Browse Source

Refactor and suggestions

upstream-pthread
Andrea Ciliberti 2 years ago
parent
commit
bed08695a0
  1. 1
      src/lib.rs
  2. 13
      src/misc.rs
  3. 345
      src/thread.rs
  4. 2
      src/thread/attr.rs
  5. 8
      src/thread_keys.rs

1
src/lib.rs

@ -2,6 +2,5 @@
#![allow(non_camel_case_types)] #![allow(non_camel_case_types)]
#![allow(clippy::missing_safety_doc)] #![allow(clippy::missing_safety_doc)]
mod misc;
mod thread; mod thread;
mod thread_keys; mod thread_keys;

13
src/misc.rs

@ -1,13 +0,0 @@
//! Miscellaneous pthread functions
// The implementation within `newlib` stubs this out *entirely*. It's not possible to use a "syscall".
/*
#[no_mangle]
pub unsafe extern "C" fn sched_yield() -> libc::c_int {
ctru_sys::svcSleepThread(0);
0
}
*/
// `pthread_sigmask` and `pthread_atfork` are stubbed out by `newlib`

345
src/thread.rs

@ -1,345 +0,0 @@
//! PThread threads implemented using libctru.
use spin::rwlock::RwLock;
use std::collections::BTreeMap;
use std::ptr;
use std::sync::atomic::{AtomicUsize, Ordering};
pub mod attr;
/// The main thread's pthread ID
const MAIN_THREAD_ID: libc::pthread_t = 0;
// We initialize to 1 since 0 is reserved for the main thread.
static NEXT_ID: AtomicUsize = AtomicUsize::new(1);
// This is a spin-lock RwLock which yields the thread every loop.
static THREADS: RwLock<BTreeMap<libc::pthread_t, PThread>, spin::Yield> =
RwLock::new(BTreeMap::new());
// Initialize to zero (main thread ID) since the main thread will have the
// default value in this thread local.
#[thread_local]
static mut THREAD_ID: libc::pthread_t = MAIN_THREAD_ID;
#[derive(Copy, Clone)]
struct PThread {
thread: ShareablePtr<ctru_sys::Thread_tag>,
os_thread_id: u32,
is_detached: bool,
is_finished: bool,
result: ShareablePtr<libc::c_void>,
}
/// Pointers are not Send or Sync, though it's really just a lint. This struct
/// lets us ignore that "lint".
struct ShareablePtr<T>(*mut T);
unsafe impl<T> Send for ShareablePtr<T> {}
unsafe impl<T> Sync for ShareablePtr<T> {}
// We can't use the derives because they add an unnecessary T: Copy/Clone bound.
impl<T> Copy for ShareablePtr<T> {}
impl<T> Clone for ShareablePtr<T> {
fn clone(&self) -> Self {
*self
}
}
#[no_mangle]
pub unsafe extern "C" fn __syscall_thread_create(
native: *mut libc::pthread_t,
entrypoint: extern "C" fn(_: *mut libc::c_void) -> *mut libc::c_void,
arg: *mut libc::c_void,
_stackaddr: *const libc::c_void,
stacksize: libc::size_t,
) -> libc::c_int {
let stack_size = stacksize;
// let priority = (*attr).schedparam.sched_priority;
let mut priority = 0;
let _ = ctru_sys::svcGetThreadPriority(&mut priority, ctru_sys::CUR_THREAD_HANDLE);
// let processor_id = (*attr).processor_id;
let thread_id = NEXT_ID.fetch_add(1, Ordering::SeqCst) as libc::pthread_t;
extern "C" fn thread_start(main: *mut libc::c_void) {
unsafe {
Box::from_raw(main as *mut Box<dyn FnOnce()>)();
}
}
// The closure needs a fat pointer (64 bits) to work since it captures a variable and is thus a
// trait object, but *mut void is only 32 bits. We need double indirection to pass along the
// full closure data.
// We make this closure in the first place because threadCreate expects a void return type, but
// entrypoint returns a pointer so the types are incompatible.
let main: *mut Box<dyn FnOnce()> = Box::into_raw(Box::new(Box::new(move || {
THREAD_ID = thread_id;
let result = entrypoint(arg);
// Update the threads map with the result, and remove this thread if
// it's detached.
// We hold the lock the whole time so there isn't a race condition.
// (ex. we copy out the thread data, pthread_detach is called, we
// check is_detached and it's still false, so thread is never
// removed from the map)
let mut thread_map = THREADS.write();
if let Some(mut pthread) = thread_map.get_mut(&thread_id) {
pthread.is_finished = true;
pthread.result.0 = result;
if pthread.is_detached {
// libctru will call threadFree once this thread dies
thread_map.remove(&thread_id);
}
}
})));
let thread = ctru_sys::threadCreate(
Some(thread_start),
main as *mut libc::c_void,
stack_size as usize,
priority,
0, // Normal CPU
false,
);
if thread.is_null() {
// There was some error, but libctru doesn't expose the result.
// We assume there was permissions issue (such as too low of a priority).
// We also need to clean up the closure at this time.
drop(Box::from_raw(main));
return libc::EPERM;
}
// Get the OS thread ID
let os_handle = ctru_sys::threadGetHandle(thread);
let mut os_thread_id = 0;
let result = ctru_sys::svcGetThreadId(&mut os_thread_id, os_handle);
if ctru_sys::R_FAILED(result) {
// TODO: improve error handling? Different codes?
return libc::EPERM;
}
// Insert the thread into our map
THREADS.write().insert(
thread_id,
PThread {
thread: ShareablePtr(thread),
os_thread_id,
is_detached: false,
is_finished: false,
result: ShareablePtr(ptr::null_mut()),
},
);
*native = thread_id;
0
}
#[no_mangle]
pub unsafe extern "C" fn __syscall_thread_join(
thread_id: libc::pthread_t,
) -> *const libc::c_void { // The return value is returned as a pointer
if thread_id == MAIN_THREAD_ID {
// This is not a valid thread to join on
return std::ptr::null();
}
let thread_map = THREADS.read();
let Some(&pthread) = thread_map.get(&thread_id) else {
// This is not a valid thread ID
return std::ptr::null();
};
// We need to drop our read guard so it doesn't stay locked while joining
// the thread.
drop(thread_map);
if pthread.is_detached {
// Cannot join on a detached thread
return std::ptr::null();
}
let result = ctru_sys::threadJoin(pthread.thread.0, u64::MAX);
if ctru_sys::R_FAILED(result) {
// TODO: improve the error code by checking the result further?
return std::ptr::null();
}
ctru_sys::threadFree(pthread.thread.0);
let thread_data = THREADS.write().remove(&thread_id);
// This should always be Some, but we use an if let just in case.
if let Some(thread_data) = thread_data {
return thread_data.result.0;
}
std::ptr::null()
}
#[no_mangle]
pub unsafe extern "C" fn __syscall_thread_detach(thread_id: libc::pthread_t) -> libc::c_int {
if thread_id == MAIN_THREAD_ID {
// This is not a valid thread to detach
return libc::EINVAL;
}
let mut thread_map = THREADS.write();
let Some(mut pthread) = thread_map.get_mut(&thread_id) else {
// This is not a valid thread ID
return libc::ESRCH
};
if pthread.is_detached {
// Cannot detach an already detached thread
return libc::EINVAL;
}
pthread.is_detached = true;
ctru_sys::threadDetach(pthread.thread.0);
if pthread.is_finished {
// threadFree was already called by threadDetach
thread_map.remove(&thread_id);
}
0
}
#[no_mangle]
pub unsafe extern "C" fn __syscall_thread_self() -> libc::pthread_t {
let thread_id = THREAD_ID;
if thread_id == MAIN_THREAD_ID {
// Take this opportunity to populate the main thread's data in the map.
// They can only "legally" get the main thread's ID by calling this
// function, so this will run before they do anything that needs this.
// We have to ignore the possible error from svcGetThreadId since
// pthread_self cannot fail... But there shouldn't be an error anyways.
let mut os_thread_id = 0;
let _ = ctru_sys::svcGetThreadId(&mut os_thread_id, ctru_sys::CUR_THREAD_HANDLE);
THREADS.write().insert(
MAIN_THREAD_ID,
PThread {
// This null pointer is safe because we return before ever using
// it (in pthread_join and pthread_detach).
thread: ShareablePtr(ptr::null_mut()),
os_thread_id,
is_detached: true,
is_finished: false,
result: ShareablePtr(ptr::null_mut()),
},
);
}
thread_id
}
/// Closes a kernel handle on drop.
struct Handle(ctru_sys::Handle);
impl TryFrom<libc::pthread_t> for Handle {
type Error = libc::c_int;
fn try_from(thread_id: libc::pthread_t) -> Result<Self, Self::Error> {
let Some(&pthread) = THREADS.read().get(&thread_id) else {
// This is not a valid thread ID
return Err(libc::ESRCH)
};
if pthread.is_finished {
return Err(libc::ESRCH);
}
let mut os_handle = 0;
let ret = unsafe {
ctru_sys::svcOpenThread(
&mut os_handle,
ctru_sys::CUR_PROCESS_HANDLE,
pthread.os_thread_id,
)
};
if ctru_sys::R_FAILED(ret) || os_handle == u32::MAX {
// Either there was an error or the thread already finished
Err(libc::ESRCH)
} else {
Ok(Self(os_handle))
}
}
}
impl Drop for Handle {
fn drop(&mut self) {
unsafe {
// We ignore the error because we can't return it or panic.
let _ = ctru_sys::svcCloseHandle(self.0);
}
}
}
#[no_mangle]
pub unsafe extern "C" fn pthread_getschedparam(
thread_id: libc::pthread_t,
policy: *mut libc::c_int,
param: *mut libc::sched_param,
) -> libc::c_int {
let handle = match Handle::try_from(thread_id) {
Ok(handle) => handle,
Err(code) => return code,
};
let mut priority = 0;
let result = ctru_sys::svcGetThreadPriority(&mut priority, handle.0);
if ctru_sys::R_FAILED(result) {
// Some error occurred. This is the only error defined for this
// function, so return it. Maybe the thread exited while this function
// was exiting?
return libc::ESRCH;
}
(*param).sched_priority = priority;
// SCHED_FIFO is closest to how the cooperative app core works, while
// SCHED_RR is closest to how the preemptive sys core works.
// However, we don't have an API to get the current processor ID of a chosen
// thread (only the current), so we just always return SCHED_FIFO.
(*policy) = libc::SCHED_FIFO;
0
}
#[no_mangle]
pub unsafe extern "C" fn pthread_setschedparam(
thread_id: libc::pthread_t,
policy: libc::c_int,
param: *const libc::sched_param,
) -> libc::c_int {
if policy != libc::SCHED_FIFO {
// We only accept SCHED_FIFO. See the note in pthread_getschedparam.
return libc::EINVAL;
}
let handle = match Handle::try_from(thread_id) {
Ok(handle) => handle,
Err(code) => return code,
};
let result = ctru_sys::svcSetThreadPriority(handle.0, (*param).sched_priority);
if ctru_sys::R_FAILED(result) {
// Probably the priority is out of the permissible bounds
// TODO: improve the error code by checking the result further?
return libc::EPERM;
}
0
}
/*
#[no_mangle]
pub unsafe extern "C" fn pthread_getprocessorid_np() -> libc::c_int {
ctru_sys::svcGetProcessorID()
}
*/

2
src/thread/attr.rs

@ -28,4 +28,4 @@ pub unsafe extern "C" fn pthread_attr_setprocessorid_np(
0 0
} }
*/ */

8
src/thread_keys.rs

@ -20,7 +20,7 @@ fn is_valid_key(key: Key) -> bool {
} }
#[no_mangle] #[no_mangle]
pub unsafe extern "C" fn __syscall_tls_create( pub unsafe extern "C" fn __syscall_tls_create(
key: *mut libc::pthread_key_t, key: *mut libc::pthread_key_t,
destructor: Option<Destructor>, destructor: Option<Destructor>,
) -> libc::c_int { ) -> libc::c_int {
@ -33,7 +33,7 @@ pub unsafe extern "C" fn __syscall_tls_create(
} }
#[no_mangle] #[no_mangle]
pub unsafe extern "C" fn __syscall_tls_delete(key: libc::pthread_key_t) -> libc::c_int { pub unsafe extern "C" fn __syscall_tls_delete(key: libc::pthread_key_t) -> libc::c_int {
match KEYS.write().remove(&(key as Key)) { match KEYS.write().remove(&(key as Key)) {
// We had a entry, so it was a valid key. // We had a entry, so it was a valid key.
// It's officially undefined behavior if they use the key after this, // It's officially undefined behavior if they use the key after this,
@ -46,7 +46,7 @@ pub unsafe extern "C" fn __syscall_tls_delete(key: libc::pthread_key_t) -> libc
} }
#[no_mangle] #[no_mangle]
pub unsafe extern "C" fn __syscall_tls_get(key: libc::pthread_key_t) -> *mut libc::c_void { pub unsafe extern "C" fn __syscall_tls_get(key: libc::pthread_key_t) -> *mut libc::c_void {
if let Some(&value) = LOCALS.get(&(key as Key)) { if let Some(&value) = LOCALS.get(&(key as Key)) {
value as _ value as _
} else { } else {
@ -56,7 +56,7 @@ pub unsafe extern "C" fn __syscall_tls_get(key: libc::pthread_key_t) -> *mut li
} }
#[no_mangle] #[no_mangle]
pub unsafe extern "C" fn __syscall_tls_set( pub unsafe extern "C" fn __syscall_tls_set(
key: libc::pthread_key_t, key: libc::pthread_key_t,
value: *const libc::c_void, value: *const libc::c_void,
) -> libc::c_int { ) -> libc::c_int {

Loading…
Cancel
Save