Adding post-execute scripts to cyclictest execution
[kvmfornfv.git] / qemu / scripts / dump-guest-memory.py
index dc8e44a..c0a2e99 100644 (file)
-# This python script adds a new gdb command, "dump-guest-memory". It
-# should be loaded with "source dump-guest-memory.py" at the (gdb)
-# prompt.
-#
-# Copyright (C) 2013, Red Hat, Inc.
-#
-# Authors:
-#   Laszlo Ersek <lersek@redhat.com>
-#
-# This work is licensed under the terms of the GNU GPL, version 2 or later. See
-# the COPYING file in the top-level directory.
-#
+"""
+This python script adds a new gdb command, "dump-guest-memory". It
+should be loaded with "source dump-guest-memory.py" at the (gdb)
+prompt.
+
+Copyright (C) 2013, Red Hat, Inc.
+
+Authors:
+   Laszlo Ersek <lersek@redhat.com>
+   Janosch Frank <frankja@linux.vnet.ibm.com>
+
+This work is licensed under the terms of the GNU GPL, version 2 or later. See
+the COPYING file in the top-level directory.
+"""
+
+import ctypes
+
+UINTPTR_T = gdb.lookup_type("uintptr_t")
+
+TARGET_PAGE_SIZE = 0x1000
+TARGET_PAGE_MASK = 0xFFFFFFFFFFFFF000
+
+# Special value for e_phnum. This indicates that the real number of
+# program headers is too large to fit into e_phnum. Instead the real
+# value is in the field sh_info of section 0.
+PN_XNUM = 0xFFFF
+
+EV_CURRENT = 1
+
+ELFCLASS32 = 1
+ELFCLASS64 = 2
+
+ELFDATA2LSB = 1
+ELFDATA2MSB = 2
+
+ET_CORE = 4
+
+PT_LOAD = 1
+PT_NOTE = 4
+
+EM_386 = 3
+EM_PPC = 20
+EM_PPC64 = 21
+EM_S390 = 22
+EM_AARCH = 183
+EM_X86_64 = 62
+
+class ELF(object):
+    """Representation of a ELF file."""
+
+    def __init__(self, arch):
+        self.ehdr = None
+        self.notes = []
+        self.segments = []
+        self.notes_size = 0
+        self.endianess = None
+        self.elfclass = ELFCLASS64
+
+        if arch == 'aarch64-le':
+            self.endianess = ELFDATA2LSB
+            self.elfclass = ELFCLASS64
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_AARCH
+
+        elif arch == 'aarch64-be':
+            self.endianess = ELFDATA2MSB
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_AARCH
+
+        elif arch == 'X86_64':
+            self.endianess = ELFDATA2LSB
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_X86_64
+
+        elif arch == '386':
+            self.endianess = ELFDATA2LSB
+            self.elfclass = ELFCLASS32
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_386
+
+        elif arch == 's390':
+            self.endianess = ELFDATA2MSB
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_S390
+
+        elif arch == 'ppc64-le':
+            self.endianess = ELFDATA2LSB
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_PPC64
+
+        elif arch == 'ppc64-be':
+            self.endianess = ELFDATA2MSB
+            self.ehdr = get_arch_ehdr(self.endianess, self.elfclass)
+            self.ehdr.e_machine = EM_PPC64
+
+        else:
+            raise gdb.GdbError("No valid arch type specified.\n"
+                               "Currently supported types:\n"
+                               "aarch64-be, aarch64-le, X86_64, 386, s390, "
+                               "ppc64-be, ppc64-le")
+
+        self.add_segment(PT_NOTE, 0, 0)
+
+    def add_note(self, n_name, n_desc, n_type):
+        """Adds a note to the ELF."""
+
+        note = get_arch_note(self.endianess, len(n_name), len(n_desc))
+        note.n_namesz = len(n_name) + 1
+        note.n_descsz = len(n_desc)
+        note.n_name = n_name.encode()
+        note.n_type = n_type
+
+        # Desc needs to be 4 byte aligned (although the 64bit spec
+        # specifies 8 byte). When defining n_desc as uint32 it will be
+        # automatically aligned but we need the memmove to copy the
+        # string into it.
+        ctypes.memmove(note.n_desc, n_desc.encode(), len(n_desc))
+
+        self.notes.append(note)
+        self.segments[0].p_filesz += ctypes.sizeof(note)
+        self.segments[0].p_memsz += ctypes.sizeof(note)
+
+    def add_segment(self, p_type, p_paddr, p_size):
+        """Adds a segment to the elf."""
+
+        phdr = get_arch_phdr(self.endianess, self.elfclass)
+        phdr.p_type = p_type
+        phdr.p_paddr = p_paddr
+        phdr.p_filesz = p_size
+        phdr.p_memsz = p_size
+        self.segments.append(phdr)
+        self.ehdr.e_phnum += 1
+
+    def to_file(self, elf_file):
+        """Writes all ELF structures to the the passed file.
+
+        Structure:
+        Ehdr
+        Segment 0:PT_NOTE
+        Segment 1:PT_LOAD
+        Segment N:PT_LOAD
+        Note    0..N
+        Dump contents
+        """
+        elf_file.write(self.ehdr)
+        off = ctypes.sizeof(self.ehdr) + \
+              len(self.segments) * ctypes.sizeof(self.segments[0])
+
+        for phdr in self.segments:
+            phdr.p_offset = off
+            elf_file.write(phdr)
+            off += phdr.p_filesz
+
+        for note in self.notes:
+            elf_file.write(note)
+
+
+def get_arch_note(endianess, len_name, len_desc):
+    """Returns a Note class with the specified endianess."""
+
+    if endianess == ELFDATA2LSB:
+        superclass = ctypes.LittleEndianStructure
+    else:
+        superclass = ctypes.BigEndianStructure
+
+    len_name = len_name + 1
+
+    class Note(superclass):
+        """Represents an ELF note, includes the content."""
+
+        _fields_ = [("n_namesz", ctypes.c_uint32),
+                    ("n_descsz", ctypes.c_uint32),
+                    ("n_type", ctypes.c_uint32),
+                    ("n_name", ctypes.c_char * len_name),
+                    ("n_desc", ctypes.c_uint32 * ((len_desc + 3) // 4))]
+    return Note()
+
+
+class Ident(ctypes.Structure):
+    """Represents the ELF ident array in the ehdr structure."""
+
+    _fields_ = [('ei_mag0', ctypes.c_ubyte),
+                ('ei_mag1', ctypes.c_ubyte),
+                ('ei_mag2', ctypes.c_ubyte),
+                ('ei_mag3', ctypes.c_ubyte),
+                ('ei_class', ctypes.c_ubyte),
+                ('ei_data', ctypes.c_ubyte),
+                ('ei_version', ctypes.c_ubyte),
+                ('ei_osabi', ctypes.c_ubyte),
+                ('ei_abiversion', ctypes.c_ubyte),
+                ('ei_pad', ctypes.c_ubyte * 7)]
+
+    def __init__(self, endianess, elfclass):
+        self.ei_mag0 = 0x7F
+        self.ei_mag1 = ord('E')
+        self.ei_mag2 = ord('L')
+        self.ei_mag3 = ord('F')
+        self.ei_class = elfclass
+        self.ei_data = endianess
+        self.ei_version = EV_CURRENT
+
+
+def get_arch_ehdr(endianess, elfclass):
+    """Returns a EHDR64 class with the specified endianess."""
+
+    if endianess == ELFDATA2LSB:
+        superclass = ctypes.LittleEndianStructure
+    else:
+        superclass = ctypes.BigEndianStructure
+
+    class EHDR64(superclass):
+        """Represents the 64 bit ELF header struct."""
+
+        _fields_ = [('e_ident', Ident),
+                    ('e_type', ctypes.c_uint16),
+                    ('e_machine', ctypes.c_uint16),
+                    ('e_version', ctypes.c_uint32),
+                    ('e_entry', ctypes.c_uint64),
+                    ('e_phoff', ctypes.c_uint64),
+                    ('e_shoff', ctypes.c_uint64),
+                    ('e_flags', ctypes.c_uint32),
+                    ('e_ehsize', ctypes.c_uint16),
+                    ('e_phentsize', ctypes.c_uint16),
+                    ('e_phnum', ctypes.c_uint16),
+                    ('e_shentsize', ctypes.c_uint16),
+                    ('e_shnum', ctypes.c_uint16),
+                    ('e_shstrndx', ctypes.c_uint16)]
+
+        def __init__(self):
+            super(superclass, self).__init__()
+            self.e_ident = Ident(endianess, elfclass)
+            self.e_type = ET_CORE
+            self.e_version = EV_CURRENT
+            self.e_ehsize = ctypes.sizeof(self)
+            self.e_phoff = ctypes.sizeof(self)
+            self.e_phentsize = ctypes.sizeof(get_arch_phdr(endianess, elfclass))
+            self.e_phnum = 0
+
+
+    class EHDR32(superclass):
+        """Represents the 32 bit ELF header struct."""
+
+        _fields_ = [('e_ident', Ident),
+                    ('e_type', ctypes.c_uint16),
+                    ('e_machine', ctypes.c_uint16),
+                    ('e_version', ctypes.c_uint32),
+                    ('e_entry', ctypes.c_uint32),
+                    ('e_phoff', ctypes.c_uint32),
+                    ('e_shoff', ctypes.c_uint32),
+                    ('e_flags', ctypes.c_uint32),
+                    ('e_ehsize', ctypes.c_uint16),
+                    ('e_phentsize', ctypes.c_uint16),
+                    ('e_phnum', ctypes.c_uint16),
+                    ('e_shentsize', ctypes.c_uint16),
+                    ('e_shnum', ctypes.c_uint16),
+                    ('e_shstrndx', ctypes.c_uint16)]
+
+        def __init__(self):
+            super(superclass, self).__init__()
+            self.e_ident = Ident(endianess, elfclass)
+            self.e_type = ET_CORE
+            self.e_version = EV_CURRENT
+            self.e_ehsize = ctypes.sizeof(self)
+            self.e_phoff = ctypes.sizeof(self)
+            self.e_phentsize = ctypes.sizeof(get_arch_phdr(endianess, elfclass))
+            self.e_phnum = 0
+
+    # End get_arch_ehdr
+    if elfclass == ELFCLASS64:
+        return EHDR64()
+    else:
+        return EHDR32()
+
+
+def get_arch_phdr(endianess, elfclass):
+    """Returns a 32 or 64 bit PHDR class with the specified endianess."""
+
+    if endianess == ELFDATA2LSB:
+        superclass = ctypes.LittleEndianStructure
+    else:
+        superclass = ctypes.BigEndianStructure
+
+    class PHDR64(superclass):
+        """Represents the 64 bit ELF program header struct."""
+
+        _fields_ = [('p_type', ctypes.c_uint32),
+                    ('p_flags', ctypes.c_uint32),
+                    ('p_offset', ctypes.c_uint64),
+                    ('p_vaddr', ctypes.c_uint64),
+                    ('p_paddr', ctypes.c_uint64),
+                    ('p_filesz', ctypes.c_uint64),
+                    ('p_memsz', ctypes.c_uint64),
+                    ('p_align', ctypes.c_uint64)]
+
+    class PHDR32(superclass):
+        """Represents the 32 bit ELF program header struct."""
+
+        _fields_ = [('p_type', ctypes.c_uint32),
+                    ('p_offset', ctypes.c_uint32),
+                    ('p_vaddr', ctypes.c_uint32),
+                    ('p_paddr', ctypes.c_uint32),
+                    ('p_filesz', ctypes.c_uint32),
+                    ('p_memsz', ctypes.c_uint32),
+                    ('p_flags', ctypes.c_uint32),
+                    ('p_align', ctypes.c_uint32)]
+
+    # End get_arch_phdr
+    if elfclass == ELFCLASS64:
+        return PHDR64()
+    else:
+        return PHDR32()
+
+
+def int128_get64(val):
+    """Returns low 64bit part of Int128 struct."""
+
+    assert val["hi"] == 0
+    return val["lo"]
+
+
+def qlist_foreach(head, field_str):
+    """Generator for qlists."""
+
+    var_p = head["lh_first"]
+    while var_p != 0:
+        var = var_p.dereference()
+        var_p = var[field_str]["le_next"]
+        yield var
+
+
+def qemu_get_ram_block(ram_addr):
+    """Returns the RAMBlock struct to which the given address belongs."""
+
+    ram_blocks = gdb.parse_and_eval("ram_list.blocks")
+
+    for block in qlist_foreach(ram_blocks, "next"):
+        if (ram_addr - block["offset"]) < block["used_length"]:
+            return block
+
+    raise gdb.GdbError("Bad ram offset %x" % ram_addr)
+
+
+def qemu_get_ram_ptr(ram_addr):
+    """Returns qemu vaddr for given guest physical address."""
+
+    block = qemu_get_ram_block(ram_addr)
+    return block["host"] + (ram_addr - block["offset"])
+
+
+def memory_region_get_ram_ptr(memory_region):
+    if memory_region["alias"] != 0:
+        return (memory_region_get_ram_ptr(memory_region["alias"].dereference())
+                + memory_region["alias_offset"])
+
+    return qemu_get_ram_ptr(memory_region["ram_block"]["offset"])
+
+
+def get_guest_phys_blocks():
+    """Returns a list of ram blocks.
+
+    Each block entry contains:
+    'target_start': guest block phys start address
+    'target_end':   guest block phys end address
+    'host_addr':    qemu vaddr of the block's start
+    """
+
+    guest_phys_blocks = []
+
+    print("guest RAM blocks:")
+    print("target_start     target_end       host_addr        message "
+          "count")
+    print("---------------- ---------------- ---------------- ------- "
+          "-----")
+
+    current_map_p = gdb.parse_and_eval("address_space_memory.current_map")
+    current_map = current_map_p.dereference()
+
+    # Conversion to int is needed for python 3
+    # compatibility. Otherwise range doesn't cast the value itself and
+    # breaks.
+    for cur in range(int(current_map["nr"])):
+        flat_range = (current_map["ranges"] + cur).dereference()
+        memory_region = flat_range["mr"].dereference()
+
+        # we only care about RAM
+        if not memory_region["ram"]:
+            continue
+
+        section_size = int128_get64(flat_range["addr"]["size"])
+        target_start = int128_get64(flat_range["addr"]["start"])
+        target_end = target_start + section_size
+        host_addr = (memory_region_get_ram_ptr(memory_region)
+                     + flat_range["offset_in_region"])
+        predecessor = None
+
+        # find continuity in guest physical address space
+        if len(guest_phys_blocks) > 0:
+            predecessor = guest_phys_blocks[-1]
+            predecessor_size = (predecessor["target_end"] -
+                                predecessor["target_start"])
+
+            # the memory API guarantees monotonically increasing
+            # traversal
+            assert predecessor["target_end"] <= target_start
+
+            # we want continuity in both guest-physical and
+            # host-virtual memory
+            if (predecessor["target_end"] < target_start or
+                predecessor["host_addr"] + predecessor_size != host_addr):
+                predecessor = None
+
+        if predecessor is None:
+            # isolated mapping, add it to the list
+            guest_phys_blocks.append({"target_start": target_start,
+                                      "target_end":   target_end,
+                                      "host_addr":    host_addr})
+            message = "added"
+        else:
+            # expand predecessor until @target_end; predecessor's
+            # start doesn't change
+            predecessor["target_end"] = target_end
+            message = "joined"
+
+        print("%016x %016x %016x %-7s %5u" %
+              (target_start, target_end, host_addr.cast(UINTPTR_T),
+               message, len(guest_phys_blocks)))
+
+    return guest_phys_blocks
+
+
 # The leading docstring doesn't have idiomatic Python formatting. It is
 # printed by gdb's "help" command (the first line is printed in the
 # "help data" summary), and it should match how other help texts look in
 # gdb.
 # The leading docstring doesn't have idiomatic Python formatting. It is
 # printed by gdb's "help" command (the first line is printed in the
 # "help data" summary), and it should match how other help texts look in
 # gdb.
-
-import struct
-
 class DumpGuestMemory(gdb.Command):
     """Extract guest vmcore from qemu process coredump.
 
 class DumpGuestMemory(gdb.Command):
     """Extract guest vmcore from qemu process coredump.
 
-The sole argument is FILE, identifying the target file to write the
-guest vmcore to.
+The two required arguments are FILE and ARCH:
+FILE identifies the target file to write the guest vmcore to.
+ARCH specifies the architecture for which the core will be generated.
 
 This GDB command reimplements the dump-guest-memory QMP command in
 python, using the representation of guest memory as captured in the qemu
 coredump. The qemu process that has been dumped must have had the
 
 This GDB command reimplements the dump-guest-memory QMP command in
 python, using the representation of guest memory as captured in the qemu
 coredump. The qemu process that has been dumped must have had the
-command line option "-machine dump-guest-core=on".
+command line option "-machine dump-guest-core=on" which is the default.
 
 For simplicity, the "paging", "begin" and "end" parameters of the QMP
 command are not supported -- no attempt is made to get the guest's
 internal paging structures (ie. paging=false is hard-wired), and guest
 memory is always fully dumped.
 
 
 For simplicity, the "paging", "begin" and "end" parameters of the QMP
 command are not supported -- no attempt is made to get the guest's
 internal paging structures (ie. paging=false is hard-wired), and guest
 memory is always fully dumped.
 
-Only x86_64 guests are supported.
+Currently aarch64-be, aarch64-le, X86_64, 386, s390, ppc64-be,
+ppc64-le guests are supported.
 
 The CORE/NT_PRSTATUS and QEMU notes (that is, the VCPUs' statuses) are
 not written to the vmcore. Preparing these would require context that is
 
 The CORE/NT_PRSTATUS and QEMU notes (that is, the VCPUs' statuses) are
 not written to the vmcore. Preparing these would require context that is
@@ -47,293 +464,66 @@ deliberately called abort(), or it was dumped in response to a signal at
 a halfway fortunate point, then its coredump should be in reasonable
 shape and this command should mostly work."""
 
 a halfway fortunate point, then its coredump should be in reasonable
 shape and this command should mostly work."""
 
-    TARGET_PAGE_SIZE = 0x1000
-    TARGET_PAGE_MASK = 0xFFFFFFFFFFFFF000
-
-    # Various ELF constants
-    EM_X86_64   = 62        # AMD x86-64 target machine
-    ELFDATA2LSB = 1         # little endian
-    ELFCLASS64  = 2
-    ELFMAG      = "\x7FELF"
-    EV_CURRENT  = 1
-    ET_CORE     = 4
-    PT_LOAD     = 1
-    PT_NOTE     = 4
-
-    # Special value for e_phnum. This indicates that the real number of
-    # program headers is too large to fit into e_phnum. Instead the real
-    # value is in the field sh_info of section 0.
-    PN_XNUM = 0xFFFF
-
-    # Format strings for packing and header size calculation.
-    ELF64_EHDR = ("4s" # e_ident/magic
-                  "B"  # e_ident/class
-                  "B"  # e_ident/data
-                  "B"  # e_ident/version
-                  "B"  # e_ident/osabi
-                  "8s" # e_ident/pad
-                  "H"  # e_type
-                  "H"  # e_machine
-                  "I"  # e_version
-                  "Q"  # e_entry
-                  "Q"  # e_phoff
-                  "Q"  # e_shoff
-                  "I"  # e_flags
-                  "H"  # e_ehsize
-                  "H"  # e_phentsize
-                  "H"  # e_phnum
-                  "H"  # e_shentsize
-                  "H"  # e_shnum
-                  "H"  # e_shstrndx
-                 )
-    ELF64_PHDR = ("I"  # p_type
-                  "I"  # p_flags
-                  "Q"  # p_offset
-                  "Q"  # p_vaddr
-                  "Q"  # p_paddr
-                  "Q"  # p_filesz
-                  "Q"  # p_memsz
-                  "Q"  # p_align
-                 )
-
     def __init__(self):
         super(DumpGuestMemory, self).__init__("dump-guest-memory",
                                               gdb.COMMAND_DATA,
                                               gdb.COMPLETE_FILENAME)
     def __init__(self):
         super(DumpGuestMemory, self).__init__("dump-guest-memory",
                                               gdb.COMMAND_DATA,
                                               gdb.COMPLETE_FILENAME)
-        self.uintptr_t     = gdb.lookup_type("uintptr_t")
-        self.elf64_ehdr_le = struct.Struct("<%s" % self.ELF64_EHDR)
-        self.elf64_phdr_le = struct.Struct("<%s" % self.ELF64_PHDR)
-
-    def int128_get64(self, val):
-        assert (val["hi"] == 0)
-        return val["lo"]
-
-    def qlist_foreach(self, head, field_str):
-        var_p = head["lh_first"]
-        while (var_p != 0):
-            var = var_p.dereference()
-            yield var
-            var_p = var[field_str]["le_next"]
-
-    def qemu_get_ram_block(self, ram_addr):
-        ram_blocks = gdb.parse_and_eval("ram_list.blocks")
-        for block in self.qlist_foreach(ram_blocks, "next"):
-            if (ram_addr - block["offset"] < block["length"]):
-                return block
-        raise gdb.GdbError("Bad ram offset %x" % ram_addr)
-
-    def qemu_get_ram_ptr(self, ram_addr):
-        block = self.qemu_get_ram_block(ram_addr)
-        return block["host"] + (ram_addr - block["offset"])
-
-    def memory_region_get_ram_ptr(self, mr):
-        if (mr["alias"] != 0):
-            return (self.memory_region_get_ram_ptr(mr["alias"].dereference()) +
-                    mr["alias_offset"])
-        return self.qemu_get_ram_ptr(mr["ram_addr"] & self.TARGET_PAGE_MASK)
-
-    def guest_phys_blocks_init(self):
-        self.guest_phys_blocks = []
-
-    def guest_phys_blocks_append(self):
-        print "guest RAM blocks:"
-        print ("target_start     target_end       host_addr        message "
-               "count")
-        print ("---------------- ---------------- ---------------- ------- "
-               "-----")
-
-        current_map_p = gdb.parse_and_eval("address_space_memory.current_map")
-        current_map = current_map_p.dereference()
-        for cur in range(current_map["nr"]):
-            flat_range   = (current_map["ranges"] + cur).dereference()
-            mr           = flat_range["mr"].dereference()
-
-            # we only care about RAM
-            if (not mr["ram"]):
-                continue
-
-            section_size = self.int128_get64(flat_range["addr"]["size"])
-            target_start = self.int128_get64(flat_range["addr"]["start"])
-            target_end   = target_start + section_size
-            host_addr    = (self.memory_region_get_ram_ptr(mr) +
-                            flat_range["offset_in_region"])
-            predecessor = None
-
-            # find continuity in guest physical address space
-            if (len(self.guest_phys_blocks) > 0):
-                predecessor = self.guest_phys_blocks[-1]
-                predecessor_size = (predecessor["target_end"] -
-                                    predecessor["target_start"])
-
-                # the memory API guarantees monotonically increasing
-                # traversal
-                assert (predecessor["target_end"] <= target_start)
-
-                # we want continuity in both guest-physical and
-                # host-virtual memory
-                if (predecessor["target_end"] < target_start or
-                    predecessor["host_addr"] + predecessor_size != host_addr):
-                    predecessor = None
-
-            if (predecessor is None):
-                # isolated mapping, add it to the list
-                self.guest_phys_blocks.append({"target_start": target_start,
-                                               "target_end"  : target_end,
-                                               "host_addr"   : host_addr})
-                message = "added"
-            else:
-                # expand predecessor until @target_end; predecessor's
-                # start doesn't change
-                predecessor["target_end"] = target_end
-                message = "joined"
-
-            print ("%016x %016x %016x %-7s %5u" %
-                   (target_start, target_end, host_addr.cast(self.uintptr_t),
-                    message, len(self.guest_phys_blocks)))
-
-    def cpu_get_dump_info(self):
-        # We can't synchronize the registers with KVM post-mortem, and
-        # the bits in (first_x86_cpu->env.hflags) seem to be stale; they
-        # may not reflect long mode for example. Hence just assume the
-        # most common values. This also means that instruction pointer
-        # etc. will be bogus in the dump, but at least the RAM contents
-        # should be valid.
-        self.dump_info = {"d_machine": self.EM_X86_64,
-                          "d_endian" : self.ELFDATA2LSB,
-                          "d_class"  : self.ELFCLASS64}
-
-    def encode_elf64_ehdr_le(self):
-        return self.elf64_ehdr_le.pack(
-                                 self.ELFMAG,                 # e_ident/magic
-                                 self.dump_info["d_class"],   # e_ident/class
-                                 self.dump_info["d_endian"],  # e_ident/data
-                                 self.EV_CURRENT,             # e_ident/version
-                                 0,                           # e_ident/osabi
-                                 "",                          # e_ident/pad
-                                 self.ET_CORE,                # e_type
-                                 self.dump_info["d_machine"], # e_machine
-                                 self.EV_CURRENT,             # e_version
-                                 0,                           # e_entry
-                                 self.elf64_ehdr_le.size,     # e_phoff
-                                 0,                           # e_shoff
-                                 0,                           # e_flags
-                                 self.elf64_ehdr_le.size,     # e_ehsize
-                                 self.elf64_phdr_le.size,     # e_phentsize
-                                 self.phdr_num,               # e_phnum
-                                 0,                           # e_shentsize
-                                 0,                           # e_shnum
-                                 0                            # e_shstrndx
-                                )
-
-    def encode_elf64_note_le(self):
-        return self.elf64_phdr_le.pack(self.PT_NOTE,         # p_type
-                                       0,                    # p_flags
-                                       (self.memory_offset -
-                                        len(self.note)),     # p_offset
-                                       0,                    # p_vaddr
-                                       0,                    # p_paddr
-                                       len(self.note),       # p_filesz
-                                       len(self.note),       # p_memsz
-                                       0                     # p_align
-                                      )
-
-    def encode_elf64_load_le(self, offset, start_hwaddr, range_size):
-        return self.elf64_phdr_le.pack(self.PT_LOAD, # p_type
-                                       0,            # p_flags
-                                       offset,       # p_offset
-                                       0,            # p_vaddr
-                                       start_hwaddr, # p_paddr
-                                       range_size,   # p_filesz
-                                       range_size,   # p_memsz
-                                       0             # p_align
-                                      )
-
-    def note_init(self, name, desc, type):
-        # name must include a trailing NUL
-        namesz = (len(name) + 1 + 3) / 4 * 4
-        descsz = (len(desc)     + 3) / 4 * 4
-        fmt = ("<"   # little endian
-               "I"   # n_namesz
-               "I"   # n_descsz
-               "I"   # n_type
-               "%us" # name
-               "%us" # desc
-               % (namesz, descsz))
-        self.note = struct.pack(fmt,
-                                len(name) + 1, len(desc), type, name, desc)
-
-    def dump_init(self):
-        self.guest_phys_blocks_init()
-        self.guest_phys_blocks_append()
-        self.cpu_get_dump_info()
-        # we have no way to retrieve the VCPU status from KVM
-        # post-mortem
-        self.note_init("NONE", "EMPTY", 0)
-
-        # Account for PT_NOTE.
-        self.phdr_num = 1
-
-        # We should never reach PN_XNUM for paging=false dumps: there's
-        # just a handful of discontiguous ranges after merging.
-        self.phdr_num += len(self.guest_phys_blocks)
-        assert (self.phdr_num < self.PN_XNUM)
-
-        # Calculate the ELF file offset where the memory dump commences:
-        #
-        #   ELF header
-        #   PT_NOTE
-        #   PT_LOAD: 1
-        #   PT_LOAD: 2
-        #   ...
-        #   PT_LOAD: len(self.guest_phys_blocks)
-        #   ELF note
-        #   memory dump
-        self.memory_offset = (self.elf64_ehdr_le.size +
-                              self.elf64_phdr_le.size * self.phdr_num +
-                              len(self.note))
-
-    def dump_begin(self, vmcore):
-        vmcore.write(self.encode_elf64_ehdr_le())
-        vmcore.write(self.encode_elf64_note_le())
-        running = self.memory_offset
+        self.elf = None
+        self.guest_phys_blocks = None
+
+    def dump_init(self, vmcore):
+        """Prepares and writes ELF structures to core file."""
+
+        # Needed to make crash happy, data for more useful notes is
+        # not available in a qemu core.
+        self.elf.add_note("NONE", "EMPTY", 0)
+
+        # We should never reach PN_XNUM for paging=false dumps,
+        # there's just a handful of discontiguous ranges after
+        # merging.
+        # The constant is needed to account for the PT_NOTE segment.
+        phdr_num = len(self.guest_phys_blocks) + 1
+        assert phdr_num < PN_XNUM
+
         for block in self.guest_phys_blocks:
         for block in self.guest_phys_blocks:
-            range_size = block["target_end"] - block["target_start"]
-            vmcore.write(self.encode_elf64_load_le(running,
-                                                   block["target_start"],
-                                                   range_size))
-            running += range_size
-        vmcore.write(self.note)
+            block_size = block["target_end"] - block["target_start"]
+            self.elf.add_segment(PT_LOAD, block["target_start"], block_size)
+
+        self.elf.to_file(vmcore)
 
     def dump_iterate(self, vmcore):
 
     def dump_iterate(self, vmcore):
+        """Writes guest core to file."""
+
         qemu_core = gdb.inferiors()[0]
         for block in self.guest_phys_blocks:
         qemu_core = gdb.inferiors()[0]
         for block in self.guest_phys_blocks:
-            cur  = block["host_addr"]
+            cur = block["host_addr"]
             left = block["target_end"] - block["target_start"]
             left = block["target_end"] - block["target_start"]
-            print ("dumping range at %016x for length %016x" %
-                   (cur.cast(self.uintptr_t), left))
-            while (left > 0):
-                chunk_size = min(self.TARGET_PAGE_SIZE, left)
+            print("dumping range at %016x for length %016x" %
+                  (cur.cast(UINTPTR_T), left))
+
+            while left > 0:
+                chunk_size = min(TARGET_PAGE_SIZE, left)
                 chunk = qemu_core.read_memory(cur, chunk_size)
                 vmcore.write(chunk)
                 chunk = qemu_core.read_memory(cur, chunk_size)
                 vmcore.write(chunk)
-                cur  += chunk_size
+                cur += chunk_size
                 left -= chunk_size
 
                 left -= chunk_size
 
-    def create_vmcore(self, filename):
-        vmcore = open(filename, "wb")
-        self.dump_begin(vmcore)
-        self.dump_iterate(vmcore)
-        vmcore.close()
-
     def invoke(self, args, from_tty):
     def invoke(self, args, from_tty):
+        """Handles command invocation from gdb."""
+
         # Unwittingly pressing the Enter key after the command should
         # not dump the same multi-gig coredump to the same file.
         self.dont_repeat()
 
         argv = gdb.string_to_argv(args)
         # Unwittingly pressing the Enter key after the command should
         # not dump the same multi-gig coredump to the same file.
         self.dont_repeat()
 
         argv = gdb.string_to_argv(args)
-        if (len(argv) != 1):
-            raise gdb.GdbError("usage: dump-guest-memory FILE")
+        if len(argv) != 2:
+            raise gdb.GdbError("usage: dump-guest-memory FILE ARCH")
+
+        self.elf = ELF(argv[1])
+        self.guest_phys_blocks = get_guest_phys_blocks()
 
 
-        self.dump_init()
-        self.create_vmcore(argv[0])
+        with open(argv[0], "wb") as vmcore:
+            self.dump_init(vmcore)
+            self.dump_iterate(vmcore)
 
 DumpGuestMemory()
 
 DumpGuestMemory()