diff options
Diffstat (limited to 'tools')
l--------- | tools/microcode-tool | 1 | ||||
-rwxr-xr-x | tools/microcode-tool.py | 253 |
2 files changed, 254 insertions, 0 deletions
diff --git a/tools/microcode-tool b/tools/microcode-tool new file mode 120000 index 0000000..8be8507 --- /dev/null +++ b/tools/microcode-tool @@ -0,0 +1 @@ +microcode-tool.py
\ No newline at end of file diff --git a/tools/microcode-tool.py b/tools/microcode-tool.py new file mode 100755 index 0000000..003716d --- /dev/null +++ b/tools/microcode-tool.py @@ -0,0 +1,253 @@ +#!/usr/bin/env python +# +# Copyright (c) 2014 Google, Inc +# +# SPDX-License-Identifier: GPL-2.0+ +# +# Intel microcode update tool + +from optparse import OptionParser +import os +import re +import struct +import sys + +MICROCODE_DIR = 'arch/x86/dts/microcode' + +class Microcode: + """Holds information about the microcode for a particular model of CPU. + + Attributes: + name: Name of the CPU this microcode is for, including any version + information (e.g. 'm12206a7_00000029') + model: Model code string (this is cpuid(1).eax, e.g. '206a7') + words: List of hex words containing the microcode. The first 16 words + are the public header. + """ + def __init__(self, name, data): + self.name = name + # Convert data into a list of hex words + self.words = [] + for value in ''.join(data).split(','): + hexval = value.strip() + if hexval: + self.words.append(int(hexval, 0)) + + # The model is in the 4rd hex word + self.model = '%x' % self.words[3] + +def ParseFile(fname): + """Parse a micrcode.dat file and return the component parts + + Args: + fname: Filename to parse + Returns: + 3-Tuple: + date: String containing date from the file's header + license_text: List of text lines for the license file + microcodes: List of Microcode objects from the file + """ + re_date = re.compile('/\* *(.* [0-9]{4}) *\*/$') + re_license = re.compile('/[^-*+] *(.*)$') + re_name = re.compile('/\* *(.*)\.inc *\*/', re.IGNORECASE) + microcodes = {} + license_text = [] + date = '' + data = [] + name = None + with open(fname) as fd: + for line in fd: + line = line.rstrip() + m_date = re_date.match(line) + m_license = re_license.match(line) + m_name = re_name.match(line) + if m_name: + if name: + microcodes[name] = Microcode(name, data) + name = m_name.group(1).lower() + data = [] + elif m_license: + license_text.append(m_license.group(1)) + elif m_date: + date = m_date.group(1) + else: + data.append(line) + if name: + microcodes[name] = Microcode(name, data) + return date, license_text, microcodes + +def List(date, microcodes, model): + """List the available microcode chunks + + Args: + date: Date of the microcode file + microcodes: Dict of Microcode objects indexed by name + model: Model string to search for, or None + """ + print 'Date: %s' % date + if model: + mcode_list, tried = FindMicrocode(microcodes, model.lower()) + print 'Matching models %s:' % (', '.join(tried)) + else: + print 'All models:' + mcode_list = [microcodes[m] for m in microcodes.keys()] + for mcode in mcode_list: + print '%-20s: model %s' % (mcode.name, mcode.model) + +def FindMicrocode(microcodes, model): + """Find all the microcode chunks which match the given model. + + This model is something like 306a9 (the value returned in eax from + cpuid(1) when running on Intel CPUs). But we allow a partial match, + omitting the last 1 or two characters to allow many families to have the + same microcode. + + If the model name is ambiguous we return a list of matches. + + Args: + microcodes: Dict of Microcode objects indexed by name + model: String containing model name to find + Returns: + Tuple: + List of matching Microcode objects + List of abbreviations we tried + """ + # Allow a full name to be used + mcode = microcodes.get(model) + if mcode: + return [mcode], [] + + tried = [] + found = [] + for i in range(3): + abbrev = model[:-i] if i else model + tried.append(abbrev) + for mcode in microcodes.values(): + if mcode.model.startswith(abbrev): + found.append(mcode) + if found: + break + return found, tried + +def CreateFile(date, license_text, mcode, outfile): + """Create a microcode file in U-Boot's .dtsi format + + Args: + date: String containing date of original microcode file + license: List of text lines for the license file + mcode: Microcode object to write + outfile: Filename to write to ('-' for stdout) + """ + out = '''/*%s + * --- + * This is a device tree fragment. Use #include to add these properties to a + * node. + * + * Date: %s + */ + +compatible = "intel,microcode"; +intel,header-version = <%d>; +intel,update-revision = <%#x>; +intel,date-code = <%#x>; +intel,processor-signature = <%#x>; +intel,checksum = <%#x>; +intel,loader-revision = <%d>; +intel,processor-flags = <%#x>; + +/* The first 48-bytes are the public header which repeats the above data */ +data = <%s +\t>;''' + words = '' + for i in range(len(mcode.words)): + if not (i & 3): + words += '\n' + val = mcode.words[i] + # Change each word so it will be little-endian in the FDT + # This data is needed before RAM is available on some platforms so we + # cannot do an endianness swap on boot. + val = struct.unpack("<I", struct.pack(">I", val))[0] + words += '\t%#010x' % val + + # Take care to avoid adding a space before a tab + text = '' + for line in license_text: + if line[0] == '\t': + text += '\n *' + line + else: + text += '\n * ' + line + args = [text, date] + args += [mcode.words[i] for i in range(7)] + args.append(words) + if outfile == '-': + print out % tuple(args) + else: + if not outfile: + if not os.path.exists(MICROCODE_DIR): + print >> sys.stderr, "Creating directory '%s'" % MICROCODE_DIR + os.makedirs(MICROCODE_DIR) + outfile = os.path.join(MICROCODE_DIR, mcode.name + '.dtsi') + print >> sys.stderr, "Writing microcode for '%s' to '%s'" % ( + mcode.name, outfile) + with open(outfile, 'w') as fd: + print >> fd, out % tuple(args) + +def MicrocodeTool(): + """Run the microcode tool""" + commands = 'create,license,list'.split(',') + parser = OptionParser() + parser.add_option('-d', '--mcfile', type='string', action='store', + help='Name of microcode.dat file') + parser.add_option('-m', '--model', type='string', action='store', + help='Model name to extract') + parser.add_option('-o', '--outfile', type='string', action='store', + help='Filename to use for output (- for stdout), default is' + ' %s/<name>.dtsi' % MICROCODE_DIR) + parser.usage += """ command + + Process an Intel microcode file (use -h for help). Commands: + + create Create microcode .dtsi file for a model + list List available models in microcode file + license Print the license + + Typical usage: + + ./tools/microcode-tool -d microcode.dat -m 306a create + + This will find the appropriate file and write it to %s.""" % MICROCODE_DIR + + (options, args) = parser.parse_args() + if not args: + parser.error('Please specify a command') + cmd = args[0] + if cmd not in commands: + parser.error("Unknown command '%s'" % cmd) + + if not options.mcfile: + parser.error('You must specify a microcode file') + date, license_text, microcodes = ParseFile(options.mcfile) + + if cmd == 'list': + List(date, microcodes, options.model) + elif cmd == 'license': + print '\n'.join(license_text) + elif cmd == 'create': + if not options.model: + parser.error('You must specify a model to create') + model = options.model.lower() + mcode_list, tried = FindMicrocode(microcodes, model) + if not mcode_list: + parser.error("Unknown model '%s' (%s) - try 'list' to list" % + (model, ', '.join(tried))) + if len(mcode_list) > 1: + parser.error("Ambiguous model '%s' (%s) matched %s - try 'list' " + "to list or specify a particular file" % + (model, ', '.join(tried), + ', '.join([m.name for m in mcode_list]))) + CreateFile(date, license_text, mcode_list[0], options.outfile) + else: + parser.error("Unknown command '%s'" % cmd) + +if __name__ == "__main__": + MicrocodeTool() |