u-boot/tools/binman/elf.py

302 lines
9.8 KiB
Python
Raw Normal View History

# SPDX-License-Identifier: GPL-2.0+
# Copyright (c) 2016 Google, Inc
# Written by Simon Glass <sjg@chromium.org>
#
# Handle various things related to ELF images
#
from __future__ import print_function
from collections import namedtuple, OrderedDict
import command
import io
import os
import re
import shutil
import struct
import tempfile
import tools
import tout
ELF_TOOLS = True
try:
from elftools.elf.elffile import ELFFile
from elftools.elf.sections import SymbolTableSection
except: # pragma: no cover
ELF_TOOLS = False
Symbol = namedtuple('Symbol', ['section', 'address', 'size', 'weak'])
# Information about an ELF file:
# data: Extracted program contents of ELF file (this would be loaded by an
# ELF loader when reading this file
# load: Load address of code
# entry: Entry address of code
# memsize: Number of bytes in memory occupied by loading this ELF file
ElfInfo = namedtuple('ElfInfo', ['data', 'load', 'entry', 'memsize'])
def GetSymbols(fname, patterns):
"""Get the symbols from an ELF file
Args:
fname: Filename of the ELF file to read
patterns: List of regex patterns to search for, each a string
Returns:
None, if the file does not exist, or Dict:
key: Name of symbol
value: Hex value of symbol
"""
stdout = command.Output('objdump', '-t', fname, raise_on_error=False)
lines = stdout.splitlines()
if patterns:
re_syms = re.compile('|'.join(patterns))
else:
re_syms = None
syms = {}
syms_started = False
for line in lines:
if not line or not syms_started:
if 'SYMBOL TABLE' in line:
syms_started = True
line = None # Otherwise code coverage complains about 'continue'
continue
if re_syms and not re_syms.search(line):
continue
space_pos = line.find(' ')
value, rest = line[:space_pos], line[space_pos + 1:]
flags = rest[:7]
parts = rest[7:].split()
section, size = parts[:2]
if len(parts) > 2:
name = parts[2]
syms[name] = Symbol(section, int(value, 16), int(size,16),
flags[1] == 'w')
# Sort dict by address
return OrderedDict(sorted(syms.items(), key=lambda x: x[1].address))
def GetSymbolAddress(fname, sym_name):
"""Get a value of a symbol from an ELF file
Args:
fname: Filename of the ELF file to read
patterns: List of regex patterns to search for, each a string
Returns:
Symbol value (as an integer) or None if not found
"""
syms = GetSymbols(fname, [sym_name])
sym = syms.get(sym_name)
if not sym:
return None
return sym.address
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
def LookupAndWriteSymbols(elf_fname, entry, section):
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
"""Replace all symbols in an entry with their correct values
The entry contents is updated so that values for referenced symbols will be
visible at run time. This is done by finding out the symbols offsets in the
entry (using the ELF file) and replacing them with values from binman's data
structures.
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
Args:
elf_fname: Filename of ELF image containing the symbol information for
entry
entry: Entry to process
section: Section which can be used to lookup symbol values
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
"""
fname = tools.GetInputFilename(elf_fname)
syms = GetSymbols(fname, ['image', 'binman'])
if not syms:
return
base = syms.get('__image_copy_start')
if not base:
return
for name, sym in syms.items():
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
if name.startswith('_binman'):
msg = ("Section '%s': Symbol '%s'\n in entry '%s'" %
(section.GetPath(), name, entry.GetPath()))
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
offset = sym.address - base.address
if offset < 0 or offset + sym.size > entry.contents_size:
raise ValueError('%s has offset %x (size %x) but the contents '
'size is %x' % (entry.GetPath(), offset,
sym.size, entry.contents_size))
if sym.size == 4:
pack_string = '<I'
elif sym.size == 8:
pack_string = '<Q'
else:
raise ValueError('%s has size %d: only 4 and 8 are supported' %
(msg, sym.size))
# Look up the symbol in our entry tables.
value = section.LookupSymbol(name, sym.weak, msg)
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
if value is not None:
value += base.address
else:
value = -1
pack_string = pack_string.lower()
value_bytes = struct.pack(pack_string, value)
tout.Debug('%s:\n insert %s, offset %x, value %x, length %d' %
(msg, name, offset, value, len(value_bytes)))
binman: Support accessing binman tables at run time Binman construct images consisting of multiple binary files. These files sometimes need to know (at run timme) where their peers are located. For example, SPL may want to know where U-Boot is located in the image, so that it can jump to U-Boot correctly on boot. In general the positions where the binaries end up after binman has finished packing them cannot be known at compile time. One reason for this is that binman does not know the size of the binaries until everything is compiled, linked and converted to binaries with objcopy. To make this work, we add a feature to binman which checks each binary for symbol names starting with '_binman'. These are then decoded to figure out which entry and property they refer to. Then binman writes the value of this symbol into the appropriate binary. With this, the symbol will have the correct value at run time. Macros are used to make this easier to use. As an example, this declares a symbol that will access the 'u-boot-spl' entry to find the 'pos' value (i.e. the position of SPL in the image): binman_sym_declare(unsigned long, u_boot_spl, pos); This converts to a symbol called '_binman_u_boot_spl_prop_pos' in any binary that includes it. Binman then updates the value in that binary, ensuring that it can be accessed at runtime with: ulong u_boot_pos = binman_sym(ulong, u_boot_spl, pos); This assigns the variable u_boot_pos to the position of SPL in the image. Signed-off-by: Simon Glass <sjg@chromium.org>
2017-11-14 01:55:01 +00:00
entry.data = (entry.data[:offset] + value_bytes +
entry.data[offset + sym.size:])
def MakeElf(elf_fname, text, data):
"""Make an elf file with the given data in a single section
The output file has a several section including '.text' and '.data',
containing the info provided in arguments.
Args:
elf_fname: Output filename
text: Text (code) to put in the file's .text section
data: Data to put in the file's .data section
"""
outdir = tempfile.mkdtemp(prefix='binman.elf.')
s_file = os.path.join(outdir, 'elf.S')
# Spilt the text into two parts so that we can make the entry point two
# bytes after the start of the text section
text_bytes1 = ['\t.byte\t%#x' % tools.ToByte(byte) for byte in text[:2]]
text_bytes2 = ['\t.byte\t%#x' % tools.ToByte(byte) for byte in text[2:]]
data_bytes = ['\t.byte\t%#x' % tools.ToByte(byte) for byte in data]
with open(s_file, 'w') as fd:
print('''/* Auto-generated C program to produce an ELF file for testing */
.section .text
.code32
.globl _start
.type _start, @function
%s
_start:
%s
.ident "comment"
.comm fred,8,4
.section .empty
.globl _empty
_empty:
.byte 1
.globl ernie
.data
.type ernie, @object
.size ernie, 4
ernie:
%s
''' % ('\n'.join(text_bytes1), '\n'.join(text_bytes2), '\n'.join(data_bytes)),
file=fd)
lds_file = os.path.join(outdir, 'elf.lds')
# Use a linker script to set the alignment and text address.
with open(lds_file, 'w') as fd:
print('''/* Auto-generated linker script to produce an ELF file for testing */
PHDRS
{
text PT_LOAD ;
data PT_LOAD ;
empty PT_LOAD FLAGS ( 6 ) ;
note PT_NOTE ;
}
SECTIONS
{
. = 0xfef20000;
ENTRY(_start)
.text . : SUBALIGN(0)
{
*(.text)
} :text
.data : {
*(.data)
} :data
_bss_start = .;
.empty : {
*(.empty)
} :empty
.note : {
*(.comment)
} :note
.bss _bss_start (OVERLAY) : {
*(.bss)
}
}
''', file=fd)
# -static: Avoid requiring any shared libraries
# -nostdlib: Don't link with C library
# -Wl,--build-id=none: Don't generate a build ID, so that we just get the
# text section at the start
# -m32: Build for 32-bit x86
# -T...: Specifies the link script, which sets the start address
stdout = command.Output('cc', '-static', '-nostdlib', '-Wl,--build-id=none',
'-m32','-T', lds_file, '-o', elf_fname, s_file)
shutil.rmtree(outdir)
def DecodeElf(data, location):
"""Decode an ELF file and return information about it
Args:
data: Data from ELF file
location: Start address of data to return
Returns:
ElfInfo object containing information about the decoded ELF file
"""
file_size = len(data)
with io.BytesIO(data) as fd:
elf = ELFFile(fd)
data_start = 0xffffffff;
data_end = 0;
mem_end = 0;
virt_to_phys = 0;
for i in range(elf.num_segments()):
segment = elf.get_segment(i)
if segment['p_type'] != 'PT_LOAD' or not segment['p_memsz']:
skipped = 1 # To make code-coverage see this line
continue
start = segment['p_paddr']
mend = start + segment['p_memsz']
rend = start + segment['p_filesz']
data_start = min(data_start, start)
data_end = max(data_end, rend)
mem_end = max(mem_end, mend)
if not virt_to_phys:
virt_to_phys = segment['p_paddr'] - segment['p_vaddr']
output = bytearray(data_end - data_start)
for i in range(elf.num_segments()):
segment = elf.get_segment(i)
if segment['p_type'] != 'PT_LOAD' or not segment['p_memsz']:
skipped = 1 # To make code-coverage see this line
continue
start = segment['p_paddr']
offset = 0
if start < location:
offset = location - start
start = location
# A legal ELF file can have a program header with non-zero length
# but zero-length file size and a non-zero offset which, added
# together, are greater than input->size (i.e. the total file size).
# So we need to not even test in the case that p_filesz is zero.
# Note: All of this code is commented out since we don't have a test
# case for it.
size = segment['p_filesz']
#if not size:
#continue
#end = segment['p_offset'] + segment['p_filesz']
#if end > file_size:
#raise ValueError('Underflow copying out the segment. File has %#x bytes left, segment end is %#x\n',
#file_size, end)
output[start - data_start:start - data_start + size] = (
segment.data()[offset:])
return ElfInfo(output, data_start, elf.header['e_entry'] + virt_to_phys,
mem_end - data_start)