Execution of system tasks in context of another process

A process requesting a system task to be executed in the context of
another process will be notified by a message when the task has
executed. This message will be on the form:
  {RequestType, RequestId, Pid, Result}.

A process requesting a system task to be executed can set priority
on the system task. The requester typically set the same priority
on the task as its own process priority, and by this avoiding
priority inversion. A request for execution of a system task is
made by calling the statically linked in NIF
erts_internal:request_system_task(Pid, Prio, Request). This is an
undocumented ERTS internal function that should remain so. It
should *only* be called from BIF implementations.

Currently defined system tasks are:
* garbage_collect
* check_process_code

Further system tasks can and will be implemented in the future.

The erlang:garbage_collect/[1,2] and erlang:check_process_code/[2,3]
BIFs are now implemented using system tasks. Both the
'garbage_collect' and the 'check_process_code' operations perform
or may perform garbage_collections. By doing these via the
system task functionality all garbage collect operations in the
system will be performed solely in the context of the process
being garbage collected. This makes it possible to later implement
functionality for disabling garbage collection of a process over
context switches.

Newly introduced BIFs:

* erlang:garbage_collect/2 - The new second argument is an option
  list. Introduced option:
  * {async, RequestId} - making it possible for users to issue
    asynchronous garbage collect requests.

* erlang:check_process_code/3 -  The new third argument is an
  option list. Introduced options:
  * {async, RequestId} - making it possible for users to issue
    asynchronous check process code requests.
  * {allow_gc, boolean()} - making it possible to issue requests
    that aren't allowed to garbage collect (operation will abort
    if gc should be needed).
  These options have been introduced as a preparation for
  parallelization of check_process_code operations when the
  code_server is about to purge a module.

21 files changed, 1787 insertions, 316 deletions

diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml
index 5ee40823bc..a9642be6fa 100644
--- a/erts/doc/src/erlang.xml
+++ b/erts/doc/src/erlang.xml
@@ -501,16 +501,87 @@
       <name name="check_process_code" arity="2"/>
       <fsummary>Check if a process is executing old code for a module</fsummary>
       <desc>
-        <p>Returns <c>true</c> if the process <c><anno>Pid</anno></c> is executing
-          old code for <c><anno>Module</anno></c>. That is, if the current call of
-          the process executes old code for this module, or if the
-          process has references to old code for this module, or if the
-          process contains funs that references old code for this
-          module. Otherwise, it returns <c>false</c>.</p>
-        <pre>
-> <input>check_process_code(Pid, lists).</input>
-false</pre>
+        <p>The same as
+	<seealso marker="#check_process_code/3"><c>erlang:check_process_code(<anno>Pid</anno>,
+	<anno>Module</anno>, [])</c></seealso>.</p>
+      </desc>
+    </func>
+    <func>
+      <name name="check_process_code" arity="3"/>
+      <fsummary>Check if a process is executing old code for a module</fsummary>
+      <desc>
+	<p>Check if the node local process identified by <c><anno>Pid</anno></c>
+	is executing old code for <c><anno>Module</anno></c>.</p>
+	<p>Currently available <c><anno>Option</anno>s</c>:</p>
+        <taglist>
+          <tag><c>{allow_gc, boolean()}</c></tag>
+          <item>
+	    Determines if garbage collection is allowed when performing
+	    the operation. If <c>{allow_gc, false}</c> is passed, and
+	    a garbage collection is needed in order to determine the
+	    result of the operation, the operation will be aborted
+	    (see information on <c><anno>CheckResult</anno></c> below).
+	    The default is to allow garbage collection, i.e.,
+	    <c>{allow_gc, true}</c>.
+	  </item>
+          <tag><c>{async, RequestId}</c></tag>
+          <item>
+	    The <c>check_process_code/3</c> function will return
+	    the value <c>async</c> immediately after the request
+	    has been sent. When the request has been processed, the
+	    process that called this function will be passed a
+	    message on the form:<br/>
+	    <c>{check_process_code, <anno>RequestId</anno>, <anno>CheckResult</anno>}</c>.
+	  </item>
+        </taglist>
+	<p>If <c><anno>Pid</anno></c> equals <c>self()</c>, and
+	no <c>async</c> option has been passed, the operation will
+	be performed at once. In all other cases a request for
+	the operation will be sent to the process identified by
+	<c><anno>Pid</anno></c>, and will be handled when
+	appropriate. If no <c>async</c> option has been passed,
+	the caller will block until <c><anno>CheckResult</anno></c>
+	is available and can be returned.</p>
+	<p><c><anno>CheckResult</anno></c> informs about the result of
+	the request:</p>
+        <taglist>
+          <tag><c>true</c></tag>
+          <item>
+	    The process identified by <c><anno>Pid</anno></c> is
+	    executing old code for <c><anno>Module</anno></c>.
+	    That is, the current call of the process executes old
+	    code for this module, or the process has references
+	    to old code for this module, or the process contains
+	    funs that references old code for this module.
+	  </item>
+          <tag><c>false</c></tag>
+          <item>
+	    The process identified by <c><anno>Pid</anno></c> is
+	    not executing old code for <c><anno>Module</anno></c>.
+	  </item>
+          <tag><c>aborted</c></tag>
+          <item>
+	    The operation was aborted since the process needed to
+	    be garbage collected in order to determine the result
+	    of the operation, and the operation was requested
+	    by passing the <c>{allow_gc, false}</c> option.</item>
+        </taglist>
         <p>See also <seealso marker="kernel:code">code(3)</seealso>.</p>
+        <p>Failures:</p>
+        <taglist>
+          <tag><c>badarg</c></tag>
+          <item>
+	    If <c><anno>Pid</anno></c> is not a node local process identifier.
+	  </item>
+          <tag><c>badarg</c></tag>
+          <item>
+	    If <c><anno>Module</anno></c> is not an atom.
+	  </item>
+          <tag><c>badarg</c></tag>
+          <item>
+	    If <c><anno>OptionList</anno></c> is not a valid list of options.
+	  </item>
+        </taglist>
       </desc>
     </func>
     <func>
@@ -1197,20 +1268,74 @@ true
           that the spontaneous garbage collection will occur too late
           or not at all. Improper use may seriously degrade system
           performance.</p>
-        <p>Compatibility note: In versions of OTP prior to R7,
-          the garbage collection took place at the next context switch,
-          not immediately. To force a context switch after a call to
-          <c>erlang:garbage_collect()</c>, it was sufficient to make
-          any function call.</p>
       </desc>
     </func>
     <func>
       <name name="garbage_collect" arity="1"/>
-      <fsummary>Force an immediate garbage collection of a process</fsummary>
+      <fsummary>Garbage collect a process</fsummary>
+      <desc>
+        <p>The same as
+	<seealso marker="#garbage_collect/2"><c>garbage_collect(<anno>Pid</anno>, [])</c></seealso>.</p>
+      </desc>
+    </func>
+    <func>
+      <name name="garbage_collect" arity="2"/>
+      <fsummary>Garbage collect a process</fsummary>
       <desc>
-        <p>Works like <c>erlang:garbage_collect()</c> but on any
-          process. The same caveats apply. Returns <c>false</c> if
-          <c><anno>Pid</anno></c> refers to a dead process; <c>true</c> otherwise.</p>
+	<p>Garbage collect the node local process identified by
+	<c><anno>Pid</anno></c>.</p>
+	<p>Currently available <c><anno>Option</anno></c>s:</p>
+        <taglist>
+          <tag><c>{async, RequestId}</c></tag>
+          <item>
+	    The <c>garbage_collect/2</c> function will return
+	    the value <c>async</c> immediately after the request
+	    has been sent. When the request has been processed, the
+	    process that called this function will be passed a
+	    message on the form:<br/>
+	    <c>{garbage_collect, <anno>RequestId</anno>, <anno>GCResult</anno>}</c>.
+	  </item>
+        </taglist>
+	<p>If <c><anno>Pid</anno></c> equals <c>self()</c>, and
+	no <c>async</c> option has been passed, the garbage
+	collection will be performed at once, i.e. the same as
+	calling
+	<seealso marker="#garbage_collect/0">garbage_collect/0</seealso>.
+	In all other cases a request for garbage collection will
+	be sent to the process identified by <c><anno>Pid</anno></c>,
+	and will be handled when appropriate. If no <c>async</c>
+	option has been passed, the caller will block until
+	<c><anno>GCResult</anno></c> is available and can be
+	returned.</p>
+	<p><c><anno>GCResult</anno></c> informs about the result of
+	the garbage collection request:</p>
+        <taglist>
+          <tag><c>true</c></tag>
+          <item>
+	    The process identified by <c><anno>Pid</anno></c> has
+	    been garbage collected.
+	  </item>
+          <tag><c>false</c></tag>
+          <item>
+	    No garbage collection was performed. This since the
+	    the process identified by <c><anno>Pid</anno></c>
+	    terminated before the request could be satisfied.
+	  </item>
+        </taglist>
+        <p>Note that the same caveats as for
+	<seealso marker="#garbage_collect/0">garbage_collect/0</seealso>
+	apply.</p>
+        <p>Failures:</p>
+        <taglist>
+          <tag><c>badarg</c></tag>
+          <item>
+	    If <c><anno>Pid</anno></c> is not a node local process identifier.
+	  </item>
+          <tag><c>badarg</c></tag>
+          <item>
+	    If <c><anno>OptionList</anno></c> is not a valid list of options.
+	  </item>
+        </taglist>
       </desc>
     </func>
     <func>
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names
index c2f32ba089..8433fd826a 100644
--- a/erts/emulator/beam/atom.names
+++ b/erts/emulator/beam/atom.names
@@ -78,6 +78,7 @@ atom allocated_areas
 atom allocator
 atom allocator_sizes
 atom alloc_util_allocators
+atom allow_gc
 atom allow_passive_connect
 atom already_loaded
 atom amd64
diff --git a/erts/emulator/beam/beam_bif_load.c b/erts/emulator/beam/beam_bif_load.c
index 649594a334..5239940a50 100644
--- a/erts/emulator/beam/beam_bif_load.c
+++ b/erts/emulator/beam/beam_bif_load.c
@@ -36,7 +36,7 @@
 #include "erl_thr_progress.h"
 
 static void set_default_trace_pattern(Eterm module);
-static Eterm check_process_code(Process* rp, Module* modp);
+static Eterm check_process_code(Process* rp, Module* modp, int allow_gc, int *redsp);
 static void delete_code(Module* modp);
 static void decrement_refc(BeamInstr* code);
 static int is_native(BeamInstr* code);
@@ -427,69 +427,80 @@ check_old_code_1(BIF_ALIST_1)
 }
 
 Eterm
-check_process_code_2(BIF_ALIST_2)
+erts_check_process_code(Process *c_p, Eterm module, int allow_gc, int *redsp)
 {
-    Process* rp;
     Module* modp;
+    Eterm res;
+    ErtsCodeIndex code_ix;
 
-    if (is_not_atom(BIF_ARG_2)) {
-	goto error;
-    }
-    if (is_internal_pid(BIF_ARG_1)) {
-	Eterm res;
-	ErtsCodeIndex code_ix;
-
-	code_ix = erts_active_code_ix();
-	modp = erts_get_module(BIF_ARG_2, code_ix);
-	if (modp == NULL) {		/* Doesn't exist. */
-	    return am_false;
-	}
-	erts_rlock_old_code(code_ix);
-	if (modp->old.code == NULL) { /* No old code. */
-	    erts_runlock_old_code(code_ix);
-	    return am_false;
-	}
-	erts_runlock_old_code(code_ix);
-	
-#ifdef ERTS_SMP
-	rp = erts_pid2proc_suspend(BIF_P, ERTS_PROC_LOCK_MAIN,
-				   BIF_ARG_1, ERTS_PROC_LOCK_MAIN);
-#else
-	rp = erts_pid2proc(BIF_P, 0, BIF_ARG_1, 0);
-#endif
-	if (!rp) {
-	    BIF_RET(am_false);
-	}
-	if (rp == ERTS_PROC_LOCK_BUSY) {
-	    ERTS_BIF_YIELD2(bif_export[BIF_check_process_code_2], BIF_P,
-			    BIF_ARG_1, BIF_ARG_2);
-	}
-	erts_rlock_old_code(code_ix);
-	if (modp->old.code != NULL) { /* must check again */
-	    res = check_process_code(rp, modp);
-	}
-	else {
-	    res = am_false;
-	}
-	erts_runlock_old_code(code_ix);
-#ifdef ERTS_SMP
-	if (BIF_P != rp) {
-	    erts_resume(rp, ERTS_PROC_LOCK_MAIN);
-	    erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MAIN);
+    (*redsp)++;
+
+    ASSERT(is_atom(module));
+
+    code_ix = erts_active_code_ix();
+    modp = erts_get_module(module, code_ix);
+    if (!modp)
+	return am_false;
+    erts_rlock_old_code(code_ix);
+    res = modp->old.code ? check_process_code(c_p, modp, allow_gc, redsp) : am_false;
+    erts_runlock_old_code(code_ix);
+
+    return res;
+}
+
+BIF_RETTYPE erts_internal_check_process_code_2(BIF_ALIST_2)
+{
+    int reds = 0;
+    Eterm res;
+    Eterm olist = BIF_ARG_2;
+    int allow_gc = 1;
+
+    if (is_not_atom(BIF_ARG_1))
+	goto badarg;
+
+    while (is_list(olist)) {
+	Eterm *lp = list_val(olist);
+	Eterm opt = CAR(lp);
+	if (is_tuple(opt)) {
+	    Eterm* tp = tuple_val(opt);
+	    switch (arityval(tp[0])) {
+	    case 2:
+		switch (tp[1]) {
+		case am_allow_gc:
+		    switch (tp[2]) {
+		    case am_false:
+			allow_gc = 0;
+			break;
+		    case am_true:
+			allow_gc = 1;
+			break;
+		    default:
+			goto badarg;
+		    }
+		    break;
+		default:
+		    goto badarg;
+		}
+		break;
+	    default:
+		goto badarg;
+	    }
 	}
-#endif
-	BIF_RET(res);
-    }
-    else if (is_external_pid(BIF_ARG_1)
-	     && external_pid_dist_entry(BIF_ARG_1) == erts_this_dist_entry) {
-	BIF_RET(am_false);
+	else
+	    goto badarg;
+	olist = CDR(lp);
     }
+    if (is_not_nil(olist))
+	goto badarg;
 
- error:
+    res = erts_check_process_code(BIF_P, BIF_ARG_1, allow_gc, &reds);
+
+    BIF_RET2(res, reds);
+
+badarg:
     BIF_ERROR(BIF_P, BADARG);
 }
 
-
 BIF_RETTYPE delete_module_1(BIF_ALIST_1)
 {
     ErtsCodeIndex code_ix;
@@ -710,7 +721,7 @@ set_default_trace_pattern(Eterm module)
 }
 
 static Eterm
-check_process_code(Process* rp, Module* modp)
+check_process_code(Process* rp, Module* modp, int allow_gc, int *redsp)
 {
     BeamInstr* start;
     char* mod_start;
@@ -786,6 +797,8 @@ check_process_code(Process* rp, Module* modp)
 		if (done_gc) {
 		    return am_true;
 		} else {
+		    if (!allow_gc)
+			return am_aborted;
 		    /*
 		    * Try to get rid of this fun by garbage collecting.
 		    * Clear both fvalue and ftrace to make sure they
@@ -796,7 +809,7 @@ check_process_code(Process* rp, Module* modp)
 		    rp->ftrace = NIL;
 		    done_gc = 1;
 		    FLAGS(rp) |= F_NEED_FULLSWEEP;
-		    (void) erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity);
+		    *redsp += erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity);
 		    goto rescan;
 		}
 	    }
@@ -850,6 +863,9 @@ check_process_code(Process* rp, Module* modp)
 	    Uint lit_size;
 	    struct erl_off_heap_header* oh;
 
+	    if (!allow_gc)
+		return am_aborted;
+
 	    /*
 	     * Try to get rid of constants by by garbage collecting.
 	     * Clear both fvalue and ftrace.
@@ -859,11 +875,12 @@ check_process_code(Process* rp, Module* modp)
 	    rp->ftrace = NIL;
 	    done_gc = 1;
 	    FLAGS(rp) |= F_NEED_FULLSWEEP;
-	    (void) erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity);
+	    *redsp += erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity);
 	    literals = (Eterm *) modp->old.code[MI_LITERALS_START];
 	    lit_size = (Eterm *) modp->old.code[MI_LITERALS_END] - literals;
 	    oh = (struct erl_off_heap_header *)
 		modp->old.code[MI_LITERALS_OFF_HEAP];
+	    *redsp += lit_size / 10; /* Need, better value... */
 	    erts_garbage_collect_literals(rp, literals, lit_size, oh);
 	}
     }
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index 755c5e6882..58bd6aac0b 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -3741,45 +3741,6 @@ BIF_RETTYPE now_0(BIF_ALIST_0)
 
 /**********************************************************************/
 
-BIF_RETTYPE garbage_collect_1(BIF_ALIST_1)
-{
-    int reds;
-    Process *rp;
-
-    if (is_not_pid(BIF_ARG_1)) {
-	BIF_ERROR(BIF_P, BADARG);
-    }
-
-    if (BIF_P->common.id == BIF_ARG_1)
-	rp = BIF_P;
-    else {
-#ifdef ERTS_SMP
-	rp = erts_pid2proc_suspend(BIF_P, ERTS_PROC_LOCK_MAIN,
-				   BIF_ARG_1, ERTS_PROC_LOCK_MAIN);
-	if (rp == ERTS_PROC_LOCK_BUSY)
-	    ERTS_BIF_YIELD1(bif_export[BIF_garbage_collect_1], BIF_P, BIF_ARG_1);
-#else
-	rp = erts_proc_lookup(BIF_ARG_1);
-#endif
-	if (!rp)
-	    BIF_RET(am_false);
-    }
-
-    /* The GC cost is taken for the process executing this BIF. */
-
-    FLAGS(rp) |= F_NEED_FULLSWEEP;
-    reds = erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity);
-
-#ifdef ERTS_SMP
-    if (BIF_P != rp) {
-	erts_resume(rp, ERTS_PROC_LOCK_MAIN);
-	erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MAIN);
-    }
-#endif
-
-    BIF_RET2(am_true, reds);
-}
-
 BIF_RETTYPE garbage_collect_0(BIF_ALIST_0)
 {
     int reds;
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index dc8e9101de..3b7bb56885 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -46,7 +46,6 @@ bif erlang:atom_to_list/1
 bif erlang:binary_to_list/1
 bif erlang:binary_to_list/3
 bif erlang:binary_to_term/1	
-bif erlang:check_process_code/2
 bif erlang:crc32/1
 bif erlang:crc32/2
 bif erlang:crc32_combine/3
@@ -67,7 +66,6 @@ bif erlang:float_to_list/1
 bif erlang:float_to_list/2
 bif erlang:fun_info/2
 bif erlang:garbage_collect/0
-bif erlang:garbage_collect/1
 bif erlang:get/0
 bif erlang:get/1
 bif erlang:get_keys/1
@@ -155,6 +153,10 @@ bif erts_internal:port_control/3
 bif erts_internal:port_close/1
 bif erts_internal:port_connect/2
 
+bif erts_internal:request_system_task/3
+bif erts_internal:check_process_code/2
+
+
 # inet_db support
 bif erlang:port_set_data/2
 bif erlang:port_get_data/1
diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c
index ad9a89b642..acbd700cc4 100644
--- a/erts/emulator/beam/break.c
+++ b/erts/emulator/beam/break.c
@@ -109,10 +109,12 @@ process_killer(void)
 		    erts_smp_proc_lock(rp, rp_locks);
 		    state = erts_smp_atomic32_read_acqb(&rp->state);
 		    if (state & (ERTS_PSFLG_FREE
-				  | ERTS_PSFLG_EXITING
-				  | ERTS_PSFLG_ACTIVE
-				  | ERTS_PSFLG_IN_RUNQ
-				  | ERTS_PSFLG_RUNNING)) {
+				 | ERTS_PSFLG_EXITING
+				 | ERTS_PSFLG_ACTIVE
+				 | ERTS_PSFLG_ACTIVE_SYS
+				 | ERTS_PSFLG_IN_RUNQ
+				 | ERTS_PSFLG_RUNNING
+				 | ERTS_PSFLG_RUNNING_SYS)) {
 			erts_printf("Can only kill WAITING processes this way\n");
 		    }
 		    else {
diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types
index bb5eba80be..32308fae9b 100644
--- a/erts/emulator/beam/erl_alloc.types
+++ b/erts/emulator/beam/erl_alloc.types
@@ -268,6 +268,8 @@ type	PROC_INTERVAL	LONG_LIVED	SYSTEM		process_interval
 type    BUSY_CALLER_TAB	SHORT_LIVED	SYSTEM		busy_caller_table
 type    BUSY_CALLER	SHORT_LIVED	SYSTEM		busy_caller
 type	PORT_DATA_HEAP	STANDARD	SYSTEM		port_data_heap
+type	PROC_SYS_TSK	SHORT_LIVED	PROCESSES	proc_sys_task
+type	PROC_SYS_TSK_QS	SHORT_LIVED	PROCESSES	proc_sys_task_queues
 
 +if threads_no_smp
 # Need thread safe allocs, but std_alloc and fix_alloc are not;
diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h
index 771eba431f..0f3bb8d281 100644
--- a/erts/emulator/beam/erl_message.h
+++ b/erts/emulator/beam/erl_message.h
@@ -46,6 +46,11 @@ typedef struct erl_off_heap {
     Uint64 overhead;     /* Administrative overhead (used to force GC). */
 } ErlOffHeap;
 
+#define ERTS_INIT_OFF_HEAP(OHP)			\
+    do {					\
+	(OHP)->first = NULL;			\
+	(OHP)->overhead = 0;			\
+    } while (0)
 #include "external.h"
 #include "erl_process.h"
 
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 434d5ca147..ad806da660 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -283,6 +283,40 @@ struct erts_system_profile_flags_t erts_system_profile_flags;
 #error "Need to store process_count in another type"
 #endif
 
+typedef enum {
+    ERTS_PSTT_GC,	/* Garbage Collect */
+    ERTS_PSTT_CPC	/* Check Process Code */
+} ErtsProcSysTaskType;
+
+#define ERTS_MAX_PROC_SYS_TASK_ARGS 2
+
+struct ErtsProcSysTask_ {
+    ErtsProcSysTask *next;
+    ErtsProcSysTask *prev;
+    ErtsProcSysTaskType type;
+    Eterm requester;
+    Eterm reply_tag;
+    Eterm req_id;
+    Uint req_id_sz;
+    Eterm arg[ERTS_MAX_PROC_SYS_TASK_ARGS];
+    ErlOffHeap off_heap;
+    Eterm heap[1];
+};
+
+#define ERTS_PROC_SYS_TASK_SIZE(HSz) \
+    (sizeof(ErtsProcSysTask) - sizeof(Eterm) + sizeof(Eterm)*(HSz))
+
+struct ErtsProcSysTaskQs_ {
+    int qmask;
+    int ncount;
+    ErtsProcSysTask *q[ERTS_NO_PROC_PRIO_LEVELS];
+};
+
+ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proc_sys_task_queues,
+				 ErtsProcSysTaskQs,
+				 50,
+				 ERTS_ALC_T_PROC_SYS_TSK_QS)
+
 ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(misc_op_list,
 				 ErtsMiscOpList,
 				 10,
@@ -353,6 +387,14 @@ static void aux_work_timeout_early_init(int no_schedulers);
 static void aux_work_timeout_late_init(void);
 static void setup_aux_work_timer(void);
 
+static int execute_sys_tasks(Process *c_p,
+			     erts_aint32_t *statep,
+			     int in_reds);
+static int cleanup_sys_tasks(Process *c_p,
+			     erts_aint32_t in_state,
+			     int in_reds);
+
+
 #if defined(DEBUG) || 0
 #define ERTS_DBG_CHK_AUX_WORK_VAL(V) dbg_chk_aux_work_val((V))
 static void
@@ -2848,7 +2890,7 @@ dequeue_process(ErtsRunQueue *runq, int prio_q, erts_aint32_t *statep)
     if (statep)
 	*statep = state;
 
-    prio = (int) (ERTS_PSFLG_PRIO_MASK & state);
+    prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state);
 
     rqi = &runq->procs.prio_info[prio];
 
@@ -2997,7 +3039,7 @@ immigrate(ErtsRunQueue *c_rq, ErtsMigrationPath *mp)
 		    erts_aint32_t state;
 		    state = erts_smp_atomic32_read_acqb(&proc->state);
 		    if (!(ERTS_PSFLG_BOUND & state)
-			&& (prio == (int) (ERTS_PSFLG_PRIO_MASK & state))) {
+			&& (prio == (int) ERTS_PSFLGS_GET_PRQ_PRIO(state))) {
 			ErtsRunQueueInfo *rqi = &rq->procs.prio_info[prio];
 			unqueue_process(rq, rpq, rqi, prio, prev_proc, proc);
 			erts_smp_runq_unlock(rq);
@@ -3079,7 +3121,7 @@ schedule_bound_processes(ErtsRunQueue *rq,
     while (proc) {
 	erts_aint32_t state = erts_smp_atomic32_read_acqb(&proc->state);
 	next = proc->next;
-	enqueue_process(rq, (int) (ERTS_PSFLG_PRIO_MASK & state), proc);
+	enqueue_process(rq, (int) ERTS_PSFLGS_GET_PRQ_PRIO(state), proc);
 	proc = next;
     }
 }
@@ -3184,7 +3226,7 @@ evacuate_run_queue(ErtsRunQueue *rq,
 		sbpp->last = proc;
 	    }
 	    else {
-		int prio = (int) (ERTS_PSFLG_PRIO_MASK & state);
+		int prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state);
 		erts_smp_runq_unlock(rq);
 
 		to_rq = mp->prio[prio].runq;
@@ -3257,7 +3299,7 @@ try_steal_task_from_victim(ErtsRunQueue *rq, int *rq_lockedp, ErtsRunQueue *vrq,
 	    erts_aint32_t state = erts_smp_atomic32_read_acqb(&proc->state);
 	    if (!(ERTS_PSFLG_BOUND & state)) {
 		/* Steal process */
-		int prio = (int) (ERTS_PSFLG_PRIO_MASK & state);
+		int prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state);
 		ErtsRunQueueInfo *rqi = &vrq->procs.prio_info[prio];
 		unqueue_process(vrq, rpq, rqi, prio, prev_proc, proc);
 		erts_smp_runq_unlock(vrq);
@@ -4575,6 +4617,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online)
 #endif
 
     init_misc_op_list_alloc();
+    init_proc_sys_task_queues_alloc();
 
 #ifdef ERTS_SMP
     set_wakeup_other_data();
@@ -4858,46 +4901,166 @@ erts_get_scheduler_data(void)
 
 #endif
 
+static Process *
+make_proxy_proc(Process *prev_proxy, Process *proc, erts_aint32_t prio)
+{
+    erts_aint32_t state;
+    Process *proxy;
+#ifdef ERTS_SMP
+    ErtsRunQueue *rq = RUNQ_READ_RQ(&proc->run_queue);
+#endif
+
+    state = (ERTS_PSFLG_PROXY
+	     | ERTS_PSFLG_IN_RUNQ
+	     | (((erts_aint32_t) 1) << (prio + ERTS_PSFLGS_IN_PRQ_MASK_OFFSET))
+	     | (prio << ERTS_PSFLGS_PRQ_PRIO_OFFSET)
+	     | (prio << ERTS_PSFLGS_USR_PRIO_OFFSET)
+	     | (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET));
+
+    if (prev_proxy) {
+	proxy = prev_proxy;
+	ASSERT(erts_smp_atomic32_read_nob(&proxy->state) & ERTS_PSFLG_PROXY);
+	erts_smp_atomic32_set_nob(&proxy->state, state);
+#ifdef ERTS_SMP
+	RUNQ_SET_RQ(&proc->run_queue, rq);
+#endif
+    }
+    else {
+	proxy = erts_alloc(ERTS_ALC_T_PROC, sizeof(Process));
+#ifdef DEBUG
+	{
+	    int i;
+	    Uint32 *ui32 = (Uint32 *) (char *) proxy;
+	    for (i = 0; i < sizeof(Process)/sizeof(Uint32); i++)
+		ui32[i] = (Uint32) 0xdeadbeef;
+	}
+#endif
+	erts_smp_atomic32_init_nob(&proxy->state, state);
+#ifdef ERTS_SMP
+	erts_smp_atomic_init_nob(&proxy->run_queue,
+				 erts_smp_atomic_read_nob(&proc->run_queue));
+#endif
+    }
+
+    proxy->common.id = proc->common.id;
+
+    return proxy;
+}
+
+static ERTS_INLINE void
+free_proxy_proc(Process *proxy)
+{
+    ASSERT(erts_smp_atomic32_read_nob(&proxy->state) & ERTS_PSFLG_PROXY);
+    erts_free(ERTS_ALC_T_PROC, proxy);
+}
+
+
+static ERTS_INLINE int
+check_enqueue_in_prio_queue(erts_aint32_t *prq_prio_p,
+			    erts_aint32_t *newp,
+			    erts_aint32_t actual)
+{
+    erts_aint32_t aprio, qbit, max_qbit;
+
+    aprio = (actual >> ERTS_PSFLGS_ACT_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK;
+    qbit = 1 << aprio;
+
+    *prq_prio_p = aprio;
+
+    max_qbit = (actual >> ERTS_PSFLGS_IN_PRQ_MASK_OFFSET) & ERTS_PSFLGS_QMASK;
+    max_qbit |= 1 << ERTS_PSFLGS_QMASK_BITS;
+    max_qbit &= -max_qbit;
+    /*
+     * max_qbit now either contain bit set for highest prio queue or a bit
+     * out of range (which will have a value larger than valid range).
+     */
+
+    if (qbit >= max_qbit)
+	return 0; /* Already queued in higher or equal prio */
+
+    /* Need to enqueue (if already enqueued, it is in lower prio) */
+    *newp |= qbit << ERTS_PSFLGS_IN_PRQ_MASK_OFFSET;
+
+    if ((actual & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLGS_USR_PRIO_MASK))
+	!= (aprio << ERTS_PSFLGS_USR_PRIO_OFFSET)) {
+	/*
+	 * Process struct already enqueued, or actual prio not
+	 * equal to user prio, i.e., enqueue using proxy.
+	 */
+	return -1;
+    }
+
+    /*
+     * Enqueue using process struct.
+     */
+    *newp &= ~ERTS_PSFLGS_PRQ_PRIO_MASK;
+    *newp |= ERTS_PSFLG_IN_RUNQ | (aprio << ERTS_PSFLGS_PRQ_PRIO_OFFSET);
+    return 1;
+}
+
 /*
  * scheduler_out_process() return with c_rq locked.
  */
 static ERTS_INLINE int
-schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p)
+schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Process *proxy)
 {
-    erts_aint32_t a, e, n;
+    erts_aint32_t a, e, n, enq_prio = -1;
     int res = 0;
+    int enqueue; /* < 0 -> use proxy */
 
     a = state;
 
     while (1) {
 	n = e = a;
 
-	ASSERT(a & ERTS_PSFLG_RUNNING);
-	ASSERT(!(a & ERTS_PSFLG_IN_RUNQ));
+	ASSERT(a & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS));
 
-	n &= ~ERTS_PSFLG_RUNNING;
-	if ((a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE)
-	    n |= ERTS_PSFLG_IN_RUNQ;
+	enqueue = 0;
+
+	n &= ~(ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS);
+	if (a & ERTS_PSFLG_ACTIVE_SYS
+	    || (a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) {
+	    enqueue = check_enqueue_in_prio_queue(&enq_prio, &n, a);
+	}
 	a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
 	if (a == e)
 	    break;
     }
 
-    if (!(n & ERTS_PSFLG_IN_RUNQ)) {
-	if (erts_system_profile_flags.runnable_procs)
-	    profile_runnable_proc(p, am_inactive);
+    if (!enqueue) {
+
+	if (erts_system_profile_flags.runnable_procs) {
+
+	    if (!(a & ERTS_PSFLG_ACTIVE_SYS)
+		&& (!(a & ERTS_PSFLG_ACTIVE)
+		    || (a & ERTS_PSFLG_SUSPENDED))) {
+		/* Process inactive */
+		profile_runnable_proc(p, am_inactive);
+	    }
+	}
+
+	if (proxy)
+	    free_proxy_proc(proxy);
     }
     else {
-	int prio = (int) (ERTS_PSFLG_PRIO_MASK & n);
+	Process *sched_p;
 	ErtsRunQueue *runq = erts_get_runq_proc(p);
 
-	ASSERT(!(n & ERTS_PSFLG_SUSPENDED));
+	ASSERT(!(n & ERTS_PSFLG_SUSPENDED) || (n & ERTS_PSFLG_ACTIVE_SYS));
+
+	if (enqueue < 0)
+	    sched_p = make_proxy_proc(proxy, p, enq_prio);
+	else {
+	    sched_p = p;
+	    if (proxy)
+		free_proxy_proc(proxy);
+	}
 
 #ifdef ERTS_SMP
 	if (!(ERTS_PSFLG_BOUND & n)) {
-	    ErtsRunQueue *new_runq = erts_check_emigration_need(runq, prio);
+	    ErtsRunQueue *new_runq = erts_check_emigration_need(runq, enq_prio);
 	    if (new_runq) {
-		RUNQ_SET_RQ(&p->run_queue, new_runq);
+		RUNQ_SET_RQ(&sched_p->run_queue, new_runq);
 		runq = new_runq;
 	    }
 	}
@@ -4908,7 +5071,7 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p)
 	erts_smp_runq_lock(runq);
 
 	/* Enqueue the process */
-	enqueue_process(runq, prio, p);
+	enqueue_process(runq, (int) enq_prio, sched_p);
 
 	if (runq == c_rq)
 	    return res;
@@ -4920,14 +5083,13 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p)
 }
 
 static ERTS_INLINE void
-add2runq(Process *p, erts_aint32_t state)
+add2runq(Process *p, erts_aint32_t state, erts_aint32_t prio)
 {
-    int prio = (int) (ERTS_PSFLG_PRIO_MASK & state);
     ErtsRunQueue *runq = erts_get_runq_proc(p);
 
 #ifdef ERTS_SMP
     if (!(ERTS_PSFLG_BOUND & state)) {
-	ErtsRunQueue *new_runq = erts_check_emigration_need(runq, prio);
+	ErtsRunQueue *new_runq = erts_check_emigration_need(runq, (int) prio);
 	if (new_runq) {
 	    RUNQ_SET_RQ(&p->run_queue, new_runq);
 	    runq = new_runq;
@@ -4939,101 +5101,236 @@ add2runq(Process *p, erts_aint32_t state)
     erts_smp_runq_lock(runq);
 
     /* Enqueue the process */
-    enqueue_process(runq, prio, p);
+    enqueue_process(runq, (int) prio, p);
 
     erts_smp_runq_unlock(runq);
     smp_notify_inc_runq(runq);
 
 }
 
-static ERTS_INLINE void
-schedule_process(Process *p, erts_aint32_t state, int active_enq)
+static ERTS_INLINE int
+change_proc_schedule_state(Process *p,
+			   erts_aint32_t clear_state_flags,
+			   erts_aint32_t set_state_flags,
+			   erts_aint32_t *statep,
+			   erts_aint32_t *enq_prio_p)
 {
-    erts_aint32_t a = state, n;
+    /*
+     * NOTE: ERTS_PSFLG_RUNNING, ERTS_PSFLG_RUNNING_SYS and
+     *       ERTS_PSFLG_ACTIVE_SYS are not allowed to be
+     *       altered by this function!
+     */
+    erts_aint32_t a = *statep, n;
+    int enqueue; /* < 0 -> use proxy */
+
+    ASSERT(!(a & ERTS_PSFLG_PROXY));
+    ASSERT((clear_state_flags & (ERTS_PSFLG_RUNNING
+				 | ERTS_PSFLG_RUNNING_SYS
+				 | ERTS_PSFLG_ACTIVE_SYS)) == 0);
+    ASSERT((set_state_flags & (ERTS_PSFLG_RUNNING
+			       | ERTS_PSFLG_RUNNING_SYS
+			       | ERTS_PSFLG_ACTIVE_SYS)) == 0);
 
     while (1) {
 	erts_aint32_t e;
 	n = e = a;
 
+	enqueue = 0;
+
 	if (a & ERTS_PSFLG_FREE)
-	    return; /* We don't want to schedule free processes... */
-	n |= ERTS_PSFLG_ACTIVE;
-	if (!(a & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_RUNNING)))
-	    n |= ERTS_PSFLG_IN_RUNQ;
+	    break; /* We don't want to schedule free processes... */
+
+	if (clear_state_flags)
+	    n &= ~clear_state_flags;
+
+	if (set_state_flags)
+	    n |= set_state_flags;
+
+	if ((n & (ERTS_PSFLG_SUSPENDED
+		  | ERTS_PSFLG_RUNNING
+		  | ERTS_PSFLG_RUNNING_SYS
+		  | ERTS_PSFLG_IN_RUNQ
+		  | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) {
+	    /*
+	     * Active and seemingly need to be enqueued, but
+	     * process may be in a run queue via proxy, need
+	     * further inspection...
+	     */
+	    enqueue = check_enqueue_in_prio_queue(enq_prio_p, &n, a);
+	}
+
 	a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
 	if (a == e)
 	    break;
-	if (!active_enq && (a & ERTS_PSFLG_ACTIVE))
-	    return; /* Someone else activated process ... */
+	if (enqueue == 0 && n == a)
+	    break;
     }
 
-    if (erts_system_profile_flags.runnable_procs
-	&& !(a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED))) {
-    	profile_runnable_proc(p, am_active);
+    if (erts_system_profile_flags.runnable_procs) {
+
+	if (((n & (ERTS_PSFLG_SUSPENDED
+		   | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE)
+	    && (!(a & (ERTS_PSFLG_ACTIVE_SYS
+		       | ERTS_PSFLG_RUNNING
+		       | ERTS_PSFLG_RUNNING_SYS)
+		  && (!(a & ERTS_PSFLG_ACTIVE)
+		      || (a & ERTS_PSFLG_SUSPENDED))))) {
+	    /* We activated a prevously inactive process */
+	    profile_runnable_proc(p, am_active);
+	}
+
     }
 
-    if ((n & ERTS_PSFLG_IN_RUNQ) && !(a & ERTS_PSFLG_IN_RUNQ))
-	add2runq(p, n);
+    *statep = a;
+
+    return enqueue;
+}
+
+static ERTS_INLINE void
+schedule_process(Process *p, erts_aint32_t in_state)
+{
+    erts_aint32_t enq_prio  = -1;
+    erts_aint32_t state = in_state;
+    int enqueue = change_proc_schedule_state(p,
+					     0,
+					     ERTS_PSFLG_ACTIVE,
+					     &state,
+					     &enq_prio);
+    if (enqueue)
+	add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio),
+		 state,
+		 enq_prio);
 }
 
 void
 erts_schedule_process(Process *p, erts_aint32_t state)
 {
-    schedule_process(p, state, 0);
+    schedule_process(p, state);
+}
+
+static void
+schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy)
+{
+    erts_aint32_t a = state, n, enq_prio = -1;
+    int enqueue; /* < 0 -> use proxy */
+
+    ASSERT(!(state & ERTS_PSFLG_PROXY));
+
+    while (1) {
+	erts_aint32_t e;
+	n = e = a;
+
+	if (a & ERTS_PSFLG_FREE)
+	    return; /* We don't want to schedule free processes... */
+
+	enqueue = 0;
+	n |= ERTS_PSFLG_ACTIVE_SYS;
+	if (!(a & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)))
+	    enqueue = check_enqueue_in_prio_queue(&enq_prio, &n, a);
+	a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
+	if (a == e)
+	    break;
+	if (a == n && !enqueue)
+	    goto cleanup;
+    }
+
+    if (erts_system_profile_flags.runnable_procs) {
+
+	if (!(a & (ERTS_PSFLG_ACTIVE_SYS
+		   | ERTS_PSFLG_RUNNING
+		   | ERTS_PSFLG_RUNNING_SYS))
+	    && (!(a & ERTS_PSFLG_ACTIVE) || (a & ERTS_PSFLG_SUSPENDED))) {
+	    /* We activated a prevously inactive process */
+	    profile_runnable_proc(p, am_active);
+	}
+
+    }
+
+    if (enqueue) {
+	Process *sched_p;
+	if (enqueue > 0)
+	    sched_p = p;
+	else {
+	    sched_p = make_proxy_proc(proxy, p, enq_prio);
+	    proxy = NULL;
+	}
+	add2runq(sched_p, n, enq_prio);
+    }
+
+cleanup:
+    if (proxy)
+	free_proxy_proc(proxy);
 }
 
 static ERTS_INLINE int
 suspend_process(Process *c_p, Process *p)
 {
-    erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state);
+    erts_aint32_t state;
     int suspended = 0;
     ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p));
 
+    state = erts_smp_atomic32_read_acqb(&p->state);
+
     if ((state & ERTS_PSFLG_SUSPENDED))
 	suspended = -1;
     else {
 	if (c_p == p) {
 	    state = erts_smp_atomic32_read_bor_relb(&p->state,
 						    ERTS_PSFLG_SUSPENDED);
-	    state |= ERTS_PSFLG_SUSPENDED;
 	    ASSERT(state & ERTS_PSFLG_RUNNING);
-	    suspended = 1;
+	    suspended = (state & ERTS_PSFLG_SUSPENDED) ? -1: 1;
 	}
 	else {
 	    while (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_EXITING))) {
-		erts_aint32_t e, n;
+		erts_aint32_t n, e;
+
 		n = e = state;
 		n |= ERTS_PSFLG_SUSPENDED;
 		state = erts_smp_atomic32_cmpxchg_relb(&p->state, n, e);
 		if (state == e) {
-		    state = n;
 		    suspended = 1;
 		    break;
 		}
+		if (state & ERTS_PSFLG_SUSPENDED) {
+		    suspended = -1;
+		    break;
+		}
 	    }
 	}
     }
 
-    if (state & ERTS_PSFLG_SUSPENDED) {
+    if (suspended) {
 
 	ASSERT(!(ERTS_PSFLG_RUNNING & state)
 	       || p == erts_get_current_process());
 
-	if (erts_system_profile_flags.runnable_procs
-	    && (p->rcount == 0)
-	    && (state & ERTS_PSFLG_ACTIVE)) {
-	    profile_runnable_proc(p, am_inactive);
+	if (suspended > 0 && erts_system_profile_flags.runnable_procs) {
+
+	    /* 'state' is before our change... */
+
+	    if ((state & (ERTS_PSFLG_ACTIVE
+			  | ERTS_PSFLG_ACTIVE_SYS
+			  | ERTS_PSFLG_RUNNING
+			  | ERTS_PSFLG_RUNNING_SYS
+			  | ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) {
+		/* We made process inactive */
+		profile_runnable_proc(p, am_inactive);
+	    }
+
 	}
 
 	p->rcount++;  /* count number of suspend */
     }
+
     return suspended;
 }
 
 static ERTS_INLINE void
 resume_process(Process *p)
 {
-    erts_aint32_t state;
+    erts_aint32_t state, enq_prio = -1;
+    int enqueue;
+
     ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p));
 
     ASSERT(p->rcount > 0);
@@ -5041,14 +5338,16 @@ resume_process(Process *p)
     if (--p->rcount > 0)  /* multiple suspend */
 	return;
 
-    state = erts_smp_atomic32_read_band_mb(&p->state, ~ERTS_PSFLG_SUSPENDED);
-    state &= ~ERTS_PSFLG_SUSPENDED;
-    if ((state & (ERTS_PSFLG_EXITING
-		  | ERTS_PSFLG_ACTIVE
-		  | ERTS_PSFLG_IN_RUNQ
-		  | ERTS_PSFLG_RUNNING)) == ERTS_PSFLG_ACTIVE) {
-	schedule_process(p, state, 1);
-    }
+    state = erts_smp_atomic32_read_nob(&p->state);
+    enqueue = change_proc_schedule_state(p,
+					 ERTS_PSFLG_SUSPENDED,
+					 0,
+					 &state,
+					 &enq_prio);
+    if (enqueue)
+	add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio),
+		 state,
+		 enq_prio);
 }
 
 int
@@ -6032,7 +6331,8 @@ pid2proc_not_running(Process *c_p, ErtsProcLocks c_p_locks,
 		goto done;
 	}
 	else {
-	    if (!(ERTS_PSFLG_RUNNING & erts_smp_atomic32_read_acqb(&rp->state)))
+	    if (!((ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)
+		  & erts_smp_atomic32_read_acqb(&rp->state)))
 		goto done;
 
 	}
@@ -6695,7 +6995,7 @@ Eterm
 erts_get_process_priority(Process *p)
 {
     erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state);
-    switch (state & ERTS_PSFLG_PRIO_MASK) {
+    switch (ERTS_PSFLGS_GET_USR_PRIO(state)) {
     case PRIORITY_MAX:		return am_max;
     case PRIORITY_HIGH:		return am_high;
     case PRIORITY_NORMAL:	return am_normal;
@@ -6718,18 +7018,60 @@ erts_set_process_priority(Process *p, Eterm value)
     }
 
     a = erts_smp_atomic32_read_nob(&p->state);
-    if (nprio == (a & ERTS_PSFLG_PRIO_MASK))
+    if (nprio == ERTS_PSFLGS_GET_USR_PRIO(a))
 	oprio = nprio;
     else {
-	erts_aint32_t e, n;
+	int slocked = 0;
+	erts_aint32_t e, n, aprio;
+
+	if (a & ERTS_PSFLG_ACTIVE_SYS) {
+	    erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS);
+	    slocked = 1;
+	}
+
 	do {
-	    oprio = a & ERTS_PSFLG_PRIO_MASK;
+	    oprio = ERTS_PSFLGS_GET_USR_PRIO(a);
 	    n = e = a;
 
-	    ASSERT(!(a & ERTS_PSFLG_IN_RUNQ));
+	    if (!(a & ERTS_PSFLG_ACTIVE_SYS))
+		aprio = nprio;
+	    else {
+		int max_qbit;
+
+		if (!slocked) {
+		    erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS);
+		    slocked = 1;
+		}
+
+		max_qbit = p->sys_task_qs->qmask;
+		max_qbit &= -max_qbit;
+		switch (max_qbit) {
+		case MAX_BIT:
+		    aprio = PRIORITY_MAX;
+		    break;
+		case HIGH_BIT:
+		    aprio = PRIORITY_HIGH;
+		    break;
+		case NORMAL_BIT:
+		    aprio = PRIORITY_NORMAL;
+		    break;
+		case LOW_BIT:
+		    aprio = PRIORITY_LOW;
+		    break;
+		default:
+		    ERTS_INTERNAL_ERROR("Invalid qmask");
+		    aprio = -1;
+		}
+
+		if (aprio > nprio) /* low value -> high prio */
+		    aprio = nprio;
+	    }
+
+	    n &= ~(ERTS_PSFLGS_USR_PRIO_MASK
+		   | ERTS_PSFLGS_ACT_PRIO_MASK);
+	    n |= ((nprio << ERTS_PSFLGS_USR_PRIO_OFFSET)
+		  | (aprio << ERTS_PSFLGS_ACT_PRIO_OFFSET));
 
-	    n &= ~ERTS_PSFLG_PRIO_MASK;
-	    n |= nprio;
 	    a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
 	} while (a != e);
     }
@@ -6763,6 +7105,7 @@ erts_set_process_priority(Process *p, Eterm value)
 
 Process *schedule(Process *p, int calls)
 {
+    Process *proxy_p = NULL;
     ErtsRunQueue *rq;
     erts_aint_t dt;
     ErtsSchedulerData *esdp;
@@ -6805,6 +7148,8 @@ Process *schedule(Process *p, int calls)
 	actual_reds = reds = 0;
 	erts_smp_runq_lock(rq);
     } else {
+    sched_out_proc:
+
 #ifdef ERTS_SMP
 	ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p);
 	esdp = p->scheduler_data;
@@ -6858,10 +7203,11 @@ Process *schedule(Process *p, int calls)
 
 	esdp->reductions += reds;
 
-	schedule_out_process(rq, state, p); /* Returns with rq locked! */
+	schedule_out_process(rq, state, p, proxy_p); /* Returns with rq locked! */
+	proxy_p = NULL;
 
 	ERTS_PROC_REDUCTIONS_EXECUTED(rq,
-				      (int) (state & ERTS_PSFLG_PRIO_MASK),
+				      (int) ERTS_PSFLGS_GET_USR_PRIO(state),
 				      reds,
 				      actual_reds);
 
@@ -6870,18 +7216,19 @@ Process *schedule(Process *p, int calls)
 	p->scheduler_data = NULL;
 #endif
 
+	erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS);
 
 	if (state & ERTS_PSFLG_FREE) {
 #ifdef ERTS_SMP
 	    ASSERT(esdp->free_process == p);
 	    esdp->free_process = NULL;
-#else	    
-	    erts_free_proc(p);
+#else
+	    state = erts_smp_atomic32_read_nob(&p->state);
+	    if (!(state & ERTS_PSFLG_IN_RUNQ))
+		erts_free_proc(p);
 #endif
 	}
 
-	erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS);
-
 #ifdef ERTS_SMP
 	ASSERT(!esdp->free_process);
 #endif
@@ -7088,6 +7435,8 @@ Process *schedule(Process *p, int calls)
 	 * Find a new process to run.
 	 */
     pick_next_process: {
+	    erts_aint32_t psflg_band_mask;
+	    erts_aint32_t running_flag;
 	    int prio_q;
 	    int qmask;
 
@@ -7121,18 +7470,62 @@ Process *schedule(Process *p, int calls)
 
 	    ASSERT(p); /* Wrong qmask in rq->flags? */
 
+	    psflg_band_mask = ~(((erts_aint32_t) 1) << (ERTS_PSFLGS_GET_PRQ_PRIO(state)
+							+ ERTS_PSFLGS_IN_PRQ_MASK_OFFSET));
+
+	    if (!(state & ERTS_PSFLG_PROXY))
+		psflg_band_mask &= ~ERTS_PSFLG_IN_RUNQ;
+	    else {
+		proxy_p = p;
+		p = erts_proc_lookup_raw(proxy_p->common.id);
+		if (!p) {
+		    free_proxy_proc(proxy_p);
+		    proxy_p = NULL;
+		    goto pick_next_process;
+		}
+		state = erts_smp_atomic32_read_nob(&p->state);
+	    }
+
+
+	    if (state & ERTS_PSFLG_ACTIVE_SYS)
+		running_flag = ERTS_PSFLG_RUNNING_SYS;
+	    else
+		running_flag = ERTS_PSFLG_RUNNING;
+
 	    while (1) {
 		erts_aint32_t exp, new, tmp;
 		tmp = new = exp = state;
-		new &= ~ERTS_PSFLG_IN_RUNQ;
-		tmp = state & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT);
-		if (tmp != ERTS_PSFLG_SUSPENDED)
-		    new |= ERTS_PSFLG_RUNNING;
+		new &= psflg_band_mask;
+		if (!(state & (ERTS_PSFLG_RUNNING
+			       | ERTS_PSFLG_RUNNING_SYS))) {
+		    tmp = state & (ERTS_PSFLG_SUSPENDED
+				   | ERTS_PSFLG_PENDING_EXIT
+				   | ERTS_PSFLG_ACTIVE_SYS);
+		    if (tmp != ERTS_PSFLG_SUSPENDED)
+			new |= running_flag;
+		}
 		state = erts_smp_atomic32_cmpxchg_relb(&p->state, new, exp);
 		if (state == exp) {
-		    tmp = state & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT);
-		    if (tmp == ERTS_PSFLG_SUSPENDED)
+		    if ((state & (ERTS_PSFLG_RUNNING
+				  | ERTS_PSFLG_RUNNING_SYS
+				  | ERTS_PSFLG_FREE))
+			|| ((state & (ERTS_PSFLG_SUSPENDED
+				      | ERTS_PSFLG_PENDING_EXIT
+				      | ERTS_PSFLG_ACTIVE_SYS))
+			    == ERTS_PSFLG_SUSPENDED)) {
+			if (state & ERTS_PSFLG_FREE) {
+#ifdef ERTS_SMP
+			    erts_smp_proc_dec_refc(p);
+#else
+			    erts_free_proc(p);
+#endif
+			}
+			if (proxy_p) {
+			    free_proxy_proc(proxy_p);
+			    proxy_p = NULL;
+			}
 			goto pick_next_process;
+		    }
 		    state = new;
 		    break;
 		}
@@ -7162,7 +7555,7 @@ Process *schedule(Process *p, int calls)
 					  (UWord) esdp->no);
 	    int migrated = old && old != esdp->no;
 
-	    prio = (int) (state & ERTS_PSFLG_PRIO_MASK);
+	    prio = (int) ERTS_PSFLGS_GET_USR_PRIO(state);
 
 	    erts_smp_spin_lock(&erts_sched_stat.lock);
 	    erts_sched_stat.prio[prio].total_executed++;
@@ -7182,9 +7575,6 @@ Process *schedule(Process *p, int calls)
 	ASSERT(!p->scheduler_data);
 	p->scheduler_data = esdp;
 #endif
-	/* Never run a suspended process */
-	ASSERT(!(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&p->state)));
-
 	reds = context_reds;
 
 	if (IS_TRACED(p)) {
@@ -7210,21 +7600,574 @@ Process *schedule(Process *p, int calls)
 	    erts_check_my_tracer_proc(p);
 #endif
 
+	if (state & ERTS_PSFLG_RUNNING_SYS) {
+	    reds -= execute_sys_tasks(p, &state, reds);
+	    if (reds <= 0) {
+		p->fcalls = reds;
+		goto sched_out_proc;
+	    }
+
+	    ASSERT(state & ERTS_PSFLG_RUNNING_SYS);
+	    ASSERT(!(state & ERTS_PSFLG_RUNNING));
+
+	    while (1) {
+		erts_aint32_t n, e;
+
+		if (((state & (ERTS_PSFLG_SUSPENDED
+			       | ERTS_PSFLG_ACTIVE)) != ERTS_PSFLG_ACTIVE)
+		    && !(state & ERTS_PSFLG_EXITING))
+		    goto sched_out_proc;
+
+		n = e = state;
+		n &= ~ERTS_PSFLG_RUNNING_SYS;
+		n |= ERTS_PSFLG_RUNNING;
+
+		state = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
+		if (state == e) {
+		    state = n;
+		    break;
+		}
+
+		ASSERT(state & ERTS_PSFLG_RUNNING_SYS);
+		ASSERT(!(state & ERTS_PSFLG_RUNNING));
+	    }
+	}
+
 	if (!(state & ERTS_PSFLG_EXITING)
 	    && ((FLAGS(p) & F_FORCE_GC)
 		|| (MSO(p).overhead > BIN_VHEAP_SZ(p)))) {
 	    reds -= erts_garbage_collect(p, 0, p->arg_reg, p->arity);
-	    if (reds < 0) {
-		reds = 1;
+	    if (reds <= 0) {
+		p->fcalls = reds;
+		goto sched_out_proc;
 	    }
 	}
+
+	if (proxy_p) {
+	    free_proxy_proc(proxy_p);
+	    proxy_p = NULL;
+	}
 	    
 	p->fcalls = reds;
 	ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p);
+
+	/* Never run a suspended process */
+	ASSERT(!(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&p->state)));
+
 	return p;
     }
 }
 
+static int
+notify_sys_task_executed(Process *c_p, ErtsProcSysTask *st, Eterm st_result)
+{
+    Process *rp = erts_proc_lookup(st->requester);
+    if (rp) {
+	ErtsProcLocks rp_locks;
+	ErlOffHeap *ohp;
+	ErlHeapFragment* bp;
+	Eterm *hp, msg, req_id, result;
+	Uint st_result_sz, hsz;
+#ifdef DEBUG
+	Eterm *hp_start;
+#endif
+
+	rp_locks = (c_p == rp) ? ERTS_PROC_LOCK_MAIN : 0;
+
+	st_result_sz = is_immed(st_result) ? 0 : size_object(st_result);
+	hsz = st->req_id_sz + st_result_sz + 4 /* 3-tuple */;
+
+	hp = erts_alloc_message_heap(hsz,
+				     &bp,
+				     &ohp,
+				     rp,
+				     &rp_locks);
+
+#ifdef DEBUG
+	hp_start = hp;
+#endif
+
+	req_id = st->req_id_sz == 0 ? st->req_id : copy_struct(st->req_id,
+							       st->req_id_sz,
+							       &hp,
+							       ohp);
+
+	result = st_result_sz == 0 ? st_result :  copy_struct(st_result,
+							      st_result_sz,
+							      &hp,
+							      ohp);
+
+	ASSERT(is_immed(st->reply_tag));
+
+	msg = TUPLE3(hp, st->reply_tag, req_id, result);
+
+#ifdef DEBUG
+	hp += 4;
+	ASSERT(hp_start + hsz == hp);
+#endif
+
+	erts_queue_message(rp,
+			   &rp_locks,
+			   bp,
+			   msg,
+			   NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+	    );
+
+	if (c_p == rp)
+	    rp_locks &= ~ERTS_PROC_LOCK_MAIN;
+
+	if (rp_locks)
+	    erts_smp_proc_unlock(rp, rp_locks);
+    }
+
+    erts_cleanup_offheap(&st->off_heap);
+
+    erts_free(ERTS_ALC_T_PROC_SYS_TSK, st);
+
+    return rp ? 1 : 0;
+}
+
+static ERTS_INLINE ErtsProcSysTask *
+fetch_sys_task(Process *c_p, erts_aint32_t state, int *qmaskp)
+{
+    ErtsProcSysTaskQs *unused_qs = NULL;
+    int qbit, qmask;
+    ErtsProcSysTask *st, **qp;
+
+    erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_STATUS);
+
+    if (!c_p->sys_task_qs) {
+	qmask = 0;
+	st = NULL;
+	goto update_state;
+    }
+
+    qmask = c_p->sys_task_qs->qmask;
+
+    if ((state & (ERTS_PSFLG_ACTIVE
+		  | ERTS_PSFLG_EXITING
+		  | ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) {
+	/* No sys tasks if we got exclusively higher prio user work to do */
+	st = NULL;
+	switch (ERTS_PSFLGS_GET_USR_PRIO(state)) {
+	case PRIORITY_MAX:
+	    if (!(qmask & MAX_BIT))
+		goto done;
+	    break;
+	case PRIORITY_HIGH:
+	    if (!(qmask & (MAX_BIT|HIGH_BIT)))
+		goto done;
+	    break;
+	default:
+	    break;
+	}
+    }
+
+    qbit = qmask & -qmask;
+    switch (qbit) {
+    case MAX_BIT:
+	qp = &c_p->sys_task_qs->q[PRIORITY_MAX];
+	break;
+    case HIGH_BIT:
+	qp = &c_p->sys_task_qs->q[PRIORITY_HIGH];
+	break;
+    case NORMAL_BIT:
+	if (!(qmask & PRIORITY_LOW)
+	    || ++c_p->sys_task_qs->ncount <= RESCHEDULE_LOW) {
+	    qp = &c_p->sys_task_qs->q[PRIORITY_NORMAL];
+	    break;
+	}
+	c_p->sys_task_qs->ncount = 0;
+	/* Fall through */
+    case LOW_BIT:
+	qp = &c_p->sys_task_qs->q[PRIORITY_LOW];
+	break;
+    default:
+	ERTS_INTERNAL_ERROR("Invalid qmask");
+    }
+
+    st = *qp;
+    ASSERT(st);
+    if (st->next != st) {
+	*qp = st->next;
+	st->next->prev = st->prev;
+	st->prev->next = st->next;
+    }
+    else {
+	erts_aint32_t a, e, n, st_prio, qmask2;
+
+	*qp = NULL;
+	qmask &= ~qbit;
+	c_p->sys_task_qs->qmask = qmask;
+
+    update_state:
+
+	qmask2 = qmask;
+
+	switch (qmask2 & -qmask2) {
+	case MAX_BIT:
+	    st_prio = PRIORITY_MAX;
+	    break;
+	case HIGH_BIT:
+	    st_prio = PRIORITY_HIGH;
+	    break;
+	case NORMAL_BIT:
+	    st_prio = PRIORITY_NORMAL;
+	    break;
+	case LOW_BIT:
+	    st_prio = PRIORITY_LOW;
+	    break;
+	case 0:
+	    st_prio = PRIORITY_LOW;
+	    unused_qs = c_p->sys_task_qs;
+	    c_p->sys_task_qs = NULL;
+	    break;
+	default:
+	    ERTS_INTERNAL_ERROR("Invalid qmask");
+	}
+
+	a = state;
+	do {
+	    erts_aint32_t prio = ERTS_PSFLGS_GET_USR_PRIO(a);
+
+	    if (prio > st_prio)
+		prio = st_prio;
+
+	    n = e = a;
+
+	    n &= ~ERTS_PSFLGS_ACT_PRIO_MASK;
+	    n |= (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET);
+
+	    if (!qmask)
+		n &= ~ERTS_PSFLG_ACTIVE_SYS;
+
+	    if (a == n)
+		break;
+	    a = erts_smp_atomic32_cmpxchg_nob(&c_p->state, n, e);
+	} while (a != e);
+    }
+
+done:
+
+    erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_STATUS);
+
+    if (unused_qs)
+	proc_sys_task_queues_free(unused_qs);
+
+    *qmaskp = qmask;
+
+    return st;
+}
+
+static int
+execute_sys_tasks(Process *c_p, erts_aint32_t *statep, int in_reds)
+{
+    int garbage_collected = 0;
+    erts_aint32_t state = *statep;
+    int max_reds = in_reds;
+    int reds = 0;
+    int qmask = 0;
+
+    ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCK_MAIN);
+
+    do {
+	ErtsProcSysTask *st;
+	Eterm st_res;
+
+	if (state & (ERTS_PSFLG_EXITING|ERTS_PSFLG_PENDING_EXIT)) {
+#ifdef ERTS_SMP
+	    if (state & ERTS_PSFLG_PENDING_EXIT)
+		erts_handle_pending_exit(c_p, ERTS_PROC_LOCK_MAIN);
+#endif
+	    ASSERT(ERTS_PROC_IS_EXITING(c_p));
+	    break;
+	}
+
+	st = fetch_sys_task(c_p, state, &qmask);
+	if (!st)
+	    break;
+
+	switch (st->type) {
+	case ERTS_PSTT_GC:
+	    if (!garbage_collected) {
+		FLAGS(c_p) |= F_NEED_FULLSWEEP;
+		reds += erts_garbage_collect(c_p, 0, c_p->arg_reg, c_p->arity);
+		garbage_collected = 1;
+	    }
+	    st_res = am_true;
+	    break;
+	case ERTS_PSTT_CPC:
+	    st_res = erts_check_process_code(c_p,
+					     st->arg[0],
+					     st->arg[1] == am_true,
+					     &reds);
+	    break;
+	default:
+	    ERTS_INTERNAL_ERROR("Invalid process sys task type");
+	    st_res = am_false;
+	}
+
+	if (st)
+	    reds += notify_sys_task_executed(c_p, st, st_res);
+
+	state = erts_smp_atomic32_read_acqb(&c_p->state);
+    } while (qmask && reds < max_reds);
+
+    *statep = state;
+
+    return reds;
+}
+
+static int
+cleanup_sys_tasks(Process *c_p, erts_aint32_t in_state, int in_reds)
+{
+    erts_aint32_t state = in_state;
+    int max_reds = in_reds;
+    int reds = 0;
+    int qmask = 0;
+
+    ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCK_MAIN);
+
+    do {
+	ErtsProcSysTask *st;
+	Eterm st_res;
+
+	st = fetch_sys_task(c_p, state, &qmask);
+	if (!st)
+	    break;
+
+	switch (st->type) {
+	case ERTS_PSTT_GC:
+	    st_res = am_false;
+	    break;
+	case ERTS_PSTT_CPC:
+	    st_res = am_false;
+	    break;
+	default:
+	    ERTS_INTERNAL_ERROR("Invalid process sys task type");
+	    st_res = am_false;
+	    break;
+	}
+
+	reds += notify_sys_task_executed(c_p, st, st_res);
+
+	state = erts_smp_atomic32_read_acqb(&c_p->state);
+    } while (qmask && reds < max_reds);
+
+    return reds;
+}
+
+BIF_RETTYPE
+erts_internal_request_system_task_3(BIF_ALIST_3)
+{
+    Process *rp = erts_proc_lookup(BIF_ARG_1);
+    ErtsProcSysTaskQs *stqs, *free_stqs = NULL;
+    ErtsProcSysTask *st = NULL;
+    erts_aint32_t prio, rp_state;
+    int rp_locked;
+    Eterm noproc_res, req_type;
+
+    if (!rp && !is_internal_pid(BIF_ARG_1)) {
+	if (!is_external_pid(BIF_ARG_1))
+	    goto badarg;
+	if (external_pid_dist_entry(BIF_ARG_1) != erts_this_dist_entry)
+	    goto badarg;
+    }
+
+    switch (BIF_ARG_2) {
+    case am_max:	prio = PRIORITY_MAX;	break;
+    case am_high:	prio = PRIORITY_HIGH;	break;
+    case am_normal:	prio = PRIORITY_NORMAL;	break;
+    case am_low:	prio = PRIORITY_LOW;	break;
+    default: goto badarg;
+    }
+
+    if (is_not_tuple(BIF_ARG_3))
+	goto badarg;
+    else {
+	int i;
+	Eterm *tp = tuple_val(BIF_ARG_3);
+	Uint arity = arityval(*tp);
+	Eterm req_id;
+	Uint req_id_sz;
+	Eterm arg[ERTS_MAX_PROC_SYS_TASK_ARGS];
+	Uint arg_sz[ERTS_MAX_PROC_SYS_TASK_ARGS];
+	Uint tot_sz;
+	Eterm *hp;
+
+	if (arity < 2)
+	    goto badarg;
+	if (arity > 2 + ERTS_MAX_PROC_SYS_TASK_ARGS)
+	    goto badarg;
+	req_type = tp[1];
+	req_id = tp[2];
+	req_id_sz = is_immed(req_id) ? req_id : size_object(req_id);
+	tot_sz = req_id_sz;
+	for (i = 0; i < ERTS_MAX_PROC_SYS_TASK_ARGS; i++) {
+	    int tix = 3 + i;
+	    if (tix > arity) {
+		arg[i] = THE_NON_VALUE;
+		arg_sz[i] = 0;
+	    }
+	    else {
+		arg[i] = tp[tix];
+		if (is_immed(arg[i]))
+		    arg_sz[i] = 0;
+		else {
+		    arg_sz[i] = size_object(arg[i]);
+		    tot_sz += arg_sz[i];
+		}
+	    }
+	}
+	st = erts_alloc(ERTS_ALC_T_PROC_SYS_TSK,
+			ERTS_PROC_SYS_TASK_SIZE(tot_sz));
+	st->next = st->prev = st; /* Prep for empty prio queue */
+	ERTS_INIT_OFF_HEAP(&st->off_heap);
+	hp = &st->heap[0];
+
+	st->requester = BIF_P->common.id;
+	st->reply_tag = req_type;
+	st->req_id_sz = req_id_sz;
+	st->req_id = req_id_sz == 0 ? req_id : copy_struct(req_id,
+							   req_id_sz,
+							   &hp,
+							   &st->off_heap);
+
+	for (i = 0; i < ERTS_MAX_PROC_SYS_TASK_ARGS; i++)
+	    st->arg[i] = arg_sz[i] == 0 ? arg[i] : copy_struct(arg[i],
+							       arg_sz[i],
+							       &hp,
+							       &st->off_heap);
+	ASSERT(&st->heap[0] + tot_sz == hp);
+    }
+
+    switch (req_type) {
+
+    case am_garbage_collect:
+	st->type = ERTS_PSTT_GC;
+	noproc_res = am_false;
+	if (!rp)
+	    goto noproc;
+	break;
+
+    case am_check_process_code:
+	if (is_not_atom(st->arg[0]))
+	    goto badarg;
+	if (st->arg[1] != am_true && st->arg[1] != am_false)
+	    goto badarg;
+	noproc_res = am_false;
+	st->type = ERTS_PSTT_CPC;
+	if (!rp)
+	    goto noproc;
+	break;
+
+    default:
+	goto badarg;
+    }
+
+    rp_state = erts_smp_atomic32_read_nob(&rp->state);
+
+    rp_locked = 0;
+
+    free_stqs = NULL;
+    if (rp_state & ERTS_PSFLG_ACTIVE_SYS)
+	stqs = NULL;
+    else {
+    alloc_qs:
+	stqs = proc_sys_task_queues_alloc();
+	stqs->qmask = 1 << prio;
+	stqs->ncount = 0;
+	stqs->q[PRIORITY_MAX] = NULL;
+	stqs->q[PRIORITY_HIGH] = NULL;
+	stqs->q[PRIORITY_NORMAL] = NULL;
+	stqs->q[PRIORITY_LOW] = NULL;
+	stqs->q[prio] = st;
+    }
+
+    if (!rp_locked) {
+	rp_locked = 1;
+	erts_smp_proc_lock(rp, ERTS_PROC_LOCK_STATUS);
+
+	rp_state = erts_smp_atomic32_read_nob(&rp->state);
+	if (rp_state & ERTS_PSFLG_EXITING) {
+	    erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS);
+	    rp = NULL;
+	    free_stqs = stqs;
+	    goto noproc;
+	}
+    }
+
+    if (!rp->sys_task_qs) {
+	if (stqs)
+	    rp->sys_task_qs = stqs;
+	else
+	    goto alloc_qs;
+    }
+    else {
+	if (stqs)
+	    free_stqs = stqs;
+	stqs = rp->sys_task_qs;
+	if (!stqs->q[prio]) {
+	    stqs->q[prio] = st;
+	    stqs->qmask |= 1 << prio;
+	}
+	else {
+	    st->next = stqs->q[prio];
+	    st->prev = stqs->q[prio]->prev;
+	    st->next->prev = st;
+	    st->prev->next = st;
+	    ASSERT(stqs->qmask & (1 << prio));
+	}
+    }
+
+    if (ERTS_PSFLGS_GET_ACT_PRIO(rp_state) > prio) {
+	erts_aint32_t n, a, e;
+	/* Need to elevate actual prio */
+
+	a = rp_state;
+	do {
+	    if (ERTS_PSFLGS_GET_ACT_PRIO(a) <= prio) {
+		n = a;
+		break;
+	    }
+	    n = e = a;
+	    n &= ~ERTS_PSFLGS_ACT_PRIO_MASK;
+	    n |= (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET);
+	    a = erts_smp_atomic32_cmpxchg_nob(&rp->state, n, e);    
+	} while (a != e);
+	rp_state = n;
+    }
+
+    erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS);
+
+    schedule_process_sys_task(rp, rp_state, NULL);
+
+    if (free_stqs)
+	proc_sys_task_queues_free(free_stqs);
+
+    BIF_RET(am_ok);
+
+noproc:
+
+    notify_sys_task_executed(BIF_P, st, noproc_res);
+    if (free_stqs)
+	proc_sys_task_queues_free(free_stqs);
+    BIF_RET(am_ok);
+
+badarg:
+
+    if (st) {
+	erts_cleanup_offheap(&st->off_heap);
+	erts_free(ERTS_ALC_T_PROC_SYS_TSK, st);
+    }
+    if (free_stqs)
+	proc_sys_task_queues_free(free_stqs);
+    BIF_ERROR(BIF_P, BADARG);
+}
+
 void
 erts_sched_stat_modify(int what)
 {
@@ -7520,7 +8463,8 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
 	prio = (erts_aint32_t) so->priority;
     }
 
-    state |= (prio & ERTS_PSFLG_PRIO_MASK);
+    state |= (((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_ACT_PRIO_OFFSET)
+	      | ((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_USR_PRIO_OFFSET));
 
     if (!rq)
 	rq = erts_get_runq_proc(parent);
@@ -7599,6 +8543,8 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     p->bin_old_vheap    = 0;
     p->bin_vheap_mature = 0;
 
+    p->sys_task_qs = NULL;
+
     /* No need to initialize p->fcalls. */
 
     p->current = p->initial+INITIAL_MOD;
@@ -7764,7 +8710,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
      * Schedule process for execution.
      */
 
-    schedule_process(p, state, 0);
+    schedule_process(p, state);
 
     VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->common.id));
 
@@ -7815,6 +8761,7 @@ void erts_init_empty_process(Process *p)
     p->bin_vheap_sz = BIN_VH_MIN_SIZE;
     p->bin_old_vheap_sz = BIN_VH_MIN_SIZE;
     p->bin_old_vheap = 0;
+    p->sys_task_qs = NULL;
     p->bin_vheap_mature = 0;
 #ifdef ERTS_SMP
     p->common.u.alive.ptimer = NULL;
@@ -8070,33 +9017,21 @@ delete_process(Process* p)
     p->fvalue = NIL;
 }
 
-static ERTS_INLINE erts_aint32_t
-set_proc_exiting_state(Process *p, erts_aint32_t state)
-{
-    erts_aint32_t a, n, e;
-    a = state;
-    while (1) {
-	n = e = a;
-	n &= ~(ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT);
-	n |= ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE;
-	if (!(a & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLG_RUNNING)))
-	    n |= ERTS_PSFLG_IN_RUNQ;
-	a = erts_smp_atomic32_cmpxchg_relb(&p->state, n, e);
-	if (a == e)
-	    break;
-    }
-    return a;
-}
-
 static ERTS_INLINE void
 set_proc_exiting(Process *p,
-		 erts_aint32_t state,
+		 erts_aint32_t in_state,
 		 Eterm reason,
 		 ErlHeapFragment *bp)
 {
+    erts_aint32_t state = in_state, enq_prio = -1;
+    int enqueue;
     ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(p) == ERTS_PROC_LOCKS_ALL);
 
-    state = set_proc_exiting_state(p, state);
+    enqueue = change_proc_schedule_state(p,
+					 ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT,
+					 ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE,
+					 &state,
+					 &enq_prio);
 
     p->fvalue = reason;
     if (bp)
@@ -8111,15 +9046,37 @@ set_proc_exiting(Process *p,
     cancel_timer(p);
     p->i = (BeamInstr *) beam_exit;
 
-    if (erts_system_profile_flags.runnable_procs
-	&& !(state & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED))) {
-    	profile_runnable_proc(p, am_active);
-    }
-
-    if (!(state & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLG_RUNNING)))
-	add2runq(p, state);
+    if (enqueue)
+	add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio),
+		 state,
+		 enq_prio);
 }
 
+static ERTS_INLINE erts_aint32_t
+set_proc_self_exiting(Process *c_p)
+{
+#ifdef DEBUG
+    int enqueue;
+#endif
+    erts_aint32_t state, enq_prio = -1;
+
+    ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCKS_ALL);
+
+    state = erts_smp_atomic32_read_nob(&c_p->state);
+    ASSERT(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS));
+
+#ifdef DEBUG
+    enqueue =
+#endif
+	change_proc_schedule_state(c_p,
+				   ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT,
+				   ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE,
+				   &state,
+				   &enq_prio);
+
+    ASSERT(!enqueue);
+    return state;
+}
 
 #ifdef ERTS_SMP
 
@@ -8167,7 +9124,7 @@ handle_pending_exiters(ErtsProcList *pnd_xtrs)
 	if (p) {
 	    if (erts_proclist_same(plp, p)) {
 		erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state);
-		if (!(state & ERTS_PSFLG_RUNNING)) {
+		if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))) {
 		    ASSERT(state & ERTS_PSFLG_PENDING_EXIT);
 		    erts_handle_pending_exit(p, ERTS_PROC_LOCKS_ALL);
 		}
@@ -8388,7 +9345,7 @@ send_exit_signal(Process *c_p,		/* current process if and only
 		}
 		set_proc_exiting(c_p, state, rsn, NULL);
 	    }
-	    else if (!(state & ERTS_PSFLG_RUNNING)) {
+	    else if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))) {
 		/* Process not running ... */
 		ErtsProcLocks need_locks = ~(*rp_locks) & ERTS_PROC_LOCKS_ALL;
 		if (need_locks
@@ -8765,9 +9722,6 @@ resume_suspend_monitor(ErtsSuspendMonitor *smon, void *vc_p)
 void 
 erts_do_exit_process(Process* p, Eterm reason)
 {
-#ifdef ERTS_SMP
-    erts_aint32_t state;
-#endif
     p->arity = 0;		/* No live registers */
     p->fvalue = reason;
 
@@ -8792,10 +9746,9 @@ erts_do_exit_process(Process* p, Eterm reason)
 #endif
 
 #ifndef ERTS_SMP
-    set_proc_exiting_state(p, erts_smp_atomic32_read_nob(&p->state));
+    set_proc_self_exiting(p);
 #else
-    state = set_proc_exiting_state(p, erts_smp_atomic32_read_nob(&p->state));
-    if (state & ERTS_PSFLG_PENDING_EXIT) {
+    if (ERTS_PSFLG_PENDING_EXIT & set_proc_self_exiting(p)) {
 	/* Process exited before pending exit was received... */
 	p->pending_exit.reason = THE_NON_VALUE;
 	if (p->pending_exit.bp) {
@@ -8849,6 +9802,7 @@ erts_continue_exit_process(Process *p)
     DistEntry *dep;
     struct saved_calls *scb;
     process_breakpoint_time_t *pbt;
+    erts_aint32_t state;
 
 #ifdef DEBUG
     int yield_allowed = 1;
@@ -8885,6 +9839,12 @@ erts_continue_exit_process(Process *p)
 	p->flags &= ~F_USING_DB;
     }
 
+    state = erts_smp_atomic32_read_acqb(&p->state);
+    if (state & ERTS_PSFLG_ACTIVE_SYS) {
+	if (cleanup_sys_tasks(p, state, CONTEXT_REDS) >= CONTEXT_REDS/2)
+	    goto yield;
+    }
+
     if (p->flags & F_USING_DDLL) {
 	erts_ddll_proc_dead(p, ERTS_PROC_LOCK_MAIN);
 	p->flags &= ~F_USING_DDLL;
@@ -8962,17 +9922,31 @@ erts_continue_exit_process(Process *p)
     {
 	/* Inactivate and notify free */
 	erts_aint32_t n, e, a = erts_smp_atomic32_read_nob(&p->state);
+#ifdef ERTS_SMP
+	int refc_inced = 0;
+#endif
 	while (1) {
 	    n = e = a;
 	    ASSERT(a & ERTS_PSFLG_EXITING);
 	    n |= ERTS_PSFLG_FREE;
 	    n &= ~ERTS_PSFLG_ACTIVE;
+#ifdef ERTS_SMP
+	    if ((n & ERTS_PSFLG_IN_RUNQ) && !refc_inced) {
+		erts_smp_proc_inc_refc(p);
+		refc_inced = 1;
+	    }
+#endif
 	    a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
 	    if (a == e)
 		break;
 	}
-    }
 
+#ifdef ERTS_SMP
+	if (refc_inced && !(n & ERTS_PSFLG_IN_RUNQ))
+	    erts_smp_proc_dec_refc(p);
+#endif
+    }
+    
     dep = ((p->flags & F_DISTRIBUTION)
 	   ? ERTS_PROC_SET_DIST_ENTRY(p, ERTS_PROC_LOCKS_ALL, NULL)
 	   : NULL);
@@ -9075,7 +10049,7 @@ timeout_proc(Process* p)
 
     state = erts_smp_atomic32_read_acqb(&p->state);
     if (!(state & ERTS_PSFLG_ACTIVE))
-	schedule_process(p, state, 0);
+	schedule_process(p, state);
 }
 
 
@@ -9153,7 +10127,9 @@ erts_program_counter_info(int to, void *to_arg, Process *p)
     print_function_from_pc(to, to_arg, p->cp);
     erts_print(to, to_arg, ")\n");
     state = erts_smp_atomic32_read_acqb(&p->state);
-    if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_GC))) {
+    if (!(state & (ERTS_PSFLG_RUNNING
+		   | ERTS_PSFLG_RUNNING_SYS
+		   | ERTS_PSFLG_GC))) {
         erts_print(to, to_arg, "arity = %d\n",p->arity);
 	if (!ERTS_IS_CRASH_DUMPING) {
 	    /*
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index 8e5467f196..814e3d34df 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -688,6 +688,9 @@ typedef struct {
     ErlHeapFragment *bp;
 } ErtsPendExit;
 
+typedef struct ErtsProcSysTask_ ErtsProcSysTask;
+typedef struct ErtsProcSysTaskQs_ ErtsProcSysTaskQs;
+
 #ifdef ERTS_SMP
 
 typedef struct ErtsPendingSuspend_ ErtsPendingSuspend;
@@ -855,6 +858,7 @@ struct process {
     Uint64 bin_old_vheap_sz;	/* Virtual old heap block size for binaries */
     Uint64 bin_old_vheap;	/* Virtual old heap size for binaries */
 
+    ErtsProcSysTaskQs *sys_task_qs;
     erts_smp_atomic32_t state;  /* Process state flags (see ERTS_PSFLG_*) */
 
 #ifdef ERTS_SMP
@@ -924,24 +928,66 @@ void erts_check_for_holes(Process* p);
 #  error "Need to increase ERTS_PSFLG_PRIO_SHIFT"
 #endif
 
-#define ERTS_PSFLG_PRIO_SHIFT 2
+#define ERTS_PSFLGS_PRIO_BITS 2
+#define ERTS_PSFLGS_PRIO_MASK \
+    ((((erts_aint32_t) 1) << ERTS_PSFLGS_PRIO_BITS) - 1)
 
-#define ERTS_PSFLG_BIT(N) \
-    (((erts_aint32_t) 1) << (ERTS_PSFLG_PRIO_SHIFT + (N)))
+#define ERTS_PSFLGS_ACT_PRIO_OFFSET (0*ERTS_PSFLGS_PRIO_BITS)
+#define ERTS_PSFLGS_USR_PRIO_OFFSET (1*ERTS_PSFLGS_PRIO_BITS)
+#define ERTS_PSFLGS_PRQ_PRIO_OFFSET (2*ERTS_PSFLGS_PRIO_BITS)
+#define ERTS_PSFLGS_ZERO_BIT_OFFSET (3*ERTS_PSFLGS_PRIO_BITS)
 
-#define ERTS_PSFLG_PRIO_MASK 		(ERTS_PSFLG_BIT(0) - 1)
+#define ERTS_PSFLGS_QMASK_BITS 4
+#define ERTS_PSFLGS_QMASK \
+    ((((erts_aint32_t) 1) << ERTS_PSFLGS_QMASK_BITS) - 1)
+#define ERTS_PSFLGS_IN_PRQ_MASK_OFFSET \
+    ERTS_PSFLGS_ZERO_BIT_OFFSET
 
-#define ERTS_PSFLG_FREE			ERTS_PSFLG_BIT(0)
-#define ERTS_PSFLG_EXITING		ERTS_PSFLG_BIT(1)
-#define ERTS_PSFLG_PENDING_EXIT		ERTS_PSFLG_BIT(2)
-#define ERTS_PSFLG_ACTIVE		ERTS_PSFLG_BIT(3)
-#define ERTS_PSFLG_IN_RUNQ		ERTS_PSFLG_BIT(4)
-#define ERTS_PSFLG_RUNNING		ERTS_PSFLG_BIT(5)
-#define ERTS_PSFLG_SUSPENDED		ERTS_PSFLG_BIT(6)
-#define ERTS_PSFLG_GC			ERTS_PSFLG_BIT(7)
-#define ERTS_PSFLG_BOUND		ERTS_PSFLG_BIT(8)
-#define ERTS_PSFLG_TRAP_EXIT		ERTS_PSFLG_BIT(9)
+#define ERTS_PSFLG_BIT(N) \
+    (((erts_aint32_t) 1) << (ERTS_PSFLGS_ZERO_BIT_OFFSET + (N)))
 
+/*
+ * ACT_PRIO -> Active prio, i.e., currently active prio. This
+ *             prio may be higher than user prio.
+ * USR_PRIO -> User prio. i.e., prio the user has set.
+ * PRQ_PRIO -> Prio queue prio, i.e., prio queue currently
+ *             enqueued in. 
+ */
+#define ERTS_PSFLGS_ACT_PRIO_MASK \
+    (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_ACT_PRIO_OFFSET)
+#define ERTS_PSFLGS_USR_PRIO_MASK \
+    (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_USR_PRIO_OFFSET)
+#define ERTS_PSFLGS_PRQ_PRIO_MASK \
+    (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_PRQ_PRIO_OFFSET)
+#define ERTS_PSFLG_IN_PRQ_MAX 		ERTS_PSFLG_BIT(0)
+#define ERTS_PSFLG_IN_PRQ_HIGH		ERTS_PSFLG_BIT(1)
+#define ERTS_PSFLG_IN_PRQ_NORMAL	ERTS_PSFLG_BIT(2)
+#define ERTS_PSFLG_IN_PRQ_LOW 		ERTS_PSFLG_BIT(3)
+#define ERTS_PSFLG_FREE			ERTS_PSFLG_BIT(4)
+#define ERTS_PSFLG_EXITING		ERTS_PSFLG_BIT(5)
+#define ERTS_PSFLG_PENDING_EXIT		ERTS_PSFLG_BIT(6)
+#define ERTS_PSFLG_ACTIVE		ERTS_PSFLG_BIT(7)
+#define ERTS_PSFLG_IN_RUNQ		ERTS_PSFLG_BIT(8)
+#define ERTS_PSFLG_RUNNING		ERTS_PSFLG_BIT(9)
+#define ERTS_PSFLG_SUSPENDED		ERTS_PSFLG_BIT(10)
+#define ERTS_PSFLG_GC			ERTS_PSFLG_BIT(11)
+#define ERTS_PSFLG_BOUND		ERTS_PSFLG_BIT(12)
+#define ERTS_PSFLG_TRAP_EXIT		ERTS_PSFLG_BIT(13)
+#define ERTS_PSFLG_ACTIVE_SYS		ERTS_PSFLG_BIT(14)
+#define ERTS_PSFLG_RUNNING_SYS		ERTS_PSFLG_BIT(15)
+#define ERTS_PSFLG_PROXY		ERTS_PSFLG_BIT(16)
+
+#define ERTS_PSFLGS_IN_PRQ_MASK 	(ERTS_PSFLG_IN_PRQ_MAX		\
+					 | ERTS_PSFLG_IN_PRQ_HIGH	\
+					 | ERTS_PSFLG_IN_PRQ_NORMAL	\
+					 | ERTS_PSFLG_IN_PRQ_LOW)
+
+#define ERTS_PSFLGS_GET_ACT_PRIO(PSFLGS) \
+    (((PSFLGS) >> ERTS_PSFLGS_ACT_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK)
+#define ERTS_PSFLGS_GET_USR_PRIO(PSFLGS) \
+    (((PSFLGS) >> ERTS_PSFLGS_USR_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK)
+#define ERTS_PSFLGS_GET_PRQ_PRIO(PSFLGS) \
+    (((PSFLGS) >> ERTS_PSFLGS_USR_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK)
 
 /* The sequential tracing token is a tuple of size 5:
  *
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index bacd5a5752..885f092aca 100755
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -655,6 +655,10 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg);
 
 Eterm erl_is_function(Process* p, Eterm arg1, Eterm arg2);
 
+/* beam_bif_load.c */
+Eterm erts_check_process_code(Process *c_p, Eterm module, int allow_gc, int *redsp);
+
+
 /* beam_load.c */
 typedef struct {
     BeamInstr* current;		/* Pointer to: Mod, Name, Arity */
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 1e5ae46bfa..c29f3f9b1b 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -763,17 +763,17 @@ allocate_init t I y
 #################################################################
 
 #
-# The BIFs erlang:check_process_code/2 must be called like a function,
+# The BIFs erts_internal:check_process_code/2 must be called like a function,
 # to ensure that c_p->i (program counter) is set correctly (an ordinary
 # BIF call doesn't set it).
 #
 
-call_ext u==2 Bif=u$bif:erlang:check_process_code/2 => i_call_ext Bif
-call_ext_last u==2 Bif=u$bif:erlang:check_process_code/2 D => i_call_ext_last Bif D
-call_ext_only u==2 Bif=u$bif:erlang:check_process_code/2 => i_call_ext_only Bif
+call_ext u==2 Bif=u$bif:erts_internal:check_process_code/2 => i_call_ext Bif
+call_ext_last u==2 Bif=u$bif:erts_internal:check_process_code/2 D => i_call_ext_last Bif D
+call_ext_only u==2 Bif=u$bif:erts_internal:check_process_code/2 => i_call_ext_only Bif
 
 #
-# The BIFs erlang:garbage_collect/0,1 must be called like functions,
+# The BIFs erlang:garbage_collect/0 must be called like a function,
 # to allow them to invoke the garbage collector. (The stack pointer must
 # be saved and p->arity must be zeroed, which is not done on ordinary BIF calls.)
 #
@@ -782,10 +782,6 @@ call_ext u==0 Bif=u$bif:erlang:garbage_collect/0 => i_call_ext Bif
 call_ext_last u==0 Bif=u$bif:erlang:garbage_collect/0 D => i_call_ext_last Bif D
 call_ext_only u==0 Bif=u$bif:erlang:garbage_collect/0 => i_call_ext_only Bif
 
-call_ext u==1 Bif=u$bif:erlang:garbage_collect/1 => i_call_ext Bif
-call_ext_last u==1 Bif=u$bif:erlang:garbage_collect/1 D => i_call_ext_last Bif D
-call_ext_only u==1 Bif=u$bif:erlang:garbage_collect/1 => i_call_ext_only Bif
-
 #
 # put/2 and erase/1 must be able to do garbage collection, so we must call
 # them like functions.
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index bd2be7afca..a5b2cc589c 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -1675,7 +1675,7 @@ static int do_send_to_logger(Eterm tag, Eterm gleader, char *buf, int len)
 	p = erts_whereis_process(NULL, 0, am_error_logger, 0, 0);
 	if (p) {
 	    erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state);
-	    if (state & ERTS_PSFLG_RUNNING)
+	    if (state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))
 		p = NULL;
 	}
     }
diff --git a/erts/emulator/hipe/hipe_bif_list.m4 b/erts/emulator/hipe/hipe_bif_list.m4
index 764b8d180c..b1fedf4838 100644
--- a/erts/emulator/hipe/hipe_bif_list.m4
+++ b/erts/emulator/hipe/hipe_bif_list.m4
@@ -151,10 +151,9 @@ standard_bif_interface_0(nbif_ports_0, ports_0)
  * BIFs and primops that may do a GC (change heap limit and walk the native stack).
  * XXX: erase/1 and put/2 cannot fail
  */
-gc_bif_interface_2(nbif_check_process_code_2, hipe_check_process_code_2)
+gc_bif_interface_2(nbif_erts_internal_check_process_code_2, hipe_erts_internal_check_process_code_2)
 gc_bif_interface_1(nbif_erase_1, erase_1)
 gc_bif_interface_0(nbif_garbage_collect_0, garbage_collect_0)
-gc_bif_interface_1(nbif_garbage_collect_1, hipe_garbage_collect_1)
 gc_nofail_primop_interface_1(nbif_gc_1, hipe_gc)
 gc_bif_interface_2(nbif_put_2, put_2)
 
diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c
index 1f76268934..7d343dd91e 100644
--- a/erts/emulator/hipe/hipe_native_bif.c
+++ b/erts/emulator/hipe/hipe_native_bif.c
@@ -41,8 +41,7 @@
  */
 
 /* for -Wmissing-prototypes :-( */
-extern Eterm hipe_check_process_code_2(BIF_ALIST_2);
-extern Eterm hipe_garbage_collect_1(BIF_ALIST_1);
+extern Eterm hipe_erts_internal_check_process_code_2(BIF_ALIST_2);
 extern Eterm hipe_show_nstack_1(BIF_ALIST_1);
 
 /* Used when a BIF can trigger a stack walk. */
@@ -51,22 +50,12 @@ static __inline__ void hipe_set_narity(Process *p, unsigned int arity)
     p->hipe.narity = arity;
 }
 
-Eterm hipe_check_process_code_2(BIF_ALIST_2)
+Eterm hipe_erts_internal_check_process_code_2(BIF_ALIST_2)
 {
     Eterm ret;
 
     hipe_set_narity(BIF_P, 2);
-    ret = check_process_code_2(BIF_P, BIF__ARGS);
-    hipe_set_narity(BIF_P, 0);
-    return ret;
-}
-
-Eterm hipe_garbage_collect_1(BIF_ALIST_1)
-{
-    Eterm ret;
-
-    hipe_set_narity(BIF_P, 1);
-    ret = garbage_collect_1(BIF_P, BIF__ARGS);
+    ret = erts_internal_check_process_code_2(BIF_P, BIF__ARGS);
     hipe_set_narity(BIF_P, 0);
     return ret;
 }
diff --git a/erts/emulator/test/process_SUITE.erl b/erts/emulator/test/process_SUITE.erl
index e3aae17df4..e66c6f09b6 100644
--- a/erts/emulator/test/process_SUITE.erl
+++ b/erts/emulator/test/process_SUITE.erl
@@ -51,7 +51,11 @@
 	 processes_term_proc_list/1,
 	 otp_7738_waiting/1, otp_7738_suspended/1,
 	 otp_7738_resume/1,
-	 garb_other_running/1]).
+	 garb_other_running/1,
+	 no_priority_inversion/1,
+	 no_priority_inversion2/1,
+	 system_task_blast/1,
+	 system_task_on_suspended/1]).
 -export([prio_server/2, prio_client/2]).
 
 -export([init_per_testcase/2, end_per_testcase/2]).
@@ -73,7 +77,8 @@ all() ->
      bad_register, garbage_collect, process_info_messages,
      process_flag_badarg, process_flag_heap_size,
      spawn_opt_heap_size, otp_6237, {group, processes_bif},
-     {group, otp_7738}, garb_other_running].
+     {group, otp_7738}, garb_other_running,
+     {group, system_task}].
 
 groups() -> 
     [{t_exit_2, [],
@@ -87,7 +92,10 @@ groups() ->
        processes_gc_trap, processes_term_proc_list]},
      {otp_7738, [],
       [otp_7738_waiting, otp_7738_suspended,
-       otp_7738_resume]}].
+       otp_7738_resume]},
+     {system_task, [],
+      [no_priority_inversion, no_priority_inversion2,
+       system_task_blast, system_task_on_suspended]}].
 
 init_per_suite(Config) ->
     A0 = case application:start(sasl) of
@@ -2214,6 +2222,154 @@ garb_other_running(Config) when is_list(Config) ->
     receive {'DOWN', Mon, process, Pid, normal} -> ok end,
     ok.
 
+no_priority_inversion(Config) when is_list(Config) ->
+    Prio = process_flag(priority, max),
+    HTLs = lists:map(fun (_) ->
+			     spawn_opt(fun () ->
+					       tok_loop()
+				       end,
+				       [{priority, high}, monitor, link])
+		     end,
+		     lists:seq(1, 2*erlang:system_info(schedulers))),
+    receive after 500 -> ok end,
+    LTL = spawn_opt(fun () ->
+			    tok_loop()
+		    end,
+		    [{priority, low}, monitor, link]),
+    false = erlang:check_process_code(element(1, LTL), nonexisting_module),
+    true = erlang:garbage_collect(element(1, LTL)),
+    lists:foreach(fun ({P, _}) ->
+			  unlink(P),
+			  exit(P, kill)
+		  end, [LTL | HTLs]),
+    lists:foreach(fun ({P, M}) ->
+			  receive
+			      {'DOWN', M, process, P, killed} ->
+				  ok
+			  end
+		  end, [LTL | HTLs]),
+    process_flag(priority, Prio),
+    ok.
+
+no_priority_inversion2(Config) when is_list(Config) ->
+    Prio = process_flag(priority, max),
+    MTLs = lists:map(fun (_) ->
+			     spawn_opt(fun () ->
+					       tok_loop()
+				       end,
+				       [{priority, max}, monitor, link])
+		     end,
+		     lists:seq(1, 2*erlang:system_info(schedulers))),
+    receive after 500 -> ok end,
+    {PL, ML} = spawn_opt(fun () ->
+			       tok_loop()
+		       end,
+		       [{priority, low}, monitor, link]),
+    RL = request_gc(PL, low),
+    RN = request_gc(PL, normal),
+    RH = request_gc(PL, high),
+    receive
+	{garbage_collect, _, _} ->
+	    ?t:fail(unexpected_gc)
+    after 1000 ->
+	    ok
+    end,
+    RM = request_gc(PL, max),
+    receive
+	{garbage_collect, RM, true} ->
+	    ok
+    end,
+    lists:foreach(fun ({P, _}) ->
+			  unlink(P),
+			  exit(P, kill)
+		  end, MTLs),
+    lists:foreach(fun ({P, M}) ->
+			  receive
+			      {'DOWN', M, process, P, killed} ->
+				  ok
+			  end
+		  end, MTLs),
+    receive
+	{garbage_collect, RH, true} ->
+	    ok
+    end,
+    receive
+	{garbage_collect, RN, true} ->
+	    ok
+    end,
+    receive
+	{garbage_collect, RL, true} ->
+	    ok
+    end,
+    unlink(PL),
+    exit(PL, kill),
+    receive
+	{'DOWN', ML, process, PL, killed} ->
+	    ok
+    end,
+    process_flag(priority, Prio),
+    ok.
+
+request_gc(Pid, Prio) ->
+    Ref = make_ref(),
+    erts_internal:request_system_task(Pid, Prio, {garbage_collect, Ref}),
+    Ref.
+
+system_task_blast(Config) when is_list(Config) ->
+    Me = self(),
+    GCReq = fun () ->
+		    RL = gc_req(Me, 100),
+		    lists:foreach(fun (R) ->
+					  receive
+					      {garbage_collect, R, true} ->
+						  ok
+					  end
+				  end, RL),
+		    exit(it_worked)
+	    end,
+    HTLs = lists:map(fun (_) -> spawn_monitor(GCReq) end, lists:seq(1, 1000)),
+    lists:foreach(fun ({P, M}) ->
+			  receive
+			      {'DOWN', M, process, P, it_worked} ->
+				  ok
+			  end
+		  end, HTLs),
+    ok.
+
+gc_req(_Pid, 0) ->
+    [];
+gc_req(Pid, N) ->
+    R0 = request_gc(Pid, low),
+    R1 = request_gc(Pid, normal),
+    R2 = request_gc(Pid, high),
+    R3 = request_gc(Pid, max),
+    [R0, R1, R2, R3 | gc_req(Pid, N-1)].
+
+system_task_on_suspended(Config) when is_list(Config) ->
+    {P, M} = spawn_monitor(fun () ->
+				   tok_loop()
+			   end),
+    true = erlang:suspend_process(P),
+    {status, suspended} = process_info(P, status),
+    true = erlang:garbage_collect(P),
+    {status, suspended} = process_info(P, status),
+    true = erlang:resume_process(P),
+    false = ({status, suspended} == process_info(P, status)),
+    exit(P, kill),
+    receive
+	{'DOWN', M, process, P, killed} ->
+	    ok
+    end.
+
+gc_req(_Pid, 0) ->
+    [];
+gc_req(Pid, N) ->
+    R0 = erts_internal:request_system_task(Pid, low, garbage_collect),
+    R1 = erts_internal:request_system_task(Pid, normal, garbage_collect),
+    R2 = erts_internal:request_system_task(Pid, high, garbage_collect),
+    R3 = erts_internal:request_system_task(Pid, max, garbage_collect),
+    [R0, R1, R2, R3 | gc_req(Pid, N-1)].
+		  
 %% Internal functions
 
 wait_until(Fun) ->
diff --git a/erts/etc/unix/etp-commands.in b/erts/etc/unix/etp-commands.in
index 54ff7b3e3a..6e3466b8c0 100644
--- a/erts/etc/unix/etp-commands.in
+++ b/erts/etc/unix/etp-commands.in
@@ -1316,49 +1316,99 @@ end
 define etp-proc-state-int
 # Args: int
 #
-  if ($arg0 & 0xfffff000)
+  if ($arg0 & 0xff800000)
     printf "GARBAGE | "
   end
-  if ($arg0 & 0x800)
+  if ($arg0 & 0x400000)
+    printf "proxy | "
+    set $proxy_process = 1
+  else
+    set $proxy_process = 0
+  end
+  if ($arg0 & 0x200000)
+    printf "running-sys | "
+  end
+  if ($arg0 & 0x100000)
+    printf "active-sys | "
+  end
+  if ($arg0 & 0x80000)
     printf "trapping-exit | "
   end
-  if ($arg0 & 0x400)
+  if ($arg0 & 0x40000)
     printf "bound | "
   end
-  if ($arg0 & 0x200)
+  if ($arg0 & 0x20000)
     printf "garbage-collecting | "
   end
-  if ($arg0 & 0x100)
+  if ($arg0 & 0x10000)
     printf "suspended | "
   end
-  if ($arg0 & 0x80)
+  if ($arg0 & 0x8000)
     printf "running | "
   end
-  if ($arg0 & 0x40)
+  if ($arg0 & 0x4000)
     printf "in-run-queue | "
   end
-  if ($arg0 & 0x20)
+  if ($arg0 & 0x2000)
     printf "active | "
   end
-  if ($arg0 & 0x10)
+  if ($arg0 & 0x1000)
     printf "pending-exit | "
   end
-  if ($arg0 & 0x8)
+  if ($arg0 & 0x800)
     printf "exiting | "
   end
-  if ($arg0 & 0x4)
+  if ($arg0 & 0x400)
     printf "free | "
   end
-  if ($arg0 & 0x3) == 0
-    printf "prio-max\n"
+  if ($arg0 & 0x200)
+    printf "in-prq-low | "
+  end
+  if ($arg0 & 0x100)
+    printf "in-prq-normal | "
+  end
+  if ($arg0 & 0x80)
+    printf "in-prq-high | "
+  end
+  if ($arg0 & 0x40)
+    printf "in-prq-max | "
+  end
+  if ($arg0 & 0x30) == 0x0
+    printf "prq-prio-max | "
   else
-    if ($arg0 & 0x3) == 1
-    printf "prio-high\n"
+    if ($arg0 & 0x30) == 0x10
+      printf "prq-prio-high | "
     else
-      if ($arg0 & 0x3) == 2
-        printf "prio-normal\n"
+      if ($arg0 & 0x30) == 0x20
+        printf "prq-prio-normal | "
       else
-        printf "prio-low\n"
+        printf "prq-prio-low | "
+      end
+    end
+  end
+  if ($arg0 & 0xc) == 0x0
+    printf "usr-prio-max | "
+  else
+    if ($arg0 & 0xc) == 0x4
+      printf "usr-prio-high | "
+    else
+      if ($arg0 & 0xc) == 0x8
+        printf "usr-prio-normal | "
+      else
+        printf "usr-prio-low | "
+      end
+    end
+  end
+  if ($arg0 & 0x3) == 0x0
+    printf "act-prio-max\n"
+  else
+    if ($arg0 & 0x3) == 0x1
+      printf "act-prio-high\n"
+    else
+      if ($arg0 & 0x3) == 0x2
+        printf "act-prio-normal\n"
+      else
+        printf "act-prio-low\n"
       end
     end
   end
@@ -1395,6 +1445,12 @@ define etp-process-info
   etp-1 $arg0->common.id
   printf "\n  State: "
   etp-proc-state $arg0
+  if $proxy_process != 0
+    printf "  Pointer: (Process *) %p\n", $arg0
+    printf "  *** PROXY process struct *** refer to: \n"
+    etp-pid2proc-1 $arg0->common.id
+    etp-process-info $proc
+  else
   if (*(((Uint32 *) &(((Process *) $arg0)->state))) & 0x4) == 0
     if ($arg0->common.u.alive.reg)
       printf "  Registered name: "
@@ -1432,6 +1488,7 @@ define etp-process-info
   printf "  Parent: "
   etp-1 $arg0->parent
   printf "\n  Pointer: (Process *) %p\n", $arg0
+  end
 end
 
 document etp-process-info
diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam
index 9da4f3cd00..159c91e37c 100644
--- a/erts/preloaded/ebin/erlang.beam
+++ b/erts/preloaded/ebin/erlang.beam
diff --git a/erts/preloaded/ebin/erts_internal.beam b/erts/preloaded/ebin/erts_internal.beam
index 3ab52615ab..9d8a4cfd00 100644
--- a/erts/preloaded/ebin/erts_internal.beam
+++ b/erts/preloaded/ebin/erts_internal.beam
diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl
index a969ef91dc..e521c6fc3d 100644
--- a/erts/preloaded/src/erlang.erl
+++ b/erts/preloaded/src/erlang.erl
@@ -81,7 +81,8 @@
 -export([binary_to_list/3, binary_to_term/1, binary_to_term/2]).
 -export([bit_size/1, bitsize/1, bitstr_to_list/1, bitstring_to_list/1]).
 -export([bump_reductions/1, byte_size/1, call_on_load_function/1]).
--export([cancel_timer/1, check_old_code/1, check_process_code/2, crc32/1]).
+-export([cancel_timer/1, check_old_code/1, check_process_code/2,
+	 check_process_code/3, crc32/1]).
 -export([crc32/2, crc32_combine/3, date/0, decode_packet/3]).
 -export([delete_element/2]).
 -export([delete_module/1, demonitor/1, demonitor/2, display/1]).
@@ -91,7 +92,7 @@
 -export([float_to_binary/1, float_to_binary/2,
 	 float_to_list/1, float_to_list/2]).
 -export([fun_info/2, fun_to_list/1, function_exported/3]).
--export([garbage_collect/0, garbage_collect/1]).
+-export([garbage_collect/0, garbage_collect/1, garbage_collect/2]).
 -export([garbage_collect_message_area/0, get/0, get/1, get_keys/1]).
 -export([get_module_info/1, get_stacktrace/0, group_leader/0]).
 -export([group_leader/2, halt/0, halt/1, halt/2, hash/2, hibernate/3]).
@@ -429,11 +430,71 @@ check_old_code(_Module) ->
     erlang:nif_error(undefined).
 
 %% check_process_code/2
--spec check_process_code(Pid, Module) -> boolean() when
+-spec check_process_code(Pid, Module) -> CheckResult when
       Pid :: pid(),
-      Module :: module().
-check_process_code(_Pid, _Module) ->
-    erlang:nif_error(undefined).
+      Module :: module(),
+      CheckResult :: boolean().
+check_process_code(Pid, Module) ->
+    try
+	erlang:check_process_code(Pid, Module, [{allow_gc, true}])
+    catch
+	error:Error -> erlang:error(Error, [Pid, Module])
+    end.
+
+%% check_process_code/3
+-spec check_process_code(Pid, Module, OptionList) -> CheckResult | async when
+      Pid :: pid(),
+      Module :: module(),
+      RequestId :: term(),
+      Option :: {async, RequestId} | {allow_gc, boolean()},
+      OptionList :: [Option],
+      CheckResult :: boolean() | aborted.
+check_process_code(Pid, Module, OptionList)  ->
+    try
+	{Async, AllowGC} = get_cpc_opts(OptionList, sync, true),
+	case Async of
+	    {async, ReqId} ->
+		{priority, Prio} = erlang:process_info(erlang:self(),
+						       priority),
+		erts_internal:request_system_task(Pid,
+						  Prio,
+						  {check_process_code,
+						   ReqId,
+						   Module,
+						   AllowGC}),
+		async;
+	    sync ->
+		case Pid == erlang:self() of
+		    true ->
+			erts_internal:check_process_code(Module,
+							 [{allow_gc, AllowGC}]);
+		    false ->
+			{priority, Prio} = erlang:process_info(erlang:self(),
+							       priority),
+			ReqId = erlang:make_ref(),
+			erts_internal:request_system_task(Pid,
+							  Prio,
+							  {check_process_code,
+							   ReqId,
+							   Module,
+							   AllowGC}),
+			receive
+			    {check_process_code, ReqId, CheckResult} ->
+				CheckResult
+			end
+		end
+	end
+    catch
+	error:Error -> erlang:error(Error, [Pid, Module, OptionList])
+    end.
+
+% gets async and allow_gc opts and verify valid option list
+get_cpc_opts([{async, _ReqId} = AsyncTuple | Options], _OldAsync, AllowGC) ->
+    get_cpc_opts(Options, AsyncTuple, AllowGC);
+get_cpc_opts([{allow_gc, AllowGC} | Options], Async, _OldAllowGC) ->
+    get_cpc_opts(Options, Async, AllowGC);
+get_cpc_opts([], Async, AllowGC) ->
+    {Async, AllowGC}.
 
 %% crc32/1
 -spec erlang:crc32(Data) -> non_neg_integer() when
@@ -793,10 +854,61 @@ garbage_collect() ->
     erlang:nif_error(undefined).
 
 %% garbage_collect/1
--spec garbage_collect(Pid) -> boolean() when
-      Pid :: pid().
-garbage_collect(_Pid) ->
-    erlang:nif_error(undefined).
+-spec garbage_collect(Pid) -> GCResult when
+      Pid :: pid(),
+      GCResult :: boolean().
+garbage_collect(Pid) ->
+    try
+	erlang:garbage_collect(Pid, [])
+    catch
+	error:Error -> erlang:error(Error, [Pid])
+    end.
+
+%% garbage_collect/2
+-spec garbage_collect(Pid, OptionList) -> GCResult | async when
+      Pid :: pid(),
+      RequestId :: term(),
+      Option :: {async, RequestId},
+      OptionList :: [Option],
+      GCResult :: boolean().
+garbage_collect(Pid, OptionList)  ->
+    try
+	Async = get_gc_opts(OptionList, sync),
+	case Async of
+	    {async, ReqId} ->
+		{priority, Prio} = erlang:process_info(erlang:self(),
+						       priority),
+		erts_internal:request_system_task(Pid,
+						  Prio,
+						  {garbage_collect, ReqId}),
+		async;
+	    sync ->
+		case Pid == erlang:self() of
+		    true ->
+			erlang:garbage_collect();
+		    false ->
+			{priority, Prio} = erlang:process_info(erlang:self(),
+							       priority),
+			ReqId = erlang:make_ref(),
+			erts_internal:request_system_task(Pid,
+							  Prio,
+							  {garbage_collect,
+							   ReqId}),
+			receive
+			    {garbage_collect, ReqId, GCResult} ->
+				GCResult
+			end
+		end
+	end
+    catch
+	error:Error -> erlang:error(Error, [Pid, OptionList])
+    end.
+
+% gets async opt and verify valid option list
+get_gc_opts([{async, _ReqId} = AsyncTuple | Options], _OldAsync) ->
+    get_gc_opts(Options, AsyncTuple);
+get_gc_opts([], Async) ->
+    Async.
 
 %% garbage_collect_message_area/0
 -spec erlang:garbage_collect_message_area() -> boolean().
diff --git a/erts/preloaded/src/erts_internal.erl b/erts/preloaded/src/erts_internal.erl
index 8a8cd52d64..c8e8e7e069 100644
--- a/erts/preloaded/src/erts_internal.erl
+++ b/erts/preloaded/src/erts_internal.erl
@@ -33,6 +33,10 @@
 -export([port_command/3, port_connect/2, port_close/1,
 	 port_control/3, port_call/3, port_info/1, port_info/2]).
 
+-export([request_system_task/3]).
+
+-export([check_process_code/2]).
+
 %%
 %% Await result of send to port
 %%
@@ -139,3 +143,20 @@ port_info(_Result) ->
 
 port_info(_Result, _Item) ->
     erlang:nif_error(undefined).
+
+-spec request_system_task(Pid, Prio, Request) -> 'ok' when
+      Prio :: 'max' | 'high' | 'normal' | 'low',
+      Request :: {'garbage_collect', term()}
+	       | {'check_process_code', term(), module(), boolean()},
+      Pid :: pid().
+
+request_system_task(_Pid, _Prio, _Request) ->
+    erlang:nif_error(undefined).
+
+-spec check_process_code(Module, OptionList) -> boolean() when
+      Module :: module(),
+      Option :: {allow_gc, boolean()},
+      OptionList :: [Option].
+check_process_code(_Module, _OptionList) ->
+    erlang:nif_error(undefined).
+