add enif_schedule_nif() to NIF API

In the #erlang IRC channel Anthony Ramine once mentioned the idea of
allowing a NIF to use an emulator trap, similar to a BIF trap, to schedule
another NIF for execution. This is exactly how dirty NIFs were implemented
for Erlang/OTP 17.0, so this commit refactors and generalizes that dirty
NIF code to support a new enif_schedule_nif() API function.

The enif_schedule_nif() function allows a long-running NIF to be broken
into separate NIF invocations. The NIF first executes part of the
long-running task, then calls enif_schedule_nif() to schedule a NIF for
later execution to continue the task. Any number of NIFs can be scheduled
in this manner, one after another. Since the emulator regains control
between invocations, this helps avoid problems caused by native code tying
up scheduler threads for too long.

The enif_schedule_nif() function also replaces the original experimental
dirty NIF API. The function takes a flags parameter that a caller can use
to indicate the NIF should be scheduled onto either a dirty CPU scheduler
thread, a dirty I/O scheduler thread, or scheduled as a regular NIF on a
regular scheduler thread. With this change, the original experimental
enif_schedule_dirty_nif(), enif_schedule_dirty_nif_finalizer() and
enif_dirty_nif_finalizer() API functions are no longer needed and have been
removed. Explicit scheduling of a dirty NIF finalization function is no
longer necessary; if an application wants similar functionality, it can
have a dirty NIF just invoke enif_schedule_nif() to schedule a non-dirty
NIF to complete its task.

Lift the restriction that dirty NIFs can't call enif_make_badarg() to raise
an exception. This was a problem with the original dirty NIF API because it
forced developers to get and check all incoming arguments in a regular NIF,
and then schedule the dirty NIF which then had to get all the arguments
again. Now, the argument checking can be done in the dirty NIF and it can
call enif_make_badarg() itself to flag incorrect arguments.

Extend the ErlNifFunc struct with a new flags field that allows NIFs to be
declared as dirty. The default value for this field is 0, indicating a
regular NIF, so it's backwards compatible with all existing statically
initialized ErlNifFunc struct instances, and so such instances require no
code changes. Defining the flags field with a value of
ERL_NIF_DIRTY_JOB_CPU_BOUND indicates that the NIF should execute on a
dirty CPU scheduler thread, or defining it with a value of
ERL_NIF_DIRTY_JOB_IO_BOUND indicates that the NIF should execute on a dirty
I/O scheduler thread. Any other flags field value causes a NIF library
loading error.

Extend the ErlNifEntry struct with a new options field that indicates
whether a NIF library was built with support for optional features such as
dirty NIFs. When a NIF library is loaded, the runtime checks the options
field to ensure compatibility. If a NIF library built with dirty NIF
support is loaded into a runtime that does not support dirty NIFs, and the
library defines one or more ErlNifFunc entries with non-zero flags fields
indicating dirty NIFs, a NIF library loading error results. There is no
error if a NIF library built with dirty NIF support is loaded into a
runtime that does not support dirty NIFs but the library does not have any
dirty NIFs. It is also not an error if a library without dirty NIF support
is loaded into a runtime built with dirty NIF support.

Add documentation and tests for enif_schedule_nif().

10 files changed, 564 insertions, 182 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 1026e5f649..8bfb7d2ad2 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -3503,6 +3503,7 @@ get_map_elements_fail:
 	     * 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;
 
@@ -3523,13 +3524,6 @@ get_map_elements_fail:
 		reg[0] = r(0);
 		nif_bif_result = (*fp)(&env, bif_nif_arity, reg);
 		erts_post_nif(&env);
-#ifdef ERTS_DIRTY_SCHEDULERS
-		if (is_non_value(nif_bif_result) && c_p->freason == TRAP) {
-		    Export* ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(c_p);
-		    ep->code[0] = I[-3];
-		    ep->code[1] = I[-2];
-		}
-#endif
 	    }
 	    ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result));
 	    PROCESS_MAIN_CHK_LOCKS(c_p);
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index e96177cfd9..cfc6146b0a 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2363,7 +2363,11 @@ load_code(LoaderState* stp)
 
 		if (stp->may_load_nif) {
 		    const int finfo_ix = ci - FUNC_INFO_SZ;
-		    enum { MIN_FUNC_SZ = 3 };		    
+#ifdef ERTS_DIRTY_SCHEDULERS
+		    enum { MIN_FUNC_SZ = 4 };
+#else
+		    enum { MIN_FUNC_SZ = 3 };
+#endif
 		    if (finfo_ix - last_func_start < MIN_FUNC_SZ && last_func_start) {		   
 			/* Must make room for call_nif op */
 			int pad = MIN_FUNC_SZ - (finfo_ix - last_func_start);
diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c
index aa15d2cc57..0db42d4325 100644
--- a/erts/emulator/beam/erl_gc.c
+++ b/erts/emulator/beam/erl_gc.c
@@ -2018,6 +2018,20 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset)
 	roots[n].sz = 1;
 	n++;
     }
+
+    /*
+     * If a NIF has saved arguments, they need to be added
+     */
+    if (ERTS_PROC_GET_NIF_TRAP_EXPORT(p)) {
+	Eterm* argv;
+	int argc;
+	if (erts_setup_nif_gc(p, &argv, &argc)) {
+	    roots[n].v = argv;
+	    roots[n].sz = argc;
+	    n++;
+	}
+    }
+
     ASSERT(n <= rootset->size);
 
     mp = p->msg.first;
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index ff551ea3af..1414744763 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -1513,72 +1513,251 @@ int enif_consume_timeslice(ErlNifEnv* env, int percent)
     return ERTS_BIF_REDS_LEFT(env->proc) == 0;
 }
 
-#ifdef ERTS_DIRTY_SCHEDULERS
-
-/* NIFs exports need one more item than the Export struct provides, the
- * erl_module_nif*, so the DirtyNifExport below adds that. The Export
- * member must be first in the struct.
+/*
+ * NIF exports need a few more items than the Export struct provides,
+ * including the erl_module_nif* and a NIF function pointer, so the
+ * NifExport below adds those. The Export member must be first in the
+ * struct. The saved_mfa, saved_argc, nif_level, alloced_argv_sz and argv
+ * members are used to track the MFA and arguments of the top NIF in case a
+ * chain of one or more enif_schedule_nif() calls results in an exception,
+ * since in that case the original MFA and registers have to be restored
+ * before returning to Erlang to ensure stacktrace information associated
+ * with the exception is correct.
  */
+typedef ERL_NIF_TERM (*NativeFunPtr)(ErlNifEnv*, int, const ERL_NIF_TERM[]);
+
 typedef struct {
     Export exp;
     struct erl_module_nif* m;
-} DirtyNifExport;
+    NativeFunPtr fp;
+    Eterm saved_mfa[3];
+    int saved_argc;
+    int alloced_argv_sz;
+    Eterm argv[1];
+} NifExport;
 
-static void
-alloc_proc_psd(Process* proc, DirtyNifExport **ep)
+/*
+ * If a process has saved arguments, they need to be part of the GC
+ * rootset. The function below is called from setup_rootset() in
+ * erl_gc.c. This function is declared in erl_process.h.
+ */
+int
+erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj)
+{
+    NifExport* ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    int gc = (ep && ep->saved_argc > 0);
+
+    if (gc) {
+	*objv = ep->argv;
+	*nobj = ep->saved_argc;
+    }
+    return gc;
+}
+
+/*
+ * Allocate a NifExport and set it in proc specific data
+ */
+static NifExport*
+allocate_nif_sched_data(Process* proc, int argc)
 {
+    NifExport* ep;
+    size_t argv_extra, total;
     int i;
-    if (!*ep) {
-	*ep = erts_alloc(ERTS_ALC_T_PSD, sizeof(DirtyNifExport));
-	sys_memset((void*) *ep, 0, sizeof(DirtyNifExport));
-	for (i=0; i<ERTS_NUM_CODE_IX; i++) {
-	    (*ep)->exp.addressv[i] = &(*ep)->exp.code[3];
-	}
-	(*ep)->exp.code[3] = (BeamInstr) em_call_nif;
+
+    argv_extra = argc > 1 ? sizeof(Eterm)*(argc-1) : 0;
+    total = sizeof(NifExport) + argv_extra;
+    ep = erts_alloc(ERTS_ALC_T_PSD, total);
+    sys_memset((void*) ep, 0, total);
+    ep->alloced_argv_sz = argc;
+    for (i=0; i<ERTS_NUM_CODE_IX; i++) {
+	ep->exp.addressv[i] = &ep->exp.code[3];
     }
-    (void) ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, &(*ep)->exp);
+    ep->exp.code[3] = (BeamInstr) em_call_nif;
+    (void) ERTS_PROC_SET_NIF_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, &ep->exp);
+    return ep;
 }
 
+/*
+ * Initialize a NifExport struct. Create it if needed and store it in the
+ * proc. The direct_fp function is what will be invoked by op_call_nif, and
+ * the indirect_fp function, if not NULL, is what the direct_fp function
+ * will call. If the allocated NifExport isn't enough to hold all of argv,
+ * allocate a larger one. Save MFA and registers only if the need_save
+ * parameter is true.
+ */
 static ERL_NIF_TERM
-execute_dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+init_nif_sched_data(ErlNifEnv* env, NativeFunPtr direct_fp, NativeFunPtr indirect_fp,
+		    int need_save, int argc, const ERL_NIF_TERM argv[])
 {
-    Eterm* reg = ERTS_PROC_GET_SCHDATA(env->proc)->x_reg_array;
-    ERL_NIF_TERM result, dirty_result = (ERL_NIF_TERM) reg[0];
-    typedef ERL_NIF_TERM (*FinalizerFP)(ErlNifEnv*, ERL_NIF_TERM);
-    FinalizerFP fp;
-#if HAVE_INT64 && SIZEOF_LONG != 8
-    ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64));
-    enif_get_uint64(env, reg[1], (ErlNifUInt64 *) &fp);
-#else
-    ASSERT(sizeof(fp) <= sizeof(unsigned long));
-    enif_get_ulong(env, reg[1], (unsigned long *) &fp);
-#endif
-    result = (*fp)(env, dirty_result);
-    if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0
-	&& env->mod_nif->mod == NULL)
-	close_lib(env->mod_nif);
-    return result;
+    Process* proc = env->proc;
+    Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array;
+    NifExport* ep;
+    int i;
+
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    if (!ep)
+	ep = allocate_nif_sched_data(proc, argc);
+    else if (need_save && ep->alloced_argv_sz < argc) {
+	NifExport* new_ep = allocate_nif_sched_data(proc, argc);
+	erts_free(ERTS_ALC_T_PSD, (void*) ep);
+	ep = new_ep;
+    }
+    ERTS_VBUMP_ALL_REDS(proc);
+    for (i = 0; i < argc; i++) {
+	if (need_save)
+	    ep->argv[i] = reg[i];
+	reg[i] = (Eterm) argv[i];
+    }
+    if (need_save) {
+	ep->saved_mfa[0] = proc->current[0];
+	ep->saved_mfa[1] = proc->current[1];
+	ep->saved_mfa[2] = proc->current[2];
+	ep->saved_argc = argc;
+    }
+    proc->i = (BeamInstr*) ep->exp.addressv[0];
+    ep->exp.code[0] = (BeamInstr) proc->current[0];
+    ep->exp.code[1] = (BeamInstr) proc->current[1];
+    ep->exp.code[2] = argc;
+    ep->exp.code[4] = (BeamInstr) direct_fp;
+    ep->m = env->mod_nif;
+    ep->fp = indirect_fp;
+    proc->freason = TRAP;
+    return THE_NON_VALUE;
 }
 
-#endif /* ERTS_DIRTY_SCHEDULERS */
+/*
+ * Restore saved MFA and registers. Registers are restored only when the
+ * exception flag is true.
+ */
+static void
+restore_nif_mfa(Process* proc, NifExport* ep, int exception)
+{
+    int i;
+    Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array;
 
-#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
+    proc->current[0] = ep->saved_mfa[0];
+    proc->current[1] = ep->saved_mfa[1];
+    proc->current[2] = ep->saved_mfa[2];
+    if (exception)
+	for (i = 0; i < ep->saved_argc; i++)
+	    reg[i] = ep->argv[i];
+    ep->saved_argc = 0;
+    ep->saved_mfa[0] = THE_NON_VALUE;
+}
 
-ERL_NIF_TERM
-enif_schedule_dirty_nif(ErlNifEnv* env, int flags,
-			ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]),
-			int argc, const ERL_NIF_TERM argv[])
+#ifdef ERTS_DIRTY_SCHEDULERS
+
+/*
+ * Finalize a dirty NIF call. This function is scheduled to cause the VM to
+ * switch the process off a dirty scheduler thread and back onto a regular
+ * scheduler thread, and then return the result from the dirty NIF. It also
+ * restores the original NIF MFA when necessary based on the value of
+ * ep->fp set by execute_dirty_nif via init_nif_sched_data -- non-NULL
+ * means restore, NULL means do not restore.
+ */
+static ERL_NIF_TERM
+dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    Process* proc = env->proc;
+    NifExport* ep;
+
+    ASSERT(argc == 1);
+    ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data));
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    if (ep->fp)
+	restore_nif_mfa(proc, ep, 0);
+    return argv[0];
+}
+
+/* Finalize a dirty NIF call that raised an exception.  Otherwise same as
+ * the dirty_nif_finalizer() function.
+ */
+static ERL_NIF_TERM
+dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    Process* proc = env->proc;
+    NifExport* ep;
+
+    ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data));
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    if (ep->fp)
+	restore_nif_mfa(proc, ep, 1);
+    return enif_make_badarg(env);
+}
+
+/*
+ * Dirty NIF execution wrapper function. Invoke an application's dirty NIF,
+ * then check the result and schedule the appropriate finalizer function
+ * where needed. Also restore the original NIF MFA when appropriate.
+ */
+static ERL_NIF_TERM
+execute_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    Process* proc = env->proc;
+    NativeFunPtr fp = (NativeFunPtr) proc->current[6];
+    NifExport* ep;
+    ERL_NIF_TERM result;
+
+    ASSERT(ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data));
+
+    /*
+     * Set ep->fp to NULL before the native call so we know later whether it scheduled another NIF for execution
+     */
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    ep->fp = NULL;
+    result = (*fp)(env, argc, argv);
+    erts_smp_atomic32_read_band_mb(&proc->state,
+				   ~(ERTS_PSFLG_DIRTY_CPU_PROC
+				     |ERTS_PSFLG_DIRTY_IO_PROC
+				     |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q
+				     |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q));
+    if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 && env->mod_nif->mod == NULL)
+	close_lib(env->mod_nif);
+    /*
+     * If no more NIFs were scheduled by the native call via
+     * enif_schedule_nif(), then ep->fp will still be NULL as set above, in
+     * which case we need to restore the original NIF calling
+     * context. Reuse fp essentially as a boolean for this, passing it to
+     * init_nif_sched_data below. Both dirty_nif_exception and
+     * dirty_nif_finalizer then check ep->fp to decide whether or not to
+     * restore the original calling context.
+     */
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    if (ep->fp)
+	fp = NULL;
+    if (is_non_value(result)) {
+	if (proc->freason != TRAP) {
+	    ASSERT(proc->freason == BADARG);
+	    return init_nif_sched_data(env, dirty_nif_exception, fp, 0, argc, argv);
+	} else {
+	    if (ep->fp == NULL)
+		restore_nif_mfa(proc, ep, 1);
+	    return result;
+	}
+    }
+    else
+	return init_nif_sched_data(env, dirty_nif_finalizer, fp, 0, 1, &result);
+}
+
+/*
+ * Dirty NIF scheduling wrapper function. Schedule a dirty NIF to execute
+ * via the execute_dirty_nif() wrapper function. The dirty scheduler thread
+ * type (CPU or I/O) is indicated in flags parameter.
+ */
+static ERTS_INLINE ERL_NIF_TERM
+schedule_dirty_nif(ErlNifEnv* env, int flags, int argc, const ERL_NIF_TERM argv[])
 {
-#ifdef USE_THREADS
     erts_aint32_t state, n, a;
     Process* proc = env->proc;
-    Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array;
-    DirtyNifExport* ep = NULL;
-    int i;
+    NativeFunPtr fp = (NativeFunPtr) proc->current[6];
+    NifExport* ep;
+    int need_save;
 
-    int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND));
-    if (chkflgs != ERL_NIF_DIRTY_JOB_IO_BOUND && chkflgs != ERL_NIF_DIRTY_JOB_CPU_BOUND)
-	return enif_make_badarg(env);
+    ASSERT(flags==ERL_NIF_DIRTY_JOB_IO_BOUND || flags==ERL_NIF_DIRTY_JOB_CPU_BOUND);
 
     a = erts_smp_atomic32_read_acqb(&proc->state);
     while (1) {
@@ -1590,7 +1769,7 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags,
 	 */
 	n &= ~(ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC
 	       |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q);
-	if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND)
+	if (flags == ERL_NIF_DIRTY_JOB_CPU_BOUND)
 	    n |= ERTS_PSFLG_DIRTY_CPU_PROC;
 	else
 	    n |= ERTS_PSFLG_DIRTY_IO_PROC;
@@ -1598,69 +1777,100 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags,
 	if (a == state)
 	    break;
     }
-    if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc)))
-	alloc_proc_psd(proc, &ep);
-    ERTS_VBUMP_ALL_REDS(proc);
-    ep->exp.code[2] = argc;
-    for (i = 0; i < argc; i++) {
-	reg[i] = (Eterm) argv[i];
-    }
-    proc->i = (BeamInstr*) ep->exp.addressv[0];
-    ep->exp.code[4] = (BeamInstr) fp;
-    ep->m = env->mod_nif;
-    proc->freason = TRAP;
-
     erts_refc_inc(&env->mod_nif->rt_dtor_cnt, 1);
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    need_save = (ep == NULL || is_non_value(ep->saved_mfa[0]));
+    return init_nif_sched_data(env, execute_dirty_nif, fp, need_save, argc, argv);
+}
 
-    return THE_NON_VALUE;
-#else
-    return (*fp)(env, argc, argv);
-#endif
+static ERL_NIF_TERM
+schedule_dirty_io_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_IO_BOUND, argc, argv);
+}
+
+static ERL_NIF_TERM
+schedule_dirty_cpu_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, argc, argv);
+}
+
+#endif /* ERTS_DIRTY_SCHEDULERS */
+
+/*
+ * NIF execution wrapper used by enif_schedule_nif() for regular NIFs. It
+ * calls the actual NIF, restores original NIF MFA if necessary, and
+ * then returns the NIF result.
+ */
+static ERL_NIF_TERM
+execute_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    Process* proc = env->proc;
+    NativeFunPtr fp = (NativeFunPtr) proc->current[6];
+    NifExport* ep;
+    ERL_NIF_TERM result;
+
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    ep->fp = NULL;
+    result = (*fp)(env, argc, argv);
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    /*
+     * If no NIFs were scheduled by the native call via
+     * enif_schedule_nif(), then ep->fp will still be NULL as set above, in
+     * which case we need to restore the original NIF MFA.
+     */
+    if (ep->fp == NULL)
+	restore_nif_mfa(proc, ep, is_non_value(result) && proc->freason != TRAP);
+    return result;
 }
 
 ERL_NIF_TERM
-enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result,
-				  ERL_NIF_TERM (*fp)(ErlNifEnv*, ERL_NIF_TERM))
+enif_schedule_nif(ErlNifEnv* env, const char* fun_name, int flags,
+		  ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]),
+		  int argc, const ERL_NIF_TERM argv[])
 {
-#ifdef USE_THREADS
     Process* proc = env->proc;
-    Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array;
-    DirtyNifExport* ep;
+    NifExport* ep;
+    ERL_NIF_TERM fun_name_atom, result;
+    int need_save;
 
-    erts_smp_atomic32_read_band_mb(&proc->state,
-				   ~(ERTS_PSFLG_DIRTY_CPU_PROC
-				     |ERTS_PSFLG_DIRTY_IO_PROC
-				     |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q
-				     |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q));
-    if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc)))
-	alloc_proc_psd(proc, &ep);
-    ERTS_VBUMP_ALL_REDS(proc);
-    ep->exp.code[2] = 2;
-    reg[0] = (Eterm) result;
-#if HAVE_INT64 && SIZEOF_LONG != 8
-    ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64));
-    reg[1] = (Eterm) enif_make_uint64(env, (ErlNifUInt64) fp);
-#else
-    ASSERT(sizeof(fp) <= sizeof(unsigned long));
-    reg[1] = (Eterm) enif_make_ulong(env, (unsigned long) fp);
-#endif
-    proc->i = (BeamInstr*) ep->exp.addressv[0];
-    ep->exp.code[4] = (BeamInstr) execute_dirty_nif_finalizer;
-    proc->freason = TRAP;
+    if (argc > MAX_ARG)
+	return enif_make_badarg(env);
+    fun_name_atom = enif_make_atom(env, fun_name);
+    if (enif_is_exception(env, fun_name_atom))
+	return fun_name_atom;
 
-    return THE_NON_VALUE;
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    need_save = (ep == NULL || is_non_value(ep->saved_mfa[0]));
+
+    if (flags) {
+#ifdef ERTS_DIRTY_SCHEDULERS
+	NativeFunPtr sched_fun;
+	int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND));
+	if (chkflgs == ERL_NIF_DIRTY_JOB_IO_BOUND)
+	    sched_fun = schedule_dirty_io_nif;
+	else if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND)
+	    sched_fun = schedule_dirty_cpu_nif;
+	else
+	    return enif_make_badarg(env);
+	result = init_nif_sched_data(env, sched_fun, fp, need_save, argc, argv);
 #else
-    return (*fp)(env, result);
+	return enif_make_badarg(env);
 #endif
-}
+    }
+    else
+	result = init_nif_sched_data(env, execute_nif, fp, need_save, argc, argv);
 
-/* A simple finalizer that just returns its result argument */
-ERL_NIF_TERM
-enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result)
-{
+    ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc);
+    ASSERT(ep);
+    ep->exp.code[1] = (BeamInstr) fun_name_atom;
     return result;
 }
 
+#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
+
 int
 enif_is_on_dirty_scheduler(ErlNifEnv* env)
 {
@@ -1977,6 +2187,35 @@ static Eterm load_nif_error(Process* p, const char* atom, const char* format, ..
     return ret;
 }
 
+/*
+ * The function below is for looping through ErlNifFunc arrays, helping
+ * provide backwards compatibility across the version 2.7 change that added
+ * the "flags" field to ErlNifFunc.
+ */
+static ErlNifFunc* next_func(ErlNifEntry* entry, int* incrp, ErlNifFunc* func)
+{
+    ASSERT(incrp);
+    if (!*incrp) {
+	if (entry->major > 2 || (entry->major == 2 && entry->minor >= 7))
+	    *incrp = sizeof(ErlNifFunc);
+	else {
+	    /*
+	     * ErlNifFuncV1 below is what ErlNifFunc was before the
+	     * addition of the flags field for 2.7, and is needed to handle
+	     * backward compatibility.
+	     */
+	    typedef struct {
+		const char* name;
+		unsigned arity;
+		ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]);
+	    }ErlNifFuncV1;
+	    *incrp = sizeof(ErlNifFuncV1);
+	}
+    }
+    return (ErlNifFunc*) ((char*)func + *incrp);
+}
+
+
 BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 {
     static const char bad_lib[] = "bad_lib";
@@ -2086,22 +2325,48 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
     }
     else {
 	/*erts_fprintf(stderr, "Found module %T\r\n", mod_atom);*/
-    
+
+	int maybe_dirty_nifs = ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7))
+				&& (entry->options & ERL_NIF_DIRTY_NIF_OPTION));
+	int incr = 0;
+	ErlNifFunc* f = entry->funcs;
 	for (i=0; i < entry->num_of_funcs && ret==am_ok; i++) {
 	    BeamInstr** code_pp;
-	    ErlNifFunc* f = &entry->funcs[i];
 	    if (!erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1)
 		|| (code_pp = get_func_pp(mod->curr.code, f_atom, f->arity))==NULL) {
 		ret = load_nif_error(BIF_P,bad_lib,"Function not found %T:%s/%u",
 				     mod_atom, f->name, f->arity);
-	    }    
-	    else if (code_pp[1] - code_pp[0] < (5+3)) {
+	    }
+	    else if (maybe_dirty_nifs && f->flags) {
+		/*
+		 * If the flags field is non-zero and this emulator was
+		 * built with dirty scheduler support, check that the flags
+		 * value is legal. But if this emulator was built without
+		 * dirty scheduler support, treat a non-zero flags field as
+		 * a load error.
+		 */
+#ifdef ERTS_DIRTY_SCHEDULERS
+		if (f->flags != ERL_NIF_DIRTY_JOB_IO_BOUND && f->flags != ERL_NIF_DIRTY_JOB_CPU_BOUND)
+		    ret = load_nif_error(BIF_P, bad_lib, "Illegal flags field value %d for NIF %T:%s/%u",
+					 f->flags, mod_atom, f->name, f->arity);
+#else
+		ret = load_nif_error(BIF_P, bad_lib, "NIF %T:%s/%u requires a runtime with dirty scheduler support.",
+				     mod_atom, f->name, f->arity);
+#endif
+	    }
+#ifdef ERTS_DIRTY_SCHEDULERS
+	    else if (code_pp[1] - code_pp[0] < (5+4))
+#else
+	    else if (code_pp[1] - code_pp[0] < (5+3))
+#endif
+	    {
 		ret = load_nif_error(BIF_P,bad_lib,"No explicit call to load_nif"
-				     " in module (%T:%s/%u to small)",
-				     mod_atom, entry->funcs[i].name, entry->funcs[i].arity);
+				     " in module (%T:%s/%u too small)",
+				     mod_atom, f->name, f->arity);
 	    }
 	    /*erts_fprintf(stderr, "Found NIF %T:%s/%u\r\n",
-			 mod_atom, entry->funcs[i].name, entry->funcs[i].arity);*/
+	      mod_atom, f->name, f->arity);*/
+	    f = next_func(entry, &incr, f);
 	}
     }
 
@@ -2127,7 +2392,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 	 * is deprecated and was only ment as a development feature not to
 	 * be used in production systems. (See warning below)
 	 */
-	int k;
+	int k, old_incr = 0;
+	ErlNifFunc* old_func;
         lib->priv_data = mod->curr.nif->priv_data;
 
 	ASSERT(mod->curr.nif->entry != NULL);
@@ -2136,13 +2402,16 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 	    goto error;
 	}
 	/* Check that no NIF is removed */
+	old_func = mod->curr.nif->entry->funcs;
 	for (k=0; k < mod->curr.nif->entry->num_of_funcs; k++) {
-	    ErlNifFunc* old_func = &mod->curr.nif->entry->funcs[k];
+	    int incr = 0;
+	    ErlNifFunc* f = entry->funcs;
 	    for (i=0; i < entry->num_of_funcs; i++) {
-		if (old_func->arity == entry->funcs[i].arity
-		    && sys_strcmp(old_func->name, entry->funcs[i].name) == 0) {			   
+		if (old_func->arity == f->arity
+		    && sys_strcmp(old_func->name, f->name) == 0) {
 		    break;
 		}
+		f = next_func(entry, &incr, f);
 	    }
 	    if (i == entry->num_of_funcs) {
 		ret = load_nif_error(BIF_P,reload,"Reloaded library missing "
@@ -2150,7 +2419,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 				     old_func->name, old_func->arity);
 		goto error;
 	    }
-	}       
+	    old_func = next_func(mod->curr.nif->entry, &old_incr, old_func);
+	}
 	erts_pre_nif(&env, BIF_P, lib);
 	veto = entry->reload(&env, &lib->priv_data, BIF_ARG_2);
 	erts_post_nif(&env);
@@ -2197,13 +2467,17 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 	/*
 	** Everything ok, patch the beam code with op_call_nif
 	*/
-        mod->curr.nif = lib;
+
+	int incr = 0;
+	ErlNifFunc* f = entry->funcs;
+
+	mod->curr.nif = lib;
 	for (i=0; i < entry->num_of_funcs; i++)
 	{
 	    BeamInstr* code_ptr;
-	    erts_atom_get(entry->funcs[i].name, sys_strlen(entry->funcs[i].name), &f_atom, ERTS_ATOM_ENC_LATIN1); 
-	    code_ptr = *get_func_pp(mod->curr.code, f_atom, entry->funcs[i].arity);
-	    
+	    erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1);
+	    code_ptr = *get_func_pp(mod->curr.code, f_atom, f->arity);
+
 	    if (code_ptr[1] == 0) {
 		code_ptr[5+0] = (BeamInstr) BeamOp(op_call_nif);
 	    }
@@ -2211,10 +2485,21 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2)
 		GenericBp* g = (GenericBp *) code_ptr[1];
 		ASSERT(code_ptr[5+0] ==
 		       (BeamInstr) BeamOp(op_i_generic_breakpoint));
-	        g->orig_instr = (BeamInstr) BeamOp(op_call_nif);
-	    }	    
-	    code_ptr[5+1] = (BeamInstr) entry->funcs[i].fptr;
+		g->orig_instr = (BeamInstr) BeamOp(op_call_nif);
+	    }
+	    if ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7))
+		&& (entry->options & ERL_NIF_DIRTY_NIF_OPTION) && f->flags) {
+#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
+		code_ptr[5+3] = (BeamInstr) f->fptr;
+		code_ptr[5+1] = (f->flags == ERL_NIF_DIRTY_JOB_IO_BOUND) ?
+		    (BeamInstr) schedule_dirty_io_nif :
+		    (BeamInstr) schedule_dirty_cpu_nif;
+#endif
+	    }
+	    else
+		code_ptr[5+1] = (BeamInstr) f->fptr;
 	    code_ptr[5+2] = (BeamInstr) lib;
+	    f = next_func(entry, &incr, f);
 	}
     }
     else {
diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h
index 5b93c2398e..226fc199a1 100644
--- a/erts/emulator/beam/erl_nif.h
+++ b/erts/emulator/beam/erl_nif.h
@@ -42,9 +42,13 @@
 ** 2.5: R17 Maps API additions
 ** 2.6: R17 with maps
 **      R17 dirty schedulers
+** 2.7: 17.3 add enif_schedule_nif
+**           remove enif_schedule_dirty_nif, enif_schedule_dirty_nif_finalizer, enif_dirty_nif_finalizer
+**           add ErlNifEntry options
+**           add ErlNifFunc flags
 */
 #define ERL_NIF_MAJOR_VERSION 2
-#define ERL_NIF_MINOR_VERSION 6
+#define ERL_NIF_MINOR_VERSION 7
 
 /*
  * The emulator will refuse to load a nif-lib with a major version
@@ -125,8 +129,10 @@ typedef struct
     const char* name;
     unsigned arity;
     ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]);
+    unsigned flags;
 }ErlNifFunc;
 
+
 typedef struct enif_entry_t
 {
     int major;
@@ -139,8 +145,11 @@ typedef struct enif_entry_t
     int  (*upgrade)(ErlNifEnv*, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info);
     void (*unload) (ErlNifEnv*, void* priv_data);
     const char* vm_variant;
+    unsigned options;
 }ErlNifEntry;
 
+/* Field bits for ErlNifEntry options */
+#define ERL_NIF_DIRTY_NIF_OPTION 1
 
 
 typedef struct
@@ -232,10 +241,21 @@ extern TWinDynNifCallbacks WinDynNifCallbacks;
 #  else
 #    define ERL_NIF_INIT_DECL(MODNAME) __declspec(dllexport) ErlNifEntry* nif_init(TWinDynNifCallbacks* callbacks)
 #  endif
-#  define ERL_NIF_INIT_BODY memcpy(&WinDynNifCallbacks,callbacks,sizeof(TWinDynNifCallbacks))
+#  ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
+#    define ERL_NIF_INIT_BODY do {					\
+	memcpy(&WinDynNifCallbacks,callbacks,sizeof(TWinDynNifCallbacks)); \
+	entry.options = ERL_NIF_DIRTY_NIF_OPTION;			\
+     } while(0)
+#  else
+#    define ERL_NIF_INIT_BODY memcpy(&WinDynNifCallbacks,callbacks,sizeof(TWinDynNifCallbacks))
+#  endif
 #else 
 #  define ERL_NIF_INIT_GLOB
-#  define ERL_NIF_INIT_BODY
+#  ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
+#    define ERL_NIF_INIT_BODY entry.options = ERL_NIF_DIRTY_NIF_OPTION
+#  else
+#    define ERL_NIF_INIT_BODY
+#  endif
 #  ifdef STATIC_ERLANG_NIF
 #    define ERL_NIF_INIT_DECL(MODNAME)  ErlNifEntry* MODNAME ## _nif_init(void)
 #  else
diff --git a/erts/emulator/beam/erl_nif_api_funcs.h b/erts/emulator/beam/erl_nif_api_funcs.h
index d7c554e60b..be39816a64 100644
--- a/erts/emulator/beam/erl_nif_api_funcs.h
+++ b/erts/emulator/beam/erl_nif_api_funcs.h
@@ -141,10 +141,8 @@ ERL_NIF_API_FUNC_DECL(int,enif_is_number,(ErlNifEnv*, ERL_NIF_TERM term));
 ERL_NIF_API_FUNC_DECL(void*,enif_dlopen,(const char* lib, void (*err_handler)(void*,const char*), void* err_arg));
 ERL_NIF_API_FUNC_DECL(void*,enif_dlsym,(void* handle, const char* symbol, void (*err_handler)(void*,const char*), void* err_arg));
 ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent));
+ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_nif,(ErlNifEnv*,const char*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[]));
 #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
-ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif,(ErlNifEnv*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[]));
-ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM,ERL_NIF_TERM (*)(ErlNifEnv*,ERL_NIF_TERM)));
-ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM));
 ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*));
 ERL_NIF_API_FUNC_DECL(int,enif_have_dirty_schedulers,(void));
 #endif
@@ -289,10 +287,8 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa
 #  define enif_dlopen ERL_NIF_API_FUNC_MACRO(enif_dlopen)
 #  define enif_dlsym ERL_NIF_API_FUNC_MACRO(enif_dlsym)
 #  define enif_consume_timeslice ERL_NIF_API_FUNC_MACRO(enif_consume_timeslice)
+#  define enif_schedule_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_nif)
 #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
-#  define enif_schedule_dirty_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif)
-#  define enif_schedule_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif_finalizer)
-#  define enif_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_dirty_nif_finalizer)
 #  define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler)
 #  define enif_have_dirty_schedulers ERL_NIF_API_FUNC_MACRO(enif_have_dirty_schedulers)
 #endif
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 1606ad119d..e96d5c4ccd 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -590,12 +590,10 @@ erts_pre_init_process(void)
      erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks
 	 = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS;
 
-#ifdef ERTS_DIRTY_SCHEDULERS
-     erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].get_locks
-	 = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS;
-     erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].set_locks
-	 = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS;
-#endif
+     erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].get_locks
+	 = ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS;
+     erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].set_locks
+	 = ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS;
 
      /* Check that we have locks for all entries */
      for (ix = 0; ix < ERTS_PSD_SIZE; ix++) {
@@ -3758,17 +3756,25 @@ evacuate_run_queue(ErtsRunQueue *rq,
 	    }
 #ifdef ERTS_DIRTY_SCHEDULERS
 	    else if (state & ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q) {
-		erts_aint32_t old;
-		old = erts_smp_atomic32_read_band_nob(&proc->state,
-						      ~(ERTS_PSFLG_DIRTY_CPU_PROC
-							| ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q));
+#ifdef DEBUG
+		erts_aint32_t old =
+#else
+		(void)
+#endif
+		    erts_smp_atomic32_read_band_nob(&proc->state,
+						    ~(ERTS_PSFLG_DIRTY_CPU_PROC
+						      | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q));
 		/* assert that no other dirty flags are set */
 		ASSERT(!(old & (ERTS_PSFLG_DIRTY_IO_PROC|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)));
 	    } else if (state & ERTS_PSFLG_DIRTY_IO_PROC_IN_Q) {
-		erts_aint32_t old;
-		old = erts_smp_atomic32_read_band_nob(&proc->state,
-						      ~(ERTS_PSFLG_DIRTY_IO_PROC
-							| ERTS_PSFLG_DIRTY_IO_PROC_IN_Q));
+#ifdef DEBUG
+		erts_aint32_t old =
+#else
+		(void)
+#endif
+		    erts_smp_atomic32_read_band_nob(&proc->state,
+						    ~(ERTS_PSFLG_DIRTY_IO_PROC
+						      | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q));
 		/* assert that no other dirty flags are set */
 		ASSERT(!(old & (ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)));
 	    }
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index ed6dadbffa..31f4a09c94 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -734,13 +734,9 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi)
 #define ERTS_PSD_DIST_ENTRY			3
 #define ERTS_PSD_CALL_TIME_BP			4
 #define ERTS_PSD_DELAYED_GC_TASK_QS		5
-#ifdef ERTS_DIRTY_SCHEDULERS
-#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT	6
+#define ERTS_PSD_NIF_TRAP_EXPORT		6
 
 #define ERTS_PSD_SIZE				7
-#else
-#define ERTS_PSD_SIZE				6
-#endif
 
 typedef struct {
     void *data[ERTS_PSD_SIZE];
@@ -767,10 +763,8 @@ typedef struct {
 #define ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS ERTS_PROC_LOCK_MAIN
 #define ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS ERTS_PROC_LOCK_MAIN
 
-#ifdef ERTS_DIRTY_SCHEDULERS
-#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN
-#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN
-#endif
+#define ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN
+#define ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN
 
 typedef struct {
     ErtsProcLocks get_locks;
@@ -1367,6 +1361,8 @@ Uint64 erts_get_proc_interval(void);
 Uint64 erts_ensure_later_proc_interval(Uint64);
 Uint64 erts_step_proc_interval(void);
 
+int erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj); /* see erl_nif.c */
+
 ErtsProcList *erts_proclist_create(Process *);
 void erts_proclist_destroy(ErtsProcList *);
 
@@ -1817,12 +1813,10 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data)
 #define ERTS_PROC_SET_DELAYED_GC_TASK_QS(P, L, PBT) \
     ((ErtsProcSysTaskQs *) erts_psd_set((P), (L), ERTS_PSD_DELAYED_GC_TASK_QS, (void *) (PBT)))
 
-#ifdef ERTS_DIRTY_SCHEDULERS
-#define ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(P) \
-    ((Export *) erts_psd_get((P), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT))
-#define ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(P, L, DSTE) \
-    ((Export *) erts_psd_set((P), (L), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT, (void *) (DSTE)))
-#endif
+#define ERTS_PROC_GET_NIF_TRAP_EXPORT(P) \
+    ((Export *) erts_psd_get((P), ERTS_PSD_NIF_TRAP_EXPORT))
+#define ERTS_PROC_SET_NIF_TRAP_EXPORT(P, L, DSTE) \
+    ((Export *) erts_psd_set((P), (L), ERTS_PSD_NIF_TRAP_EXPORT, (void *) (DSTE)))
 
 
 ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p);
diff --git a/erts/emulator/test/nif_SUITE.erl b/erts/emulator/test/nif_SUITE.erl
index b2da6f58af..01ee71c4fa 100644
--- a/erts/emulator/test/nif_SUITE.erl
+++ b/erts/emulator/test/nif_SUITE.erl
@@ -37,7 +37,8 @@
 	 threading/1, send/1, send2/1, send3/1, send_threaded/1, neg/1, 
 	 is_checks/1,
 	 get_length/1, make_atom/1, make_string/1, reverse_list_test/1,
-	 otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1
+	 otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1,
+	 dirty_nif_exception/1, nif_schedule/1
 	]).
 
 -export([many_args_100/100]).
@@ -64,7 +65,8 @@ all() ->
      resource_takeover, threading, send, send2, send3,
      send_threaded, neg, is_checks, get_length, make_atom,
      make_string,reverse_list_test,
-     otp_9668, consume_timeslice, dirty_nif, dirty_nif_send
+     otp_9668, consume_timeslice,
+     nif_schedule, dirty_nif, dirty_nif_send, dirty_nif_exception
     ].
 
 groups() -> 
@@ -1524,6 +1526,20 @@ consume_timeslice(Config) when is_list(Config) ->
 
     ok.
 
+nif_schedule(Config) when is_list(Config) ->
+    ensure_lib_loaded(Config),
+    A = "this is a string",
+    B = {this,is,a,tuple},
+    {B,A} = call_nif_schedule(A, B),
+    ok = try call_nif_schedule(1, 2)
+	 catch
+	     error:badarg ->
+		 [{?MODULE,call_nif_schedule,[1,2],_}|_] =
+		     erlang:get_stacktrace(),
+		 ok
+	 end,
+    ok.
+
 dirty_nif(Config) when is_list(Config) ->
     try erlang:system_info(dirty_cpu_schedulers) of
 	N when is_integer(N) ->
@@ -1556,6 +1572,24 @@ dirty_nif_send(Config) when is_list(Config) ->
 	    {skipped,"No dirty scheduler support"}
     end.
 
+dirty_nif_exception(Config) when is_list(Config) ->
+    try erlang:system_info(dirty_cpu_schedulers) of
+	N when is_integer(N) ->
+	    ensure_lib_loaded(Config),
+	    try
+	        call_dirty_nif_exception(),
+	        ?t:fail(expected_badarg)
+	    catch
+	        error:badarg ->
+		    [{?MODULE,call_dirty_nif_exception,[],_}|_] =
+			erlang:get_stacktrace(),
+		    ok
+	    end
+    catch
+	error:badarg ->
+	    {skipped,"No dirty scheduler support"}
+    end.
+
 next_msg(_Pid) ->
     receive
 	M -> M
@@ -1685,8 +1719,10 @@ echo_int(_) -> ?nif_stub.
 type_sizes() -> ?nif_stub.
 otp_9668_nif(_) -> ?nif_stub.
 consume_timeslice_nif(_,_) -> ?nif_stub.
+call_nif_schedule(_,_) -> ?nif_stub.
 call_dirty_nif(_,_,_) -> ?nif_stub.
 send_from_dirty_nif(_) -> ?nif_stub.
+call_dirty_nif_exception() -> ?nif_stub.
 
 %% maps
 is_map_nif(_) -> ?nif_stub.
diff --git a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c
index 955dc64189..ad9d5d9254 100644
--- a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c
+++ b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c
@@ -1493,6 +1493,31 @@ static ERL_NIF_TERM consume_timeslice_nif(ErlNifEnv* env, int argc, const ERL_NI
     }
 }
 
+static ERL_NIF_TERM nif_sched2(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    char s[64];
+    if (!enif_get_string(env, argv[2], s, sizeof s, ERL_NIF_LATIN1))
+	return enif_make_badarg(env);
+    return enif_make_tuple2(env, argv[3], argv[2]);
+}
+
+static ERL_NIF_TERM nif_sched1(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    ERL_NIF_TERM new_argv[4];
+    new_argv[0] = enif_make_atom(env, "garbage0");
+    new_argv[1] = enif_make_atom(env, "garbage1");
+    new_argv[2] = argv[0];
+    new_argv[3] = argv[1];
+    return enif_schedule_nif(env, "nif_sched2", 0, nif_sched2, 4, new_argv);
+}
+
+static ERL_NIF_TERM call_nif_schedule(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+{
+    if (argc != 2)
+	return enif_make_atom(env, "false");
+    return enif_schedule_nif(env, "nif_sched1", 0, nif_sched1, argc, argv);
+}
+
 #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
 static ERL_NIF_TERM dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
 {
@@ -1507,11 +1532,10 @@ static ERL_NIF_TERM dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[
     enif_get_int(env, argv[0], &n);
     enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1);
     enif_inspect_binary(env, argv[2], &b);
-    result = enif_make_tuple3(env,
-			      enif_make_int(env, n),
-			      enif_make_string(env, s, ERL_NIF_LATIN1),
-			      enif_make_binary(env, &b));
-    return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer);
+    return enif_make_tuple3(env,
+			    enif_make_int(env, n),
+			    enif_make_string(env, s, ERL_NIF_LATIN1),
+			    enif_make_binary(env, &b));
 }
 
 static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
@@ -1526,7 +1550,7 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM
 	if (enif_get_int(env, argv[0], &n) &&
 	    enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1) &&
 	    enif_inspect_binary(env, argv[2], &b))
-	    return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv);
+	    return enif_schedule_nif(env, "call_dirty_nif", ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv);
 	else
 	    return enif_make_badarg(env);
     } else {
@@ -1534,35 +1558,42 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM
     }
 }
 
-static ERL_NIF_TERM dirty_sender(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
 {
     ERL_NIF_TERM result;
     ErlNifPid pid;
     ErlNifEnv* menv;
     int res;
 
-    enif_get_local_pid(env, argv[0], &pid);
+    if (!enif_get_local_pid(env, argv[0], &pid))
+	return enif_make_badarg(env);
     result = enif_make_tuple2(env, enif_make_atom(env, "ok"), enif_make_pid(env, &pid));
     menv = enif_alloc_env();
     res = enif_send(env, &pid, menv, result);
     enif_free_env(menv);
     if (!res)
-	/* Note the next line will crash, since dirty nifs can't return exceptions.
-	 * This is intentional, since enif_send should not fail if the test succeeds.
-	 */
-	return enif_schedule_dirty_nif_finalizer(env, enif_make_badarg(env), enif_dirty_nif_finalizer);
+	return enif_make_badarg(env);
     else
-	return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer);
+	return result;
 }
 
-static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
+static ERL_NIF_TERM call_dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[])
 {
-    ERL_NIF_TERM result;
-    ErlNifPid pid;
-
-    if (!enif_get_local_pid(env, argv[0], &pid))
+    switch (argc) {
+    case 0: {
+	ERL_NIF_TERM args[255];
+	int i;
+	for (i = 0; i < 255; i++)
+	    args[i] = enif_make_int(env, i);
+	return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND,
+				 call_dirty_nif_exception, 255, argv);
+    }
+    case 1:
 	return enif_make_badarg(env);
-    return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_sender, argc, argv);
+    default:
+	return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND,
+				 call_dirty_nif_exception, argc-1, argv);
+    }
 }
 #endif
 
@@ -1742,9 +1773,11 @@ static ErlNifFunc nif_funcs[] =
     {"type_sizes", 0, type_sizes},
     {"otp_9668_nif", 1, otp_9668_nif},
     {"consume_timeslice_nif", 2, consume_timeslice_nif},
+    {"call_nif_schedule", 2, call_nif_schedule},
 #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT
     {"call_dirty_nif", 3, call_dirty_nif},
-    {"send_from_dirty_nif", 1, send_from_dirty_nif},
+    {"send_from_dirty_nif", 1, send_from_dirty_nif, ERL_NIF_DIRTY_JOB_CPU_BOUND},
+    {"call_dirty_nif_exception", 0, call_dirty_nif_exception, ERL_NIF_DIRTY_JOB_IO_BOUND},
 #endif
     {"is_map_nif", 1, is_map_nif},
     {"get_map_size_nif", 1, get_map_size_nif},