[kernel] Add thread_join syscall

Thread joining is an important primitive that I seem to have totally
forgotten to implement previously.

definitions/objects/thread.def

@@ -3,6 +3,7 @@ object thread : object {
 
     capabilities [
         kill
+        join
     ]
 
     method create [constructor] {
@@ -13,6 +14,8 @@ object thread : object {
 
     method kill [destructor cap:kill]
 
+    method join [cap:join]
+
     method exit [static]
 
     method sleep [static] {

src/kernel/objects/thread.cpp

@@ -52,6 +52,20 @@ thread::block()
     return m_wake_value;
 }
 
+j6_status_t
+thread::join()
+{
+    if (has_state(state::exited))
+        return j6_status_ok;
+
+    thread &caller = current();
+    if (&caller == this)
+        return j6_err_invalid_arg;
+
+    m_join_queue.add_thread(&caller);
+    return caller.block();
+}
+
 void
 thread::wake(uint64_t value)
 {

src/kernel/objects/thread.h

@@ -8,6 +8,7 @@
 #include <util/spinlock.h>
 
 #include "objects/kobject.h"
+#include "wait_queue.h"
 
 struct cpu_data;
 struct page_table;
@@ -98,6 +99,9 @@ public:
     /// \returns  The value passed to wake()
     uint64_t block();
 
+    /// Block the calling thread until this thread exits
+    j6_status_t join();
+
     /// Wake this thread, giving it a value
     /// \arg value  The value that block() should return
     void wake(uint64_t value = 0);
@@ -123,15 +127,16 @@ public:
     message_data & get_message_data() { return m_message_data; }
 
     inline bool has_state(state s) const {
-        return static_cast<uint8_t>(m_state) & static_cast<uint8_t>(s);
+        return __atomic_load_n(reinterpret_cast<const uint8_t*>(&m_state), __ATOMIC_ACQUIRE) &
+                static_cast<uint8_t>(s);
     }
 
     inline void set_state(state s) {
-        m_state = static_cast<state>(static_cast<uint8_t>(m_state) | static_cast<uint8_t>(s));
+        __atomic_or_fetch(reinterpret_cast<uint8_t*>(&m_state), static_cast<uint8_t>(s), __ATOMIC_ACQ_REL);
     }
 
     inline void clear_state(state s) {
-        m_state = static_cast<state>(static_cast<uint8_t>(m_state) & ~static_cast<uint8_t>(s));
+        __atomic_and_fetch(reinterpret_cast<uint8_t*>(&m_state), ~static_cast<uint8_t>(s), __ATOMIC_ACQ_REL);
     }
 
     inline tcb_node * tcb() { return &m_tcb; }
@@ -191,6 +196,7 @@ private:
     uint64_t m_wake_timeout;
     message_data m_message_data;
 
+    wait_queue m_join_queue;
     j6_handle_t m_self_handle;
 };
 

src/kernel/syscalls/thread.cpp

@@ -36,6 +36,12 @@ thread_kill(thread *self)
     return j6_status_ok;
 }
 
+j6_status_t
+thread_join(thread *self)
+{
+    return self->join();
+}
+
 j6_status_t
 thread_exit()
 {