xnu-7195.60.75.tar.gz

[apple/xnu.git] / osfmk / kern / syscall_subr.c

diff --git a/osfmk/kern/syscall_subr.c b/osfmk/kern/syscall_subr.c
index 3daf1ec3874eccb8884bcb60508ff74ba14ffe8b..4d65fe2ae3c3bdb4ca5a0ba2b7089041ec3d3495 100644 (file)
--- a/osfmk/kern/syscall_subr.c
+++ b/osfmk/kern/syscall_subr.c
@@ -1,8 +1,8 @@
 /*
 /*

- * Copyright (c) 2000-2009 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2017 Apple Inc. All rights reserved.

  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@

- * 
+ *

  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in


@@ -11,10 +11,10 @@
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.

- * 
+ *

  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.

- * 
+ *

  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,


@@ -22,39 +22,37 @@
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.

- * 
+ *

  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
  */
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
  */

-/* 
+/*

  * Mach Operating System
  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  * All Rights Reserved.
  * Mach Operating System
  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  * All Rights Reserved.

- * 
+ *

  * Permission to use, copy, modify and distribute this software and its
  * documentation is hereby granted, provided that both the copyright
  * notice and this permission notice appear in all copies of the
  * software, derivative works or modified versions, and any portions
  * thereof, and that both notices appear in supporting documentation.
  * Permission to use, copy, modify and distribute this software and its
  * documentation is hereby granted, provided that both the copyright
  * notice and this permission notice appear in all copies of the
  * software, derivative works or modified versions, and any portions
  * thereof, and that both notices appear in supporting documentation.

- * 
+ *

  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.

- * 
+ *

  * Carnegie Mellon requests users of this software to return to
  * Carnegie Mellon requests users of this software to return to

- * 
+ *

  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  *  School of Computer Science
  *  Carnegie Mellon University
  *  Pittsburgh PA 15213-3890
  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  *  School of Computer Science
  *  Carnegie Mellon University
  *  Pittsburgh PA 15213-3890

- * 
+ *

  * any improvements or extensions that they make and grant Carnegie Mellon
  * the rights to redistribute these changes.
  */
  * any improvements or extensions that they make and grant Carnegie Mellon
  * the rights to redistribute these changes.
  */

-/*
- */

 
 #include <mach/boolean.h>
 #include <mach/thread_switch.h>
 
 #include <mach/boolean.h>
 #include <mach/thread_switch.h>


@@ -68,28 +66,29 @@
 #include <kern/spl.h>
 #include <kern/task.h>
 #include <kern/thread.h>
 #include <kern/spl.h>
 #include <kern/task.h>
 #include <kern/thread.h>

+#include <kern/policy_internal.h>
+

 #include <mach/policy.h>
 
 #include <kern/syscall_subr.h>
 #include <mach/mach_host_server.h>
 #include <mach/mach_syscalls.h>
 #include <mach/policy.h>
 
 #include <kern/syscall_subr.h>
 #include <mach/mach_host_server.h>
 #include <mach/mach_syscalls.h>

+#include <sys/kdebug.h>
+#include <kern/ast.h>

 
 

-
-#ifdef MACH_BSD
-extern void workqueue_thread_yielded(void);
-#endif /* MACH_BSD */
-
+static void thread_depress_abstime(uint64_t interval);
+static void thread_depress_ms(mach_msg_timeout_t interval);

 
 /* Called from commpage to take a delayed preemption when exiting
  * the "Preemption Free Zone" (PFZ).
  */
 kern_return_t
 pfz_exit(
 
 /* Called from commpage to take a delayed preemption when exiting
  * the "Preemption Free Zone" (PFZ).
  */
 kern_return_t
 pfz_exit(

-__unused       struct pfz_exit_args *args)
+       __unused        struct pfz_exit_args *args)

 {
        /* For now, nothing special to do.  We'll pick up the ASTs on kernel exit. */
 
 {
        /* For now, nothing special to do.  We'll pick up the ASTs on kernel exit. */
 

-       return (KERN_SUCCESS);
+       return KERN_SUCCESS;

 }
 
 
 }
 
 


@@ -97,27 +96,22 @@ __unused    struct pfz_exit_args *args)
  *     swtch and swtch_pri both attempt to context switch (logic in
  *     thread_block no-ops the context switch if nothing would happen).
  *     A boolean is returned that indicates whether there is anything
  *     swtch and swtch_pri both attempt to context switch (logic in
  *     thread_block no-ops the context switch if nothing would happen).
  *     A boolean is returned that indicates whether there is anything

- *     else runnable.
- *
- *     This boolean can be used by a thread waiting on a
- *     lock or condition:  If FALSE is returned, the thread is justified
- *     in becoming a resource hog by continuing to spin because there's
- *     nothing else useful that the processor could do.  If TRUE is
- *     returned, the thread should make one more check on the
- *     lock and then be a good citizen and really suspend.
+ *     else runnable.  That's no excuse to spin, though.

  */
 
 static void
 swtch_continue(void)
 {
  */
 
 static void
 swtch_continue(void)
 {

-       register processor_t    myprocessor;
-    boolean_t                          result;
+       processor_t     myprocessor;
+       boolean_t       result;

 
 

-    disable_preemption();
+       disable_preemption();

        myprocessor = current_processor();
        myprocessor = current_processor();

-       result = myprocessor->runq.count > 0 || rt_runq.count > 0;
+       result = SCHED(thread_should_yield)(myprocessor, current_thread());

        enable_preemption();
 
        enable_preemption();
 

+       ml_delay_on_yield();
+

        thread_syscall_return(result);
        /*NOTREACHED*/
 }
        thread_syscall_return(result);
        /*NOTREACHED*/
 }


@@ -126,87 +120,74 @@ boolean_t
 swtch(
        __unused struct swtch_args *args)
 {
 swtch(
        __unused struct swtch_args *args)
 {

-       register processor_t    myprocessor;
-       boolean_t                               result;
+       processor_t     myprocessor;

 
        disable_preemption();
        myprocessor = current_processor();
 
        disable_preemption();
        myprocessor = current_processor();

-       if (myprocessor->runq.count == 0 &&     rt_runq.count == 0) {
+       if (!SCHED(thread_should_yield)(myprocessor, current_thread())) {

                mp_enable_preemption();
 
                mp_enable_preemption();
 

-               return (FALSE);
+               return FALSE;

        }
        enable_preemption();
 
        counter(c_swtch_block++);
 
        }
        enable_preemption();
 
        counter(c_swtch_block++);
 

-       thread_block_reason((thread_continue_t)swtch_continue, NULL, AST_YIELD);
-
-       disable_preemption();
-       myprocessor = current_processor();
-       result = myprocessor->runq.count > 0 || rt_runq.count > 0;
-       enable_preemption();
-
-       return (result);
+       thread_yield_with_continuation((thread_continue_t)swtch_continue, NULL);

 }
 
 static void
 swtch_pri_continue(void)
 {
 }
 
 static void
 swtch_pri_continue(void)
 {

-       register processor_t    myprocessor;
-    boolean_t                          result;
+       processor_t     myprocessor;
+       boolean_t       result;

 
 

-       thread_depress_abort_internal(current_thread());
+       thread_depress_abort(current_thread());

 
 

-    disable_preemption();
+       disable_preemption();

        myprocessor = current_processor();
        myprocessor = current_processor();

-       result = myprocessor->runq.count > 0 || rt_runq.count > 0;
+       result = SCHED(thread_should_yield)(myprocessor, current_thread());

        mp_enable_preemption();
 
        mp_enable_preemption();
 

+       ml_delay_on_yield();
+

        thread_syscall_return(result);
        /*NOTREACHED*/
 }
 
 boolean_t
 swtch_pri(
        thread_syscall_return(result);
        /*NOTREACHED*/
 }
 
 boolean_t
 swtch_pri(

-__unused       struct swtch_pri_args *args)
+       __unused        struct swtch_pri_args *args)

 {
 {

-       register processor_t    myprocessor;
-       boolean_t                               result;
+       processor_t     myprocessor;

 
        disable_preemption();
        myprocessor = current_processor();
 
        disable_preemption();
        myprocessor = current_processor();

-       if (myprocessor->runq.count == 0 && rt_runq.count == 0) {
+       if (!SCHED(thread_should_yield)(myprocessor, current_thread())) {

                mp_enable_preemption();
 
                mp_enable_preemption();
 

-               return (FALSE);
+               return FALSE;

        }
        enable_preemption();
 
        counter(c_swtch_pri_block++);
 
        }
        enable_preemption();
 
        counter(c_swtch_pri_block++);
 

-       thread_depress_abstime(std_quantum);
-
-       thread_block_reason((thread_continue_t)swtch_pri_continue, NULL, AST_YIELD);
-
-       thread_depress_abort_internal(current_thread());
+       thread_depress_abstime(thread_depress_time);

 
 

-       disable_preemption();
-       myprocessor = current_processor();
-       result = myprocessor->runq.count > 0 || rt_runq.count > 0;
-       enable_preemption();
-
-       return (result);
+       thread_yield_with_continuation((thread_continue_t)swtch_pri_continue, NULL);

 }
 
 static void
 }
 
 static void

-thread_switch_continue(void)
+thread_switch_continue(void *parameter, __unused int ret)

 {
 {

-       register thread_t       self = current_thread();
-       int                                     option = self->saved.swtch.option;
+       thread_t self = current_thread();
+       int option = (int)(intptr_t)parameter;

 
 

-       if (option == SWITCH_OPTION_DEPRESS)
-               thread_depress_abort_internal(self);
+       if (option == SWITCH_OPTION_DEPRESS || option == SWITCH_OPTION_OSLOCK_DEPRESS) {
+               thread_depress_abort(self);
+       }
+
+       ml_delay_on_yield();

 
        thread_syscall_return(KERN_SUCCESS);
        /*NOTREACHED*/
 
        thread_syscall_return(KERN_SUCCESS);
        /*NOTREACHED*/


@@ -221,157 +202,261 @@ kern_return_t
 thread_switch(
        struct thread_switch_args *args)
 {
 thread_switch(
        struct thread_switch_args *args)
 {

-       register thread_t               thread, self = current_thread();
-       mach_port_name_t                thread_name = args->thread_name;
-       int                                             option = args->option;
-       mach_msg_timeout_t              option_time = args->option_time;
-
-    /*
-     * Process option.
-     */
-    switch (option) {
+       thread_t                        thread = THREAD_NULL;
+       thread_t                        self = current_thread();
+       mach_port_name_t                thread_name = args->thread_name;
+       int                             option = args->option;
+       mach_msg_timeout_t              option_time = args->option_time;
+       uint32_t                        scale_factor = NSEC_PER_MSEC;
+       boolean_t                       depress_option = FALSE;
+       boolean_t                       wait_option = FALSE;
+       wait_interrupt_t                interruptible = THREAD_ABORTSAFE;
+       port_to_thread_options_t        ptt_options = PORT_TO_THREAD_NOT_CURRENT_THREAD;

 
 

+       /*
+        *      Validate and process option.
+        *
+        * OSLock boosting only applies to other threads
+        * in your same task (even if you have a port for
+        * a thread in another task)
+        */
+       switch (option) {

        case SWITCH_OPTION_NONE:
        case SWITCH_OPTION_NONE:

-       case SWITCH_OPTION_DEPRESS:
+               break;

        case SWITCH_OPTION_WAIT:
        case SWITCH_OPTION_WAIT:

-           break;
-
+               wait_option = TRUE;
+               break;
+       case SWITCH_OPTION_DEPRESS:
+               depress_option = TRUE;
+               break;
+       case SWITCH_OPTION_DISPATCH_CONTENTION:
+               scale_factor = NSEC_PER_USEC;
+               wait_option = TRUE;
+               interruptible |= THREAD_WAIT_NOREPORT;
+               break;
+       case SWITCH_OPTION_OSLOCK_DEPRESS:
+               depress_option = TRUE;
+               interruptible |= THREAD_WAIT_NOREPORT;
+               ptt_options |= PORT_TO_THREAD_IN_CURRENT_TASK;
+               break;
+       case SWITCH_OPTION_OSLOCK_WAIT:
+               wait_option = TRUE;
+               interruptible |= THREAD_WAIT_NOREPORT;
+               ptt_options |= PORT_TO_THREAD_IN_CURRENT_TASK;
+               break;

        default:
        default:

-           return (KERN_INVALID_ARGUMENT);
-    }
-
-    workqueue_thread_yielded();
+               return KERN_INVALID_ARGUMENT;
+       }

 
        /*
         * Translate the port name if supplied.
         */
 
        /*
         * Translate the port name if supplied.
         */

-    if (thread_name != MACH_PORT_NULL) {
-               ipc_port_t                      port;
-
-               if (ipc_port_translate_send(self->task->itk_space,
-                                                                       thread_name, &port) == KERN_SUCCESS) {
-                       ip_reference(port);
-                       ip_unlock(port);
+       if (thread_name != MACH_PORT_NULL) {
+               thread = port_name_to_thread(thread_name, ptt_options);
+       }

 
 

-                       thread = convert_port_to_thread(port);
-                       ipc_port_release(port);
+       if (option == SWITCH_OPTION_OSLOCK_DEPRESS || option == SWITCH_OPTION_OSLOCK_WAIT) {
+               if (thread != THREAD_NULL) {
+                       /*
+                        * Attempt to kick the lock owner up to our same IO throttling tier.
+                        * If the thread is currently blocked in throttle_lowpri_io(),
+                        * it will immediately break out.
+                        *
+                        * TODO: SFI break out?
+                        */
+                       int new_policy = proc_get_effective_thread_policy(self, TASK_POLICY_IO);

 
 

-                       if (thread == self) {
-                               (void)thread_deallocate_internal(thread);
-                               thread = THREAD_NULL;
-                       }
+                       set_thread_iotier_override(thread, new_policy);

                }
                }

-               else
-                       thread = THREAD_NULL;

        }
        }

-       else
-               thread = THREAD_NULL;

 
        /*
         * Try to handoff if supplied.
         */
        if (thread != THREAD_NULL) {
 
        /*
         * Try to handoff if supplied.
         */
        if (thread != THREAD_NULL) {

-               processor_t             processor;
-               spl_t                   s;
-
-               s = splsched();
-               thread_lock(thread);
-
-               /*
-                *      Check that the thread is not bound
-                *      to a different processor, and that realtime
-                *      is not involved.
-                *
-                *      Next, pull it off its run queue.  If it
-                *      doesn't come, it's not eligible.
-                */
-               processor = current_processor();
-               if (processor->current_pri < BASEPRI_RTQUEUES                   &&
-                       thread->sched_pri < BASEPRI_RTQUEUES                            &&
-                       (thread->bound_processor == PROCESSOR_NULL      ||
-                        thread->bound_processor == processor)                          &&
-                               run_queue_remove(thread)                                                        ) {
-                       /*
-                        *      Hah, got it!!
-                        */
-                       thread_unlock(thread);
+               spl_t s = splsched();

 
 

-                       (void)thread_deallocate_internal(thread);
+               /* This may return a different thread if the target is pushing on something */
+               thread_t pulled_thread = thread_run_queue_remove_for_handoff(thread);

 
 

-                       if (option == SWITCH_OPTION_WAIT)
-                               assert_wait_timeout((event_t)assert_wait_timeout, THREAD_ABORTSAFE,
-                                                                                                               option_time, 1000*NSEC_PER_USEC);
-                       else
-                       if (option == SWITCH_OPTION_DEPRESS)
-                               thread_depress_ms(option_time);
+               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_THREAD_SWITCH) | DBG_FUNC_NONE,
+                   thread_tid(thread), thread->state,
+                   pulled_thread ? TRUE : FALSE, 0, 0);

 
 

-                       self->saved.swtch.option = option;
+               if (pulled_thread != THREAD_NULL) {
+                       /* We can't be dropping the last ref here */
+                       thread_deallocate_safe(thread);

 
 

-                       thread_run(self, (thread_continue_t)thread_switch_continue, NULL, thread);
-                       /* NOTREACHED */
+                       if (wait_option) {
+                               assert_wait_timeout((event_t)assert_wait_timeout, interruptible,
+                                   option_time, scale_factor);
+                       } else if (depress_option) {
+                               thread_depress_ms(option_time);
+                       }
+
+                       thread_run(self, thread_switch_continue, (void *)(intptr_t)option, pulled_thread);
+                       __builtin_unreachable();

                }
 
                }
 

-               thread_unlock(thread);

                splx(s);
 
                thread_deallocate(thread);
        }
                splx(s);
 
                thread_deallocate(thread);
        }

-               
-       if (option == SWITCH_OPTION_WAIT)
-               assert_wait_timeout((event_t)assert_wait_timeout, THREAD_ABORTSAFE, option_time, 1000*NSEC_PER_USEC);
-       else
-       if (option == SWITCH_OPTION_DEPRESS)
-               thread_depress_ms(option_time);
-         
-       self->saved.swtch.option = option;

 
 

-       thread_block_reason((thread_continue_t)thread_switch_continue, NULL, AST_YIELD);
+       if (wait_option) {
+               assert_wait_timeout((event_t)assert_wait_timeout, interruptible, option_time, scale_factor);
+       } else {
+               disable_preemption();
+               bool should_yield = SCHED(thread_should_yield)(current_processor(), current_thread());
+               enable_preemption();
+
+               if (should_yield == false) {
+                       /* Early-return if yielding to the scheduler will not be beneficial */
+                       return KERN_SUCCESS;
+               }
+
+               if (depress_option) {
+                       thread_depress_ms(option_time);
+               }
+       }
+
+       thread_yield_with_continuation(thread_switch_continue, (void *)(intptr_t)option);
+       __builtin_unreachable();
+}
+
+void
+thread_yield_with_continuation(
+       thread_continue_t       continuation,
+       void                            *parameter)
+{
+       assert(continuation);
+       thread_block_reason(continuation, parameter, AST_YIELD);
+       __builtin_unreachable();
+}
+
+/* This function is called after an assert_wait(), therefore it must not
+ * cause another wait until after the thread_run() or thread_block()
+ *
+ * Following are the calling convention for thread ref deallocation.
+ *
+ * 1) If no continuation is provided, then thread ref is consumed.
+ * (thread_handoff_deallocate convention).
+ *
+ * 2) If continuation is provided with option THREAD_HANDOFF_SETRUN_NEEDED
+ * then thread ref is always consumed.
+ *
+ * 3) If continuation is provided with option THREAD_HANDOFF_NONE then thread
+ * ref is not consumed and it is upto the continuation to deallocate
+ * the thread reference.
+ */
+static wait_result_t
+thread_handoff_internal(thread_t thread, thread_continue_t continuation,
+    void *parameter, thread_handoff_option_t option)
+{
+       thread_t self = current_thread();

 
 

-       if (option == SWITCH_OPTION_DEPRESS)
-               thread_depress_abort_internal(self);
+       /*
+        * Try to handoff if supplied.
+        */
+       if (thread != THREAD_NULL) {
+               spl_t s = splsched();
+
+               thread_t pulled_thread = thread_prepare_for_handoff(thread, option);
+
+               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_THREAD_SWITCH) | DBG_FUNC_NONE,
+                   thread_tid(thread), thread->state,
+                   pulled_thread ? TRUE : FALSE, 0, 0);
+
+               /* Deallocate thread ref if needed */
+               if (continuation == NULL || (option & THREAD_HANDOFF_SETRUN_NEEDED)) {
+                       /* Use the safe version of thread deallocate */
+                       thread_deallocate_safe(thread);
+               }
+
+               if (pulled_thread != THREAD_NULL) {
+                       int result = thread_run(self, continuation, parameter, pulled_thread);
+
+                       splx(s);
+                       return result;
+               }
+
+               splx(s);
+       }

 
 

-    return (KERN_SUCCESS);
+       int result = thread_block_parameter(continuation, parameter);
+       return result;
+}
+
+void
+thread_handoff_parameter(thread_t thread, thread_continue_t continuation,
+    void *parameter, thread_handoff_option_t option)
+{
+       thread_handoff_internal(thread, continuation, parameter, option);
+       panic("NULL continuation passed to %s", __func__);
+       __builtin_unreachable();
+}
+
+wait_result_t
+thread_handoff_deallocate(thread_t thread, thread_handoff_option_t option)
+{
+       return thread_handoff_internal(thread, NULL, NULL, option);

 }
 
 }
 

+/*
+ * Thread depression
+ *
+ * This mechanism drops a thread to priority 0 in order for it to yield to
+ * all other runnnable threads on the system.  It can be canceled or timed out,
+ * whereupon the thread goes back to where it was.
+ *
+ * Note that TH_SFLAG_DEPRESS and TH_SFLAG_POLLDEPRESS are never set at the
+ * same time.  DEPRESS always defers to POLLDEPRESS.
+ *
+ * DEPRESS only lasts across a single thread_block call, and never returns
+ * to userspace.
+ * POLLDEPRESS can be active anywhere up until thread termination.
+ */
+

 /*
  * Depress thread's priority to lowest possible for the specified interval,
 /*
  * Depress thread's priority to lowest possible for the specified interval,

- * with a value of zero resulting in no timeout being scheduled.
+ * with an interval of zero resulting in no timeout being scheduled.
+ *
+ * Must block with AST_YIELD afterwards to take effect

  */
 void
  */
 void

-thread_depress_abstime(
-       uint64_t                                interval)
+thread_depress_abstime(uint64_t interval)

 {
 {

-       register thread_t               self = current_thread();
-       uint64_t                                deadline;
-    spl_t                                      s;
+       thread_t self = current_thread();

 
 

-    s = splsched();
-    thread_lock(self);
-       if (!(self->sched_mode & TH_MODE_ISDEPRESSED)) {
-               processor_t             myprocessor = self->last_processor;
+       spl_t s = splsched();
+       thread_lock(self);

 
 

-               self->sched_pri = DEPRESSPRI;
-               myprocessor->current_pri = self->sched_pri;
-               self->sched_mode |= TH_MODE_DEPRESS;
+       assert((self->sched_flags & TH_SFLAG_DEPRESS) == 0);
+
+       if ((self->sched_flags & TH_SFLAG_POLLDEPRESS) == 0) {
+               self->sched_flags |= TH_SFLAG_DEPRESS;
+               thread_recompute_sched_pri(self, SETPRI_LAZY);

 
                if (interval != 0) {
 
                if (interval != 0) {

+                       uint64_t deadline;
+

                        clock_absolutetime_interval_to_deadline(interval, &deadline);
                        clock_absolutetime_interval_to_deadline(interval, &deadline);

-                       if (!timer_call_enter(&self->depress_timer, deadline))
+                       if (!timer_call_enter(&self->depress_timer, deadline, TIMER_CALL_USER_CRITICAL)) {

                                self->depress_timer_active++;
                                self->depress_timer_active++;

+                       }

                }
        }
                }
        }

+

        thread_unlock(self);
        thread_unlock(self);

-    splx(s);
+       splx(s);

 }
 
 void
 }
 
 void

-thread_depress_ms(
-       mach_msg_timeout_t              interval)
+thread_depress_ms(mach_msg_timeout_t interval)

 {
 {

-       uint64_t                abstime;
+       uint64_t abstime;

 
 

-       clock_interval_to_absolutetime_interval(
-                                                       interval, 1000*NSEC_PER_USEC, &abstime);
+       clock_interval_to_absolutetime_interval(interval, NSEC_PER_MSEC, &abstime);

        thread_depress_abstime(abstime);
 }
 
        thread_depress_abstime(abstime);
 }
 


@@ -379,101 +464,142 @@ thread_depress_ms(
  *     Priority depression expiration.
  */
 void
  *     Priority depression expiration.
  */
 void

-thread_depress_expire(
-       void                    *p0,
-       __unused void   *p1)
+thread_depress_expire(void      *p0,
+    __unused void      *p1)

 {
 {

-       thread_t                thread = p0;
-    spl_t                      s;
+       thread_t thread = (thread_t)p0;
+
+       spl_t s = splsched();
+       thread_lock(thread);
+
+       assert((thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) != TH_SFLAG_DEPRESSED_MASK);

 
 

-    s = splsched();
-    thread_lock(thread);

        if (--thread->depress_timer_active == 0) {
        if (--thread->depress_timer_active == 0) {

-               thread->sched_mode &= ~TH_MODE_ISDEPRESSED;
-               compute_priority(thread, FALSE);
+               thread->sched_flags &= ~TH_SFLAG_DEPRESSED_MASK;
+               if ((thread->state & TH_RUN) == TH_RUN) {
+                       thread->last_basepri_change_time = mach_absolute_time();
+               }
+               thread_recompute_sched_pri(thread, SETPRI_DEFAULT);

        }
        }

-    thread_unlock(thread);
-    splx(s);
+
+       thread_unlock(thread);
+       splx(s);

 }
 
 /*
 }
 
 /*

- *     Prematurely abort priority depression if there is one.
+ * Prematurely abort priority depression if there is one.

  */
 kern_return_t
  */
 kern_return_t

-thread_depress_abort_internal(
-       thread_t                                thread)
+thread_depress_abort(thread_t thread)

 {
 {

-    kern_return_t                      result = KERN_NOT_DEPRESSED;
-    spl_t                                      s;
-
-    s = splsched();
-    thread_lock(thread);
-       if (!(thread->sched_mode & TH_MODE_POLLDEPRESS)) {
-               if (thread->sched_mode & TH_MODE_ISDEPRESSED) {
-                       thread->sched_mode &= ~TH_MODE_ISDEPRESSED;
-                       compute_priority(thread, FALSE);
-                       result = KERN_SUCCESS;
-               }
+       kern_return_t result = KERN_NOT_DEPRESSED;
+
+       spl_t s = splsched();
+       thread_lock(thread);

 
 

-               if (timer_call_cancel(&thread->depress_timer))
-                       thread->depress_timer_active--;
+       assert((thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) != TH_SFLAG_DEPRESSED_MASK);
+
+       /*
+        * User-triggered depress-aborts should not get out
+        * of the poll-depress, but they should cancel a regular depress.
+        */
+       if ((thread->sched_flags & TH_SFLAG_POLLDEPRESS) == 0) {
+               result = thread_depress_abort_locked(thread);

        }
        }

+

        thread_unlock(thread);
        thread_unlock(thread);

-    splx(s);
+       splx(s);

 
 

-    return (result);
+       return result;

 }
 
 }
 

-void
-thread_poll_yield(
-       thread_t                self)
+/*
+ * Prematurely abort priority depression or poll depression if one is active.
+ * Called with the thread locked.
+ */
+kern_return_t
+thread_depress_abort_locked(thread_t thread)

 {
 {

-       spl_t                   s;
+       if ((thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) == 0) {
+               return KERN_NOT_DEPRESSED;
+       }

 
 

+       assert((thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) != TH_SFLAG_DEPRESSED_MASK);
+
+       thread->sched_flags &= ~TH_SFLAG_DEPRESSED_MASK;
+       if ((thread->state & TH_RUN) == TH_RUN) {
+               thread->last_basepri_change_time = mach_absolute_time();
+       }
+
+       thread_recompute_sched_pri(thread, SETPRI_LAZY);
+
+       if (timer_call_cancel(&thread->depress_timer)) {
+               thread->depress_timer_active--;
+       }
+
+       return KERN_SUCCESS;
+}
+
+/*
+ * Invoked as part of a polling operation like a no-timeout port receive
+ *
+ * Forces a fixpri thread to yield if it is detected polling without blocking for too long.
+ */
+void
+thread_poll_yield(thread_t self)
+{

        assert(self == current_thread());
        assert(self == current_thread());

+       assert((self->sched_flags & TH_SFLAG_DEPRESS) == 0);

 
 

-       s = splsched();
-       if (!(self->sched_mode & (TH_MODE_REALTIME|TH_MODE_TIMESHARE))) {
-               uint64_t                        total_computation, abstime;
-
-               abstime = mach_absolute_time();
-               total_computation = abstime - self->computation_epoch;
-               total_computation += self->computation_metered;
-               if (total_computation >= max_poll_computation) {
-                       processor_t             myprocessor = current_processor();
-                       ast_t                   preempt;
-
-                       thread_lock(self);
-                       if (!(self->sched_mode & TH_MODE_ISDEPRESSED)) {
-                               self->sched_pri = DEPRESSPRI;
-                               myprocessor->current_pri = self->sched_pri;
-                       }
-                       self->computation_epoch = abstime;
-                       self->computation_metered = 0;
-                       self->sched_mode |= TH_MODE_POLLDEPRESS;
+       if (self->sched_mode != TH_MODE_FIXED) {
+               return;
+       }

 
 

-                       abstime += (total_computation >> sched_poll_yield_shift);
-                       if (!timer_call_enter(&self->depress_timer, abstime))
-                               self->depress_timer_active++;
-                       thread_unlock(self);
+       spl_t s = splsched();
+
+       uint64_t abstime = mach_absolute_time();
+       uint64_t total_computation = abstime -
+           self->computation_epoch + self->computation_metered;
+
+       if (total_computation >= max_poll_computation) {
+               thread_lock(self);
+
+               self->computation_epoch   = abstime;
+               self->computation_metered = 0;
+
+               uint64_t yield_expiration = abstime +
+                   (total_computation >> sched_poll_yield_shift);

 
 

-                       if ((preempt = csw_check(myprocessor)) != AST_NONE)
-                               ast_on(preempt);
+               if (!timer_call_enter(&self->depress_timer, yield_expiration,
+                   TIMER_CALL_USER_CRITICAL)) {
+                       self->depress_timer_active++;

                }
                }

+
+               self->sched_flags |= TH_SFLAG_POLLDEPRESS;
+               thread_recompute_sched_pri(self, SETPRI_DEFAULT);
+
+               thread_unlock(self);

        }
        splx(s);
 }
 
        }
        splx(s);
 }
 

-
+/*
+ * Kernel-internal interface to yield for a specified period
+ *
+ * WARNING: Will still yield to priority 0 even if the thread is holding a contended lock!
+ */

 void
 void

-thread_yield_internal(
-       mach_msg_timeout_t      ms)
+thread_yield_internal(mach_msg_timeout_t ms)

 {
 {

-       processor_t     myprocessor;
+       thread_t self = current_thread();
+
+       assert((self->sched_flags & TH_SFLAG_DEPRESSED_MASK) != TH_SFLAG_DEPRESSED_MASK);
+
+       processor_t     myprocessor;

 
        disable_preemption();
        myprocessor = current_processor();
 
        disable_preemption();
        myprocessor = current_processor();

-       if (myprocessor->runq.count == 0 && rt_runq.count == 0) {
+       if (!SCHED(thread_should_yield)(myprocessor, self)) {

                mp_enable_preemption();
 
                return;
                mp_enable_preemption();
 
                return;


@@ -484,6 +610,42 @@ thread_yield_internal(
 
        thread_block_reason(THREAD_CONTINUE_NULL, NULL, AST_YIELD);
 
 
        thread_block_reason(THREAD_CONTINUE_NULL, NULL, AST_YIELD);
 

-       thread_depress_abort_internal(current_thread());
+       thread_depress_abort(self);

 }
 
 }
 

+/*
+ * This yields to a possible non-urgent preemption pending on the current processor.
+ *
+ * This is useful when doing a long computation in the kernel without returning to userspace.
+ *
+ * As opposed to other yielding mechanisms, this does not drop the priority of the current thread.
+ */
+void
+thread_yield_to_preemption()
+{
+       /*
+        * ast_pending() should ideally be called with interrupts disabled, but
+        * the check here is fine because csw_check() will do the right thing.
+        */
+       ast_t *pending_ast = ast_pending();
+       ast_t ast = AST_NONE;
+       processor_t p;
+
+       if (*pending_ast & AST_PREEMPT) {
+               thread_t self = current_thread();
+
+               spl_t s = splsched();
+
+               p = current_processor();
+               thread_lock(self);
+               ast = csw_check(self, p, AST_YIELD);
+               ast_on(ast);
+               thread_unlock(self);
+
+               if (ast != AST_NONE) {
+                       (void)thread_block_reason(THREAD_CONTINUE_NULL, NULL, ast);
+               }
+
+               splx(s);
+       }
+}