M-Labs
/
nac3
8
0
Fork 5
Code Issues 87 Pull Requests 3 Releases Wiki Activity

Replace ld.lld with nac3ld for non-host targets #292

Merged
sb10q merged 4 commits from occheung/nac3:nac3ld into master 2022-06-06 17:13:25 +08:00
Conversation 4 Commits 4 Files Changed 8
8 changed files with 5113 additions and 241 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Cargo.lock generated
View File

@ -489,6 +489,7 @@ version = "0.1.0"
dependencies = [
"inkwell",
"nac3core",
"nac3ld",
"nac3parser",
"parking_lot 0.11.2",
"pyo3",
@ -521,6 +522,13 @@ dependencies = [
"test-case",
]
[[package]]
name = "nac3ld"
version = "0.1.0"
dependencies = [
"byteorder",
]
[[package]]
name = "nac3parser"
version = "0.1.2"

1
Cargo.toml
View File

@ -1,5 +1,6 @@
[workspace]
members = [
"nac3ld",
"nac3ast",
"nac3parser",
"nac3core",

1
nac3artiq/Cargo.toml
View File

@ -14,6 +14,7 @@ parking_lot = "0.11"
tempfile = "3"
nac3parser = { path = "../nac3parser" }
nac3core = { path = "../nac3core" }
nac3ld = { path = "../nac3ld" }
[dependencies.inkwell]
git = "https://github.com/TheDan64/inkwell.git"

561
nac3artiq/src/lib.rs
View File

@ -1,12 +1,13 @@
use std::collections::{HashMap, HashSet};
use std::fs;
use std::io::Write;
use std::process::Command;
use std::rc::Rc;
use std::sync::Arc;
use inkwell::{
memory_buffer::MemoryBuffer,
module::Linkage,
module::{Linkage, Module},
passes::{PassManager, PassManagerBuilder},
targets::*,
OptimizationLevel,
@ -36,6 +37,8 @@ use nac3core::{
typecheck::{type_inferencer::PrimitiveStore, typedef::Type},
};
use nac3ld::Linker;
use tempfile::{self, TempDir};
use crate::codegen::attributes_writeback;
@ -263,213 +266,16 @@ impl Nac3 {
}
None
}
}
fn add_exceptions(
composer: &mut TopLevelComposer,
builtin_def: &mut HashMap<StrRef, DefinitionId>,
builtin_ty: &mut HashMap<StrRef, Type>,
error_names: &[&str]
) -> Vec<Type> {
let mut types = Vec::new();
// note: this is only for builtin exceptions, i.e. the exception name is "0:{exn}"
for name in error_names {
let def_id = composer.definition_ast_list.len();
let (exception_fn, exception_class, exception_cons, exception_type) = get_exn_constructor(
name,
// class id
def_id,
// constructor id
def_id + 1,
&mut composer.unifier,
&composer.primitives_ty
);
composer.definition_ast_list.push((Arc::new(RwLock::new(exception_class)), None));
composer.definition_ast_list.push((Arc::new(RwLock::new(exception_fn)), None));
builtin_ty.insert((*name).into(), exception_cons);
builtin_def.insert((*name).into(), DefinitionId(def_id));
types.push(exception_type);
}
types
}
#[pymethods]
impl Nac3 {
#[new]
fn new(isa: &str, py: Python) -> PyResult<Self> {
let isa = match isa {
"host" => Isa::Host,
"rv32g" => Isa::RiscV32G,
"rv32ima" => Isa::RiscV32IMA,
"cortexa9" => Isa::CortexA9,
_ => return Err(exceptions::PyValueError::new_err("invalid ISA")),
};
let time_fns: &(dyn TimeFns + Sync) = match isa {
Isa::Host => &timeline::EXTERN_TIME_FNS,
Isa::RiscV32G => &timeline::NOW_PINNING_TIME_FNS_64,
Isa::RiscV32IMA => &timeline::NOW_PINNING_TIME_FNS,
Isa::CortexA9 => &timeline::EXTERN_TIME_FNS,
};
let primitive: PrimitiveStore = TopLevelComposer::make_primitives().0;
let builtins = vec![
(
"now_mu".into(),
FunSignature { args: vec![], ret: primitive.int64, vars: HashMap::new() },
Arc::new(GenCall::new(Box::new(move |ctx, _, _, _, _| {
Ok(Some(time_fns.emit_now_mu(ctx)))
}))),
),
(
"at_mu".into(),
FunSignature {
args: vec![FuncArg {
name: "t".into(),
ty: primitive.int64,
default_value: None,
}],
ret: primitive.none,
vars: HashMap::new(),
},
Arc::new(GenCall::new(Box::new(move |ctx, _, fun, args, generator| {
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty).unwrap();
time_fns.emit_at_mu(ctx, arg);
Ok(None)
}))),
),
(
"delay_mu".into(),
FunSignature {
args: vec![FuncArg {
name: "dt".into(),
ty: primitive.int64,
default_value: None,
}],
ret: primitive.none,
vars: HashMap::new(),
},
Arc::new(GenCall::new(Box::new(move |ctx, _, fun, args, generator| {
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty).unwrap();
time_fns.emit_delay_mu(ctx, arg);
Ok(None)
}))),
),
];
let builtins_mod = PyModule::import(py, "builtins").unwrap();
let id_fn = builtins_mod.getattr("id").unwrap();
let numpy_mod = PyModule::import(py, "numpy").unwrap();
let typing_mod = PyModule::import(py, "typing").unwrap();
let types_mod = PyModule::import(py, "types").unwrap();
let get_id = |x| id_fn.call1((x,)).unwrap().extract().unwrap();
let get_attr_id = |obj: &PyModule, attr| id_fn.call1((obj.getattr(attr).unwrap(),))
.unwrap().extract().unwrap();
let primitive_ids = PrimitivePythonId {
virtual_id: get_id(
builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("virtual")
.unwrap(
)),
generic_alias: (
get_attr_id(typing_mod, "_GenericAlias"),
get_attr_id(types_mod, "GenericAlias"),
),
none: id_fn
.call1((builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("none")
.unwrap(),))
.unwrap()
.extract()
.unwrap(),
typevar: get_attr_id(typing_mod, "TypeVar"),
int: get_attr_id(builtins_mod, "int"),
int32: get_attr_id(numpy_mod, "int32"),
int64: get_attr_id(numpy_mod, "int64"),
uint32: get_attr_id(numpy_mod, "uint32"),
uint64: get_attr_id(numpy_mod, "uint64"),
bool: get_attr_id(builtins_mod, "bool"),
float: get_attr_id(builtins_mod, "float"),
float64: get_attr_id(numpy_mod, "float64"),
list: get_attr_id(builtins_mod, "list"),
tuple: get_attr_id(builtins_mod, "tuple"),
exception: get_attr_id(builtins_mod, "Exception"),
option: id_fn
.call1((builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("Option")
.unwrap(),))
.unwrap()
.extract()
.unwrap(),
};
let working_directory = tempfile::Builder::new().prefix("nac3-").tempdir().unwrap();
fs::write(working_directory.path().join("kernel.ld"), include_bytes!("kernel.ld")).unwrap();
Ok(Nac3 {
isa,
time_fns,
primitive,
builtins,
primitive_ids,
top_levels: Default::default(),
pyid_to_def: Default::default(),
working_directory,
string_store: Default::default(),
exception_ids: Default::default(),
deferred_eval_store: DeferredEvaluationStore::new(),
})
}
fn analyze(&mut self, functions: &PySet, classes: &PySet) -> PyResult<()> {
let (modules, class_ids) =
Python::with_gil(|py| -> PyResult<(HashMap<u64, PyObject>, HashSet<u64>)> {
let mut modules: HashMap<u64, PyObject> = HashMap::new();
let mut class_ids: HashSet<u64> = HashSet::new();
let id_fn = PyModule::import(py, "builtins")?.getattr("id")?;
let getmodule_fn = PyModule::import(py, "inspect")?.getattr("getmodule")?;
for function in functions.iter() {
let module = getmodule_fn.call1((function,))?.extract()?;
modules.insert(id_fn.call1((&module,))?.extract()?, module);
}
for class in classes.iter() {
let module = getmodule_fn.call1((class,))?.extract()?;
modules.insert(id_fn.call1((&module,))?.extract()?, module);
class_ids.insert(id_fn.call1((class,))?.extract()?);
}
Ok((modules, class_ids))
})?;
for module in modules.into_values() {
self.register_module(module, &class_ids)?;
}
Ok(())
}
fn compile_method_to_file(
&mut self,
fn compile_method<T>(
&self,
obj: &PyAny,
method_name: &str,
args: Vec<&PyAny>,
filename: &str,
embedding_map: &PyAny,
py: Python,
) -> PyResult<()> {
link_fn: &dyn Fn(&Module) -> PyResult<T>,
) -> PyResult<T> {
let (mut composer, mut builtins_def, mut builtins_ty) = TopLevelComposer::new(
self.builtins.clone(),
ComposerConfig { kernel_ann: Some("Kernel"), kernel_invariant_ann: "KernelInvariant" },
@ -756,8 +562,6 @@ impl Nac3 {
calls: Arc::new(Default::default()),
id: 0,
};
let isa = self.isa;
let working_directory = self.working_directory.path().to_owned();
let membuffers: Arc<Mutex<Vec<Vec<u8>>>> = Default::default();
@ -843,7 +647,13 @@ impl Nac3 {
builder.populate_module_pass_manager(&passes);
passes.run_on(&main);
let (triple, features) = match isa {
link_fn(&main)
}
fn get_llvm_target_machine(
&self,
) -> TargetMachine {
let (triple, features) = match self.isa {
Isa::Host => (
TargetMachine::get_default_triple(),
TargetMachine::get_host_cpu_features().to_string(),
@ -854,12 +664,12 @@ impl Nac3 {
Isa::RiscV32IMA => (TargetTriple::create("riscv32-unknown-linux"), "+a,+m".to_string()),
Isa::CortexA9 => (
TargetTriple::create("armv7-unknown-linux-gnueabihf"),
"+dsp,+fp16,+neon,+vfp3".to_string(),
"+dsp,+fp16,+neon,+vfp3,+long-calls".to_string(),
),
};
let target =
Target::from_triple(&triple).expect("couldn't create target from target triple");
let target_machine = target
target
.create_target_machine(
&triple,
"",
@ -868,43 +678,282 @@ impl Nac3 {
RelocMode::PIC,
CodeModel::Default,
)
.expect("couldn't create target machine");
target_machine
.write_to_file(&main, FileType::Object, &working_directory.join("module.o"))
.expect("couldn't write module to file");
.expect("couldn't create target machine")
}
}
let mut linker_args = vec![
"-shared".to_string(),
"--eh-frame-hdr".to_string(),
"-x".to_string(),
"-o".to_string(),
filename.to_string(),
working_directory.join("module.o").to_string_lossy().to_string(),
fn link_with_lld(
elf_filename: String,
obj_filename: String,
) -> PyResult<()>{
let linker_args = vec![
"-shared".to_string(),
"--eh-frame-hdr".to_string(),
"-x".to_string(),
"-o".to_string(),
elf_filename,
obj_filename,
];
#[cfg(not(windows))]
let lld_command = "ld.lld";
#[cfg(windows)]
let lld_command = "ld.lld.exe";
if let Ok(linker_status) = Command::new(lld_command).args(linker_args).status() {
if !linker_status.success() {
return Err(CompileError::new_err("failed to start linker"));
}
} else {
return Err(CompileError::new_err(
"linker returned non-zero status code",
));
}
Ok(())
}
fn add_exceptions(
composer: &mut TopLevelComposer,
builtin_def: &mut HashMap<StrRef, DefinitionId>,
builtin_ty: &mut HashMap<StrRef, Type>,
error_names: &[&str]
) -> Vec<Type> {
let mut types = Vec::new();
// note: this is only for builtin exceptions, i.e. the exception name is "0:{exn}"
for name in error_names {
let def_id = composer.definition_ast_list.len();
let (exception_fn, exception_class, exception_cons, exception_type) = get_exn_constructor(
name,
// class id
def_id,
// constructor id
def_id + 1,
&mut composer.unifier,
&composer.primitives_ty
);
composer.definition_ast_list.push((Arc::new(RwLock::new(exception_class)), None));
composer.definition_ast_list.push((Arc::new(RwLock::new(exception_fn)), None));
builtin_ty.insert((*name).into(), exception_cons);
builtin_def.insert((*name).into(), DefinitionId(def_id));
types.push(exception_type);
}
types
}
#[pymethods]
impl Nac3 {
#[new]
fn new(isa: &str, py: Python) -> PyResult<Self> {
let isa = match isa {
"host" => Isa::Host,
"rv32g" => Isa::RiscV32G,
"rv32ima" => Isa::RiscV32IMA,
"cortexa9" => Isa::CortexA9,
_ => return Err(exceptions::PyValueError::new_err("invalid ISA")),
};
let time_fns: &(dyn TimeFns + Sync) = match isa {
Isa::Host => &timeline::EXTERN_TIME_FNS,
Isa::RiscV32G => &timeline::NOW_PINNING_TIME_FNS_64,
Isa::RiscV32IMA => &timeline::NOW_PINNING_TIME_FNS,
Isa::CortexA9 => &timeline::EXTERN_TIME_FNS,
};
let primitive: PrimitiveStore = TopLevelComposer::make_primitives().0;
let builtins = vec![
(
"now_mu".into(),
FunSignature { args: vec![], ret: primitive.int64, vars: HashMap::new() },
Arc::new(GenCall::new(Box::new(move |ctx, _, _, _, _| {
Ok(Some(time_fns.emit_now_mu(ctx)))
}))),
),
(
"at_mu".into(),
FunSignature {
args: vec![FuncArg {
name: "t".into(),
ty: primitive.int64,
default_value: None,
}],
ret: primitive.none,
vars: HashMap::new(),
},
Arc::new(GenCall::new(Box::new(move |ctx, _, fun, args, generator| {
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty).unwrap();
time_fns.emit_at_mu(ctx, arg);
Ok(None)
}))),
),
(
"delay_mu".into(),
FunSignature {
args: vec![FuncArg {
name: "dt".into(),
ty: primitive.int64,
default_value: None,
}],
ret: primitive.none,
vars: HashMap::new(),
},
Arc::new(GenCall::new(Box::new(move |ctx, _, fun, args, generator| {
let arg_ty = fun.0.args[0].ty;
let arg = args[0].1.clone().to_basic_value_enum(ctx, generator, arg_ty).unwrap();
time_fns.emit_delay_mu(ctx, arg);
Ok(None)
}))),
),
];
if isa != Isa::Host {
linker_args.push(
"-T".to_string()
+ self.working_directory.path().join("kernel.ld").to_str().unwrap(),
);
}
#[cfg(not(windows))]
let lld_command = "ld.lld";
#[cfg(windows)]
let lld_command = "ld.lld.exe";
if let Ok(linker_status) = Command::new(lld_command).args(linker_args).status() {
if !linker_status.success() {
return Err(CompileError::new_err("failed to start linker"));
}
} else {
return Err(CompileError::new_err(
"linker returned non-zero status code",
));
}
let builtins_mod = PyModule::import(py, "builtins").unwrap();
let id_fn = builtins_mod.getattr("id").unwrap();
let numpy_mod = PyModule::import(py, "numpy").unwrap();
let typing_mod = PyModule::import(py, "typing").unwrap();
let types_mod = PyModule::import(py, "types").unwrap();
let get_id = |x| id_fn.call1((x,)).unwrap().extract().unwrap();
let get_attr_id = |obj: &PyModule, attr| id_fn.call1((obj.getattr(attr).unwrap(),))
.unwrap().extract().unwrap();
let primitive_ids = PrimitivePythonId {
virtual_id: get_id(
builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("virtual")
.unwrap(
)),
generic_alias: (
get_attr_id(typing_mod, "_GenericAlias"),
get_attr_id(types_mod, "GenericAlias"),
),
none: id_fn
.call1((builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("none")
.unwrap(),))
.unwrap()
.extract()
.unwrap(),
typevar: get_attr_id(typing_mod, "TypeVar"),
int: get_attr_id(builtins_mod, "int"),
int32: get_attr_id(numpy_mod, "int32"),
int64: get_attr_id(numpy_mod, "int64"),
uint32: get_attr_id(numpy_mod, "uint32"),
uint64: get_attr_id(numpy_mod, "uint64"),
bool: get_attr_id(builtins_mod, "bool"),
float: get_attr_id(builtins_mod, "float"),
float64: get_attr_id(numpy_mod, "float64"),
list: get_attr_id(builtins_mod, "list"),
tuple: get_attr_id(builtins_mod, "tuple"),
exception: get_attr_id(builtins_mod, "Exception"),
option: id_fn
.call1((builtins_mod
.getattr("globals")
.unwrap()
.call0()
.unwrap()
.get_item("Option")
.unwrap(),))
.unwrap()
.extract()
.unwrap(),
};
let working_directory = tempfile::Builder::new().prefix("nac3-").tempdir().unwrap();
fs::write(working_directory.path().join("kernel.ld"), include_bytes!("kernel.ld")).unwrap();
Ok(Nac3 {
isa,
time_fns,
primitive,
builtins,
primitive_ids,
top_levels: Default::default(),
pyid_to_def: Default::default(),
working_directory,
string_store: Default::default(),
exception_ids: Default::default(),
deferred_eval_store: DeferredEvaluationStore::new(),
})
}
fn analyze(&mut self, functions: &PySet, classes: &PySet) -> PyResult<()> {
let (modules, class_ids) =
Python::with_gil(|py| -> PyResult<(HashMap<u64, PyObject>, HashSet<u64>)> {
let mut modules: HashMap<u64, PyObject> = HashMap::new();
let mut class_ids: HashSet<u64> = HashSet::new();
let id_fn = PyModule::import(py, "builtins")?.getattr("id")?;
let getmodule_fn = PyModule::import(py, "inspect")?.getattr("getmodule")?;
for function in functions.iter() {
let module = getmodule_fn.call1((function,))?.extract()?;
modules.insert(id_fn.call1((&module,))?.extract()?, module);
}
for class in classes.iter() {
let module = getmodule_fn.call1((class,))?.extract()?;
modules.insert(id_fn.call1((&module,))?.extract()?, module);
class_ids.insert(id_fn.call1((class,))?.extract()?);
}
Ok((modules, class_ids))
})?;
for module in modules.into_values() {
self.register_module(module, &class_ids)?;
}
Ok(())
}
fn compile_method_to_file(
&mut self,
obj: &PyAny,
method_name: &str,
args: Vec<&PyAny>,
filename: &str,
embedding_map: &PyAny,
py: Python,
) -> PyResult<()> {
let target_machine = self.get_llvm_target_machine();
if self.isa == Isa::Host {
let link_fn = |module: &Module| {
let working_directory = self.working_directory.path().to_owned();
target_machine
.write_to_file(module, FileType::Object, &working_directory.join("module.o"))
.expect("couldn't write module to file");
link_with_lld(
filename.to_string(),
working_directory.join("module.o").to_string_lossy().to_string()
)?;
Ok(())
};
self.compile_method(obj, method_name, args, embedding_map, py, &link_fn)
} else {
let link_fn = |module: &Module| {
let object_mem = target_machine
.write_to_memory_buffer(module, FileType::Object)
.expect("couldn't write module to object file buffer");
if let Ok(dyn_lib) = Linker::ld(object_mem.as_slice()) {
if let Ok(mut file) = fs::File::create(filename) {
file.write_all(&dyn_lib).expect("couldn't write linked library to file");
Ok(())
} else {
Err(CompileError::new_err("failed to create file"))
}
} else {
Err(CompileError::new_err("linker failed to process object file"))
}
};
self.compile_method(obj, method_name, args, embedding_map, py, &link_fn)
}
}
fn compile_method_to_mem(
&mut self,
obj: &PyAny,
@ -913,10 +962,40 @@ impl Nac3 {
embedding_map: &PyAny,
py: Python,
) -> PyResult<PyObject> {
let filename_path = self.working_directory.path().join("module.elf");
let filename = filename_path.to_str().unwrap();
self.compile_method_to_file(obj, method_name, args, filename, embedding_map, py)?;
Ok(PyBytes::new(py, &fs::read(filename).unwrap()).into())
let target_machine = self.get_llvm_target_machine();
if self.isa == Isa::Host {
let link_fn = |module: &Module| {
let working_directory = self.working_directory.path().to_owned();
target_machine
.write_to_file(&module, FileType::Object, &working_directory.join("module.o"))
.expect("couldn't write module to file");
let filename_path = self.working_directory.path().join("module.elf");
let filename = filename_path.to_str().unwrap();
link_with_lld(
filename.to_string(),
working_directory.join("module.o").to_string_lossy().to_string()
)?;
Ok(PyBytes::new(py, &fs::read(filename).unwrap()).into())
};
self.compile_method(obj, method_name, args, embedding_map, py, &link_fn)
} else {
let link_fn = |module: &Module| {
let object_mem = target_machine
.write_to_memory_buffer(&module, FileType::Object)
.expect("couldn't write module to object file buffer");
if let Ok(dyn_lib) = Linker::ld(object_mem.as_slice()) {
Ok(PyBytes::new(py, &dyn_lib).into())
} else {
Err(CompileError::new_err("linker failed to process object file"))
}
};
self.compile_method(obj, method_name, args, embedding_map, py, &link_fn)
}
}
}

8
nac3ld/Cargo.toml Normal file
View File

@ -0,0 +1,8 @@
[package]
name = "nac3ld"
version = "0.1.0"
authors = ["M-Labs"]
edition = "2018"
[dependencies]
byteorder = { version = "1.0", default-features = false }

392
nac3ld/src/dwarf.rs Normal file
View File

@ -0,0 +1,392 @@
#![allow(non_camel_case_types, non_upper_case_globals)]
use std::mem;
use byteorder::{ByteOrder, LittleEndian};
pub const DW_EH_PE_omit: u8 = 0xFF;
pub const DW_EH_PE_absptr: u8 = 0x00;
pub const DW_EH_PE_uleb128: u8 = 0x01;
pub const DW_EH_PE_udata2: u8 = 0x02;
pub const DW_EH_PE_udata4: u8 = 0x03;
pub const DW_EH_PE_udata8: u8 = 0x04;
pub const DW_EH_PE_sleb128: u8 = 0x09;
pub const DW_EH_PE_sdata2: u8 = 0x0A;
pub const DW_EH_PE_sdata4: u8 = 0x0B;
pub const DW_EH_PE_sdata8: u8 = 0x0C;
pub const DW_EH_PE_pcrel: u8 = 0x10;
pub const DW_EH_PE_textrel: u8 = 0x20;
pub const DW_EH_PE_datarel: u8 = 0x30;
pub const DW_EH_PE_funcrel: u8 = 0x40;
pub const DW_EH_PE_aligned: u8 = 0x50;
pub const DW_EH_PE_indirect: u8 = 0x80;
pub struct DwarfReader<'a> {
pub slice: &'a [u8],
pub virt_addr: u32,
}
impl<'a> DwarfReader<'a> {
pub fn new(slice: &[u8], virt_addr: u32) -> DwarfReader {
DwarfReader { slice, virt_addr }
}
pub fn offset(&mut self, offset: i32) {
self.slice = &self.slice[offset as usize..];
self.virt_addr = self.virt_addr.wrapping_add(offset as u32);
}
// ULEB128 and SLEB128 encodings are defined in Section 7.6 - "Variable
// Length Data".
pub fn read_uleb128(&mut self) -> u64 {
let mut shift: usize = 0;
let mut result: u64 = 0;
let mut byte: u8;
loop {
byte = self.read_u8();
result |= ((byte & 0x7F) as u64) << shift;
shift += 7;
if byte & 0x80 == 0 {
break;
}
}
result
}
pub fn read_sleb128(&mut self) -> i64 {
let mut shift: u32 = 0;
let mut result: u64 = 0;
let mut byte: u8;
loop {
byte = self.read_u8();
result |= ((byte & 0x7F) as u64) << shift;
shift += 7;
if byte & 0x80 == 0 {
break;
}
}
// sign-extend
if shift < u64::BITS && (byte & 0x40) != 0 {
result |= (!0 as u64) << shift;
}
result as i64
}
pub fn read_u8(&mut self) -> u8 {
let val = self.slice[0];
self.slice = &self.slice[1..];
val
}
}
macro_rules! impl_read_fn {
( $($type: ty, $byteorder_fn: ident);* ) => {
impl<'a> DwarfReader<'a> {
$(
pub fn $byteorder_fn(&mut self) -> $type {
let val = LittleEndian::$byteorder_fn(self.slice);
self.slice = &self.slice[mem::size_of::<$type>()..];
val
}
)*
}
}
}
impl_read_fn!(
u16, read_u16;
u32, read_u32;
u64, read_u64;
i16, read_i16;
i32, read_i32;
i64, read_i64
);
pub struct DwarfWriter<'a> {
pub slice: &'a mut [u8],
pub offset: usize,
}
impl<'a> DwarfWriter<'a> {
pub fn new(slice: &mut [u8]) -> DwarfWriter {
DwarfWriter { slice, offset: 0 }
}
pub fn write_u8(&mut self, data: u8) {
self.slice[self.offset] = data;
self.offset += 1;
}
pub fn write_u32(&mut self, data: u32) {
LittleEndian::write_u32(&mut self.slice[self.offset..], data);
self.offset += 4;
}
}
fn read_encoded_pointer(reader: &mut DwarfReader, encoding: u8) -> Result<usize, ()> {
if encoding == DW_EH_PE_omit {
return Err(());
}
// DW_EH_PE_aligned implies it's an absolute pointer value
// However, we are linking library for 32-bits architecture
// The size of variable should be 4 bytes instead
if encoding == DW_EH_PE_aligned {
let shifted_virt_addr = round_up(reader.virt_addr as usize, mem::size_of::<u32>())?;
let addr_inc = shifted_virt_addr - reader.virt_addr as usize;
reader.slice = &reader.slice[addr_inc..];
reader.virt_addr = shifted_virt_addr as u32;
return Ok(reader.read_u32() as usize);
}
match encoding & 0x0F {
DW_EH_PE_absptr => Ok(reader.read_u32() as usize),
DW_EH_PE_uleb128 => Ok(reader.read_uleb128() as usize),
DW_EH_PE_udata2 => Ok(reader.read_u16() as usize),
DW_EH_PE_udata4 => Ok(reader.read_u32() as usize),
DW_EH_PE_udata8 => Ok(reader.read_u64() as usize),
DW_EH_PE_sleb128 => Ok(reader.read_sleb128() as usize),
DW_EH_PE_sdata2 => Ok(reader.read_i16() as usize),
DW_EH_PE_sdata4 => Ok(reader.read_i32() as usize),
DW_EH_PE_sdata8 => Ok(reader.read_i64() as usize),
_ => Err(()),
}
}
fn read_encoded_pointer_with_pc(
reader: &mut DwarfReader,
encoding: u8,
) -> Result<usize, ()> {
let entry_virt_addr = reader.virt_addr;
let mut result = read_encoded_pointer(reader, encoding)?;
// DW_EH_PE_aligned implies it's an absolute pointer value
if encoding == DW_EH_PE_aligned {
return Ok(result);
}
result = match encoding & 0x70 {
DW_EH_PE_pcrel => result.wrapping_add(entry_virt_addr as usize),
// .eh_frame normally would not have these kinds of relocations
// These would not be supported by a dedicated linker relocation schemes for RISC-V
DW_EH_PE_textrel | DW_EH_PE_datarel | DW_EH_PE_funcrel | DW_EH_PE_aligned => {
unimplemented!()
}
// Other values should be impossible
_ => unreachable!(),
};
if encoding & DW_EH_PE_indirect != 0 {
// There should not be a need for indirect addressing, as assembly code from
// the dynamic library should not be freely moved relative to the EH frame.
unreachable!()
}
Ok(result)
}
#[inline]
fn round_up(unrounded: usize, align: usize) -> Result<usize, ()> {
if align.is_power_of_two() {
Ok((unrounded + align - 1) & !(align - 1))
} else {
Err(())
}
}
// Minimalistic structure to store everything needed for parsing FDEs to synthesize
// .eh_frame_hdr section. Since we are only linking 1 object file, there should only be 1 call
// frame information (CFI) record, so there should be only 1 common information entry (CIE).
// So the class parses the only CIE on init, cache the encoding info, then parse the FDE on
// iterations based on the cached encoding format.
pub struct EH_Frame<'a> {
// It refers to the augmentation data that corresponds to 'R' in the augmentation string
pub fde_pointer_encoding: u8,
pub fde_reader: DwarfReader<'a>,
pub fde_sz: usize,
}
impl<'a> EH_Frame<'a> {
pub fn new(eh_frame_slice: &[u8], eh_frame_addr: u32) -> Result<EH_Frame, ()> {
let mut cie_reader = DwarfReader::new(eh_frame_slice, eh_frame_addr);
let eh_frame_size = eh_frame_slice.len();
let length = cie_reader.read_u32();
let fde_reader = match length {
// eh_frame with 0 lengths means the CIE is terminated
// while length == u32::MAX means that the length is only representable with 64 bits,
// which does not make sense in a system with 32-bit address.
0 | 0xFFFFFFFF => unimplemented!(),
_ => {
let mut fde_reader = DwarfReader::new(cie_reader.slice, cie_reader.virt_addr);
fde_reader.offset(length as i32);
fde_reader
}
};
let fde_sz = eh_frame_size - mem::size_of::<u32>() - length as usize;
// Routine check on the .eh_frame well-formness, in terms of CIE ID & Version args.
assert_eq!(cie_reader.read_u32(), 0);
assert_eq!(cie_reader.read_u8(), 1);
// Parse augmentation string
// The first character must be 'z', there is no way to proceed otherwise
assert_eq!(cie_reader.read_u8(), b'z');
// Establish a pointer that skips ahead of the string
// Skip code/data alignment factors & return address register along the way as well
// We only tackle the case where 'z' and 'R' are part of the augmentation string, otherwise
// we cannot get the addresses to make .eh_frame_hdr
let mut aug_data_reader = DwarfReader::new(cie_reader.slice, cie_reader.virt_addr);
let mut aug_str_len = 0;
loop {
if aug_data_reader.read_u8() == b'\0' {
break;
}
aug_str_len += 1;
}
if aug_str_len == 0 {
unimplemented!();
}
aug_data_reader.read_uleb128(); // Code alignment factor
aug_data_reader.read_sleb128(); // Data alignment factor
aug_data_reader.read_uleb128(); // Return address register
aug_data_reader.read_uleb128(); // Augmentation data length
let mut fde_pointer_encoding = DW_EH_PE_omit;
for _ in 0..aug_str_len {
match cie_reader.read_u8() {
b'L' => {
aug_data_reader.read_u8();
}
b'P' => {
let encoding = aug_data_reader.read_u8();
read_encoded_pointer(&mut aug_data_reader, encoding)?;
}
b'R' => {
fde_pointer_encoding = aug_data_reader.read_u8();
}
// Other characters are not supported
_ => unimplemented!(),
}
}
assert_ne!(fde_pointer_encoding, DW_EH_PE_omit);
Ok(EH_Frame { fde_pointer_encoding, fde_reader, fde_sz })
}
pub fn iterate_fde(&self, callback: &mut dyn FnMut(u32, u32)) -> Result<(), ()> {
// Parse each FDE to obtain the starting address that the FDE applies to
// Send the FDE offset and the mentioned address to a callback that write up the
// .eh_frame_hdr section
let mut remaining_len = self.fde_sz;
let mut reader = DwarfReader::new(self.fde_reader.slice, self.fde_reader.virt_addr);
loop {
if remaining_len == 0 {
break;
}
let fde_virt_addr = reader.virt_addr;
let length = match reader.read_u32() {
0 | 0xFFFFFFFF => unimplemented!(),
other => other,
};
// Remove the length of the header and the content from the counter
remaining_len -= length as usize + mem::size_of::<u32>();
let mut next_fde_reader = DwarfReader::new(reader.slice, reader.virt_addr);
next_fde_reader.offset(length as i32);
// Skip CIE pointer offset
reader.read_u32();
// Parse PC Begin using the encoding scheme mentioned in the CIE
let pc_begin = read_encoded_pointer_with_pc(&mut reader, self.fde_pointer_encoding)?;
callback(pc_begin as u32, fde_virt_addr);
reader = next_fde_reader;
}
Ok(())
}
}
pub struct EH_Frame_Hdr<'a> {
fde_writer: DwarfWriter<'a>,
eh_frame_hdr_addr: u32,
fdes: Vec<(u32, u32)>,
}
impl<'a> EH_Frame_Hdr<'a> {
// Create a EH_Frame_Hdr object, and write out the fixed fields of .eh_frame_hdr to memory
// eh_frame_ptr_enc will be 0x1B (PC-relative, 4 bytes)
// table_enc will be 0x3B (Relative to the start of .eh_frame_hdr, 4 bytes)
// Load address is not known at this point.
pub fn new(
eh_frame_hdr_slice: &mut [u8],
eh_frame_hdr_addr: u32,
eh_frame_addr: u32,
) -> EH_Frame_Hdr {
let mut writer = DwarfWriter::new(eh_frame_hdr_slice);
writer.write_u8(1);
writer.write_u8(0x1B);
writer.write_u8(0x03);
writer.write_u8(0x3B);
let eh_frame_offset =
(eh_frame_addr).wrapping_sub(eh_frame_hdr_addr + ((mem::size_of::<u8>() as u32) * 4));
writer.write_u32(eh_frame_offset);
writer.write_u32(0);
EH_Frame_Hdr { fde_writer: writer, eh_frame_hdr_addr, fdes: Vec::new() }
}
fn fde_count_offset() -> usize {
8
}
pub fn add_fde(&mut self, init_loc: u32, addr: u32) {
self.fdes.push((
init_loc.wrapping_sub(self.eh_frame_hdr_addr),
addr.wrapping_sub(self.eh_frame_hdr_addr),
));
}
pub fn finalize_fde(mut self) {
self.fdes
.sort_by(|(left_init_loc, _), (right_init_loc, _)| left_init_loc.cmp(right_init_loc));
for (init_loc, addr) in &self.fdes {
self.fde_writer.write_u32(*init_loc);
self.fde_writer.write_u32(*addr);
}
LittleEndian::write_u32(&mut self.fde_writer.slice[Self::fde_count_offset()..], self.fdes.len() as u32);
}
pub fn size_from_eh_frame(eh_frame: &[u8]) -> usize {
// The virtual address of the EH frame does not matter in this case
// Calculation of size does not involve modifying any headers
let mut reader = DwarfReader::new(eh_frame, 0);
let mut fde_count = 0;
while !reader.slice.is_empty() {
// The original length field should be able to hold the entire value.
// The device memory space is limited to 32-bits addresses anyway.
let entry_length = reader.read_u32();
if entry_length == 0 || entry_length == 0xFFFFFFFF {
unimplemented!()
}
if reader.read_u32() != 0 {
fde_count += 1;
}
reader.offset(entry_length as i32 - mem::size_of::<u32>() as i32)
}
12 + fde_count * 8
}
}

2893
nac3ld/src/elf.rs Normal file
View File

File diff suppressed because it is too large Load Diff

1490
nac3ld/src/lib.rs Normal file
View File

File diff suppressed because it is too large Load Diff