390 lines
14 KiB
Python
Executable File
390 lines
14 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
import argparse
|
|
import elftools.elf.elffile
|
|
import elftools.dwarf.descriptions
|
|
from collections import namedtuple
|
|
from struct import unpack
|
|
import os
|
|
|
|
from lib.dump import decode_dump
|
|
from lib.avr import *
|
|
|
|
|
|
Entry = namedtuple('Entry', ['name', 'loc', 'size', 'declpos'])
|
|
Member = namedtuple('Member', ['name', 'off', 'size'])
|
|
|
|
|
|
def array_inc(loc, dim, idx=0):
|
|
if idx == len(dim):
|
|
return True
|
|
loc[idx] += 1
|
|
if loc[idx] == dim[idx]:
|
|
loc[idx] = 0
|
|
return array_inc(loc, dim, idx+1)
|
|
return False
|
|
|
|
def get_type_size(type_DIE):
|
|
while True:
|
|
if 'DW_AT_byte_size' in type_DIE.attributes:
|
|
return type_DIE, type_DIE.attributes.get('DW_AT_byte_size').value
|
|
if 'DW_AT_type' not in type_DIE.attributes:
|
|
return None
|
|
type_DIE = type_DIE.get_DIE_from_attribute('DW_AT_type')
|
|
|
|
def get_type_arrsize(type_DIE):
|
|
size = get_type_size(type_DIE)
|
|
if size is None:
|
|
return None
|
|
byte_size = size[1]
|
|
if size[0].tag != 'DW_TAG_pointer_type':
|
|
array_DIE = get_type_def(type_DIE, 'DW_TAG_array_type')
|
|
if array_DIE is not None:
|
|
for range_DIE in array_DIE.iter_children():
|
|
if range_DIE.tag == 'DW_TAG_subrange_type' and \
|
|
'DW_AT_upper_bound' in range_DIE.attributes:
|
|
dim = range_DIE.attributes['DW_AT_upper_bound'].value + 1
|
|
byte_size *= dim
|
|
return byte_size
|
|
|
|
def get_type_def(type_DIE, type_tag):
|
|
while True:
|
|
if type_DIE.tag == type_tag:
|
|
return type_DIE
|
|
if 'DW_AT_type' not in type_DIE.attributes:
|
|
return None
|
|
type_DIE = type_DIE.get_DIE_from_attribute('DW_AT_type')
|
|
|
|
def get_FORM_block1(attr):
|
|
if attr.form != 'DW_FORM_block1':
|
|
return None
|
|
if attr.value[0] == 3: # OP_addr
|
|
return int.from_bytes(attr.value[1:], 'little')
|
|
if attr.value[0] == 35: # OP_plus_uconst (ULEB128)
|
|
v = 0
|
|
s = 0
|
|
for b in attr.value[1:]:
|
|
v |= (b & 0x7f) << s
|
|
if b & 0x80 == 0:
|
|
break
|
|
s += 7
|
|
return v
|
|
return None
|
|
|
|
|
|
def get_array_dims(DIE):
|
|
array_DIE = get_type_def(DIE, 'DW_TAG_array_type')
|
|
if array_DIE is None:
|
|
return []
|
|
|
|
array_dim = []
|
|
for range_DIE in array_DIE.iter_children():
|
|
if range_DIE.tag == 'DW_TAG_subrange_type' and \
|
|
'DW_AT_upper_bound' in range_DIE.attributes:
|
|
array_dim.append(range_DIE.attributes['DW_AT_upper_bound'].value + 1)
|
|
return array_dim
|
|
|
|
|
|
def get_struct_members(DIE, entry, expand_structs, struct_gaps):
|
|
struct_DIE = get_type_def(DIE, 'DW_TAG_structure_type')
|
|
if struct_DIE is None:
|
|
return []
|
|
|
|
members = []
|
|
for member_DIE in struct_DIE.iter_children():
|
|
if member_DIE.tag == 'DW_TAG_member' and 'DW_AT_name' in member_DIE.attributes:
|
|
m_name = member_DIE.attributes['DW_AT_name'].value.decode('ascii')
|
|
m_off = get_FORM_block1(member_DIE.attributes['DW_AT_data_member_location'])
|
|
m_byte_size = get_type_size(member_DIE)[1]
|
|
|
|
# still expand member arrays
|
|
m_array_dim = get_array_dims(member_DIE)
|
|
|
|
if m_byte_size == 1 and len(m_array_dim) > 1:
|
|
# likely string, remove one dimension
|
|
m_byte_size *= m_array_dim.pop()
|
|
if len(m_array_dim) == 0 or (len(m_array_dim) == 1 and m_array_dim[0] == 1):
|
|
# plain entry
|
|
members.append(Member(m_name, m_off, m_byte_size))
|
|
elif len(m_array_dim) == 1 and m_byte_size == 1:
|
|
# likely string, avoid expansion
|
|
members.append(Member(m_name + '[]', m_off, m_array_dim[0]))
|
|
else:
|
|
# expand array entries
|
|
m_array_pos = m_off
|
|
m_array_loc = [0] * len(m_array_dim)
|
|
while True:
|
|
# location index
|
|
sfx = ''
|
|
for d in range(len(m_array_dim)):
|
|
sfx += '[{}]'.format(str(m_array_loc[d]).rjust(len(str(m_array_dim[d]-1)), '0'))
|
|
members.append(Member(m_name + sfx, m_array_pos, m_byte_size))
|
|
# advance
|
|
if array_inc(m_array_loc, m_array_dim):
|
|
break
|
|
m_array_pos += m_byte_size
|
|
|
|
if struct_gaps and len(members):
|
|
# fill gaps in the middle
|
|
members = list(sorted(members, key=lambda x: x.off))
|
|
last_end = 0
|
|
for n in range(len(members)):
|
|
member = members[n]
|
|
if member.off > last_end:
|
|
members.append(Member('*UNKNOWN*', last_end, member.off - last_end))
|
|
last_end = member.off + member.size
|
|
|
|
if struct_gaps and len(members):
|
|
# fill gap at the end
|
|
members = list(sorted(members, key=lambda x: x.off))
|
|
last = members[-1]
|
|
last_end = last.off + last.size
|
|
if entry.size > last_end:
|
|
members.append(Member('*UNKNOWN*', last_end, entry.size - last_end))
|
|
|
|
return members
|
|
|
|
|
|
def get_elf_globals(path, expand_structs, struct_gaps=True):
|
|
fd = open(path, "rb")
|
|
if fd is None:
|
|
return
|
|
elffile = elftools.elf.elffile.ELFFile(fd)
|
|
if elffile is None or not elffile.has_dwarf_info():
|
|
return
|
|
|
|
# probably not needed, since we're decoding expressions manually
|
|
elftools.dwarf.descriptions.set_global_machine_arch(elffile.get_machine_arch())
|
|
dwarfinfo = elffile.get_dwarf_info()
|
|
|
|
grefs = []
|
|
for CU in dwarfinfo.iter_CUs():
|
|
file_entries = dwarfinfo.line_program_for_CU(CU).header["file_entry"]
|
|
|
|
for DIE in CU.iter_DIEs():
|
|
# handle only variable types
|
|
if DIE.tag != 'DW_TAG_variable':
|
|
continue
|
|
if 'DW_AT_location' not in DIE.attributes:
|
|
continue
|
|
if 'DW_AT_name' not in DIE.attributes and \
|
|
'DW_AT_abstract_origin' not in DIE.attributes:
|
|
continue
|
|
|
|
# handle locations encoded directly as DW_OP_addr (leaf globals)
|
|
loc = get_FORM_block1(DIE.attributes['DW_AT_location'])
|
|
if loc is None or loc < SRAM_OFFSET or loc >= EEPROM_OFFSET:
|
|
continue
|
|
loc -= SRAM_OFFSET
|
|
|
|
# variable name/type
|
|
if 'DW_AT_name' not in DIE.attributes and \
|
|
'DW_AT_abstract_origin' in DIE.attributes:
|
|
DIE = DIE.get_DIE_from_attribute('DW_AT_abstract_origin')
|
|
if 'DW_AT_location' in DIE.attributes:
|
|
# duplicate reference (handled directly), skip
|
|
continue
|
|
if 'DW_AT_name' not in DIE.attributes:
|
|
continue
|
|
if 'DW_AT_type' not in DIE.attributes:
|
|
continue
|
|
|
|
name = DIE.attributes['DW_AT_name'].value.decode('ascii')
|
|
|
|
# get final storage size
|
|
size = get_type_size(DIE)
|
|
if size is None:
|
|
continue
|
|
byte_size = size[1]
|
|
|
|
# location of main definition
|
|
declpos = ''
|
|
if 'DW_AT_decl_file' in DIE.attributes and \
|
|
'DW_AT_decl_line' in DIE.attributes:
|
|
line = DIE.attributes['DW_AT_decl_line'].value
|
|
fname = DIE.attributes['DW_AT_decl_file'].value
|
|
if fname and fname - 1 < len(file_entries):
|
|
fname = file_entries[fname-1].name.decode('ascii')
|
|
declpos = '{}:{}'.format(fname, line)
|
|
|
|
# fetch array dimensions (if known)
|
|
array_dim = get_array_dims(DIE)
|
|
|
|
# fetch structure members (one level only)
|
|
entry = Entry(name, loc, byte_size, declpos)
|
|
if not expand_structs or size[0].tag == 'DW_TAG_pointer_type':
|
|
members = []
|
|
else:
|
|
members = get_struct_members(DIE, entry, expand_structs, struct_gaps)
|
|
|
|
def expand_members(entry, members):
|
|
if len(members) == 0:
|
|
grefs.append(entry)
|
|
else:
|
|
for member in members:
|
|
grefs.append(Entry(entry.name + '.' + member.name,
|
|
entry.loc + member.off, member.size,
|
|
entry.declpos))
|
|
|
|
if byte_size == 1 and len(array_dim) > 1:
|
|
# likely string, remove one dimension
|
|
byte_size *= array_dim.pop()
|
|
if len(array_dim) == 0 or (len(array_dim) == 1 and array_dim[0] == 1):
|
|
# plain entry
|
|
expand_members(entry, members)
|
|
elif len(array_dim) == 1 and byte_size == 1:
|
|
# likely string, avoid expansion
|
|
grefs.append(Entry(entry.name + '[]', entry.loc,
|
|
array_dim[0], entry.declpos))
|
|
else:
|
|
# expand array entries
|
|
array_pos = loc
|
|
array_loc = [0] * len(array_dim)
|
|
while True:
|
|
# location index
|
|
sfx = ''
|
|
for d in range(len(array_dim)):
|
|
sfx += '[{}]'.format(str(array_loc[d]).rjust(len(str(array_dim[d]-1)), '0'))
|
|
expand_members(Entry(entry.name + sfx, array_pos,
|
|
byte_size, entry.declpos), members)
|
|
# advance
|
|
if array_inc(array_loc, array_dim):
|
|
break
|
|
array_pos += byte_size
|
|
|
|
return grefs
|
|
|
|
|
|
def annotate_refs(grefs, addr, data, width, gaps=True, overlaps=True):
|
|
last_end = None
|
|
for entry in grefs:
|
|
if entry.loc < addr:
|
|
continue
|
|
if entry.loc + entry.size > addr + len(data):
|
|
continue
|
|
|
|
pos = entry.loc-addr
|
|
end_pos = pos + entry.size
|
|
buf = data[pos:end_pos]
|
|
|
|
buf_repr = ''
|
|
if len(buf) in [1, 2, 4]:
|
|
# attempt to decode as integers
|
|
buf_repr += ' I:' + str(int.from_bytes(buf, 'little')).rjust(10)
|
|
if len(buf) in [4, 8]:
|
|
# attempt to decode as floats
|
|
typ = 'f' if len(buf) == 4 else 'd'
|
|
buf_repr += ' F:' + '{:10.3f}'.format(unpack(typ, buf)[0])
|
|
|
|
if last_end is not None:
|
|
if gaps and last_end < pos:
|
|
# decode gaps
|
|
gap_size = pos - last_end
|
|
gap_buf = data[last_end:pos]
|
|
print('{:04x} {} {:4} R:{}'.format(addr+last_end, "*UNKNOWN*".ljust(width),
|
|
gap_size, gap_buf.hex()))
|
|
if overlaps and last_end > pos + 1:
|
|
gap_size = pos - last_end
|
|
print('{:04x} {} {:4}'.format(addr+last_end, "*OVERLAP*".ljust(width), gap_size))
|
|
|
|
print('{:04x} {} {:4}{} R:{}'.format(entry.loc, entry.name.ljust(width),
|
|
entry.size, buf_repr, buf.hex()))
|
|
last_end = end_pos
|
|
|
|
|
|
def print_map(grefs):
|
|
print('OFFSET\tSIZE\tNAME\tDECLPOS')
|
|
for entry in grefs:
|
|
print('{:x}\t{}\t{}\t{}'.format(entry.loc, entry.size, entry.name, entry.declpos))
|
|
|
|
|
|
def print_qdirstat(grefs):
|
|
print('[qdirstat 1.0 cache file]')
|
|
|
|
entries = {}
|
|
for entry in grefs:
|
|
# do not output registers when looking at space usage
|
|
if entry.loc < SRAM_START:
|
|
continue
|
|
|
|
paths = list(filter(None, re.split(r'[\[\].]', entry.name)))
|
|
base = entries
|
|
for i in range(len(paths) - 1):
|
|
name = paths[i]
|
|
if name not in base:
|
|
base[name] = {}
|
|
base = base[name]
|
|
name = paths[-1]
|
|
if name in base:
|
|
name = '{}_{:x}'.format(entry.name, entry.loc)
|
|
base[name] = entry.size
|
|
|
|
def walker(root, prefix):
|
|
files = []
|
|
dirs = []
|
|
|
|
for name, entries in root.items():
|
|
if type(entries) == int:
|
|
files.append([name, entries])
|
|
else:
|
|
dirs.append([name, entries])
|
|
|
|
# print files
|
|
print('D\t{}\t{}\t0x0'.format(prefix, 0))
|
|
for name, size in files:
|
|
print('F\t{}\t{}\t0x0'.format(name, size))
|
|
|
|
# recurse directories
|
|
for name, entries in dirs:
|
|
walker(entries, prefix + '/' + name)
|
|
|
|
walker(entries, '/')
|
|
|
|
|
|
def main():
|
|
ap = argparse.ArgumentParser(description="""
|
|
Generate a symbol table map starting directly from an ELF
|
|
firmware with DWARF3 debugging information.
|
|
When used along with a memory dump obtained from the D2/D21/D23 g-code,
|
|
show the value of each symbol which is within the address range.
|
|
""")
|
|
ap.add_argument('elf', help='ELF file containing DWARF debugging information')
|
|
ap.add_argument('--no-gaps', action='store_true',
|
|
help='do not dump memory inbetween known symbols')
|
|
ap.add_argument('--no-expand-structs', action='store_true',
|
|
help='do not decode structure data')
|
|
ap.add_argument('--overlaps', action='store_true',
|
|
help='annotate overlaps greater than 1 byte')
|
|
ap.add_argument('--name-width', type=int, default=50,
|
|
help='set name column width')
|
|
g = ap.add_mutually_exclusive_group(required=True)
|
|
g.add_argument('dump', nargs='?', help='RAM dump obtained from D2 g-code')
|
|
g.add_argument('--map', action='store_true', help='dump global memory map')
|
|
g.add_argument('--qdirstat', action='store_true',
|
|
help='dump qdirstat-compatible size usage map')
|
|
args = ap.parse_args()
|
|
|
|
grefs = get_elf_globals(args.elf, expand_structs=not args.no_expand_structs)
|
|
grefs = list(sorted(grefs, key=lambda x: x.loc))
|
|
if args.map:
|
|
print_map(grefs)
|
|
elif args.qdirstat:
|
|
print_qdirstat(grefs)
|
|
else:
|
|
# fetch the memory data
|
|
dump = decode_dump(args.dump)
|
|
if dump is None:
|
|
return os.EX_DATAERR
|
|
|
|
# strip padding, if present
|
|
addr_start = dump.ranges[0][0]
|
|
addr_end = dump.ranges[-1][0]+dump.ranges[-1][1]
|
|
data = dump.data[addr_start:addr_end]
|
|
|
|
annotate_refs(grefs, addr_start, data,
|
|
width=args.name_width,
|
|
gaps=not args.no_gaps,
|
|
overlaps=args.overlaps)
|
|
|
|
if __name__ == '__main__':
|
|
exit(main())
|