) > 0):
432 regions.extend(IOMemRegionTree.find_hpet_regions(tree))
436 # recurse down the tree
438 def parse_iomem_tree(tree):
441 for tree in tree.children:
445 # System RAM on the first level will be added completely,
446 # if they don't contain the kernel itself, if they do,
448 if (tree.level == 1 and s == 'System RAM'):
449 regions.extend(tree.regions_split_by_kernel())
452 # blacklisted on all levels
454 (s.find('PCI MMCONFIG') >= 0) or
455 (s.find('APIC') >= 0) or # covers both APIC and IOAPIC
456 (s.find('dmar') >= 0)
460 # generally blacklisted, unless we find an HPET behind it
461 if (s == 'reserved'):
462 regions.extend(IOMemRegionTree.find_hpet_regions(tree))
465 # if the tree continues recurse further down ...
466 if (len(tree.children) > 0):
467 regions.extend(IOMemRegionTree.parse_iomem_tree(tree))
470 # add all remaining leaves
476 def parse_iomem(pcidevices):
477 regions = IOMemRegionTree.parse_iomem_tree(
478 IOMemRegionTree.parse_iomem_file())
480 # filter the list for MSI-X pages
484 if d.msix_address >= r.start and d.msix_address <= r.stop:
485 if d.msix_address > r.start:
486 head_r = MemRegion(r.start, d.msix_address - 1,
487 r.typestr, r.comments)
489 if d.msix_address + d.msix_region_size < r.stop:
490 tail_r = MemRegion(d.msix_address + d.msix_region_size,
491 r.stop, r.typestr, r.comments)
498 # newer Linux kernels will report the first page as reserved
499 # it is needed for CPU init so include it anyways
500 if (ret[0].typestr == 'System RAM' and ret[0].start == 0x1000):
506 def parse_pcidevices():
509 basedir = '/sys/bus/pci/devices'
510 list = input_listdir(basedir, ['*/config'])
512 d = PCIDevice.parse_pcidevice_sysfsdir(basedir, dir)
516 # look for duplicate capability patterns
518 if d2.caps == d.caps:
519 # reused existing capability list, but record all users
520 d2.caps[0].comments.append(str(d))
521 d.caps_start = d2.caps_start
525 d.caps[0].comments.append(str(d))
526 d.caps_start = len(caps)
529 return (devices, caps)
532 def parse_kernel_cmdline():
533 line = input_readline('/proc/cmdline')
534 m = re.match(r'.*memmap=([0-9a-fA-FxX]+)([KMG]?)\$'
535 '([0-9a-fA-FxX]+)([KMG]?).*',
538 size = kmg_multiply(int(m.group(1), 0), m.group(2))
539 start = kmg_multiply(int(m.group(3), 0), m.group(4))
544 def alloc_mem(regions, size):
545 mem = [0x3b000000, size]
548 r.typestr == 'System RAM' and
549 r.start <= mem[0] and
550 r.stop + 1 >= mem[0] + mem[1]
553 head_r = MemRegion(r.start, mem[0] - 1, r.typestr, r.comments)
554 regions.insert(regions.index(r), head_r)
555 if r.stop + 1 > mem[0] + mem[1]:
556 tail_r = MemRegion(mem[0] + mem[1], r.stop, r.typestr,
558 regions.insert(regions.index(r), tail_r)
561 for r in reversed(regions):
562 if (r.typestr == 'System RAM' and r.size() >= mem[1]):
566 raise RuntimeError('failed to allocate memory')
570 list = input_listdir('/sys/devices/system/cpu', ['cpu*/uevent'])
573 if re.match(r'cpu[0-9]+', f):
579 f = input_open('/sys/firmware/acpi/tables/APIC', 'rb')
580 signature = f.read(4)
581 if signature != b'APIC':
582 raise RuntimeError('MADT: incorrect input file format %s' % signature)
583 (length,) = struct.unpack('<I', f.read(4))
590 (struct_type, struct_len) = struct.unpack('<BB', f.read(2))
595 (id, address, gsi_base) = struct.unpack('<BxII', f.read(10))
597 ioapics.append(IOAPIC(id, address, gsi_base))
599 f.seek(struct_len - offset, os.SEEK_CUR)
605 def parse_dmar_devscope(f):
606 (scope_type, scope_len, id, bus, dev, fn) = \
607 struct.unpack('<BBxxBBBB', f.read(8))
609 raise RuntimeError('Unsupported DMAR Device Scope Structure')
610 return (scope_type, scope_len, id, bus, dev, fn)
613 # parsing of DMAR ACPI Table
614 # see Intel VT-d Spec chapter 8
615 def parse_dmar(pcidevices, ioapics):
616 f = input_open('/sys/firmware/acpi/tables/DMAR', 'rb')
617 signature = f.read(4)
618 if signature != b'DMAR':
619 raise RuntimeError('DMAR: incorrect input file format %s' % signature)
620 (length,) = struct.unpack('<I', f.read(4))
628 (struct_type, struct_len) = struct.unpack('<HH', f.read(4))
632 # DMA Remapping Hardware Unit Definition
634 (flags, segment, base) = struct.unpack('<BxHQ', f.read(12))
636 raise RuntimeError('We do not support multiple PCI segments')
638 raise RuntimeError('Too many DMAR units. '
639 'Raise JAILHOUSE_MAX_DMAR_UNITS.')
644 d.iommu = len(units) - 1
645 offset += 16 - offset
646 while offset < struct_len:
647 (scope_type, scope_len, id, bus, dev, fn) =\
648 parse_dmar_devscope(f)
649 # PCI Endpoint Device
651 assert not (flags & 1)
653 if d.bus == bus and d.dev == dev and d.fn == fn:
654 d.iommu = len(units) - 1
657 elif scope_type == 2:
658 assert not (flags & 1)
660 if d.bus == bus and d.dev == dev and d.fn == fn:
661 (secondbus, subordinate) = \
662 PCIPCIBridge.get_2nd_busses(d)
663 for d2 in pcidevices:
665 d2.bus >= secondbus and
666 d2.bus <= subordinate
668 d2.iommu = len(units) - 1
671 elif scope_type == 3:
672 ioapic = next(chip for chip in ioapics if chip.id == id)
673 bdf = (bus << 8) | (dev << 3) | fn
676 raise RuntimeError('IOAPICs with identical BDF')
678 ioapic.dmar_unit = len(units) - 1
681 # Reserved Memory Region Reporting Structure
683 f.seek(8 - offset, os.SEEK_CUR)
685 (base, limit) = struct.unpack('<QQ', f.read(16))
689 while offset < struct_len:
690 (scope_type, scope_len, id, bus, dev, fn) =\
691 parse_dmar_devscope(f)
693 comments.append('PCI device: %02x:%02x.%x' %
696 comments.append('DMAR parser could not decode device path')
699 reg = MemRegion(base, limit, 'ACPI DMAR RMRR', comments)
702 f.seek(struct_len - offset, os.SEEK_CUR)
705 return units, regions
710 f = input_open('/proc/ioports')
712 if line.endswith('ACPI PM_TMR\n'):
713 pm_timer_base = int(line.split('-')[0], 16)
720 def __init__(self, base, end_bus):
722 self.end_bus = end_bus
726 f = input_open('/sys/firmware/acpi/tables/MCFG', 'rb')
727 signature = f.read(4)
728 if signature != b'MCFG':
729 raise RuntimeError('MCFG: incorrect input file format %s' %
731 (length,) = struct.unpack('<I', f.read(4))
733 raise RuntimeError('Multiple MMCONFIG regions found! '
734 'This is not supported')
736 (base, segment, start_bus, end_bus) = \
737 struct.unpack('<QHBB', f.read(12))
738 if segment != 0 or start_bus != 0:
739 raise RuntimeError('Invalid MCFG structure found')
740 return MMConfig(base, end_bus)
743 def get_cpu_vendor():
745 if cpuvendor is not None:
747 with input_open('/proc/cpuinfo', 'r') as f:
751 key, value = line.split(':')
752 if key.strip() == 'vendor_id':
753 cpuvendor = value.strip()
757 if options.generate_collector:
758 f = open(options.file, 'w')
759 filelist = ' '.join(inputs['files'])
760 filelist_opt = ' '.join(inputs['files_opt'])
761 filelist_intel = ' '.join(inputs['files_intel'])
763 tmpl = Template(filename=os.path.join(options.template_dir,
764 'jailhouse-config-collect.tmpl'))
765 f.write(tmpl.render(filelist=filelist, filelist_opt=filelist_opt,
766 filelist_intel=filelist_intel))
770 if ((options.root is '/') and (os.geteuid() is not 0)):
771 print('ERROR: You have to be root to work on "/"!', file=sys.stderr)
774 jh_enabled = input_readline('/sys/devices/jailhouse/enabled', True).rstrip()
775 if jh_enabled == '1':
776 print('ERROR: Jailhouse was enabled when collecting input files! '
777 'Disable jailhouse and try again.',
781 (pcidevices, pcicaps) = parse_pcidevices()
783 product = [input_readline('/sys/class/dmi/id/sys_vendor',
785 input_readline('/sys/class/dmi/id/product_name',
789 inmatemem = kmg_multiply_str(options.mem_inmates)
790 hvmem = [0, kmg_multiply_str(options.mem_hv)]
792 regions = parse_iomem(pcidevices)
793 ourmem = parse_kernel_cmdline()
794 total = hvmem[1] + inmatemem
796 mmconfig = MMConfig.parse()
798 ioapics = parse_madt()
800 if get_cpu_vendor() == 'GenuineIntel':
801 (dmar_units, rmrr_regs) = parse_dmar(pcidevices, ioapics)
803 (dmar_units, rmrr_regs) = [], []
807 if get_cpu_vendor() == 'AuthenticAMD':
808 d.iommu = 0 # temporary workaround
810 raise RuntimeError('PCI device %02x:%02x.%x outside the scope of an '
811 'IOMMU' % (d.bus, d.dev, d.fn))
813 # kernel does not have memmap region, pick one
815 ourmem = alloc_mem(regions, total)
816 elif (total > ourmem[1]):
817 raise RuntimeError('Your memmap reservation is too small you need >="' +
822 inmatereg = MemRegion(ourmem[0] + hvmem[1],
823 ourmem[0] + hvmem[1] + inmatemem - 1,
824 'JAILHOUSE Inmate Memory')
825 regions.append(inmatereg)
827 cpucount = count_cpus()
829 pm_timer_base = parse_ioports()
832 f = open(options.file, 'w')
833 tmpl = Template(filename=os.path.join(options.template_dir,
834 'root-cell-config.c.tmpl'))
835 f.write(tmpl.render(regions=regions,
837 argstr=' '.join(sys.argv),
840 pcidevices=pcidevices,
844 pm_timer_base=pm_timer_base,
846 dmar_units=dmar_units))