Improved scheduler suspend functionality

- The calling process is now suspended while synchronizing
  scheduler suspends via erlang:system_flag(schedulers_online, _)
  and erlang:system_flag(multi_scheduling, _), instead of blocking
  the scheduler thread in the BIF call waiting for the operation
  to synchronize. Besides releasing the scheduler for other work
  (or immediate suspend) it also makes it possible to abort the
  operation by killing the process.
- erlang:system_flag(schedulers_online, _) now only wait for normal
  schedulers to complete before it returns. This since it may take
  a very long time before all dirty schedulers suspends.
- erlang:system_flag(multi_scheduling, block_normal|unblock_normal)
  which only operate on normal schedulers has been introduced. This
  since there are use cases where suspend of dirty schedulers are
  not of interest (hipe loader).
- erlang:system_flag(multi_scheduling, block) still blocks all
  dirty schedulers as well as all normal schedulers except one since
  it is hard to redefine what multi scheduling block means.
- The three operations:
  - changing amount of schedulers online
  - blocking/unblocking normal multi scheduling
  - blocking/unblocking full multi scheduling
  can now be done in parallel. This is important since otherwise
  a full multi scheduling block would potentially delay the other
  operations for a very long time.

1 files changed, 71 insertions, 28 deletions

diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml
index 1f8e89768c..350a8506f5 100644
--- a/erts/doc/src/erlang.xml
+++ b/erts/doc/src/erlang.xml
@@ -6469,32 +6469,44 @@ ok
         <p><marker id="system_flag_multi_scheduling"></marker>
           If multi-scheduling is enabled, more than one scheduler
           thread is used by the emulator. Multi-scheduling can be
-          blocked. When multi-scheduling is blocked, only
-          one scheduler thread schedules Erlang processes.</p>
+          blocked in two different ways. Either all schedulers but
+	  one is blocked, or all <em>normal</em> schedulers but
+	  one is blocked. When only normal schedulers are blocked
+	  dirty schedulers are free to continue to schedule
+	  processes.</p>
         <p>If <c><anno>BlockState</anno> =:= block</c>, multi-scheduling is
-          blocked. If <c><anno>BlockState</anno> =:= unblock</c> and no one
+          blocked. That is, one and only one scheduler thread will
+	  execute. If <c><anno>BlockState</anno> =:= unblock</c> and no one
           else blocks multi-scheduling, and this process has
           blocked only once, multi-scheduling is unblocked.</p>
-        <p>One process can block multi-scheduling multiple times.
-          If a process has blocked multiple times, it must
-          unblock exactly as many times as it has blocked before it
-          has released its multi-scheduling block. If a process that
-          has blocked multi-scheduling exits, it releases its
-          blocking of multi-scheduling.</p>
+        <p>If <c><anno>BlockState</anno> =:= block_normal</c>, normal
+	  multi-scheduling is blocked. That is, only one normal scheduler
+	  thread will execute, but multiple dirty schedulers may execute.
+	  If <c><anno>BlockState</anno> =:= unblock_normal</c> and no one
+          else blocks normal multi-scheduling, and this process has
+          blocked only once, normal multi-scheduling is unblocked.</p>
+        <p>One process can block multi-scheduling as well as normal
+	  multi-scheduling multiple times. If a process has blocked
+	  multiple times, it must unblock exactly as many times as it
+	  has blocked before it has released its multi-scheduling
+	  block. If a process that has blocked multi-scheduling or normal
+	  multi scheduling exits, it automatically releases its blocking
+	  of multi-scheduling and normal multi-scheduling.</p>
         <p>The return values are <c>disabled</c>, <c>blocked</c>,
-          or <c>enabled</c>. The returned value describes the
-          state just after the call to
+          <c>blocked_normal</c>, or <c>enabled</c>. The returned value
+	  describes the state just after the call to
           <c>erlang:system_flag(multi_scheduling, <anno>BlockState</anno>)</c>
           has been made. For information about the return values, see
           <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>.</p>
-        <note><p>Blocking of multi-scheduling is normally not needed.
-          If you feel that you need to block multi-scheduling,
-          consider it a few more times again. Blocking multi-scheduling
-          is only to be used as a last resort, as it is most likely
-          a <em>very inefficient</em> way to solve the problem.</p>
+        <note><p>Blocking of multi-scheduling and normal multi-scheduling
+	  is normally not needed. If you feel that you need to use these
+	  features, consider it a few more times again. Blocking
+	  multi-scheduling is only to be used as a last resort, as it is
+	  most likely a <em>very inefficient</em> way to solve the problem.</p>
         </note>
         <p>See also
           <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>,
+          <seealso marker="#system_info_normal_multi_scheduling_blockers">erlang:system_info(normal_multi_scheduling_blockers)</seealso>,
           <seealso marker="#system_info_multi_scheduling_blockers">erlang:system_info(multi_scheduling_blockers)</seealso>, and
           <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>.</p>
       </desc>
@@ -7329,6 +7341,16 @@ ok
               where <c><anno>MinHeapSize</anno></c> is the current
               system-wide minimum heap size for spawned processes.</p>
           </item>
+          <tag><marker id="system_info_message_queue_data"><c>message_queue_data</c></marker></tag>
+          <item>
+            <p>Returns the default value of the <c>message_queue_data</c>
+	    process flag which is either <c>off_heap</c>, <c>on_heap</c>, or <c>mixed</c>.
+	    This default is set by the <c>erl</c> command line argument
+	    <seealso marker="erl#+xmqd"><c>+xmqd</c></seealso>. For more information on the
+	    <c>message_queue_data</c> process flag, see documentation of
+            <seealso marker="#process_flag_message_queue_data"><c>process_flag(message_queue_data,
+	    MQD)</c></seealso>.</p>
+          </item>
 	  <tag><c>min_bin_vheap_size</c></tag>
           <item>
 	      <p>Returns <c>{min_bin_vheap_size,
@@ -7348,7 +7370,8 @@ ok
           <tag><c>multi_scheduling</c></tag>
           <item>
           <marker id="system_info_multi_scheduling"></marker>
-            <p>Returns <c>disabled</c>, <c>blocked</c>, or <c>enabled</c>:</p>
+            <p>Returns <c>disabled</c>, <c>blocked</c>, <c>blocked_normal</c>,
+	    or <c>enabled</c>:</p>
             <taglist>
               <tag><c>disabled</c></tag>
               <item>
@@ -7359,14 +7382,22 @@ ok
               <tag><c>blocked</c></tag>
               <item>
                 <p>The emulator has more than one scheduler thread,
-                  but all scheduler threads except one are blocked,
-                  that is, only one scheduler thread schedules
+                  but all scheduler threads except one are blocked.
+                  That is, only one scheduler thread schedules
                   Erlang processes and executes Erlang code.</p>
               </item>
+              <tag><c>blocked_normal</c></tag>
+              <item>
+                <p>The emulator has more than one scheduler thread,
+                  but all normal scheduler threads except one are
+		  blocked. Note that dirty schedulers are not
+		  blocked, and may schedule Erlang processes and
+		  execute native code.</p>
+              </item>
               <tag><c>enabled</c></tag>
               <item>
                 <p>The emulator has more than one scheduler thread,
-                  and no scheduler threads are blocked, that is,
+                  and no scheduler threads are blocked. That is,
                   all available scheduler threads schedule
                   Erlang processes and execute Erlang code.</p>
               </item>
@@ -7374,6 +7405,7 @@ ok
             <p>See also
               <seealso marker="#system_flag_multi_scheduling">erlang:system_flag(multi_scheduling, BlockState)</seealso>,
               <seealso marker="#system_info_multi_scheduling_blockers">erlang:system_info(multi_scheduling_blockers)</seealso>,
+              <seealso marker="#system_info_normal_multi_scheduling_blockers">erlang:system_info(normal_multi_scheduling_blockers)</seealso>,
               and
               <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>.</p>
           </item>
@@ -7390,6 +7422,8 @@ ok
             <p>See also
               <seealso marker="#system_flag_multi_scheduling">erlang:system_flag(multi_scheduling, BlockState)</seealso>,
               <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>,
+	      <seealso marker="#system_info_normal_multi_scheduling_blockers">erlang:system_info(normal_multi_scheduling_blockers)</seealso>,
+
               and
               <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>.</p>
           </item>
@@ -7399,15 +7433,23 @@ ok
               used by the runtime system. It is on the form
 	      "&lt;major ver&gt;.&lt;minor ver&gt;".</p>
           </item>
-          <tag><marker id="system_info_message_queue_data"><c>message_queue_data</c></marker></tag>
+          <tag><c>normal_multi_scheduling_blockers</c></tag>
           <item>
-            <p>Returns the default value of the <c>message_queue_data</c>
-	    process flag which is either <c>off_heap</c>, <c>on_heap</c>, or <c>mixed</c>.
-	    This default is set by the <c>erl</c> command line argument
-	    <seealso marker="erl#+xmqd"><c>+xmqd</c></seealso>. For more information on the
-	    <c>message_queue_data</c> process flag, see documentation of
-            <seealso marker="#process_flag_message_queue_data"><c>process_flag(message_queue_data,
-	    MQD)</c></seealso>.</p>
+          <marker id="system_info_normal_multi_scheduling_blockers"></marker>
+            <p>Returns a list of <c><anno>Pid</anno></c>s when
+              normal multi-scheduling is blocked (i.e. all normal schedulers
+	      but one is blocked), otherwise the empty list is returned.
+	      The <c><anno>Pid</anno></c>s in the list represent all the
+	      processes currently blocking normal multi-scheduling.
+	      A <c><anno>Pid</anno></c> occurs only once in the list, even if
+	      the corresponding process has blocked multiple times.</p>
+            <p>See also
+              <seealso marker="#system_flag_multi_scheduling">erlang:system_flag(multi_scheduling, BlockState)</seealso>,
+              <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>,
+	      <seealso marker="#system_info_multi_scheduling_blockers">erlang:system_info(multi_scheduling_blockers)</seealso>,
+
+              and
+              <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>.</p>
           </item>
           <tag><marker id="system_info_otp_release"><c>otp_release</c></marker></tag>
           <item>
@@ -7650,6 +7692,7 @@ ok
 	      <seealso marker="#system_info_scheduler_id">erlang:system_info(scheduler_id)</seealso>,
               <seealso marker="#system_flag_multi_scheduling">erlang:system_flag(multi_scheduling, BlockState)</seealso>,
               <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>,
+	      <seealso marker="#system_info_normal_multi_scheduling_blockers">erlang:system_info(normal_multi_scheduling_blockers)</seealso>
               and
               <seealso marker="#system_info_multi_scheduling_blockers">erlang:system_info(multi_scheduling_blockers)</seealso>.</p>
           </item>