#!/usr/bin/env python import fileinput import re import sys # Linker address generation validity checker. By default, GNU ld is # broken when faced with sections where the load address (i.e. the # spot in the XIP program binary where initialized data lives) differs # from the virtual address (i.e. the location in RAM where that data # will live at runtime. We need to be sure we're using the # ALIGN_WITH_INPUT feature correctly everywhere, which is hard -- # especially so given that many of these bugs are semi-invisible at # runtime (most initialized data is still a bunch of zeros and often # "works" even if it's wrong). # # This quick test just checks the offsets between sequential segments # with separate VMA/LMA addresses and verifies that the size deltas # are identical. # # Note that this is assuming that the address generation is always # in-order and that there is only one "ROM" LMA block. It's possible # to write a valid linker script that will fail this script, but we # don't have such a use case and one isn't forseen. section_re = re.compile('(?x)' # (allow whitespace) '^([a-zA-Z0-9_\.]+) \s+' # name ' (0x[0-9a-f]+) \s+' # addr ' (0x[0-9a-f]+) \s+') # size load_addr_re = re.compile('load address (0x[0-9a-f]+)') in_prologue = True lma = 0 last_sec = None for line in fileinput.input(): # Skip the header junk if line.find("Linker script and memory map") >= 0: in_prologue = False continue match = section_re.match(line.rstrip()) if match: sec = match.group(1) vma = int(match.group(2), 16) size = int(match.group(3), 16) lmatch = load_addr_re.search(line.rstrip()) if lmatch: lma = int(lmatch.group(1), 16) else: last_sec = None continue if last_sec != None: dv = vma - last_vma dl = lma - last_lma if dv != dl: sys.stderr.write("ERROR: section %s is %d bytes " "in the virtual/runtime address space, " "but only %d in the loaded/XIP section!\n" % (last_sec, dv, dl)) sys.exit(1) last_sec = sec last_vma = vma last_lma = lma