You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1057 lines
39 KiB
1057 lines
39 KiB
#!/usr/bin/env python3 |
|
|
|
# Copyright (c) 2019 - 2020 Nordic Semiconductor ASA |
|
# Copyright (c) 2019 Linaro Limited |
|
# SPDX-License-Identifier: BSD-3-Clause |
|
|
|
# This script uses edtlib to generate a header file from a devicetree |
|
# (.dts) file. Information from binding files in YAML format is used |
|
# as well. |
|
# |
|
# Bindings are files that describe devicetree nodes. Devicetree nodes are |
|
# usually mapped to bindings via their 'compatible = "..."' property. |
|
# |
|
# See Zephyr's Devicetree user guide for details. |
|
# |
|
# Note: Do not access private (_-prefixed) identifiers from edtlib here (and |
|
# also note that edtlib is not meant to expose the dtlib API directly). |
|
# Instead, think of what API you need, and add it as a public documented API in |
|
# edtlib. This will keep this script simple. |
|
|
|
import argparse |
|
from collections import defaultdict |
|
import logging |
|
import os |
|
import pathlib |
|
import pickle |
|
import re |
|
import sys |
|
|
|
sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'python-devicetree', |
|
'src')) |
|
|
|
from devicetree import edtlib |
|
|
|
# The set of binding types whose values can be iterated over with |
|
# DT_FOREACH_PROP_ELEM(). If you change this, make sure to update the |
|
# doxygen string for that macro. |
|
FOREACH_PROP_ELEM_TYPES = set(['string', 'array', 'uint8-array', 'string-array', |
|
'phandles', 'phandle-array']) |
|
|
|
class LogFormatter(logging.Formatter): |
|
'''A log formatter that prints the level name in lower case, |
|
for compatibility with earlier versions of edtlib.''' |
|
|
|
def __init__(self): |
|
super().__init__(fmt='%(levelnamelower)s: %(message)s') |
|
|
|
def format(self, record): |
|
record.levelnamelower = record.levelname.lower() |
|
return super().format(record) |
|
|
|
def main(): |
|
global header_file |
|
global flash_area_num |
|
|
|
args = parse_args() |
|
|
|
setup_edtlib_logging() |
|
|
|
vendor_prefixes = {} |
|
for prefixes_file in args.vendor_prefixes: |
|
vendor_prefixes.update(edtlib.load_vendor_prefixes_txt(prefixes_file)) |
|
|
|
try: |
|
edt = edtlib.EDT(args.dts, args.bindings_dirs, |
|
# Suppress this warning if it's suppressed in dtc |
|
warn_reg_unit_address_mismatch= |
|
"-Wno-simple_bus_reg" not in args.dtc_flags, |
|
default_prop_types=True, |
|
infer_binding_for_paths=["/zephyr,user"], |
|
werror=args.edtlib_Werror, |
|
vendor_prefixes=vendor_prefixes) |
|
except edtlib.EDTError as e: |
|
sys.exit(f"devicetree error: {e}") |
|
|
|
flash_area_num = 0 |
|
|
|
# Save merged DTS source, as a debugging aid |
|
with open(args.dts_out, "w", encoding="utf-8") as f: |
|
print(edt.dts_source, file=f) |
|
|
|
# The raw index into edt.compat2nodes[compat] is used for node |
|
# instance numbering within a compatible. |
|
# |
|
# As a way to satisfy people's intuitions about instance numbers, |
|
# though, we sort this list so enabled instances come first. |
|
# |
|
# This might look like a hack, but it keeps drivers and |
|
# applications which don't use instance numbers carefully working |
|
# as expected, since e.g. instance number 0 is always the |
|
# singleton instance if there's just one enabled node of a |
|
# particular compatible. |
|
# |
|
# This doesn't violate any devicetree.h API guarantees about |
|
# instance ordering, since we make no promises that instance |
|
# numbers are stable across builds. |
|
for compat, nodes in edt.compat2nodes.items(): |
|
edt.compat2nodes[compat] = sorted( |
|
nodes, key=lambda node: 0 if node.status == "okay" else 1) |
|
|
|
# Create the generated header. |
|
with open(args.header_out, "w", encoding="utf-8") as header_file: |
|
write_top_comment(edt) |
|
|
|
write_utils() |
|
|
|
# populate all z_path_id first so any children references will |
|
# work correctly. |
|
for node in sorted(edt.nodes, key=lambda node: node.dep_ordinal): |
|
node.z_path_id = node_z_path_id(node) |
|
|
|
# Check to see if we have duplicate "zephyr,memory-region" property values. |
|
regions = dict() |
|
for node in sorted(edt.nodes, key=lambda node: node.dep_ordinal): |
|
if 'zephyr,memory-region' in node.props: |
|
region = node.props['zephyr,memory-region'].val |
|
if region in regions: |
|
sys.exit(f"ERROR: Duplicate 'zephyr,memory-region' ({region}) properties " |
|
f"between {regions[region].path} and {node.path}") |
|
regions[region] = node |
|
|
|
for node in sorted(edt.nodes, key=lambda node: node.dep_ordinal): |
|
write_node_comment(node) |
|
|
|
out_comment("Node's full path:") |
|
out_dt_define(f"{node.z_path_id}_PATH", f'"{escape(node.path)}"') |
|
|
|
out_comment("Node's name with unit-address:") |
|
out_dt_define(f"{node.z_path_id}_FULL_NAME", |
|
f'"{escape(node.name)}"') |
|
|
|
if node.parent is not None: |
|
out_comment(f"Node parent ({node.parent.path}) identifier:") |
|
out_dt_define(f"{node.z_path_id}_PARENT", |
|
f"DT_{node.parent.z_path_id}") |
|
|
|
out_comment(f"Node's index in its parent's list of children:") |
|
out_dt_define(f"{node.z_path_id}_CHILD_IDX", |
|
node.parent.child_index(node)) |
|
|
|
write_children(node) |
|
write_dep_info(node) |
|
write_idents_and_existence(node) |
|
write_bus(node) |
|
write_special_props(node) |
|
write_vanilla_props(node) |
|
|
|
write_chosen(edt) |
|
write_global_macros(edt) |
|
|
|
if args.edt_pickle_out: |
|
write_pickled_edt(edt, args.edt_pickle_out) |
|
|
|
|
|
def setup_edtlib_logging(): |
|
# The edtlib module emits logs using the standard 'logging' module. |
|
# Configure it so that warnings and above are printed to stderr, |
|
# using the LogFormatter class defined above to format each message. |
|
|
|
handler = logging.StreamHandler(sys.stderr) |
|
handler.setFormatter(LogFormatter()) |
|
|
|
logger = logging.getLogger('edtlib') |
|
logger.setLevel(logging.WARNING) |
|
logger.addHandler(handler) |
|
|
|
def node_z_path_id(node): |
|
# Return the node specific bit of the node's path identifier: |
|
# |
|
# - the root node's path "/" has path identifier "N" |
|
# - "/foo" has "N_S_foo" |
|
# - "/foo/bar" has "N_S_foo_S_bar" |
|
# - "/foo/bar@123" has "N_S_foo_S_bar_123" |
|
# |
|
# This is used throughout this file to generate macros related to |
|
# the node. |
|
|
|
components = ["N"] |
|
if node.parent is not None: |
|
components.extend(f"S_{str2ident(component)}" for component in |
|
node.path.split("/")[1:]) |
|
|
|
return "_".join(components) |
|
|
|
def parse_args(): |
|
# Returns parsed command-line arguments |
|
|
|
parser = argparse.ArgumentParser(allow_abbrev=False) |
|
parser.add_argument("--dts", required=True, help="DTS file") |
|
parser.add_argument("--dtc-flags", |
|
help="'dtc' devicetree compiler flags, some of which " |
|
"might be respected here") |
|
parser.add_argument("--bindings-dirs", nargs='+', required=True, |
|
help="directory with bindings in YAML format, " |
|
"we allow multiple") |
|
parser.add_argument("--header-out", required=True, |
|
help="path to write header to") |
|
parser.add_argument("--dts-out", required=True, |
|
help="path to write merged DTS source code to (e.g. " |
|
"as a debugging aid)") |
|
parser.add_argument("--edt-pickle-out", |
|
help="path to write pickled edtlib.EDT object to") |
|
parser.add_argument("--vendor-prefixes", action='append', default=[], |
|
help="vendor-prefixes.txt path; used for validation; " |
|
"may be given multiple times") |
|
parser.add_argument("--edtlib-Werror", action="store_true", |
|
help="if set, edtlib-specific warnings become errors. " |
|
"(this does not apply to warnings shared " |
|
"with dtc.)") |
|
|
|
return parser.parse_args() |
|
|
|
|
|
def write_top_comment(edt): |
|
# Writes an overview comment with misc. info at the top of the header and |
|
# configuration file |
|
|
|
s = f"""\ |
|
Generated by gen_defines.py |
|
|
|
DTS input file: |
|
{edt.dts_path} |
|
|
|
Directories with bindings: |
|
{", ".join(map(relativize, edt.bindings_dirs))} |
|
|
|
Node dependency ordering (ordinal and path): |
|
""" |
|
|
|
for scc in edt.scc_order: |
|
if len(scc) > 1: |
|
err("cycle in devicetree involving " |
|
+ ", ".join(node.path for node in scc)) |
|
s += f" {scc[0].dep_ordinal:<3} {scc[0].path}\n" |
|
|
|
s += """ |
|
Definitions derived from these nodes in dependency order are next, |
|
followed by /chosen nodes. |
|
""" |
|
|
|
out_comment(s, blank_before=False) |
|
|
|
|
|
def write_utils(): |
|
# Writes utility macros |
|
|
|
out_comment("Used to remove brackets from around a single argument") |
|
out_define("DT_DEBRACKET_INTERNAL(...)", "__VA_ARGS__") |
|
|
|
|
|
def write_node_comment(node): |
|
# Writes a comment describing 'node' to the header and configuration file |
|
|
|
s = f"""\ |
|
Devicetree node: {node.path} |
|
|
|
Node identifier: DT_{node.z_path_id} |
|
""" |
|
|
|
if node.matching_compat: |
|
if node.binding_path: |
|
s += f""" |
|
Binding (compatible = {node.matching_compat}): |
|
{relativize(node.binding_path)} |
|
""" |
|
else: |
|
s += f""" |
|
Binding (compatible = {node.matching_compat}): |
|
No yaml (bindings inferred from properties) |
|
""" |
|
|
|
if node.description: |
|
# We used to put descriptions in the generated file, but |
|
# devicetree bindings now have pages in the HTML |
|
# documentation. Let users who are accustomed to digging |
|
# around in the generated file where to find the descriptions |
|
# now. |
|
# |
|
# Keeping them here would mean that the descriptions |
|
# themselves couldn't contain C multi-line comments, which is |
|
# inconvenient when we want to do things like quote snippets |
|
# of .dtsi files within the descriptions, or otherwise |
|
# include the string "*/". |
|
s += ("\n(Descriptions have moved to the Devicetree Bindings Index\n" |
|
"in the documentation.)\n") |
|
|
|
out_comment(s) |
|
|
|
|
|
def relativize(path): |
|
# If 'path' is within $ZEPHYR_BASE, returns it relative to $ZEPHYR_BASE, |
|
# with a "$ZEPHYR_BASE/..." hint at the start of the string. Otherwise, |
|
# returns 'path' unchanged. |
|
|
|
zbase = os.getenv("ZEPHYR_BASE") |
|
if zbase is None: |
|
return path |
|
|
|
try: |
|
return str("$ZEPHYR_BASE" / pathlib.Path(path).relative_to(zbase)) |
|
except ValueError: |
|
# Not within ZEPHYR_BASE |
|
return path |
|
|
|
|
|
def write_idents_and_existence(node): |
|
# Writes macros related to the node's aliases, labels, etc., |
|
# as well as existence flags. |
|
|
|
# Aliases |
|
idents = [f"N_ALIAS_{str2ident(alias)}" for alias in node.aliases] |
|
# Instances |
|
for compat in node.compats: |
|
instance_no = node.edt.compat2nodes[compat].index(node) |
|
idents.append(f"N_INST_{instance_no}_{str2ident(compat)}") |
|
# Node labels |
|
idents.extend(f"N_NODELABEL_{str2ident(label)}" for label in node.labels) |
|
|
|
out_comment("Existence and alternate IDs:") |
|
out_dt_define(node.z_path_id + "_EXISTS", 1) |
|
|
|
# Only determine maxlen if we have any idents |
|
if idents: |
|
maxlen = max(len("DT_" + ident) for ident in idents) |
|
for ident in idents: |
|
out_dt_define(ident, "DT_" + node.z_path_id, width=maxlen) |
|
|
|
|
|
def write_bus(node): |
|
# Macros about the node's bus controller, if there is one |
|
|
|
bus = node.bus_node |
|
if not bus: |
|
return |
|
|
|
out_comment(f"Bus info (controller: '{bus.path}', type: '{node.on_buses}')") |
|
|
|
for one_bus in node.on_buses: |
|
out_dt_define(f"{node.z_path_id}_BUS_{str2ident(one_bus)}", 1) |
|
|
|
out_dt_define(f"{node.z_path_id}_BUS", f"DT_{bus.z_path_id}") |
|
|
|
|
|
def write_special_props(node): |
|
# Writes required macros for special case properties, when the |
|
# data cannot otherwise be obtained from write_vanilla_props() |
|
# results |
|
|
|
# Macros that are special to the devicetree specification |
|
out_comment("Macros for properties that are special in the specification:") |
|
write_regs(node) |
|
write_ranges(node) |
|
write_interrupts(node) |
|
write_compatibles(node) |
|
write_status(node) |
|
|
|
# Macros that are special to bindings inherited from Linux, which |
|
# we can't capture with the current bindings language. |
|
write_pinctrls(node) |
|
write_fixed_partitions(node) |
|
|
|
def write_ranges(node): |
|
# ranges property: edtlib knows the right #address-cells and |
|
# #size-cells of parent and child, and can therefore pack the |
|
# child & parent addresses and sizes correctly |
|
|
|
idx_vals = [] |
|
path_id = node.z_path_id |
|
|
|
if node.ranges is not None: |
|
idx_vals.append((f"{path_id}_RANGES_NUM", len(node.ranges))) |
|
|
|
for i,range in enumerate(node.ranges): |
|
idx_vals.append((f"{path_id}_RANGES_IDX_{i}_EXISTS", 1)) |
|
|
|
if "pcie" in node.buses: |
|
idx_vals.append((f"{path_id}_RANGES_IDX_{i}_VAL_CHILD_BUS_FLAGS_EXISTS", 1)) |
|
idx_macro = f"{path_id}_RANGES_IDX_{i}_VAL_CHILD_BUS_FLAGS" |
|
idx_value = range.child_bus_addr >> ((range.child_bus_cells - 1) * 32) |
|
idx_vals.append((idx_macro, |
|
f"{idx_value} /* {hex(idx_value)} */")) |
|
if range.child_bus_addr is not None: |
|
idx_macro = f"{path_id}_RANGES_IDX_{i}_VAL_CHILD_BUS_ADDRESS" |
|
if "pcie" in node.buses: |
|
idx_value = range.child_bus_addr & ((1 << (range.child_bus_cells - 1) * 32) - 1) |
|
else: |
|
idx_value = range.child_bus_addr |
|
idx_vals.append((idx_macro, |
|
f"{idx_value} /* {hex(idx_value)} */")) |
|
if range.parent_bus_addr is not None: |
|
idx_macro = f"{path_id}_RANGES_IDX_{i}_VAL_PARENT_BUS_ADDRESS" |
|
idx_vals.append((idx_macro, |
|
f"{range.parent_bus_addr} /* {hex(range.parent_bus_addr)} */")) |
|
if range.length is not None: |
|
idx_macro = f"{path_id}_RANGES_IDX_{i}_VAL_LENGTH" |
|
idx_vals.append((idx_macro, |
|
f"{range.length} /* {hex(range.length)} */")) |
|
|
|
for macro, val in idx_vals: |
|
out_dt_define(macro, val) |
|
|
|
out_dt_define(f"{path_id}_FOREACH_RANGE(fn)", |
|
" ".join(f"fn(DT_{path_id}, {i})" for i,range in enumerate(node.ranges))) |
|
|
|
def write_regs(node): |
|
# reg property: edtlib knows the right #address-cells and |
|
# #size-cells, and can therefore pack the register base addresses |
|
# and sizes correctly |
|
|
|
idx_vals = [] |
|
name_vals = [] |
|
path_id = node.z_path_id |
|
|
|
if node.regs is not None: |
|
idx_vals.append((f"{path_id}_REG_NUM", len(node.regs))) |
|
|
|
for i, reg in enumerate(node.regs): |
|
idx_vals.append((f"{path_id}_REG_IDX_{i}_EXISTS", 1)) |
|
if reg.addr is not None: |
|
idx_macro = f"{path_id}_REG_IDX_{i}_VAL_ADDRESS" |
|
idx_vals.append((idx_macro, |
|
f"{reg.addr} /* {hex(reg.addr)} */")) |
|
if reg.name: |
|
name_macro = f"{path_id}_REG_NAME_{reg.name}_VAL_ADDRESS" |
|
name_vals.append((name_macro, f"DT_{idx_macro}")) |
|
|
|
if reg.size is not None: |
|
idx_macro = f"{path_id}_REG_IDX_{i}_VAL_SIZE" |
|
idx_vals.append((idx_macro, |
|
f"{reg.size} /* {hex(reg.size)} */")) |
|
if reg.name: |
|
name_macro = f"{path_id}_REG_NAME_{reg.name}_VAL_SIZE" |
|
name_vals.append((name_macro, f"DT_{idx_macro}")) |
|
|
|
for macro, val in idx_vals: |
|
out_dt_define(macro, val) |
|
for macro, val in name_vals: |
|
out_dt_define(macro, val) |
|
|
|
def write_interrupts(node): |
|
# interrupts property: we have some hard-coded logic for interrupt |
|
# mapping here. |
|
# |
|
# TODO: can we push map_arm_gic_irq_type() and |
|
# encode_zephyr_multi_level_irq() out of Python and into C with |
|
# macro magic in devicetree.h? |
|
|
|
def map_arm_gic_irq_type(irq, irq_num): |
|
# Maps ARM GIC IRQ (type)+(index) combo to linear IRQ number |
|
if "type" not in irq.data: |
|
err(f"Expected binding for {irq.controller!r} to have 'type' in " |
|
"interrupt-cells") |
|
irq_type = irq.data["type"] |
|
|
|
if irq_type == 0: # GIC_SPI |
|
return irq_num + 32 |
|
if irq_type == 1: # GIC_PPI |
|
return irq_num + 16 |
|
err(f"Invalid interrupt type specified for {irq!r}") |
|
|
|
def encode_zephyr_multi_level_irq(irq, irq_num): |
|
# See doc/reference/kernel/other/interrupts.rst for details |
|
# on how this encoding works |
|
|
|
irq_ctrl = irq.controller |
|
# Look for interrupt controller parent until we have none |
|
while irq_ctrl.interrupts: |
|
irq_num = (irq_num + 1) << 8 |
|
if "irq" not in irq_ctrl.interrupts[0].data: |
|
err(f"Expected binding for {irq_ctrl!r} to have 'irq' in " |
|
"interrupt-cells") |
|
irq_num |= irq_ctrl.interrupts[0].data["irq"] |
|
irq_ctrl = irq_ctrl.interrupts[0].controller |
|
return irq_num |
|
|
|
idx_vals = [] |
|
name_vals = [] |
|
path_id = node.z_path_id |
|
|
|
if node.interrupts is not None: |
|
idx_vals.append((f"{path_id}_IRQ_NUM", len(node.interrupts))) |
|
|
|
for i, irq in enumerate(node.interrupts): |
|
for cell_name, cell_value in irq.data.items(): |
|
name = str2ident(cell_name) |
|
|
|
if cell_name == "irq": |
|
if "arm,gic" in irq.controller.compats: |
|
cell_value = map_arm_gic_irq_type(irq, cell_value) |
|
cell_value = encode_zephyr_multi_level_irq(irq, cell_value) |
|
|
|
idx_vals.append((f"{path_id}_IRQ_IDX_{i}_EXISTS", 1)) |
|
idx_macro = f"{path_id}_IRQ_IDX_{i}_VAL_{name}" |
|
idx_vals.append((idx_macro, cell_value)) |
|
idx_vals.append((idx_macro + "_EXISTS", 1)) |
|
if irq.name: |
|
name_macro = \ |
|
f"{path_id}_IRQ_NAME_{str2ident(irq.name)}_VAL_{name}" |
|
name_vals.append((name_macro, f"DT_{idx_macro}")) |
|
name_vals.append((name_macro + "_EXISTS", 1)) |
|
|
|
for macro, val in idx_vals: |
|
out_dt_define(macro, val) |
|
for macro, val in name_vals: |
|
out_dt_define(macro, val) |
|
|
|
|
|
def write_compatibles(node): |
|
# Writes a macro for each of the node's compatibles. We don't care |
|
# about whether edtlib / Zephyr's binding language recognizes |
|
# them. The compatibles the node provides are what is important. |
|
|
|
for i, compat in enumerate(node.compats): |
|
out_dt_define( |
|
f"{node.z_path_id}_COMPAT_MATCHES_{str2ident(compat)}", 1) |
|
|
|
if node.edt.compat2vendor[compat]: |
|
out_dt_define(f"{node.z_path_id}_COMPAT_VENDOR_IDX_{i}_EXISTS", 1) |
|
out_dt_define(f"{node.z_path_id}_COMPAT_VENDOR_IDX_{i}", |
|
quote_str(node.edt.compat2vendor[compat])) |
|
|
|
if node.edt.compat2model[compat]: |
|
out_dt_define(f"{node.z_path_id}_COMPAT_MODEL_IDX_{i}_EXISTS", 1) |
|
out_dt_define(f"{node.z_path_id}_COMPAT_MODEL_IDX_{i}", |
|
quote_str(node.edt.compat2model[compat])) |
|
|
|
def write_children(node): |
|
# Writes helper macros for dealing with node's children. |
|
|
|
out_comment("Helper macros for child nodes of this node.") |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD(fn)", |
|
" ".join(f"fn(DT_{child.z_path_id})" for child in |
|
node.children.values())) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_SEP(fn, sep)", |
|
" DT_DEBRACKET_INTERNAL sep ".join(f"fn(DT_{child.z_path_id})" |
|
for child in node.children.values())) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_VARGS(fn, ...)", |
|
" ".join(f"fn(DT_{child.z_path_id}, __VA_ARGS__)" |
|
for child in node.children.values())) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_SEP_VARGS(fn, sep, ...)", |
|
" DT_DEBRACKET_INTERNAL sep ".join(f"fn(DT_{child.z_path_id}, __VA_ARGS__)" |
|
for child in node.children.values())) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_STATUS_OKAY(fn)", |
|
" ".join(f"fn(DT_{child.z_path_id})" |
|
for child in node.children.values() if child.status == "okay")) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_STATUS_OKAY_SEP(fn, sep)", |
|
" DT_DEBRACKET_INTERNAL sep ".join(f"fn(DT_{child.z_path_id})" |
|
for child in node.children.values() if child.status == "okay")) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_STATUS_OKAY_VARGS(fn, ...)", |
|
" ".join(f"fn(DT_{child.z_path_id}, __VA_ARGS__)" |
|
for child in node.children.values() if child.status == "okay")) |
|
|
|
out_dt_define(f"{node.z_path_id}_FOREACH_CHILD_STATUS_OKAY_SEP_VARGS(fn, sep, ...)", |
|
" DT_DEBRACKET_INTERNAL sep ".join(f"fn(DT_{child.z_path_id}, __VA_ARGS__)" |
|
for child in node.children.values() if child.status == "okay")) |
|
|
|
|
|
def write_status(node): |
|
out_dt_define(f"{node.z_path_id}_STATUS_{str2ident(node.status)}", 1) |
|
|
|
|
|
def write_pinctrls(node): |
|
# Write special macros for pinctrl-<index> and pinctrl-names properties. |
|
|
|
out_comment("Pin control (pinctrl-<i>, pinctrl-names) properties:") |
|
|
|
out_dt_define(f"{node.z_path_id}_PINCTRL_NUM", len(node.pinctrls)) |
|
|
|
if not node.pinctrls: |
|
return |
|
|
|
for pc_idx, pinctrl in enumerate(node.pinctrls): |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_IDX_{pc_idx}_EXISTS", 1) |
|
|
|
if not pinctrl.name: |
|
continue |
|
|
|
name = pinctrl.name_as_token |
|
|
|
# Below we rely on the fact that edtlib ensures the |
|
# pinctrl-<pc_idx> properties are contiguous, start from 0, |
|
# and contain only phandles. |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_IDX_{pc_idx}_TOKEN", name) |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_IDX_{pc_idx}_UPPER_TOKEN", name.upper()) |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_NAME_{name}_EXISTS", 1) |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_NAME_{name}_IDX", pc_idx) |
|
for idx, ph in enumerate(pinctrl.conf_nodes): |
|
out_dt_define(f"{node.z_path_id}_PINCTRL_NAME_{name}_IDX_{idx}_PH", |
|
f"DT_{ph.z_path_id}") |
|
|
|
|
|
def write_fixed_partitions(node): |
|
# Macros for child nodes of each fixed-partitions node. |
|
|
|
if not (node.parent and "fixed-partitions" in node.parent.compats): |
|
return |
|
|
|
global flash_area_num |
|
out_comment("fixed-partitions identifier:") |
|
out_dt_define(f"{node.z_path_id}_PARTITION_ID", flash_area_num) |
|
flash_area_num += 1 |
|
|
|
|
|
def write_vanilla_props(node): |
|
# Writes macros for any and all properties defined in the |
|
# "properties" section of the binding for the node. |
|
# |
|
# This does generate macros for special properties as well, like |
|
# regs, etc. Just let that be rather than bothering to add |
|
# never-ending amounts of special case code here to skip special |
|
# properties. This function's macros can't conflict with |
|
# write_special_props() macros, because they're in different |
|
# namespaces. Special cases aren't special enough to break the rules. |
|
|
|
macro2val = {} |
|
for prop_name, prop in node.props.items(): |
|
prop_id = str2ident(prop_name) |
|
macro = f"{node.z_path_id}_P_{prop_id}" |
|
val = prop2value(prop) |
|
if val is not None: |
|
# DT_N_<node-id>_P_<prop-id> |
|
macro2val[macro] = val |
|
|
|
if prop.spec.type == 'string': |
|
macro2val[macro + "_STRING_TOKEN"] = prop.val_as_token |
|
macro2val[macro + "_STRING_UPPER_TOKEN"] = prop.val_as_token.upper() |
|
|
|
if prop.enum_index is not None: |
|
# DT_N_<node-id>_P_<prop-id>_ENUM_IDX |
|
macro2val[macro + "_ENUM_IDX"] = prop.enum_index |
|
spec = prop.spec |
|
|
|
if spec.enum_tokenizable: |
|
as_token = prop.val_as_token |
|
|
|
# DT_N_<node-id>_P_<prop-id>_ENUM_TOKEN |
|
macro2val[macro + "_ENUM_TOKEN"] = as_token |
|
|
|
if spec.enum_upper_tokenizable: |
|
# DT_N_<node-id>_P_<prop-id>_ENUM_UPPER_TOKEN |
|
macro2val[macro + "_ENUM_UPPER_TOKEN"] = as_token.upper() |
|
|
|
if "phandle" in prop.type: |
|
macro2val.update(phandle_macros(prop, macro)) |
|
elif "array" in prop.type: |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i> |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_EXISTS |
|
for i, subval in enumerate(prop.val): |
|
if isinstance(subval, str): |
|
macro2val[macro + f"_IDX_{i}"] = quote_str(subval) |
|
subval_as_token = edtlib.str_as_token(subval) |
|
macro2val[macro + f"_IDX_{i}_STRING_TOKEN"] = subval_as_token |
|
macro2val[macro + f"_IDX_{i}_STRING_UPPER_TOKEN"] = subval_as_token.upper() |
|
else: |
|
macro2val[macro + f"_IDX_{i}"] = subval |
|
macro2val[macro + f"_IDX_{i}_EXISTS"] = 1 |
|
|
|
if prop.type in FOREACH_PROP_ELEM_TYPES: |
|
# DT_N_<node-id>_P_<prop-id>_FOREACH_PROP_ELEM |
|
macro2val[f"{macro}_FOREACH_PROP_ELEM(fn)"] = \ |
|
' \\\n\t'.join( |
|
f'fn(DT_{node.z_path_id}, {prop_id}, {i})' |
|
for i in range(len(prop.val))) |
|
|
|
macro2val[f"{macro}_FOREACH_PROP_ELEM_SEP(fn, sep)"] = \ |
|
' DT_DEBRACKET_INTERNAL sep \\\n\t'.join( |
|
f'fn(DT_{node.z_path_id}, {prop_id}, {i})' |
|
for i in range(len(prop.val))) |
|
|
|
macro2val[f"{macro}_FOREACH_PROP_ELEM_VARGS(fn, ...)"] = \ |
|
' \\\n\t'.join( |
|
f'fn(DT_{node.z_path_id}, {prop_id}, {i}, __VA_ARGS__)' |
|
for i in range(len(prop.val))) |
|
|
|
macro2val[f"{macro}_FOREACH_PROP_ELEM_SEP_VARGS(fn, sep, ...)"] = \ |
|
' DT_DEBRACKET_INTERNAL sep \\\n\t'.join( |
|
f'fn(DT_{node.z_path_id}, {prop_id}, {i}, __VA_ARGS__)' |
|
for i in range(len(prop.val))) |
|
|
|
plen = prop_len(prop) |
|
if plen is not None: |
|
# DT_N_<node-id>_P_<prop-id>_LEN |
|
macro2val[macro + "_LEN"] = plen |
|
|
|
macro2val[f"{macro}_EXISTS"] = 1 |
|
|
|
if macro2val: |
|
out_comment("Generic property macros:") |
|
for macro, val in macro2val.items(): |
|
out_dt_define(macro, val) |
|
else: |
|
out_comment("(No generic property macros)") |
|
|
|
|
|
def write_dep_info(node): |
|
# Write dependency-related information about the node. |
|
|
|
def fmt_dep_list(dep_list): |
|
if dep_list: |
|
# Sort the list by dependency ordinal for predictability. |
|
sorted_list = sorted(dep_list, key=lambda node: node.dep_ordinal) |
|
return "\\\n\t" + \ |
|
" \\\n\t".join(f"{n.dep_ordinal}, /* {n.path} */" |
|
for n in sorted_list) |
|
else: |
|
return "/* nothing */" |
|
|
|
out_comment("Node's dependency ordinal:") |
|
out_dt_define(f"{node.z_path_id}_ORD", node.dep_ordinal) |
|
|
|
out_comment("Ordinals for what this node depends on directly:") |
|
out_dt_define(f"{node.z_path_id}_REQUIRES_ORDS", |
|
fmt_dep_list(node.depends_on)) |
|
|
|
out_comment("Ordinals for what depends directly on this node:") |
|
out_dt_define(f"{node.z_path_id}_SUPPORTS_ORDS", |
|
fmt_dep_list(node.required_by)) |
|
|
|
|
|
def prop2value(prop): |
|
# Gets the macro value for property 'prop', if there is |
|
# a single well-defined C rvalue that it can be represented as. |
|
# Returns None if there isn't one. |
|
|
|
if prop.type == "string": |
|
return quote_str(prop.val) |
|
|
|
if prop.type == "int": |
|
return prop.val |
|
|
|
if prop.type == "boolean": |
|
return 1 if prop.val else 0 |
|
|
|
if prop.type in ["array", "uint8-array"]: |
|
return list2init(f"{val} /* {hex(val)} */" for val in prop.val) |
|
|
|
if prop.type == "string-array": |
|
return list2init(quote_str(val) for val in prop.val) |
|
|
|
# phandle, phandles, phandle-array, path, compound: nothing |
|
return None |
|
|
|
|
|
def prop_len(prop): |
|
# Returns the property's length if and only if we should generate |
|
# a _LEN macro for the property. Otherwise, returns None. |
|
# |
|
# This deliberately excludes ranges, dma-ranges, reg and interrupts. |
|
# While they have array type, their lengths as arrays are |
|
# basically nonsense semantically due to #address-cells and |
|
# #size-cells for "reg", #interrupt-cells for "interrupts" |
|
# and #address-cells, #size-cells and the #address-cells from the |
|
# parent node for "ranges" and "dma-ranges". |
|
# |
|
# We have special purpose macros for the number of register blocks |
|
# / interrupt specifiers. Excluding them from this list means |
|
# DT_PROP_LEN(node_id, ...) fails fast at the devicetree.h layer |
|
# with a build error. This forces users to switch to the right |
|
# macros. |
|
|
|
if prop.type == "phandle": |
|
return 1 |
|
|
|
if (prop.type in ["array", "uint8-array", "string-array", |
|
"phandles", "phandle-array"] and |
|
prop.name not in ["ranges", "dma-ranges", "reg", "interrupts"]): |
|
return len(prop.val) |
|
|
|
return None |
|
|
|
|
|
def phandle_macros(prop, macro): |
|
# Returns a dict of macros for phandle or phandles property 'prop'. |
|
# |
|
# The 'macro' argument is the N_<node-id>_P_<prop-id> bit. |
|
# |
|
# These are currently special because we can't serialize their |
|
# values without using label properties, which we're trying to get |
|
# away from needing in Zephyr. (Label properties are great for |
|
# humans, but have drawbacks for code size and boot time.) |
|
# |
|
# The names look a bit weird to make it easier for devicetree.h |
|
# to use the same macros for phandle, phandles, and phandle-array. |
|
|
|
ret = {} |
|
|
|
if prop.type == "phandle": |
|
# A phandle is treated as a phandles with fixed length 1. |
|
ret[f"{macro}"] = f"DT_{prop.val.z_path_id}" |
|
ret[f"{macro}_IDX_0"] = f"DT_{prop.val.z_path_id}" |
|
ret[f"{macro}_IDX_0_PH"] = f"DT_{prop.val.z_path_id}" |
|
ret[f"{macro}_IDX_0_EXISTS"] = 1 |
|
elif prop.type == "phandles": |
|
for i, node in enumerate(prop.val): |
|
ret[f"{macro}_IDX_{i}"] = f"DT_{node.z_path_id}" |
|
ret[f"{macro}_IDX_{i}_PH"] = f"DT_{node.z_path_id}" |
|
ret[f"{macro}_IDX_{i}_EXISTS"] = 1 |
|
elif prop.type == "phandle-array": |
|
for i, entry in enumerate(prop.val): |
|
if entry is None: |
|
# Unspecified element. The phandle-array at this index |
|
# does not point at a ControllerAndData value, but |
|
# subsequent indices in the array may. |
|
ret[f"{macro}_IDX_{i}_EXISTS"] = 0 |
|
continue |
|
|
|
ret.update(controller_and_data_macros(entry, i, macro)) |
|
|
|
return ret |
|
|
|
|
|
def controller_and_data_macros(entry, i, macro): |
|
# Helper procedure used by phandle_macros(). |
|
# |
|
# Its purpose is to write the "controller" (i.e. label property of |
|
# the phandle's node) and associated data macros for a |
|
# ControllerAndData. |
|
|
|
ret = {} |
|
data = entry.data |
|
|
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_EXISTS |
|
ret[f"{macro}_IDX_{i}_EXISTS"] = 1 |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_PH |
|
ret[f"{macro}_IDX_{i}_PH"] = f"DT_{entry.controller.z_path_id}" |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_VAL_<VAL> |
|
for cell, val in data.items(): |
|
ret[f"{macro}_IDX_{i}_VAL_{str2ident(cell)}"] = val |
|
ret[f"{macro}_IDX_{i}_VAL_{str2ident(cell)}_EXISTS"] = 1 |
|
|
|
if not entry.name: |
|
return ret |
|
|
|
name = str2ident(entry.name) |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_EXISTS |
|
ret[f"{macro}_IDX_{i}_EXISTS"] = 1 |
|
# DT_N_<node-id>_P_<prop-id>_IDX_<i>_NAME |
|
ret[f"{macro}_IDX_{i}_NAME"] = quote_str(entry.name) |
|
# DT_N_<node-id>_P_<prop-id>_NAME_<NAME>_PH |
|
ret[f"{macro}_NAME_{name}_PH"] = f"DT_{entry.controller.z_path_id}" |
|
# DT_N_<node-id>_P_<prop-id>_NAME_<NAME>_EXISTS |
|
ret[f"{macro}_NAME_{name}_EXISTS"] = 1 |
|
# DT_N_<node-id>_P_<prop-id>_NAME_<NAME>_VAL_<VAL> |
|
for cell, val in data.items(): |
|
cell_ident = str2ident(cell) |
|
ret[f"{macro}_NAME_{name}_VAL_{cell_ident}"] = \ |
|
f"DT_{macro}_IDX_{i}_VAL_{cell_ident}" |
|
ret[f"{macro}_NAME_{name}_VAL_{cell_ident}_EXISTS"] = 1 |
|
|
|
return ret |
|
|
|
|
|
def write_chosen(edt): |
|
# Tree-wide information such as chosen nodes is printed here. |
|
|
|
out_comment("Chosen nodes\n") |
|
chosen = {} |
|
for name, node in edt.chosen_nodes.items(): |
|
chosen[f"DT_CHOSEN_{str2ident(name)}"] = f"DT_{node.z_path_id}" |
|
chosen[f"DT_CHOSEN_{str2ident(name)}_EXISTS"] = 1 |
|
max_len = max(map(len, chosen), default=0) |
|
for macro, value in chosen.items(): |
|
out_define(macro, value, width=max_len) |
|
|
|
|
|
def write_global_macros(edt): |
|
# Global or tree-wide information, such as number of instances |
|
# with status "okay" for each compatible, is printed here. |
|
|
|
|
|
out_comment("Macros for iterating over all nodes and enabled nodes") |
|
out_dt_define("FOREACH_HELPER(fn)", |
|
" ".join(f"fn(DT_{node.z_path_id})" for node in edt.nodes)) |
|
out_dt_define("FOREACH_OKAY_HELPER(fn)", |
|
" ".join(f"fn(DT_{node.z_path_id})" for node in edt.nodes |
|
if node.status == "okay")) |
|
|
|
n_okay_macros = {} |
|
for_each_macros = {} |
|
compat2buses = defaultdict(list) # just for "okay" nodes |
|
for compat, okay_nodes in edt.compat2okay.items(): |
|
for node in okay_nodes: |
|
buses = node.on_buses |
|
for bus in buses: |
|
if bus is not None and bus not in compat2buses[compat]: |
|
compat2buses[compat].append(bus) |
|
|
|
ident = str2ident(compat) |
|
n_okay_macros[f"DT_N_INST_{ident}_NUM_OKAY"] = len(okay_nodes) |
|
|
|
# Helpers for non-INST for-each macros that take node |
|
# identifiers as arguments. |
|
for_each_macros[f"DT_FOREACH_OKAY_{ident}(fn)"] = \ |
|
" ".join(f"fn(DT_{node.z_path_id})" |
|
for node in okay_nodes) |
|
for_each_macros[f"DT_FOREACH_OKAY_VARGS_{ident}(fn, ...)"] = \ |
|
" ".join(f"fn(DT_{node.z_path_id}, __VA_ARGS__)" |
|
for node in okay_nodes) |
|
|
|
# Helpers for INST versions of for-each macros, which take |
|
# instance numbers. We emit separate helpers for these because |
|
# avoiding an intermediate node_id --> instance number |
|
# conversion in the preprocessor helps to keep the macro |
|
# expansions simpler. That hopefully eases debugging. |
|
for_each_macros[f"DT_FOREACH_OKAY_INST_{ident}(fn)"] = \ |
|
" ".join(f"fn({edt.compat2nodes[compat].index(node)})" |
|
for node in okay_nodes) |
|
for_each_macros[f"DT_FOREACH_OKAY_INST_VARGS_{ident}(fn, ...)"] = \ |
|
" ".join(f"fn({edt.compat2nodes[compat].index(node)}, __VA_ARGS__)" |
|
for node in okay_nodes) |
|
|
|
for compat, nodes in edt.compat2nodes.items(): |
|
for node in nodes: |
|
if compat == "fixed-partitions": |
|
for child in node.children.values(): |
|
if "label" in child.props: |
|
label = child.props["label"].val |
|
macro = f"COMPAT_{str2ident(compat)}_LABEL_{str2ident(label)}" |
|
val = f"DT_{child.z_path_id}" |
|
|
|
out_dt_define(macro, val) |
|
out_dt_define(macro + "_EXISTS", 1) |
|
|
|
out_comment('Macros for compatibles with status "okay" nodes\n') |
|
for compat, okay_nodes in edt.compat2okay.items(): |
|
if okay_nodes: |
|
out_define(f"DT_COMPAT_HAS_OKAY_{str2ident(compat)}", 1) |
|
|
|
out_comment('Macros for status "okay" instances of each compatible\n') |
|
for macro, value in n_okay_macros.items(): |
|
out_define(macro, value) |
|
for macro, value in for_each_macros.items(): |
|
out_define(macro, value) |
|
|
|
out_comment('Bus information for status "okay" nodes of each compatible\n') |
|
for compat, buses in compat2buses.items(): |
|
for bus in buses: |
|
out_define( |
|
f"DT_COMPAT_{str2ident(compat)}_BUS_{str2ident(bus)}", 1) |
|
|
|
def str2ident(s): |
|
# Converts 's' to a form suitable for (part of) an identifier |
|
|
|
return re.sub('[-,.@/+]', '_', s.lower()) |
|
|
|
|
|
def list2init(l): |
|
# Converts 'l', a Python list (or iterable), to a C array initializer |
|
|
|
return "{" + ", ".join(l) + "}" |
|
|
|
|
|
def out_dt_define(macro, val, width=None, deprecation_msg=None): |
|
# Writes "#define DT_<macro> <val>" to the header file |
|
# |
|
# The macro will be left-justified to 'width' characters if that |
|
# is specified, and the value will follow immediately after in |
|
# that case. Otherwise, this function decides how to add |
|
# whitespace between 'macro' and 'val'. |
|
# |
|
# If a 'deprecation_msg' string is passed, the generated identifiers will |
|
# generate a warning if used, via __WARN(<deprecation_msg>)). |
|
# |
|
# Returns the full generated macro for 'macro', with leading "DT_". |
|
ret = "DT_" + macro |
|
out_define(ret, val, width=width, deprecation_msg=deprecation_msg) |
|
return ret |
|
|
|
|
|
def out_define(macro, val, width=None, deprecation_msg=None): |
|
# Helper for out_dt_define(). Outputs "#define <macro> <val>", |
|
# adds a deprecation message if given, and allocates whitespace |
|
# unless told not to. |
|
|
|
warn = fr' __WARN("{deprecation_msg}")' if deprecation_msg else "" |
|
|
|
if width: |
|
s = f"#define {macro.ljust(width)}{warn} {val}" |
|
else: |
|
s = f"#define {macro}{warn} {val}" |
|
|
|
print(s, file=header_file) |
|
|
|
|
|
def out_comment(s, blank_before=True): |
|
# Writes 's' as a comment to the header and configuration file. 's' is |
|
# allowed to have multiple lines. blank_before=True adds a blank line |
|
# before the comment. |
|
|
|
if blank_before: |
|
print(file=header_file) |
|
|
|
if "\n" in s: |
|
# Format multi-line comments like |
|
# |
|
# /* |
|
# * first line |
|
# * second line |
|
# * |
|
# * empty line before this line |
|
# */ |
|
res = ["/*"] |
|
for line in s.splitlines(): |
|
# Avoid an extra space after '*' for empty lines. They turn red in |
|
# Vim if space error checking is on, which is annoying. |
|
res.append(" *" if not line.strip() else " * " + line) |
|
res.append(" */") |
|
print("\n".join(res), file=header_file) |
|
else: |
|
# Format single-line comments like |
|
# |
|
# /* foo bar */ |
|
print("/* " + s + " */", file=header_file) |
|
|
|
|
|
def escape(s): |
|
# Backslash-escapes any double quotes and backslashes in 's' |
|
|
|
# \ must be escaped before " to avoid double escaping |
|
return s.replace("\\", "\\\\").replace('"', '\\"') |
|
|
|
|
|
def quote_str(s): |
|
# Puts quotes around 's' and escapes any double quotes and |
|
# backslashes within it |
|
|
|
return f'"{escape(s)}"' |
|
|
|
|
|
def write_pickled_edt(edt, out_file): |
|
# Writes the edt object in pickle format to out_file. |
|
|
|
with open(out_file, 'wb') as f: |
|
# Pickle protocol version 4 is the default as of Python 3.8 |
|
# and was introduced in 3.4, so it is both available and |
|
# recommended on all versions of Python that Zephyr supports |
|
# (at time of writing, Python 3.6 was Zephyr's minimum |
|
# version, and 3.8 the most recent CPython release). |
|
# |
|
# Using a common protocol version here will hopefully avoid |
|
# reproducibility issues in different Python installations. |
|
pickle.dump(edt, f, protocol=4) |
|
|
|
|
|
def err(s): |
|
raise Exception(s) |
|
|
|
|
|
if __name__ == "__main__": |
|
main()
|
|
|