uth: add got_posix_signal() to the 2LS ops

[akaros.git] / user / pthread / pthread.c

diff --git a/user/pthread/pthread.c b/user/pthread/pthread.c
index 59faf38..1a4d3f8 100644 (file)
--- a/user/pthread/pthread.c
+++ b/user/pthread/pthread.c
@@ -5,7 +5,6 @@
 #include <stdlib.h>
 #include <string.h>
 #include <parlib/assert.h>
 #include <stdlib.h>
 #include <string.h>
 #include <parlib/assert.h>

-#include <stdio.h>

 #include <errno.h>
 #include <parlib/parlib.h>
 #include <ros/event.h>
 #include <errno.h>
 #include <parlib/parlib.h>
 #include <ros/event.h>


@@ -17,6 +16,24 @@
 #include <parlib/ucq.h>
 #include <parlib/signal.h>
 #include <parlib/arch/trap.h>
 #include <parlib/ucq.h>
 #include <parlib/signal.h>
 #include <parlib/arch/trap.h>

+#include <parlib/ros_debug.h>
+#include <parlib/stdio.h>
+#include <sys/fork_cb.h>
+
+#include <parlib/alarm.h>
+#include <futex.h>
+#include <parlib/serialize.h>
+
+/* TODO: eventually, we probably want to split this into the pthreads interface
+ * and a default 2LS.  That way, apps can use the pthreads interface and use any
+ * 2LS.  Here's a few blockers:
+ * - pthread_cleanup(): probably support at the uthread level
+ * - attrs and creation: probably use a default stack size and handle detached
+ * - getattrs_np: return -1, mostly due to the stackaddr.  Callers probably want
+ *   a real 2LS operation.
+ * Then we can split pthreads into parlib/default_sched.c (replaces thread0) and
+ * pthread.c.  After that, we can have a signal handling thread (even for
+ * 'thread0'), which allows us to close() or do other vcore-ctx-unsafe ops. */

 
 struct pthread_queue ready_queue = TAILQ_HEAD_INITIALIZER(ready_queue);
 struct pthread_queue active_queue = TAILQ_HEAD_INITIALIZER(active_queue);
 
 struct pthread_queue ready_queue = TAILQ_HEAD_INITIALIZER(ready_queue);
 struct pthread_queue active_queue = TAILQ_HEAD_INITIALIZER(active_queue);


@@ -25,6 +42,8 @@ int threads_ready = 0;
 int threads_active = 0;
 atomic_t threads_total;
 bool need_tls = TRUE;
 int threads_active = 0;
 atomic_t threads_total;
 bool need_tls = TRUE;

+static uint64_t fork_generation;
+#define INIT_FORK_GENERATION 1

 
 /* Array of per-vcore structs to manage waiting on syscalls and handling
  * overflow.  Init'd in pth_init(). */
 
 /* Array of per-vcore structs to manage waiting on syscalls and handling
  * overflow.  Init'd in pth_init(). */


@@ -32,10 +51,10 @@ struct sysc_mgmt *sysc_mgmt = 0;
 
 /* Helper / local functions */
 static int get_next_pid(void);
 
 /* Helper / local functions */
 static int get_next_pid(void);

-static inline void spin_to_sleep(unsigned int spins, unsigned int *spun);

 static inline void pthread_exit_no_cleanup(void *ret);
 
 /* Pthread 2LS operations */
 static inline void pthread_exit_no_cleanup(void *ret);
 
 /* Pthread 2LS operations */

+static void pth_sched_init(void);

 static void pth_sched_entry(void);
 static void pth_thread_runnable(struct uthread *uthread);
 static void pth_thread_paused(struct uthread *uthread);
 static void pth_sched_entry(void);
 static void pth_thread_runnable(struct uthread *uthread);
 static void pth_thread_paused(struct uthread *uthread);


@@ -43,20 +62,31 @@ static void pth_thread_blockon_sysc(struct uthread *uthread, void *sysc);
 static void pth_thread_has_blocked(struct uthread *uthread, int flags);
 static void pth_thread_refl_fault(struct uthread *uth,
                                   struct user_context *ctx);
 static void pth_thread_has_blocked(struct uthread *uthread, int flags);
 static void pth_thread_refl_fault(struct uthread *uth,
                                   struct user_context *ctx);

+static void pth_thread_exited(struct uthread *uth);
+static struct uthread *pth_thread_create(void *(*func)(void *), void *arg);
+static void pth_got_posix_signal(int sig_nr, struct siginfo *info);
+static void pth_thread_bulk_runnable(uth_sync_t *wakees);

 
 /* Event Handlers */
 static void pth_handle_syscall(struct event_msg *ev_msg, unsigned int ev_type,
                                void *data);
 
 struct schedule_ops pthread_sched_ops = {
 
 /* Event Handlers */
 static void pth_handle_syscall(struct event_msg *ev_msg, unsigned int ev_type,
                                void *data);
 
 struct schedule_ops pthread_sched_ops = {

+       .sched_init = pth_sched_init,

        .sched_entry = pth_sched_entry,
        .thread_runnable = pth_thread_runnable,
        .thread_paused = pth_thread_paused,
        .thread_blockon_sysc = pth_thread_blockon_sysc,
        .thread_has_blocked = pth_thread_has_blocked,
        .thread_refl_fault = pth_thread_refl_fault,
        .sched_entry = pth_sched_entry,
        .thread_runnable = pth_thread_runnable,
        .thread_paused = pth_thread_paused,
        .thread_blockon_sysc = pth_thread_blockon_sysc,
        .thread_has_blocked = pth_thread_has_blocked,
        .thread_refl_fault = pth_thread_refl_fault,

+       .thread_exited = pth_thread_exited,
+       .thread_create = pth_thread_create,
+       .got_posix_signal = pth_got_posix_signal,
+       .thread_bulk_runnable = pth_thread_bulk_runnable,

 };
 
 };
 

+struct schedule_ops *sched_ops = &pthread_sched_ops;
+

 /* Static helpers */
 static void __pthread_free_stack(struct pthread_tcb *pt);
 static int __pthread_allocate_stack(struct pthread_tcb *pt);
 /* Static helpers */
 static void __pthread_free_stack(struct pthread_tcb *pt);
 static int __pthread_allocate_stack(struct pthread_tcb *pt);


@@ -69,8 +99,8 @@ static void __attribute__((noreturn)) pth_sched_entry(void)
 {
        uint32_t vcoreid = vcore_id();
        if (current_uthread) {
 {
        uint32_t vcoreid = vcore_id();
        if (current_uthread) {

-               /* Prep the pthread to run any pending posix signal handlers registered
-         * via pthread_kill once it is restored. */
+               /* Prep the pthread to run any pending posix signal handlers
+                * registered via pthread_kill once it is restored. */

                uthread_prep_pending_signals(current_uthread);
                /* Run the thread itself */
                run_current_uthread();
                uthread_prep_pending_signals(current_uthread);
                /* Run the thread itself */
                run_current_uthread();


@@ -78,14 +108,20 @@ static void __attribute__((noreturn)) pth_sched_entry(void)
        }
        /* no one currently running, so lets get someone from the ready queue */
        struct pthread_tcb *new_thread = NULL;
        }
        /* no one currently running, so lets get someone from the ready queue */
        struct pthread_tcb *new_thread = NULL;

-       /* Try to get a thread.  If we get one, we'll break out and run it.  If not,
-        * we'll try to yield.  vcore_yield() might return, if we lost a race and
-        * had a new event come in, one that may make us able to get a new_thread */
+
+       /* Try to get a thread.  If we get one, we'll break out and run it.  If
+        * not, we'll try to yield.  vcore_yield() might return, if we lost a
+        * race and had a new event come in, one that may make us able to get a
+        * new_thread */

        do {
                handle_events(vcoreid);
                __check_preempt_pending(vcoreid);
                mcs_pdr_lock(&queue_lock);
        do {
                handle_events(vcoreid);
                __check_preempt_pending(vcoreid);
                mcs_pdr_lock(&queue_lock);

-               new_thread = TAILQ_FIRST(&ready_queue);
+               TAILQ_FOREACH(new_thread, &ready_queue, tq_next) {
+                       if (new_thread->fork_generation < fork_generation)
+                               continue;
+                       break;
+               }

                if (new_thread) {
                        TAILQ_REMOVE(&ready_queue, new_thread, tq_next);
                        assert(new_thread->state == PTH_RUNNABLE);
                if (new_thread) {
                        TAILQ_REMOVE(&ready_queue, new_thread, tq_next);
                        assert(new_thread->state == PTH_RUNNABLE);


@@ -94,8 +130,9 @@ static void __attribute__((noreturn)) pth_sched_entry(void)
                        threads_active++;
                        threads_ready--;
                        mcs_pdr_unlock(&queue_lock);
                        threads_active++;
                        threads_ready--;
                        mcs_pdr_unlock(&queue_lock);

-                       /* If you see what looks like the same uthread running in multiple
-                        * places, your list might be jacked up.  Turn this on. */
+                       /* If you see what looks like the same uthread running
+                        * in multiple places, your list might be jacked up.
+                        * Turn this on. */

                        printd("[P] got uthread %08p on vc %d state %08p flags %08p\n",
                               new_thread, vcoreid,
                               ((struct uthread*)new_thread)->state,
                        printd("[P] got uthread %08p on vc %d state %08p flags %08p\n",
                               new_thread, vcoreid,
                               ((struct uthread*)new_thread)->state,


@@ -105,12 +142,12 @@ static void __attribute__((noreturn)) pth_sched_entry(void)
                mcs_pdr_unlock(&queue_lock);
                /* no new thread, try to yield */
                printd("[P] No threads, vcore %d is yielding\n", vcore_id());
                mcs_pdr_unlock(&queue_lock);
                /* no new thread, try to yield */
                printd("[P] No threads, vcore %d is yielding\n", vcore_id());

-               /* TODO: you can imagine having something smarter here, like spin for a
-                * bit before yielding. */
+               /* TODO: you can imagine having something smarter here, like
+                * spin for a bit before yielding. */

                vcore_yield(FALSE);
        } while (1);
        /* Prep the pthread to run any pending posix signal handlers registered
                vcore_yield(FALSE);
        } while (1);
        /* Prep the pthread to run any pending posix signal handlers registered

-     * via pthread_kill once it is restored. */
+        * via pthread_kill once it is restored. */

        uthread_prep_pending_signals((struct uthread*)new_thread);
        /* Run the thread itself */
        run_uthread((struct uthread*)new_thread);
        uthread_prep_pending_signals((struct uthread*)new_thread);
        /* Run the thread itself */
        run_uthread((struct uthread*)new_thread);


@@ -129,33 +166,36 @@ static void __pthread_run(void)
 static void pth_thread_runnable(struct uthread *uthread)
 {
        struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
 static void pth_thread_runnable(struct uthread *uthread)
 {
        struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;

-       /* At this point, the 2LS can see why the thread blocked and was woken up in
-        * the first place (coupling these things together).  On the yield path, the
-        * 2LS was involved and was able to set the state.  Now when we get the
-        * thread back, we can take a look. */
-       printd("pthread %08p runnable, state was %d\n", pthread, pthread->state);
+
+       /* At this point, the 2LS can see why the thread blocked and was woken
+        * up in the first place (coupling these things together).  On the yield
+        * path, the 2LS was involved and was able to set the state.  Now when
+        * we get the thread back, we can take a look. */
+       printd("pthread %08p runnable, state was %d\n", pthread,
+              pthread->state);

        switch (pthread->state) {
        switch (pthread->state) {

-               case (PTH_CREATED):
-               case (PTH_BLK_YIELDING):
-               case (PTH_BLK_JOINING):
-               case (PTH_BLK_SYSC):
-               case (PTH_BLK_PAUSED):
-               case (PTH_BLK_MUTEX):
-                       /* can do whatever for each of these cases */
-                       break;
-               default:
-                       panic("Odd state %d for pthread %08p\n", pthread->state, pthread);
+       case (PTH_CREATED):
+       case (PTH_BLK_YIELDING):
+       case (PTH_BLK_SYSC):
+       case (PTH_BLK_PAUSED):
+       case (PTH_BLK_MUTEX):
+       case (PTH_BLK_MISC):
+               /* can do whatever for each of these cases */
+               break;
+       default:
+               panic("Odd state %d for pthread %08p\n", pthread->state,
+                     pthread);

        }
        pthread->state = PTH_RUNNABLE;
        }
        pthread->state = PTH_RUNNABLE;

-       /* Insert the newly created thread into the ready queue of threads.
-        * It will be removed from this queue later when vcore_entry() comes up */
+       /* Insert the newly created thread into the ready queue of threads.  It
+        * will be removed from this queue later when vcore_entry() comes up */

        mcs_pdr_lock(&queue_lock);
        /* Again, GIANT WARNING: if you change this, change batch wakeup code */
        TAILQ_INSERT_TAIL(&ready_queue, pthread, tq_next);
        threads_ready++;
        mcs_pdr_unlock(&queue_lock);
        mcs_pdr_lock(&queue_lock);
        /* Again, GIANT WARNING: if you change this, change batch wakeup code */
        TAILQ_INSERT_TAIL(&ready_queue, pthread, tq_next);
        threads_ready++;
        mcs_pdr_unlock(&queue_lock);

-       /* Smarter schedulers should look at the num_vcores() and how much work is
-        * going on to make a decision about how many vcores to request. */
+       /* Smarter schedulers should look at the num_vcores() and how much work
+        * is going on to make a decision about how many vcores to request. */

        vcore_request_more(threads_ready);
 }
 
        vcore_request_more(threads_ready);
 }
 


@@ -176,9 +216,9 @@ static void pth_thread_paused(struct uthread *uthread)
        __pthread_generic_yield(pthread);
        /* communicate to pth_thread_runnable */
        pthread->state = PTH_BLK_PAUSED;
        __pthread_generic_yield(pthread);
        /* communicate to pth_thread_runnable */
        pthread->state = PTH_BLK_PAUSED;

-       /* At this point, you could do something clever, like put it at the front of
-        * the runqueue, see if it was holding a lock, do some accounting, or
-        * whatever. */
+       /* At this point, you could do something clever, like put it at the
+        * front of the runqueue, see if it was holding a lock, do some
+        * accounting, or whatever. */

        pth_thread_runnable(uthread);
 }
 
        pth_thread_runnable(uthread);
 }
 


@@ -203,14 +243,16 @@ static void pth_handle_syscall(struct event_msg *ev_msg, unsigned int ev_type,
        struct syscall *sysc;
        assert(in_vcore_context());
        /* if we just got a bit (not a msg), it should be because the process is
        struct syscall *sysc;
        assert(in_vcore_context());
        /* if we just got a bit (not a msg), it should be because the process is

-        * still an SCP and hasn't started using the MCP ev_q yet (using the simple
-        * ev_q and glibc's blockon) or because the bit is still set from an old
-        * ev_q (blocking syscalls from before we could enter vcore ctx).  Either
-        * way, just return.  Note that if you screwed up the pth ev_q and made it
-        * NO_MSG, you'll never notice (we used to assert(ev_msg)). */
+        * still an SCP and hasn't started using the MCP ev_q yet (using the
+        * simple ev_q and glibc's blockon) or because the bit is still set from
+        * an old ev_q (blocking syscalls from before we could enter vcore ctx).
+        * Either way, just return.  Note that if you screwed up the pth ev_q
+        * and made it NO_MSG, you'll never notice (we used to assert(ev_msg)).
+        * */

        if (!ev_msg)
                return;
        if (!ev_msg)
                return;

-       /* It's a bug if we don't have a msg (we're handling a syscall bit-event) */
+       /* It's a bug if we don't have a msg (we're handling a syscall
+        * bit-event) */

        assert(ev_msg);
        /* Get the sysc from the message and just restart it */
        sysc = ev_msg->ev_arg3;
        assert(ev_msg);
        /* Get the sysc from the message and just restart it */
        sysc = ev_msg->ev_arg3;


@@ -235,8 +277,8 @@ static void pth_thread_blockon_sysc(struct uthread *uthread, void *syscall)
        sysc->u_data = uthread;
        /* Register our vcore's syscall ev_q to hear about this syscall. */
        if (!register_evq(sysc, sysc_mgmt[vcoreid].ev_q)) {
        sysc->u_data = uthread;
        /* Register our vcore's syscall ev_q to hear about this syscall. */
        if (!register_evq(sysc, sysc_mgmt[vcoreid].ev_q)) {

-               /* Lost the race with the call being done.  The kernel won't send the
-                * event.  Just restart him. */
+               /* Lost the race with the call being done.  The kernel won't
+                * send the event.  Just restart him. */

                restart_thread(sysc);
        }
        /* GIANT WARNING: do not touch the thread after this point. */
                restart_thread(sysc);
        }
        /* GIANT WARNING: do not touch the thread after this point. */


@@ -247,14 +289,19 @@ static void pth_thread_has_blocked(struct uthread *uthread, int flags)
        struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
 
        __pthread_generic_yield(pthread);
        struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
 
        __pthread_generic_yield(pthread);

-       /* could imagine doing something with the flags.  For now, we just treat all
-        * externally blocked reasons as 'MUTEX'.  Whatever we do here, we are
-        * mostly communicating to our future selves in pth_thread_runnable(), which
-        * gets called by whoever triggered this callback */
-       pthread->state = PTH_BLK_MUTEX;
-       /* Just for yucks: */
-       if (flags == UTH_EXT_BLK_JUSTICE)
-               printf("For great justice!\n");
+       /* Whatever we do here, we are mostly communicating to our future selves
+        * in pth_thread_runnable(), which gets called by whoever triggered this
+        * callback */
+       switch (flags) {
+       case UTH_EXT_BLK_YIELD:
+               pthread->state = PTH_BLK_YIELDING;
+               break;
+       case UTH_EXT_BLK_MUTEX:
+               pthread->state = PTH_BLK_MUTEX;
+               break;
+       default:
+               pthread->state = PTH_BLK_MISC;
+       };

 }
 
 static void __signal_and_restart(struct uthread *uthread,
 }
 
 static void __signal_and_restart(struct uthread *uthread,


@@ -270,9 +317,84 @@ static void handle_div_by_zero(struct uthread *uthread, unsigned int err,
        __signal_and_restart(uthread, SIGFPE, FPE_INTDIV, (void*)aux);
 }
 
        __signal_and_restart(uthread, SIGFPE, FPE_INTDIV, (void*)aux);
 }
 

+// checks that usys in go passes its arguments correctly
+// it only automatically checks with 7 arguments, print is for the rest
+int go_usys_tester(uint64_t a, uint64_t b, uint64_t c, uint64_t d, uint64_t e,
+                   uint64_t f, uint64_t g, uint64_t h, uint64_t i, uint64_t j,
+                   uint64_t k, uint64_t l)
+{
+       printf("a = %lu, b = %lu, c = %lu, d = %lu, e = %lu, f = %lu, g = %lu, h = %lu, i = %lu, j = %lu, k = %lu, l = %lu\n",
+               a, b, c, d, e, f, g, h, i, j, k, l);
+       uint64_t ret_val = 0;
+
+       ret_val |= a;
+       ret_val |= (b << 8);
+       ret_val |= (c << 16);
+       ret_val |= (d << 24);
+       ret_val |= (e << 32);
+       ret_val |= (f << 40);
+       ret_val |= (g << 48);
+       return ret_val;
+}
+
+struct alarm_waiter *abort_syscall_at_abs_unix(uint64_t deadline)
+{
+       // note the malloc of waiter instead of it going on the stack
+       struct alarm_waiter *waiter = malloc(sizeof(struct alarm_waiter));
+
+       init_awaiter(waiter, alarm_abort_sysc);
+       waiter->data = current_uthread;
+       set_awaiter_abs_unix(waiter, deadline);
+       set_alarm(waiter);
+       return waiter;
+}
+
+bool unset_alarm_with_free(struct alarm_waiter *waiter)
+{
+       // we need to free the waiter we created in abort_syscall_at_abs_unix
+       bool ret = unset_alarm(waiter);
+
+       free(waiter);
+       return ret;
+}
+
+// ros_syscall_sync, but makes sure errors are zeros if there is no error
+void go_syscall(struct syscall *sysc)
+{
+       ros_syscall_sync(sysc);
+       if (!syscall_retval_is_error(sysc->num, sysc->retval)) {
+               sysc->err = 0;
+               sysc->errstr[0] = 0;
+       }
+}
+
+static void set_up_go_table(void **table)
+{
+       table[0] = abort_syscall_at_abs_unix;
+       table[1] = unset_alarm_with_free;
+       table[2] = go_syscall;
+       table[3] = go_usys_tester;
+       table[4] = futex;
+       table[5] = serialize_argv_envp;
+       table[6] = free;
+       assert(table[7] == (void*) 0xDEADBEEF);
+}
+

 static void handle_gp_fault(struct uthread *uthread, unsigned int err,
                             unsigned long aux)
 {
 static void handle_gp_fault(struct uthread *uthread, unsigned int err,
                             unsigned long aux)
 {

+       //TODO this code is x86-64 only
+       uint64_t rax = uthread->u_ctx.tf.hw_tf.tf_rax;
+
+       // we fault with a known high 16 bits in go to set up a function pointer
+       // table, the address of the table is the low 48 bits
+       if (rax >> 48 == 0xDEAD) {
+               set_up_go_table((void **)(0xFFFFFFFFFFFFUL & rax));
+               // we jump over the call instruction which is 2 bytes
+               uthread->u_ctx.tf.hw_tf.tf_rip += 2;
+               pth_thread_runnable(uthread);
+               return;
+       }

        __signal_and_restart(uthread, SIGSEGV, SEGV_ACCERR, (void*)aux);
 }
 
        __signal_and_restart(uthread, SIGSEGV, SEGV_ACCERR, (void*)aux);
 }
 


@@ -332,8 +454,74 @@ static void pth_thread_refl_fault(struct uthread *uth,
        }
 }
 
        }
 }
 

+static void pth_thread_exited(struct uthread *uth)
+{
+       struct pthread_tcb *pthread = (struct pthread_tcb*)uth;
+
+       __pthread_generic_yield(pthread);
+       /* Catch some bugs */
+       pthread->state = PTH_EXITING;
+       /* Destroy the pthread */
+       uthread_cleanup(uth);
+       /* Cleanup, mirroring pthread_create() */
+       __pthread_free_stack(pthread);
+       /* If we were the last pthread, we exit for the whole process.  Keep in
+        * mind that thread0 is counted in this, so this will only happen if
+        * that thread calls pthread_exit(). */
+       if ((atomic_fetch_and_add(&threads_total, -1) == 1))
+               exit(0);
+}
+
+/* Careful, if someone used the pthread_need_tls() hack to turn off TLS, it will
+ * also be turned off for these threads. */
+static struct uthread *pth_thread_create(void *(*func)(void *), void *arg)
+{
+       struct pthread_tcb *pth;
+       int ret;
+
+       ret = pthread_create(&pth, NULL, func, arg);
+       return ret == 0 ? (struct uthread*)pth : NULL;
+}
+
+/* Careful, that fake_uctx takes up a lot of stack space.  We could call
+ * pthread_kill too.  Note the VMM 2LS has similar code. */
+static void pth_got_posix_signal(int sig_nr, struct siginfo *info)
+{
+       struct user_context fake_uctx;
+
+       /* If we happen to have a current uthread, we can use that - perhaps
+        * that's what the user wants.  If not, we'll build a fake one
+        * representing our current call stack. */
+       if (current_uthread) {
+               trigger_posix_signal(sig_nr, info, get_cur_uth_ctx());
+       } else {
+               init_user_ctx(&fake_uctx, (uintptr_t)pth_got_posix_signal,
+                             get_stack_pointer());
+               trigger_posix_signal(sig_nr, info, &fake_uctx);
+       }
+}
+
+static void pth_thread_bulk_runnable(uth_sync_t *wakees)
+{
+       struct uthread *uth_i;
+       struct pthread_tcb *pth_i;
+
+       /* Amortize the lock grabbing over all restartees */
+       mcs_pdr_lock(&queue_lock);
+       while ((uth_i = __uth_sync_get_next(wakees))) {
+               pth_i = (struct pthread_tcb*)uth_i;
+               pth_i->state = PTH_RUNNABLE;
+               TAILQ_INSERT_TAIL(&ready_queue, pth_i, tq_next);
+               threads_ready++;
+       }
+       mcs_pdr_unlock(&queue_lock);
+       vcore_request_more(threads_ready);
+}
+

 /* Akaros pthread extensions / hacks */
 
 /* Akaros pthread extensions / hacks */
 

+/* Careful using this - glibc and gcc are likely to use TLS without you knowing
+ * it. */

 void pthread_need_tls(bool need)
 {
        need_tls = need;
 void pthread_need_tls(bool need)
 {
        need_tls = need;


@@ -369,8 +557,8 @@ static int __pthread_allocate_stack(struct pthread_tcb *pt)
        int force_a_page_fault;
        assert(pt->stacksize);
        void* stackbot = mmap(0, pt->stacksize,
        int force_a_page_fault;
        assert(pt->stacksize);
        void* stackbot = mmap(0, pt->stacksize,

-                             PROT_READ|PROT_WRITE|PROT_EXEC,
-                             MAP_ANONYMOUS, -1, 0);
+                             PROT_READ | PROT_WRITE | PROT_EXEC,
+                             MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);

        if (stackbot == MAP_FAILED)
                return -1; // errno set by mmap
        pt->stacktop = stackbot + pt->stacksize;
        if (stackbot == MAP_FAILED)
                return -1; // errno set by mmap
        pt->stacktop = stackbot + pt->stacksize;


@@ -412,7 +600,7 @@ int pthread_attr_getguardsize(pthread_attr_t *attr, size_t *guardsize)
 }
 
 int pthread_attr_getstack(const pthread_attr_t *__restrict __attr,
 }
 
 int pthread_attr_getstack(const pthread_attr_t *__restrict __attr,

-                                                  void **__stackaddr, size_t *__stacksize)
+                         void **__stackaddr, size_t *__stacksize)

 {
        *__stackaddr = __attr->stackaddr;
        *__stacksize = __attr->stacksize;
 {
        *__stackaddr = __attr->stackaddr;
        *__stacksize = __attr->stacksize;


@@ -421,59 +609,86 @@ int pthread_attr_getstack(const pthread_attr_t *__restrict __attr,
 
 int pthread_getattr_np(pthread_t __th, pthread_attr_t *__attr)
 {
 
 int pthread_getattr_np(pthread_t __th, pthread_attr_t *__attr)
 {

+       struct uthread *uth = (struct uthread*)__th;
+

        __attr->stackaddr = __th->stacktop - __th->stacksize;
        __attr->stacksize = __th->stacksize;
        __attr->stackaddr = __th->stacktop - __th->stacksize;
        __attr->stacksize = __th->stacksize;

-       if (__th->detached)
+       if (atomic_read(&uth->join_ctl.state) == UTH_JOIN_DETACHED)

                __attr->detachstate = PTHREAD_CREATE_DETACHED;
        else
                __attr->detachstate = PTHREAD_CREATE_JOINABLE;
        return 0;
 }
 
                __attr->detachstate = PTHREAD_CREATE_DETACHED;
        else
                __attr->detachstate = PTHREAD_CREATE_JOINABLE;
        return 0;
 }
 

+/* All multi-threading is suspended during a fork.  Thread0 will continue to
+ * run, which could come up if SYS_fork blocks or we get interrupted.  Parents
+ * will continue threading after the fork, like normal.  Old threads in the
+ * child will never run again.  New threads in the child will run. */
+static void pth_pre_fork(void)
+{
+       struct pthread_tcb *pth_0 = (struct pthread_tcb*)current_uthread;
+
+       if (!uthread_is_thread0(current_uthread))
+               panic("Tried to fork from a non-thread0 thread!");
+       if (in_multi_mode())
+               panic("Tried to fork from an MCP!");
+       pth_0->fork_generation = fork_generation + 1;
+       /* in case we get interrupted after incrementing the global gen */
+       cmb();
+       /* We're single-core and thread0 here, so we can modify fork_generation
+        */
+       fork_generation++;
+       /* At this point, whether we come back as the child or the parent, no
+        * old thread (from the previous generation) will run. */
+}
+
+static void pth_post_fork(pid_t ret)
+{
+       struct pthread_tcb *pth_0 = (struct pthread_tcb*)current_uthread;
+
+       if (ret) {
+               fork_generation--;
+               pth_0->fork_generation = fork_generation;
+       }
+}
+

 /* Do whatever init you want.  At some point call uthread_2ls_init() and pass it
  * a uthread representing thread0 (int main()) */
 /* Do whatever init you want.  At some point call uthread_2ls_init() and pass it
  * a uthread representing thread0 (int main()) */

-void __attribute__((constructor)) pthread_lib_init(void)
+void pth_sched_init(void)

 {
        uintptr_t mmap_block;
        struct pthread_tcb *t;
        int ret;
 
 {
        uintptr_t mmap_block;
        struct pthread_tcb *t;
        int ret;
 

-       /* Only run once, but make sure that uthread_lib_init() has run already. */
-       init_once_racy(return);
-       uthread_lib_init();
-

        mcs_pdr_init(&queue_lock);
        mcs_pdr_init(&queue_lock);

+       fork_generation = INIT_FORK_GENERATION;

        /* Create a pthread_tcb for the main thread */
        ret = posix_memalign((void**)&t, __alignof__(struct pthread_tcb),
                             sizeof(struct pthread_tcb));
        assert(!ret);
        /* Create a pthread_tcb for the main thread */
        ret = posix_memalign((void**)&t, __alignof__(struct pthread_tcb),
                             sizeof(struct pthread_tcb));
        assert(!ret);

-       memset(t, 0, sizeof(struct pthread_tcb));       /* aggressively 0 for bugs */
+       /* aggressively 0 for bugs */
+       memset(t, 0, sizeof(struct pthread_tcb));

        t->id = get_next_pid();
        t->id = get_next_pid();

+       t->fork_generation = fork_generation;

        t->stacksize = USTACK_NUM_PAGES * PGSIZE;
        t->stacktop = (void*)USTACKTOP;
        t->stacksize = USTACK_NUM_PAGES * PGSIZE;
        t->stacktop = (void*)USTACKTOP;

-       t->detached = TRUE;

        t->state = PTH_RUNNING;
        t->state = PTH_RUNNING;

-       t->joiner = 0;

        /* implies that sigmasks are longs, which they are. */
        assert(t->id == 0);
        /* implies that sigmasks are longs, which they are. */
        assert(t->id == 0);

-       t->sched_policy = SCHED_FIFO;
-       t->sched_priority = 0;

        SLIST_INIT(&t->cr_stack);
        /* Put the new pthread (thread0) on the active queue */
        mcs_pdr_lock(&queue_lock);
        threads_active++;
        TAILQ_INSERT_TAIL(&active_queue, t, tq_next);
        mcs_pdr_unlock(&queue_lock);
        SLIST_INIT(&t->cr_stack);
        /* Put the new pthread (thread0) on the active queue */
        mcs_pdr_lock(&queue_lock);
        threads_active++;
        TAILQ_INSERT_TAIL(&active_queue, t, tq_next);
        mcs_pdr_unlock(&queue_lock);

-       /* Tell the kernel where and how we want to receive events.  This is just an
-        * example of what to do to have a notification turned on.  We're turning on
-        * USER_IPIs, posting events to vcore 0's vcpd, and telling the kernel to
-        * send to vcore 0.  Note sys_self_notify will ignore the vcoreid and
-        * private preference.  Also note that enable_kevent() is just an example,
-        * and you probably want to use parts of event.c to do what you want. */
+       /* Tell the kernel where and how we want to receive events.  This is
+        * just an example of what to do to have a notification turned on.
+        * We're turning on USER_IPIs, posting events to vcore 0's vcpd, and
+        * telling the kernel to send to vcore 0.  Note sys_self_notify will
+        * ignore the vcoreid and private preference.  Also note that
+        * enable_kevent() is just an example, and you probably want to use
+        * parts of event.c to do what you want. */

        enable_kevent(EV_USER_IPI, 0, EVENT_IPI | EVENT_VCORE_PRIVATE);
        enable_kevent(EV_USER_IPI, 0, EVENT_IPI | EVENT_VCORE_PRIVATE);

-
-       /* Handle syscall events. */
-       register_ev_handler(EV_SYSCALL, pth_handle_syscall, 0);

        /* Set up the per-vcore structs to track outstanding syscalls */
        sysc_mgmt = malloc(sizeof(struct sysc_mgmt) * max_vcores());
        assert(sysc_mgmt);
        /* Set up the per-vcore structs to track outstanding syscalls */
        sysc_mgmt = malloc(sizeof(struct sysc_mgmt) * max_vcores());
        assert(sysc_mgmt);


@@ -481,10 +696,11 @@ void __attribute__((constructor)) pthread_lib_init(void)
        /* Get a block of pages for our per-vcore (but non-VCPD) ev_qs */
        mmap_block = (uintptr_t)mmap(0, PGSIZE * 2 * max_vcores(),
                                     PROT_WRITE | PROT_READ,
        /* Get a block of pages for our per-vcore (but non-VCPD) ev_qs */
        mmap_block = (uintptr_t)mmap(0, PGSIZE * 2 * max_vcores(),
                                     PROT_WRITE | PROT_READ,

-                                    MAP_POPULATE | MAP_ANONYMOUS, -1, 0);
+                                    MAP_POPULATE | MAP_ANONYMOUS | MAP_PRIVATE,
+                                    -1, 0);

        assert(mmap_block);
        assert(mmap_block);

-       /* Could be smarter and do this on demand (in case we don't actually want
-        * max_vcores()). */
+       /* Could be smarter and do this on demand (in case we don't actually
+        * want max_vcores()). */

        for (int i = 0; i < max_vcores(); i++) {
                /* Each vcore needs to point to a non-VCPD ev_q */
                sysc_mgmt[i].ev_q = get_eventq_raw();
        for (int i = 0; i < max_vcores(); i++) {
                /* Each vcore needs to point to a non-VCPD ev_q */
                sysc_mgmt[i].ev_q = get_eventq_raw();


@@ -501,8 +717,10 @@ void __attribute__((constructor)) pthread_lib_init(void)
 #endif 
 #if 0   /* One global ev_mbox, separate ev_q per vcore */
        struct event_mbox *sysc_mbox = malloc(sizeof(struct event_mbox));
 #endif 
 #if 0   /* One global ev_mbox, separate ev_q per vcore */
        struct event_mbox *sysc_mbox = malloc(sizeof(struct event_mbox));

-       uintptr_t two_pages = (uintptr_t)mmap(0, PGSIZE * 2, PROT_WRITE | PROT_READ,
-                                             MAP_POPULATE | MAP_ANONYMOUS, -1, 0);
+       uintptr_t two_pages = (uintptr_t)mmap(0, PGSIZE * 2, PROT_WRITE |
+                                             PROT_READ, MAP_POPULATE |
+                                             MAP_ANONYMOUS | MAP_PRIVATE, -1,
+                                             0);

        printd("Global ucq: %08p\n", &sysc_mbox->ev_msgs);
        assert(sysc_mbox);
        assert(two_pages);
        printd("Global ucq: %08p\n", &sysc_mbox->ev_msgs);
        assert(sysc_mbox);
        assert(two_pages);


@@ -517,21 +735,22 @@ void __attribute__((constructor)) pthread_lib_init(void)
                sysc_mgmt[i].ev_q->ev_mbox = sysc_mbox;
        }
 #endif
                sysc_mgmt[i].ev_q->ev_mbox = sysc_mbox;
        }
 #endif

-       /* Sched ops is set by 2ls_init */
-       uthread_2ls_init((struct uthread*)t, &pthread_sched_ops);
+       uthread_2ls_init((struct uthread*)t, pth_handle_syscall, NULL);

        atomic_init(&threads_total, 1);                 /* one for thread0 */
        atomic_init(&threads_total, 1);                 /* one for thread0 */

+       pre_fork_2ls = pth_pre_fork;
+       post_fork_2ls = pth_post_fork;

 }
 
 /* Make sure our scheduler runs inside an MCP rather than an SCP. */
 void pthread_mcp_init()
 {
        /* Prevent this from happening more than once. */
 }
 
 /* Make sure our scheduler runs inside an MCP rather than an SCP. */
 void pthread_mcp_init()
 {
        /* Prevent this from happening more than once. */

-       init_once_racy(return);
+       parlib_init_once_racy(return);

 
        uthread_mcp_init();
 
        uthread_mcp_init();

-       /* From here forward we are an MCP running on vcore 0. Could consider doing
-        * other pthread specific initialization based on knowing we are an mcp
-        * after this point. */
+       /* From here forward we are an MCP running on vcore 0. Could consider
+        * doing other pthread specific initialization based on knowing we are
+        * an mcp after this point. */

 }
 
 int __pthread_create(pthread_t *thread, const pthread_attr_t *attr,
 }
 
 int __pthread_create(pthread_t *thread, const pthread_attr_t *attr,


@@ -542,42 +761,35 @@ int __pthread_create(pthread_t *thread, const pthread_attr_t *attr,
        struct pthread_tcb *pthread;
        int ret;
 
        struct pthread_tcb *pthread;
        int ret;
 

-       /* For now, unconditionally become an mcp when creating a pthread (if not
-        * one already). This may change in the future once we support 2LSs in an
-        * SCP. */
+       /* For now, unconditionally become an mcp when creating a pthread (if
+        * not one already). This may change in the future once we support 2LSs
+        * in an SCP. */

        pthread_mcp_init();
 
        parent = (struct pthread_tcb*)current_uthread;
        ret = posix_memalign((void**)&pthread, __alignof__(struct pthread_tcb),
                             sizeof(struct pthread_tcb));
        assert(!ret);
        pthread_mcp_init();
 
        parent = (struct pthread_tcb*)current_uthread;
        ret = posix_memalign((void**)&pthread, __alignof__(struct pthread_tcb),
                             sizeof(struct pthread_tcb));
        assert(!ret);

-       memset(pthread, 0, sizeof(struct pthread_tcb)); /* aggressively 0 for bugs*/
+       /* aggressively 0 for bugs*/
+       memset(pthread, 0, sizeof(struct pthread_tcb));

        pthread->stacksize = PTHREAD_STACK_SIZE;        /* default */
        pthread->state = PTH_CREATED;
        pthread->id = get_next_pid();
        pthread->stacksize = PTHREAD_STACK_SIZE;        /* default */
        pthread->state = PTH_CREATED;
        pthread->id = get_next_pid();

-       pthread->detached = FALSE;                              /* default */
-       pthread->joiner = 0;
-       /* Might override these later, based on attr && EXPLICIT_SCHED */
-       pthread->sched_policy = parent->sched_policy;
-       pthread->sched_priority = parent->sched_priority;
+       pthread->fork_generation = fork_generation;

        SLIST_INIT(&pthread->cr_stack);
        /* Respect the attributes */
        if (attr) {
        SLIST_INIT(&pthread->cr_stack);
        /* Respect the attributes */
        if (attr) {

-               if (attr->stacksize)                                    /* don't set a 0 stacksize */
+               if (attr->stacksize)    /* don't set a 0 stacksize */

                        pthread->stacksize = attr->stacksize;
                if (attr->detachstate == PTHREAD_CREATE_DETACHED)
                        pthread->stacksize = attr->stacksize;
                if (attr->detachstate == PTHREAD_CREATE_DETACHED)

-                       pthread->detached = TRUE;
-               if (attr->sched_inherit == PTHREAD_EXPLICIT_SCHED) {
-                       pthread->sched_policy = attr->sched_policy;
-                       pthread->sched_priority = attr->sched_priority;
-               }
+                       uth_attr.detached = TRUE;

        }
        /* allocate a stack */
        if (__pthread_allocate_stack(pthread))
                printf("We're fucked\n");
        /* Set the u_tf to start up in __pthread_run, which will call the real
        }
        /* allocate a stack */
        if (__pthread_allocate_stack(pthread))
                printf("We're fucked\n");
        /* Set the u_tf to start up in __pthread_run, which will call the real

-        * start_routine and pass it the arg.  Note those aren't set until later in
-        * pthread_create(). */
+        * start_routine and pass it the arg.  Note those aren't set until later
+        * in pthread_create(). */

        init_user_ctx(&pthread->uthread.u_ctx, (uintptr_t)&__pthread_run,
                      (uintptr_t)(pthread->stacktop));
        pthread->start_routine = start_routine;
        init_user_ctx(&pthread->uthread.u_ctx, (uintptr_t)&__pthread_run,
                      (uintptr_t)(pthread->stacktop));
        pthread->start_routine = start_routine;


@@ -610,100 +822,20 @@ void __pthread_generic_yield(struct pthread_tcb *pthread)
        mcs_pdr_unlock(&queue_lock);
 }
 
        mcs_pdr_unlock(&queue_lock);
 }
 

-/* Callback/bottom half of join, called from __uthread_yield (vcore context).
- * join_target is who we are trying to join on (and who is calling exit). */
-static void __pth_join_cb(struct uthread *uthread, void *arg)
-{
-       struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
-       struct pthread_tcb *join_target = (struct pthread_tcb*)arg;
-       struct pthread_tcb *temp_pth = 0;
-       __pthread_generic_yield(pthread);
-       /* We're trying to join, yield til we get woken up */
-       pthread->state = PTH_BLK_JOINING;       /* could do this front-side */
-       /* Put ourselves in the join target's joiner slot.  If we get anything back,
-        * we lost the race and need to wake ourselves.  Syncs with __pth_exit_cb.*/
-       temp_pth = atomic_swap_ptr((void**)&join_target->joiner, pthread);
-       /* After that atomic swap, the pthread might be woken up (if it succeeded),
-        * so don't touch pthread again after that (this following if () is okay).*/
-       if (temp_pth) {         /* temp_pth != 0 means they exited first */
-               assert(temp_pth == join_target);        /* Sanity */
-               /* wake ourselves, not the exited one! */
-               printd("[pth] %08p already exit, rewaking ourselves, joiner %08p\n",
-                      temp_pth, pthread);
-               pth_thread_runnable(uthread);   /* wake ourselves */
-       }
-}
-

 int pthread_join(struct pthread_tcb *join_target, void **retval)
 {
 int pthread_join(struct pthread_tcb *join_target, void **retval)
 {

-       /* Not sure if this is the right semantics.  There is a race if we deref
-        * join_target and he is already freed (which would have happened if he was
-        * detached. */
-       if (join_target->detached) {
-               printf("[pthread] trying to join on a detached pthread");
-               return -1;
-       }
-       /* See if it is already done, to avoid the pain of a uthread_yield() (the
-        * early check is an optimization, pth_thread_yield() handles the race). */
-       if (!join_target->joiner) {
-               uthread_yield(TRUE, __pth_join_cb, join_target);
-               /* When we return/restart, the thread will be done */
-       } else {
-               assert(join_target->joiner == join_target);     /* sanity check */
-       }
-       if (retval)
-               *retval = join_target->retval;
-       free(join_target);
+       uthread_join((struct uthread*)join_target, retval);

        return 0;
 }
 
        return 0;
 }
 

-/* Callback/bottom half of exit.  Syncs with __pth_join_cb.  Here's how it
- * works: the slot for joiner is initially 0.  Joiners try to swap themselves
- * into that spot.  Exiters try to put 'themselves' into it.  Whoever gets 0
- * back won the race.  If the exiter lost the race, it must wake up the joiner
- * (which was the value from temp_pth).  If the joiner lost the race, it must
- * wake itself up, and for sanity reasons can ensure the value from temp_pth is
- * the join target). */
-static void __pth_exit_cb(struct uthread *uthread, void *junk)
-{
-       struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
-       struct pthread_tcb *temp_pth = 0;
-       __pthread_generic_yield(pthread);
-       /* Catch some bugs */
-       pthread->state = PTH_EXITING;
-       /* Destroy the pthread */
-       uthread_cleanup(uthread);
-       /* Cleanup, mirroring pthread_create() */
-       __pthread_free_stack(pthread);
-       /* TODO: race on detach state (see join) */
-       if (pthread->detached) {
-               free(pthread);
-       } else {
-               /* See if someone is joining on us.  If not, we're done (and the
-                * joiner will wake itself when it saw us there instead of 0). */
-               temp_pth = atomic_swap_ptr((void**)&pthread->joiner, pthread);
-               if (temp_pth) {
-                       /* they joined before we exited, we need to wake them */
-                       printd("[pth] %08p exiting, waking joiner %08p\n",
-                              pthread, temp_pth);
-                       pth_thread_runnable((struct uthread*)temp_pth);
-               }
-       }
-       /* If we were the last pthread, we exit for the whole process.  Keep in mind
-        * that thread0 is counted in this, so this will only happen if that thread
-        * calls pthread_exit(). */
-       if ((atomic_fetch_and_add(&threads_total, -1) == 1))
-               exit(0);
-}
-

 static inline void pthread_exit_no_cleanup(void *ret)
 {
        struct pthread_tcb *pthread = pthread_self();
 static inline void pthread_exit_no_cleanup(void *ret)
 {
        struct pthread_tcb *pthread = pthread_self();

-       pthread->retval = ret;
-       destroy_dtls();
+

        while (SLIST_FIRST(&pthread->cr_stack))
                pthread_cleanup_pop(FALSE);
        while (SLIST_FIRST(&pthread->cr_stack))
                pthread_cleanup_pop(FALSE);

-       uthread_yield(FALSE, __pth_exit_cb, 0);
+       destroy_dtls();
+       uth_2ls_thread_exit(ret);

 }
 
 void pthread_exit(void *ret)
 }
 
 void pthread_exit(void *ret)


@@ -714,22 +846,10 @@ void pthread_exit(void *ret)
        pthread_exit_no_cleanup(ret);
 }
 
        pthread_exit_no_cleanup(ret);
 }
 

-/* Callback/bottom half of yield.  For those writing these pth callbacks, the
- * minimum is call generic, set state (communicate with runnable), then do
- * something that causes it to be runnable in the future (or right now). */
-static void __pth_yield_cb(struct uthread *uthread, void *junk)
-{
-       struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
-       __pthread_generic_yield(pthread);
-       pthread->state = PTH_BLK_YIELDING;
-       /* just immediately restart it */
-       pth_thread_runnable(uthread);
-}
-

 /* Cooperative yielding of the processor, to allow other threads to run */
 int pthread_yield(void)
 {
 /* Cooperative yielding of the processor, to allow other threads to run */
 int pthread_yield(void)
 {

-       uthread_yield(TRUE, __pth_yield_cb, 0);
+       uthread_sched_yield();

        return 0;
 }
 
        return 0;
 }
 


@@ -744,6 +864,7 @@ void pthread_cleanup_push(void (*routine)(void *), void *arg)
 {
        struct pthread_tcb *p = pthread_self();
        struct pthread_cleanup_routine *r = malloc(sizeof(*r));
 {
        struct pthread_tcb *p = pthread_self();
        struct pthread_cleanup_routine *r = malloc(sizeof(*r));

+

        r->routine = routine;
        r->arg = arg;
        SLIST_INSERT_HEAD(&p->cr_stack, r, cr_next);
        r->routine = routine;
        r->arg = arg;
        SLIST_INSERT_HEAD(&p->cr_stack, r, cr_next);


@@ -753,6 +874,7 @@ void pthread_cleanup_pop(int execute)
 {
        struct pthread_tcb *p = pthread_self();
        struct pthread_cleanup_routine *r = SLIST_FIRST(&p->cr_stack);
 {
        struct pthread_tcb *p = pthread_self();
        struct pthread_cleanup_routine *r = SLIST_FIRST(&p->cr_stack);

+

        if (r) {
                SLIST_REMOVE_HEAD(&p->cr_stack, cr_next);
                if (execute)
        if (r) {
                SLIST_REMOVE_HEAD(&p->cr_stack, cr_next);
                if (execute)


@@ -761,15 +883,15 @@ void pthread_cleanup_pop(int execute)
        }
 }
 
        }
 }
 

-int pthread_mutexattr_init(pthread_mutexattr_t* attr)
+int pthread_mutexattr_init(pthread_mutexattr_t *attr)

 {
 {

-  attr->type = PTHREAD_MUTEX_DEFAULT;
-  return 0;
+       attr->type = PTHREAD_MUTEX_DEFAULT;
+       return 0;

 }
 
 }
 

-int pthread_mutexattr_destroy(pthread_mutexattr_t* attr)
+int pthread_mutexattr_destroy(pthread_mutexattr_t *attr)

 {
 {

-  return 0;
+       return 0;

 }
 
 int pthread_attr_setdetachstate(pthread_attr_t *__attr, int __detachstate)
 }
 
 int pthread_attr_setdetachstate(pthread_attr_t *__attr, int __detachstate)


@@ -778,141 +900,150 @@ int pthread_attr_setdetachstate(pthread_attr_t *__attr, int __detachstate)
        return 0;
 }
 
        return 0;
 }
 

-int pthread_mutexattr_gettype(const pthread_mutexattr_t* attr, int* type)
+int pthread_mutexattr_gettype(const pthread_mutexattr_t *attr, int *type)

 {
 {

-  *type = attr ? attr->type : PTHREAD_MUTEX_DEFAULT;
-  return 0;
+       *type = attr ? attr->type : PTHREAD_MUTEX_DEFAULT;
+       return 0;

 }
 
 }
 

-int pthread_mutexattr_settype(pthread_mutexattr_t* attr, int type)
+static bool __pthread_mutex_type_ok(int type)

 {
 {

-  if(type != PTHREAD_MUTEX_NORMAL)
-    return EINVAL;
-  attr->type = type;
-  return 0;
+       switch (type) {
+       case PTHREAD_MUTEX_NORMAL:
+       case PTHREAD_MUTEX_RECURSIVE:
+               return TRUE;
+       }
+       return FALSE;

 }
 
 }
 

-int pthread_mutex_init(pthread_mutex_t* m, const pthread_mutexattr_t* attr)
+int pthread_mutexattr_settype(pthread_mutexattr_t *attr, int type)

 {
 {

-  m->attr = attr;
-  atomic_init(&m->lock, 0);
-  return 0;
+       if (!__pthread_mutex_type_ok(type))
+               return EINVAL;
+       attr->type = type;
+       return 0;

 }
 
 }
 

-/* Helper for spinning sync, returns TRUE if it is okay to keep spinning.
- *
- * Alternatives include:
- *             old_count <= num_vcores() (barrier code, pass in old_count as *state, 
- *                                        but this only works if every awake pthread
- *                                        will belong to the barrier).
- *             just spin for a bit       (use *state to track spins)
- *             FALSE                     (always is safe)
- *             etc...
- * 'threads_ready' isn't too great since sometimes it'll be non-zero when it is
- * about to become 0.  We really want "I have no threads waiting to run that
- * aren't going to run on their on unless this core yields instead of spins". */
-/* TODO: consider making this a 2LS op */
-static inline bool safe_to_spin(unsigned int *state)
+int pthread_mutex_init(pthread_mutex_t *m, const pthread_mutexattr_t *attr)

 {
 {

-       return !threads_ready;
+       if (attr) {
+               if (!__pthread_mutex_type_ok(attr->type))
+                       return EINVAL;
+               m->type = attr->type;
+       } else {
+               m->type = PTHREAD_MUTEX_NORMAL;
+       }
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               uth_mutex_init(&m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               uth_recurse_mutex_init(&m->r_mtx);
+               break;
+       }
+       return 0;

 }
 
 }
 

-/* Set *spun to 0 when calling this the first time.  It will yield after 'spins'
- * calls.  Use this for adaptive mutexes and such. */
-static inline void spin_to_sleep(unsigned int spins, unsigned int *spun)
+int pthread_mutex_lock(pthread_mutex_t *m)

 {
 {

-       if ((*spun)++ == spins) {
-               pthread_yield();
-               *spun = 0;
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               uth_mutex_lock(&m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               uth_recurse_mutex_lock(&m->r_mtx);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);

        }
        }

+       return 0;

 }
 
 }
 

-int pthread_mutex_lock(pthread_mutex_t* m)
+int pthread_mutex_trylock(pthread_mutex_t *m)

 {
 {

-       unsigned int spinner = 0;
-       while(pthread_mutex_trylock(m))
-               while(*(volatile size_t*)&m->lock) {
-                       cpu_relax();
-                       spin_to_sleep(PTHREAD_MUTEX_SPINS, &spinner);
-               }
-       /* normally we'd need a wmb() and a wrmb() after locking, but the
-        * atomic_swap handles the CPU mb(), so just a cmb() is necessary. */
-       cmb();
-       return 0;
+       bool got_it;
+
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               got_it = uth_mutex_trylock(&m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               got_it = uth_recurse_mutex_trylock(&m->r_mtx);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return got_it ? 0 : EBUSY;

 }
 
 }
 

-int pthread_mutex_trylock(pthread_mutex_t* m)
+int pthread_mutex_unlock(pthread_mutex_t *m)

 {
 {

-  return atomic_swap(&m->lock, 1) == 0 ? 0 : EBUSY;
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               uth_mutex_unlock(&m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               uth_recurse_mutex_unlock(&m->r_mtx);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return 0;

 }
 
 }
 

-int pthread_mutex_unlock(pthread_mutex_t* m)
+int pthread_mutex_destroy(pthread_mutex_t *m)

 {
 {

-  /* keep reads and writes inside the protected region */
-  rwmb();
-  wmb();
-  atomic_set(&m->lock, 0);
-  return 0;
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               uth_mutex_destroy(&m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               uth_recurse_mutex_destroy(&m->r_mtx);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return 0;

 }
 
 }
 

-int pthread_mutex_destroy(pthread_mutex_t* m)
+int pthread_mutex_timedlock(pthread_mutex_t *m, const struct timespec *abstime)

 {
 {

-  return 0;
+       bool got_it;
+
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               got_it = uth_mutex_timed_lock(&m->mtx, abstime);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               got_it = uth_recurse_mutex_timed_lock(&m->r_mtx, abstime);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return got_it ? 0 : ETIMEDOUT;

 }
 
 int pthread_cond_init(pthread_cond_t *c, const pthread_condattr_t *a)
 {
 }
 
 int pthread_cond_init(pthread_cond_t *c, const pthread_condattr_t *a)
 {

-       SLIST_INIT(&c->waiters);
-       spin_pdr_init(&c->spdr_lock);

        if (a) {
        if (a) {

-               c->attr_pshared = a->pshared;
-               c->attr_clock = a->clock;
-       } else {
-               c->attr_pshared = PTHREAD_PROCESS_PRIVATE;
-               c->attr_clock = 0;
+               if (a->pshared != PTHREAD_PROCESS_PRIVATE)
+                       fprintf(stderr,
+                               "pthreads only supports private condvars");
+               /* We also ignore clock_id */

        }
        }

+       uth_cond_var_init(c);

        return 0;
 }
 
 int pthread_cond_destroy(pthread_cond_t *c)
 {
        return 0;
 }
 
 int pthread_cond_destroy(pthread_cond_t *c)
 {

+       uth_cond_var_destroy(c);

        return 0;
 }
 
        return 0;
 }
 

-static void swap_slists(struct pthread_list *a, struct pthread_list *b)
-{
-       struct pthread_list temp;
-       temp = *a;
-       *a = *b;
-       *b = temp;
-}
-
-static void wake_slist(struct pthread_list *to_wake)
-{
-       unsigned int nr_woken = 0;      /* assuming less than 4 bil threads */
-       struct pthread_tcb *pthread_i, *pth_temp;
-       /* Amortize the lock grabbing over all restartees */
-       mcs_pdr_lock(&queue_lock);
-       /* Do the work of pth_thread_runnable().  We're in uth context here, but I
-        * think it's okay.  When we need to (when locking) we drop into VC ctx, as
-        * far as the kernel and other cores are concerned. */
-       SLIST_FOREACH_SAFE(pthread_i, to_wake, sl_next, pth_temp) {
-               pthread_i->state = PTH_RUNNABLE;
-               nr_woken++;
-               TAILQ_INSERT_TAIL(&ready_queue, pthread_i, tq_next);
-       }
-       threads_ready += nr_woken;
-       mcs_pdr_unlock(&queue_lock);
-       vcore_request_more(threads_ready);
-}
-

 int pthread_cond_broadcast(pthread_cond_t *c)
 {
 int pthread_cond_broadcast(pthread_cond_t *c)
 {

-       struct pthread_list restartees = SLIST_HEAD_INITIALIZER(restartees);
-       spin_pdr_lock(&c->spdr_lock);
-       swap_slists(&restartees, &c->waiters);
-       spin_pdr_unlock(&c->spdr_lock);
-       wake_slist(&restartees);
+       uth_cond_var_broadcast(c);

        return 0;
 }
 
        return 0;
 }
 


@@ -920,50 +1051,41 @@ int pthread_cond_broadcast(pthread_cond_t *c)
  * already. */
 int pthread_cond_signal(pthread_cond_t *c)
 {
  * already. */
 int pthread_cond_signal(pthread_cond_t *c)
 {

-       struct pthread_tcb *pthread;
-       spin_pdr_lock(&c->spdr_lock);
-       pthread = SLIST_FIRST(&c->waiters);
-       if (!pthread) {
-               spin_pdr_unlock(&c->spdr_lock);
-               return 0;
-       }
-       SLIST_REMOVE_HEAD(&c->waiters, sl_next);
-       spin_pdr_unlock(&c->spdr_lock);
-       pth_thread_runnable((struct uthread*)pthread);
+       uth_cond_var_signal(c);

        return 0;
 }
 
        return 0;
 }
 

-/* Communicate btw cond_wait and its callback */
-struct cond_junk {
-       pthread_cond_t                          *c;
-       pthread_mutex_t                         *m;
-};
-
-/* Callback/bottom half of cond wait.  For those writing these pth callbacks,
- * the minimum is call generic, set state (communicate with runnable), then do
- * something that causes it to be runnable in the future (or right now). */
-static void __pth_wait_cb(struct uthread *uthread, void *junk)
+int pthread_cond_wait(pthread_cond_t *c, pthread_mutex_t *m)

 {
 {

-       struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;
-       pthread_cond_t *c = ((struct cond_junk*)junk)->c;
-       pthread_mutex_t *m = ((struct cond_junk*)junk)->m;
-       /* this removes us from the active list; we can reuse next below */
-       __pthread_generic_yield(pthread);
-       pthread->state = PTH_BLK_MUTEX;
-       spin_pdr_lock(&c->spdr_lock);
-       SLIST_INSERT_HEAD(&c->waiters, pthread, sl_next);
-       spin_pdr_unlock(&c->spdr_lock);
-       pthread_mutex_unlock(m);
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               uth_cond_var_wait(c, &m->mtx);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               uth_cond_var_wait_recurse(c, &m->r_mtx);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return 0;

 }
 
 }
 

-int pthread_cond_wait(pthread_cond_t *c, pthread_mutex_t *m)
+int pthread_cond_timedwait(pthread_cond_t *c, pthread_mutex_t *m,
+                           const struct timespec *abstime)

 {
 {

-       struct cond_junk local_junk;
-       local_junk.c = c;
-       local_junk.m = m;
-       uthread_yield(TRUE, __pth_wait_cb, &local_junk);
-       pthread_mutex_lock(m);
-       return 0;
+       bool got_it;
+
+       switch (m->type) {
+       case PTHREAD_MUTEX_NORMAL:
+               got_it = uth_cond_var_timed_wait(c, &m->mtx, abstime);
+               break;
+       case PTHREAD_MUTEX_RECURSIVE:
+               got_it = uth_cond_var_timed_wait_recurse(c, &m->r_mtx, abstime);
+               break;
+       default:
+               panic("Bad pth mutex type %d!", m->type);
+       }
+       return got_it ? 0 : ETIMEDOUT;

 }
 
 int pthread_condattr_init(pthread_condattr_t *a)
 }
 
 int pthread_condattr_init(pthread_condattr_t *a)


@@ -1008,21 +1130,66 @@ int pthread_condattr_setclock(pthread_condattr_t *attr, clockid_t clock_id)
        return 0;
 }
 
        return 0;
 }
 

-pthread_t pthread_self()
+int pthread_rwlock_init(pthread_rwlock_t *rwl, const pthread_rwlockattr_t *a)

 {
 {

-  return (struct pthread_tcb*)current_uthread;
+       uth_rwlock_init(rwl);
+       return 0;
+}
+
+int pthread_rwlock_destroy(pthread_rwlock_t *rwl)
+{
+       uth_rwlock_destroy(rwl);
+       return 0;
+}
+
+int pthread_rwlock_rdlock(pthread_rwlock_t *rwl)
+{
+       uth_rwlock_rdlock(rwl);
+       return 0;
+}
+
+int pthread_rwlock_tryrdlock(pthread_rwlock_t *rwl)
+{
+       return uth_rwlock_try_rdlock(rwl) ? 0 : EBUSY;
+}
+
+int pthread_rwlock_wrlock(pthread_rwlock_t *rwl)
+{
+       uth_rwlock_wrlock(rwl);
+       return 0;
+}
+
+int pthread_rwlock_trywrlock(pthread_rwlock_t *rwl)
+{
+       return uth_rwlock_try_wrlock(rwl) ? 0 : EBUSY;
+}
+
+int pthread_rwlock_unlock(pthread_rwlock_t *rwl)
+{
+       uth_rwlock_unlock(rwl);
+       return 0;
+}
+
+pthread_t pthread_self(void)
+{
+       return (struct pthread_tcb*)uthread_self();

 }
 
 int pthread_equal(pthread_t t1, pthread_t t2)
 {
 }
 
 int pthread_equal(pthread_t t1, pthread_t t2)
 {

-  return t1 == t2;
+       return t1 == t2;

 }
 
 }
 

-int pthread_once(pthread_once_t* once_control, void (*init_routine)(void))
+int pthread_once(pthread_once_t *once_control, void (*init_routine)(void))

 {
 {

-  if (atomic_swap_u32(once_control, 1) == 0)
-    init_routine();
-  return 0;
+       /* pthread_once's init routine doesn't take an argument, like parlibs.
+        * This means the func will be run with an argument passed to it, but
+        * it'll be ignored. */
+       parlib_run_once(once_control, (void (*)(void *))init_routine, NULL);
+       /* The return for pthread_once isn't an error from the function, it's
+        * just an overall error.  Note pthread's init_routine() has no return
+        * value. */
+       return 0;

 }
 
 int pthread_barrier_init(pthread_barrier_t *b,
 }
 
 int pthread_barrier_init(pthread_barrier_t *b,


@@ -1032,37 +1199,52 @@ int pthread_barrier_init(pthread_barrier_t *b,
        b->sense = 0;
        atomic_set(&b->count, count);
        spin_pdr_init(&b->lock);
        b->sense = 0;
        atomic_set(&b->count, count);
        spin_pdr_init(&b->lock);

-       SLIST_INIT(&b->waiters);
+       __uth_sync_init(&b->waiters);

        b->nr_waiters = 0;
        return 0;
 }
 
 struct barrier_junk {
        b->nr_waiters = 0;
        return 0;
 }
 
 struct barrier_junk {

-       pthread_barrier_t                               *b;
-       int                                                             ls;
+       pthread_barrier_t               *b;
+       int                             ls;

 };
 
 };
 

+/* Helper for spinning sync, returns TRUE if it is okay to keep spinning.
+ *
+ * Alternatives include:
+ *     old_count <= num_vcores() (barrier code, pass in old_count as *state,
+ *                                but this only works if every awake pthread
+ *                                will belong to the barrier).
+ *     just spin for a bit       (use *state to track spins)
+ *     FALSE                     (always is safe)
+ *     etc...
+ * 'threads_ready' isn't too great since sometimes it'll be non-zero when it is
+ * about to become 0.  We really want "I have no threads waiting to run that
+ * aren't going to run on their on unless this core yields instead of spins". */
+/* TODO: consider making this a 2LS op */
+static inline bool safe_to_spin(unsigned int *state)
+{
+       return (*state)++ % PTHREAD_BARRIER_SPINS;
+}
+

 /* Callback/bottom half of barrier. */
 static void __pth_barrier_cb(struct uthread *uthread, void *junk)
 {
 /* Callback/bottom half of barrier. */
 static void __pth_barrier_cb(struct uthread *uthread, void *junk)
 {

-       struct pthread_tcb *pthread = (struct pthread_tcb*)uthread;

        pthread_barrier_t *b = ((struct barrier_junk*)junk)->b;
        int ls = ((struct barrier_junk*)junk)->ls;
        pthread_barrier_t *b = ((struct barrier_junk*)junk)->b;
        int ls = ((struct barrier_junk*)junk)->ls;

-       /* Removes from active list, we can reuse.  must also restart */
-       __pthread_generic_yield(pthread);
+
+       uthread_has_blocked(uthread, UTH_EXT_BLK_MUTEX);

        /* TODO: if we used a trylock, we could bail as soon as we see sense */
        spin_pdr_lock(&b->lock);
        /* TODO: if we used a trylock, we could bail as soon as we see sense */
        spin_pdr_lock(&b->lock);

-       /* If sense is ls (our free value), we lost the race and shouldn't sleep */
+       /* If sense is ls (our free value), we lost the race and shouldn't sleep
+        */

        if (b->sense == ls) {
        if (b->sense == ls) {

-               /* TODO: i'd like to fast-path the wakeup, skipping pth_runnable */
-               pthread->state = PTH_BLK_YIELDING;      /* not sure which state for this */

                spin_pdr_unlock(&b->lock);
                spin_pdr_unlock(&b->lock);

-               pth_thread_runnable(uthread);
+               uthread_runnable(uthread);

                return;
        }
        /* otherwise, we sleep */
                return;
        }
        /* otherwise, we sleep */

-       pthread->state = PTH_BLK_MUTEX; /* TODO: consider ignoring this */
-       SLIST_INSERT_HEAD(&b->waiters, pthread, sl_next);
+       __uth_sync_enqueue(uthread, &b->waiters);

        b->nr_waiters++;
        spin_pdr_unlock(&b->lock);
 }
        b->nr_waiters++;
        spin_pdr_unlock(&b->lock);
 }


@@ -1084,22 +1266,22 @@ static void __pth_barrier_cb(struct uthread *uthread, void *junk)
 int pthread_barrier_wait(pthread_barrier_t *b)
 {
        unsigned int spin_state = 0;
 int pthread_barrier_wait(pthread_barrier_t *b)
 {
        unsigned int spin_state = 0;

-       int ls = !b->sense;     /* when b->sense is the value we read, then we're free*/
-       struct pthread_list restartees = SLIST_HEAD_INITIALIZER(restartees);
-       struct pthread_tcb *pthread_i;
+       /* when b->sense is the value we read, then we're free*/
+       int ls = !b->sense;
+       uth_sync_t restartees;
+       struct uthread *uth_i;

        struct barrier_junk local_junk;
        
        long old_count = atomic_fetch_and_add(&b->count, -1);
 
        if (old_count == 1) {
        struct barrier_junk local_junk;
        
        long old_count = atomic_fetch_and_add(&b->count, -1);
 
        if (old_count == 1) {

-               printd("Thread %d is last to hit the barrier, resetting...\n",
-                      pthread_self()->id);
-               /* TODO: we might want to grab the lock right away, so a few short
-                * circuit faster? */
+               /* TODO: we might want to grab the lock right away, so a few
+                * short circuit faster? */

                atomic_set(&b->count, b->total_threads);
                atomic_set(&b->count, b->total_threads);

-               /* we still need to maintain ordering btw count and sense, in case
-                * another thread doesn't sleep (if we wrote sense first, they could
-                * break out, race around, and muck with count before it is time) */
+               /* we still need to maintain ordering btw count and sense, in
+                * case another thread doesn't sleep (if we wrote sense first,
+                * they could break out, race around, and muck with count before
+                * it is time) */

                /* wmb(); handled by the spin lock */
                spin_pdr_lock(&b->lock);
                /* Sense is only protected in addition to decisions to sleep */
                /* wmb(); handled by the spin lock */
                spin_pdr_lock(&b->lock);
                /* Sense is only protected in addition to decisions to sleep */


@@ -1109,13 +1291,15 @@ int pthread_barrier_wait(pthread_barrier_t *b)
                        spin_pdr_unlock(&b->lock);
                        return PTHREAD_BARRIER_SERIAL_THREAD;
                }
                        spin_pdr_unlock(&b->lock);
                        return PTHREAD_BARRIER_SERIAL_THREAD;
                }

-               swap_slists(&restartees, &b->waiters);
+               __uth_sync_init(&restartees);
+               __uth_sync_swap(&restartees, &b->waiters);

                b->nr_waiters = 0;
                spin_pdr_unlock(&b->lock);
                b->nr_waiters = 0;
                spin_pdr_unlock(&b->lock);

-               wake_slist(&restartees);
+               __uth_sync_wake_all(&restartees);

                return PTHREAD_BARRIER_SERIAL_THREAD;
        } else {
                return PTHREAD_BARRIER_SERIAL_THREAD;
        } else {

-               /* Spin if there are no other threads to run.  No sense sleeping */
+               /* Spin if there are no other threads to run.  No sense sleeping
+                */

                do {
                        if (b->sense == ls)
                                return 0;
                do {
                        if (b->sense == ls)
                                return 0;


@@ -1133,16 +1317,15 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 
 int pthread_barrier_destroy(pthread_barrier_t *b)
 {
 
 int pthread_barrier_destroy(pthread_barrier_t *b)
 {

-       assert(SLIST_EMPTY(&b->waiters));

        assert(!b->nr_waiters);
        assert(!b->nr_waiters);

+       __uth_sync_destroy(&b->waiters);

        /* Free any locks (if we end up using an MCS) */
        return 0;
 }
 
 int pthread_detach(pthread_t thread)
 {
        /* Free any locks (if we end up using an MCS) */
        return 0;
 }
 
 int pthread_detach(pthread_t thread)
 {

-       /* TODO: race on this state.  Someone could be trying to join now */
-       thread->detached = TRUE;
+       uthread_detach((struct uthread*)thread);

        return 0;
 }
 
        return 0;
 }
 


@@ -1192,26 +1375,16 @@ int pthread_setspecific(pthread_key_t key, const void *value)
 }
 
 
 }
 
 

-/* Scheduling Stuff */
-
-static bool policy_is_supported(int policy)
-{
-       /* As our scheduler changes, we can add more policies here */
-       switch (policy) {
-               case SCHED_FIFO:
-                       return TRUE;
-               default:
-                       return FALSE;
-       }
-}
+/* Scheduling Stuff.  Actually, these don't tell the 2LS anything - they just
+ * pretend to muck with attrs and params, as expected by pthreads apps. */

 
 int pthread_attr_setschedparam(pthread_attr_t *attr,
                                const struct sched_param *param)
 {
        /* The set of acceptable priorities are based on the scheduling policy.
         * We'll just accept any old number, since we might not know the policy
 
 int pthread_attr_setschedparam(pthread_attr_t *attr,
                                const struct sched_param *param)
 {
        /* The set of acceptable priorities are based on the scheduling policy.
         * We'll just accept any old number, since we might not know the policy

-        * yet.  I didn't see anything in the man pages saying attr had to have a
-        * policy set before setting priority. */
+        * yet.  I didn't see anything in the man pages saying attr had to have
+        * a policy set before setting priority. */

        attr->sched_priority = param->sched_priority;
        return 0;
 }
        attr->sched_priority = param->sched_priority;
        return 0;
 }


@@ -1225,8 +1398,6 @@ int pthread_attr_getschedparam(pthread_attr_t *attr,
 
 int pthread_attr_setschedpolicy(pthread_attr_t *attr, int policy)
 {
 
 int pthread_attr_setschedpolicy(pthread_attr_t *attr, int policy)
 {

-       if (!policy_is_supported(policy))
-               return -EINVAL;

        attr->sched_policy = policy;
        return 0;
 }
        attr->sched_policy = policy;
        return 0;
 }


@@ -1276,40 +1447,14 @@ int pthread_attr_getinheritsched(const pthread_attr_t *attr,
 int pthread_setschedparam(pthread_t thread, int policy,
                            const struct sched_param *param)
 {
 int pthread_setschedparam(pthread_t thread, int policy,
                            const struct sched_param *param)
 {

-       if (!policy_is_supported(policy))
-               return -EINVAL;
-       thread->sched_policy = policy;
-       /* We actually could check if the priority falls in the range of the
-        * specified policy here, since we have both policy and priority. */
-       thread->sched_priority = param->sched_priority;

        return 0;
 }
 
 int pthread_getschedparam(pthread_t thread, int *policy,
                            struct sched_param *param)
 {
        return 0;
 }
 
 int pthread_getschedparam(pthread_t thread, int *policy,
                            struct sched_param *param)
 {

-       *policy = thread->sched_policy;
-       param->sched_priority = thread->sched_priority;
+       /* Faking {FIFO, 0}.  It's up to the 2LS to do whatever it wants. */
+       *policy = SCHED_FIFO;
+       param->sched_priority = 0;

        return 0;
 }
        return 0;
 }

-
-
-/* Unsupported Stuff */
-
-int pthread_mutex_timedlock (pthread_mutex_t *__restrict __mutex,
-                                       const struct timespec *__restrict
-                                       __abstime)
-{
-       fprintf(stderr, "Unsupported %s!", __FUNCTION__);
-       abort();
-       return -1;
-}
-
-int pthread_cond_timedwait (pthread_cond_t *__restrict __cond,
-                                  pthread_mutex_t *__restrict __mutex,
-                                  const struct timespec *__restrict __abstime)
-{
-       fprintf(stderr, "Unsupported %s!", __FUNCTION__);
-       abort();
-       return -1;
-}