u-boot/tools/rkmux.py
Simon Glass 002c634c1f rockchip: Add a script to parse datasheets
This script has proved useful for parsing datasheets and creating register
shift/mask values for use in header files. Include it in case it is useful
for others.

Signed-off-by: Simon Glass <sjg@chromium.org>
2016-01-21 20:42:37 -07:00

219 lines
6.0 KiB
Python
Executable File

#!/usr/bin/python
# Script to create enums from datasheet register tables
#
# Usage:
#
# First, create a text file from the datasheet:
# pdftotext -layout /path/to/rockchip-3288-trm.pdf /tmp/asc
#
# Then use this script to output the #defines for a particular register:
# ./tools/rkmux.py GRF_GPIO4C_IOMUX
#
# It will create output suitable for putting in a header file, with SHIFT and
# MASK values for each bitfield in the register.
#
# Note: this tool is not perfect and you may need to edit the resulting code.
# But it should speed up the process.
import csv
import re
import sys
tab_to_col = 3
class RegField:
def __init__(self, cols=None):
if cols:
self.bits, self.attr, self.reset_val, self.desc = (
[x.strip() for x in cols])
self.desc = [self.desc]
else:
self.bits = ''
self.attr = ''
self.reset_val = ''
self.desc = []
def Setup(self, cols):
self.bits, self.attr, self.reset_val = cols[0:3]
if len(cols) > 3:
self.desc.append(cols[3])
def AddDesc(self, desc):
self.desc.append(desc)
def Show(self):
print self
print
self.__init__()
def __str__(self):
return '%s,%s,%s,%s' % (self.bits, self.attr, self.reset_val,
'\n'.join(self.desc))
class Printer:
def __init__(self, name):
self.first = True
self.name = name
self.re_sel = re.compile("[1-9]'b([01]+): (.*)")
def __enter__(self):
return self
def __exit__(self, type, value, traceback):
if not self.first:
self.output_footer()
def output_header(self):
print '/* %s */' % self.name
print 'enum {'
def output_footer(self):
print '};';
def output_regfield(self, regfield):
lines = regfield.desc
field = lines[0]
#print 'field:', field
if field in ['reserved', 'reserve', 'write_enable', 'write_mask']:
return
if field.endswith('_sel') or field.endswith('_con'):
field = field[:-4]
elif field.endswith(' iomux'):
field = field[:-6]
elif field.endswith('_mode') or field.endswith('_mask'):
field = field[:-5]
#else:
#print 'bad field %s' % field
#return
field = field.upper()
if ':' in regfield.bits:
bit_high, bit_low = [int(x) for x in regfield.bits.split(':')]
else:
bit_high = bit_low = int(regfield.bits)
bit_width = bit_high - bit_low + 1
mask = (1 << bit_width) - 1
if self.first:
self.first = False
self.output_header()
else:
print
out_enum(field, 'shift', bit_low)
out_enum(field, 'mask', mask)
next_val = -1
#print 'lines: %s', lines
for line in lines:
m = self.re_sel.match(line)
if m:
val, enum = int(m.group(1), 2), m.group(2)
if enum not in ['reserved', 'reserve']:
out_enum(field, enum, val, val == next_val)
next_val = val + 1
def process_file(name, fd):
field = RegField()
reg = ''
fields = []
def add_it(field):
if field.bits:
if reg == name:
fields.append(field)
field = RegField()
return field
def is_field_start(line):
if '=' in line or '+' in line:
return False
if (line.startswith('gpio') or line.startswith('peri_') or
line.endswith('_sel') or line.endswith('_con')):
return True
if not ' ' in line: # and '_' in line:
return True
return False
for line in fd:
line = line.rstrip()
if line[:4] in ['GRF_', 'PMU_', 'CRU_']:
field = add_it(field)
reg = line
do_this = name == reg
elif not line or not line.startswith(' '):
continue
line = line.replace('\xe2\x80\x99', "'")
leading = len(line) - len(line.lstrip())
line = line.lstrip()
cols = re.split(' *', line, 3)
if leading > 15 or (len(cols) > 3 and is_field_start(cols[3])):
if is_field_start(line):
field = add_it(field)
field.AddDesc(line)
else:
if cols[0] == 'Bit' or len(cols) < 3:
continue
#print
#print field
field = add_it(field)
field.Setup(cols)
field = add_it(field)
with Printer(name) as printer:
for field in fields:
#print field
printer.output_regfield(field)
#print
def out_enum(field, suffix, value, skip_val=False):
str = '%s_%s' % (field.upper(), suffix.upper())
if not skip_val:
tabs = tab_to_col - len(str) / 8
if value > 9:
val_str = '%#x' % value
else:
val_str = '%d' % value
str += '%s= %s' % ('\t' * tabs, val_str)
print '\t%s,' % str
# Process a CSV file, e.g. from tabula
def process_csv(name, fd):
reader = csv.reader(fd)
rows = []
field = RegField()
for row in reader:
#print field.desc
if not row[0]:
field.desc.append(row[3])
continue
if field.bits:
if field.bits != 'Bit':
rows.append(field)
#print row
field = RegField(row)
with Printer(name) as printer:
for row in rows:
#print field
printer.output_regfield(row)
#print
fname = sys.argv[1]
name = sys.argv[2]
# Read output from pdftotext -layout
if 1:
with open(fname, 'r') as fd:
process_file(name, fd)
# Use tabula
# It seems to be better at outputting text for an entire cell in one cell.
# But it does not always work. E.g. GRF_GPIO7CH_IOMUX.
# So there is no point in using it.
if 0:
with open(fname, 'r') as fd:
process_csv(name, fd)