Add dirty_process_main function

Dirty schedulers only execute NIFs, so having them execute the full
process_main function isn't necessary. Add dirty_process_main for
dirty schedulers to execute instead.

Add erts_pre_dirty_nif(), called when preparing to execute a dirty
nif.

Add more dirty NIF tests to verify that activities requiring the
process main lock can succeed when the process is executing a dirty
NIF.

6 files changed, 492 insertions, 52 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index f8f2e29c95..25cc41323b 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -3559,12 +3559,8 @@ do {						\
 		typedef Eterm NifF(struct enif_environment_t*, int argc, Eterm argv[]);
 		NifF* fp = vbf = (NifF*) I[1];
 		struct enif_environment_t env;
-#ifdef ERTS_DIRTY_SCHEDULERS
-		if (!c_p->scheduler_data)
-		    live_hf_end = ERTS_INVALID_HFRAG_PTR; /* On dirty scheduler */
-		else
-#endif
-		    live_hf_end = c_p->mbuf;
+		ASSERT(c_p->scheduler_data);
+		live_hf_end = c_p->mbuf;
 		erts_pre_nif(&env, c_p, (struct erl_module_nif*)I[2], NULL);
 		nif_bif_result = (*fp)(&env, bif_nif_arity, reg);
 		if (env.exception_thrown)
@@ -3574,10 +3570,7 @@ do {						\
 		PROCESS_MAIN_CHK_LOCKS(c_p);
 		ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		ERTS_MSACC_SET_STATE_CACHED_M_X(ERTS_MSACC_STATE_EMULATOR);
-		if (env.exiting) {
-		    ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
-		    goto do_schedule;
-		}
+		ASSERT(!env.exiting);
 		ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	    }
 
@@ -5162,6 +5155,300 @@ do {						\
     }
 }
 
+/*
+ * dirty_process_main() is what dirty schedulers execute. Since they handle
+ * only NIF calls they do not need to be able to execute all BEAM
+ * instructions.
+ */
+void dirty_process_main(void)
+{
+    Process* c_p = NULL;
+    int reds_used;
+#ifdef DEBUG
+    ERTS_DECLARE_DUMMY(Eterm pid);
+#endif
+    BeamInstr *do_call_nif = OpCode(call_nif);
+
+    /* Pointer to X registers: x(1)..x(N); reg[0] is used when doing GC,
+     * in all other cases x0 is used.
+     */
+    register Eterm* reg REG_xregs = NULL;
+
+    /*
+     * Top of heap (next free location); grows upwards.
+     */
+    register Eterm* HTOP REG_htop = NULL;
+
+    /* Stack pointer.  Grows downwards; points
+     * to last item pushed (normally a saved
+     * continuation pointer).
+     */
+    register Eterm* E REG_stop = NULL;
+
+    /*
+     * Pointer to next threaded instruction.
+     */
+    register BeamInstr *I REG_I = NULL;
+
+    /* Number of reductions left.  This function
+     * returns to the scheduler when FCALLS reaches zero.
+     */
+    register Sint FCALLS REG_fcalls = 0;
+
+    /*
+     * X registers and floating point registers are located in
+     * scheduler specific data.
+     */
+    register FloatDef *freg = NULL;
+
+    /*
+     * For keeping the negative old value of 'reds' when call saving is active.
+     */
+    int neg_o_reds = 0;
+
+    ERTS_MSACC_DECLARE_CACHE_X() /* a cached value of the tsd pointer for msacc */
+
+    ERL_BITS_DECLARE_STATEP; /* Has to be last declaration */
+
+    c_p = NULL;
+    reds_used = 0;
+
+    goto do_dirty_schedule1;
+
+ context_switch:
+    c_p->arity = I[-1];
+    c_p->current = I-3;		/* Pointer to Mod, Func, Arity */
+
+    {
+	Eterm* argp;
+	int i;
+
+	/*
+	 * Make sure that there is enough room for the argument registers to be saved.
+	 */
+	if (c_p->arity > c_p->max_arg_reg) {
+	    /*
+	     * Yes, this is an expensive operation, but you only pay it the first
+	     * time you call a function with more than 6 arguments which is
+	     * scheduled out.  This is better than paying for 26 words of wasted
+	     * space for most processes which never call functions with more than
+	     * 6 arguments.
+	     */
+	    Uint size = c_p->arity * sizeof(c_p->arg_reg[0]);
+	    if (c_p->arg_reg != c_p->def_arg_reg) {
+		c_p->arg_reg = (Eterm *) erts_realloc(ERTS_ALC_T_ARG_REG,
+						      (void *) c_p->arg_reg,
+						      size);
+	    } else {
+		c_p->arg_reg = (Eterm *) erts_alloc(ERTS_ALC_T_ARG_REG, size);
+	    }
+	    c_p->max_arg_reg = c_p->arity;
+	}
+
+	/*
+	 * Since REDS_IN(c_p) is stored in the save area (c_p->arg_reg) we must read it
+	 * now before saving registers.
+	 */
+
+	ASSERT(c_p->debug_reds_in == REDS_IN(c_p));
+	if (!ERTS_PROC_GET_SAVED_CALLS_BUF(c_p))
+	    reds_used = REDS_IN(c_p) - FCALLS;
+	else
+	    reds_used = REDS_IN(c_p) - (CONTEXT_REDS + FCALLS);
+	ASSERT(reds_used >= 0);
+
+	/*
+	 * Save the argument registers and everything else.
+	 */
+
+	argp = c_p->arg_reg;
+	for (i = c_p->arity - 1; i >= 0; i--) {
+	    argp[i] = reg[i];
+	}
+	SWAPOUT;
+	c_p->i = I;
+	goto do_dirty_schedule1;
+    }
+
+ do_dirty_schedule:
+    ASSERT(c_p->debug_reds_in == REDS_IN(c_p));
+    if (!ERTS_PROC_GET_SAVED_CALLS_BUF(c_p))
+	reds_used = REDS_IN(c_p) - FCALLS;
+    else
+	reds_used = REDS_IN(c_p) - (CONTEXT_REDS + FCALLS);
+    ASSERT(reds_used >= 0);
+ do_dirty_schedule1:
+
+    PROCESS_MAIN_CHK_LOCKS(c_p);
+    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
+    c_p = schedule(c_p, reds_used);
+    ASSERT(!(c_p->flags & F_HIPE_MODE));
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
+#ifdef DEBUG
+    pid = c_p->common.id; /* Save for debugging purposes */
+#endif
+    ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
+    PROCESS_MAIN_CHK_LOCKS(c_p);
+
+    ERTS_MSACC_UPDATE_CACHE_X();
+
+    reg = erts_proc_sched_data(c_p)->x_reg_array;
+    freg = erts_proc_sched_data(c_p)->f_reg_array;
+    ERL_BITS_RELOAD_STATEP(c_p);
+    {
+	int reds;
+	Eterm* argp;
+	int i;
+
+	argp = c_p->arg_reg;
+	for (i = c_p->arity - 1; i >= 0; i--) {
+	    reg[i] = argp[i];
+	    CHECK_TERM(reg[i]);
+	}
+
+	/*
+	 * We put the original reduction count in the process structure, to reduce
+	 * the code size (referencing a field in a struct through a pointer stored
+	 * in a register gives smaller code than referencing a global variable).
+	 */
+
+	I = c_p->i;
+
+	REDS_IN(c_p) = reds = c_p->fcalls;
+#ifdef DEBUG
+	c_p->debug_reds_in = reds;
+#endif
+
+	if (ERTS_PROC_GET_SAVED_CALLS_BUF(c_p)) {
+	    neg_o_reds = -CONTEXT_REDS;
+	    FCALLS = neg_o_reds + reds;
+	} else {
+	    neg_o_reds = 0;
+	    FCALLS = reds;
+	}
+
+	ERTS_DBG_CHK_REDS(c_p, FCALLS);
+
+	SWAPIN;
+
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(process_scheduled)) {
+            DTRACE_CHARBUF(process_buf, DTRACE_TERM_BUF_SIZE);
+            DTRACE_CHARBUF(fun_buf, DTRACE_TERM_BUF_SIZE);
+            dtrace_proc_str(c_p, process_buf);
+
+            if (ERTS_PROC_IS_EXITING(c_p)) {
+                strcpy(fun_buf, "<exiting>");
+            } else {
+                BeamInstr *fptr = find_function_from_pc(c_p->i);
+                if (fptr) {
+                    dtrace_fun_decode(c_p, (Eterm)fptr[0],
+                                      (Eterm)fptr[1], (Uint)fptr[2],
+                                      NULL, fun_buf);
+                } else {
+                    erts_snprintf(fun_buf, sizeof(DTRACE_CHARBUF_NAME(fun_buf)),
+                                  "<unknown/%p>", next);
+                }
+            }
+
+            DTRACE2(process_scheduled, process_buf, fun_buf);
+        }
+#endif
+    }
+
+    {
+	Eterm nif_bif_result;
+	Eterm bif_nif_arity;
+
+    OpCase(call_nif):
+	{
+	    /*
+	     * call_nif is always first instruction in function:
+	     *
+	     * I[-3]: Module
+	     * I[-2]: Function
+	     * I[-1]: Arity
+	     * I[0]: &&call_nif
+	     * I[1]: Function pointer to NIF function
+	     * I[2]: Pointer to erl_module_nif
+	     * I[3]: Function pointer to dirty NIF
+	     */
+	    BifFunction vbf;
+
+            if (!((FCALLS - 1) > 0 || (FCALLS - 1) > neg_o_reds)) {
+                /* If we have run out of reductions, we do a context
+                   switch before calling the nif */
+                goto context_switch;
+            }
+
+	    ERTS_MSACC_SET_STATE_CACHED_M_X(ERTS_MSACC_STATE_NIF);
+
+	    DTRACE_NIF_ENTRY(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
+	    c_p->current = I-3; /* current and vbf set to please handle_error */
+	    SWAPOUT;
+	    c_p->fcalls = FCALLS - 1;
+	    PROCESS_MAIN_CHK_LOCKS(c_p);
+	    bif_nif_arity = I[-1];
+	    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
+
+	    ASSERT(!ERTS_PROC_IS_EXITING(c_p));
+	    {
+		typedef Eterm NifF(struct enif_environment_t*, int argc, Eterm argv[]);
+		NifF* fp = vbf = (NifF*) I[1];
+		struct enif_environment_t env;
+		ASSERT(!c_p->scheduler_data);
+		erts_pre_dirty_nif(&env, c_p, (struct erl_module_nif*)I[2], NULL);
+		nif_bif_result = (*fp)(&env, bif_nif_arity, reg);
+		if (env.exception_thrown)
+		    nif_bif_result = THE_NON_VALUE;
+		erts_post_nif(&env);
+		PROCESS_MAIN_CHK_LOCKS(c_p);
+		ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
+		ERTS_MSACC_SET_STATE_CACHED_M_X(ERTS_MSACC_STATE_EMULATOR);
+		if (env.exiting) {
+		    ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
+		    goto do_dirty_schedule;
+		}
+		ASSERT(!ERTS_PROC_IS_EXITING(c_p));
+	    }
+	    DTRACE_NIF_RETURN(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
+	    ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
+	    ERTS_HOLE_CHECK(c_p);
+	    if (ERTS_IS_GC_DESIRED(c_p)) {
+		nif_bif_result = erts_gc_after_bif_call(c_p,
+							nif_bif_result,
+							reg, bif_nif_arity);
+	    }
+	    SWAPIN;  /* There might have been a garbage collection. */
+	    FCALLS = c_p->fcalls;
+	    ERTS_DBG_CHK_REDS(c_p, FCALLS);
+	    if (is_value(nif_bif_result)) {
+		r(0) = nif_bif_result;
+		CHECK_TERM(r(0));
+		I = c_p->cp;
+		c_p->cp = 0;
+		Goto(*I);
+	    } else if (c_p->freason == TRAP) {
+		I = c_p->i;
+		if (c_p->flags & F_HIBERNATE_SCHED) {
+		    c_p->flags &= ~F_HIBERNATE_SCHED;
+		    goto do_dirty_schedule;
+		}
+		DispatchMacro();
+	    }
+	    I = handle_error(c_p, c_p->cp, reg, vbf);
+	}
+    }
+    if (I == 0) {
+	goto do_dirty_schedule;
+    } else {
+	ASSERT(!is_value(r(0)));
+	SWAPIN;
+	Goto(do_call_nif);
+    }
+}
+
 static BifFunction
 translate_gc_bif(void* gcf)
 {
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index 2bbb8e3c91..4437adb57b 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -178,9 +178,11 @@ static ERTS_INLINE void ensure_heap(ErlNifEnv* env, size_t may_need)
 void erts_pre_nif(ErlNifEnv* env, Process* p, struct erl_module_nif* mod_nif,
                   Process* tracee)
 {
+#ifdef DEBUG
 #ifdef ERTS_DIRTY_SCHEDULERS
     ErtsSchedulerData *esdp;
 #endif
+#endif
     env->mod_nif = mod_nif;
     env->proc = p;
     env->hp = HEAP_TOP(p);
@@ -193,12 +195,12 @@ void erts_pre_nif(ErlNifEnv* env, Process* p, struct erl_module_nif* mod_nif,
 
     ASSERT(p->common.id != ERTS_INVALID_PID);
 
+#ifdef DEBUG
 #ifdef ERTS_DIRTY_SCHEDULERS
     esdp = erts_get_scheduler_data();
     ASSERT(esdp);
 
     if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) {
-#ifdef DEBUG
 	erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state);
 
 	ASSERT(p->scheduler_data == esdp);
@@ -206,44 +208,57 @@ void erts_pre_nif(ErlNifEnv* env, Process* p, struct erl_module_nif* mod_nif,
 			 | ERTS_PSFLG_RUNNING_SYS))
 	       && !(state & (ERTS_PSFLG_DIRTY_RUNNING
 			     | ERTS_PSFLG_DIRTY_RUNNING_SYS)));
+    }
 #endif
+#endif
+}
 
-    }
-    else {
-	Process *sproc;
+void erts_pre_dirty_nif(ErlNifEnv* env, Process* p, struct erl_module_nif* mod_nif,
+			Process* tracee)
+{
+#ifdef ERTS_DIRTY_SCHEDULERS
 #ifdef DEBUG
-	erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state);
+    erts_aint32_t state;
+#endif
+    Process *sproc;
+    ErtsSchedulerData *esdp;
+    esdp = erts_get_scheduler_data();
+    ASSERT(esdp);
+
+    erts_pre_nif(env, p, mod_nif, tracee);
 
-	ASSERT(!p->scheduler_data);
-	ASSERT((state & ERTS_PSFLG_DIRTY_RUNNING)
-	       && !(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)));
+#ifdef DEBUG
+    state = erts_smp_atomic32_read_nob(&p->state);
+
+    ASSERT(!p->scheduler_data);
+    ASSERT((state & ERTS_PSFLG_DIRTY_RUNNING)
+	   && !(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)));
 #endif
 
-	sproc = esdp->dirty_shadow_process;
-	ASSERT(sproc);
-	ASSERT(sproc->static_flags & ERTS_STC_FLG_SHADOW_PROC);
-	ASSERT(erts_smp_atomic32_read_nob(&sproc->state)
-	       == (ERTS_PSFLG_ACTIVE
-		   | ERTS_PSFLG_DIRTY_RUNNING
-		   | ERTS_PSFLG_PROXY));
-
-	sproc->next = p;
-	sproc->common.id = p->common.id;
-	sproc->htop = p->htop;
-	sproc->stop = p->stop;
-	sproc->hend = p->hend;
-	sproc->heap = p->heap;
-	sproc->abandoned_heap = p->abandoned_heap;
-	sproc->heap_sz = p->heap_sz;
-	sproc->high_water = p->high_water;
-	sproc->old_hend = p->old_hend;
-	sproc->old_htop = p->old_htop;
-	sproc->old_heap = p->old_heap;
-	sproc->mbuf = NULL;
-	sproc->mbuf_sz = 0;
-	ERTS_INIT_OFF_HEAP(&sproc->off_heap);
-	env->proc = sproc;
-    }
+    sproc = esdp->dirty_shadow_process;
+    ASSERT(sproc);
+    ASSERT(sproc->static_flags & ERTS_STC_FLG_SHADOW_PROC);
+    ASSERT(erts_smp_atomic32_read_nob(&sproc->state)
+	   == (ERTS_PSFLG_ACTIVE
+	       | ERTS_PSFLG_DIRTY_RUNNING
+	       | ERTS_PSFLG_PROXY));
+
+    sproc->next = p;
+    sproc->common.id = p->common.id;
+    sproc->htop = p->htop;
+    sproc->stop = p->stop;
+    sproc->hend = p->hend;
+    sproc->heap = p->heap;
+    sproc->abandoned_heap = p->abandoned_heap;
+    sproc->heap_sz = p->heap_sz;
+    sproc->high_water = p->high_water;
+    sproc->old_hend = p->old_hend;
+    sproc->old_htop = p->old_htop;
+    sproc->old_heap = p->old_heap;
+    sproc->mbuf = NULL;
+    sproc->mbuf_sz = 0;
+    ERTS_INIT_OFF_HEAP(&sproc->off_heap);
+    env->proc = sproc;
 #endif
 }
 
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index f8cbe60e76..e245c9e6bb 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -8197,7 +8197,7 @@ sched_dirty_cpu_thread_func(void *vesdp)
 #endif
     erts_thread_init_float();
 
-    process_main();
+    dirty_process_main();
     /* No schedulers should *ever* terminate */
     erts_exit(ERTS_ABORT_EXIT,
 	     "Dirty CPU scheduler thread number %beu terminated\n",
@@ -8242,7 +8242,7 @@ sched_dirty_io_thread_func(void *vesdp)
 #endif
     erts_thread_init_float();
 
-    process_main();
+    dirty_process_main();
     /* No schedulers should *ever* terminate */
     erts_exit(ERTS_ABORT_EXIT,
 	     "Dirty I/O scheduler thread number %beu terminated\n",
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index b76b9cd874..15253bb53e 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -62,6 +62,8 @@ struct enif_environment_t /* ErlNifEnv */
 extern void erts_pre_nif(struct enif_environment_t*, Process*,
 			 struct erl_module_nif*, Process* tracee);
 extern void erts_post_nif(struct enif_environment_t* env);
+extern void erts_pre_dirty_nif(struct enif_environment_t*, Process*,
+			       struct erl_module_nif*, Process* tracee);
 extern Eterm erts_nif_taints(Process* p);
 extern void erts_print_nif_taints(int to, void* to_arg);
 void erts_unload_nif(struct erl_module_nif* nif);
@@ -1152,6 +1154,7 @@ void print_pass_through(int, byte*, int);
 int catchlevel(Process*);
 void init_emulator(void);
 void process_main(void);
+void dirty_process_main(void);
 Eterm build_stacktrace(Process* c_p, Eterm exc);
 Eterm expand_error_value(Process* c_p, Uint freason, Eterm Value);
 void erts_save_stacktrace(Process* p, struct StackTrace* s, int depth);
diff --git a/erts/emulator/test/dirty_nif_SUITE.erl b/erts/emulator/test/dirty_nif_SUITE.erl
index c3afbc0803..83b098a704 100644
--- a/erts/emulator/test/dirty_nif_SUITE.erl
+++ b/erts/emulator/test/dirty_nif_SUITE.erl
@@ -32,19 +32,23 @@
 	 dirty_nif/1, dirty_nif_send/1,
 	 dirty_nif_exception/1, call_dirty_nif_exception/1,
 	 dirty_scheduler_exit/1, dirty_call_while_terminated/1,
-	 dirty_heap_access/1]).
+	 dirty_heap_access/1, dirty_process_info/1,
+	 dirty_process_register/1, dirty_process_trace/1]).
 
 -define(nif_stub,nif_stub_error(?LINE)).
 
 suite() -> [{ct_hooks,[ts_install_cth]}].
 
-all() -> 
+all() ->
     [dirty_nif,
      dirty_nif_send,
      dirty_nif_exception,
      dirty_scheduler_exit,
      dirty_call_while_terminated,
-     dirty_heap_access].
+     dirty_heap_access,
+     dirty_process_info,
+     dirty_process_register,
+     dirty_process_trace].
 
 init_per_suite(Config) ->
     try erlang:system_info(dirty_cpu_schedulers) of
@@ -187,7 +191,7 @@ dirty_call_while_terminated(Config) when is_list(Config) ->
 				    blipp:blupp(Bin)
 			    end,
 			    [monitor,link]),
-    receive {dirty_alive, Pid} -> ok end,
+    receive {dirty_alive, _Pid} -> ok end,
     {value, {BinAddr, 4711, 2}} = lists:keysearch(4711, 2,
 						  element(2,
 							  process_info(self(),
@@ -241,7 +245,7 @@ dirty_heap_access(Config) when is_list(Config) ->
 		       end),
     {N, R} = access_dirty_heap(Dirty, RGL, 0, 0),
     receive
-	{Pid, Res} ->
+	{_Pid, Res} ->
 	    1000 = length(Res),
 	    lists:foreach(fun (X) -> Ref = X end, Res)
     end,
@@ -269,12 +273,123 @@ access_dirty_heap(Dirty, RGL, N, R) ->
 					       end)
     end.
 
+%% These tests verify that processes that access a process executing a
+%% dirty NIF where the main lock is needed for that access do not get
+%% blocked. Each test passes its pid to dirty_sleeper, which sends a
+%% 'ready' message when it's running on a dirty scheduler and just before
+%% it starts a 6 second sleep. When it receives the message, it verifies
+%% that access to the dirty process is as it expects.  After the dirty
+%% process finishes its 6 second sleep but before it returns from the dirty
+%% scheduler, it sends a 'done' message. If the tester already received
+%% that message, the test fails because it means attempting to access the
+%% dirty process waited for that process to return to a regular scheduler,
+%% so verify that we haven't received that message, and also verify that
+%% the dirty process is still alive immediately after accessing it.
+dirty_process_info(Config) when is_list(Config) ->
+    access_dirty_process(
+      Config,
+      fun() -> ok end,
+      fun(NifPid) ->
+	      PI = process_info(NifPid),
+	      {current_function,{?MODULE,dirty_sleeper,1}} =
+		  lists:keyfind(current_function, 1, PI),
+	      ok
+      end,
+      fun(_) -> ok end).
+
+dirty_process_register(Config) when is_list(Config) ->
+    access_dirty_process(
+      Config,
+      fun() -> ok end,
+      fun(NifPid) ->
+	      register(test_dirty_process_register, NifPid),
+	      NifPid = whereis(test_dirty_process_register),
+	      unregister(test_dirty_process_register),
+	      false = lists:member(test_dirty_process_register,
+				   registered()),
+	      ok
+      end,
+      fun(_) -> ok end).
+
+dirty_process_trace(Config) when is_list(Config) ->
+    access_dirty_process(
+      Config,
+      fun() ->
+	      erlang:trace_pattern({?MODULE,dirty_sleeper,1},
+				   [{'_',[],[{return_trace}]}],
+				   [local,meta]),
+	      ok
+      end,
+      fun(NifPid) ->
+	      erlang:trace(NifPid, true, [call,timestamp]),
+	      ok
+      end,
+      fun(NifPid) ->
+	      receive
+		  done ->
+		      receive
+			  {trace_ts,NifPid,call,{?MODULE,dirty_sleeper,_},_} ->
+			      ok
+		      after
+			  0 ->
+			      error(missing_trace_call_message)
+		      end,
+		      receive
+			  {trace_ts,NifPid,return_from,{?MODULE,dirty_sleeper,1},
+			   ok,_} ->
+			      ok
+		      after
+			  100 ->
+			      error(missing_trace_return_message)
+		      end
+	      after
+		  6500 ->
+		      error(missing_done_message)
+	      end,
+	      ok
+      end).
+
 %%
 %% Internal...
 %%
 
+access_dirty_process(Config, Start, Test, Finish) ->
+    {ok, Node} = start_node(Config, ""),
+    [ok] = mcall(Node,
+		 [fun() ->
+                          Path = ?config(data_dir, Config),
+                          Lib = atom_to_list(?MODULE),
+                          ok = erlang:load_nif(filename:join(Path,Lib), []),
+			  ok = test_dirty_process_access(Start, Test, Finish)
+		  end]),
+    stop_node(Node),
+    ok.
+
+test_dirty_process_access(Start, Test, Finish) ->
+    ok = Start(),
+    Self = self(),
+    NifPid = spawn_link(fun() ->
+				ok = dirty_sleeper(Self)
+			end),
+    ok = receive
+	     ready ->
+		 ok = Test(NifPid),
+		 receive
+		     done ->
+			 error(dirty_process_info_blocked)
+		 after
+		     0 ->
+			 true = erlang:is_process_alive(NifPid),
+			 ok
+		 end
+	 after
+	     3000 ->
+		 error(timeout)
+	 end,
+    ok = Finish(NifPid).
+
 receive_any() ->
-    receive M -> M end.	     
+    receive M -> M end.
 
 start_node(Config) ->
     start_node(Config, "").
@@ -314,13 +429,13 @@ mcall(Node, Funs) ->
 
 %% The NIFs:
 lib_loaded() -> false.
-call_nif_schedule(_,_) -> ?nif_stub.
 call_dirty_nif(_,_,_) -> ?nif_stub.
 send_from_dirty_nif(_) -> ?nif_stub.
 call_dirty_nif_exception(_) -> ?nif_stub.
 call_dirty_nif_zero_args() -> ?nif_stub.
 dirty_call_while_terminated_nif(_) -> ?nif_stub.
 dirty_sleeper() -> ?nif_stub.
+dirty_sleeper(_) -> ?nif_stub.
 dirty_heap_access_nif(_) -> ?nif_stub.
 
 nif_stub_error(Line) ->
diff --git a/erts/emulator/test/dirty_nif_SUITE_data/dirty_nif_SUITE.c b/erts/emulator/test/dirty_nif_SUITE_data/dirty_nif_SUITE.c
index 2013c88167..1d2a099186 100644
--- a/erts/emulator/test/dirty_nif_SUITE_data/dirty_nif_SUITE.c
+++ b/erts/emulator/test/dirty_nif_SUITE_data/dirty_nif_SUITE.c
@@ -146,12 +146,31 @@ static ERL_NIF_TERM call_dirty_nif_zero_args(ErlNifEnv* env, int argc, const ERL
 static ERL_NIF_TERM
 dirty_sleeper(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
 {
+    ErlNifPid pid;
+    ErlNifEnv* msg_env = NULL;
+
     assert(enif_is_on_dirty_scheduler(env));
+
+    /* If we get a pid argument, it indicates a process involved in the
+       test wants a message from us. Prior to the sleep we send a 'ready'
+       message, and then after the sleep, send a 'done' message. */
+    if (argc == 1 && enif_get_local_pid(env, argv[0], &pid)) {
+        msg_env = enif_alloc_env();
+        enif_send(env, &pid, msg_env, enif_make_atom(msg_env, "ready"));
+    }
+
 #ifdef __WIN32__
     Sleep(6000);
 #else
     sleep(6);
 #endif
+
+    if (argc == 1) {
+        assert(msg_env != NULL);
+        enif_send(env, &pid, msg_env, enif_make_atom(msg_env, "done"));
+        enif_free_env(msg_env);
+    }
+
     return enif_make_atom(env, "ok");
 }
 
@@ -216,6 +235,7 @@ static ErlNifFunc nif_funcs[] =
     {"call_dirty_nif_exception", 1, call_dirty_nif_exception, ERL_NIF_DIRTY_JOB_IO_BOUND},
     {"call_dirty_nif_zero_args", 0, call_dirty_nif_zero_args, ERL_NIF_DIRTY_JOB_CPU_BOUND},
     {"dirty_sleeper", 0, dirty_sleeper, ERL_NIF_DIRTY_JOB_IO_BOUND},
+    {"dirty_sleeper", 1, dirty_sleeper, ERL_NIF_DIRTY_JOB_CPU_BOUND},
     {"dirty_call_while_terminated_nif", 1, dirty_call_while_terminated_nif, ERL_NIF_DIRTY_JOB_CPU_BOUND},
     {"dirty_heap_access_nif", 1, dirty_heap_access_nif, ERL_NIF_DIRTY_JOB_CPU_BOUND}
 };