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/bus/pci/devices/*/class')
81 inputs['files'].add('/sys/devices/system/cpu/cpu*/uevent')
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,
220 def __init__(self, type, domain, bus, dev, fn, caps):
229 self.num_caps = len(caps)
230 self.num_msi_vectors = 0
232 self.num_msix_vectors = 0
233 self.msix_region_size = 0
234 self.msix_address = 0
236 if c.id in (0x05, 0x11):
237 msg_ctrl = struct.unpack('<H', c.content[:2])[0]
238 if c.id == 0x05: # MSI
239 self.num_msi_vectors = 1 << ((msg_ctrl >> 1) & 0x7)
240 self.msi_64bits = (msg_ctrl >> 7) & 1
242 vectors = (msg_ctrl & 0x7ff) + 1
243 self.num_msix_vectors = vectors
244 self.msix_region_size = (vectors * 16 + 0xfff) & 0xf000
245 self.msix_address = c.msix_address
248 return 'PCIDevice: %02x:%02x.%x' % (self.bus, self.dev, self.fn)
251 return self.bus << 8 | self.dev << 3 | self.fn
254 def parse_pcidevice_sysfsdir(basedir, dir):
255 dpath = os.path.join(basedir, dir)
256 dclass = input_readline(os.path.join(dpath, 'class'))
257 if re.match(r'0x0604..', dclass):
258 type = 'JAILHOUSE_PCI_TYPE_BRIDGE'
260 type = 'JAILHOUSE_PCI_TYPE_DEVICE'
262 domain = int(a[0], 16)
265 caps = PCICapability.parse_pcicaps(dpath)
266 return PCIDevice(type, domain, bus, int(df[0], 16), int(df[1], 16),
271 def __init__(self, start, stop, typestr, comments=None):
274 self.typestr = typestr
278 self.comments = comments
281 return 'MemRegion: %08x-%08x : %s' % \
282 (self.start, self.stop, self.typestr)
285 # round up to full PAGE_SIZE
286 return int((self.stop - self.start + 0xfff) / 0x1000) * 0x1000
288 def flagstr(self, p=''):
290 self.typestr == 'System RAM' or
291 self.typestr == 'Kernel' or
292 self.typestr == 'RAM buffer' or
293 self.typestr == 'ACPI DMAR RMRR'
295 s = 'JAILHOUSE_MEM_READ | JAILHOUSE_MEM_WRITE |\n'
296 s += p + '\t\tJAILHOUSE_MEM_EXECUTE | JAILHOUSE_MEM_DMA'
298 return 'JAILHOUSE_MEM_READ | JAILHOUSE_MEM_WRITE'
301 class IOMemRegionTree:
302 def __init__(self, region, level):
311 s = (' ' * (self.level - 1)) + str(self.region)
312 if self.parent and self.parent.region:
313 s += ' --> ' + self.parent.region.typestr
315 for c in self.children:
319 def regions_split_by_kernel(self):
320 kernel = [x for x in self.children if
321 x.region.typestr.startswith('Kernel ')]
323 if (len(kernel) == 0):
329 kernel_start = kernel[0].region.start
330 kernel_stop = kernel[len(kernel) - 1].region.stop
332 # align this for 16M, but only if we have enough space
333 kernel_stop = (kernel_stop & ~0xFFFFFF) + 0xFFFFFF
334 if (kernel_stop > r.stop):
340 # before Kernel if any
341 if (r.start < kernel_start):
342 before_kernel = MemRegion(r.start, kernel_start - 1, s)
344 kernel_region = MemRegion(kernel_start, kernel_stop, "Kernel")
346 # after Kernel if any
347 if (r.stop > kernel_stop):
348 after_kernel = MemRegion(kernel_stop + 1, r.stop, s)
350 return [before_kernel, kernel_region, after_kernel]
353 def parse_iomem_line(line):
354 a = line.split(':', 1)
355 level = int(a[0].count(' ') / 2) + 1
356 region = a[0].split('-', 1)
358 return level, MemRegion(int(region[0], 16), int(region[1], 16), a[1])
361 def parse_iomem_file():
362 root = IOMemRegionTree(None, 0)
363 f = input_open('/proc/iomem')
367 (level, r) = IOMemRegionTree.parse_iomem_line(line)
368 t = IOMemRegionTree(r, level)
369 if (t.level > lastlevel):
371 if (t.level == lastlevel):
372 t.parent = lastnode.parent
373 if (t.level < lastlevel):
375 while(t.level < p.level):
379 t.parent.children.append(t)
386 # find HPET regions in tree
388 def find_hpet_regions(tree):
391 for tree in tree.children:
395 if (s.find('HPET') >= 0):
398 # if the tree continues recurse further down ...
399 if (len(tree.children) > 0):
400 regions.extend(IOMemRegionTree.find_hpet_regions(tree))
404 # recurse down the tree
406 def parse_iomem_tree(tree):
409 for tree in tree.children:
413 # System RAM on the first level will be added completely,
414 # if they don't contain the kernel itself, if they do,
416 if (tree.level == 1 and s == 'System RAM'):
417 regions.extend(tree.regions_split_by_kernel())
420 # blacklisted on all levels
422 (s.find('PCI MMCONFIG') >= 0) or
423 (s.find('APIC') >= 0) or # covers both APIC and IOAPIC
424 (s.find('dmar') >= 0)
428 # generally blacklisted, unless we find an HPET behind it
429 if (s == 'reserved'):
430 regions.extend(IOMemRegionTree.find_hpet_regions(tree))
433 # if the tree continues recurse further down ...
434 if (len(tree.children) > 0):
435 regions.extend(IOMemRegionTree.parse_iomem_tree(tree))
438 # add all remaining leaves
444 def parse_iomem(pcidevices):
445 regions = IOMemRegionTree.parse_iomem_tree(
446 IOMemRegionTree.parse_iomem_file())
448 # filter the list for MSI-X pages
452 if d.msix_address >= r.start and d.msix_address <= r.stop:
453 if d.msix_address > r.start:
454 head_r = MemRegion(r.start, d.msix_address - 1,
455 r.typestr, r.comments)
457 if d.msix_address + d.msix_region_size < r.stop:
458 tail_r = MemRegion(d.msix_address + d.msix_region_size,
459 r.stop, r.typestr, r.comments)
466 # newer Linux kernels will report the first page as reserved
467 # it is needed for CPU init so include it anyways
468 if (ret[0].typestr == 'System RAM' and ret[0].start == 0x1000):
474 def parse_pcidevices():
477 basedir = '/sys/bus/pci/devices'
478 list = input_listdir(basedir, ['*/class', '*/config'])
480 d = PCIDevice.parse_pcidevice_sysfsdir(basedir, dir)
484 # look for duplicate capability patterns
486 if d2.caps == d.caps:
487 # reused existing capability list, but record all users
488 d2.caps[0].comments.append(str(d))
489 d.caps_start = d2.caps_start
493 d.caps[0].comments.append(str(d))
494 d.caps_start = len(caps)
497 return (devices, caps)
500 def parse_kernel_cmdline():
501 line = input_readline('/proc/cmdline')
502 m = re.match(r'.*memmap=([0-9a-fA-FxX]+)([KMG]?)\$'
503 '([0-9a-fA-FxX]+)([KMG]?).*',
506 size = kmg_multiply(int(m.group(1), 0), m.group(2))
507 start = kmg_multiply(int(m.group(3), 0), m.group(4))
512 def alloc_mem(regions, size):
513 mem = [0x3b000000, size]
516 r.typestr == 'System RAM' and
517 r.start <= mem[0] and
518 r.stop + 1 >= mem[0] + mem[1]
521 head_r = MemRegion(r.start, mem[0] - 1, r.typestr, r.comments)
522 regions.insert(regions.index(r), head_r)
523 if r.stop + 1 > mem[0] + mem[1]:
524 tail_r = MemRegion(mem[0] + mem[1], r.stop, r.typestr,
526 regions.insert(regions.index(r), tail_r)
529 for r in reversed(regions):
530 if (r.typestr == 'System RAM' and r.size() >= mem[1]):
534 raise RuntimeError('failed to allocate memory')
538 list = input_listdir('/sys/devices/system/cpu', ['cpu*/uevent'])
541 if re.match(r'cpu[0-9]+', f):
546 def parse_dmar_devscope(f):
547 (scope_type, scope_len, bus, dev, fn) = \
548 struct.unpack('<BBxxxBBB', f.read(8))
550 raise RuntimeError('Unsupported DMAR Device Scope Structure')
551 return (scope_type, scope_len, bus, dev, fn)
554 # parsing of DMAR ACPI Table
555 # see Intel VT-d Spec chapter 8
556 def parse_dmar(pcidevices):
557 f = input_open('/sys/firmware/acpi/tables/DMAR', 'rb')
558 signature = f.read(4)
559 if signature != b'DMAR':
560 raise RuntimeError('DMAR: incorrect input file format %s' % signature)
561 (length,) = struct.unpack('<I', f.read(4))
570 (struct_type, struct_len) = struct.unpack('<HH', f.read(4))
574 # DMA Remapping Hardware Unit Definition
576 (flags, segment, base) = struct.unpack('<BxHQ', f.read(12))
578 raise RuntimeError('We do not support multiple PCI segments')
580 raise RuntimeError('Too many DMAR units. '
581 'Raise JAILHOUSE_MAX_DMAR_UNITS.')
586 d.iommu = len(units) - 1
587 offset += 16 - offset
588 while offset < struct_len:
589 (scope_type, scope_len, bus, dev, fn) =\
590 parse_dmar_devscope(f)
593 if d.bus == bus and d.dev == dev and d.fn == fn:
594 d.iommu = len(units) - 1
595 elif scope_type == 2:
596 raise RuntimeError('Unsupported DMAR Device Scope type')
597 elif scope_type == 3:
599 raise RuntimeError('We do not support more '
601 # encode the DMAR unit number into the device ID
602 ioapic_id = ((len(units) - 1) << 16) | \
603 (bus << 8) | (dev << 3) | fn
606 # Reserved Memory Region Reporting Structure
608 f.seek(8 - offset, os.SEEK_CUR)
610 (base, limit) = struct.unpack('<QQ', f.read(16))
614 while offset < struct_len:
615 (scope_type, scope_len, bus, dev, fn) =\
616 parse_dmar_devscope(f)
618 comments.append('PCI device: %02x:%02x.%x' %
621 comments.append('DMAR parser could not decode device path')
624 reg = MemRegion(base, limit, 'ACPI DMAR RMRR', comments)
627 f.seek(struct_len - offset, os.SEEK_CUR)
629 return units, ioapic_id, regions
634 f = input_open('/proc/ioports')
636 if line.endswith('ACPI PM_TMR\n'):
637 pm_timer_base = int(line.split('-')[0], 16)
644 def __init__(self, base, end_bus):
646 self.end_bus = end_bus
650 f = input_open('/sys/firmware/acpi/tables/MCFG', 'rb')
651 signature = f.read(4)
652 if signature != b'MCFG':
653 raise RuntimeError('MCFG: incorrect input file format %s' %
655 (length,) = struct.unpack('<I', f.read(4))
657 raise RuntimeError('Multiple MMCONFIG regions found! '
658 'This is not supported')
660 (base, segment, start_bus, end_bus) = \
661 struct.unpack('<QHBB', f.read(12))
662 if segment != 0 or start_bus != 0:
663 raise RuntimeError('Invalid MCFG structure found')
664 return MMConfig(base, end_bus)
667 def get_cpu_vendor():
669 if cpuvendor is not None:
671 with input_open('/proc/cpuinfo', 'r') as f:
675 key, value = line.split(':')
676 if key.strip() == 'vendor_id':
677 cpuvendor = value.strip()
681 if options.generate_collector:
682 f = open(options.file, 'w')
683 filelist = ' '.join(inputs['files'])
684 filelist_opt = ' '.join(inputs['files_opt'])
685 filelist_intel = ' '.join(inputs['files_intel'])
687 tmpl = Template(filename=os.path.join(options.template_dir,
688 'jailhouse-config-collect.tmpl'))
689 f.write(tmpl.render(filelist=filelist, filelist_opt=filelist_opt,
690 filelist_intel=filelist_intel))
694 if ((options.root is '/') and (os.geteuid() is not 0)):
695 print('ERROR: You have to be root to work on "/"!', file=sys.stderr)
698 jh_enabled = input_readline('/sys/devices/jailhouse/enabled', True).rstrip()
699 if jh_enabled == '1':
700 print('ERROR: Jailhouse was enabled when collecting input files! '
701 'Disable jailhouse and try again.',
705 (pcidevices, pcicaps) = parse_pcidevices()
707 product = [input_readline('/sys/class/dmi/id/sys_vendor',
709 input_readline('/sys/class/dmi/id/product_name',
713 inmatemem = kmg_multiply_str(options.mem_inmates)
714 hvmem = [0, kmg_multiply_str(options.mem_hv)]
716 regions = parse_iomem(pcidevices)
717 ourmem = parse_kernel_cmdline()
718 total = hvmem[1] + inmatemem
720 mmconfig = MMConfig.parse()
722 if get_cpu_vendor() == 'GenuineIntel':
723 (dmar_units, ioapic_id, rmrr_regs) = parse_dmar(pcidevices)
725 (dmar_units, ioapic_id, rmrr_regs) = [], 0, []
729 if get_cpu_vendor() == 'AuthenticAMD':
730 d.iommu = 0 # temporary workaround
732 raise RuntimeError('PCI device %02x:%02x.%x outside the scope of an '
733 'IOMMU' % (d.bus, d.dev, d.fn))
735 # kernel does not have memmap region, pick one
737 ourmem = alloc_mem(regions, total)
738 elif (total > ourmem[1]):
739 raise RuntimeError('Your memmap reservation is too small you need >="' +
744 inmatereg = MemRegion(ourmem[0] + hvmem[1],
745 ourmem[0] + hvmem[1] + inmatemem - 1,
746 'JAILHOUSE Inmate Memory')
747 regions.append(inmatereg)
749 cpucount = count_cpus()
751 pm_timer_base = parse_ioports()
754 f = open(options.file, 'w')
755 tmpl = Template(filename=os.path.join(options.template_dir,
756 'root-cell-config.c.tmpl'))
757 f.write(tmpl.render(regions=regions,
759 argstr=' '.join(sys.argv),
762 pcidevices=pcidevices,
766 pm_timer_base=pm_timer_base,
768 dmar_units=dmar_units))