Files
Odin/core/thread/thread_unix.odin
Laytan Laats 71afb8e7a5 removes the darwin specific paths from thread_unix
I know I left a note there about deadlocks but it doesn't seem to happen
anymore, especially now that the test runner creates cancellation points.

Also, what was this `can_set_thread_cancel_state` for? It does not make
sense to me that it enables cancellation, and then does it again later.
With the comment it seems like it should be `.DISABLE` to first disable
and then wait for the thread to start. But even that seems weird, why
do you need to do that? I removed it.
2025-11-07 21:40:25 +01:00

183 lines
5.0 KiB
Odin

#+build linux, darwin, freebsd, openbsd, netbsd, haiku
#+private
package thread
import "base:runtime"
import "core:sync"
import "core:sys/posix"
_IS_SUPPORTED :: true
// NOTE(tetra): Aligned here because of core/unix/pthread_linux.odin/pthread_t.
// Also see core/sys/darwin/mach_darwin.odin/semaphore_t.
Thread_Os_Specific :: struct #align(16) {
unix_thread: posix.pthread_t, // NOTE: very large on Darwin, small on Linux.
start_ok: sync.Sema,
}
//
// Creates a thread which will run the given procedure.
// It then waits for `start` to be called.
//
_create :: proc(procedure: Thread_Proc, priority: Thread_Priority) -> ^Thread {
__unix_thread_entry_proc :: proc "c" (t: rawptr) -> rawptr {
t := (^Thread)(t)
t.id = sync.current_thread_id()
for (.Started not_in sync.atomic_load(&t.flags)) {
sync.wait(&t.start_ok)
}
// Enable thread's cancelability.
err := posix.pthread_setcancelstate(.ENABLE, nil)
assert_contextless(err == nil)
// NOTE(laytan): .ASYNCHRONOUS should make `pthread_cancel` cancel immediately
// instead of waiting for a cancellation point.
// This does not seem to work on at least Darwin and NetBSD though.
err = posix.pthread_setcanceltype(.ASYNCHRONOUS, nil)
assert_contextless(err == nil)
{
init_context := t.init_context
// NOTE(tetra, 2023-05-31): Must do this AFTER thread.start() is called, so that the user can set the init_context, etc!
// Here on Unix, we start the OS thread in a running state, and so we manually have it wait on a condition
// variable above. We must perform that waiting BEFORE we select the context!
context = _select_context_for_thread(init_context)
defer {
_maybe_destroy_default_temp_allocator(init_context)
runtime.run_thread_local_cleaners()
}
t.procedure(t)
}
sync.atomic_or(&t.flags, { .Done })
if .Self_Cleanup in sync.atomic_load(&t.flags) {
res := posix.pthread_detach(t.unix_thread)
assert_contextless(res == nil)
t.unix_thread = {}
// NOTE(ftphikari): It doesn't matter which context 'free' received, right?
context = {}
free(t, t.creation_allocator)
}
return nil
}
attrs: posix.pthread_attr_t
if posix.pthread_attr_init(&attrs) != nil {
return nil // NOTE(tetra, 2019-11-01): POSIX OOM.
}
defer posix.pthread_attr_destroy(&attrs)
stacksize: posix.rlimit
if res := posix.getrlimit(.STACK, &stacksize); res == .OK && stacksize.rlim_cur > 0 {
_ = posix.pthread_attr_setstacksize(&attrs, uint(stacksize.rlim_cur))
}
res: posix.Errno
// NOTE(tetra, 2019-11-01): These only fail if their argument is invalid.
res = posix.pthread_attr_setdetachstate(&attrs, .CREATE_JOINABLE)
assert(res == nil)
when ODIN_OS != .Haiku && ODIN_OS != .NetBSD {
res = posix.pthread_attr_setinheritsched(&attrs, .EXPLICIT_SCHED)
assert(res == nil)
}
thread := new(Thread)
if thread == nil {
return nil
}
thread.creation_allocator = context.allocator
// Set thread priority.
policy: posix.Sched_Policy
when ODIN_OS != .Haiku && ODIN_OS != .NetBSD {
res = posix.pthread_attr_getschedpolicy(&attrs, &policy)
assert(res == nil)
}
params: posix.sched_param
res = posix.pthread_attr_getschedparam(&attrs, &params)
assert(res == nil)
low := posix.sched_get_priority_min(policy)
high := posix.sched_get_priority_max(policy)
switch priority {
case .Normal: // Okay
case .Low: params.sched_priority = low + 1
case .High: params.sched_priority = high
}
res = posix.pthread_attr_setschedparam(&attrs, &params)
assert(res == nil)
thread.procedure = procedure
if posix.pthread_create(&thread.unix_thread, &attrs, __unix_thread_entry_proc, thread) != nil {
free(thread, thread.creation_allocator)
return nil
}
return thread
}
_start :: proc(t: ^Thread) {
sync.atomic_or(&t.flags, { .Started })
sync.post(&t.start_ok)
}
_is_done :: proc(t: ^Thread) -> bool {
return .Done in sync.atomic_load(&t.flags)
}
_join :: proc(t: ^Thread) {
if posix.pthread_equal(posix.pthread_self(), t.unix_thread) {
return
}
// If the previous value was already `Joined`, then we can return.
if .Joined in sync.atomic_or(&t.flags, {.Joined}) {
return
}
// Prevent non-started threads from blocking main thread with initial wait
// condition.
for (.Started not_in sync.atomic_load(&t.flags)) {
_start(t)
}
posix.pthread_join(t.unix_thread, nil)
t.flags += {.Joined}
}
_join_multiple :: proc(threads: ..^Thread) {
for t in threads {
_join(t)
}
}
_destroy :: proc(t: ^Thread) {
_join(t)
t.unix_thread = {}
free(t, t.creation_allocator)
}
_terminate :: proc(t: ^Thread, exit_code: int) {
// NOTE(Feoramund): For thread cancellation to succeed on BSDs and
// possibly Darwin systems, the thread must call one of the pthread
// cancelation points at some point after this.
//
// The most obvious one of these is `pthread_cancel`, but there is an
// entire list of functions that act as cancelation points available in the
// pthreads manual page.
//
// This is in contrast to behavior I have seen on Linux where the thread is
// just terminated.
posix.pthread_cancel(t.unix_thread)
}
_yield :: proc() {
posix.sched_yield()
}