freebsd-skq/usr.bin/dtc/fdt.cc

1467 lines
31 KiB
C++

/*-
* Copyright (c) 2013 David Chisnall
* All rights reserved.
*
* This software was developed by SRI International and the University of
* Cambridge Computer Laboratory under DARPA/AFRL contract (FA8750-10-C-0237)
* ("CTSRD"), as part of the DARPA CRASH research programme.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
#define __STDC_LIMIT_MACROS 1
#include "fdt.hh"
#include <algorithm>
#include <ctype.h>
#include <fcntl.h>
#include <inttypes.h>
#include <libgen.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include "dtb.hh"
namespace dtc
{
namespace fdt
{
uint32_t
property_value::get_as_uint32()
{
if (byte_data.size() != 4)
{
return 0;
}
uint32_t v = 0;
v &= byte_data[0] << 24;
v &= byte_data[1] << 16;
v &= byte_data[2] << 8;
v &= byte_data[3] << 0;
return v;
}
void
property_value::push_to_buffer(byte_buffer &buffer)
{
if (!byte_data.empty())
{
buffer.insert(buffer.end(), byte_data.begin(), byte_data.end());
}
else
{
string_data.push_to_buffer(buffer, true);
// Trailing nul
buffer.push_back(0);
}
}
void
property_value::write_dts(FILE *file)
{
resolve_type();
switch (type)
{
default:
assert(0 && "Invalid type");
case STRING:
case STRING_LIST:
case CROSS_REFERENCE:
write_as_string(file);
break;
case PHANDLE:
write_as_cells(file);
break;
case BINARY:
if (byte_data.size() % 4 == 0)
{
write_as_cells(file);
break;
}
write_as_bytes(file);
break;
}
}
void
property_value::resolve_type()
{
if (type != UNKNOWN)
{
return;
}
if (byte_data.empty())
{
type = STRING;
return;
}
if (byte_data.back() == 0)
{
bool is_all_printable = true;
int nuls = 0;
int bytes = 0;
for (byte_buffer::iterator i=byte_data.begin(), e=byte_data.end()-1; i<e ; i++)
{
bytes++;
is_all_printable &= (*i == '\0') || isprint(*i);
if (*i == '\0')
{
nuls++;
}
if (!is_all_printable)
{
break;
}
}
if ((is_all_printable && (bytes > nuls)) || bytes == 0)
{
type = STRING;
if (nuls > 0)
{
type = STRING_LIST;
}
return;
}
}
type = BINARY;
}
void
property_value::write_as_string(FILE *file)
{
putc('"', file);
if (byte_data.empty())
{
string_data.print(file);
}
else
{
for (byte_buffer::iterator i=byte_data.begin(), e=byte_data.end()-1; i!=e ; ++i)
{
// FIXME Escape tabs, newlines, and so on.
if (*i == '\0')
{
fputs("\", \"", file);
continue;
}
putc(*i, file);
}
}
putc('"', file);
}
void
property_value::write_as_cells(FILE *file)
{
putc('<', file);
assert((byte_data.size() % 4) == 0);
for (byte_buffer::iterator i=byte_data.begin(), e=byte_data.end(); i!=e ; ++i)
{
uint32_t v = 0;
v = (v << 8) | *i;
++i;
v = (v << 8) | *i;
++i;
v = (v << 8) | *i;
++i;
v = (v << 8) | *i;
fprintf(file, "0x%" PRIx32, v);
if (i+1 != e)
{
putc(' ', file);
}
}
putc('>', file);
}
void
property_value::write_as_bytes(FILE *file)
{
putc('[', file);
for (byte_buffer::iterator i=byte_data.begin(), e=byte_data.end(); i!=e ; i++)
{
fprintf(file, "%02hhx", *i);
if (i+1 != e)
{
putc(' ', file);
}
}
putc(']', file);
}
void
property::parse_string(input_buffer &input)
{
property_value v;
assert(input[0] == '"');
++input;
const char *start = (const char*)input;
int length = 0;
while (char c = input[0])
{
if (c == '"' && input[-1] != '\\')
{
input.consume('"');
break;
}
++input;
++length;
}
v.string_data = string(start, length);
values.push_back(v);
}
void
property::parse_cells(input_buffer &input)
{
assert(input[0] == '<');
++input;
property_value v;
input.next_token();
while (!input.consume('>'))
{
input.next_token();
// If this is a phandle then we need to get the name of the
// referenced node
if (input.consume('&'))
{
input.next_token();
// FIXME: We should support full paths here, but we
// don't.
string referenced = string::parse_node_name(input);
if (referenced.empty())
{
input.parse_error("Expected node name");
valid = false;
return;
}
input.next_token();
// If we already have some bytes, make the phandle a
// separate component.
if (!v.byte_data.empty())
{
values.push_back(v);
v = property_value();
}
v.string_data = referenced;
v.type = property_value::PHANDLE;
values.push_back(v);
v = property_value();
}
else
{
//FIXME: We should support labels in the middle
//of these, but we don't.
long long val;
if (!input.consume_integer(val))
{
input.parse_error("Expected numbers in array of cells");
valid = false;
return;
}
if ((val < 0) || (val > UINT32_MAX))
{
input.parse_error("Value out of range");
valid = false;
return;
}
push_big_endian(v.byte_data, (uint32_t)val);
input.next_token();
}
}
// Don't store an empty string value here.
if (v.byte_data.size() > 0)
{
values.push_back(v);
}
}
void
property::parse_bytes(input_buffer &input)
{
assert(input[0] == '[');
++input;
property_value v;
input.next_token();
while (!input.consume(']'))
{
{
//FIXME: We should support
//labels in the middle of
//these, but we don't.
uint8_t val;
if (!input.consume_hex_byte(val))
{
input.parse_error("Expected hex bytes in array of bytes");
valid = false;
return;
}
v.byte_data.push_back(val);
input.next_token();
}
}
values.push_back(v);
}
void
property::parse_reference(input_buffer &input)
{
assert(input[0] == '&');
++input;
input.next_token();
property_value v;
v.string_data = string::parse_node_name(input);
if (v.string_data.empty())
{
input.parse_error("Expected node name");
valid = false;
return;
}
v.type = property_value::CROSS_REFERENCE;
values.push_back(v);
}
property::property(input_buffer &structs, input_buffer &strings)
{
uint32_t name_offset;
uint32_t length;
valid = structs.consume_binary(length) &&
structs.consume_binary(name_offset);
if (!valid)
{
fprintf(stderr, "Failed to read property\n");
return;
}
// Find the name
input_buffer name_buffer = strings.buffer_from_offset(name_offset);
if (name_buffer.empty())
{
fprintf(stderr, "Property name offset %" PRIu32
" is past the end of the strings table\n",
name_offset);
valid = false;
return;
}
key = string(name_buffer);
// If we're empty, do not push anything as value.
if (!length)
return;
// Read the value
uint8_t byte;
property_value v;
for (uint32_t i=0 ; i<length ; i++)
{
if (!(valid = structs.consume_binary(byte)))
{
fprintf(stderr, "Failed to read property value\n");
return;
}
v.byte_data.push_back(byte);
}
values.push_back(v);
}
void property::parse_define(input_buffer &input, define_map *defines)
{
input.consume('$');
if (!defines)
{
input.parse_error("No predefined properties to match name\n");
valid = false;
return;
}
string name = string::parse_property_name(input);
define_map::iterator found;
if ((name == string()) ||
((found = defines->find(name)) == defines->end()))
{
input.parse_error("Undefined property name\n");
valid = false;
return;
}
values.push_back((*found).second->values[0]);
}
property::property(input_buffer &input,
string k,
string l,
bool semicolonTerminated,
define_map *defines) : key(k), label(l), valid(true)
{
do {
input.next_token();
switch (input[0])
{
case '$':
{
parse_define(input, defines);
if (valid)
{
break;
}
}
default:
input.parse_error("Invalid property value.");
valid = false;
return;
case '"':
parse_string(input);
break;
case '<':
parse_cells(input);
break;
case '[':
parse_bytes(input);
break;
case '&':
parse_reference(input);
break;
case ';':
{
break;
}
}
input.next_token();
} while (input.consume(','));
if (semicolonTerminated && !input.consume(';'))
{
input.parse_error("Expected ; at end of property");
valid = false;
}
}
property*
property::parse_dtb(input_buffer &structs, input_buffer &strings)
{
property *p = new property(structs, strings);
if (!p->valid)
{
delete p;
p = 0;
}
return p;
}
property*
property::parse(input_buffer &input, string key, string label,
bool semicolonTerminated, define_map *defines)
{
property *p = new property(input, key, label, semicolonTerminated, defines);
if (!p->valid)
{
delete p;
p = 0;
}
return p;
}
void
property::write(dtb::output_writer &writer, dtb::string_table &strings)
{
writer.write_token(dtb::FDT_PROP);
byte_buffer value_buffer;
for (value_iterator i=begin(), e=end() ; i!=e ; ++i)
{
i->push_to_buffer(value_buffer);
}
writer.write_data((uint32_t)value_buffer.size());
writer.write_comment(key);
writer.write_data(strings.add_string(key));
writer.write_data(value_buffer);
}
void
property::write_dts(FILE *file, int indent)
{
for (int i=0 ; i<indent ; i++)
{
putc('\t', file);
}
if (label != string())
{
label.print(file);
fputs(": ", file);
}
if (key != string())
{
key.print(file);
}
if (!values.empty())
{
fputs(" = ", file);
for (value_iterator i=begin(), e=end() ; i!=e ; ++i)
{
i->write_dts(file);
if (i+1 != e)
{
putc(',', file);
putc(' ', file);
}
}
}
fputs(";\n", file);
}
string
node::parse_name(input_buffer &input, bool &is_property, const char *error)
{
if (!valid)
{
return string();
}
input.next_token();
if (is_property)
{
return string::parse_property_name(input);
}
string n = string::parse_node_or_property_name(input, is_property);
if (n.empty())
{
if (n.empty())
{
input.parse_error(error);
valid = false;
}
}
return n;
}
node::node(input_buffer &structs, input_buffer &strings) : valid(true)
{
const char *name_start = (const char*)structs;
int name_length = 0;
while (structs[0] != '\0' && structs[0] != '@')
{
name_length++;
++structs;
}
name = string(name_start, name_length);
if (structs[0] == '@')
{
++structs;
name_start = (const char*)structs;
name_length = 0;
while (structs[0] != '\0')
{
name_length++;
++structs;
}
unit_address = string(name_start, name_length);
}
++structs;
uint32_t token;
while (structs.consume_binary(token))
{
switch (token)
{
default:
fprintf(stderr, "Unexpected token 0x%" PRIx32
" while parsing node.\n", token);
valid = false;
return;
// Child node, parse it.
case dtb::FDT_BEGIN_NODE:
{
node *child = node::parse_dtb(structs, strings);
if (child == 0)
{
valid = false;
return;
}
children.push_back(child);
break;
}
// End of this node, no errors.
case dtb::FDT_END_NODE:
return;
// Property, parse it.
case dtb::FDT_PROP:
{
property *prop = property::parse_dtb(structs, strings);
if (prop == 0)
{
valid = false;
return;
}
properties.push_back(prop);
break;
}
break;
// End of structs table. Should appear after
// the end of the last node.
case dtb::FDT_END:
fprintf(stderr, "Unexpected FDT_END token while parsing node.\n");
valid = false;
return;
// NOPs are padding. Ignore them.
case dtb::FDT_NOP:
break;
}
}
fprintf(stderr, "Failed to read token from structs table while parsing node.\n");
valid = false;
return;
}
node::node(input_buffer &input, string n, string l, string a, define_map *defines) :
label(l), name(n), unit_address(a), valid(true)
{
if (!input.consume('{'))
{
input.parse_error("Expected { to start new device tree node.\n");
}
input.next_token();
while (valid && !input.consume('}'))
{
// flag set if we find any characters that are only in
// the property name character set, not the node
bool is_property = false;
string child_name, child_label, child_address;
child_name = parse_name(input, is_property,
"Expected property or node name");
if (input.consume(':'))
{
// Node labels can contain any characters? The
// spec doesn't say, so we guess so...
is_property = false;
child_label = child_name;
child_name = parse_name(input, is_property, "Expected property or node name");
}
if (input.consume('@'))
{
child_address = parse_name(input, is_property, "Expected unit address");
}
if (!valid)
{
return;
}
input.next_token();
// If we're parsing a property, then we must actually do that.
if (input.consume('='))
{
property *p= property::parse(input, child_name,
child_label, true, defines);
if (p == 0)
{
valid = false;
}
else
{
properties.push_back(p);
}
}
else if (!is_property && input[0] == ('{'))
{
node *child = node::parse(input, child_name,
child_label, child_address, defines);
if (child)
{
children.push_back(child);
}
else
{
valid = false;
}
}
else if (input.consume(';'))
{
properties.push_back(new property(child_name, child_label));
}
else
{
input.parse_error("Error parsing property.");
valid = false;
}
input.next_token();
}
input.consume(';');
}
bool
node::cmp_properties(property *p1, property *p2)
{
return p1->get_key() < p2->get_key();
}
bool
node::cmp_children(node *c1, node *c2)
{
if (c1->name == c2->name)
{
return c1->unit_address < c2->unit_address;
}
return c1->name < c2->name;
}
void
node::sort()
{
std::sort(property_begin(), property_end(), cmp_properties);
std::sort(child_begin(), child_end(), cmp_children);
for (child_iterator i=child_begin(), e=child_end() ; i!=e ; ++i)
{
(*i)->sort();
}
}
node*
node::parse(input_buffer &input,
string name,
string label,
string address,
define_map *defines)
{
node *n = new node(input, name, label, address, defines);
if (!n->valid)
{
delete n;
n = 0;
}
return n;
}
node*
node::parse_dtb(input_buffer &structs, input_buffer &strings)
{
node *n = new node(structs, strings);
if (!n->valid)
{
delete n;
n = 0;
}
return n;
}
node::~node()
{
while (!children.empty())
{
delete children.back();
children.pop_back();
}
while (!properties.empty())
{
delete properties.back();
properties.pop_back();
}
}
property*
node::get_property(string key)
{
for (property_iterator i=property_begin(), e=property_end() ; i!=e ; ++i)
{
if ((*i)->get_key() == key)
{
return *i;
}
}
return 0;
}
void
node::merge_node(node *other)
{
if (!other->label.empty())
{
label = other->label;
}
// Note: this is an O(n*m) operation. It might be sensible to
// optimise this if we find that there are nodes with very
// large numbers of properties, but for typical usage the
// entire vector will fit (easily) into cache, so iterating
// over it repeatedly isn't that expensive.
while (!other->properties.empty())
{
property *p = other->properties.front();
for (property_iterator i=property_begin(), e=property_end() ; i!=e ; ++i)
{
if ((*i)->get_key() == p->get_key())
{
delete *i;
properties.erase(i);
break;
}
}
add_property(p);
other->properties.erase(other->properties.begin());
}
while (!other->children.empty())
{
node *c = other->children.front();
bool found = false;
for (child_iterator i=child_begin(), e=child_end() ; i!=e ; ++i)
{
if ((*i)->name == c->name && (*i)->unit_address == c->unit_address)
{
(*i)->merge_node(c);
delete c;
found = true;
break;
}
}
if (!found)
{
children.push_back(c);
}
other->children.erase(other->children.begin());
}
}
void
node::write(dtb::output_writer &writer, dtb::string_table &strings)
{
writer.write_token(dtb::FDT_BEGIN_NODE);
byte_buffer name_buffer;
name.push_to_buffer(name_buffer);
if (unit_address != string())
{
name_buffer.push_back('@');
unit_address.push_to_buffer(name_buffer);
}
writer.write_comment(name);
writer.write_data(name_buffer);
writer.write_data((uint8_t)0);
for (property_iterator i=property_begin(), e=property_end() ; i!=e ; ++i)
{
(*i)->write(writer, strings);
}
for (child_iterator i=child_begin(), e=child_end() ; i!=e ; ++i)
{
(*i)->write(writer, strings);
}
writer.write_token(dtb::FDT_END_NODE);
}
void
node::write_dts(FILE *file, int indent)
{
for (int i=0 ; i<indent ; i++)
{
putc('\t', file);
}
if (label != string())
{
label.print(file);
fputs(": ", file);
}
if (name != string())
{
name.print(file);
}
if (unit_address != string())
{
putc('@', file);
unit_address.print(file);
}
fputs(" {\n\n", file);
for (property_iterator i=property_begin(), e=property_end() ; i!=e ; ++i)
{
(*i)->write_dts(file, indent+1);
}
for (child_iterator i=child_begin(), e=child_end() ; i!=e ; ++i)
{
(*i)->write_dts(file, indent+1);
}
for (int i=0 ; i<indent ; i++)
{
putc('\t', file);
}
fputs("};\n", file);
}
void
device_tree::collect_names_recursive(node* n, node_path &path)
{
string name = n->label;
path.push_back(std::make_pair(n->name, n->unit_address));
if (name != string())
{
if (node_names.find(name) == node_names.end())
{
node_names.insert(std::make_pair(name, n));
node_paths.insert(std::make_pair(name, path));
}
else
{
node_names[name] = (node*)-1;
std::map<string, node_path>::iterator i = node_paths.find(name);
if (i != node_paths.end())
{
node_paths.erase(name);
}
fprintf(stderr, "Label not unique: ");
name.dump();
fprintf(stderr, ". References to this label will not be resolved.");
}
}
for (node::child_iterator i=n->child_begin(), e=n->child_end() ; i!=e ; ++i)
{
collect_names_recursive(*i, path);
}
path.pop_back();
// Now we collect the phandles and properties that reference
// other nodes.
for (node::property_iterator i=n->property_begin(), e=n->property_end() ; i!=e ; ++i)
{
for (property::value_iterator p=(*i)->begin(),pe=(*i)->end() ; p!=pe ; ++p)
{
if (p->is_phandle())
{
phandles.push_back(&*p);
}
if (p->is_cross_reference())
{
cross_references.push_back(&*p);
}
}
if ((*i)->get_key() == string("phandle") ||
(*i)->get_key() == string("linux,phandle"))
{
if ((*i)->begin()->byte_data.size() != 4)
{
fprintf(stderr, "Invalid phandle value for node ");
n->name.dump();
fprintf(stderr, ". Should be a 4-byte value.\n");
valid = false;
}
else
{
uint32_t phandle = (*i)->begin()->get_as_uint32();
used_phandles.insert(std::make_pair(phandle, n));
}
}
}
}
void
device_tree::collect_names()
{
node_path p;
collect_names_recursive(root, p);
}
void
device_tree::resolve_cross_references()
{
for (std::vector<property_value*>::iterator i=cross_references.begin(), e=cross_references.end() ; i!=e ; ++i)
{
property_value* pv = *i;
node_path path = node_paths[pv->string_data];
// Skip the first name in the path. It's always "", and implicitly /
for (node_path::iterator p=path.begin()+1, pe=path.end() ; p!=pe ; ++p)
{
pv->byte_data.push_back('/');
p->first.push_to_buffer(pv->byte_data);
if (!(p->second.empty()))
{
pv->byte_data.push_back('@');
p->second.push_to_buffer(pv->byte_data);
}
}
pv->byte_data.push_back(0);
}
uint32_t phandle = 1;
for (std::vector<property_value*>::iterator i=phandles.begin(), e=phandles.end() ; i!=e ; ++i)
{
string target_name = (*i)->string_data;
node *target = node_names[target_name];
if (target == 0)
{
fprintf(stderr, "Failed to find node with label:");
target_name.dump();
fprintf(stderr, "\n");
valid = 0;
return;
}
// If there is an existing phandle, use it
property *p = target->get_property("phandle");
if (p == 0)
{
p = target->get_property("linux,phandle");
}
if (p == 0)
{
// Otherwise insert a new phandle node
property_value v;
while (used_phandles.find(phandle) != used_phandles.end())
{
// Note that we only don't need to
// store this phandle in the set,
// because we are monotonically
// increasing the value of phandle and
// so will only ever revisit this value
// if we have used 2^32 phandles, at
// which point our blob won't fit in
// any 32-bit system and we've done
// something badly wrong elsewhere
// already.
phandle++;
}
push_big_endian(v.byte_data, phandle++);
if (phandle_node_name == BOTH || phandle_node_name == LINUX)
{
p = new property(string("linux,phandle"));
p->add_value(v);
target->add_property(p);
}
if (phandle_node_name == BOTH || phandle_node_name == EPAPR)
{
p = new property(string("phandle"));
p->add_value(v);
target->add_property(p);
}
}
p->begin()->push_to_buffer((*i)->byte_data);
assert((*i)->byte_data.size() == 4);
}
}
void
device_tree::parse_roots(input_buffer &input, std::vector<node*> &roots)
{
input.next_token();
while (valid && input.consume('/'))
{
input.next_token();
node *n = node::parse(input, string("", 1), string(), string(), &defines);
if (n)
{
roots.push_back(n);
}
else
{
valid = false;
}
input.next_token();
}
}
input_buffer*
device_tree::buffer_for_file(const char *path)
{
if (string(path) == string("-"))
{
input_buffer *b = new stream_input_buffer();
buffers.push_back(b);
return b;
}
int source = open(path, O_RDONLY);
if (source == -1)
{
fprintf(stderr, "Unable to open file %s\n", path);
return 0;
}
struct stat st;
if (fstat(source, &st) == 0 && S_ISDIR(st.st_mode))
{
fprintf(stderr, "File %s is a directory\n", path);
close(source);
return 0;
}
input_buffer *b = new mmap_input_buffer(source);
// Keep the buffer that owns the memory around for the lifetime
// of this FDT. Ones simply referring to it may have shorter
// lifetimes.
buffers.push_back(b);
close(source);
return b;
}
template<class writer> void
device_tree::write(int fd)
{
dtb::string_table st;
dtb::header head;
writer head_writer;
writer reservation_writer;
writer struct_writer;
writer strings_writer;
// Build the reservation table
reservation_writer.write_comment(string("Memory reservations"));
reservation_writer.write_label(string("dt_reserve_map"));
for (std::vector<reservation>::iterator i=reservations.begin(),
e=reservations.end() ; i!=e ; ++i)
{
reservation_writer.write_comment(string("Reservation start"));
reservation_writer.write_data(i->first);
reservation_writer.write_comment(string("Reservation length"));
reservation_writer.write_data(i->first);
}
// Write n spare reserve map entries, plus the trailing 0.
for (uint32_t i=0 ; i<=spare_reserve_map_entries ; i++)
{
reservation_writer.write_data((uint64_t)0);
reservation_writer.write_data((uint64_t)0);
}
struct_writer.write_comment(string("Device tree"));
struct_writer.write_label(string("dt_struct_start"));
root->write(struct_writer, st);
struct_writer.write_token(dtb::FDT_END);
struct_writer.write_label(string("dt_struct_end"));
st.write(strings_writer);
// Find the strings size before we stick padding on the end.
// Note: We should possibly use a new writer for the padding.
head.size_dt_strings = strings_writer.size();
// Stick the padding in the strings writer, but after the
// marker indicating that it's the end.
// Note: We probably should add a padding call to the writer so
// that the asm back end can write padding directives instead
// of a load of 0 bytes.
for (uint32_t i=0 ; i<blob_padding ; i++)
{
strings_writer.write_data((uint8_t)0);
}
head.totalsize = sizeof(head) + strings_writer.size() +
struct_writer.size() + reservation_writer.size();
while (head.totalsize < minimum_blob_size)
{
head.totalsize++;
strings_writer.write_data((uint8_t)0);
}
head.off_dt_struct = sizeof(head) + reservation_writer.size();;
head.off_dt_strings = head.off_dt_struct + struct_writer.size();
head.off_mem_rsvmap = sizeof(head);
head.boot_cpuid_phys = boot_cpu;
head.size_dt_struct = struct_writer.size();
head.write(head_writer);
head_writer.write_to_file(fd);
reservation_writer.write_to_file(fd);
struct_writer.write_to_file(fd);
strings_writer.write_label(string("dt_blob_end"));
strings_writer.write_to_file(fd);
}
node*
device_tree::referenced_node(property_value &v)
{
if (v.is_phandle())
{
return node_names[v.string_data];
}
if (v.is_binary())
{
return used_phandles[v.get_as_uint32()];
}
return 0;
}
void
device_tree::write_binary(int fd)
{
write<dtb::binary_writer>(fd);
}
void
device_tree::write_asm(int fd)
{
write<dtb::asm_writer>(fd);
}
void
device_tree::write_dts(int fd)
{
FILE *file = fdopen(fd, "w");
fputs("/dts-v1/;\n\n", file);
if (!reservations.empty())
{
const char msg[] = "/memreserve/";
fwrite(msg, sizeof(msg), 1, file);
for (std::vector<reservation>::iterator i=reservations.begin(),
e=reservations.end() ; i!=e ; ++i)
{
fprintf(file, " %" PRIx64 " %" PRIx64, i->first, i->second);
}
fputs(";\n\n", file);
}
putc('/', file);
putc(' ', file);
root->write_dts(file, 0);
fclose(file);
}
void
device_tree::parse_dtb(const char *fn, FILE *depfile)
{
input_buffer *in = buffer_for_file(fn);
if (in == 0)
{
valid = false;
return;
}
input_buffer &input = *in;
dtb::header h;
valid = h.read_dtb(input);
boot_cpu = h.boot_cpuid_phys;
if (h.last_comp_version > 17)
{
fprintf(stderr, "Don't know how to read this version of the device tree blob");
valid = false;
}
if (!valid)
{
return;
}
input_buffer reservation_map =
input.buffer_from_offset(h.off_mem_rsvmap, 0);
uint64_t start, length;
do
{
if (!(reservation_map.consume_binary(start) &&
reservation_map.consume_binary(length)))
{
fprintf(stderr, "Failed to read memory reservation table\n");
valid = false;
return;
}
} while (!((start == 0) && (length == 0)));
input_buffer struct_table =
input.buffer_from_offset(h.off_dt_struct, h.size_dt_struct);
input_buffer strings_table =
input.buffer_from_offset(h.off_dt_strings, h.size_dt_strings);
uint32_t token;
if (!(struct_table.consume_binary(token) &&
(token == dtb::FDT_BEGIN_NODE)))
{
fprintf(stderr, "Expected FDT_BEGIN_NODE token.\n");
valid = false;
return;
}
root = node::parse_dtb(struct_table, strings_table);
if (!(struct_table.consume_binary(token) && (token == dtb::FDT_END)))
{
fprintf(stderr, "Expected FDT_END token after parsing root node.\n");
valid = false;
return;
}
valid = (root != 0);
}
void
device_tree::parse_dts(const char *fn, FILE *depfile)
{
input_buffer *in = buffer_for_file(fn);
if (in == 0)
{
valid = false;
return;
}
std::vector<node*> roots;
input_buffer &input = *in;
input.next_token();
bool read_header = false;
// Read the header
if (input.consume("/dts-v1/;"))
{
read_header = true;
}
input.next_token();
while(input.consume("/include/"))
{
bool reallyInclude = true;
if (input.consume("if "))
{
input.next_token();
string name = string::parse_property_name(input);
// XXX: Error handling
if (defines.find(name) == defines.end())
{
reallyInclude = false;
}
input.consume('/');
}
input.next_token();
if (!input.consume('"'))
{
input.parse_error("Expected quoted filename");
valid = false;
return;
}
int length = 0;
while (input[length] != '"') length++;
const char *file = (const char*)input;
const char *dir = dirname(fn);
int dir_length = strlen(dir);
char *include_file = (char*)malloc(strlen(dir) + length + 2);
memcpy(include_file, dir, dir_length);
include_file[dir_length] = '/';
memcpy(include_file+dir_length+1, file, length);
include_file[dir_length+length+1] = 0;
input.consume(include_file+dir_length+1);
input.consume('"');
if (!reallyInclude)
{
continue;
}
input_buffer *include_buffer = buffer_for_file(include_file);
if (include_buffer == 0)
{
for (std::vector<const char*>::iterator i=include_paths.begin(), e=include_paths.end() ; e!=i ; ++i)
{
free(include_file);
dir = *i;
dir_length = strlen(dir);
include_file = (char*)malloc(strlen(dir) +
length + 2);
memcpy(include_file, dir, dir_length);
include_file[dir_length] = '/';
memcpy(include_file+dir_length+1, file, length);
include_file[dir_length+length+1] = 0;
include_buffer = buffer_for_file(include_file);
if (include_buffer != 0)
{
break;
}
}
}
if (depfile != 0)
{
putc(' ', depfile);
fputs(include_file, depfile);
}
if (include_buffer == 0)
{
valid = false;
return;
}
input_buffer &include = *include_buffer;
free((void*)include_file);
if (!read_header)
{
include.next_token();
read_header = include.consume("/dts-v1/;");
}
parse_roots(include, roots);
}
input.next_token();
if (!read_header)
{
input.parse_error("Expected /dts-v1/; version string");
}
// Read any memory reservations
while(input.consume("/memreserve/"))
{
long long start, len;
input.next_token();
// Read the start and length.
if (!(input.consume_integer(start) &&
(input.next_token(),
input.consume_integer(len))))
{
input.parse_error("Expected size on /memreserve/ node.");
}
input.next_token();
input.consume(';');
reservations.push_back(reservation(start, len));
}
parse_roots(input, roots);
switch (roots.size())
{
case 0:
valid = false;
input.parse_error("Failed to find root node /.");
return;
case 1:
root = roots[0];
break;
default:
{
root = roots[0];
for (std::vector<node*>::iterator i=roots.begin()+1,
e=roots.end() ; i!=e ; ++i)
{
root->merge_node(*i);
delete *i;
}
roots.resize(1);
}
}
collect_names();
resolve_cross_references();
}
device_tree::~device_tree()
{
if (root != 0)
{
delete root;
}
while (!buffers.empty())
{
delete buffers.back();
buffers.pop_back();
}
for (define_map::iterator i=defines.begin(), e=defines.end() ;
i!=e ; ++i)
{
delete i->second;
}
}
bool device_tree::parse_define(const char *def)
{
char *val = strchr(def, '=');
if (!val)
{
if (strlen(def) != 0)
{
string name(def);
defines[name];
return true;
}
return false;
}
string name(def, val-def);
val++;
input_buffer in = input_buffer(val, strlen(val));
property *p = property::parse(in, name, string(), false);
if (p)
defines[name] = p;
return p;
}
} // namespace fdt
} // namespace dtc