]> git.proxmox.com Git - mirror_qemu.git/blame - scripts/dump-guest-memory.py
tcg/sparc: Zero extend address argument to ld/st helpers
[mirror_qemu.git] / scripts / dump-guest-memory.py
CommitLineData
28fbf8f6
JF
1"""
2This python script adds a new gdb command, "dump-guest-memory". It
3should be loaded with "source dump-guest-memory.py" at the (gdb)
4prompt.
5
6Copyright (C) 2013, Red Hat, Inc.
7
8Authors:
9 Laszlo Ersek <lersek@redhat.com>
10 Janosch Frank <frankja@linux.vnet.ibm.com>
11
12This work is licensed under the terms of the GNU GPL, version 2 or later. See
13the COPYING file in the top-level directory.
14"""
3e16d14f 15
368e3adc 16import ctypes
3e16d14f 17
47890203
JF
18UINTPTR_T = gdb.lookup_type("uintptr_t")
19
ca81ce72
JF
20TARGET_PAGE_SIZE = 0x1000
21TARGET_PAGE_MASK = 0xFFFFFFFFFFFFF000
22
ca81ce72
JF
23# Special value for e_phnum. This indicates that the real number of
24# program headers is too large to fit into e_phnum. Instead the real
25# value is in the field sh_info of section 0.
26PN_XNUM = 0xFFFF
27
368e3adc
JF
28EV_CURRENT = 1
29
30ELFCLASS32 = 1
31ELFCLASS64 = 2
32
33ELFDATA2LSB = 1
34ELFDATA2MSB = 2
35
36ET_CORE = 4
37
38PT_LOAD = 1
39PT_NOTE = 4
40
41EM_386 = 3
42EM_PPC = 20
43EM_PPC64 = 21
44EM_S390 = 22
45EM_AARCH = 183
46EM_X86_64 = 62
47
48class ELF(object):
49 """Representation of a ELF file."""
50
51 def __init__(self, arch):
52 self.ehdr = None
53 self.notes = []
54 self.segments = []
55 self.notes_size = 0
1d817db3 56 self.endianness = None
368e3adc
JF
57 self.elfclass = ELFCLASS64
58
59 if arch == 'aarch64-le':
1d817db3 60 self.endianness = ELFDATA2LSB
368e3adc 61 self.elfclass = ELFCLASS64
1d817db3 62 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
63 self.ehdr.e_machine = EM_AARCH
64
65 elif arch == 'aarch64-be':
1d817db3
SW
66 self.endianness = ELFDATA2MSB
67 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
68 self.ehdr.e_machine = EM_AARCH
69
70 elif arch == 'X86_64':
1d817db3
SW
71 self.endianness = ELFDATA2LSB
72 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
73 self.ehdr.e_machine = EM_X86_64
74
75 elif arch == '386':
1d817db3 76 self.endianness = ELFDATA2LSB
368e3adc 77 self.elfclass = ELFCLASS32
1d817db3 78 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
79 self.ehdr.e_machine = EM_386
80
81 elif arch == 's390':
1d817db3
SW
82 self.endianness = ELFDATA2MSB
83 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
84 self.ehdr.e_machine = EM_S390
85
86 elif arch == 'ppc64-le':
1d817db3
SW
87 self.endianness = ELFDATA2LSB
88 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
89 self.ehdr.e_machine = EM_PPC64
90
91 elif arch == 'ppc64-be':
1d817db3
SW
92 self.endianness = ELFDATA2MSB
93 self.ehdr = get_arch_ehdr(self.endianness, self.elfclass)
368e3adc
JF
94 self.ehdr.e_machine = EM_PPC64
95
96 else:
97 raise gdb.GdbError("No valid arch type specified.\n"
98 "Currently supported types:\n"
99 "aarch64-be, aarch64-le, X86_64, 386, s390, "
100 "ppc64-be, ppc64-le")
101
102 self.add_segment(PT_NOTE, 0, 0)
103
104 def add_note(self, n_name, n_desc, n_type):
105 """Adds a note to the ELF."""
106
1d817db3 107 note = get_arch_note(self.endianness, len(n_name), len(n_desc))
368e3adc
JF
108 note.n_namesz = len(n_name) + 1
109 note.n_descsz = len(n_desc)
110 note.n_name = n_name.encode()
111 note.n_type = n_type
112
113 # Desc needs to be 4 byte aligned (although the 64bit spec
114 # specifies 8 byte). When defining n_desc as uint32 it will be
115 # automatically aligned but we need the memmove to copy the
116 # string into it.
117 ctypes.memmove(note.n_desc, n_desc.encode(), len(n_desc))
118
119 self.notes.append(note)
120 self.segments[0].p_filesz += ctypes.sizeof(note)
121 self.segments[0].p_memsz += ctypes.sizeof(note)
122
123 def add_segment(self, p_type, p_paddr, p_size):
124 """Adds a segment to the elf."""
125
1d817db3 126 phdr = get_arch_phdr(self.endianness, self.elfclass)
368e3adc
JF
127 phdr.p_type = p_type
128 phdr.p_paddr = p_paddr
129 phdr.p_filesz = p_size
130 phdr.p_memsz = p_size
131 self.segments.append(phdr)
132 self.ehdr.e_phnum += 1
133
134 def to_file(self, elf_file):
135 """Writes all ELF structures to the the passed file.
136
137 Structure:
138 Ehdr
139 Segment 0:PT_NOTE
140 Segment 1:PT_LOAD
141 Segment N:PT_LOAD
142 Note 0..N
143 Dump contents
144 """
145 elf_file.write(self.ehdr)
146 off = ctypes.sizeof(self.ehdr) + \
147 len(self.segments) * ctypes.sizeof(self.segments[0])
148
149 for phdr in self.segments:
150 phdr.p_offset = off
151 elf_file.write(phdr)
152 off += phdr.p_filesz
153
154 for note in self.notes:
155 elf_file.write(note)
156
157
1d817db3
SW
158def get_arch_note(endianness, len_name, len_desc):
159 """Returns a Note class with the specified endianness."""
368e3adc 160
1d817db3 161 if endianness == ELFDATA2LSB:
368e3adc
JF
162 superclass = ctypes.LittleEndianStructure
163 else:
164 superclass = ctypes.BigEndianStructure
165
166 len_name = len_name + 1
167
168 class Note(superclass):
169 """Represents an ELF note, includes the content."""
170
171 _fields_ = [("n_namesz", ctypes.c_uint32),
172 ("n_descsz", ctypes.c_uint32),
173 ("n_type", ctypes.c_uint32),
174 ("n_name", ctypes.c_char * len_name),
175 ("n_desc", ctypes.c_uint32 * ((len_desc + 3) // 4))]
176 return Note()
177
178
179class Ident(ctypes.Structure):
180 """Represents the ELF ident array in the ehdr structure."""
181
182 _fields_ = [('ei_mag0', ctypes.c_ubyte),
183 ('ei_mag1', ctypes.c_ubyte),
184 ('ei_mag2', ctypes.c_ubyte),
185 ('ei_mag3', ctypes.c_ubyte),
186 ('ei_class', ctypes.c_ubyte),
187 ('ei_data', ctypes.c_ubyte),
188 ('ei_version', ctypes.c_ubyte),
189 ('ei_osabi', ctypes.c_ubyte),
190 ('ei_abiversion', ctypes.c_ubyte),
191 ('ei_pad', ctypes.c_ubyte * 7)]
192
1d817db3 193 def __init__(self, endianness, elfclass):
368e3adc
JF
194 self.ei_mag0 = 0x7F
195 self.ei_mag1 = ord('E')
196 self.ei_mag2 = ord('L')
197 self.ei_mag3 = ord('F')
198 self.ei_class = elfclass
1d817db3 199 self.ei_data = endianness
368e3adc
JF
200 self.ei_version = EV_CURRENT
201
202
1d817db3
SW
203def get_arch_ehdr(endianness, elfclass):
204 """Returns a EHDR64 class with the specified endianness."""
368e3adc 205
1d817db3 206 if endianness == ELFDATA2LSB:
368e3adc
JF
207 superclass = ctypes.LittleEndianStructure
208 else:
209 superclass = ctypes.BigEndianStructure
210
211 class EHDR64(superclass):
212 """Represents the 64 bit ELF header struct."""
213
214 _fields_ = [('e_ident', Ident),
215 ('e_type', ctypes.c_uint16),
216 ('e_machine', ctypes.c_uint16),
217 ('e_version', ctypes.c_uint32),
218 ('e_entry', ctypes.c_uint64),
219 ('e_phoff', ctypes.c_uint64),
220 ('e_shoff', ctypes.c_uint64),
221 ('e_flags', ctypes.c_uint32),
222 ('e_ehsize', ctypes.c_uint16),
223 ('e_phentsize', ctypes.c_uint16),
224 ('e_phnum', ctypes.c_uint16),
225 ('e_shentsize', ctypes.c_uint16),
226 ('e_shnum', ctypes.c_uint16),
227 ('e_shstrndx', ctypes.c_uint16)]
228
229 def __init__(self):
230 super(superclass, self).__init__()
1d817db3 231 self.e_ident = Ident(endianness, elfclass)
368e3adc
JF
232 self.e_type = ET_CORE
233 self.e_version = EV_CURRENT
234 self.e_ehsize = ctypes.sizeof(self)
235 self.e_phoff = ctypes.sizeof(self)
1d817db3 236 self.e_phentsize = ctypes.sizeof(get_arch_phdr(endianness, elfclass))
368e3adc
JF
237 self.e_phnum = 0
238
239
240 class EHDR32(superclass):
241 """Represents the 32 bit ELF header struct."""
242
243 _fields_ = [('e_ident', Ident),
244 ('e_type', ctypes.c_uint16),
245 ('e_machine', ctypes.c_uint16),
246 ('e_version', ctypes.c_uint32),
247 ('e_entry', ctypes.c_uint32),
248 ('e_phoff', ctypes.c_uint32),
249 ('e_shoff', ctypes.c_uint32),
250 ('e_flags', ctypes.c_uint32),
251 ('e_ehsize', ctypes.c_uint16),
252 ('e_phentsize', ctypes.c_uint16),
253 ('e_phnum', ctypes.c_uint16),
254 ('e_shentsize', ctypes.c_uint16),
255 ('e_shnum', ctypes.c_uint16),
256 ('e_shstrndx', ctypes.c_uint16)]
257
258 def __init__(self):
259 super(superclass, self).__init__()
1d817db3 260 self.e_ident = Ident(endianness, elfclass)
368e3adc
JF
261 self.e_type = ET_CORE
262 self.e_version = EV_CURRENT
263 self.e_ehsize = ctypes.sizeof(self)
264 self.e_phoff = ctypes.sizeof(self)
1d817db3 265 self.e_phentsize = ctypes.sizeof(get_arch_phdr(endianness, elfclass))
368e3adc
JF
266 self.e_phnum = 0
267
268 # End get_arch_ehdr
269 if elfclass == ELFCLASS64:
270 return EHDR64()
271 else:
272 return EHDR32()
273
274
1d817db3
SW
275def get_arch_phdr(endianness, elfclass):
276 """Returns a 32 or 64 bit PHDR class with the specified endianness."""
368e3adc 277
1d817db3 278 if endianness == ELFDATA2LSB:
368e3adc
JF
279 superclass = ctypes.LittleEndianStructure
280 else:
281 superclass = ctypes.BigEndianStructure
282
283 class PHDR64(superclass):
284 """Represents the 64 bit ELF program header struct."""
285
286 _fields_ = [('p_type', ctypes.c_uint32),
287 ('p_flags', ctypes.c_uint32),
288 ('p_offset', ctypes.c_uint64),
289 ('p_vaddr', ctypes.c_uint64),
290 ('p_paddr', ctypes.c_uint64),
291 ('p_filesz', ctypes.c_uint64),
292 ('p_memsz', ctypes.c_uint64),
293 ('p_align', ctypes.c_uint64)]
294
295 class PHDR32(superclass):
296 """Represents the 32 bit ELF program header struct."""
297
298 _fields_ = [('p_type', ctypes.c_uint32),
299 ('p_offset', ctypes.c_uint32),
300 ('p_vaddr', ctypes.c_uint32),
301 ('p_paddr', ctypes.c_uint32),
302 ('p_filesz', ctypes.c_uint32),
303 ('p_memsz', ctypes.c_uint32),
304 ('p_flags', ctypes.c_uint32),
305 ('p_align', ctypes.c_uint32)]
306
307 # End get_arch_phdr
308 if elfclass == ELFCLASS64:
309 return PHDR64()
310 else:
311 return PHDR32()
312
ca81ce72 313
47890203 314def int128_get64(val):
6782c0e7
JF
315 """Returns low 64bit part of Int128 struct."""
316
9b4b157e
MAL
317 try:
318 assert val["hi"] == 0
319 return val["lo"]
320 except gdb.error:
321 u64t = gdb.lookup_type('uint64_t').array(2)
322 u64 = val.cast(u64t)
323 if sys.byteorder == 'little':
324 assert u64[1] == 0
325 return u64[0]
326 else:
327 assert u64[0] == 0
328 return u64[1]
47890203 329
6782c0e7 330
47890203 331def qlist_foreach(head, field_str):
6782c0e7
JF
332 """Generator for qlists."""
333
47890203 334 var_p = head["lh_first"]
6782c0e7 335 while var_p != 0:
47890203 336 var = var_p.dereference()
47890203 337 var_p = var[field_str]["le_next"]
6782c0e7
JF
338 yield var
339
47890203 340
0878d0e1 341def qemu_map_ram_ptr(block, offset):
6782c0e7
JF
342 """Returns qemu vaddr for given guest physical address."""
343
0878d0e1 344 return block["host"] + offset
47890203 345
6782c0e7
JF
346
347def memory_region_get_ram_ptr(memory_region):
348 if memory_region["alias"] != 0:
349 return (memory_region_get_ram_ptr(memory_region["alias"].dereference())
350 + memory_region["alias_offset"])
351
0878d0e1 352 return qemu_map_ram_ptr(memory_region["ram_block"], 0)
6782c0e7 353
47890203
JF
354
355def get_guest_phys_blocks():
6782c0e7
JF
356 """Returns a list of ram blocks.
357
358 Each block entry contains:
359 'target_start': guest block phys start address
360 'target_end': guest block phys end address
361 'host_addr': qemu vaddr of the block's start
362 """
363
47890203 364 guest_phys_blocks = []
6782c0e7 365
7cb1089d
JF
366 print("guest RAM blocks:")
367 print("target_start target_end host_addr message "
368 "count")
369 print("---------------- ---------------- ---------------- ------- "
370 "-----")
47890203
JF
371
372 current_map_p = gdb.parse_and_eval("address_space_memory.current_map")
373 current_map = current_map_p.dereference()
7cb1089d
JF
374
375 # Conversion to int is needed for python 3
376 # compatibility. Otherwise range doesn't cast the value itself and
377 # breaks.
378 for cur in range(int(current_map["nr"])):
6782c0e7
JF
379 flat_range = (current_map["ranges"] + cur).dereference()
380 memory_region = flat_range["mr"].dereference()
47890203
JF
381
382 # we only care about RAM
6782c0e7 383 if not memory_region["ram"]:
47890203
JF
384 continue
385
386 section_size = int128_get64(flat_range["addr"]["size"])
387 target_start = int128_get64(flat_range["addr"]["start"])
6782c0e7
JF
388 target_end = target_start + section_size
389 host_addr = (memory_region_get_ram_ptr(memory_region)
390 + flat_range["offset_in_region"])
47890203
JF
391 predecessor = None
392
393 # find continuity in guest physical address space
6782c0e7 394 if len(guest_phys_blocks) > 0:
47890203
JF
395 predecessor = guest_phys_blocks[-1]
396 predecessor_size = (predecessor["target_end"] -
397 predecessor["target_start"])
398
399 # the memory API guarantees monotonically increasing
400 # traversal
6782c0e7 401 assert predecessor["target_end"] <= target_start
47890203
JF
402
403 # we want continuity in both guest-physical and
404 # host-virtual memory
405 if (predecessor["target_end"] < target_start or
406 predecessor["host_addr"] + predecessor_size != host_addr):
407 predecessor = None
408
6782c0e7 409 if predecessor is None:
47890203
JF
410 # isolated mapping, add it to the list
411 guest_phys_blocks.append({"target_start": target_start,
6782c0e7
JF
412 "target_end": target_end,
413 "host_addr": host_addr})
47890203
JF
414 message = "added"
415 else:
416 # expand predecessor until @target_end; predecessor's
417 # start doesn't change
418 predecessor["target_end"] = target_end
419 message = "joined"
420
7cb1089d
JF
421 print("%016x %016x %016x %-7s %5u" %
422 (target_start, target_end, host_addr.cast(UINTPTR_T),
423 message, len(guest_phys_blocks)))
47890203
JF
424
425 return guest_phys_blocks
426
427
28fbf8f6
JF
428# The leading docstring doesn't have idiomatic Python formatting. It is
429# printed by gdb's "help" command (the first line is printed in the
430# "help data" summary), and it should match how other help texts look in
431# gdb.
3e16d14f
LE
432class DumpGuestMemory(gdb.Command):
433 """Extract guest vmcore from qemu process coredump.
434
368e3adc
JF
435The two required arguments are FILE and ARCH:
436FILE identifies the target file to write the guest vmcore to.
437ARCH specifies the architecture for which the core will be generated.
3e16d14f
LE
438
439This GDB command reimplements the dump-guest-memory QMP command in
440python, using the representation of guest memory as captured in the qemu
441coredump. The qemu process that has been dumped must have had the
368e3adc 442command line option "-machine dump-guest-core=on" which is the default.
3e16d14f
LE
443
444For simplicity, the "paging", "begin" and "end" parameters of the QMP
445command are not supported -- no attempt is made to get the guest's
446internal paging structures (ie. paging=false is hard-wired), and guest
447memory is always fully dumped.
448
368e3adc
JF
449Currently aarch64-be, aarch64-le, X86_64, 386, s390, ppc64-be,
450ppc64-le guests are supported.
3e16d14f
LE
451
452The CORE/NT_PRSTATUS and QEMU notes (that is, the VCPUs' statuses) are
453not written to the vmcore. Preparing these would require context that is
454only present in the KVM host kernel module when the guest is alive. A
455fake ELF note is written instead, only to keep the ELF parser of "crash"
456happy.
457
458Dependent on how busted the qemu process was at the time of the
459coredump, this command might produce unpredictable results. If qemu
460deliberately called abort(), or it was dumped in response to a signal at
461a halfway fortunate point, then its coredump should be in reasonable
462shape and this command should mostly work."""
463
3e16d14f
LE
464 def __init__(self):
465 super(DumpGuestMemory, self).__init__("dump-guest-memory",
466 gdb.COMMAND_DATA,
467 gdb.COMPLETE_FILENAME)
368e3adc 468 self.elf = None
47890203 469 self.guest_phys_blocks = None
3e16d14f 470
368e3adc
JF
471 def dump_init(self, vmcore):
472 """Prepares and writes ELF structures to core file."""
473
474 # Needed to make crash happy, data for more useful notes is
475 # not available in a qemu core.
476 self.elf.add_note("NONE", "EMPTY", 0)
477
478 # We should never reach PN_XNUM for paging=false dumps,
479 # there's just a handful of discontiguous ranges after
480 # merging.
481 # The constant is needed to account for the PT_NOTE segment.
482 phdr_num = len(self.guest_phys_blocks) + 1
483 assert phdr_num < PN_XNUM
484
3e16d14f 485 for block in self.guest_phys_blocks:
368e3adc
JF
486 block_size = block["target_end"] - block["target_start"]
487 self.elf.add_segment(PT_LOAD, block["target_start"], block_size)
488
489 self.elf.to_file(vmcore)
3e16d14f
LE
490
491 def dump_iterate(self, vmcore):
368e3adc
JF
492 """Writes guest core to file."""
493
3e16d14f
LE
494 qemu_core = gdb.inferiors()[0]
495 for block in self.guest_phys_blocks:
6782c0e7 496 cur = block["host_addr"]
3e16d14f 497 left = block["target_end"] - block["target_start"]
7cb1089d
JF
498 print("dumping range at %016x for length %016x" %
499 (cur.cast(UINTPTR_T), left))
368e3adc 500
6782c0e7 501 while left > 0:
ca81ce72 502 chunk_size = min(TARGET_PAGE_SIZE, left)
3e16d14f
LE
503 chunk = qemu_core.read_memory(cur, chunk_size)
504 vmcore.write(chunk)
6782c0e7 505 cur += chunk_size
3e16d14f
LE
506 left -= chunk_size
507
3e16d14f 508 def invoke(self, args, from_tty):
368e3adc
JF
509 """Handles command invocation from gdb."""
510
3e16d14f
LE
511 # Unwittingly pressing the Enter key after the command should
512 # not dump the same multi-gig coredump to the same file.
513 self.dont_repeat()
514
515 argv = gdb.string_to_argv(args)
368e3adc
JF
516 if len(argv) != 2:
517 raise gdb.GdbError("usage: dump-guest-memory FILE ARCH")
518
519 self.elf = ELF(argv[1])
520 self.guest_phys_blocks = get_guest_phys_blocks()
3e16d14f 521
368e3adc
JF
522 with open(argv[0], "wb") as vmcore:
523 self.dump_init(vmcore)
524 self.dump_iterate(vmcore)
3e16d14f
LE
525
526DumpGuestMemory()