[illumos-gate merge]

commit b39b008f8a57ea7ddfd0f69b24529deba7c25ae1
    12783 async unlinked drain races with ZFS unmount
commit 94bce860c16a04a3d8eaeaa18807723c026e23b5
    12559 zs: NULL pointer errors
commit fdf1a8710a94b4953ba782ed5bdc1549b0356ddc
    12804 sysevent.h: C++11 requires a space between string literal and macro
commit e86372a01d2d16a5dd4a64e144ed978ba17fe7dd
    12668 ZFS support for vectorized algorithms on x86 (initial support)
commit 82049ff560eed6fbdf4cf222d894467f5809f9b3
    12119 ndmpd: cast between incompatible function types
commit 8e3a263e496f0b896b62a8ebe101c0b2b56c9a15
    12750 bhyve should not hold HMA registration with no running VMs
commit d0b3c59ba652f183eeec1414dd9fbdc56bf05cc8
    12775 reorganize bhyve contrib headers
commit e213fe2d75a0e47cc0b3f67c473e24dfa9304e0b
    12545 ixgbe: IXGBE_LE32_TO_CPUS macro is missing assignment

166 files changed, 4865 insertions, 507 deletions

diff --git a/exception_lists/check_rtime b/exception_lists/check_rtime
index 3d5e7bf491..308d0e35d4 100644
--- a/exception_lists/check_rtime
+++ b/exception_lists/check_rtime
@@ -232,6 +232,7 @@ FORBIDDEN libfakekernel\.so
 FORBIDDEN_DEP usr/MACH(lib)/libzpool.so.1
 FORBIDDEN_DEP usr/bin/amd64/ztest
 FORBIDDEN_DEP usr/bin/i86/ztest
+FORBIDDEN_DEP usr/bin/raidz_test
 FORBIDDEN_DEP usr/bin/sparcv7/ztest
 FORBIDDEN_DEP usr/bin/sparcv9/ztest
 FORBIDDEN_DEP usr/lib/MACH(smbfs)/libfknsmb.so.1
diff --git a/exception_lists/copyright b/exception_lists/copyright
index 354f7fb236..45c53c793c 100644
--- a/exception_lists/copyright
+++ b/exception_lists/copyright
@@ -546,14 +546,9 @@ usr/src/cmd/bhyve/vga.[ch]
 usr/src/cmd/bhyve/virtio.[ch]
 usr/src/cmd/bhyve/xmsr.[ch]
 usr/src/cmd/bhyvectl/bhyvectl.c
-usr/src/compat/freebsd/*.h
-usr/src/compat/freebsd/*/*.h
-usr/src/compat/freebsd/amd64/machine/*.h
-usr/contrib/freebsd/*/*.h
-usr/contrib/freebsd/*/*/*.h
-usr/contrib/freebsd/lib/libutil/*.c
+usr/src/compat/bhyve/*
+usr/src/contrib/bhyve/*
 usr/src/lib/libvmmapi/common/vmmapi.[ch]
-usr/src/tools/scripts/gensetdefs.pl
 usr/src/uts/i86pc/io/vmm/amd/*.[chs]
 usr/src/uts/i86pc/io/vmm/intel/*.[chs]
 usr/src/uts/i86pc/io/vmm/intel/offsets.in
diff --git a/exception_lists/cstyle b/exception_lists/cstyle
index b7f0327ffe..35d79e7c69 100644
--- a/exception_lists/cstyle
+++ b/exception_lists/cstyle
@@ -1422,12 +1422,8 @@ usr/src/cmd/bhyve/virtio.[ch]
 usr/src/cmd/bhyve/xmsr.[ch]
 usr/src/cmd/bhyveconsole/bhyveconsole.c
 usr/src/cmd/bhyvectl/bhyvectl.c
-usr/src/compat/freebsd/*.h
-usr/src/compat/freebsd/*/*.h
-usr/src/compat/freebsd/amd64/machine/*.h
-usr/contrib/freebsd/*/*.h
-usr/contrib/freebsd/*/*/*.h
-usr/contrib/freebsd/lib/libutil/*.c
+usr/src/compat/bhyve/*
+usr/src/contrib/bhyve/*
 usr/src/lib/libvmmapi/common/vmmapi.[ch]
 usr/src/uts/i86pc/io/vmm/amd/*.[ch]
 usr/src/uts/i86pc/io/vmm/intel/*.[chs]
diff --git a/exception_lists/hdrchk b/exception_lists/hdrchk
index c6a5ffe4d3..e80a93515d 100644
--- a/exception_lists/hdrchk
+++ b/exception_lists/hdrchk
@@ -418,11 +418,8 @@ usr/src/cmd/bhyve/uart_emul.h
 usr/src/cmd/bhyve/vga.h
 usr/src/cmd/bhyve/virtio.h
 usr/src/cmd/bhyve/xmsr.h
-usr/src/compat/freebsd/*.h
-usr/src/compat/freebsd/*/*.h
-usr/src/compat/freebsd/amd64/machine/*.h
-usr/contrib/freebsd/*/*.h
-usr/contrib/freebsd/*/*/*.h
+usr/src/compat/bhyve/*
+usr/src/contrib/bhyve/*
 usr/src/lib/libvmmapi/common/vmmapi.h
 usr/src/uts/i86pc/io/vmm/intel/*.h
 usr/src/uts/i86pc/io/vmm/io/*.h
diff --git a/exception_lists/wscheck b/exception_lists/wscheck
index ea93389a01..cfba871041 100644
--- a/exception_lists/wscheck
+++ b/exception_lists/wscheck
@@ -176,12 +176,8 @@ usr/src/cmd/bhyve/virtio.[ch]
 usr/src/cmd/bhyve/xmsr.[ch]
 usr/src/cmd/bhyveconsole/bhyveconsole.c
 usr/src/cmd/bhyvectl/bhyvectl.c
-usr/src/compat/freebsd/*.h
-usr/src/compat/freebsd/*/*.h
-usr/src/compat/freebsd/amd64/machine/*.h
-usr/contrib/freebsd/*/*.h
-usr/contrib/freebsd/*/*/*.h
-usr/contrib/freebsd/lib/libutil/*.c
+usr/src/compat/bhyve/*
+usr/src/contrib/bhyve/*
 usr/src/lib/libvmmapi/common/vmmapi.[ch]
 usr/src/uts/i86pc/io/vmm/amd/*.[ch]
 usr/src/uts/i86pc/io/vmm/intel/*.[chs]
diff --git a/usr/src/Makefile.master b/usr/src/Makefile.master
index 7c036ac21a..7dc553209d 100644
--- a/usr/src/Makefile.master
+++ b/usr/src/Makefile.master
@@ -61,7 +61,7 @@ NATIVE_ADJUNCT=	/usr
 # Compatibility code for FreeBSD etc.
 #
 COMPAT=	$(SRC)/compat
-CONTRIB= $(SRC)/../contrib
+CONTRIB= $(SRC)/contrib
 
 #
 # RELEASE_BUILD should be cleared for final release builds.
diff --git a/usr/src/cmd/Makefile b/usr/src/cmd/Makefile
index dd035124e9..acb3717b0a 100644
--- a/usr/src/cmd/Makefile
+++ b/usr/src/cmd/Makefile
@@ -21,7 +21,7 @@
 
 #
 # Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright 2019 Joyent, Inc.
+# Copyright 2020 Joyent, Inc.
 # Copyright (c) 2012, 2015 by Delphix. All rights reserved.
 # Copyright (c) 2013 DEY Storage Systems, Inc. All rights reserved.
 # Copyright 2014 Garrett D'Amore <garrett@damore.org>
@@ -343,6 +343,7 @@ COMMON_SUBDIRS=		\
 	pwconv		\
 	pwd		\
 	raidctl		\
+	raidz_test	\
 	ramdiskadm	\
 	rcap		\
 	rcm_daemon	\
diff --git a/usr/src/cmd/bhyve/Makefile b/usr/src/cmd/bhyve/Makefile
index ea4537596a..b766cb8357 100644
--- a/usr/src/cmd/bhyve/Makefile
+++ b/usr/src/cmd/bhyve/Makefile
@@ -111,10 +111,10 @@ MEVENT_TEST_OBJS = $(MEVENT_TEST_SRCS:.c=.o)
 CLEANFILES =	$(PROG) $(ZHYVE_PROG) $(MEVENT_TEST_PROG) $(MEVENT_TEST_OBJS)
 
 CFLAGS +=	$(CCVERBOSE) -_gcc=-Wimplicit-function-declaration -_gcc=-Wno-parentheses
-CPPFLAGS =	-I$(COMPAT)/freebsd -I$(CONTRIB)/freebsd \
-		-I$(COMPAT)/freebsd/amd64 -I$(CONTRIB)/freebsd/amd64 \
-		-I$(CONTRIB)/freebsd/dev/usb/controller \
-		-I$(CONTRIB)/freebsd/dev/mii \
+CPPFLAGS =	-I$(COMPAT)/bhyve -I$(CONTRIB)/bhyve \
+		-I$(COMPAT)/bhyve/amd64 -I$(CONTRIB)/bhyve/amd64 \
+		-I$(CONTRIB)/bhyve/dev/usb/controller \
+		-I$(CONTRIB)/bhyve/dev/mii \
 		-I$(SRC)/uts/common/io/e1000api \
 		$(CPPFLAGS.master) \
 		-I$(ROOT)/usr/platform/i86pc/include \
diff --git a/usr/src/cmd/bhyve/test/Makefile.com b/usr/src/cmd/bhyve/test/Makefile.com
index 2bd67140f5..aeba956e12 100644
--- a/usr/src/cmd/bhyve/test/Makefile.com
+++ b/usr/src/cmd/bhyve/test/Makefile.com
@@ -28,9 +28,9 @@ CFLAGS +=	$(CCVERBOSE) -_gcc=-Wimplicit-function-declaration \
 CFLAGS64 +=	$(CCVERBOSE) -_gcc=-Wimplicit-function-declaration \
 		-_gcc=-Wno-parentheses
 CPPFLAGS =	-I$(SRC)/cmd/bhyve \
-		-I$(COMPAT)/freebsd -I$(CONTRIB)/freebsd \
-		-I$(CONTRIB)/freebsd/dev/usb/controller \
-		-I$(CONTRIB)/freebsd/dev/mii \
+		-I$(COMPAT)/bhyve -I$(CONTRIB)/bhyve \
+		-I$(CONTRIB)/bhyve/dev/usb/controller \
+		-I$(CONTRIB)/bhyve/dev/mii \
 		$(CPPFLAGS.master) \
 		-I$(ROOT)/usr/platform/i86pc/include \
 		-I$(SRC)/uts/i86pc/io/vmm \
@@ -38,7 +38,7 @@ CPPFLAGS =	-I$(SRC)/cmd/bhyve \
 		-I$(SRC)/uts/i86pc \
 		-I$(SRC)/lib/libdladm/common \
 		-DWITHOUT_CAPSICUM
-CPPFLAGS +=	-I$(COMPAT)/freebsd/amd64 -I$(CONTRIB)/freebsd/amd64
+CPPFLAGS +=	-I$(COMPAT)/bhyve/amd64 -I$(CONTRIB)/bhyve/amd64
 
 SMOFF += all_func_returns
 
diff --git a/usr/src/cmd/bhyvectl/Makefile b/usr/src/cmd/bhyvectl/Makefile
index 72b3cd913f..c35b590196 100644
--- a/usr/src/cmd/bhyvectl/Makefile
+++ b/usr/src/cmd/bhyvectl/Makefile
@@ -28,8 +28,8 @@ CLOBBERFILES +=	$(ROOTUSRSBINPROG)
 .KEEP_STATE:
 
 CFLAGS +=	$(CCVERBOSE)
-CPPFLAGS =	-I$(COMPAT)/freebsd -I$(CONTRIB)/freebsd \
-		-I$(COMPAT)/freebsd/amd64 -I$(CONTRIB)/freebsd/amd64 \
+CPPFLAGS =	-I$(COMPAT)/bhyve -I$(CONTRIB)/bhyve \
+		-I$(COMPAT)/bhyve/amd64 -I$(CONTRIB)/bhyve/amd64 \
 		$(CPPFLAGS.master) \
 		-I$(ROOT)/usr/platform/i86pc/include \
 		-I$(SRC)/uts/i86pc/io/vmm
@@ -53,6 +53,6 @@ clean:
 
 include ../Makefile.targ
 
-%.o: $(CONTRIB)/freebsd/lib/libutil/%.c
+%.o: $(CONTRIB)/bhyve/lib/libutil/%.c
 	$(COMPILE.c) -o $@ $<
 	$(POST_PROCESS_O)
diff --git a/usr/src/cmd/ndmpd/include/tlm.h b/usr/src/cmd/ndmpd/include/tlm.h
index f382d2eca3..8c667a334b 100644
--- a/usr/src/cmd/ndmpd/include/tlm.h
+++ b/usr/src/cmd/ndmpd/include/tlm.h
@@ -11,10 +11,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -113,9 +113,9 @@ typedef struct fs_fhandle {
 } fs_fhandle_t;
 
 typedef struct scsi_link {
-	struct scsi_link 	*sl_next;
-	struct scsi_link 	*sl_prev;
-	struct scsi_adapter 	*sl_sa;
+	struct scsi_link	*sl_next;
+	struct scsi_link	*sl_prev;
+	struct scsi_adapter	*sl_sa;
 	unsigned int		sl_sid;
 	unsigned int		sl_lun;
 	unsigned int		sl_requested_max_active;
@@ -427,7 +427,7 @@ typedef struct tm_ops {
 	int (*tm_putfile)();
 	int (*tm_putdir)();
 	int (*tm_putvol)();	/* Reserved */
-	int (*tm_getfile)();
+	void * (*tm_getfile)(void *);
 	int (*tm_getdir)();
 	int (*tm_getvol)();	/* Reserved */
 } tm_ops_t;
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd.h b/usr/src/cmd/ndmpd/ndmp/ndmpd.h
index 4f9fcd9a09..929a833c1c 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd.h
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd.h
@@ -11,10 +11,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -998,7 +998,7 @@ extern int ndmp_backup_extract_params(ndmpd_session_t *,
     ndmpd_module_params_t *);
 extern int ndmp_restore_extract_params(ndmpd_session_t *,
     ndmpd_module_params_t *);
-extern int ndmp_tar_reader(ndmp_tar_reader_arg_t *);
+extern void *ndmp_tar_reader(void *);
 
 extern int tape_open(char *, int);
 extern int tape_is_at_bot(ndmpd_session_t *);
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_common.h b/usr/src/cmd/ndmpd/ndmp/ndmpd_common.h
index 846f64e66f..383d5a673c 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_common.h
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_common.h
@@ -10,10 +10,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -281,6 +281,6 @@ extern int ndmp_log_msg_id;
 /*
  * Module function prototypes.
  */
-typedef int module_start_func_t(void *);
+typedef void *module_start_func_t(void *);
 typedef int module_abort_func_t(void *);
 #endif	/* _NDMP_COMMON_H */
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_data.c b/usr/src/cmd/ndmpd/ndmp/ndmpd_data.c
index 9350cb05f7..2713f8bec2 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_data.c
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_data.c
@@ -11,10 +11,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -1466,7 +1466,7 @@ ndmpd_tar_start_backup_v3(ndmpd_session_t *session, char *bu_type,
 	 * client request here.
 	 */
 	err = pthread_create(NULL, NULL,
-	    (funct_t)session->ns_data.dd_module.dm_start_func,
+	    session->ns_data.dd_module.dm_start_func,
 	    params);
 	if (err != 0) {
 		NDMP_LOG(LOG_ERR, "Can't start backup session.");
@@ -1595,7 +1595,7 @@ ndmpd_tar_start_recover_v3(ndmpd_session_t *session,
 	 * client request here.
 	 */
 	err = pthread_create(NULL, NULL,
-	    (funct_t)session->ns_data.dd_module.dm_start_func,
+	    session->ns_data.dd_module.dm_start_func,
 	    params);
 
 	if (err != 0) {
@@ -1692,7 +1692,8 @@ ndmpd_zfs_start_op(ndmpd_session_t *session, ndmp_pval *env_val,
 	}
 
 	err = pthread_create(&tid, NULL,
-	    (funct_t)session->ns_data.dd_module.dm_start_func, ndmpd_zfs_args);
+	    session->ns_data.dd_module.dm_start_func,
+	    ndmpd_zfs_args);
 
 	if (err) {
 		NDMP_LOG(LOG_ERR, "Can't start %s session (errno %d)",
@@ -2185,7 +2186,7 @@ ndmpd_tar_start_backup_v2(ndmpd_session_t *session, char *bu_type,
 	 * client request here.
 	 */
 	(void) pthread_create(NULL, NULL,
-	    (funct_t)session->ns_data.dd_module.dm_start_func,
+	    session->ns_data.dd_module.dm_start_func,
 	    params);
 
 	return (NDMP_NO_ERR);
@@ -2322,7 +2323,7 @@ ndmpd_tar_start_recover_v2(ndmpd_session_t *session, char *bu_type,
 	 * client request here.
 	 */
 	(void) pthread_create(NULL, NULL,
-	    (funct_t)session->ns_data.dd_module.dm_start_func,
+	    session->ns_data.dd_module.dm_start_func,
 	    params);
 
 	return (NDMP_NO_ERR);
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_mover.c b/usr/src/cmd/ndmpd/ndmp/ndmpd_mover.c
index 2f7f311e75..8428679e93 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_mover.c
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_mover.c
@@ -10,10 +10,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -2503,8 +2503,8 @@ mover_tape_read_one_buf(ndmpd_session_t *session, tlm_buffer_t *buf)
  *   0: on success
  *  -1: otherwise
  */
-int
-mover_tape_reader(ndmpd_session_t *session)
+void *
+mover_tape_reader(void *ptr)
 {
 	int bidx;	/* buffer index */
 	int rv;
@@ -2513,10 +2513,11 @@ mover_tape_reader(ndmpd_session_t *session)
 	tlm_buffers_t *bufs;
 	tlm_cmd_t *lcmd;	/* Local command */
 	tlm_commands_t *cmds;	/* Commands structure */
+	ndmpd_session_t *session = ptr;
 
 	if ((nlp = ndmp_get_nlp(session)) == NULL) {
 		NDMP_LOG(LOG_DEBUG, "nlp == NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	cmds = &nlp->nlp_cmds;
@@ -2582,7 +2583,7 @@ mover_tape_reader(ndmpd_session_t *session)
 	cmds->tcs_reader_count--;
 	lcmd->tc_ref--;
 	lcmd->tc_writer = TLM_STOP;
-	return (0);
+	return (NULL);
 }
 
 
@@ -2652,8 +2653,8 @@ mover_socket_write_one_buf(ndmpd_session_t *session, tlm_buffer_t *buf)
  *   0: on success
  *  -1: otherwise
  */
-int
-mover_socket_writer(ndmpd_session_t *session)
+void *
+mover_socket_writer(void *ptr)
 {
 	int bidx;	/* buffer index */
 	ndmp_lbr_params_t *nlp;
@@ -2661,10 +2662,11 @@ mover_socket_writer(ndmpd_session_t *session)
 	tlm_buffers_t *bufs;
 	tlm_cmd_t *lcmd;	/* Local command */
 	tlm_commands_t *cmds;	/* Commands structure */
+	ndmpd_session_t *session = ptr;
 
 	if ((nlp = ndmp_get_nlp(session)) == NULL) {
 		NDMP_LOG(LOG_DEBUG, "nlp == NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	cmds = &nlp->nlp_cmds;
@@ -2725,7 +2727,7 @@ mover_socket_writer(ndmpd_session_t *session)
 	cmds->tcs_writer_count--;
 	lcmd->tc_ref--;
 	lcmd->tc_reader = TLM_STOP;
-	return (0);
+	return (NULL);
 }
 
 
@@ -2772,7 +2774,7 @@ start_mover_for_restore(ndmpd_session_t *session)
 	 * must be sent to the client before probable errors are sent
 	 * to the client.
 	 */
-	rc = pthread_create(NULL, NULL, (funct_t)mover_tape_reader, session);
+	rc = pthread_create(NULL, NULL, mover_tape_reader, session);
 	if (rc == 0) {
 		tlm_cmd_wait(cmds->tcs_command, TLM_TAPE_READER);
 	} else {
@@ -2781,7 +2783,7 @@ start_mover_for_restore(ndmpd_session_t *session)
 		return (-1);
 	}
 
-	rc = pthread_create(NULL, NULL, (funct_t)mover_socket_writer, session);
+	rc = pthread_create(NULL, NULL, mover_socket_writer, session);
 	if (rc == 0) {
 		tlm_cmd_wait(cmds->tcs_command, TLM_SOCK_WRITER);
 	} else {
@@ -2874,8 +2876,8 @@ mover_socket_read_one_buf(ndmpd_session_t *session, tlm_buffer_t *buf,
  *   0: on success
  *  -1: otherwise
  */
-int
-mover_socket_reader(ndmpd_session_t *session)
+void *
+mover_socket_reader(void *ptr)
 {
 	int bidx;	/* buffer index */
 	ndmp_lbr_params_t *nlp;
@@ -2883,11 +2885,12 @@ mover_socket_reader(ndmpd_session_t *session)
 	tlm_buffers_t *bufs;
 	tlm_cmd_t *lcmd;	/* Local command */
 	tlm_commands_t *cmds;	/* Commands structure */
+	ndmpd_session_t *session = ptr;
 	static int nr = 0;
 
 	if ((nlp = ndmp_get_nlp(session)) == NULL) {
 		NDMP_LOG(LOG_DEBUG, "nlp == NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	cmds = &nlp->nlp_cmds;
@@ -2951,7 +2954,7 @@ mover_socket_reader(ndmpd_session_t *session)
 	cmds->tcs_reader_count--;
 	lcmd->tc_ref--;
 	lcmd->tc_writer = TLM_STOP;
-	return (0);
+	return (NULL);
 }
 
 
@@ -3014,8 +3017,8 @@ mover_tape_write_one_buf(ndmpd_session_t *session, tlm_buffer_t *buf)
  *   0: on success
  *  -1: otherwise
  */
-int
-mover_tape_writer(ndmpd_session_t *session)
+void *
+mover_tape_writer(void *ptr)
 {
 	int bidx;
 	ndmp_lbr_params_t *nlp;
@@ -3023,11 +3026,12 @@ mover_tape_writer(ndmpd_session_t *session)
 	tlm_buffers_t *bufs;
 	tlm_cmd_t *lcmd;
 	tlm_commands_t *cmds;
+	ndmpd_session_t *session = ptr;
 	static int nw = 0;
 
 	if ((nlp = ndmp_get_nlp(session)) == NULL) {
 		NDMP_LOG(LOG_DEBUG, "nlp == NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	cmds = &nlp->nlp_cmds;
@@ -3096,7 +3100,7 @@ mover_tape_writer(ndmpd_session_t *session)
 	cmds->tcs_writer_count--;
 	lcmd->tc_ref--;
 	lcmd->tc_reader = TLM_STOP;
-	return (0);
+	return (NULL);
 }
 
 
@@ -3143,7 +3147,7 @@ start_mover_for_backup(ndmpd_session_t *session)
 	 * must be sent to the client before probable errors are sent
 	 * to the client.
 	 */
-	rc = pthread_create(NULL, NULL, (funct_t)mover_socket_reader, session);
+	rc = pthread_create(NULL, NULL, mover_socket_reader, session);
 	if (rc == 0) {
 		tlm_cmd_wait(cmds->tcs_command, TLM_SOCK_READER);
 	} else {
@@ -3152,7 +3156,7 @@ start_mover_for_backup(ndmpd_session_t *session)
 		return (-1);
 	}
 
-	rc = pthread_create(NULL, NULL, (funct_t)mover_tape_writer, session);
+	rc = pthread_create(NULL, NULL, mover_tape_writer, session);
 	if (rc == 0) {
 		tlm_cmd_wait(cmds->tcs_command, TLM_TAPE_WRITER);
 	} else {
@@ -3180,7 +3184,7 @@ start_mover_for_backup(ndmpd_session_t *session)
  *	Note: non-zero is also returned if the backup type is
  *		neither TAR nor DUMP.  I.e. the is_writer_running()
  *		check does not apply in this case and things should
- * 		appear successful.
+ *		appear successful.
  */
 static boolean_t
 is_writer_running(ndmpd_session_t *session)
@@ -3216,7 +3220,7 @@ is_writer_running(ndmpd_session_t *session)
  *	Note: non-zero is also returned if the backup type is
  *		neither TAR nor DUMP.  I.e. the is_writer_running()
  *		check does not apply in this case and things should
- * 		appear successful.
+ *		appear successful.
  */
 static boolean_t
 is_writer_running_v3(ndmpd_session_t *session)
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_tar.c b/usr/src/cmd/ndmpd/ndmp/ndmpd_tar.c
index 8040355eb0..89dc343051 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_tar.c
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_tar.c
@@ -10,10 +10,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -370,8 +370,7 @@ ndmp_write_utf8magic(tlm_cmd_t *cmd)
  *
  */
 static boolean_t
-timecmp(bk_selector_t *bksp,
-		struct stat64 *attr)
+timecmp(bk_selector_t *bksp, struct stat64 *attr)
 {
 	ndmp_lbr_params_t *nlp;
 
@@ -476,7 +475,7 @@ get_dir_acl_info(char *dir, tlm_acls_t *tlm_acls, tlm_job_stats_t *js)
 	char	*spot;
 	char	*fil;
 	acl_t	*aclp = NULL;
-	char 	*acltp;
+	char	*acltp;
 
 	checkpointed_dir = ndmp_malloc(TLM_MAX_PATH_NAME);
 	if (checkpointed_dir == NULL)
@@ -656,7 +655,7 @@ backup_work(char *bk_path, tlm_job_stats_t *job_stats,
 	longlong_t fsize;
 	bk_selector_t bks;
 	tlm_cmd_t *local_commands;
-	long 	dpos;
+	long	dpos;
 
 	NDMP_LOG(LOG_DEBUG, "nr_chkpnted %d nr_ldate: %u bk_path: \"%s\"",
 	    NLP_ISCHKPNTED(nlp), nlp->nlp_ldate, bk_path);
@@ -1067,8 +1066,8 @@ read_one_buf(ndmpd_module_params_t *mod_params, tlm_buffers_t *bufs,
  * file from the tape and wakes up the consumer thread to extract
  * it on the disk
  */
-int
-ndmp_tar_reader(ndmp_tar_reader_arg_t *argp)
+void *
+ndmp_tar_reader(void *ptr)
 {
 	int bidx;
 	int err;
@@ -1078,9 +1077,10 @@ ndmp_tar_reader(ndmp_tar_reader_arg_t *argp)
 	ndmpd_session_t *session;
 	ndmpd_module_params_t *mod_params;
 	tlm_commands_t *cmds;
+	ndmp_tar_reader_arg_t *argp = ptr;
 
 	if (!argp)
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 
 	session = argp->tr_session;
 	mod_params = argp->tr_mod_params;
@@ -1097,7 +1097,7 @@ ndmp_tar_reader(ndmp_tar_reader_arg_t *argp)
 
 	if (err != 0) {
 		tlm_cmd_signal(cmds->tcs_command, TLM_TAR_READER);
-		return (err);
+		return ((void *)(uintptr_t)err);
 	}
 
 	lcmd = cmds->tcs_command;
@@ -1160,7 +1160,7 @@ ndmp_tar_reader(ndmp_tar_reader_arg_t *argp)
 	 */
 	cmds->tcs_reader_count--;
 	lcmd->tc_ref--;
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 
@@ -1314,8 +1314,7 @@ ndmpd_tar_restore(ndmpd_session_t *session, ndmpd_module_params_t *mod_params,
 		arg.tr_mod_params = mod_params;
 		arg.tr_cmds = cmds;
 
-		err = pthread_create(&rdtp, NULL, (funct_t)ndmp_tar_reader,
-		    (void *)&arg);
+		err = pthread_create(&rdtp, NULL, ndmp_tar_reader, &arg);
 		if (err == 0) {
 			tlm_cmd_wait(cmds->tcs_command, TLM_TAR_READER);
 		} else {
@@ -1346,8 +1345,8 @@ ndmpd_tar_restore(ndmpd_session_t *session, ndmpd_module_params_t *mod_params,
 
 
 		if (tm_tar_ops.tm_getfile != NULL) {
-			err = pthread_create(&wrtp, NULL,
-			    (funct_t)tm_tar_ops.tm_getfile, (void *)&tlm_arg);
+			err = pthread_create(&wrtp, NULL, tm_tar_ops.tm_getfile,
+			    &tlm_arg);
 		} else {
 			(void) pthread_barrier_destroy(&tlm_arg.ba_barrier);
 			NDMP_LOG(LOG_DEBUG,
@@ -1859,7 +1858,7 @@ ndmp_restore_extract_params(ndmpd_session_t *session,
  * and calls ndmp_tar_backup to perform the actual backup. It does the cleanup
  * and release the snapshot at the end.
  */
-int
+void *
 ndmpd_tar_backup_starter(void *arg)
 {
 	ndmpd_module_params_t *mod_params = arg;
@@ -1933,7 +1932,7 @@ ndmpd_tar_backup_starter(void *arg)
 
 	NS_DEC(nbk);
 	ndmp_session_unref(session);
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 
@@ -1966,7 +1965,7 @@ ndmpd_tar_backup_abort(void *module_cookie)
  * Starts the restore by running ndmpd_tar_restore function (V2 only)
  */
 
-int
+void *
 ndmpd_tar_restore_starter(void *arg)
 {
 	ndmpd_module_params_t *mod_params = arg;
@@ -1986,7 +1985,7 @@ ndmpd_tar_restore_starter(void *arg)
 
 	NS_DEC(nrs);
 	ndmp_session_unref(session);
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_tar3.c b/usr/src/cmd/ndmpd/ndmp/ndmpd_tar3.c
index d3e89d5a3d..001368ac44 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_tar3.c
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_tar3.c
@@ -10,10 +10,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -2222,8 +2222,8 @@ lbrbk_v3(void *arg, fst_node_t *pnp, fst_node_t *enp)
  *   0: on success
  *   != 0: otherwise
  */
-static int
-backup_reader_v3(backup_reader_arg_t *argp)
+static void *
+backup_reader_v3(void *ptr)
 {
 	int rv;
 	tlm_cmd_t *lcmd;
@@ -2234,9 +2234,10 @@ backup_reader_v3(backup_reader_arg_t *argp)
 	char *jname;
 	ndmp_lbr_params_t *nlp;
 	tlm_commands_t *cmds;
+	backup_reader_arg_t *argp = ptr;
 
 	if (!argp)
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 
 	jname = argp->br_jname;
 	nlp = argp->br_nlp;
@@ -2265,7 +2266,7 @@ backup_reader_v3(backup_reader_arg_t *argp)
 
 	bp.bp_tmp = ndmp_malloc(sizeof (char) * TLM_MAX_PATH_NAME);
 	if (!bp.bp_tmp)
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 
 	/*
 	 * Make the checkpointed paths for traversing the
@@ -2277,7 +2278,7 @@ backup_reader_v3(backup_reader_arg_t *argp)
 		bp.bp_chkpnm = ndmp_malloc(sizeof (char) * TLM_MAX_PATH_NAME);
 		if (!bp.bp_chkpnm) {
 			NDMP_FREE(bp.bp_tmp);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 		(void) tlm_build_snapshot_name(nlp->nlp_backup_path,
 		    bp.bp_chkpnm, nlp->nlp_jstat->js_job_name);
@@ -2341,8 +2342,7 @@ backup_reader_v3(backup_reader_arg_t *argp)
 	lcmd->tc_writer = TLM_STOP;
 	tlm_release_reader_writer_ipc(lcmd);
 	tlm_un_ref_job_stats(jname);
-	return (rv);
-
+	return ((void *)(uintptr_t)rv);
 }
 
 
@@ -2438,8 +2438,7 @@ tar_backup_v3(ndmpd_session_t *session, ndmpd_module_params_t *params,
 
 		(void) pthread_barrier_init(&arg.br_barrier, 0, 2);
 
-		err = pthread_create(&rdtp, NULL, (funct_t)backup_reader_v3,
-		    (void *)&arg);
+		err = pthread_create(&rdtp, NULL, backup_reader_v3, &arg);
 		if (err == 0) {
 			(void) pthread_barrier_wait(&arg.br_barrier);
 			(void) pthread_barrier_destroy(&arg.br_barrier);
@@ -2512,9 +2511,10 @@ backup_out:
  * Find the estimate of backup size. This is used to get an estimate
  * of the progress of backup during NDMP backup.
  */
-void
-get_backup_size(ndmp_bkup_size_arg_t *sarg)
+void *
+get_backup_size(void *ptr)
 {
+	ndmp_bkup_size_arg_t *sarg = ptr;
 	fs_traverse_t ft;
 	u_longlong_t bk_size;
 	char spath[PATH_MAX];
@@ -2543,6 +2543,7 @@ get_backup_size(ndmp_bkup_size_arg_t *sarg)
 		    bk_size, bk_size / 1024, bk_size /(1024 * 1024));
 	}
 	sarg->bs_session->ns_data.dd_data_size = bk_size;
+	return (NULL);
 }
 
 /*
@@ -3186,8 +3187,7 @@ ndmpd_dar_tar_v3(ndmpd_session_t *session, ndmpd_module_params_t *params,
 		arg.tr_mod_params = params;
 		arg.tr_cmds = cmds;
 
-		err = pthread_create(&rdtp, NULL, (funct_t)ndmp_tar_reader,
-		    (void *)&arg);
+		err = pthread_create(&rdtp, NULL, ndmp_tar_reader, &arg);
 		if (err == 0) {
 			tlm_cmd_wait(cmds->tcs_command, TLM_TAR_READER);
 		} else {
@@ -3445,8 +3445,8 @@ ndmp_plugin_pre_restore(ndmp_context_t *ctxp, ndmpd_module_params_t *params,
  * /link-to-backup-path -> /backup/path
  *
  * Returns:
- * 	Pointer to the new path (allocated)
- * 	NULL if the path doesnt exist
+ *	Pointer to the new path (allocated)
+ *	NULL if the path doesnt exist
  */
 static char *
 get_absolute_path(const char *bkpath)
@@ -3548,8 +3548,7 @@ ndmpd_rs_sar_tar_v3(ndmpd_session_t *session, ndmpd_module_params_t *params,
 		arg.tr_session = session;
 		arg.tr_mod_params = params;
 		arg.tr_cmds = cmds;
-		err = pthread_create(&rdtp, NULL, (funct_t)ndmp_tar_reader,
-		    (void *)&arg);
+		err = pthread_create(&rdtp, NULL, ndmp_tar_reader, &arg);
 		if (err == 0) {
 			tlm_cmd_wait(cmds->tcs_command, TLM_TAR_READER);
 		} else {
@@ -3746,7 +3745,7 @@ ndmp_backup_get_params_v3(ndmpd_session_t *session,
  *   0: on success
  *   != 0: otherwise
  */
-int
+void *
 ndmpd_tar_backup_starter_v3(void *arg)
 {
 	ndmpd_module_params_t *params = arg;
@@ -3780,8 +3779,7 @@ ndmpd_tar_backup_starter_v3(void *arg)
 		sarg.bs_path = nlp->nlp_backup_path;
 
 		/* Get an estimate of the data size */
-		if (pthread_create(&tid, NULL, (funct_t)get_backup_size,
-		    (void *)&sarg) == 0)
+		if (pthread_create(&tid, NULL, get_backup_size, &sarg) == 0)
 			(void) pthread_detach(tid);
 
 		err = ndmp_get_cur_bk_time(nlp, &nlp->nlp_cdate, jname);
@@ -3810,8 +3808,7 @@ ndmpd_tar_backup_starter_v3(void *arg)
 
 	NS_DEC(nbk);
 	ndmp_session_unref(session);
-	return (err);
-
+	return ((void *)(uintptr_t)err);
 }
 
 
@@ -3924,7 +3921,7 @@ ndmp_restore_get_params_v3(ndmpd_session_t *session,
  *   NDMP_NO_ERR: on success
  *   != NDMP_NO_ERR: otherwise
  */
-int
+void *
 ndmpd_tar_restore_starter_v3(void *arg)
 {
 	ndmpd_module_params_t *params = arg;
@@ -3948,8 +3945,7 @@ ndmpd_tar_restore_starter_v3(void *arg)
 	/* nlp_params is allocated in start_recover() */
 	NDMP_FREE(nlp->nlp_params);
 	ndmp_session_unref(session);
-	return (err);
-
+	return ((void *)(uintptr_t)err);
 }
 
 /*
diff --git a/usr/src/cmd/ndmpd/ndmp/ndmpd_zfs.c b/usr/src/cmd/ndmpd/ndmp/ndmpd_zfs.c
index c124794fc4..72708f0475 100644
--- a/usr/src/cmd/ndmpd/ndmp/ndmpd_zfs.c
+++ b/usr/src/cmd/ndmpd/ndmp/ndmpd_zfs.c
@@ -11,10 +11,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -74,12 +74,12 @@ static void ndmpd_zfs_close_one_fd(ndmpd_zfs_args_t *, int);
 static int ndmpd_zfs_header_write(ndmpd_session_t *);
 static int ndmpd_zfs_header_read(ndmpd_zfs_args_t *);
 
-static int ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *);
-static int ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *);
+static void *ndmpd_zfs_backup_send_read(void *);
+static void *ndmpd_zfs_backup_tape_write(void *);
 
 static int ndmpd_zfs_restore(ndmpd_zfs_args_t *);
-static int ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *);
-static int ndmpd_zfs_restore_recv_write(ndmpd_zfs_args_t *);
+static void *ndmpd_zfs_restore_tape_read(void *);
+static void *ndmpd_zfs_restore_recv_write(void *);
 
 static int ndmpd_zfs_reader_writer(ndmpd_zfs_args_t *, int **, int **);
 
@@ -143,8 +143,8 @@ static int ndmpd_zfs_backup(ndmpd_zfs_args_t *);
  *
  * Examples:
  *
- * 	0.0.n/0.bob.p
- * 	0.0.u/1.bob.p/0.jane.d
+ *	0.0.n/0.bob.p
+ *	0.0.u/1.bob.p/0.jane.d
  *
  * Note: NDMPD_ZFS_SUBPROP_MAX is calculated based on ZFS_MAXPROPLEN
  */
@@ -164,8 +164,8 @@ static int ndmpd_zfs_backup(ndmpd_zfs_args_t *);
 #define	NDMPD_ZFS_LOG_ZERR(ndmpd_zfs_args, ...) {			\
 	NDMP_LOG(LOG_ERR, __VA_ARGS__);					\
 	NDMP_LOG(LOG_ERR, "%s--%s",					\
-	    libzfs_error_action((ndmpd_zfs_args)->nz_zlibh),           	\
-	    libzfs_error_description((ndmpd_zfs_args)->nz_zlibh));     	\
+	    libzfs_error_action((ndmpd_zfs_args)->nz_zlibh),		\
+	    libzfs_error_description((ndmpd_zfs_args)->nz_zlibh));	\
 	ndmpd_zfs_zerr_dma_log((ndmpd_zfs_args));			\
 }
 
@@ -459,7 +459,7 @@ _err:
 	return (-1);
 }
 
-int
+void *
 ndmpd_zfs_backup_starter(void *arg)
 {
 	ndmpd_zfs_args_t *ndmpd_zfs_args = arg;
@@ -492,7 +492,7 @@ _done:
 	ndmp_session_unref(session);
 	ndmpd_zfs_fini(ndmpd_zfs_args);
 
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 static int
@@ -599,17 +599,18 @@ ndmpd_zfs_backup(ndmpd_zfs_args_t *ndmpd_zfs_args)
  *
  * This routine executes zfs_send() to create the backup data stream.
  * The value of ZFS_MODE determines the type of zfs_send():
- * 	dataset ('d'): Only the dataset specified (i.e., top level) is backed up
- * 	recursive ('r'): The dataset and its child file systems are backed up
- * 	package ('p'): Same as 'r', except all intermediate snapshots are also
+ *	dataset ('d'): Only the dataset specified (i.e., top level) is backed up
+ *	recursive ('r'): The dataset and its child file systems are backed up
+ *	package ('p'): Same as 'r', except all intermediate snapshots are also
  *			backed up
  *
  * Volumes do not have descednants, so 'd' and 'r' produce equivalent results.
  */
 
-static int
-ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
+static void *
+ndmpd_zfs_backup_send_read(void *ptr)
 {
+	ndmpd_zfs_args_t *ndmpd_zfs_args = ptr;
 	ndmpd_session_t *session = (ndmpd_session_t *)
 	    (ndmpd_zfs_params->mp_daemon_cookie);
 	sendflags_t flags = { 0 };
@@ -623,7 +624,7 @@ ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	if (!zhp) {
 		if (!session->ns_data.dd_abort)
 			NDMPD_ZFS_LOG_ZERR(ndmpd_zfs_args, "zfs_open");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	switch (ndmpd_zfs_args->nz_zfs_mode) {
@@ -641,14 +642,14 @@ ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 		NDMP_LOG(LOG_ERR, "unknown zfs_mode: %c",
 		    ndmpd_zfs_args->nz_zfs_mode);
 		zfs_close(zhp);
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	if (ndmpd_zfs_is_incremental(ndmpd_zfs_args)) {
 		if (ndmpd_zfs_args->nz_fromsnap[0] == '\0') {
 			NDMP_LOG(LOG_ERR, "no fromsnap");
 			zfs_close(zhp);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 		fromsnap = ndmpd_zfs_args->nz_fromsnap;
 	}
@@ -661,7 +662,7 @@ ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 
 	zfs_close(zhp);
 
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 /*
@@ -672,9 +673,10 @@ ndmpd_zfs_backup_send_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
  * ndmpd_zfs_args->nz_bufsize).
  */
 
-static int
-ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
+static void *
+ndmpd_zfs_backup_tape_write(void *ptr)
 {
+	ndmpd_zfs_args_t *ndmpd_zfs_args = ptr;
 	ndmpd_session_t *session = (ndmpd_session_t *)
 	    (ndmpd_zfs_params->mp_daemon_cookie);
 	int bufsize = ndmpd_zfs_args->nz_bufsize;
@@ -685,7 +687,7 @@ ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	buf = ndmp_malloc(bufsize);
 	if (buf == NULL) {
 		NDMP_LOG(LOG_DEBUG, "buf NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	bytes_totalp =
@@ -704,7 +706,8 @@ ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
 			    *bytes_totalp - bufsize, *bytes_totalp);
 			free(buf);
 
-			return (ndmpd_zfs_addenv_backup_size(ndmpd_zfs_args,
+			return ((void *)(uintptr_t)
+			    ndmpd_zfs_addenv_backup_size(ndmpd_zfs_args,
 			    *bytes_totalp));
 		}
 
@@ -712,7 +715,7 @@ ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
 			NDMP_LOG(LOG_DEBUG, "pipe read error (errno %d)",
 			    errno);
 			free(buf);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 		NS_ADD(rdisk, count);
 
@@ -720,7 +723,7 @@ ndmpd_zfs_backup_tape_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
 			(void) ndmpd_zfs_abort((void *) ndmpd_zfs_args);
 			NDMP_LOG(LOG_ERR, "MOD_WRITE error");
 			free(buf);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 
 		*bytes_totalp += count;
@@ -750,7 +753,7 @@ ndmpd_zfs_addenv_backup_size(ndmpd_zfs_args_t *ndmpd_zfs_args,
 	return (0);
 }
 
-int
+void *
 ndmpd_zfs_restore_starter(void *arg)
 {
 	ndmpd_zfs_args_t *ndmpd_zfs_args = arg;
@@ -770,7 +773,7 @@ ndmpd_zfs_restore_starter(void *arg)
 
 	ndmpd_zfs_fini(ndmpd_zfs_args);
 
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 static int
@@ -827,9 +830,10 @@ ndmpd_zfs_restore(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	return (err);
 }
 
-static int
-ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
+static void *
+ndmpd_zfs_restore_tape_read(void *ptr)
 {
+	ndmpd_zfs_args_t *ndmpd_zfs_args = ptr;
 	ndmpd_session_t *session = (ndmpd_session_t *)
 	    (ndmpd_zfs_params->mp_daemon_cookie);
 	int bufsize = ndmpd_zfs_args->nz_bufsize;
@@ -843,7 +847,7 @@ ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	buf = ndmp_malloc(bufsize);
 	if (buf == NULL) {
 		NDMP_LOG(LOG_DEBUG, "buf NULL");
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	bytes_totalp = &ndmpd_zfs_args->nz_nlp->nlp_bytes_total;
@@ -861,7 +865,7 @@ ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 			NDMP_LOG(LOG_ERR, "MOD_READ error: %d; returning -1",
 			    err);
 			free(buf);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 
 		count = write(ndmpd_zfs_args->nz_pipe_fd[PIPE_TAPE], buf,
@@ -880,7 +884,7 @@ ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 			}
 
 			free(buf);
-			return (-1);
+			return ((void *)(uintptr_t)-1);
 		}
 
 		NS_ADD(wdisk, count);
@@ -889,7 +893,7 @@ ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	}
 
 	free(buf);
-	return (0);
+	return (NULL);
 }
 
 /*
@@ -898,9 +902,10 @@ ndmpd_zfs_restore_tape_read(ndmpd_zfs_args_t *ndmpd_zfs_args)
  * This routine executes zfs_receive() to restore the backup.
  */
 
-static int
-ndmpd_zfs_restore_recv_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
+static void *
+ndmpd_zfs_restore_recv_write(void *ptr)
 {
+	ndmpd_zfs_args_t *ndmpd_zfs_args = ptr;
 	ndmpd_session_t *session = (ndmpd_session_t *)
 	    (ndmpd_zfs_params->mp_daemon_cookie);
 	recvflags_t flags;
@@ -921,7 +926,7 @@ ndmpd_zfs_restore_recv_write(ndmpd_zfs_args_t *ndmpd_zfs_args)
 	if (err && !session->ns_data.dd_abort)
 		NDMPD_ZFS_LOG_ZERR(ndmpd_zfs_args, "zfs_receive: %d", err);
 
-	return (err);
+	return ((void *)(uintptr_t)err);
 }
 
 /*
@@ -941,12 +946,12 @@ ndmpd_zfs_reader_writer(ndmpd_zfs_args_t *ndmpd_zfs_args,
 
 	switch (ndmpd_zfs_params->mp_operation) {
 	case NDMP_DATA_OP_BACKUP:
-		sendrecv_func = (funct_t)ndmpd_zfs_backup_send_read;
-		tape_func = (funct_t)ndmpd_zfs_backup_tape_write;
+		sendrecv_func = ndmpd_zfs_backup_send_read;
+		tape_func = ndmpd_zfs_backup_tape_write;
 		break;
 	case NDMP_DATA_OP_RECOVER:
-		sendrecv_func = (funct_t)ndmpd_zfs_restore_recv_write;
-		tape_func = (funct_t)ndmpd_zfs_restore_tape_read;
+		sendrecv_func = ndmpd_zfs_restore_recv_write;
+		tape_func = ndmpd_zfs_restore_tape_read;
 		break;
 	}
 
diff --git a/usr/src/cmd/ndmpd/tlm/tlm_proto.h b/usr/src/cmd/ndmpd/tlm/tlm_proto.h
index f37f206e6e..6bdd214a5d 100644
--- a/usr/src/cmd/ndmpd/tlm/tlm_proto.h
+++ b/usr/src/cmd/ndmpd/tlm/tlm_proto.h
@@ -10,10 +10,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -134,7 +134,7 @@ extern int tar_putdir(char *,
     tlm_cmd_t *,
     tlm_job_stats_t *);
 
-extern int tar_getfile(tlm_backup_restore_arg_t *);
+extern void * tar_getfile(void *);
 
 extern int
 tar_getdir(tlm_commands_t *,
diff --git a/usr/src/cmd/ndmpd/tlm/tlm_restore_writer.c b/usr/src/cmd/ndmpd/tlm/tlm_restore_writer.c
index 82af6967ef..f757e59deb 100644
--- a/usr/src/cmd/ndmpd/tlm/tlm_restore_writer.c
+++ b/usr/src/cmd/ndmpd/tlm/tlm_restore_writer.c
@@ -11,10 +11,10 @@
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
- * 	- Redistributions of source code must retain the above copyright
+ *	- Redistributions of source code must retain the above copyright
  *	  notice, this list of conditions and the following disclaimer.
  *
- * 	- Redistributions in binary form must reproduce the above copyright
+ *	- Redistributions in binary form must reproduce the above copyright
  *	  notice, this list of conditions and the following disclaimer in
  *	  the documentation and/or other materials provided with the
  *	  distribution.
@@ -286,8 +286,8 @@ tar_getdir(tlm_commands_t *commands,
 	long nm_end, lnk_end;
 	char	*name, *nmp;
 	cstack_t *stp;
-	char 	*bkpath;
-	char 	*parentlnk;
+	char	*bkpath;
+	char	*parentlnk;
 	int dir_dar = 0;
 
 	/*
@@ -987,9 +987,10 @@ tar_getdir(tlm_commands_t *commands,
 /*
  * Main file restore function for tar (should run as a thread)
  */
-int
-tar_getfile(tlm_backup_restore_arg_t *argp)
+void *
+tar_getfile(void *ptr)
 {
+	tlm_backup_restore_arg_t *argp = ptr;
 	tlm_job_stats_t	*job_stats;
 	char	**sels;		/* list of files desired */
 	char	**exls;		/* list of files not wanted */
@@ -1012,7 +1013,7 @@ tar_getfile(tlm_backup_restore_arg_t *argp)
 	if (dir == NULL) {
 		local_commands->tc_reader = TLM_STOP;
 		(void) pthread_barrier_wait(&argp->ba_barrier);
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	(void) strlcpy(job, argp->ba_job, TLM_MAX_BACKUP_JOB_NAME+1);
@@ -1031,7 +1032,7 @@ tar_getfile(tlm_backup_restore_arg_t *argp)
 		local_commands->tc_reader = TLM_STOP;
 		free(dir);
 		(void) pthread_barrier_wait(&argp->ba_barrier);
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 
 	sels = argp->ba_sels;
@@ -1039,7 +1040,7 @@ tar_getfile(tlm_backup_restore_arg_t *argp)
 		local_commands->tc_reader = TLM_STOP;
 		free(dir);
 		(void) pthread_barrier_wait(&argp->ba_barrier);
-		return (-1);
+		return ((void *)(uintptr_t)-1);
 	}
 	exls = &list;
 
@@ -1081,7 +1082,7 @@ tar_getfile(tlm_backup_restore_arg_t *argp)
 	local_commands->tc_reader = TLM_STOP;
 	tlm_release_reader_writer_ipc(local_commands);
 	free(dir);
-	return (erc);
+	return ((void *)(uintptr_t)erc);
 }
 
 /*
diff --git a/usr/src/cmd/raidz_test/Makefile b/usr/src/cmd/raidz_test/Makefile
new file mode 100644
index 0000000000..43e0c07829
--- /dev/null
+++ b/usr/src/cmd/raidz_test/Makefile
@@ -0,0 +1,61 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2020 Joyent, Inc.
+#
+
+include ../Makefile.cmd
+include ../Makefile.cmd.64
+
+PROG= raidz_test
+OBJS= raidz_test.o raidz_bench.o
+SRCS= $(OBJS:%.o=%.c)
+POFILES= $(PROG:%=%.po)
+
+# No msg catalog here.
+POFILE=
+
+LDLIBS += -lzpool -lfakekernel -lumem
+
+INCS += -I../../lib/libzpool/common
+INCS += -I../../uts/common/fs/zfs
+
+CPPFLAGS.first = -I$(SRC)/lib/libfakekernel/common -D_FAKE_KERNEL
+CPPFLAGS += -D_LARGEFILE64_SOURCE=1
+CPPFLAGS += $(INCS)
+
+CSTD =   $(CSTD_GNU99)
+
+CERRWARN += -_gcc=-Wno-type-limits
+
+SMATCH=off
+
+.KEEP_STATE:
+
+all: $(PROG)
+
+$(PROG): $(OBJS)
+	$(LINK.c) -o $(PROG) $(OBJS) $(LDLIBS)
+	$(POST_PROCESS)
+
+install: all $(ROOTPROG)
+
+clean:
+	$(RM) $(OBJS)
+
+_msg: $(MSGDOMAIN) $(POFILES)
+	$(CP) $(POFILES) $(MSGDOMAIN)
+
+$(MSGDOMAIN):
+	$(INS.dir)
+
+include ../Makefile.targ
diff --git a/usr/src/cmd/raidz_test/raidz_bench.c b/usr/src/cmd/raidz_test/raidz_bench.c
new file mode 100644
index 0000000000..9dc22af6fd
--- /dev/null
+++ b/usr/src/cmd/raidz_test/raidz_bench.c
@@ -0,0 +1,228 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+#include <sys/zio.h>
+#include <sys/vdev_raidz.h>
+#include <sys/vdev_raidz_impl.h>
+#include <stdio.h>
+#include <strings.h>
+
+#include <sys/time.h>
+
+#include "raidz_test.h"
+
+#define	GEN_BENCH_MEMORY	(((uint64_t)1ULL)<<32)
+#define	REC_BENCH_MEMORY	(((uint64_t)1ULL)<<29)
+#define	BENCH_ASHIFT		12
+#define	MIN_CS_SHIFT		BENCH_ASHIFT
+#define	MAX_CS_SHIFT		SPA_MAXBLOCKSHIFT
+
+static zio_t zio_bench;
+static raidz_map_t *rm_bench;
+static size_t max_data_size = SPA_MAXBLOCKSIZE;
+
+static void
+bench_init_raidz_map(void)
+{
+	zio_bench.io_offset = 0;
+	zio_bench.io_size = max_data_size;
+
+	/*
+	 * To permit larger column sizes these have to be done
+	 * allocated using aligned alloc instead of zio_abd_buf_alloc
+	 */
+	zio_bench.io_abd = raidz_alloc(max_data_size);
+
+	init_zio_abd(&zio_bench);
+}
+
+static void
+bench_fini_raidz_maps(void)
+{
+	/* tear down golden zio */
+	raidz_free(zio_bench.io_abd, max_data_size);
+	bzero(&zio_bench, sizeof (zio_t));
+}
+
+static inline void
+run_gen_bench_impl(const char *impl)
+{
+	int fn, ncols;
+	uint64_t ds, iter_cnt, iter, disksize;
+	hrtime_t start;
+	double elapsed, d_bw;
+
+	/* Benchmark generate functions */
+	for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
+
+		for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
+			/* create suitable raidz_map */
+			ncols = rto_opts.rto_dcols + fn + 1;
+			zio_bench.io_size = 1ULL << ds;
+			rm_bench = vdev_raidz_map_alloc(&zio_bench,
+			    BENCH_ASHIFT, ncols, fn+1);
+
+			/* estimate iteration count */
+			iter_cnt = GEN_BENCH_MEMORY;
+			iter_cnt /= zio_bench.io_size;
+
+			start = gethrtime();
+			for (iter = 0; iter < iter_cnt; iter++)
+				vdev_raidz_generate_parity(rm_bench);
+			elapsed = NSEC2SEC((double)(gethrtime() - start));
+
+			disksize = (1ULL << ds) / rto_opts.rto_dcols;
+			d_bw = (double)iter_cnt * (double)disksize;
+			d_bw /= (1024.0 * 1024.0 * elapsed);
+
+			LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
+			    impl,
+			    raidz_gen_name[fn],
+			    rto_opts.rto_dcols,
+			    (1ULL<<ds),
+			    d_bw,
+			    d_bw * (double)(ncols),
+			    (unsigned)iter_cnt);
+
+			vdev_raidz_map_free(rm_bench);
+		}
+	}
+}
+
+void
+run_gen_bench(void)
+{
+	char **impl_name;
+
+	LOG(D_INFO, DBLSEP "\nBenchmarking parity generation...\n\n");
+	LOG(D_ALL, "impl, math, dcols, iosize, disk_bw, total_bw, iter\n");
+
+	for (impl_name = (char **)raidz_impl_names; *impl_name != NULL;
+	    impl_name++) {
+
+		if (vdev_raidz_impl_set(*impl_name) != 0)
+			continue;
+
+		run_gen_bench_impl(*impl_name);
+	}
+}
+
+static void
+run_rec_bench_impl(const char *impl)
+{
+	int fn, ncols, nbad;
+	uint64_t ds, iter_cnt, iter, disksize;
+	hrtime_t start;
+	double elapsed, d_bw;
+	static const int tgt[7][3] = {
+		{1, 2, 3},	/* rec_p:   bad QR & D[0]	*/
+		{0, 2, 3},	/* rec_q:   bad PR & D[0]	*/
+		{0, 1, 3},	/* rec_r:   bad PQ & D[0]	*/
+		{2, 3, 4},	/* rec_pq:  bad R  & D[0][1]	*/
+		{1, 3, 4},	/* rec_pr:  bad Q  & D[0][1]	*/
+		{0, 3, 4},	/* rec_qr:  bad P  & D[0][1]	*/
+		{3, 4, 5}	/* rec_pqr: bad    & D[0][1][2] */
+	};
+
+	for (fn = 0; fn < RAIDZ_REC_NUM; fn++) {
+		for (ds = MIN_CS_SHIFT; ds <= MAX_CS_SHIFT; ds++) {
+
+			/* create suitable raidz_map */
+			ncols = rto_opts.rto_dcols + PARITY_PQR;
+			zio_bench.io_size = 1ULL << ds;
+
+			/*
+			 * raidz block is too short to test
+			 * the requested method
+			 */
+			if (zio_bench.io_size / rto_opts.rto_dcols <
+			    (1ULL << BENCH_ASHIFT))
+				continue;
+
+			rm_bench = vdev_raidz_map_alloc(&zio_bench,
+			    BENCH_ASHIFT, ncols, PARITY_PQR);
+
+			/* estimate iteration count */
+			iter_cnt = (REC_BENCH_MEMORY);
+			iter_cnt /= zio_bench.io_size;
+
+			/* calculate how many bad columns there are */
+			nbad = MIN(3, raidz_ncols(rm_bench) -
+			    raidz_parity(rm_bench));
+
+			start = gethrtime();
+			for (iter = 0; iter < iter_cnt; iter++)
+				vdev_raidz_reconstruct(rm_bench, tgt[fn], nbad);
+			elapsed = NSEC2SEC((double)(gethrtime() - start));
+
+			disksize = (1ULL << ds) / rto_opts.rto_dcols;
+			d_bw = (double)iter_cnt * (double)(disksize);
+			d_bw /= (1024.0 * 1024.0 * elapsed);
+
+			LOG(D_ALL, "%10s, %8s, %zu, %10llu, %lf, %lf, %u\n",
+			    impl,
+			    raidz_rec_name[fn],
+			    rto_opts.rto_dcols,
+			    (1ULL<<ds),
+			    d_bw,
+			    d_bw * (double)ncols,
+			    (unsigned)iter_cnt);
+
+			vdev_raidz_map_free(rm_bench);
+		}
+	}
+}
+
+void
+run_rec_bench(void)
+{
+	char **impl_name;
+
+	LOG(D_INFO, DBLSEP "\nBenchmarking data reconstruction...\n\n");
+	LOG(D_ALL, "impl, math, dcols, iosize, disk_bw, total_bw, iter\n");
+
+	for (impl_name = (char **)raidz_impl_names; *impl_name != NULL;
+	    impl_name++) {
+
+		if (vdev_raidz_impl_set(*impl_name) != 0)
+			continue;
+
+		run_rec_bench_impl(*impl_name);
+	}
+}
+
+void
+run_raidz_benchmark(void)
+{
+	bench_init_raidz_map();
+
+	run_gen_bench();
+	run_rec_bench();
+
+	bench_fini_raidz_maps();
+}
diff --git a/usr/src/cmd/raidz_test/raidz_test.c b/usr/src/cmd/raidz_test/raidz_test.c
new file mode 100644
index 0000000000..8d025b479d
--- /dev/null
+++ b/usr/src/cmd/raidz_test/raidz_test.c
@@ -0,0 +1,761 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ * Copyright 2020 Joyent, Inc.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/time.h>
+#include <sys/wait.h>
+#include <sys/zio.h>
+#include <umem.h>
+#include <sys/vdev_raidz.h>
+#include <sys/vdev_raidz_impl.h>
+#include <assert.h>
+#include <stdio.h>
+#include <strings.h>
+#include <unistd.h>
+#include "raidz_test.h"
+
+static int *rand_data;
+raidz_test_opts_t rto_opts;
+
+static char gdb[256];
+static const char gdb_tmpl[] = "gdb -ex \"set pagination 0\" -p %d";
+
+#define	boot_ncpus	(sysconf(_SC_NPROCESSORS_ONLN))
+
+static void print_opts(raidz_test_opts_t *opts, boolean_t force)
+{
+	char *verbose;
+	switch (opts->rto_v) {
+		case 0:
+			verbose = "no";
+			break;
+		case 1:
+			verbose = "info";
+			break;
+		default:
+			verbose = "debug";
+			break;
+	}
+
+	if (force || opts->rto_v >= D_INFO) {
+		(void) fprintf(stdout, DBLSEP "Running with options:\n"
+		    "  (-a) zio ashift                   : %zu\n"
+		    "  (-o) zio offset                   : 1 << %zu\n"
+		    "  (-d) number of raidz data columns : %zu\n"
+		    "  (-s) size of DATA                 : 1 << %zu\n"
+		    "  (-S) sweep parameters             : %s \n"
+		    "  (-v) verbose                      : %s \n\n",
+		    opts->rto_ashift,			/* -a */
+		    ilog2(opts->rto_offset),		/* -o */
+		    opts->rto_dcols,			/* -d */
+		    ilog2(opts->rto_dsize),		/* -s */
+		    opts->rto_sweep ? "yes" : "no",	/* -S */
+		    verbose);				/* -v */
+	}
+}
+
+static void usage(boolean_t requested)
+{
+	const raidz_test_opts_t *o = &rto_opts_defaults;
+
+	FILE *fp = requested ? stdout : stderr;
+
+	(void) fprintf(fp, "Usage:\n"
+	    "\t[-a zio ashift (default: %zu)]\n"
+	    "\t[-o zio offset, exponent radix 2 (default: %zu)]\n"
+	    "\t[-d number of raidz data columns (default: %zu)]\n"
+	    "\t[-s zio size, exponent radix 2 (default: %zu)]\n"
+	    "\t[-S parameter sweep (default: %s)]\n"
+	    "\t[-t timeout for parameter sweep test]\n"
+	    "\t[-B benchmark all raidz implementations]\n"
+	    "\t[-v increase verbosity (default: %zu)]\n"
+	    "\t[-h (print help)]\n"
+	    "\t[-T test the test, see if failure would be detected]\n"
+	    "\t[-D debug (attach gdb on SIGSEGV)]\n"
+	    "",
+	    o->rto_ashift,				/* -a */
+	    ilog2(o->rto_offset),			/* -o */
+	    o->rto_dcols,				/* -d */
+	    ilog2(o->rto_dsize),			/* -s */
+	    rto_opts.rto_sweep ? "yes" : "no",		/* -S */
+	    o->rto_v);					/* -d */
+
+	exit(requested ? 0 : 1);
+}
+
+static void process_options(int argc, char **argv)
+{
+	size_t value;
+	int opt;
+
+	raidz_test_opts_t *o = &rto_opts;
+
+	bcopy(&rto_opts_defaults, o, sizeof (*o));
+
+	while ((opt = getopt(argc, argv, "TDBSvha:o:d:s:t:")) != -1) {
+		value = 0;
+
+		switch (opt) {
+		case 'a':
+			value = strtoull(optarg, NULL, 0);
+			o->rto_ashift = MIN(13, MAX(9, value));
+			break;
+		case 'o':
+			value = strtoull(optarg, NULL, 0);
+			o->rto_offset = ((1ULL << MIN(12, value)) >> 9) << 9;
+			break;
+		case 'd':
+			value = strtoull(optarg, NULL, 0);
+			o->rto_dcols = MIN(255, MAX(1, value));
+			break;
+		case 's':
+			value = strtoull(optarg, NULL, 0);
+			o->rto_dsize = 1ULL <<  MIN(SPA_MAXBLOCKSHIFT,
+			    MAX(SPA_MINBLOCKSHIFT, value));
+			break;
+		case 't':
+			value = strtoull(optarg, NULL, 0);
+			o->rto_sweep_timeout = value;
+			break;
+		case 'v':
+			o->rto_v++;
+			break;
+		case 'S':
+			o->rto_sweep = 1;
+			break;
+		case 'B':
+			o->rto_benchmark = 1;
+			break;
+		case 'D':
+			o->rto_gdb = 1;
+			break;
+		case 'T':
+			o->rto_sanity = 1;
+			break;
+		case 'h':
+			usage(B_TRUE);
+			break;
+		case '?':
+		default:
+			usage(B_FALSE);
+			break;
+		}
+	}
+}
+
+#define	DATA_COL(rm, i) ((rm)->rm_col[raidz_parity(rm) + (i)].rc_abd)
+#define	DATA_COL_SIZE(rm, i) ((rm)->rm_col[raidz_parity(rm) + (i)].rc_size)
+
+#define	CODE_COL(rm, i) ((rm)->rm_col[(i)].rc_abd)
+#define	CODE_COL_SIZE(rm, i) ((rm)->rm_col[(i)].rc_size)
+
+static int
+cmp_code(raidz_test_opts_t *opts, const raidz_map_t *rm, const int parity)
+{
+	int i, ret = 0;
+
+	VERIFY(parity >= 1 && parity <= 3);
+
+	for (i = 0; i < parity; i++) {
+		if (abd_cmp(CODE_COL(rm, i), CODE_COL(opts->rm_golden, i),
+		    CODE_COL(rm, i)->abd_size) != 0) {
+			ret++;
+			LOG_OPT(D_DEBUG, opts,
+			    "\nParity block [%d] different!\n", i);
+		}
+	}
+	return (ret);
+}
+
+static int
+cmp_data(raidz_test_opts_t *opts, raidz_map_t *rm)
+{
+	int i, ret = 0;
+	int dcols = opts->rm_golden->rm_cols - raidz_parity(opts->rm_golden);
+
+	for (i = 0; i < dcols; i++) {
+		if (abd_cmp(DATA_COL(opts->rm_golden, i), DATA_COL(rm, i),
+		    DATA_COL(opts->rm_golden, i)->abd_size) != 0) {
+			ret++;
+
+			LOG_OPT(D_DEBUG, opts,
+			    "\nData block [%d] different!\n", i);
+		}
+	}
+	return (ret);
+}
+
+static int
+init_rand(void *data, size_t size, void *private)
+{
+	int i;
+	int *dst = (int *)data;
+
+	for (i = 0; i < size / sizeof (int); i++)
+		dst[i] = rand_data[i];
+
+	return (0);
+}
+
+static void
+corrupt_colums(raidz_map_t *rm, const int *tgts, const int cnt)
+{
+	int i;
+	raidz_col_t *col;
+
+	for (i = 0; i < cnt; i++) {
+		col = &rm->rm_col[tgts[i]];
+		(void) abd_iterate_func(col->rc_abd, 0, col->rc_size,
+		    init_rand, NULL);
+	}
+}
+
+void
+init_zio_abd(zio_t *zio)
+{
+	(void) abd_iterate_func(zio->io_abd, 0, zio->io_size, init_rand, NULL);
+}
+
+static void
+fini_raidz_map(zio_t **zio, raidz_map_t **rm)
+{
+	vdev_raidz_map_free(*rm);
+	raidz_free((*zio)->io_abd, (*zio)->io_size);
+	umem_free(*zio, sizeof (zio_t));
+
+	*zio = NULL;
+	*rm = NULL;
+}
+
+static int
+init_raidz_golden_map(raidz_test_opts_t *opts, const int parity)
+{
+	int err = 0;
+	zio_t *zio_test;
+	raidz_map_t *rm_test;
+	const size_t total_ncols = opts->rto_dcols + parity;
+
+	if (opts->rm_golden) {
+		fini_raidz_map(&opts->zio_golden, &opts->rm_golden);
+	}
+
+	opts->zio_golden = umem_zalloc(sizeof (zio_t), UMEM_NOFAIL);
+	zio_test = umem_zalloc(sizeof (zio_t), UMEM_NOFAIL);
+
+	opts->zio_golden->io_offset = zio_test->io_offset = opts->rto_offset;
+	opts->zio_golden->io_size = zio_test->io_size = opts->rto_dsize;
+
+	opts->zio_golden->io_abd = raidz_alloc(opts->rto_dsize);
+	zio_test->io_abd = raidz_alloc(opts->rto_dsize);
+
+	init_zio_abd(opts->zio_golden);
+	init_zio_abd(zio_test);
+
+	VERIFY0(vdev_raidz_impl_set("original"));
+
+	opts->rm_golden = vdev_raidz_map_alloc(opts->zio_golden,
+	    opts->rto_ashift, total_ncols, parity);
+	rm_test = vdev_raidz_map_alloc(zio_test,
+	    opts->rto_ashift, total_ncols, parity);
+
+	VERIFY(opts->zio_golden);
+	VERIFY(opts->rm_golden);
+
+	vdev_raidz_generate_parity(opts->rm_golden);
+	vdev_raidz_generate_parity(rm_test);
+
+	/* sanity check */
+	err |= cmp_data(opts, rm_test);
+	err |= cmp_code(opts, rm_test, parity);
+
+	if (err)
+		ERRMSG("initializing the golden copy ... [FAIL]!\n");
+
+	/* tear down raidz_map of test zio */
+	fini_raidz_map(&zio_test, &rm_test);
+
+	return (err);
+}
+
+static raidz_map_t *
+init_raidz_map(raidz_test_opts_t *opts, zio_t **zio, const int parity)
+{
+	raidz_map_t *rm = NULL;
+	const size_t alloc_dsize = opts->rto_dsize;
+	const size_t total_ncols = opts->rto_dcols + parity;
+	const int ccols[] = { 0, 1, 2 };
+
+	VERIFY(zio);
+	VERIFY(parity <= 3 && parity >= 1);
+
+	*zio = umem_zalloc(sizeof (zio_t), UMEM_NOFAIL);
+
+	(*zio)->io_offset = 0;
+	(*zio)->io_size = alloc_dsize;
+	(*zio)->io_abd = raidz_alloc(alloc_dsize);
+	init_zio_abd(*zio);
+
+	rm = vdev_raidz_map_alloc(*zio, opts->rto_ashift,
+	    total_ncols, parity);
+	VERIFY(rm);
+
+	/* Make sure code columns are destroyed */
+	corrupt_colums(rm, ccols, parity);
+
+	return (rm);
+}
+
+static int
+run_gen_check(raidz_test_opts_t *opts)
+{
+	char **impl_name;
+	int fn, err = 0;
+	zio_t *zio_test;
+	raidz_map_t *rm_test;
+
+	err = init_raidz_golden_map(opts, PARITY_PQR);
+	if (0 != err)
+		return (err);
+
+	LOG(D_INFO, DBLSEP);
+	LOG(D_INFO, "Testing parity generation...\n");
+
+	for (impl_name = (char **)raidz_impl_names+1; *impl_name != NULL;
+	    impl_name++) {
+
+		LOG(D_INFO, SEP);
+		LOG(D_INFO, "\tTesting [%s] implementation...", *impl_name);
+
+		if (0 != vdev_raidz_impl_set(*impl_name)) {
+			LOG(D_INFO, "[SKIP]\n");
+			continue;
+		} else {
+			LOG(D_INFO, "[SUPPORTED]\n");
+		}
+
+		for (fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
+
+			/* Check if should stop */
+			if (rto_opts.rto_should_stop)
+				return (err);
+
+			/* create suitable raidz_map */
+			rm_test = init_raidz_map(opts, &zio_test, fn+1);
+			VERIFY(rm_test);
+
+			LOG(D_INFO, "\t\tTesting method [%s] ...",
+			    raidz_gen_name[fn]);
+
+			if (!opts->rto_sanity)
+				vdev_raidz_generate_parity(rm_test);
+
+			if (cmp_code(opts, rm_test, fn+1) != 0) {
+				LOG(D_INFO, "[FAIL]\n");
+				err++;
+			} else
+				LOG(D_INFO, "[PASS]\n");
+
+			fini_raidz_map(&zio_test, &rm_test);
+		}
+	}
+
+	fini_raidz_map(&opts->zio_golden, &opts->rm_golden);
+
+	return (err);
+}
+
+static int
+run_rec_check_impl(raidz_test_opts_t *opts, raidz_map_t *rm, const int fn)
+{
+	int x0, x1, x2;
+	int tgtidx[3];
+	int err = 0;
+	static const int rec_tgts[7][3] = {
+		{1, 2, 3},	/* rec_p:   bad QR & D[0]	*/
+		{0, 2, 3},	/* rec_q:   bad PR & D[0]	*/
+		{0, 1, 3},	/* rec_r:   bad PQ & D[0]	*/
+		{2, 3, 4},	/* rec_pq:  bad R  & D[0][1]	*/
+		{1, 3, 4},	/* rec_pr:  bad Q  & D[0][1]	*/
+		{0, 3, 4},	/* rec_qr:  bad P  & D[0][1]	*/
+		{3, 4, 5}	/* rec_pqr: bad    & D[0][1][2] */
+	};
+
+	memcpy(tgtidx, rec_tgts[fn], sizeof (tgtidx));
+
+	if (fn < RAIDZ_REC_PQ) {
+		/* can reconstruct 1 failed data disk */
+		for (x0 = 0; x0 < opts->rto_dcols; x0++) {
+			if (x0 >= rm->rm_cols - raidz_parity(rm))
+				continue;
+
+			/* Check if should stop */
+			if (rto_opts.rto_should_stop)
+				return (err);
+
+			LOG(D_DEBUG, "[%d] ", x0);
+
+			tgtidx[2] = x0 + raidz_parity(rm);
+
+			corrupt_colums(rm, tgtidx+2, 1);
+
+			if (!opts->rto_sanity)
+				(void) vdev_raidz_reconstruct(rm, tgtidx, 3);
+
+			if (cmp_data(opts, rm) != 0) {
+				err++;
+				LOG(D_DEBUG, "\nREC D[%d]... [FAIL]\n", x0);
+			}
+		}
+
+	} else if (fn < RAIDZ_REC_PQR) {
+		/* can reconstruct 2 failed data disk */
+		for (x0 = 0; x0 < opts->rto_dcols; x0++) {
+			if (x0 >= rm->rm_cols - raidz_parity(rm))
+				continue;
+			for (x1 = x0 + 1; x1 < opts->rto_dcols; x1++) {
+				if (x1 >= rm->rm_cols - raidz_parity(rm))
+					continue;
+
+				/* Check if should stop */
+				if (rto_opts.rto_should_stop)
+					return (err);
+
+				LOG(D_DEBUG, "[%d %d] ", x0, x1);
+
+				tgtidx[1] = x0 + raidz_parity(rm);
+				tgtidx[2] = x1 + raidz_parity(rm);
+
+				corrupt_colums(rm, tgtidx+1, 2);
+
+				if (!opts->rto_sanity)
+					(void) vdev_raidz_reconstruct(rm,
+					    tgtidx, 3);
+
+				if (cmp_data(opts, rm) != 0) {
+					err++;
+					LOG(D_DEBUG, "\nREC D[%d %d]... "
+					    "[FAIL]\n", x0, x1);
+				}
+			}
+		}
+	} else {
+		/* can reconstruct 3 failed data disk */
+		for (x0 = 0; x0 < opts->rto_dcols; x0++) {
+			if (x0 >= rm->rm_cols - raidz_parity(rm))
+				continue;
+			for (x1 = x0 + 1; x1 < opts->rto_dcols; x1++) {
+				if (x1 >= rm->rm_cols - raidz_parity(rm))
+					continue;
+				for (x2 = x1 + 1; x2 < opts->rto_dcols; x2++) {
+					if (x2 >=
+					    rm->rm_cols - raidz_parity(rm))
+						continue;
+
+					/* Check if should stop */
+					if (rto_opts.rto_should_stop)
+						return (err);
+
+					LOG(D_DEBUG, "[%d %d %d]", x0, x1, x2);
+
+					tgtidx[0] = x0 + raidz_parity(rm);
+					tgtidx[1] = x1 + raidz_parity(rm);
+					tgtidx[2] = x2 + raidz_parity(rm);
+
+					corrupt_colums(rm, tgtidx, 3);
+
+					if (!opts->rto_sanity)
+						(void) vdev_raidz_reconstruct(
+						    rm, tgtidx, 3);
+
+					if (cmp_data(opts, rm) != 0) {
+						err++;
+						LOG(D_DEBUG,
+						    "\nREC D[%d %d %d]... "
+						    "[FAIL]\n", x0, x1, x2);
+					}
+				}
+			}
+		}
+	}
+	return (err);
+}
+
+static int
+run_rec_check(raidz_test_opts_t *opts)
+{
+	char **impl_name;
+	unsigned fn, err = 0;
+	zio_t *zio_test;
+	raidz_map_t *rm_test;
+
+	err = init_raidz_golden_map(opts, PARITY_PQR);
+	if (0 != err)
+		return (err);
+
+	LOG(D_INFO, DBLSEP);
+	LOG(D_INFO, "Testing data reconstruction...\n");
+
+	for (impl_name = (char **)raidz_impl_names+1; *impl_name != NULL;
+	    impl_name++) {
+
+		LOG(D_INFO, SEP);
+		LOG(D_INFO, "\tTesting [%s] implementation...", *impl_name);
+
+		if (vdev_raidz_impl_set(*impl_name) != 0) {
+			LOG(D_INFO, "[SKIP]\n");
+			continue;
+		} else
+			LOG(D_INFO, "[SUPPORTED]\n");
+
+
+		/* create suitable raidz_map */
+		rm_test = init_raidz_map(opts, &zio_test, PARITY_PQR);
+		/* generate parity */
+		vdev_raidz_generate_parity(rm_test);
+
+		for (fn = 0; fn < RAIDZ_REC_NUM; fn++) {
+
+			LOG(D_INFO, "\t\tTesting method [%s] ...",
+			    raidz_rec_name[fn]);
+
+			if (run_rec_check_impl(opts, rm_test, fn) != 0) {
+				LOG(D_INFO, "[FAIL]\n");
+				err++;
+
+			} else
+				LOG(D_INFO, "[PASS]\n");
+
+		}
+		/* tear down test raidz_map */
+		fini_raidz_map(&zio_test, &rm_test);
+	}
+
+	fini_raidz_map(&opts->zio_golden, &opts->rm_golden);
+
+	return (err);
+}
+
+static int
+run_test(raidz_test_opts_t *opts)
+{
+	int err = 0;
+
+	if (opts == NULL)
+		opts = &rto_opts;
+
+	print_opts(opts, B_FALSE);
+
+	err |= run_gen_check(opts);
+	err |= run_rec_check(opts);
+
+	return (err);
+}
+
+#define	SWEEP_RUNNING	0
+#define	SWEEP_FINISHED	1
+#define	SWEEP_ERROR	2
+#define	SWEEP_TIMEOUT	3
+
+static int sweep_state = 0;
+static raidz_test_opts_t failed_opts;
+
+static kmutex_t sem_mtx;
+static kcondvar_t sem_cv;
+static int max_free_slots;
+static int free_slots;
+
+static void
+sweep_thread(void *arg)
+{
+	int err = 0;
+	raidz_test_opts_t *opts = (raidz_test_opts_t *)arg;
+	VERIFY(opts != NULL);
+
+	err = run_test(opts);
+
+	if (rto_opts.rto_sanity) {
+		/* 25% chance that a sweep test fails */
+		if (rand() < (RAND_MAX/4))
+			err = 1;
+	}
+
+	if (0 != err) {
+		mutex_enter(&sem_mtx);
+		memcpy(&failed_opts, opts, sizeof (raidz_test_opts_t));
+		sweep_state = SWEEP_ERROR;
+		mutex_exit(&sem_mtx);
+	}
+
+	umem_free(opts, sizeof (raidz_test_opts_t));
+
+	/* signal the next thread */
+	mutex_enter(&sem_mtx);
+	free_slots++;
+	cv_signal(&sem_cv);
+	mutex_exit(&sem_mtx);
+
+	thread_exit();
+}
+
+static int
+run_sweep(void)
+{
+	static const size_t dcols_v[] = { 1, 2, 3, 4, 5, 6, 7, 8, 12, 15, 16 };
+	static const size_t ashift_v[] = { 9, 12, 14 };
+	static const size_t size_v[] = { 1 << 9, 21 * (1 << 9), 13 * (1 << 12),
+		1 << 17, (1 << 20) - (1 << 12), SPA_MAXBLOCKSIZE };
+
+	(void) setvbuf(stdout, NULL, _IONBF, 0);
+
+	ulong_t total_comb = ARRAY_SIZE(size_v) * ARRAY_SIZE(ashift_v) *
+	    ARRAY_SIZE(dcols_v);
+	ulong_t tried_comb = 0;
+	hrtime_t time_diff, start_time = gethrtime();
+	raidz_test_opts_t *opts;
+	int a, d, s;
+
+	max_free_slots = free_slots = MAX(2, boot_ncpus);
+
+	mutex_init(&sem_mtx, NULL, MUTEX_DEFAULT, NULL);
+	cv_init(&sem_cv, NULL, CV_DEFAULT, NULL);
+
+	for (s = 0; s < ARRAY_SIZE(size_v); s++)
+	for (a = 0; a < ARRAY_SIZE(ashift_v); a++)
+	for (d = 0; d < ARRAY_SIZE(dcols_v); d++) {
+
+		if (size_v[s] < (1 << ashift_v[a])) {
+			total_comb--;
+			continue;
+		}
+
+		if (++tried_comb % 20 == 0)
+			LOG(D_ALL, "%lu/%lu... ", tried_comb, total_comb);
+
+		/* wait for signal to start new thread */
+		mutex_enter(&sem_mtx);
+		while (cv_timedwait_sig(&sem_cv, &sem_mtx,
+		    ddi_get_lbolt() + hz)) {
+
+			/* check if should stop the test (timeout) */
+			time_diff = (gethrtime() - start_time) / NANOSEC;
+			if (rto_opts.rto_sweep_timeout > 0 &&
+			    time_diff >= rto_opts.rto_sweep_timeout) {
+				sweep_state = SWEEP_TIMEOUT;
+				rto_opts.rto_should_stop = B_TRUE;
+				mutex_exit(&sem_mtx);
+				goto exit;
+			}
+
+			/* check if should stop the test (error) */
+			if (sweep_state != SWEEP_RUNNING) {
+				mutex_exit(&sem_mtx);
+				goto exit;
+			}
+
+			/* exit loop if a slot is available */
+			if (free_slots > 0) {
+				break;
+			}
+		}
+
+		free_slots--;
+		mutex_exit(&sem_mtx);
+
+		opts = umem_zalloc(sizeof (raidz_test_opts_t), UMEM_NOFAIL);
+		opts->rto_ashift = ashift_v[a];
+		opts->rto_dcols = dcols_v[d];
+		opts->rto_offset = (1 << ashift_v[a]) * rand();
+		opts->rto_dsize = size_v[s];
+		opts->rto_v = 0; /* be quiet */
+
+		VERIFY3P(thread_create(NULL, 0, sweep_thread, (void *) opts,
+		    0, NULL, TS_RUN, maxclsyspri), !=, NULL);
+	}
+
+exit:
+	LOG(D_ALL, "\nWaiting for test threads to finish...\n");
+	mutex_enter(&sem_mtx);
+	VERIFY(free_slots <= max_free_slots);
+	while (free_slots < max_free_slots) {
+		(void) cv_wait(&sem_cv, &sem_mtx);
+	}
+	mutex_exit(&sem_mtx);
+
+	if (sweep_state == SWEEP_ERROR) {
+		ERRMSG("Sweep test failed! Failed option: \n");
+		print_opts(&failed_opts, B_TRUE);
+	} else {
+		if (sweep_state == SWEEP_TIMEOUT)
+			LOG(D_ALL, "Test timeout (%lus). Stopping...\n",
+			    (ulong_t)rto_opts.rto_sweep_timeout);
+
+		LOG(D_ALL, "Sweep test succeeded on %lu raidz maps!\n",
+		    (ulong_t)tried_comb);
+	}
+
+	mutex_destroy(&sem_mtx);
+
+	return (sweep_state == SWEEP_ERROR ? SWEEP_ERROR : 0);
+}
+
+int
+main(int argc, char **argv)
+{
+	size_t i;
+	int err = 0;
+
+	/* init gdb string early */
+	(void) sprintf(gdb, gdb_tmpl, getpid());
+
+	(void) setvbuf(stdout, NULL, _IOLBF, 0);
+
+	dprintf_setup(&argc, argv);
+
+	process_options(argc, argv);
+
+	kernel_init(FREAD);
+
+	/* setup random data because rand() is not reentrant */
+	rand_data = (int *)umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
+	srand((unsigned)time(NULL) * getpid());
+	for (i = 0; i < SPA_MAXBLOCKSIZE / sizeof (int); i++)
+		rand_data[i] = rand();
+
+	mprotect((void *)rand_data, SPA_MAXBLOCKSIZE, PROT_READ);
+
+	if (rto_opts.rto_benchmark) {
+		run_raidz_benchmark();
+	} else if (rto_opts.rto_sweep) {
+		err = run_sweep();
+	} else {
+		err = run_test(NULL);
+	}
+
+	umem_free(rand_data, SPA_MAXBLOCKSIZE);
+	kernel_fini();
+
+	return (err);
+}
diff --git a/usr/src/cmd/raidz_test/raidz_test.h b/usr/src/cmd/raidz_test/raidz_test.h
new file mode 100644
index 0000000000..c91e521436
--- /dev/null
+++ b/usr/src/cmd/raidz_test/raidz_test.h
@@ -0,0 +1,117 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ * Copyright 2020 Joyent, Inc.
+ */
+
+#ifndef	RAIDZ_TEST_H
+#define	RAIDZ_TEST_H
+
+#include <sys/spa.h>
+
+static const char *raidz_impl_names[] = {
+	"original",
+	"scalar",
+	"sse2",
+	"ssse3",
+	"avx2",
+	"avx512f",
+	"avx512bw",
+	"aarch64_neon",
+	"aarch64_neonx2",
+	"powerpc_altivec",
+	NULL
+};
+
+typedef struct raidz_test_opts {
+	size_t rto_ashift;
+	size_t rto_offset;
+	size_t rto_dcols;
+	size_t rto_dsize;
+	size_t rto_v;
+	size_t rto_sweep;
+	size_t rto_sweep_timeout;
+	size_t rto_benchmark;
+	size_t rto_sanity;
+	size_t rto_gdb;
+
+	/* non-user options */
+	boolean_t rto_should_stop;
+
+	zio_t *zio_golden;
+	raidz_map_t *rm_golden;
+} raidz_test_opts_t;
+
+static const raidz_test_opts_t rto_opts_defaults = {
+	.rto_ashift = 9,
+	.rto_offset = 1ULL << 0,
+	.rto_dcols = 8,
+	.rto_dsize = 1<<19,
+	.rto_v = 0,
+	.rto_sweep = 0,
+	.rto_benchmark = 0,
+	.rto_sanity = 0,
+	.rto_gdb = 0,
+	.rto_should_stop = B_FALSE
+};
+
+extern raidz_test_opts_t rto_opts;
+
+static inline size_t ilog2(size_t a)
+{
+	return (a > 1 ? 1 + ilog2(a >> 1) : 0);
+}
+
+
+#define	D_ALL	0
+#define	D_INFO	1
+#define	D_DEBUG	2
+
+#define	LOG(lvl, a...)				\
+{						\
+	if (rto_opts.rto_v >= lvl)		\
+		(void) fprintf(stdout, a);	\
+}						\
+
+#define	LOG_OPT(lvl, opt, a...)			\
+{						\
+	if (opt->rto_v >= lvl)			\
+		(void) fprintf(stdout, a);	\
+}						\
+
+#define	ERRMSG(a...)	(void) fprintf(stderr, a)
+
+
+#define	DBLSEP "================\n"
+#define	SEP    "----------------\n"
+
+
+#define	raidz_alloc(size)	abd_alloc(size, B_FALSE)
+#define	raidz_free(p, size)	abd_free(p)
+
+
+void init_zio_abd(zio_t *zio);
+
+void run_raidz_benchmark(void);
+
+#endif /* RAIDZ_TEST_H */
diff --git a/usr/src/compat/bhyve/README b/usr/src/compat/bhyve/README
new file mode 100644
index 0000000000..87ceb83695
--- /dev/null
+++ b/usr/src/compat/bhyve/README
@@ -0,0 +1,9 @@
+These are the compatibility headers for building bhyve on illumos.  They are
+comprised of the shims and glue needed to use native functionality as the
+backing for FreeBSD interfaces.  It often means heavy use of #include_next,
+#define renames, and forward definitions to some glue functions.
+
+For headers which would otherwise be copied verbatim from FreeBSD, the
+usr/src/contrib/bhyve area is the appropriate home.  This allows us to carry
+only the bare minimum in the 'compat' headers while making updates for the
+'contrib' headers easy (simply copy over the new version).
diff --git a/usr/src/compat/freebsd/amd64/machine/asmacros.h b/usr/src/compat/bhyve/amd64/machine/asmacros.h
index 1f6955130b..1f6955130b 100644
--- a/usr/src/compat/freebsd/amd64/machine/asmacros.h
+++ b/usr/src/compat/bhyve/amd64/machine/asmacros.h
diff --git a/usr/src/compat/freebsd/amd64/machine/atomic.h b/usr/src/compat/bhyve/amd64/machine/atomic.h
index 1da9724b7d..1da9724b7d 100644
--- a/usr/src/compat/freebsd/amd64/machine/atomic.h
+++ b/usr/src/compat/bhyve/amd64/machine/atomic.h
diff --git a/usr/src/compat/freebsd/amd64/machine/clock.h b/usr/src/compat/bhyve/amd64/machine/clock.h
index f50b42a126..f50b42a126 100644
--- a/usr/src/compat/freebsd/amd64/machine/clock.h
+++ b/usr/src/compat/bhyve/amd64/machine/clock.h
diff --git a/usr/src/compat/freebsd/amd64/machine/cpu.h b/usr/src/compat/bhyve/amd64/machine/cpu.h
index 40253af108..40253af108 100644
--- a/usr/src/compat/freebsd/amd64/machine/cpu.h
+++ b/usr/src/compat/bhyve/amd64/machine/cpu.h
diff --git a/usr/src/compat/freebsd/amd64/machine/cpufunc.h b/usr/src/compat/bhyve/amd64/machine/cpufunc.h
index 0b7bcdaa59..0b7bcdaa59 100644
--- a/usr/src/compat/freebsd/amd64/machine/cpufunc.h
+++ b/usr/src/compat/bhyve/amd64/machine/cpufunc.h
diff --git a/usr/src/compat/freebsd/amd64/machine/fpu.h b/usr/src/compat/bhyve/amd64/machine/fpu.h
index 6bc651d996..6bc651d996 100644
--- a/usr/src/compat/freebsd/amd64/machine/fpu.h
+++ b/usr/src/compat/bhyve/amd64/machine/fpu.h
diff --git a/usr/src/compat/freebsd/amd64/machine/iodev.h b/usr/src/compat/bhyve/amd64/machine/iodev.h
index c7cdddc817..c7cdddc817 100644
--- a/usr/src/compat/freebsd/amd64/machine/iodev.h
+++ b/usr/src/compat/bhyve/amd64/machine/iodev.h
diff --git a/usr/src/compat/freebsd/amd64/machine/md_var.h b/usr/src/compat/bhyve/amd64/machine/md_var.h
index ed57a8bebc..ed57a8bebc 100644
--- a/usr/src/compat/freebsd/amd64/machine/md_var.h
+++ b/usr/src/compat/bhyve/amd64/machine/md_var.h
diff --git a/usr/src/compat/freebsd/amd64/machine/param.h b/usr/src/compat/bhyve/amd64/machine/param.h
index b152f4d526..b152f4d526 100644
--- a/usr/src/compat/freebsd/amd64/machine/param.h
+++ b/usr/src/compat/bhyve/amd64/machine/param.h
diff --git a/usr/src/compat/freebsd/amd64/machine/pcb.h b/usr/src/compat/bhyve/amd64/machine/pcb.h
index 75b5de640c..75b5de640c 100644
--- a/usr/src/compat/freebsd/amd64/machine/pcb.h
+++ b/usr/src/compat/bhyve/amd64/machine/pcb.h
diff --git a/usr/src/compat/freebsd/amd64/machine/pmap.h b/usr/src/compat/bhyve/amd64/machine/pmap.h
index ce3185629b..ce3185629b 100644
--- a/usr/src/compat/freebsd/amd64/machine/pmap.h
+++ b/usr/src/compat/bhyve/amd64/machine/pmap.h
diff --git a/usr/src/compat/freebsd/amd64/machine/reg.h b/usr/src/compat/bhyve/amd64/machine/reg.h
index 4a73463603..4a73463603 100644
--- a/usr/src/compat/freebsd/amd64/machine/reg.h
+++ b/usr/src/compat/bhyve/amd64/machine/reg.h
diff --git a/usr/src/compat/freebsd/amd64/machine/segments.h b/usr/src/compat/bhyve/amd64/machine/segments.h
index d0655f4a0e..d0655f4a0e 100644
--- a/usr/src/compat/freebsd/amd64/machine/segments.h
+++ b/usr/src/compat/bhyve/amd64/machine/segments.h
diff --git a/usr/src/compat/freebsd/amd64/machine/smp.h b/usr/src/compat/bhyve/amd64/machine/smp.h
index 9c4f2d111b..9c4f2d111b 100644
--- a/usr/src/compat/freebsd/amd64/machine/smp.h
+++ b/usr/src/compat/bhyve/amd64/machine/smp.h
diff --git a/usr/src/compat/freebsd/amd64/machine/specialreg.h b/usr/src/compat/bhyve/amd64/machine/specialreg.h
index ead63aaaab..ead63aaaab 100644
--- a/usr/src/compat/freebsd/amd64/machine/specialreg.h
+++ b/usr/src/compat/bhyve/amd64/machine/specialreg.h
diff --git a/usr/src/compat/freebsd/amd64/machine/vmm.h b/usr/src/compat/bhyve/amd64/machine/vmm.h
index 1c54c0830d..1c54c0830d 100644
--- a/usr/src/compat/freebsd/amd64/machine/vmm.h
+++ b/usr/src/compat/bhyve/amd64/machine/vmm.h
diff --git a/usr/src/compat/freebsd/amd64/machine/vmm_dev.h b/usr/src/compat/bhyve/amd64/machine/vmm_dev.h
index fe9cb6c705..fe9cb6c705 100644
--- a/usr/src/compat/freebsd/amd64/machine/vmm_dev.h
+++ b/usr/src/compat/bhyve/amd64/machine/vmm_dev.h
diff --git a/usr/src/compat/freebsd/amd64/machine/vmm_instruction_emul.h b/usr/src/compat/bhyve/amd64/machine/vmm_instruction_emul.h
index 02c3f391c7..02c3f391c7 100644
--- a/usr/src/compat/freebsd/amd64/machine/vmm_instruction_emul.h
+++ b/usr/src/compat/bhyve/amd64/machine/vmm_instruction_emul.h
diff --git a/usr/src/compat/freebsd/amd64/machine/vmparam.h b/usr/src/compat/bhyve/amd64/machine/vmparam.h
index c76a3259f3..c76a3259f3 100644
--- a/usr/src/compat/freebsd/amd64/machine/vmparam.h
+++ b/usr/src/compat/bhyve/amd64/machine/vmparam.h
diff --git a/usr/src/compat/freebsd/contrib/dev/acpica/include/acpi.h b/usr/src/compat/bhyve/contrib/dev/acpica/include/acpi.h
index 2668f98ab3..2668f98ab3 100644
--- a/usr/src/compat/freebsd/contrib/dev/acpica/include/acpi.h
+++ b/usr/src/compat/bhyve/contrib/dev/acpica/include/acpi.h
diff --git a/usr/src/compat/freebsd/dev/pci/pcivar.h b/usr/src/compat/bhyve/dev/pci/pcivar.h
index 064d983117..064d983117 100644
--- a/usr/src/compat/freebsd/dev/pci/pcivar.h
+++ b/usr/src/compat/bhyve/dev/pci/pcivar.h
diff --git a/usr/src/compat/freebsd/err.h b/usr/src/compat/bhyve/err.h
index 40d144e025..40d144e025 100644
--- a/usr/src/compat/freebsd/err.h
+++ b/usr/src/compat/bhyve/err.h
diff --git a/usr/src/compat/freebsd/libutil.h b/usr/src/compat/bhyve/libutil.h
index f899d4425e..f899d4425e 100644
--- a/usr/src/compat/freebsd/libutil.h
+++ b/usr/src/compat/bhyve/libutil.h
diff --git a/usr/src/compat/freebsd/net/ethernet.h b/usr/src/compat/bhyve/net/ethernet.h
index dcd3a58925..dcd3a58925 100644
--- a/usr/src/compat/freebsd/net/ethernet.h
+++ b/usr/src/compat/bhyve/net/ethernet.h
diff --git a/usr/src/compat/freebsd/net/ieee_oui.h b/usr/src/compat/bhyve/net/ieee_oui.h
index 068328d833..068328d833 100644
--- a/usr/src/compat/freebsd/net/ieee_oui.h
+++ b/usr/src/compat/bhyve/net/ieee_oui.h
diff --git a/usr/src/compat/freebsd/paths.h b/usr/src/compat/bhyve/paths.h
index e43c963f93..e43c963f93 100644
--- a/usr/src/compat/freebsd/paths.h
+++ b/usr/src/compat/bhyve/paths.h
diff --git a/usr/src/compat/freebsd/pthread_np.h b/usr/src/compat/bhyve/pthread_np.h
index c4f76b259c..c4f76b259c 100644
--- a/usr/src/compat/freebsd/pthread_np.h
+++ b/usr/src/compat/bhyve/pthread_np.h
diff --git a/usr/src/compat/freebsd/string.h b/usr/src/compat/bhyve/string.h
index 7e0f5c7ddc..7e0f5c7ddc 100644
--- a/usr/src/compat/freebsd/string.h
+++ b/usr/src/compat/bhyve/string.h
diff --git a/usr/src/compat/freebsd/strings.h b/usr/src/compat/bhyve/strings.h
index fa3539fb96..fa3539fb96 100644
--- a/usr/src/compat/freebsd/strings.h
+++ b/usr/src/compat/bhyve/strings.h
diff --git a/usr/src/compat/freebsd/sys/_cpuset.h b/usr/src/compat/bhyve/sys/_cpuset.h
index 286d26fc00..286d26fc00 100644
--- a/usr/src/compat/freebsd/sys/_cpuset.h
+++ b/usr/src/compat/bhyve/sys/_cpuset.h
diff --git a/usr/src/compat/freebsd/sys/_iovec.h b/usr/src/compat/bhyve/sys/_iovec.h
index b755ae7e21..b755ae7e21 100644
--- a/usr/src/compat/freebsd/sys/_iovec.h
+++ b/usr/src/compat/bhyve/sys/_iovec.h
diff --git a/usr/src/compat/freebsd/sys/_pthreadtypes.h b/usr/src/compat/bhyve/sys/_pthreadtypes.h
index d746da3712..d746da3712 100644
--- a/usr/src/compat/freebsd/sys/_pthreadtypes.h
+++ b/usr/src/compat/bhyve/sys/_pthreadtypes.h
diff --git a/usr/src/compat/freebsd/sys/_types.h b/usr/src/compat/bhyve/sys/_types.h
index 62c327d216..62c327d216 100644
--- a/usr/src/compat/freebsd/sys/_types.h
+++ b/usr/src/compat/bhyve/sys/_types.h
diff --git a/usr/src/compat/freebsd/sys/bus.h b/usr/src/compat/bhyve/sys/bus.h
index e3b5e0e69d..e3b5e0e69d 100644
--- a/usr/src/compat/freebsd/sys/bus.h
+++ b/usr/src/compat/bhyve/sys/bus.h
diff --git a/usr/src/compat/freebsd/sys/callout.h b/usr/src/compat/bhyve/sys/callout.h
index 11823e6321..11823e6321 100644
--- a/usr/src/compat/freebsd/sys/callout.h
+++ b/usr/src/compat/bhyve/sys/callout.h
diff --git a/usr/src/compat/freebsd/sys/cdefs.h b/usr/src/compat/bhyve/sys/cdefs.h
index 0b857437e3..0b857437e3 100644
--- a/usr/src/compat/freebsd/sys/cdefs.h
+++ b/usr/src/compat/bhyve/sys/cdefs.h
diff --git a/usr/src/compat/freebsd/sys/clock.h b/usr/src/compat/bhyve/sys/clock.h
index ebf7f171a3..ebf7f171a3 100644
--- a/usr/src/compat/freebsd/sys/clock.h
+++ b/usr/src/compat/bhyve/sys/clock.h
diff --git a/usr/src/compat/freebsd/sys/cpuset.h b/usr/src/compat/bhyve/sys/cpuset.h
index 626b323d7d..626b323d7d 100644
--- a/usr/src/compat/freebsd/sys/cpuset.h
+++ b/usr/src/compat/bhyve/sys/cpuset.h
diff --git a/usr/src/compat/freebsd/sys/disk.h b/usr/src/compat/bhyve/sys/disk.h
index c9bdc6a2d8..c9bdc6a2d8 100644
--- a/usr/src/compat/freebsd/sys/disk.h
+++ b/usr/src/compat/bhyve/sys/disk.h
diff --git a/usr/src/compat/freebsd/sys/endian.h b/usr/src/compat/bhyve/sys/endian.h
index 24ea02d251..24ea02d251 100644
--- a/usr/src/compat/freebsd/sys/endian.h
+++ b/usr/src/compat/bhyve/sys/endian.h
diff --git a/usr/src/compat/freebsd/sys/errno.h b/usr/src/compat/bhyve/sys/errno.h
index bd37f43065..bd37f43065 100644
--- a/usr/src/compat/freebsd/sys/errno.h
+++ b/usr/src/compat/bhyve/sys/errno.h
diff --git a/usr/src/compat/freebsd/sys/eventhandler.h b/usr/src/compat/bhyve/sys/eventhandler.h
index 133aa664f0..133aa664f0 100644
--- a/usr/src/compat/freebsd/sys/eventhandler.h
+++ b/usr/src/compat/bhyve/sys/eventhandler.h
diff --git a/usr/src/compat/freebsd/sys/fcntl.h b/usr/src/compat/bhyve/sys/fcntl.h
index 062a3b84ac..062a3b84ac 100644
--- a/usr/src/compat/freebsd/sys/fcntl.h
+++ b/usr/src/compat/bhyve/sys/fcntl.h
diff --git a/usr/src/compat/freebsd/sys/ioctl.h b/usr/src/compat/bhyve/sys/ioctl.h
index 72a46b8085..72a46b8085 100644
--- a/usr/src/compat/freebsd/sys/ioctl.h
+++ b/usr/src/compat/bhyve/sys/ioctl.h
diff --git a/usr/src/compat/freebsd/sys/kernel.h b/usr/src/compat/bhyve/sys/kernel.h
index adf96f40fc..adf96f40fc 100644
--- a/usr/src/compat/freebsd/sys/kernel.h
+++ b/usr/src/compat/bhyve/sys/kernel.h
diff --git a/usr/src/compat/freebsd/sys/ktr.h b/usr/src/compat/bhyve/sys/ktr.h
index 96c499ef18..96c499ef18 100644
--- a/usr/src/compat/freebsd/sys/ktr.h
+++ b/usr/src/compat/bhyve/sys/ktr.h
diff --git a/usr/src/compat/freebsd/sys/libkern.h b/usr/src/compat/bhyve/sys/libkern.h
index 94675a0d66..94675a0d66 100644
--- a/usr/src/compat/freebsd/sys/libkern.h
+++ b/usr/src/compat/bhyve/sys/libkern.h
diff --git a/usr/src/compat/freebsd/sys/limits.h b/usr/src/compat/bhyve/sys/limits.h
index 0e66319791..0e66319791 100644
--- a/usr/src/compat/freebsd/sys/limits.h
+++ b/usr/src/compat/bhyve/sys/limits.h
diff --git a/usr/src/compat/freebsd/sys/lock.h b/usr/src/compat/bhyve/sys/lock.h
index fd6021a87e..fd6021a87e 100644
--- a/usr/src/compat/freebsd/sys/lock.h
+++ b/usr/src/compat/bhyve/sys/lock.h
diff --git a/usr/src/compat/freebsd/sys/malloc.h b/usr/src/compat/bhyve/sys/malloc.h
index 341d57b807..341d57b807 100644
--- a/usr/src/compat/freebsd/sys/malloc.h
+++ b/usr/src/compat/bhyve/sys/malloc.h
diff --git a/usr/src/compat/freebsd/sys/module.h b/usr/src/compat/bhyve/sys/module.h
index 87b73e3fa3..87b73e3fa3 100644
--- a/usr/src/compat/freebsd/sys/module.h
+++ b/usr/src/compat/bhyve/sys/module.h
diff --git a/usr/src/compat/freebsd/sys/mutex.h b/usr/src/compat/bhyve/sys/mutex.h
index 9e588cb98a..9e588cb98a 100644
--- a/usr/src/compat/freebsd/sys/mutex.h
+++ b/usr/src/compat/bhyve/sys/mutex.h
diff --git a/usr/src/compat/freebsd/sys/param.h b/usr/src/compat/bhyve/sys/param.h
index 5ba21a2809..5ba21a2809 100644
--- a/usr/src/compat/freebsd/sys/param.h
+++ b/usr/src/compat/bhyve/sys/param.h
diff --git a/usr/src/compat/freebsd/sys/pcpu.h b/usr/src/compat/bhyve/sys/pcpu.h
index 1bad53c159..1bad53c159 100644
--- a/usr/src/compat/freebsd/sys/pcpu.h
+++ b/usr/src/compat/bhyve/sys/pcpu.h
diff --git a/usr/src/compat/freebsd/sys/sched.h b/usr/src/compat/bhyve/sys/sched.h
index b426ee757e..b426ee757e 100644
--- a/usr/src/compat/freebsd/sys/sched.h
+++ b/usr/src/compat/bhyve/sys/sched.h
diff --git a/usr/src/compat/freebsd/sys/sdt.h b/usr/src/compat/bhyve/sys/sdt.h
index 32d887c0d8..32d887c0d8 100644
--- a/usr/src/compat/freebsd/sys/sdt.h
+++ b/usr/src/compat/bhyve/sys/sdt.h
diff --git a/usr/src/compat/freebsd/sys/select.h b/usr/src/compat/bhyve/sys/select.h
index fcb40c23b1..fcb40c23b1 100644
--- a/usr/src/compat/freebsd/sys/select.h
+++ b/usr/src/compat/bhyve/sys/select.h
diff --git a/usr/src/compat/freebsd/sys/sglist.h b/usr/src/compat/bhyve/sys/sglist.h
index 519c67915f..519c67915f 100644
--- a/usr/src/compat/freebsd/sys/sglist.h
+++ b/usr/src/compat/bhyve/sys/sglist.h
diff --git a/usr/src/compat/freebsd/sys/smp.h b/usr/src/compat/bhyve/sys/smp.h
index 3d6413ce16..3d6413ce16 100644
--- a/usr/src/compat/freebsd/sys/smp.h
+++ b/usr/src/compat/bhyve/sys/smp.h
diff --git a/usr/src/compat/freebsd/sys/socket.h b/usr/src/compat/bhyve/sys/socket.h
index 3bf7a8f440..3bf7a8f440 100644
--- a/usr/src/compat/freebsd/sys/socket.h
+++ b/usr/src/compat/bhyve/sys/socket.h
diff --git a/usr/src/compat/freebsd/sys/sysctl.h b/usr/src/compat/bhyve/sys/sysctl.h
index 9f6a695e34..9f6a695e34 100644
--- a/usr/src/compat/freebsd/sys/sysctl.h
+++ b/usr/src/compat/bhyve/sys/sysctl.h
diff --git a/usr/src/compat/freebsd/sys/systm.h b/usr/src/compat/bhyve/sys/systm.h
index 43fa16d450..43fa16d450 100644
--- a/usr/src/compat/freebsd/sys/systm.h
+++ b/usr/src/compat/bhyve/sys/systm.h
diff --git a/usr/src/compat/freebsd/sys/time.h b/usr/src/compat/bhyve/sys/time.h
index 4e0fbfc02c..4e0fbfc02c 100644
--- a/usr/src/compat/freebsd/sys/time.h
+++ b/usr/src/compat/bhyve/sys/time.h
diff --git a/usr/src/compat/freebsd/sys/types.h b/usr/src/compat/bhyve/sys/types.h
index 63731da42e..63731da42e 100644
--- a/usr/src/compat/freebsd/sys/types.h
+++ b/usr/src/compat/bhyve/sys/types.h
diff --git a/usr/src/compat/freebsd/sys/uio.h b/usr/src/compat/bhyve/sys/uio.h
index 05c6f2a028..05c6f2a028 100644
--- a/usr/src/compat/freebsd/sys/uio.h
+++ b/usr/src/compat/bhyve/sys/uio.h
diff --git a/usr/src/compat/freebsd/termios.h b/usr/src/compat/bhyve/termios.h
index feaa705358..feaa705358 100644
--- a/usr/src/compat/freebsd/termios.h
+++ b/usr/src/compat/bhyve/termios.h
diff --git a/usr/src/compat/freebsd/unistd.h b/usr/src/compat/bhyve/unistd.h
index b4357e1da5..b4357e1da5 100644
--- a/usr/src/compat/freebsd/unistd.h
+++ b/usr/src/compat/bhyve/unistd.h
diff --git a/usr/src/compat/freebsd/uuid.h b/usr/src/compat/bhyve/uuid.h
index 72ef2c7787..72ef2c7787 100644
--- a/usr/src/compat/freebsd/uuid.h
+++ b/usr/src/compat/bhyve/uuid.h
diff --git a/usr/src/compat/freebsd/vm/vm.h b/usr/src/compat/bhyve/vm/vm.h
index f5bb7b6eb8..f5bb7b6eb8 100644
--- a/usr/src/compat/freebsd/vm/vm.h
+++ b/usr/src/compat/bhyve/vm/vm.h
diff --git a/usr/src/compat/freebsd/vm/vm_param.h b/usr/src/compat/bhyve/vm/vm_param.h
index fd76b62a37..fd76b62a37 100644
--- a/usr/src/compat/freebsd/vm/vm_param.h
+++ b/usr/src/compat/bhyve/vm/vm_param.h
diff --git a/usr/src/compat/freebsd/x86/_types.h b/usr/src/compat/bhyve/x86/_types.h
index 8bbae549d8..8bbae549d8 100644
--- a/usr/src/compat/freebsd/x86/_types.h
+++ b/usr/src/compat/bhyve/x86/_types.h
diff --git a/usr/src/compat/freebsd/x86/segments.h b/usr/src/compat/bhyve/x86/segments.h
index 11edc582b5..11edc582b5 100644
--- a/usr/src/compat/freebsd/x86/segments.h
+++ b/usr/src/compat/bhyve/x86/segments.h
diff --git a/usr/src/contrib/bhyve/README b/usr/src/contrib/bhyve/README
new file mode 100644
index 0000000000..415048d584
--- /dev/null
+++ b/usr/src/contrib/bhyve/README
@@ -0,0 +1,12 @@
+These are headers (and sources) contributed directly from FreeBSD which are
+required to build bhyve.  Updates are made to our copies here when corresponding
+updates to bhyve require them.
+
+With few exceptions (noted below) they are verbatim copies taken from the
+FreeBSD source tree.  Files requiring extensive modification are likely a
+candidate to exist as a shim in usr/src/compat/bhyve.  Conversely, if only a
+tiny modification is needed to make a file suitable for contrib, that trade-off
+is probably worthwhile (if done with a '#ifdef _FreeBSD_' guard for clarity).
+
+Files modified for compatibility:
+- isa/rtc.h
diff --git a/usr/contrib/freebsd/amd64/machine/_types.h b/usr/src/contrib/bhyve/amd64/machine/_types.h
index 59994352b5..59994352b5 100644
--- a/usr/contrib/freebsd/amd64/machine/_types.h
+++ b/usr/src/contrib/bhyve/amd64/machine/_types.h
diff --git a/usr/contrib/freebsd/amd64/machine/pmap.h b/usr/src/contrib/bhyve/amd64/machine/pmap.h
index a0b8ee37f2..a0b8ee37f2 100644
--- a/usr/contrib/freebsd/amd64/machine/pmap.h
+++ b/usr/src/contrib/bhyve/amd64/machine/pmap.h
diff --git a/usr/contrib/freebsd/amd64/machine/psl.h b/usr/src/contrib/bhyve/amd64/machine/psl.h
index c660bfbab0..c660bfbab0 100644
--- a/usr/contrib/freebsd/amd64/machine/psl.h
+++ b/usr/src/contrib/bhyve/amd64/machine/psl.h
diff --git a/usr/contrib/freebsd/amd64/machine/timerreg.h b/usr/src/contrib/bhyve/amd64/machine/timerreg.h
index bca7b4dd19..bca7b4dd19 100644
--- a/usr/contrib/freebsd/amd64/machine/timerreg.h
+++ b/usr/src/contrib/bhyve/amd64/machine/timerreg.h
diff --git a/usr/contrib/freebsd/amd64/machine/vm.h b/usr/src/contrib/bhyve/amd64/machine/vm.h
index 885c1607ea..885c1607ea 100644
--- a/usr/contrib/freebsd/amd64/machine/vm.h
+++ b/usr/src/contrib/bhyve/amd64/machine/vm.h
diff --git a/usr/contrib/freebsd/dev/acpica/acpi_hpet.h b/usr/src/contrib/bhyve/dev/acpica/acpi_hpet.h
index df817b7a2b..df817b7a2b 100644
--- a/usr/contrib/freebsd/dev/acpica/acpi_hpet.h
+++ b/usr/src/contrib/bhyve/dev/acpica/acpi_hpet.h
diff --git a/usr/contrib/freebsd/dev/ic/i8253reg.h b/usr/src/contrib/bhyve/dev/ic/i8253reg.h
index 47568b3436..47568b3436 100644
--- a/usr/contrib/freebsd/dev/ic/i8253reg.h
+++ b/usr/src/contrib/bhyve/dev/ic/i8253reg.h
diff --git a/usr/contrib/freebsd/dev/ic/i8259.h b/usr/src/contrib/bhyve/dev/ic/i8259.h
index be523c1df4..be523c1df4 100644
--- a/usr/contrib/freebsd/dev/ic/i8259.h
+++ b/usr/src/contrib/bhyve/dev/ic/i8259.h
diff --git a/usr/contrib/freebsd/dev/ic/ns16550.h b/usr/src/contrib/bhyve/dev/ic/ns16550.h
index 5e8f30e3e8..5e8f30e3e8 100644
--- a/usr/contrib/freebsd/dev/ic/ns16550.h
+++ b/usr/src/contrib/bhyve/dev/ic/ns16550.h
diff --git a/usr/contrib/freebsd/dev/io/iodev.h b/usr/src/contrib/bhyve/dev/io/iodev.h
index d040fcccf4..d040fcccf4 100644
--- a/usr/contrib/freebsd/dev/io/iodev.h
+++ b/usr/src/contrib/bhyve/dev/io/iodev.h
diff --git a/usr/contrib/freebsd/dev/mii/mii.h b/usr/src/contrib/bhyve/dev/mii/mii.h
index fa1ec84eaa..fa1ec84eaa 100644
--- a/usr/contrib/freebsd/dev/mii/mii.h
+++ b/usr/src/contrib/bhyve/dev/mii/mii.h
diff --git a/usr/contrib/freebsd/dev/nvme/nvme.h b/usr/src/contrib/bhyve/dev/nvme/nvme.h
index c7f6496426..c7f6496426 100644
--- a/usr/contrib/freebsd/dev/nvme/nvme.h
+++ b/usr/src/contrib/bhyve/dev/nvme/nvme.h
diff --git a/usr/contrib/freebsd/dev/pci/pcireg.h b/usr/src/contrib/bhyve/dev/pci/pcireg.h
index 32a569dbd4..32a569dbd4 100644
--- a/usr/contrib/freebsd/dev/pci/pcireg.h
+++ b/usr/src/contrib/bhyve/dev/pci/pcireg.h
diff --git a/usr/contrib/freebsd/dev/usb/controller/xhcireg.h b/usr/src/contrib/bhyve/dev/usb/controller/xhcireg.h
index 0e588ecba3..0e588ecba3 100644
--- a/usr/contrib/freebsd/dev/usb/controller/xhcireg.h
+++ b/usr/src/contrib/bhyve/dev/usb/controller/xhcireg.h
diff --git a/usr/contrib/freebsd/dev/usb/usb.h b/usr/src/contrib/bhyve/dev/usb/usb.h
index bcea2ac8bd..bcea2ac8bd 100644
--- a/usr/contrib/freebsd/dev/usb/usb.h
+++ b/usr/src/contrib/bhyve/dev/usb/usb.h
diff --git a/usr/contrib/freebsd/dev/usb/usb_endian.h b/usr/src/contrib/bhyve/dev/usb/usb_endian.h
index 0bbcb9bf82..0bbcb9bf82 100644
--- a/usr/contrib/freebsd/dev/usb/usb_endian.h
+++ b/usr/src/contrib/bhyve/dev/usb/usb_endian.h
diff --git a/usr/contrib/freebsd/dev/usb/usb_freebsd.h b/usr/src/contrib/bhyve/dev/usb/usb_freebsd.h
index 3bc9d2c1eb..3bc9d2c1eb 100644
--- a/usr/contrib/freebsd/dev/usb/usb_freebsd.h
+++ b/usr/src/contrib/bhyve/dev/usb/usb_freebsd.h
diff --git a/usr/contrib/freebsd/dev/usb/usbdi.h b/usr/src/contrib/bhyve/dev/usb/usbdi.h
index 202ad89fa7..202ad89fa7 100644
--- a/usr/contrib/freebsd/dev/usb/usbdi.h
+++ b/usr/src/contrib/bhyve/dev/usb/usbdi.h
diff --git a/usr/contrib/freebsd/isa/isareg.h b/usr/src/contrib/bhyve/isa/isareg.h
index e83e34674f..e83e34674f 100644
--- a/usr/contrib/freebsd/isa/isareg.h
+++ b/usr/src/contrib/bhyve/isa/isareg.h
diff --git a/usr/contrib/freebsd/isa/rtc.h b/usr/src/contrib/bhyve/isa/rtc.h
index bb964ddf6a..bb964ddf6a 100644
--- a/usr/contrib/freebsd/isa/rtc.h
+++ b/usr/src/contrib/bhyve/isa/rtc.h
diff --git a/usr/contrib/freebsd/lib/libutil/expand_number.c b/usr/src/contrib/bhyve/lib/libutil/expand_number.c
index f3b4da89f9..f3b4da89f9 100644
--- a/usr/contrib/freebsd/lib/libutil/expand_number.c
+++ b/usr/src/contrib/bhyve/lib/libutil/expand_number.c
diff --git a/usr/contrib/freebsd/lib/libutil/humanize_number.c b/usr/src/contrib/bhyve/lib/libutil/humanize_number.c
index 675a969aaa..675a969aaa 100644
--- a/usr/contrib/freebsd/lib/libutil/humanize_number.c
+++ b/usr/src/contrib/bhyve/lib/libutil/humanize_number.c
diff --git a/usr/contrib/freebsd/sys/ata.h b/usr/src/contrib/bhyve/sys/ata.h
index 223bd7b3eb..223bd7b3eb 100644
--- a/usr/contrib/freebsd/sys/ata.h
+++ b/usr/src/contrib/bhyve/sys/ata.h
diff --git a/usr/contrib/freebsd/sys/pciio.h b/usr/src/contrib/bhyve/sys/pciio.h
index d70bfbcf6f..d70bfbcf6f 100644
--- a/usr/contrib/freebsd/sys/pciio.h
+++ b/usr/src/contrib/bhyve/sys/pciio.h
diff --git a/usr/contrib/freebsd/sys/queue.h b/usr/src/contrib/bhyve/sys/queue.h
index f26c492af1..f26c492af1 100644
--- a/usr/contrib/freebsd/sys/queue.h
+++ b/usr/src/contrib/bhyve/sys/queue.h
diff --git a/usr/contrib/freebsd/sys/tree.h b/usr/src/contrib/bhyve/sys/tree.h
index 6b47e247bb..6b47e247bb 100644
--- a/usr/contrib/freebsd/sys/tree.h
+++ b/usr/src/contrib/bhyve/sys/tree.h
diff --git a/usr/contrib/freebsd/x86/apicreg.h b/usr/src/contrib/bhyve/x86/apicreg.h
index 24006e2733..24006e2733 100644
--- a/usr/contrib/freebsd/x86/apicreg.h
+++ b/usr/src/contrib/bhyve/x86/apicreg.h
diff --git a/usr/contrib/freebsd/x86/mptable.h b/usr/src/contrib/bhyve/x86/mptable.h
index 8f3c62a295..8f3c62a295 100644
--- a/usr/contrib/freebsd/x86/mptable.h
+++ b/usr/src/contrib/bhyve/x86/mptable.h
diff --git a/usr/contrib/freebsd/x86/psl.h b/usr/src/contrib/bhyve/x86/psl.h
index 6934b4feb7..6934b4feb7 100644
--- a/usr/contrib/freebsd/x86/psl.h
+++ b/usr/src/contrib/bhyve/x86/psl.h
diff --git a/usr/contrib/freebsd/x86/segments.h b/usr/src/contrib/bhyve/x86/segments.h
index 1b8c4a3c1c..1b8c4a3c1c 100644
--- a/usr/contrib/freebsd/x86/segments.h
+++ b/usr/src/contrib/bhyve/x86/segments.h
diff --git a/usr/contrib/freebsd/x86/specialreg.h b/usr/src/contrib/bhyve/x86/specialreg.h
index f528bad55c..f528bad55c 100644
--- a/usr/contrib/freebsd/x86/specialreg.h
+++ b/usr/src/contrib/bhyve/x86/specialreg.h
diff --git a/usr/src/lib/libvmm/Makefile.com b/usr/src/lib/libvmm/Makefile.com
index bef555aed3..ab0e7bd2b8 100644
--- a/usr/src/lib/libvmm/Makefile.com
+++ b/usr/src/lib/libvmm/Makefile.com
@@ -27,8 +27,8 @@ LIBS		= $(DYNLIB)
 # The FreeBSD compat and contrib headers need to be first in the search
 # path, hence we can't just append them to CPPFLAGS. So we assign CPPFLAGS
 # directly and pull in CPPFLAGS.master at the appropriate place.
-CPPFLAGS =	-I$(COMPAT)/freebsd -I$(CONTRIB)/freebsd \
-		-I$(COMPAT)/freebsd/amd64 -I$(CONTRIB)/freebsd/amd64 \
+CPPFLAGS =	-I$(COMPAT)/bhyve -I$(CONTRIB)/bhyve \
+		-I$(COMPAT)/bhyve/amd64 -I$(CONTRIB)/bhyve/amd64 \
 		$(CPPFLAGS.master) -I$(SRC)/uts/i86pc
 
 LDLIBS +=	-lc -lvmmapi
diff --git a/usr/src/lib/libvmmapi/Makefile.com b/usr/src/lib/libvmmapi/Makefile.com
index 1653e8619c..4c1c74d5a0 100644
--- a/usr/src/lib/libvmmapi/Makefile.com
+++ b/usr/src/lib/libvmmapi/Makefile.com
@@ -29,7 +29,7 @@ SRCDIR		= ../common
 
 LIBS		= $(DYNLIB)
 
-CPPFLAGS	= -I$(COMPAT)/freebsd -I$(CONTRIB)/freebsd \
+CPPFLAGS	= -I$(COMPAT)/bhyve -I$(CONTRIB)/bhyve \
 	$(CPPFLAGS.master) -I$(SRC)/uts/i86pc
 
 # not linted
@@ -41,7 +41,7 @@ LDLIBS		+= -lc
 
 all: $(LIBS)
 
-pics/%.o: $(CONTRIB)/freebsd/lib/libutil/%.c
+pics/%.o: $(CONTRIB)/bhyve/lib/libutil/%.c
 	$(COMPILE.c) -o $@ $<
 	$(POST_PROCESS_O)
 
diff --git a/usr/src/lib/libvmmapi/amd64/Makefile b/usr/src/lib/libvmmapi/amd64/Makefile
index b5cac1ffce..d777b9d456 100644
--- a/usr/src/lib/libvmmapi/amd64/Makefile
+++ b/usr/src/lib/libvmmapi/amd64/Makefile
@@ -16,6 +16,6 @@
 include ../Makefile.com
 include ../../Makefile.lib.64
 
-CPPFLAGS += -I$(COMPAT)/freebsd/amd64 -I$(CONTRIB)/freebsd/amd64
+CPPFLAGS += -I$(COMPAT)/bhyve/amd64 -I$(CONTRIB)/bhyve/amd64
 
 install: all $(ROOTLIBS64) $(ROOTLINKS64) $(ROOTLINT64)
diff --git a/usr/src/pkg/manifests/system-file-system-zfs-tests.mf b/usr/src/pkg/manifests/system-file-system-zfs-tests.mf
index 4e2b5f1add..d39248a2e4 100644
--- a/usr/src/pkg/manifests/system-file-system-zfs-tests.mf
+++ b/usr/src/pkg/manifests/system-file-system-zfs-tests.mf
@@ -22,6 +22,7 @@
 #
 # Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
 # Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+# Copyright 2020 Joyent, Inc.
 #
 
 set name=pkg.fmri value=pkg:/system/file-system/zfs/tests@$(PKGVERS)
@@ -53,6 +54,7 @@ $(i386_ONLY)file path=usr/bin/$(ARCH32)/zlook mode=0555
 $(i386_ONLY)file path=usr/bin/$(ARCH32)/ztest mode=0555
 file path=usr/bin/$(ARCH64)/zlook mode=0555
 file path=usr/bin/$(ARCH64)/ztest mode=0555
+file path=usr/bin/raidz_test mode=0555
 file path=usr/bin/zloop mode=0555
 file path=usr/include/sys/fs/zut.h
 file path=usr/lib/devfsadm/linkmod/SUNW_zut_link.so group=sys
diff --git a/usr/src/pkg/manifests/system-test-zfstest.mf b/usr/src/pkg/manifests/system-test-zfstest.mf
index 3b4bcc373e..0f05230a9b 100644
--- a/usr/src/pkg/manifests/system-test-zfstest.mf
+++ b/usr/src/pkg/manifests/system-test-zfstest.mf
@@ -143,6 +143,7 @@ dir path=opt/zfs-tests/tests/functional/poolversion
 dir path=opt/zfs-tests/tests/functional/privilege
 dir path=opt/zfs-tests/tests/functional/projectquota
 dir path=opt/zfs-tests/tests/functional/quota
+dir path=opt/zfs-tests/tests/functional/raidz
 dir path=opt/zfs-tests/tests/functional/redundancy
 dir path=opt/zfs-tests/tests/functional/refquota
 dir path=opt/zfs-tests/tests/functional/refreserv
@@ -2800,6 +2801,10 @@ file path=opt/zfs-tests/tests/functional/quota/quota_004_pos mode=0555
 file path=opt/zfs-tests/tests/functional/quota/quota_005_pos mode=0555
 file path=opt/zfs-tests/tests/functional/quota/quota_006_neg mode=0555
 file path=opt/zfs-tests/tests/functional/quota/setup mode=0555
+file path=opt/zfs-tests/tests/functional/raidz/cleanup mode=0555
+file path=opt/zfs-tests/tests/functional/raidz/raidz_001_neg mode=0555
+file path=opt/zfs-tests/tests/functional/raidz/raidz_002_pos mode=0555
+file path=opt/zfs-tests/tests/functional/raidz/setup mode=0555
 file path=opt/zfs-tests/tests/functional/redundancy/cleanup mode=0555
 file path=opt/zfs-tests/tests/functional/redundancy/redundancy.cfg mode=0444
 file path=opt/zfs-tests/tests/functional/redundancy/redundancy.kshlib \
diff --git a/usr/src/req.flg b/usr/src/req.flg
index aaf37ddd97..f1e17bbd52 100644
--- a/usr/src/req.flg
+++ b/usr/src/req.flg
@@ -34,5 +34,3 @@ echo_file usr/src/Makefile.master.64
 echo_file usr/src/Makefile.msg.targ
 echo_file usr/src/Makefile.psm
 echo_file usr/src/Makefile.psm.targ
-
-find_files "s.*" usr/contrib/freebsd
diff --git a/usr/src/test/zfs-tests/include/commands.cfg b/usr/src/test/zfs-tests/include/commands.cfg
index 050d6caba7..f9b0bdf7ac 100644
--- a/usr/src/test/zfs-tests/include/commands.cfg
+++ b/usr/src/test/zfs-tests/include/commands.cfg
@@ -11,7 +11,7 @@
 
 #
 # Copyright (c) 2016, 2018 by Delphix. All rights reserved.
-# Copyright 2019 Joyent, Inc.
+# Copyright 2020 Joyent, Inc.
 #
 
 #
@@ -89,6 +89,7 @@ export USR_BIN_FILES='awk
     ps
     pwd
     python
+    raidz_test
     readlink
     rm
     rmdir
diff --git a/usr/src/test/zfs-tests/runfiles/delphix.run b/usr/src/test/zfs-tests/runfiles/delphix.run
index f7dc35928e..656e274985 100644
--- a/usr/src/test/zfs-tests/runfiles/delphix.run
+++ b/usr/src/test/zfs-tests/runfiles/delphix.run
@@ -582,6 +582,9 @@ tests = ['projectid_001_pos', 'projectid_002_pos', 'projectid_003_pos',
 tests = ['quota_001_pos', 'quota_002_pos', 'quota_003_pos', 'quota_004_pos',
     'quota_005_pos', 'quota_006_neg']
 
+[/opt/zfs-tests/tests/functional/raidz]
+tests = ['raidz_001_neg', 'raidz_002_pos']
+
 [/opt/zfs-tests/tests/functional/redundancy]
 tests = ['redundancy_001_pos', 'redundancy_002_pos', 'redundancy_003_pos',
     'redundancy_004_neg']
diff --git a/usr/src/test/zfs-tests/runfiles/omnios.run b/usr/src/test/zfs-tests/runfiles/omnios.run
index 4dbfde3fb0..54848ba403 100644
--- a/usr/src/test/zfs-tests/runfiles/omnios.run
+++ b/usr/src/test/zfs-tests/runfiles/omnios.run
@@ -582,6 +582,9 @@ tests = ['projectid_001_pos', 'projectid_002_pos', 'projectid_003_pos',
 tests = ['quota_001_pos', 'quota_002_pos', 'quota_003_pos', 'quota_004_pos',
     'quota_005_pos', 'quota_006_neg']
 
+[/opt/zfs-tests/tests/functional/raidz]
+tests = ['raidz_001_neg', 'raidz_002_pos']
+
 [/opt/zfs-tests/tests/functional/redundancy]
 tests = ['redundancy_001_pos', 'redundancy_002_pos', 'redundancy_003_pos',
     'redundancy_004_neg']
diff --git a/usr/src/test/zfs-tests/runfiles/openindiana.run b/usr/src/test/zfs-tests/runfiles/openindiana.run
index 21473782c2..6ae3857a0a 100644
--- a/usr/src/test/zfs-tests/runfiles/openindiana.run
+++ b/usr/src/test/zfs-tests/runfiles/openindiana.run
@@ -582,6 +582,9 @@ tests = ['projectid_001_pos', 'projectid_002_pos', 'projectid_003_pos',
 tests = ['quota_001_pos', 'quota_002_pos', 'quota_003_pos', 'quota_004_pos',
     'quota_005_pos', 'quota_006_neg']
 
+[/opt/zfs-tests/tests/functional/raidz]
+tests = ['raidz_001_neg', 'raidz_002_pos']
+
 [/opt/zfs-tests/tests/functional/redundancy]
 tests = ['redundancy_001_pos', 'redundancy_002_pos', 'redundancy_003_pos',
     'redundancy_004_neg']
diff --git a/usr/src/test/zfs-tests/runfiles/smartos.run b/usr/src/test/zfs-tests/runfiles/smartos.run
index c8d6dc4783..47059b0b52 100644
--- a/usr/src/test/zfs-tests/runfiles/smartos.run
+++ b/usr/src/test/zfs-tests/runfiles/smartos.run
@@ -510,6 +510,9 @@ tests = ['projectid_001_pos', 'projectid_002_pos', 'projectid_003_pos',
 tests = ['quota_001_pos', 'quota_002_pos', 'quota_003_pos', 'quota_004_pos',
     'quota_005_pos', 'quota_006_neg']
 
+[/opt/zfs-tests/tests/functional/raidz]
+tests = ['raidz_001_neg', 'raidz_002_pos']
+
 [/opt/zfs-tests/tests/functional/refquota]
 tests = ['refquota_001_pos', 'refquota_002_pos', 'refquota_003_pos',
     'refquota_004_pos', 'refquota_005_pos', 'refquota_006_neg']
diff --git a/usr/src/test/zfs-tests/tests/functional/raidz/Makefile b/usr/src/test/zfs-tests/tests/functional/raidz/Makefile
new file mode 100644
index 0000000000..5d0bf4506a
--- /dev/null
+++ b/usr/src/test/zfs-tests/tests/functional/raidz/Makefile
@@ -0,0 +1,21 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2020 Joyent, Inc.
+#
+
+include $(SRC)/Makefile.master
+
+ROOTOPTPKG = $(ROOT)/opt/zfs-tests
+TARGETDIR = $(ROOTOPTPKG)/tests/functional/raidz
+
+include $(SRC)/test/zfs-tests/Makefile.com
diff --git a/usr/src/test/zfs-tests/tests/functional/raidz/cleanup.ksh b/usr/src/test/zfs-tests/tests/functional/raidz/cleanup.ksh
new file mode 100755
index 0000000000..c92c54c270
--- /dev/null
+++ b/usr/src/test/zfs-tests/tests/functional/raidz/cleanup.ksh
@@ -0,0 +1,30 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2016 by Gvozden Neskovic. All rights reserved.
+# Use is subject to license terms.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+# default_cleanup
diff --git a/usr/src/test/zfs-tests/tests/functional/raidz/raidz_001_neg.ksh b/usr/src/test/zfs-tests/tests/functional/raidz/raidz_001_neg.ksh
new file mode 100755
index 0000000000..4c105b9411
--- /dev/null
+++ b/usr/src/test/zfs-tests/tests/functional/raidz/raidz_001_neg.ksh
@@ -0,0 +1,38 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2016 by Gvozden Neskovic. All rights reserved.
+# Use is subject to license terms.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+#	Call the raidz_test tool with -T options to test the infrastructure.
+#	This option should make raidz_test to return non 0.
+#
+
+log_mustnot raidz_test -T
+
+log_pass "raidz_test detects errors as espected."
diff --git a/usr/src/test/zfs-tests/tests/functional/raidz/raidz_002_pos.ksh b/usr/src/test/zfs-tests/tests/functional/raidz/raidz_002_pos.ksh
new file mode 100755
index 0000000000..e238a881b0
--- /dev/null
+++ b/usr/src/test/zfs-tests/tests/functional/raidz/raidz_002_pos.ksh
@@ -0,0 +1,41 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2016 by Gvozden Neskovic. All rights reserved.
+# Use is subject to license terms.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+#	Call the raidz_test tool with -S to test all supported raidz
+#	implementations. This options will test several raidz block geometries
+#	and several zio parameters that affect raidz block layout. Data
+#	reconstruction performs all combinations of failed disks. Wall time
+#	is set to 5min, but actual runtime might be longer.
+#
+
+log_must raidz_test -S -t 300
+
+log_pass "raidz_test parameter sweep test succeeded."
diff --git a/usr/src/test/zfs-tests/tests/functional/raidz/setup.ksh b/usr/src/test/zfs-tests/tests/functional/raidz/setup.ksh
new file mode 100755
index 0000000000..4e155d24d5
--- /dev/null
+++ b/usr/src/test/zfs-tests/tests/functional/raidz/setup.ksh
@@ -0,0 +1,32 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2016 by Gvozden Neskovic. All rights reserved.
+# Use is subject to license terms.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+verify_runnable "global"
+
+log_pass
diff --git a/usr/src/uts/common/Makefile.files b/usr/src/uts/common/Makefile.files
index 720701371d..ff108e47e5 100644
--- a/usr/src/uts/common/Makefile.files
+++ b/usr/src/uts/common/Makefile.files
@@ -1434,6 +1434,8 @@ ZFS_COMMON_OBJS +=		\
 	vdev_missing.o		\
 	vdev_queue.o		\
 	vdev_raidz.o		\
+	vdev_raidz_math.o	\
+	vdev_raidz_math_scalar.o \
 	vdev_removal.o		\
 	vdev_root.o		\
 	vdev_trim.o		\
diff --git a/usr/src/uts/common/fs/zfs/abd.c b/usr/src/uts/common/fs/zfs/abd.c
index 66a7a49d73..e980c82eee 100644
--- a/usr/src/uts/common/fs/zfs/abd.c
+++ b/usr/src/uts/common/fs/zfs/abd.c
@@ -12,6 +12,7 @@
 /*
  * Copyright (c) 2014 by Chunwei Chen. All rights reserved.
  * Copyright (c) 2019 by Delphix. All rights reserved.
+ * Copyright 2020 Joyent, Inc.
  */
 
 /*
@@ -764,7 +765,8 @@ abd_iter_map(struct abd_iter *aiter)
 	} else {
 		size_t index = abd_iter_scatter_chunk_index(aiter);
 		offset = abd_iter_scatter_chunk_offset(aiter);
-		aiter->iter_mapsize = zfs_abd_chunk_size - offset;
+		aiter->iter_mapsize = MIN(zfs_abd_chunk_size - offset,
+		    aiter->iter_abd->abd_size - aiter->iter_pos);
 		paddr = aiter->iter_abd->abd_u.abd_scatter.abd_chunks[index];
 	}
 	aiter->iter_mapaddr = (char *)paddr + offset;
@@ -993,3 +995,180 @@ abd_cmp(abd_t *dabd, abd_t *sabd, size_t size)
 {
 	return (abd_iterate_func2(dabd, sabd, 0, 0, size, abd_cmp_cb, NULL));
 }
+
+/*
+ * Iterate over code ABDs and a data ABD and call @func_raidz_gen.
+ *
+ * @cabds          parity ABDs, must have equal size
+ * @dabd           data ABD. Can be NULL (in this case @dsize = 0)
+ * @func_raidz_gen should be implemented so that its behaviour
+ *                 is the same when taking linear and when taking scatter
+ */
+void
+abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
+    ssize_t csize, ssize_t dsize, const unsigned parity,
+    void (*func_raidz_gen)(void **, const void *, size_t, size_t))
+{
+	int i;
+	ssize_t len, dlen;
+	struct abd_iter caiters[3];
+	struct abd_iter daiter = {0};
+	void *caddrs[3];
+
+	ASSERT3U(parity, <=, 3);
+
+	for (i = 0; i < parity; i++)
+		abd_iter_init(&caiters[i], cabds[i]);
+
+	if (dabd)
+		abd_iter_init(&daiter, dabd);
+
+	ASSERT3S(dsize, >=, 0);
+
+#ifdef _KERNEL
+	kpreempt_disable();
+#endif
+	while (csize > 0) {
+		len = csize;
+
+		if (dabd && dsize > 0)
+			abd_iter_map(&daiter);
+
+		for (i = 0; i < parity; i++) {
+			abd_iter_map(&caiters[i]);
+			caddrs[i] = caiters[i].iter_mapaddr;
+		}
+
+		switch (parity) {
+			case 3:
+				len = MIN(caiters[2].iter_mapsize, len);
+				/* falls through */
+			case 2:
+				len = MIN(caiters[1].iter_mapsize, len);
+				/* falls through */
+			case 1:
+				len = MIN(caiters[0].iter_mapsize, len);
+		}
+
+		/* must be progressive */
+		ASSERT3S(len, >, 0);
+
+		if (dabd && dsize > 0) {
+			/* this needs precise iter.length */
+			len = MIN(daiter.iter_mapsize, len);
+			len = MIN(dsize, len);
+			dlen = len;
+		} else
+			dlen = 0;
+
+		/* must be progressive */
+		ASSERT3S(len, >, 0);
+		/*
+		 * The iterated function likely will not do well if each
+		 * segment except the last one is not multiple of 512 (raidz).
+		 */
+		ASSERT3U(((uint64_t)len & 511ULL), ==, 0);
+
+		func_raidz_gen(caddrs, daiter.iter_mapaddr, len, dlen);
+
+		for (i = parity-1; i >= 0; i--) {
+			abd_iter_unmap(&caiters[i]);
+			abd_iter_advance(&caiters[i], len);
+		}
+
+		if (dabd && dsize > 0) {
+			abd_iter_unmap(&daiter);
+			abd_iter_advance(&daiter, dlen);
+			dsize -= dlen;
+		}
+
+		csize -= len;
+
+		ASSERT3S(dsize, >=, 0);
+		ASSERT3S(csize, >=, 0);
+	}
+#ifdef _KERNEL
+	kpreempt_enable();
+#endif
+}
+
+/*
+ * Iterate over code ABDs and data reconstruction target ABDs and call
+ * @func_raidz_rec. Function maps at most 6 pages atomically.
+ *
+ * @cabds           parity ABDs, must have equal size
+ * @tabds           rec target ABDs, at most 3
+ * @tsize           size of data target columns
+ * @func_raidz_rec  expects syndrome data in target columns. Function
+ *                  reconstructs data and overwrites target columns.
+ */
+void
+abd_raidz_rec_iterate(abd_t **cabds, abd_t **tabds,
+    ssize_t tsize, const unsigned parity,
+    void (*func_raidz_rec)(void **t, const size_t tsize, void **c,
+    const unsigned *mul),
+    const unsigned *mul)
+{
+	int i;
+	ssize_t len;
+	struct abd_iter citers[3];
+	struct abd_iter xiters[3];
+	void *caddrs[3], *xaddrs[3];
+
+	ASSERT3U(parity, <=, 3);
+
+	for (i = 0; i < parity; i++) {
+		abd_iter_init(&citers[i], cabds[i]);
+		abd_iter_init(&xiters[i], tabds[i]);
+	}
+
+#ifdef _KERNEL
+	kpreempt_disable();
+#endif
+	while (tsize > 0) {
+
+		for (i = 0; i < parity; i++) {
+			abd_iter_map(&citers[i]);
+			abd_iter_map(&xiters[i]);
+			caddrs[i] = citers[i].iter_mapaddr;
+			xaddrs[i] = xiters[i].iter_mapaddr;
+		}
+
+		len = tsize;
+		switch (parity) {
+			case 3:
+				len = MIN(xiters[2].iter_mapsize, len);
+				len = MIN(citers[2].iter_mapsize, len);
+				/* falls through */
+			case 2:
+				len = MIN(xiters[1].iter_mapsize, len);
+				len = MIN(citers[1].iter_mapsize, len);
+				/* falls through */
+			case 1:
+				len = MIN(xiters[0].iter_mapsize, len);
+				len = MIN(citers[0].iter_mapsize, len);
+		}
+		/* must be progressive */
+		ASSERT3S(len, >, 0);
+		/*
+		 * The iterated function likely will not do well if each
+		 * segment except the last one is not multiple of 512 (raidz).
+		 */
+		ASSERT3U(((uint64_t)len & 511ULL), ==, 0);
+
+		func_raidz_rec(xaddrs, len, caddrs, mul);
+
+		for (i = parity-1; i >= 0; i--) {
+			abd_iter_unmap(&xiters[i]);
+			abd_iter_unmap(&citers[i]);
+			abd_iter_advance(&xiters[i], len);
+			abd_iter_advance(&citers[i], len);
+		}
+
+		tsize -= len;
+		ASSERT3S(tsize, >=, 0);
+	}
+#ifdef _KERNEL
+	kpreempt_enable();
+#endif
+}
diff --git a/usr/src/uts/common/fs/zfs/spa_misc.c b/usr/src/uts/common/fs/zfs/spa_misc.c
index 9dac4e2ddc..c9ceeb6873 100644
--- a/usr/src/uts/common/fs/zfs/spa_misc.c
+++ b/usr/src/uts/common/fs/zfs/spa_misc.c
@@ -44,6 +44,7 @@
 #include <sys/vdev_impl.h>
 #include <sys/vdev_initialize.h>
 #include <sys/vdev_trim.h>
+#include <sys/vdev_raidz.h>
 #include <sys/metaslab.h>
 #include <sys/uberblock_impl.h>
 #include <sys/txg.h>
@@ -2253,6 +2254,7 @@ spa_init(int mode)
 	zil_init();
 	vdev_cache_stat_init();
 	vdev_mirror_stat_init();
+	vdev_raidz_math_init();
 	zfs_prop_init();
 	zpool_prop_init();
 	zpool_feature_init();
@@ -2271,6 +2273,7 @@ spa_fini(void)
 
 	vdev_cache_stat_fini();
 	vdev_mirror_stat_fini();
+	vdev_raidz_math_fini();
 	zil_fini();
 	dmu_fini();
 	zio_fini();
diff --git a/usr/src/uts/common/fs/zfs/sys/abd.h b/usr/src/uts/common/fs/zfs/sys/abd.h
index 621635933e..23699c0420 100644
--- a/usr/src/uts/common/fs/zfs/sys/abd.h
+++ b/usr/src/uts/common/fs/zfs/sys/abd.h
@@ -103,6 +103,15 @@ int abd_cmp(abd_t *, abd_t *, size_t);
 int abd_cmp_buf_off(abd_t *, const void *, size_t, size_t);
 void abd_zero_off(abd_t *, size_t, size_t);
 
+void abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
+	ssize_t csize, ssize_t dsize, const unsigned parity,
+	void (*func_raidz_gen)(void **, const void *, size_t, size_t));
+void abd_raidz_rec_iterate(abd_t **cabds, abd_t **tabds,
+	ssize_t tsize, const unsigned parity,
+	void (*func_raidz_rec)(void **t, const size_t tsize, void **c,
+	const unsigned *mul),
+	const unsigned *mul);
+
 /*
  * Wrappers for calls with offsets of 0
  */
diff --git a/usr/src/uts/common/fs/zfs/sys/simd.h b/usr/src/uts/common/fs/zfs/sys/simd.h
new file mode 100644
index 0000000000..4494c7d02a
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/sys/simd.h
@@ -0,0 +1,40 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2020 Joyent, Inc.
+ */
+
+#ifndef _SIMD_H
+#define	_SIMD_H
+
+#if defined(__amd64__) || defined(__i386__)
+
+/* Temporararily disabled until subsequent work to turn this on. */
+#define	kfpu_allowed()		0
+#define	kfpu_initialize(tsk)	do {} while (0)
+#define	kfpu_begin()		do {} while (0)
+#define	kfpu_end()		do {} while (0)
+#define	kfpu_init()		(0)
+#define	kfpu_fini()		do {} while (0)
+
+#else
+
+/* Non-x86 CPUs currently always disallow kernel FPU support */
+#define	kfpu_allowed()		0
+#define	kfpu_initialize(tsk)	do {} while (0)
+#define	kfpu_begin()		do {} while (0)
+#define	kfpu_end()		do {} while (0)
+#define	kfpu_init()		(0)
+#define	kfpu_fini()		do {} while (0)
+#endif
+
+#endif /* _SIMD_H */
diff --git a/usr/src/uts/common/fs/zfs/sys/vdev_raidz.h b/usr/src/uts/common/fs/zfs/sys/vdev_raidz.h
new file mode 100644
index 0000000000..bf5c840139
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/sys/vdev_raidz.h
@@ -0,0 +1,65 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (C) 2016 Gvozden Neskovic <neskovic@compeng.uni-frankfurt.de>.
+ * Copyright 2020 Joyent, Inc.
+ */
+
+#ifndef _SYS_VDEV_RAIDZ_H
+#define	_SYS_VDEV_RAIDZ_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct zio;
+struct raidz_map;
+#if !defined(_KERNEL)
+struct kernel_param {};
+#endif
+
+/*
+ * vdev_raidz interface
+ */
+struct raidz_map *	vdev_raidz_map_alloc(struct zio *, uint64_t,
+    uint64_t, uint64_t);
+void		vdev_raidz_map_free(struct raidz_map *);
+void 		vdev_raidz_generate_parity(struct raidz_map *);
+int 		vdev_raidz_reconstruct(struct raidz_map *, const int *, int);
+
+/*
+ * vdev_raidz_math interface
+ */
+void	vdev_raidz_math_init(void);
+void	vdev_raidz_math_fini(void);
+const struct raidz_impl_ops *vdev_raidz_math_get_ops(void);
+int	vdev_raidz_math_generate(struct raidz_map *);
+int	vdev_raidz_math_reconstruct(struct raidz_map *, const int *,
+	    const int *, const int);
+int	vdev_raidz_impl_set(const char *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _SYS_VDEV_RAIDZ_H */
diff --git a/usr/src/uts/common/fs/zfs/sys/vdev_raidz_impl.h b/usr/src/uts/common/fs/zfs/sys/vdev_raidz_impl.h
new file mode 100644
index 0000000000..d8defc04ea
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/sys/vdev_raidz_impl.h
@@ -0,0 +1,351 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ */
+
+#ifndef _VDEV_RAIDZ_H
+#define	_VDEV_RAIDZ_H
+
+#include <sys/types.h>
+#include <sys/debug.h>
+#include <sys/kstat.h>
+#include <sys/abd.h>
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+#define	CODE_P		(0U)
+#define	CODE_Q		(1U)
+#define	CODE_R		(2U)
+
+#define	PARITY_P	(1U)
+#define	PARITY_PQ	(2U)
+#define	PARITY_PQR	(3U)
+
+#define	TARGET_X	(0U)
+#define	TARGET_Y	(1U)
+#define	TARGET_Z	(2U)
+
+/*
+ * Parity generation methods indexes
+ */
+enum raidz_math_gen_op {
+	RAIDZ_GEN_P = 0,
+	RAIDZ_GEN_PQ,
+	RAIDZ_GEN_PQR,
+	RAIDZ_GEN_NUM = 3
+};
+/*
+ * Data reconstruction methods indexes
+ */
+enum raidz_rec_op {
+	RAIDZ_REC_P = 0,
+	RAIDZ_REC_Q,
+	RAIDZ_REC_R,
+	RAIDZ_REC_PQ,
+	RAIDZ_REC_PR,
+	RAIDZ_REC_QR,
+	RAIDZ_REC_PQR,
+	RAIDZ_REC_NUM = 7
+};
+
+extern const char *raidz_gen_name[RAIDZ_GEN_NUM];
+extern const char *raidz_rec_name[RAIDZ_REC_NUM];
+
+/*
+ * Methods used to define raidz implementation
+ *
+ * @raidz_gen_f	Parity generation function
+ *     @par1	pointer to raidz_map
+ * @raidz_rec_f	Data reconstruction function
+ *     @par1	pointer to raidz_map
+ *     @par2	array of reconstruction targets
+ * @will_work_f Function returns TRUE if impl. is supported on the system
+ * @init_impl_f Function is called once on init
+ * @fini_impl_f Function is called once on fini
+ */
+typedef void		(*raidz_gen_f)(void *);
+typedef int		(*raidz_rec_f)(void *, const int *);
+typedef boolean_t	(*will_work_f)(void);
+typedef void		(*init_impl_f)(void);
+typedef void		(*fini_impl_f)(void);
+
+#define	RAIDZ_IMPL_NAME_MAX	(20)
+
+typedef struct raidz_impl_ops {
+	init_impl_f init;
+	fini_impl_f fini;
+	raidz_gen_f gen[RAIDZ_GEN_NUM];	/* Parity generate functions */
+	raidz_rec_f rec[RAIDZ_REC_NUM];	/* Data reconstruction functions */
+	will_work_f is_supported;	/* Support check function */
+	char name[RAIDZ_IMPL_NAME_MAX];	/* Name of the implementation */
+} raidz_impl_ops_t;
+
+typedef struct raidz_col {
+	size_t rc_devidx;		/* child device index for I/O */
+	size_t rc_offset;		/* device offset */
+	size_t rc_size;			/* I/O size */
+	abd_t *rc_abd;			/* I/O data */
+	void *rc_gdata;			/* used to store the "good" version */
+	int rc_error;			/* I/O error for this device */
+	unsigned int rc_tried;		/* Did we attempt this I/O column? */
+	unsigned int rc_skipped;	/* Did we skip this I/O column? */
+} raidz_col_t;
+
+typedef struct raidz_map {
+	size_t rm_cols;			/* Regular column count */
+	size_t rm_scols;		/* Count including skipped columns */
+	size_t rm_bigcols;		/* Number of oversized columns */
+	size_t rm_asize;		/* Actual total I/O size */
+	size_t rm_missingdata;		/* Count of missing data devices */
+	size_t rm_missingparity;	/* Count of missing parity devices */
+	size_t rm_firstdatacol;		/* First data column/parity count */
+	size_t rm_nskip;		/* Skipped sectors for padding */
+	size_t rm_skipstart;		/* Column index of padding start */
+	void *rm_abd_copy;		/* rm_asize-buffer of copied data */
+	size_t rm_reports;		/* # of referencing checksum reports */
+	unsigned int rm_freed;		/* map no longer has referencing ZIO */
+	unsigned int rm_ecksuminjected;	/* checksum error was injected */
+	const raidz_impl_ops_t *rm_ops;	/* RAIDZ math operations */
+	raidz_col_t rm_col[1];		/* Flexible array of I/O columns */
+} raidz_map_t;
+
+#define	RAIDZ_ORIGINAL_IMPL	(INT_MAX)
+
+extern const raidz_impl_ops_t vdev_raidz_scalar_impl;
+
+/*
+ * Commonly used raidz_map helpers
+ *
+ * raidz_parity		Returns parity of the RAIDZ block
+ * raidz_ncols		Returns number of columns the block spans
+ * raidz_nbigcols	Returns number of big columns columns
+ * raidz_col_p		Returns pointer to a column
+ * raidz_col_size	Returns size of a column
+ * raidz_big_size	Returns size of big columns
+ * raidz_short_size	Returns size of short columns
+ */
+#define	raidz_parity(rm)	((rm)->rm_firstdatacol)
+#define	raidz_ncols(rm)		((rm)->rm_cols)
+#define	raidz_nbigcols(rm)	((rm)->rm_bigcols)
+#define	raidz_col_p(rm, c)	((rm)->rm_col + (c))
+#define	raidz_col_size(rm, c)	((rm)->rm_col[c].rc_size)
+#define	raidz_big_size(rm)	(raidz_col_size(rm, CODE_P))
+#define	raidz_short_size(rm)	(raidz_col_size(rm, raidz_ncols(rm)-1))
+
+/*
+ * Macro defines an RAIDZ parity generation method
+ *
+ * @code	parity the function produce
+ * @impl	name of the implementation
+ */
+#define	_RAIDZ_GEN_WRAP(code, impl) 					\
+static void								\
+impl ## _gen_ ## code(void *rmp)					\
+{									\
+	raidz_map_t *rm = (raidz_map_t *) rmp;				\
+	raidz_generate_## code ## _impl(rm); 				\
+}
+
+/*
+ * Macro defines an RAIDZ data reconstruction method
+ *
+ * @code	parity the function produce
+ * @impl	name of the implementation
+ */
+#define	_RAIDZ_REC_WRAP(code, impl) 					\
+static int 								\
+impl ## _rec_ ## code(void *rmp, const int *tgtidx)			\
+{									\
+	raidz_map_t *rm = (raidz_map_t *) rmp;				\
+	return (raidz_reconstruct_## code ## _impl(rm, tgtidx));	\
+}
+
+/*
+ * Define all gen methods for an implementation
+ *
+ * @impl	name of the implementation
+ */
+#define	DEFINE_GEN_METHODS(impl)					\
+	_RAIDZ_GEN_WRAP(p, impl);					\
+	_RAIDZ_GEN_WRAP(pq, impl);					\
+	_RAIDZ_GEN_WRAP(pqr, impl)
+
+/*
+ * Define all rec functions for an implementation
+ *
+ * @impl	name of the implementation
+ */
+#define	DEFINE_REC_METHODS(impl)					\
+	_RAIDZ_REC_WRAP(p, impl);					\
+	_RAIDZ_REC_WRAP(q, impl);					\
+	_RAIDZ_REC_WRAP(r, impl);					\
+	_RAIDZ_REC_WRAP(pq, impl);					\
+	_RAIDZ_REC_WRAP(pr, impl);					\
+	_RAIDZ_REC_WRAP(qr, impl);					\
+	_RAIDZ_REC_WRAP(pqr, impl)
+
+#define	RAIDZ_GEN_METHODS(impl)						\
+{									\
+	[RAIDZ_GEN_P] = & impl ## _gen_p,				\
+	[RAIDZ_GEN_PQ] = & impl ## _gen_pq,				\
+	[RAIDZ_GEN_PQR] = & impl ## _gen_pqr				\
+}
+
+#define	RAIDZ_REC_METHODS(impl)						\
+{									\
+	[RAIDZ_REC_P] = & impl ## _rec_p,				\
+	[RAIDZ_REC_Q] = & impl ## _rec_q,				\
+	[RAIDZ_REC_R] = & impl ## _rec_r,				\
+	[RAIDZ_REC_PQ] = & impl ## _rec_pq,				\
+	[RAIDZ_REC_PR] = & impl ## _rec_pr,				\
+	[RAIDZ_REC_QR] = & impl ## _rec_qr,				\
+	[RAIDZ_REC_PQR] = & impl ## _rec_pqr				\
+}
+
+
+typedef struct raidz_impl_kstat {
+	uint64_t gen[RAIDZ_GEN_NUM];	/* gen method speed kiB/s */
+	uint64_t rec[RAIDZ_REC_NUM];	/* rec method speed kiB/s */
+} raidz_impl_kstat_t;
+
+/*
+ * Enumerate various multiplication constants
+ * used in reconstruction methods
+ */
+typedef enum raidz_mul_info {
+	/* Reconstruct Q */
+	MUL_Q_X		= 0,
+	/* Reconstruct R */
+	MUL_R_X		= 0,
+	/* Reconstruct PQ */
+	MUL_PQ_X	= 0,
+	MUL_PQ_Y	= 1,
+	/* Reconstruct PR */
+	MUL_PR_X	= 0,
+	MUL_PR_Y	= 1,
+	/* Reconstruct QR */
+	MUL_QR_XQ	= 0,
+	MUL_QR_X	= 1,
+	MUL_QR_YQ	= 2,
+	MUL_QR_Y	= 3,
+	/* Reconstruct PQR */
+	MUL_PQR_XP	= 0,
+	MUL_PQR_XQ	= 1,
+	MUL_PQR_XR	= 2,
+	MUL_PQR_YU	= 3,
+	MUL_PQR_YP	= 4,
+	MUL_PQR_YQ	= 5,
+
+	MUL_CNT		= 6
+} raidz_mul_info_t;
+
+/*
+ * Powers of 2 in the Galois field.
+ */
+extern const uint8_t vdev_raidz_pow2[256] __attribute__((aligned(256)));
+/* Logs of 2 in the Galois field defined above. */
+extern const uint8_t vdev_raidz_log2[256] __attribute__((aligned(256)));
+
+/*
+ * Multiply a given number by 2 raised to the given power.
+ */
+static inline uint8_t
+vdev_raidz_exp2(const uint8_t a, const unsigned exp)
+{
+	if (a == 0)
+		return (0);
+
+	return (vdev_raidz_pow2[(exp + (unsigned) vdev_raidz_log2[a]) % 255]);
+}
+
+/*
+ * Galois Field operations.
+ *
+ * gf_exp2	- computes 2 raised to the given power
+ * gf_exp2	- computes 4 raised to the given power
+ * gf_mul	- multiplication
+ * gf_div	- division
+ * gf_inv	- multiplicative inverse
+ */
+typedef unsigned gf_t;
+typedef unsigned gf_log_t;
+
+static inline gf_t
+gf_mul(const gf_t a, const gf_t b)
+{
+	gf_log_t logsum;
+
+	if (a == 0 || b == 0)
+		return (0);
+
+	logsum = (gf_log_t) vdev_raidz_log2[a] + (gf_log_t) vdev_raidz_log2[b];
+
+	return ((gf_t) vdev_raidz_pow2[logsum % 255]);
+}
+
+static inline gf_t
+gf_div(const gf_t  a, const gf_t b)
+{
+	gf_log_t logsum;
+
+	ASSERT3U(b, >, 0);
+	if (a == 0)
+		return (0);
+
+	logsum = (gf_log_t) 255 + (gf_log_t) vdev_raidz_log2[a] -
+	    (gf_log_t) vdev_raidz_log2[b];
+
+	return ((gf_t) vdev_raidz_pow2[logsum % 255]);
+}
+
+static inline gf_t
+gf_inv(const gf_t a)
+{
+	gf_log_t logsum;
+
+	ASSERT3U(a, >, 0);
+
+	logsum = (gf_log_t) 255 - (gf_log_t) vdev_raidz_log2[a];
+
+	return ((gf_t) vdev_raidz_pow2[logsum]);
+}
+
+static inline gf_t
+gf_exp2(gf_log_t exp)
+{
+	return (vdev_raidz_pow2[exp % 255]);
+}
+
+static inline gf_t
+gf_exp4(gf_log_t exp)
+{
+	ASSERT3U(exp, <=, 255);
+	return ((gf_t) vdev_raidz_pow2[(2 * exp) % 255]);
+}
+
+#ifdef  __cplusplus
+}
+#endif
+
+#endif /* _VDEV_RAIDZ_H */
diff --git a/usr/src/uts/common/fs/zfs/vdev_raidz.c b/usr/src/uts/common/fs/zfs/vdev_raidz.c
index 10772d5265..e4db03ce89 100644
--- a/usr/src/uts/common/fs/zfs/vdev_raidz.c
+++ b/usr/src/uts/common/fs/zfs/vdev_raidz.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
+ * Copyright (c) 2016 Gvozden Nešković. All rights reserved.
  * Copyright 2019 Joyent, Inc.
  * Copyright (c) 2014 Integros [integros.com]
  */
@@ -35,6 +36,8 @@
 #include <sys/abd.h>
 #include <sys/fs/zfs.h>
 #include <sys/fm/fs/zfs.h>
+#include <sys/vdev_raidz.h>
+#include <sys/vdev_raidz_impl.h>
 
 #ifdef ZFS_DEBUG
 #include <sys/vdev.h>	/* For vdev_xlate() in vdev_raidz_io_verify() */
@@ -98,7 +101,7 @@
  *	R = 4^n-1 * D_0 + 4^n-2 * D_1 + ... + 4^1 * D_n-2 + 4^0 * D_n-1
  *	  = ((...((D_0) * 4 + D_1) * 4 + ...) * 4 + D_n-2) * 4 + D_n-1
  *
- * We chose 1, 2, and 4 as our generators because 1 corresponds to the trival
+ * We chose 1, 2, and 4 as our generators because 1 corresponds to the trivial
  * XOR operation, and 2 and 4 can be computed quickly and generate linearly-
  * independent coefficients. (There are no additional coefficients that have
  * this property which is why the uncorrected Plank method breaks down.)
@@ -107,34 +110,6 @@
  * or in concert to recover missing data columns.
  */
 
-typedef struct raidz_col {
-	uint64_t rc_devidx;		/* child device index for I/O */
-	uint64_t rc_offset;		/* device offset */
-	uint64_t rc_size;		/* I/O size */
-	abd_t *rc_abd;			/* I/O data */
-	void *rc_gdata;			/* used to store the "good" version */
-	int rc_error;			/* I/O error for this device */
-	uint8_t rc_tried;		/* Did we attempt this I/O column? */
-	uint8_t rc_skipped;		/* Did we skip this I/O column? */
-} raidz_col_t;
-
-typedef struct raidz_map {
-	uint64_t rm_cols;		/* Regular column count */
-	uint64_t rm_scols;		/* Count including skipped columns */
-	uint64_t rm_bigcols;		/* Number of oversized columns */
-	uint64_t rm_asize;		/* Actual total I/O size */
-	uint64_t rm_missingdata;	/* Count of missing data devices */
-	uint64_t rm_missingparity;	/* Count of missing parity devices */
-	uint64_t rm_firstdatacol;	/* First data column/parity count */
-	uint64_t rm_nskip;		/* Skipped sectors for padding */
-	uint64_t rm_skipstart;		/* Column index of padding start */
-	abd_t *rm_abd_copy;		/* rm_asize-buffer of copied data */
-	uintptr_t rm_reports;		/* # of referencing checksum reports */
-	uint8_t	rm_freed;		/* map no longer has referencing ZIO */
-	uint8_t	rm_ecksuminjected;	/* checksum error was injected */
-	raidz_col_t rm_col[1];		/* Flexible array of I/O columns */
-} raidz_map_t;
-
 #define	VDEV_RAIDZ_P		0
 #define	VDEV_RAIDZ_Q		1
 #define	VDEV_RAIDZ_R		2
@@ -153,7 +128,7 @@ typedef struct raidz_map {
 	(mask) = (x) & 0x8080808080808080ULL; \
 	(mask) = ((mask) << 1) - ((mask) >> 7); \
 	(x) = (((x) << 1) & 0xfefefefefefefefeULL) ^ \
-	    ((mask) & 0x1d1d1d1d1d1d1d1d); \
+	    ((mask) & 0x1d1d1d1d1d1d1d1dULL); \
 }
 
 #define	VDEV_RAIDZ_64MUL_4(x, mask) \
@@ -164,104 +139,7 @@ typedef struct raidz_map {
 
 #define	VDEV_LABEL_OFFSET(x)	(x + VDEV_LABEL_START_SIZE)
 
-/*
- * Force reconstruction to use the general purpose method.
- */
-int vdev_raidz_default_to_general;
-
-/* Powers of 2 in the Galois field defined above. */
-static const uint8_t vdev_raidz_pow2[256] = {
-	0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
-	0x1d, 0x3a, 0x74, 0xe8, 0xcd, 0x87, 0x13, 0x26,
-	0x4c, 0x98, 0x2d, 0x5a, 0xb4, 0x75, 0xea, 0xc9,
-	0x8f, 0x03, 0x06, 0x0c, 0x18, 0x30, 0x60, 0xc0,
-	0x9d, 0x27, 0x4e, 0x9c, 0x25, 0x4a, 0x94, 0x35,
-	0x6a, 0xd4, 0xb5, 0x77, 0xee, 0xc1, 0x9f, 0x23,
-	0x46, 0x8c, 0x05, 0x0a, 0x14, 0x28, 0x50, 0xa0,
-	0x5d, 0xba, 0x69, 0xd2, 0xb9, 0x6f, 0xde, 0xa1,
-	0x5f, 0xbe, 0x61, 0xc2, 0x99, 0x2f, 0x5e, 0xbc,
-	0x65, 0xca, 0x89, 0x0f, 0x1e, 0x3c, 0x78, 0xf0,
-	0xfd, 0xe7, 0xd3, 0xbb, 0x6b, 0xd6, 0xb1, 0x7f,
-	0xfe, 0xe1, 0xdf, 0xa3, 0x5b, 0xb6, 0x71, 0xe2,
-	0xd9, 0xaf, 0x43, 0x86, 0x11, 0x22, 0x44, 0x88,
-	0x0d, 0x1a, 0x34, 0x68, 0xd0, 0xbd, 0x67, 0xce,
-	0x81, 0x1f, 0x3e, 0x7c, 0xf8, 0xed, 0xc7, 0x93,
-	0x3b, 0x76, 0xec, 0xc5, 0x97, 0x33, 0x66, 0xcc,
-	0x85, 0x17, 0x2e, 0x5c, 0xb8, 0x6d, 0xda, 0xa9,
-	0x4f, 0x9e, 0x21, 0x42, 0x84, 0x15, 0x2a, 0x54,
-	0xa8, 0x4d, 0x9a, 0x29, 0x52, 0xa4, 0x55, 0xaa,
-	0x49, 0x92, 0x39, 0x72, 0xe4, 0xd5, 0xb7, 0x73,
-	0xe6, 0xd1, 0xbf, 0x63, 0xc6, 0x91, 0x3f, 0x7e,
-	0xfc, 0xe5, 0xd7, 0xb3, 0x7b, 0xf6, 0xf1, 0xff,
-	0xe3, 0xdb, 0xab, 0x4b, 0x96, 0x31, 0x62, 0xc4,
-	0x95, 0x37, 0x6e, 0xdc, 0xa5, 0x57, 0xae, 0x41,
-	0x82, 0x19, 0x32, 0x64, 0xc8, 0x8d, 0x07, 0x0e,
-	0x1c, 0x38, 0x70, 0xe0, 0xdd, 0xa7, 0x53, 0xa6,
-	0x51, 0xa2, 0x59, 0xb2, 0x79, 0xf2, 0xf9, 0xef,
-	0xc3, 0x9b, 0x2b, 0x56, 0xac, 0x45, 0x8a, 0x09,
-	0x12, 0x24, 0x48, 0x90, 0x3d, 0x7a, 0xf4, 0xf5,
-	0xf7, 0xf3, 0xfb, 0xeb, 0xcb, 0x8b, 0x0b, 0x16,
-	0x2c, 0x58, 0xb0, 0x7d, 0xfa, 0xe9, 0xcf, 0x83,
-	0x1b, 0x36, 0x6c, 0xd8, 0xad, 0x47, 0x8e, 0x01
-};
-/* Logs of 2 in the Galois field defined above. */
-static const uint8_t vdev_raidz_log2[256] = {
-	0x00, 0x00, 0x01, 0x19, 0x02, 0x32, 0x1a, 0xc6,
-	0x03, 0xdf, 0x33, 0xee, 0x1b, 0x68, 0xc7, 0x4b,
-	0x04, 0x64, 0xe0, 0x0e, 0x34, 0x8d, 0xef, 0x81,
-	0x1c, 0xc1, 0x69, 0xf8, 0xc8, 0x08, 0x4c, 0x71,
-	0x05, 0x8a, 0x65, 0x2f, 0xe1, 0x24, 0x0f, 0x21,
-	0x35, 0x93, 0x8e, 0xda, 0xf0, 0x12, 0x82, 0x45,
-	0x1d, 0xb5, 0xc2, 0x7d, 0x6a, 0x27, 0xf9, 0xb9,
-	0xc9, 0x9a, 0x09, 0x78, 0x4d, 0xe4, 0x72, 0xa6,
-	0x06, 0xbf, 0x8b, 0x62, 0x66, 0xdd, 0x30, 0xfd,
-	0xe2, 0x98, 0x25, 0xb3, 0x10, 0x91, 0x22, 0x88,
-	0x36, 0xd0, 0x94, 0xce, 0x8f, 0x96, 0xdb, 0xbd,
-	0xf1, 0xd2, 0x13, 0x5c, 0x83, 0x38, 0x46, 0x40,
-	0x1e, 0x42, 0xb6, 0xa3, 0xc3, 0x48, 0x7e, 0x6e,
-	0x6b, 0x3a, 0x28, 0x54, 0xfa, 0x85, 0xba, 0x3d,
-	0xca, 0x5e, 0x9b, 0x9f, 0x0a, 0x15, 0x79, 0x2b,
-	0x4e, 0xd4, 0xe5, 0xac, 0x73, 0xf3, 0xa7, 0x57,
-	0x07, 0x70, 0xc0, 0xf7, 0x8c, 0x80, 0x63, 0x0d,
-	0x67, 0x4a, 0xde, 0xed, 0x31, 0xc5, 0xfe, 0x18,
-	0xe3, 0xa5, 0x99, 0x77, 0x26, 0xb8, 0xb4, 0x7c,
-	0x11, 0x44, 0x92, 0xd9, 0x23, 0x20, 0x89, 0x2e,
-	0x37, 0x3f, 0xd1, 0x5b, 0x95, 0xbc, 0xcf, 0xcd,
-	0x90, 0x87, 0x97, 0xb2, 0xdc, 0xfc, 0xbe, 0x61,
-	0xf2, 0x56, 0xd3, 0xab, 0x14, 0x2a, 0x5d, 0x9e,
-	0x84, 0x3c, 0x39, 0x53, 0x47, 0x6d, 0x41, 0xa2,
-	0x1f, 0x2d, 0x43, 0xd8, 0xb7, 0x7b, 0xa4, 0x76,
-	0xc4, 0x17, 0x49, 0xec, 0x7f, 0x0c, 0x6f, 0xf6,
-	0x6c, 0xa1, 0x3b, 0x52, 0x29, 0x9d, 0x55, 0xaa,
-	0xfb, 0x60, 0x86, 0xb1, 0xbb, 0xcc, 0x3e, 0x5a,
-	0xcb, 0x59, 0x5f, 0xb0, 0x9c, 0xa9, 0xa0, 0x51,
-	0x0b, 0xf5, 0x16, 0xeb, 0x7a, 0x75, 0x2c, 0xd7,
-	0x4f, 0xae, 0xd5, 0xe9, 0xe6, 0xe7, 0xad, 0xe8,
-	0x74, 0xd6, 0xf4, 0xea, 0xa8, 0x50, 0x58, 0xaf,
-};
-
-static void vdev_raidz_generate_parity(raidz_map_t *rm);
-
-/*
- * Multiply a given number by 2 raised to the given power.
- */
-static uint8_t
-vdev_raidz_exp2(uint_t a, int exp)
-{
-	if (a == 0)
-		return (0);
-
-	ASSERT(exp >= 0);
-	ASSERT(vdev_raidz_log2[a] > 0 || a == 1);
-
-	exp += vdev_raidz_log2[a];
-	if (exp > 255)
-		exp -= 255;
-
-	return (vdev_raidz_pow2[exp]);
-}
-
-static void
+void
 vdev_raidz_map_free(raidz_map_t *rm)
 {
 	int c;
@@ -271,7 +149,6 @@ vdev_raidz_map_free(raidz_map_t *rm)
 
 		if (rm->rm_col[c].rc_gdata != NULL)
 			abd_free(rm->rm_col[c].rc_gdata);
-
 	}
 
 	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++)
@@ -311,7 +188,7 @@ static void
 vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const abd_t *good_data)
 {
 	raidz_map_t *rm = zcr->zcr_cbdata;
-	size_t c = zcr->zcr_cbinfo;
+	const size_t c = zcr->zcr_cbinfo;
 	size_t x, offset;
 
 	const abd_t *good = NULL;
@@ -459,19 +336,19 @@ static const zio_vsd_ops_t vdev_raidz_vsd_ops = {
  * Divides the IO evenly across all child vdevs; usually, dcols is
  * the number of children in the target vdev.
  */
-static raidz_map_t *
-vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
-    uint64_t unit_shift, uint64_t dcols, uint64_t nparity)
+raidz_map_t *
+vdev_raidz_map_alloc(zio_t *zio, uint64_t ashift, uint64_t dcols,
+    uint64_t nparity)
 {
 	raidz_map_t *rm;
 	/* The starting RAIDZ (parent) vdev sector of the block. */
-	uint64_t b = offset >> unit_shift;
+	uint64_t b = zio->io_offset >> ashift;
 	/* The zio's size in units of the vdev's minimum sector size. */
-	uint64_t s = size >> unit_shift;
+	uint64_t s = zio->io_size >> ashift;
 	/* The first column for this stripe. */
 	uint64_t f = b % dcols;
 	/* The starting byte offset on each child vdev. */
-	uint64_t o = (b / dcols) << unit_shift;
+	uint64_t o = (b / dcols) << ashift;
 	uint64_t q, r, c, bc, col, acols, scols, coff, devidx, asize, tot;
 	uint64_t off = 0;
 
@@ -530,7 +407,7 @@ vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
 		coff = o;
 		if (col >= dcols) {
 			col -= dcols;
-			coff += 1ULL << unit_shift;
+			coff += 1ULL << ashift;
 		}
 		rm->rm_col[c].rc_devidx = col;
 		rm->rm_col[c].rc_offset = coff;
@@ -543,29 +420,29 @@ vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
 		if (c >= acols)
 			rm->rm_col[c].rc_size = 0;
 		else if (c < bc)
-			rm->rm_col[c].rc_size = (q + 1) << unit_shift;
+			rm->rm_col[c].rc_size = (q + 1) << ashift;
 		else
-			rm->rm_col[c].rc_size = q << unit_shift;
+			rm->rm_col[c].rc_size = q << ashift;
 
 		asize += rm->rm_col[c].rc_size;
 	}
 
-	ASSERT3U(asize, ==, tot << unit_shift);
-	rm->rm_asize = roundup(asize, (nparity + 1) << unit_shift);
+	ASSERT3U(asize, ==, tot << ashift);
+	rm->rm_asize = roundup(asize, (nparity + 1) << ashift);
 	rm->rm_nskip = roundup(tot, nparity + 1) - tot;
-	ASSERT3U(rm->rm_asize - asize, ==, rm->rm_nskip << unit_shift);
+	ASSERT3U(rm->rm_asize - asize, ==, rm->rm_nskip << ashift);
 	ASSERT3U(rm->rm_nskip, <=, nparity);
 
 	for (c = 0; c < rm->rm_firstdatacol; c++)
 		rm->rm_col[c].rc_abd =
 		    abd_alloc_linear(rm->rm_col[c].rc_size, B_FALSE);
 
-	rm->rm_col[c].rc_abd = abd_get_offset_size(abd, 0,
+	rm->rm_col[c].rc_abd = abd_get_offset_size(zio->io_abd, 0,
 	    rm->rm_col[c].rc_size);
 	off = rm->rm_col[c].rc_size;
 
 	for (c = c + 1; c < acols; c++) {
-		rm->rm_col[c].rc_abd = abd_get_offset_size(abd, off,
+		rm->rm_col[c].rc_abd = abd_get_offset_size(zio->io_abd, off,
 		    rm->rm_col[c].rc_size);
 		off += rm->rm_col[c].rc_size;
 	}
@@ -573,7 +450,7 @@ vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
 	/*
 	 * If all data stored spans all columns, there's a danger that parity
 	 * will always be on the same device and, since parity isn't read
-	 * during normal operation, that that device's I/O bandwidth won't be
+	 * during normal operation, that device's I/O bandwidth won't be
 	 * used effectively. We therefore switch the parity every 1MB.
 	 *
 	 * ... at least that was, ostensibly, the theory. As a practical
@@ -593,7 +470,7 @@ vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
 	ASSERT(rm->rm_cols >= 2);
 	ASSERT(rm->rm_col[0].rc_size == rm->rm_col[1].rc_size);
 
-	if (rm->rm_firstdatacol == 1 && (offset & (1ULL << 20))) {
+	if (rm->rm_firstdatacol == 1 && (zio->io_offset & (1ULL << 20))) {
 		devidx = rm->rm_col[0].rc_devidx;
 		o = rm->rm_col[0].rc_offset;
 		rm->rm_col[0].rc_devidx = rm->rm_col[1].rc_devidx;
@@ -605,6 +482,9 @@ vdev_raidz_map_alloc(abd_t *abd, uint64_t size, uint64_t offset,
 			rm->rm_skipstart = 1;
 	}
 
+	/* init RAIDZ parity ops */
+	rm->rm_ops = vdev_raidz_math_get_ops();
+
 	return (rm);
 }
 
@@ -681,7 +561,6 @@ vdev_raidz_generate_parity_p(raidz_map_t *rm)
 		p = abd_to_buf(rm->rm_col[VDEV_RAIDZ_P].rc_abd);
 
 		if (c == rm->rm_firstdatacol) {
-			ASSERT3U(src->abd_size, >=, rm->rm_col[c].rc_size);
 			abd_copy_to_buf_off(p, src, 0, rm->rm_col[c].rc_size);
 		} else {
 			struct pqr_struct pqr = { p, NULL, NULL };
@@ -793,9 +672,13 @@ vdev_raidz_generate_parity_pqr(raidz_map_t *rm)
  * Generate RAID parity in the first virtual columns according to the number of
  * parity columns available.
  */
-static void
+void
 vdev_raidz_generate_parity(raidz_map_t *rm)
 {
+	/* Generate using the new math implementation */
+	if (vdev_raidz_math_generate(rm) != RAIDZ_ORIGINAL_IMPL)
+		return;
+
 	switch (rm->rm_firstdatacol) {
 	case 1:
 		vdev_raidz_generate_parity_p(rm);
@@ -873,8 +756,8 @@ vdev_raidz_reconst_q_post_func(void *buf, size_t size, void *private)
 	int cnt = size / sizeof (dst[0]);
 
 	for (int i = 0; i < cnt; i++, dst++, rq->q++) {
-		*dst ^= *rq->q;
 
+		*dst ^= *rq->q;
 		int j;
 		uint8_t *b;
 		for (j = 0, b = (uint8_t *)dst; j < 8; j++, b++) {
@@ -1159,9 +1042,12 @@ vdev_raidz_reconstruct_pq(raidz_map_t *rm, int *tgts, int ntgts)
  *           ~~                               ~~
  *           __                               __
  *           |  1   1   1   1   1   1   1   1  |
+ *           | 128  64  32  16  8   4   2   1  |
  *           |  19 205 116  29  64  16  4   1  |
  *           |  1   0   0   0   0   0   0   0  |
- *  (V|I)' = |  0   0   0   1   0   0   0   0  |
+ *           |  0   1   0   0   0   0   0   0  |
+ *  (V|I)' = |  0   0   1   0   0   0   0   0  |
+ *           |  0   0   0   1   0   0   0   0  |
  *           |  0   0   0   0   1   0   0   0  |
  *           |  0   0   0   0   0   1   0   0  |
  *           |  0   0   0   0   0   0   1   0  |
@@ -1385,8 +1271,8 @@ vdev_raidz_matrix_reconstruct(raidz_map_t *rm, int n, int nmissing,
 	int i, j, x, cc, c;
 	uint8_t *src;
 	uint64_t ccount;
-	uint8_t *dst[VDEV_RAIDZ_MAXPARITY];
-	uint64_t dcount[VDEV_RAIDZ_MAXPARITY];
+	uint8_t *dst[VDEV_RAIDZ_MAXPARITY] = { NULL };
+	uint64_t dcount[VDEV_RAIDZ_MAXPARITY] = { 0 };
 	uint8_t log = 0;
 	uint8_t val;
 	int ll;
@@ -1595,12 +1481,12 @@ vdev_raidz_reconstruct_general(raidz_map_t *rm, int *tgts, int ntgts)
 	return (code);
 }
 
-static int
-vdev_raidz_reconstruct(raidz_map_t *rm, int *t, int nt)
+int
+vdev_raidz_reconstruct(raidz_map_t *rm, const int *t, int nt)
 {
 	int tgts[VDEV_RAIDZ_MAXPARITY], *dt;
 	int ntgts;
-	int i, c;
+	int i, c, ret;
 	int code;
 	int nbadparity, nbaddata;
 	int parity_valid[VDEV_RAIDZ_MAXPARITY];
@@ -1638,34 +1524,37 @@ vdev_raidz_reconstruct(raidz_map_t *rm, int *t, int nt)
 
 	dt = &tgts[nbadparity];
 
+	/* Reconstruct using the new math implementation */
+	ret = vdev_raidz_math_reconstruct(rm, parity_valid, dt, nbaddata);
+	if (ret != RAIDZ_ORIGINAL_IMPL)
+		return (ret);
+
 	/*
 	 * See if we can use any of our optimized reconstruction routines.
 	 */
-	if (!vdev_raidz_default_to_general) {
-		switch (nbaddata) {
-		case 1:
-			if (parity_valid[VDEV_RAIDZ_P])
-				return (vdev_raidz_reconstruct_p(rm, dt, 1));
+	switch (nbaddata) {
+	case 1:
+		if (parity_valid[VDEV_RAIDZ_P])
+			return (vdev_raidz_reconstruct_p(rm, dt, 1));
 
-			ASSERT(rm->rm_firstdatacol > 1);
+		ASSERT(rm->rm_firstdatacol > 1);
 
-			if (parity_valid[VDEV_RAIDZ_Q])
-				return (vdev_raidz_reconstruct_q(rm, dt, 1));
+		if (parity_valid[VDEV_RAIDZ_Q])
+			return (vdev_raidz_reconstruct_q(rm, dt, 1));
 
-			ASSERT(rm->rm_firstdatacol > 2);
-			break;
+		ASSERT(rm->rm_firstdatacol > 2);
+		break;
 
-		case 2:
-			ASSERT(rm->rm_firstdatacol > 1);
+	case 2:
+		ASSERT(rm->rm_firstdatacol > 1);
 
-			if (parity_valid[VDEV_RAIDZ_P] &&
-			    parity_valid[VDEV_RAIDZ_Q])
-				return (vdev_raidz_reconstruct_pq(rm, dt, 2));
+		if (parity_valid[VDEV_RAIDZ_P] &&
+		    parity_valid[VDEV_RAIDZ_Q])
+			return (vdev_raidz_reconstruct_pq(rm, dt, 2));
 
-			ASSERT(rm->rm_firstdatacol > 2);
+		ASSERT(rm->rm_firstdatacol > 2);
 
-			break;
-		}
+		break;
 	}
 
 	code = vdev_raidz_reconstruct_general(rm, tgts, ntgts);
@@ -1821,11 +1710,16 @@ vdev_raidz_dumpio(vdev_t *vd, caddr_t data, size_t size,
 	 * treat the on-disk format as if the only blocks are the complete 128
 	 * KB size.
 	 */
-	abd_t *abd = abd_get_from_buf(data - (offset - origoffset),
+
+	/* First, fake a zio for vdev_raidz_map_alloc. */
+	zio_t *zio = kmem_zalloc(sizeof (zio_t), KM_SLEEP);
+	zio->io_offset = origoffset;
+	zio->io_size = SPA_OLD_MAXBLOCKSIZE;
+	zio->io_abd = abd_get_from_buf(data - (offset - origoffset),
 	    SPA_OLD_MAXBLOCKSIZE);
-	rm = vdev_raidz_map_alloc(abd,
-	    SPA_OLD_MAXBLOCKSIZE, origoffset, tvd->vdev_ashift,
-	    vd->vdev_children, vd->vdev_nparity);
+
+	rm = vdev_raidz_map_alloc(zio, tvd->vdev_ashift, vd->vdev_children,
+	    vd->vdev_nparity);
 
 	coloffset = origoffset;
 
@@ -1874,7 +1768,9 @@ vdev_raidz_dumpio(vdev_t *vd, caddr_t data, size_t size,
 	}
 
 	vdev_raidz_map_free(rm);
-	abd_put(abd);
+	abd_put(zio->io_abd);
+	kmem_free(zio, sizeof (zio_t));
+
 #endif	/* KERNEL */
 
 	return (err);
@@ -1965,8 +1861,7 @@ vdev_raidz_io_start(zio_t *zio)
 	raidz_col_t *rc;
 	int c, i;
 
-	rm = vdev_raidz_map_alloc(zio->io_abd, zio->io_size, zio->io_offset,
-	    tvd->vdev_ashift, vd->vdev_children,
+	rm = vdev_raidz_map_alloc(zio, tvd->vdev_ashift, vd->vdev_children,
 	    vd->vdev_nparity);
 
 	zio->io_vsd = rm;
@@ -2141,11 +2036,6 @@ raidz_parity_verify(zio_t *zio, raidz_map_t *rm)
 	return (ret);
 }
 
-/*
- * Keep statistics on all the ways that we used parity to correct data.
- */
-static uint64_t raidz_corrected[1 << VDEV_RAIDZ_MAXPARITY];
-
 static int
 vdev_raidz_worst_error(raidz_map_t *rm)
 {
@@ -2251,7 +2141,6 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 			 */
 			code = vdev_raidz_reconstruct(rm, tgts, n);
 			if (raidz_checksum_verify(zio) == 0) {
-				atomic_inc_64(&raidz_corrected[code]);
 
 				for (i = 0; i < n; i++) {
 					c = tgts[i];
@@ -2466,8 +2355,6 @@ vdev_raidz_io_done(zio_t *zio)
 			code = vdev_raidz_reconstruct(rm, tgts, n);
 
 			if (raidz_checksum_verify(zio) == 0) {
-				atomic_inc_64(&raidz_corrected[code]);
-
 				/*
 				 * If we read more parity disks than were used
 				 * for reconstruction, confirm that the other
@@ -2620,7 +2507,7 @@ vdev_raidz_state_change(vdev_t *vd, int faulted, int degraded)
 /*
  * Determine if any portion of the provided block resides on a child vdev
  * with a dirty DTL and therefore needs to be resilvered.  The function
- * assumes that at least one DTL is dirty which imples that full stripe
+ * assumes that at least one DTL is dirty which implies that full stripe
  * width blocks must be resilvered.
  */
 static boolean_t
diff --git a/usr/src/uts/common/fs/zfs/vdev_raidz_math.c b/usr/src/uts/common/fs/zfs/vdev_raidz_math.c
new file mode 100644
index 0000000000..2a1dac33c5
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/vdev_raidz_math.c
@@ -0,0 +1,571 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/types.h>
+#include <sys/zio.h>
+#include <sys/debug.h>
+#include <sys/zfs_debug.h>
+#include <sys/vdev_raidz.h>
+#include <sys/vdev_raidz_impl.h>
+#include <sys/simd.h>
+
+#ifndef isspace
+#define	isspace(c)	((c) == ' ' || (c) == '\t' || (c) == '\n' || \
+			(c) == '\r' || (c) == '\f' || (c) == '\013')
+#endif
+
+extern boolean_t raidz_will_scalar_work(void);
+
+/* Opaque implementation with NULL methods to represent original methods */
+static const raidz_impl_ops_t vdev_raidz_original_impl = {
+	.name = "original",
+	.is_supported = raidz_will_scalar_work,
+};
+
+/* RAIDZ parity op that contain the fastest methods */
+static raidz_impl_ops_t vdev_raidz_fastest_impl = {
+	.name = "fastest"
+};
+
+/* All compiled in implementations */
+const raidz_impl_ops_t *raidz_all_maths[] = {
+	&vdev_raidz_original_impl,
+	&vdev_raidz_scalar_impl,
+};
+
+/* Indicate that benchmark has been completed */
+static boolean_t raidz_math_initialized = B_FALSE;
+
+/* Select raidz implementation */
+#define	IMPL_FASTEST	(UINT32_MAX)
+#define	IMPL_CYCLE	(UINT32_MAX - 1)
+#define	IMPL_ORIGINAL	(0)
+#define	IMPL_SCALAR	(1)
+
+#define	RAIDZ_IMPL_READ(i)	(*(volatile uint32_t *) &(i))
+
+static uint32_t zfs_vdev_raidz_impl = IMPL_SCALAR;
+static uint32_t user_sel_impl = IMPL_FASTEST;
+
+/* Hold all supported implementations */
+static size_t raidz_supp_impl_cnt = 0;
+static raidz_impl_ops_t *raidz_supp_impl[ARRAY_SIZE(raidz_all_maths)];
+
+#if defined(_KERNEL)
+/*
+ * kstats values for supported implementations
+ * Values represent per disk throughput of 8 disk+parity raidz vdev [B/s]
+ *
+ * PORTING NOTE:
+ * On illumos this is not a kstat. OpenZFS uses their home-grown kstat code
+ * which implements a free-form kstat using additional functionality that does
+ * not exist in illumos. Because there are no software consumers of this
+ * information, we omit a kstat API. If an administrator needs to see this
+ * data for some reason, they can use mdb.
+ *
+ * The format of the kstat data on OpenZFS would be a "header" that looks like
+ * this (a column for each entry in the "raidz_gen_name" and "raidz_rec_name"
+ * arrays, starting with the parity function "implementation" name):
+ *     impl gen_p gen_pq gen_pqr rec_p rec_q rec_r rec_pq rec_pr rec_qr rec_pqr
+ * This is followed by a row for each parity function implementation, showing
+ * the "speed" values calculated for that implementation for each of the
+ * parity generation and reconstruction functions in the "raidz_all_maths"
+ * array.
+ */
+static raidz_impl_kstat_t raidz_impl_kstats[ARRAY_SIZE(raidz_all_maths) + 1];
+
+#endif
+
+/*
+ * Returns the RAIDZ operations for raidz_map() parity calculations.   When
+ * a SIMD implementation is not allowed in the current context, then fallback
+ * to the fastest generic implementation.
+ */
+const raidz_impl_ops_t *
+vdev_raidz_math_get_ops(void)
+{
+	/*
+	 * illumos porting note:
+	 * The following check from OpenZFS is disabled since we don't have
+	 * this compiled in yet and we need to be able to change the
+	 * implementation for the user-level test suite.
+	 *
+	 * if (!kfpu_allowed())
+	 *	return (&vdev_raidz_scalar_impl);
+	 */
+
+	raidz_impl_ops_t *ops = NULL;
+	const uint32_t impl = RAIDZ_IMPL_READ(zfs_vdev_raidz_impl);
+
+	switch (impl) {
+	case IMPL_FASTEST:
+		ASSERT(raidz_math_initialized);
+		ops = &vdev_raidz_fastest_impl;
+		break;
+	case IMPL_CYCLE:
+		/* Cycle through all supported implementations */
+		ASSERT(raidz_math_initialized);
+		ASSERT3U(raidz_supp_impl_cnt, >, 0);
+		static size_t cycle_impl_idx = 0;
+		size_t idx = (++cycle_impl_idx) % raidz_supp_impl_cnt;
+		ops = raidz_supp_impl[idx];
+		break;
+	case IMPL_ORIGINAL:
+		ops = (raidz_impl_ops_t *)&vdev_raidz_original_impl;
+		break;
+	case IMPL_SCALAR:
+		ops = (raidz_impl_ops_t *)&vdev_raidz_scalar_impl;
+		break;
+	default:
+		ASSERT3U(impl, <, raidz_supp_impl_cnt);
+		ASSERT3U(raidz_supp_impl_cnt, >, 0);
+		if (impl < ARRAY_SIZE(raidz_all_maths))
+			ops = raidz_supp_impl[impl];
+		break;
+	}
+
+	ASSERT3P(ops, !=, NULL);
+
+	return (ops);
+}
+
+/*
+ * Select parity generation method for raidz_map
+ */
+int
+vdev_raidz_math_generate(raidz_map_t *rm)
+{
+	raidz_gen_f gen_parity = NULL;
+
+	switch (raidz_parity(rm)) {
+		case 1:
+			gen_parity = rm->rm_ops->gen[RAIDZ_GEN_P];
+			break;
+		case 2:
+			gen_parity = rm->rm_ops->gen[RAIDZ_GEN_PQ];
+			break;
+		case 3:
+			gen_parity = rm->rm_ops->gen[RAIDZ_GEN_PQR];
+			break;
+		default:
+			gen_parity = NULL;
+			cmn_err(CE_PANIC, "invalid RAID-Z configuration %u",
+			    (uint_t)raidz_parity(rm));
+			break;
+	}
+
+	/* if method is NULL execute the original implementation */
+	if (gen_parity == NULL)
+		return (RAIDZ_ORIGINAL_IMPL);
+
+	gen_parity(rm);
+
+	return (0);
+}
+
+static raidz_rec_f
+reconstruct_fun_p_sel(raidz_map_t *rm, const int *parity_valid,
+    const int nbaddata)
+{
+	if (nbaddata == 1 && parity_valid[CODE_P]) {
+		return (rm->rm_ops->rec[RAIDZ_REC_P]);
+	}
+	return ((raidz_rec_f) NULL);
+}
+
+static raidz_rec_f
+reconstruct_fun_pq_sel(raidz_map_t *rm, const int *parity_valid,
+    const int nbaddata)
+{
+	if (nbaddata == 1) {
+		if (parity_valid[CODE_P]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_P]);
+		} else if (parity_valid[CODE_Q]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_Q]);
+		}
+	} else if (nbaddata == 2 &&
+	    parity_valid[CODE_P] && parity_valid[CODE_Q]) {
+		return (rm->rm_ops->rec[RAIDZ_REC_PQ]);
+	}
+	return ((raidz_rec_f) NULL);
+}
+
+static raidz_rec_f
+reconstruct_fun_pqr_sel(raidz_map_t *rm, const int *parity_valid,
+    const int nbaddata)
+{
+	if (nbaddata == 1) {
+		if (parity_valid[CODE_P]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_P]);
+		} else if (parity_valid[CODE_Q]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_Q]);
+		} else if (parity_valid[CODE_R]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_R]);
+		}
+	} else if (nbaddata == 2) {
+		if (parity_valid[CODE_P] && parity_valid[CODE_Q]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_PQ]);
+		} else if (parity_valid[CODE_P] && parity_valid[CODE_R]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_PR]);
+		} else if (parity_valid[CODE_Q] && parity_valid[CODE_R]) {
+			return (rm->rm_ops->rec[RAIDZ_REC_QR]);
+		}
+	} else if (nbaddata == 3 &&
+	    parity_valid[CODE_P] && parity_valid[CODE_Q] &&
+	    parity_valid[CODE_R]) {
+		return (rm->rm_ops->rec[RAIDZ_REC_PQR]);
+	}
+	return ((raidz_rec_f) NULL);
+}
+
+/*
+ * Select data reconstruction method for raidz_map
+ * @parity_valid - Parity validity flag
+ * @dt           - Failed data index array
+ * @nbaddata     - Number of failed data columns
+ */
+int
+vdev_raidz_math_reconstruct(raidz_map_t *rm, const int *parity_valid,
+    const int *dt, const int nbaddata)
+{
+	raidz_rec_f rec_fn = NULL;
+
+	switch (raidz_parity(rm)) {
+	case PARITY_P:
+		rec_fn = reconstruct_fun_p_sel(rm, parity_valid, nbaddata);
+		break;
+	case PARITY_PQ:
+		rec_fn = reconstruct_fun_pq_sel(rm, parity_valid, nbaddata);
+		break;
+	case PARITY_PQR:
+		rec_fn = reconstruct_fun_pqr_sel(rm, parity_valid, nbaddata);
+		break;
+	default:
+		cmn_err(CE_PANIC, "invalid RAID-Z configuration %u",
+		    (uint_t)raidz_parity(rm));
+		break;
+	}
+
+	if (rec_fn == NULL)
+		return (RAIDZ_ORIGINAL_IMPL);
+	else
+		return (rec_fn(rm, dt));
+}
+
+const char *raidz_gen_name[] = {
+	"gen_p", "gen_pq", "gen_pqr"
+};
+const char *raidz_rec_name[] = {
+	"rec_p", "rec_q", "rec_r",
+	"rec_pq", "rec_pr", "rec_qr", "rec_pqr"
+};
+
+#if defined(_KERNEL)
+
+#define	BENCH_D_COLS	(8ULL)
+#define	BENCH_COLS	(BENCH_D_COLS + PARITY_PQR)
+#define	BENCH_ZIO_SIZE	(1ULL << SPA_OLD_MAXBLOCKSHIFT)	/* 128 kiB */
+#define	BENCH_NS	MSEC2NSEC(25)			/* 25ms */
+
+typedef void (*benchmark_fn)(raidz_map_t *rm, const int fn);
+
+static void
+benchmark_gen_impl(raidz_map_t *rm, const int fn)
+{
+	(void) fn;
+	vdev_raidz_generate_parity(rm);
+}
+
+static void
+benchmark_rec_impl(raidz_map_t *rm, const int fn)
+{
+	static const int rec_tgt[7][3] = {
+		{1, 2, 3},	/* rec_p:   bad QR & D[0]	*/
+		{0, 2, 3},	/* rec_q:   bad PR & D[0]	*/
+		{0, 1, 3},	/* rec_r:   bad PQ & D[0]	*/
+		{2, 3, 4},	/* rec_pq:  bad R  & D[0][1]	*/
+		{1, 3, 4},	/* rec_pr:  bad Q  & D[0][1]	*/
+		{0, 3, 4},	/* rec_qr:  bad P  & D[0][1]	*/
+		{3, 4, 5}	/* rec_pqr: bad    & D[0][1][2] */
+	};
+
+	vdev_raidz_reconstruct(rm, rec_tgt[fn], 3);
+}
+
+/*
+ * Benchmarking of all supported implementations (raidz_supp_impl_cnt)
+ * is performed by setting the rm_ops pointer and calling the top level
+ * generate/reconstruct methods of bench_rm.
+ */
+static void
+benchmark_raidz_impl(raidz_map_t *bench_rm, const int fn, benchmark_fn bench_fn)
+{
+	uint64_t run_cnt, speed, best_speed = 0;
+	hrtime_t t_start, t_diff;
+	raidz_impl_ops_t *curr_impl;
+	raidz_impl_kstat_t *fstat = &raidz_impl_kstats[raidz_supp_impl_cnt];
+	int impl, i;
+
+	for (impl = 0; impl < raidz_supp_impl_cnt; impl++) {
+		/* set an implementation to benchmark */
+		curr_impl = raidz_supp_impl[impl];
+		bench_rm->rm_ops = curr_impl;
+
+		run_cnt = 0;
+		t_start = gethrtime();
+
+		do {
+			for (i = 0; i < 25; i++, run_cnt++)
+				bench_fn(bench_rm, fn);
+
+			t_diff = gethrtime() - t_start;
+		} while (t_diff < BENCH_NS);
+
+		speed = run_cnt * BENCH_ZIO_SIZE * NANOSEC;
+		speed /= (t_diff * BENCH_COLS);
+
+		if (bench_fn == benchmark_gen_impl)
+			raidz_impl_kstats[impl].gen[fn] = speed;
+		else
+			raidz_impl_kstats[impl].rec[fn] = speed;
+
+		/* Update fastest implementation method */
+		if (speed > best_speed) {
+			best_speed = speed;
+
+			if (bench_fn == benchmark_gen_impl) {
+				fstat->gen[fn] = impl;
+				vdev_raidz_fastest_impl.gen[fn] =
+				    curr_impl->gen[fn];
+			} else {
+				fstat->rec[fn] = impl;
+				vdev_raidz_fastest_impl.rec[fn] =
+				    curr_impl->rec[fn];
+			}
+		}
+	}
+}
+#endif
+
+/*
+ * Initialize and benchmark all supported implementations.
+ */
+static void
+benchmark_raidz(void)
+{
+	raidz_impl_ops_t *curr_impl;
+	int i, c;
+
+	/* Move supported impl into raidz_supp_impl */
+	for (i = 0, c = 0; i < ARRAY_SIZE(raidz_all_maths); i++) {
+		curr_impl = (raidz_impl_ops_t *)raidz_all_maths[i];
+
+		if (curr_impl->init)
+			curr_impl->init();
+
+		if (curr_impl->is_supported())
+			raidz_supp_impl[c++] = (raidz_impl_ops_t *)curr_impl;
+	}
+	membar_producer();		/* complete raidz_supp_impl[] init */
+	raidz_supp_impl_cnt = c;	/* number of supported impl */
+
+#if defined(_KERNEL)
+	zio_t *bench_zio = NULL;
+	raidz_map_t *bench_rm = NULL;
+	uint64_t bench_parity;
+
+	/* Fake a zio and run the benchmark on a warmed up buffer */
+	bench_zio = kmem_zalloc(sizeof (zio_t), KM_SLEEP);
+	bench_zio->io_offset = 0;
+	bench_zio->io_size = BENCH_ZIO_SIZE; /* only data columns */
+	bench_zio->io_abd = abd_alloc_linear(BENCH_ZIO_SIZE, B_TRUE);
+	memset(abd_to_buf(bench_zio->io_abd), 0xAA, BENCH_ZIO_SIZE);
+
+	/* Benchmark parity generation methods */
+	for (int fn = 0; fn < RAIDZ_GEN_NUM; fn++) {
+		bench_parity = fn + 1;
+		/* New raidz_map is needed for each generate_p/q/r */
+		bench_rm = vdev_raidz_map_alloc(bench_zio, SPA_MINBLOCKSHIFT,
+		    BENCH_D_COLS + bench_parity, bench_parity);
+
+		benchmark_raidz_impl(bench_rm, fn, benchmark_gen_impl);
+
+		vdev_raidz_map_free(bench_rm);
+	}
+
+	/* Benchmark data reconstruction methods */
+	bench_rm = vdev_raidz_map_alloc(bench_zio, SPA_MINBLOCKSHIFT,
+	    BENCH_COLS, PARITY_PQR);
+
+	for (int fn = 0; fn < RAIDZ_REC_NUM; fn++)
+		benchmark_raidz_impl(bench_rm, fn, benchmark_rec_impl);
+
+	vdev_raidz_map_free(bench_rm);
+
+	/* cleanup the bench zio */
+	abd_free(bench_zio->io_abd);
+	kmem_free(bench_zio, sizeof (zio_t));
+#else
+	/*
+	 * Skip the benchmark in user space to avoid impacting libzpool
+	 * consumers (zdb, zhack, zinject, ztest).  The last implementation
+	 * is assumed to be the fastest and used by default.
+	 */
+	memcpy(&vdev_raidz_fastest_impl,
+	    raidz_supp_impl[raidz_supp_impl_cnt - 1],
+	    sizeof (vdev_raidz_fastest_impl));
+	strcpy(vdev_raidz_fastest_impl.name, "fastest");
+#endif /* _KERNEL */
+}
+
+void
+vdev_raidz_math_init(void)
+{
+	/* Determine the fastest available implementation. */
+	benchmark_raidz();
+
+	/* Finish initialization */
+	atomic_swap_32(&zfs_vdev_raidz_impl, user_sel_impl);
+	raidz_math_initialized = B_TRUE;
+}
+
+void
+vdev_raidz_math_fini(void)
+{
+	raidz_impl_ops_t const *curr_impl;
+
+	for (int i = 0; i < ARRAY_SIZE(raidz_all_maths); i++) {
+		curr_impl = raidz_all_maths[i];
+		if (curr_impl->fini)
+			curr_impl->fini();
+	}
+}
+
+static const struct {
+	char *name;
+	uint32_t sel;
+} math_impl_opts[] = {
+		{ "cycle",	IMPL_CYCLE },
+		{ "fastest",	IMPL_FASTEST },
+		{ "original",	IMPL_ORIGINAL },
+		{ "scalar",	IMPL_SCALAR }
+};
+
+/*
+ * Function sets desired raidz implementation.
+ *
+ * If we are called before init(), user preference will be saved in
+ * user_sel_impl, and applied in later init() call. This occurs when module
+ * parameter is specified on module load. Otherwise, directly update
+ * zfs_vdev_raidz_impl.
+ *
+ * @val		Name of raidz implementation to use
+ * @param	Unused.
+ */
+int
+vdev_raidz_impl_set(const char *val)
+{
+	int err = -EINVAL;
+	char req_name[RAIDZ_IMPL_NAME_MAX];
+	uint32_t impl = RAIDZ_IMPL_READ(user_sel_impl);
+	size_t i;
+
+	/* sanitize input */
+	i = strnlen(val, RAIDZ_IMPL_NAME_MAX);
+	if (i == 0 || i == RAIDZ_IMPL_NAME_MAX)
+		return (err);
+
+	strlcpy(req_name, val, RAIDZ_IMPL_NAME_MAX);
+	while (i > 0 && !!isspace(req_name[i-1]))
+		i--;
+	req_name[i] = '\0';
+
+	/* Check mandatory options */
+	for (i = 0; i < ARRAY_SIZE(math_impl_opts); i++) {
+		if (strcmp(req_name, math_impl_opts[i].name) == 0) {
+			impl = math_impl_opts[i].sel;
+			err = 0;
+			break;
+		}
+	}
+
+	/* check all supported impl if init() was already called */
+	if (err != 0 && raidz_math_initialized) {
+		/* check all supported implementations */
+		for (i = 0; i < raidz_supp_impl_cnt; i++) {
+			if (strcmp(req_name, raidz_supp_impl[i]->name) == 0) {
+				impl = i;
+				err = 0;
+				break;
+			}
+		}
+	}
+
+	if (err == 0) {
+		if (raidz_math_initialized)
+			atomic_swap_32(&zfs_vdev_raidz_impl, impl);
+		else
+			atomic_swap_32(&user_sel_impl, impl);
+	}
+
+	return (err);
+}
+
+#if defined(_KERNEL) && defined(__linux__)
+
+static int
+zfs_vdev_raidz_impl_set(const char *val, zfs_kernel_param_t *kp)
+{
+	return (vdev_raidz_impl_set(val));
+}
+
+static int
+zfs_vdev_raidz_impl_get(char *buffer, zfs_kernel_param_t *kp)
+{
+	int i, cnt = 0;
+	char *fmt;
+	const uint32_t impl = RAIDZ_IMPL_READ(zfs_vdev_raidz_impl);
+
+	ASSERT(raidz_math_initialized);
+
+	/* list mandatory options */
+	for (i = 0; i < ARRAY_SIZE(math_impl_opts) - 2; i++) {
+		fmt = (impl == math_impl_opts[i].sel) ? "[%s] " : "%s ";
+		cnt += sprintf(buffer + cnt, fmt, math_impl_opts[i].name);
+	}
+
+	/* list all supported implementations */
+	for (i = 0; i < raidz_supp_impl_cnt; i++) {
+		fmt = (i == impl) ? "[%s] " : "%s ";
+		cnt += sprintf(buffer + cnt, fmt, raidz_supp_impl[i]->name);
+	}
+
+	return (cnt);
+}
+
+module_param_call(zfs_vdev_raidz_impl, zfs_vdev_raidz_impl_set,
+    zfs_vdev_raidz_impl_get, NULL, 0644);
+MODULE_PARM_DESC(zfs_vdev_raidz_impl, "Select raidz implementation.");
+#endif
diff --git a/usr/src/uts/common/fs/zfs/vdev_raidz_math_impl.h b/usr/src/uts/common/fs/zfs/vdev_raidz_math_impl.h
new file mode 100644
index 0000000000..89c2082c4a
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/vdev_raidz_math_impl.h
@@ -0,0 +1,1477 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ */
+
+#ifndef _VDEV_RAIDZ_MATH_IMPL_H
+#define	_VDEV_RAIDZ_MATH_IMPL_H
+
+#include <sys/types.h>
+
+#define	raidz_inline inline __attribute__((always_inline))
+#ifndef noinline
+#define	noinline __attribute__((noinline))
+#endif
+
+/*
+ * Functions calculate multiplication constants for data reconstruction.
+ * Coefficients depend on RAIDZ geometry, indexes of failed child vdevs, and
+ * used parity columns for reconstruction.
+ * @rm			RAIDZ map
+ * @tgtidx		array of missing data indexes
+ * @coeff		output array of coefficients. Array must be provided by
+ *         		user and must hold minimum MUL_CNT values.
+ */
+static noinline void
+raidz_rec_q_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+
+	coeff[MUL_Q_X] = gf_exp2(255 - (ncols - x - 1));
+}
+
+static noinline void
+raidz_rec_r_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+
+	coeff[MUL_R_X] = gf_exp4(255 - (ncols - x - 1));
+}
+
+static noinline void
+raidz_rec_pq_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+	const unsigned y = tgtidx[TARGET_Y];
+	gf_t a, b, e;
+
+	a = gf_exp2(x + 255 - y);
+	b = gf_exp2(255 - (ncols - x - 1));
+	e = a ^ 0x01;
+
+	coeff[MUL_PQ_X] = gf_div(a, e);
+	coeff[MUL_PQ_Y] = gf_div(b, e);
+}
+
+static noinline void
+raidz_rec_pr_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+	const unsigned y = tgtidx[TARGET_Y];
+
+	gf_t a, b, e;
+
+	a = gf_exp4(x + 255 - y);
+	b = gf_exp4(255 - (ncols - x - 1));
+	e = a ^ 0x01;
+
+	coeff[MUL_PR_X] = gf_div(a, e);
+	coeff[MUL_PR_Y] = gf_div(b, e);
+}
+
+static noinline void
+raidz_rec_qr_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+	const unsigned y = tgtidx[TARGET_Y];
+
+	gf_t nx, ny, nxxy, nxyy, d;
+
+	nx = gf_exp2(ncols - x - 1);
+	ny = gf_exp2(ncols - y - 1);
+	nxxy = gf_mul(gf_mul(nx, nx), ny);
+	nxyy = gf_mul(gf_mul(nx, ny), ny);
+	d = nxxy ^ nxyy;
+
+	coeff[MUL_QR_XQ] = ny;
+	coeff[MUL_QR_X]	= gf_div(ny, d);
+	coeff[MUL_QR_YQ] = nx;
+	coeff[MUL_QR_Y]	= gf_div(nx, d);
+}
+
+static noinline void
+raidz_rec_pqr_coeff(const raidz_map_t *rm, const int *tgtidx, unsigned *coeff)
+{
+	const unsigned ncols = raidz_ncols(rm);
+	const unsigned x = tgtidx[TARGET_X];
+	const unsigned y = tgtidx[TARGET_Y];
+	const unsigned z = tgtidx[TARGET_Z];
+
+	gf_t nx, ny, nz, nxx, nyy, nzz, nyyz, nyzz, xd, yd;
+
+	nx = gf_exp2(ncols - x - 1);
+	ny = gf_exp2(ncols - y - 1);
+	nz = gf_exp2(ncols - z - 1);
+
+	nxx = gf_exp4(ncols - x - 1);
+	nyy = gf_exp4(ncols - y - 1);
+	nzz = gf_exp4(ncols - z - 1);
+
+	nyyz = gf_mul(gf_mul(ny, nz), ny);
+	nyzz = gf_mul(nzz, ny);
+
+	xd = gf_mul(nxx, ny) ^ gf_mul(nx, nyy) ^ nyyz ^
+	    gf_mul(nxx, nz) ^ gf_mul(nzz, nx) ^  nyzz;
+
+	yd = gf_inv(ny ^ nz);
+
+	coeff[MUL_PQR_XP] = gf_div(nyyz ^ nyzz, xd);
+	coeff[MUL_PQR_XQ] = gf_div(nyy ^ nzz, xd);
+	coeff[MUL_PQR_XR] = gf_div(ny ^ nz, xd);
+	coeff[MUL_PQR_YU] = nx;
+	coeff[MUL_PQR_YP] = gf_mul(nz, yd);
+	coeff[MUL_PQR_YQ] = yd;
+}
+
+/*
+ * Method for zeroing a buffer (can be implemented using SIMD).
+ * This method is used by multiple for gen/rec functions.
+ *
+ * @dc		Destination buffer
+ * @dsize	Destination buffer size
+ * @private	Unused
+ */
+static int
+raidz_zero_abd_cb(void *dc, size_t dsize, void *private)
+{
+	v_t *dst = (v_t *)dc;
+	size_t i;
+
+	ZERO_DEFINE();
+
+	(void) private; /* unused */
+
+	ZERO(ZERO_D);
+
+	for (i = 0; i < dsize / sizeof (v_t); i += (2 * ZERO_STRIDE)) {
+		STORE(dst + i, ZERO_D);
+		STORE(dst + i + ZERO_STRIDE, ZERO_D);
+	}
+
+	return (0);
+}
+
+#define	raidz_zero(dabd, size)						\
+{									\
+	abd_iterate_func(dabd, 0, size, raidz_zero_abd_cb, NULL);	\
+}
+
+/*
+ * Method for copying two buffers (can be implemented using SIMD).
+ * This method is used by multiple for gen/rec functions.
+ *
+ * @dc		Destination buffer
+ * @sc		Source buffer
+ * @dsize	Destination buffer size
+ * @ssize	Source buffer size
+ * @private	Unused
+ */
+static int
+raidz_copy_abd_cb(void *dc, void *sc, size_t size, void *private)
+{
+	v_t *dst = (v_t *)dc;
+	const v_t *src = (v_t *)sc;
+	size_t i;
+
+	COPY_DEFINE();
+
+	(void) private; /* unused */
+
+	for (i = 0; i < size / sizeof (v_t); i += (2 * COPY_STRIDE)) {
+		LOAD(src + i, COPY_D);
+		STORE(dst + i, COPY_D);
+
+		LOAD(src + i + COPY_STRIDE, COPY_D);
+		STORE(dst + i + COPY_STRIDE, COPY_D);
+	}
+
+	return (0);
+}
+
+
+#define	raidz_copy(dabd, sabd, size)					\
+{									\
+	abd_iterate_func2(dabd, sabd, 0, 0, size, raidz_copy_abd_cb, NULL);\
+}
+
+/*
+ * Method for adding (XORing) two buffers.
+ * Source and destination are XORed together and result is stored in
+ * destination buffer. This method is used by multiple for gen/rec functions.
+ *
+ * @dc		Destination buffer
+ * @sc		Source buffer
+ * @dsize	Destination buffer size
+ * @ssize	Source buffer size
+ * @private	Unused
+ */
+static int
+raidz_add_abd_cb(void *dc, void *sc, size_t size, void *private)
+{
+	v_t *dst = (v_t *)dc;
+	const v_t *src = (v_t *)sc;
+	size_t i;
+
+	ADD_DEFINE();
+
+	(void) private; /* unused */
+
+	for (i = 0; i < size / sizeof (v_t); i += (2 * ADD_STRIDE)) {
+		LOAD(dst + i, ADD_D);
+		XOR_ACC(src + i, ADD_D);
+		STORE(dst + i, ADD_D);
+
+		LOAD(dst + i + ADD_STRIDE, ADD_D);
+		XOR_ACC(src + i + ADD_STRIDE, ADD_D);
+		STORE(dst + i + ADD_STRIDE, ADD_D);
+	}
+
+	return (0);
+}
+
+#define	raidz_add(dabd, sabd, size)					\
+{									\
+	abd_iterate_func2(dabd, sabd, 0, 0, size, raidz_add_abd_cb, NULL);\
+}
+
+/*
+ * Method for multiplying a buffer with a constant in GF(2^8).
+ * Symbols from buffer are multiplied by a constant and result is stored
+ * back in the same buffer.
+ *
+ * @dc		In/Out data buffer.
+ * @size	Size of the buffer
+ * @private	pointer to the multiplication constant (unsigned)
+ */
+static int
+raidz_mul_abd_cb(void *dc, size_t size, void *private)
+{
+	const unsigned mul = *((unsigned *)private);
+	v_t *d = (v_t *)dc;
+	size_t i;
+
+	MUL_DEFINE();
+
+	for (i = 0; i < size / sizeof (v_t); i += (2 * MUL_STRIDE)) {
+		LOAD(d + i, MUL_D);
+		MUL(mul, MUL_D);
+		STORE(d + i, MUL_D);
+
+		LOAD(d + i + MUL_STRIDE, MUL_D);
+		MUL(mul, MUL_D);
+		STORE(d + i + MUL_STRIDE, MUL_D);
+	}
+
+	return (0);
+}
+
+
+/*
+ * Syndrome generation/update macros
+ *
+ * Require LOAD(), XOR(), STORE(), MUL2(), and MUL4() macros
+ */
+#define	P_D_SYNDROME(D, T, t)		\
+{					\
+	LOAD((t), T);			\
+	XOR(D, T);			\
+	STORE((t), T);			\
+}
+
+#define	Q_D_SYNDROME(D, T, t)		\
+{					\
+	LOAD((t), T);			\
+	MUL2(T);			\
+	XOR(D, T);			\
+	STORE((t), T);			\
+}
+
+#define	Q_SYNDROME(T, t)		\
+{					\
+	LOAD((t), T);			\
+	MUL2(T);			\
+	STORE((t), T);			\
+}
+
+#define	R_D_SYNDROME(D, T, t)		\
+{					\
+	LOAD((t), T);			\
+	MUL4(T);			\
+	XOR(D, T);			\
+	STORE((t), T);			\
+}
+
+#define	R_SYNDROME(T, t)		\
+{					\
+	LOAD((t), T);			\
+	MUL4(T);			\
+	STORE((t), T);			\
+}
+
+
+/*
+ * PARITY CALCULATION
+ *
+ * Macros *_SYNDROME are used for parity/syndrome calculation.
+ * *_D_SYNDROME() macros are used to calculate syndrome between 0 and
+ * length of data column, and *_SYNDROME() macros are only for updating
+ * the parity/syndrome if data column is shorter.
+ *
+ * P parity is calculated using raidz_add_abd().
+ */
+
+/*
+ * Generate P parity (RAIDZ1)
+ *
+ * @rm	RAIDZ map
+ */
+static raidz_inline void
+raidz_generate_p_impl(raidz_map_t * const rm)
+{
+	size_t c;
+	const size_t ncols = raidz_ncols(rm);
+	const size_t psize = rm->rm_col[CODE_P].rc_size;
+	abd_t *pabd = rm->rm_col[CODE_P].rc_abd;
+	size_t size;
+	abd_t *dabd;
+
+	raidz_math_begin();
+
+	/* start with first data column */
+	raidz_copy(pabd, rm->rm_col[1].rc_abd, psize);
+
+	for (c = 2; c < ncols; c++) {
+		dabd = rm->rm_col[c].rc_abd;
+		size = rm->rm_col[c].rc_size;
+
+		/* add data column */
+		raidz_add(pabd, dabd, size);
+	}
+
+	raidz_math_end();
+}
+
+
+/*
+ * Generate PQ parity (RAIDZ2)
+ * The function is called per data column.
+ *
+ * @c		array of pointers to parity (code) columns
+ * @dc		pointer to data column
+ * @csize	size of parity columns
+ * @dsize	size of data column
+ */
+static void
+raidz_gen_pq_add(void **c, const void *dc, const size_t csize,
+    const size_t dsize)
+{
+	v_t *p = (v_t *)c[0];
+	v_t *q = (v_t *)c[1];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const qend = q + (csize / sizeof (v_t));
+
+	GEN_PQ_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += GEN_PQ_STRIDE, p += GEN_PQ_STRIDE,
+	    q += GEN_PQ_STRIDE) {
+		LOAD(d, GEN_PQ_D);
+		P_D_SYNDROME(GEN_PQ_D, GEN_PQ_C, p);
+		Q_D_SYNDROME(GEN_PQ_D, GEN_PQ_C, q);
+	}
+	for (; q < qend; q += GEN_PQ_STRIDE) {
+		Q_SYNDROME(GEN_PQ_C, q);
+	}
+}
+
+
+/*
+ * Generate PQ parity (RAIDZ2)
+ *
+ * @rm	RAIDZ map
+ */
+static raidz_inline void
+raidz_generate_pq_impl(raidz_map_t * const rm)
+{
+	size_t c;
+	const size_t ncols = raidz_ncols(rm);
+	const size_t csize = rm->rm_col[CODE_P].rc_size;
+	size_t dsize;
+	abd_t *dabd;
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_P].rc_abd,
+		rm->rm_col[CODE_Q].rc_abd
+	};
+
+	raidz_math_begin();
+
+	raidz_copy(cabds[CODE_P], rm->rm_col[2].rc_abd, csize);
+	raidz_copy(cabds[CODE_Q], rm->rm_col[2].rc_abd, csize);
+
+	for (c = 3; c < ncols; c++) {
+		dabd = rm->rm_col[c].rc_abd;
+		dsize = rm->rm_col[c].rc_size;
+
+		abd_raidz_gen_iterate(cabds, dabd, csize, dsize, 2,
+		    raidz_gen_pq_add);
+	}
+
+	raidz_math_end();
+}
+
+
+/*
+ * Generate PQR parity (RAIDZ3)
+ * The function is called per data column.
+ *
+ * @c		array of pointers to parity (code) columns
+ * @dc		pointer to data column
+ * @csize	size of parity columns
+ * @dsize	size of data column
+ */
+static void
+raidz_gen_pqr_add(void **c, const void *dc, const size_t csize,
+    const size_t dsize)
+{
+	v_t *p = (v_t *)c[0];
+	v_t *q = (v_t *)c[1];
+	v_t *r = (v_t *)c[CODE_R];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const qend = q + (csize / sizeof (v_t));
+
+	GEN_PQR_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += GEN_PQR_STRIDE, p += GEN_PQR_STRIDE,
+	    q += GEN_PQR_STRIDE, r += GEN_PQR_STRIDE) {
+		LOAD(d, GEN_PQR_D);
+		P_D_SYNDROME(GEN_PQR_D, GEN_PQR_C, p);
+		Q_D_SYNDROME(GEN_PQR_D, GEN_PQR_C, q);
+		R_D_SYNDROME(GEN_PQR_D, GEN_PQR_C, r);
+	}
+	for (; q < qend; q += GEN_PQR_STRIDE, r += GEN_PQR_STRIDE) {
+		Q_SYNDROME(GEN_PQR_C, q);
+		R_SYNDROME(GEN_PQR_C, r);
+	}
+}
+
+
+/*
+ * Generate PQR parity (RAIDZ2)
+ *
+ * @rm	RAIDZ map
+ */
+static raidz_inline void
+raidz_generate_pqr_impl(raidz_map_t * const rm)
+{
+	size_t c;
+	const size_t ncols = raidz_ncols(rm);
+	const size_t csize = rm->rm_col[CODE_P].rc_size;
+	size_t dsize;
+	abd_t *dabd;
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_P].rc_abd,
+		rm->rm_col[CODE_Q].rc_abd,
+		rm->rm_col[CODE_R].rc_abd
+	};
+
+	raidz_math_begin();
+
+	raidz_copy(cabds[CODE_P], rm->rm_col[3].rc_abd, csize);
+	raidz_copy(cabds[CODE_Q], rm->rm_col[3].rc_abd, csize);
+	raidz_copy(cabds[CODE_R], rm->rm_col[3].rc_abd, csize);
+
+	for (c = 4; c < ncols; c++) {
+		dabd = rm->rm_col[c].rc_abd;
+		dsize = rm->rm_col[c].rc_size;
+
+		abd_raidz_gen_iterate(cabds, dabd, csize, dsize, 3,
+		    raidz_gen_pqr_add);
+	}
+
+	raidz_math_end();
+}
+
+
+/*
+ * DATA RECONSTRUCTION
+ *
+ * Data reconstruction process consists of two phases:
+ * 	- Syndrome calculation
+ * 	- Data reconstruction
+ *
+ * Syndrome is calculated by generating parity using available data columns
+ * and zeros in places of erasure. Existing parity is added to corresponding
+ * syndrome value to obtain the [P|Q|R]syn values from equation:
+ * 	P = Psyn + Dx + Dy + Dz
+ * 	Q = Qsyn + 2^x * Dx + 2^y * Dy + 2^z * Dz
+ * 	R = Rsyn + 4^x * Dx + 4^y * Dy + 4^z * Dz
+ *
+ * For data reconstruction phase, the corresponding equations are solved
+ * for missing data (Dx, Dy, Dz). This generally involves multiplying known
+ * symbols by an coefficient and adding them together. The multiplication
+ * constant coefficients are calculated ahead of the operation in
+ * raidz_rec_[q|r|pq|pq|qr|pqr]_coeff() functions.
+ *
+ * IMPLEMENTATION NOTE: RAID-Z block can have complex geometry, with "big"
+ * and "short" columns.
+ * For this reason, reconstruction is performed in minimum of
+ * two steps. First, from offset 0 to short_size, then from short_size to
+ * short_size. Calculation functions REC_[*]_BLOCK() are implemented to work
+ * over both ranges. The split also enables removal of conditional expressions
+ * from loop bodies, improving throughput of SIMD implementations.
+ * For the best performance, all functions marked with raidz_inline attribute
+ * must be inlined by compiler.
+ *
+ *    parity          data
+ *    columns         columns
+ * <----------> <------------------>
+ *                   x       y  <----+ missing columns (x, y)
+ *                   |       |
+ * +---+---+---+---+-v-+---+-v-+---+   ^ 0
+ * |   |   |   |   |   |   |   |   |   |
+ * |   |   |   |   |   |   |   |   |   |
+ * | P | Q | R | D | D | D | D | D |   |
+ * |   |   |   | 0 | 1 | 2 | 3 | 4 |   |
+ * |   |   |   |   |   |   |   |   |   v
+ * |   |   |   |   |   +---+---+---+   ^ short_size
+ * |   |   |   |   |   |               |
+ * +---+---+---+---+---+               v big_size
+ * <------------------> <---------->
+ *      big columns     short columns
+ *
+ */
+
+
+
+
+/*
+ * Reconstruct single data column using P parity
+ *
+ * @syn_method	raidz_add_abd()
+ * @rec_method	not applicable
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_p_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	size_t size;
+	abd_t *dabd;
+
+	raidz_math_begin();
+
+	/* copy P into target */
+	raidz_copy(xabd, rm->rm_col[CODE_P].rc_abd, xsize);
+
+	/* generate p_syndrome */
+	for (c = firstdc; c < ncols; c++) {
+		if (c == x)
+			continue;
+
+		dabd = rm->rm_col[c].rc_abd;
+		size = MIN(rm->rm_col[c].rc_size, xsize);
+
+		raidz_add(xabd, dabd, size);
+	}
+
+	raidz_math_end();
+
+	return (1 << CODE_P);
+}
+
+
+/*
+ * Generate Q syndrome (Qsyn)
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @xsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_q_abd(void **xc, const void *dc, const size_t xsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)xc[TARGET_X];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const xend = x + (xsize / sizeof (v_t));
+
+	SYN_Q_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += SYN_STRIDE, x += SYN_STRIDE) {
+		LOAD(d, SYN_Q_D);
+		Q_D_SYNDROME(SYN_Q_D, SYN_Q_X, x);
+	}
+	for (; x < xend; x += SYN_STRIDE) {
+		Q_SYNDROME(SYN_Q_X, x);
+	}
+}
+
+
+/*
+ * Reconstruct single data column using Q parity
+ *
+ * @syn_method	raidz_add_abd()
+ * @rec_method	raidz_mul_abd_cb()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_q_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	const size_t xsize = rm->rm_col[x].rc_size;
+	abd_t *tabds[] = { xabd };
+
+	unsigned coeff[MUL_CNT];
+	raidz_rec_q_coeff(rm, tgtidx, coeff);
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+	}
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 1,
+		    raidz_syn_q_abd);
+	}
+
+	/* add Q to the syndrome */
+	raidz_add(xabd, rm->rm_col[CODE_Q].rc_abd, xsize);
+
+	/* transform the syndrome */
+	abd_iterate_func(xabd, 0, xsize, raidz_mul_abd_cb, (void*) coeff);
+
+	raidz_math_end();
+
+	return (1 << CODE_Q);
+}
+
+
+/*
+ * Generate R syndrome (Rsyn)
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @tsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_r_abd(void **xc, const void *dc, const size_t tsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)xc[TARGET_X];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+
+	SYN_R_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += SYN_STRIDE, x += SYN_STRIDE) {
+		LOAD(d, SYN_R_D);
+		R_D_SYNDROME(SYN_R_D, SYN_R_X, x);
+	}
+	for (; x < xend; x += SYN_STRIDE) {
+		R_SYNDROME(SYN_R_X, x);
+	}
+}
+
+
+/*
+ * Reconstruct single data column using R parity
+ *
+ * @syn_method	raidz_add_abd()
+ * @rec_method	raidz_mul_abd_cb()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_r_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	abd_t *tabds[] = { xabd };
+
+	unsigned coeff[MUL_CNT];
+	raidz_rec_r_coeff(rm, tgtidx, coeff);
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+	}
+
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 1,
+		    raidz_syn_r_abd);
+	}
+
+	/* add R to the syndrome */
+	raidz_add(xabd, rm->rm_col[CODE_R].rc_abd, xsize);
+
+	/* transform the syndrome */
+	abd_iterate_func(xabd, 0, xsize, raidz_mul_abd_cb, (void *)coeff);
+
+	raidz_math_end();
+
+	return (1 << CODE_R);
+}
+
+
+/*
+ * Generate P and Q syndromes
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @tsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_pq_abd(void **tc, const void *dc, const size_t tsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)tc[TARGET_X];
+	v_t *y = (v_t *)tc[TARGET_Y];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const yend = y + (tsize / sizeof (v_t));
+
+	SYN_PQ_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += SYN_STRIDE, x += SYN_STRIDE, y += SYN_STRIDE) {
+		LOAD(d, SYN_PQ_D);
+		P_D_SYNDROME(SYN_PQ_D, SYN_PQ_X, x);
+		Q_D_SYNDROME(SYN_PQ_D, SYN_PQ_X, y);
+	}
+	for (; y < yend; y += SYN_STRIDE) {
+		Q_SYNDROME(SYN_PQ_X, y);
+	}
+}
+
+/*
+ * Reconstruct data using PQ parity and PQ syndromes
+ *
+ * @tc		syndrome/result columns
+ * @tsize	size of syndrome/result columns
+ * @c		parity columns
+ * @mul		array of multiplication constants
+ */
+static void
+raidz_rec_pq_abd(void **tc, const size_t tsize, void **c,
+    const unsigned *mul)
+{
+	v_t *x = (v_t *)tc[TARGET_X];
+	v_t *y = (v_t *)tc[TARGET_Y];
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+	const v_t *p = (v_t *)c[CODE_P];
+	const v_t *q = (v_t *)c[CODE_Q];
+
+	REC_PQ_DEFINE();
+
+	for (; x < xend; x += REC_PQ_STRIDE, y += REC_PQ_STRIDE,
+	    p += REC_PQ_STRIDE, q += REC_PQ_STRIDE) {
+		LOAD(x, REC_PQ_X);
+		LOAD(y, REC_PQ_Y);
+
+		XOR_ACC(p, REC_PQ_X);
+		XOR_ACC(q, REC_PQ_Y);
+
+		/* Save Pxy */
+		COPY(REC_PQ_X,  REC_PQ_T);
+
+		/* Calc X */
+		MUL(mul[MUL_PQ_X], REC_PQ_X);
+		MUL(mul[MUL_PQ_Y], REC_PQ_Y);
+		XOR(REC_PQ_Y,  REC_PQ_X);
+		STORE(x, REC_PQ_X);
+
+		/* Calc Y */
+		XOR(REC_PQ_T,  REC_PQ_X);
+		STORE(y, REC_PQ_X);
+	}
+}
+
+
+/*
+ * Reconstruct two data columns using PQ parity
+ *
+ * @syn_method	raidz_syn_pq_abd()
+ * @rec_method	raidz_rec_pq_abd()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_pq_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	const size_t y = tgtidx[TARGET_Y];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	const size_t ysize = rm->rm_col[y].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	abd_t *yabd = rm->rm_col[y].rc_abd;
+	abd_t *tabds[2] = { xabd, yabd };
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_P].rc_abd,
+		rm->rm_col[CODE_Q].rc_abd
+	};
+
+	unsigned coeff[MUL_CNT];
+	raidz_rec_pq_coeff(rm, tgtidx, coeff);
+
+	/*
+	 * Check if some of targets is shorter then others
+	 * In this case, shorter target needs to be replaced with
+	 * new buffer so that syndrome can be calculated.
+	 */
+	if (ysize < xsize) {
+		yabd = abd_alloc(xsize, B_FALSE);
+		tabds[1] = yabd;
+	}
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+		raidz_copy(yabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+		raidz_zero(yabd, xsize);
+	}
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x || c == y) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 2,
+		    raidz_syn_pq_abd);
+	}
+
+	abd_raidz_rec_iterate(cabds, tabds, xsize, 2, raidz_rec_pq_abd, coeff);
+
+	/* Copy shorter targets back to the original abd buffer */
+	if (ysize < xsize)
+		raidz_copy(rm->rm_col[y].rc_abd, yabd, ysize);
+
+	raidz_math_end();
+
+	if (ysize < xsize)
+		abd_free(yabd);
+
+	return ((1 << CODE_P) | (1 << CODE_Q));
+}
+
+
+/*
+ * Generate P and R syndromes
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @tsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_pr_abd(void **c, const void *dc, const size_t tsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)c[TARGET_X];
+	v_t *y = (v_t *)c[TARGET_Y];
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+	const v_t * const yend = y + (tsize / sizeof (v_t));
+
+	SYN_PR_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += SYN_STRIDE, x += SYN_STRIDE, y += SYN_STRIDE) {
+		LOAD(d, SYN_PR_D);
+		P_D_SYNDROME(SYN_PR_D, SYN_PR_X, x);
+		R_D_SYNDROME(SYN_PR_D, SYN_PR_X, y);
+	}
+	for (; y < yend; y += SYN_STRIDE) {
+		R_SYNDROME(SYN_PR_X, y);
+	}
+}
+
+/*
+ * Reconstruct data using PR parity and PR syndromes
+ *
+ * @tc		syndrome/result columns
+ * @tsize	size of syndrome/result columns
+ * @c		parity columns
+ * @mul		array of multiplication constants
+ */
+static void
+raidz_rec_pr_abd(void **t, const size_t tsize, void **c,
+    const unsigned *mul)
+{
+	v_t *x = (v_t *)t[TARGET_X];
+	v_t *y = (v_t *)t[TARGET_Y];
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+	const v_t *p = (v_t *)c[CODE_P];
+	const v_t *q = (v_t *)c[CODE_Q];
+
+	REC_PR_DEFINE();
+
+	for (; x < xend; x += REC_PR_STRIDE, y += REC_PR_STRIDE,
+	    p += REC_PR_STRIDE, q += REC_PR_STRIDE) {
+		LOAD(x, REC_PR_X);
+		LOAD(y, REC_PR_Y);
+		XOR_ACC(p, REC_PR_X);
+		XOR_ACC(q, REC_PR_Y);
+
+		/* Save Pxy */
+		COPY(REC_PR_X,  REC_PR_T);
+
+		/* Calc X */
+		MUL(mul[MUL_PR_X], REC_PR_X);
+		MUL(mul[MUL_PR_Y], REC_PR_Y);
+		XOR(REC_PR_Y,  REC_PR_X);
+		STORE(x, REC_PR_X);
+
+		/* Calc Y */
+		XOR(REC_PR_T,  REC_PR_X);
+		STORE(y, REC_PR_X);
+	}
+}
+
+
+/*
+ * Reconstruct two data columns using PR parity
+ *
+ * @syn_method	raidz_syn_pr_abd()
+ * @rec_method	raidz_rec_pr_abd()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_pr_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[0];
+	const size_t y = tgtidx[1];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	const size_t ysize = rm->rm_col[y].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	abd_t *yabd = rm->rm_col[y].rc_abd;
+	abd_t *tabds[2] = { xabd, yabd };
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_P].rc_abd,
+		rm->rm_col[CODE_R].rc_abd
+	};
+	unsigned coeff[MUL_CNT];
+	raidz_rec_pr_coeff(rm, tgtidx, coeff);
+
+	/*
+	 * Check if some of targets are shorter then others.
+	 * They need to be replaced with a new buffer so that syndrome can
+	 * be calculated on full length.
+	 */
+	if (ysize < xsize) {
+		yabd = abd_alloc(xsize, B_FALSE);
+		tabds[1] = yabd;
+	}
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+		raidz_copy(yabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+		raidz_zero(yabd, xsize);
+	}
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x || c == y) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 2,
+		    raidz_syn_pr_abd);
+	}
+
+	abd_raidz_rec_iterate(cabds, tabds, xsize, 2, raidz_rec_pr_abd, coeff);
+
+	/*
+	 * Copy shorter targets back to the original abd buffer
+	 */
+	if (ysize < xsize)
+		raidz_copy(rm->rm_col[y].rc_abd, yabd, ysize);
+
+	raidz_math_end();
+
+	if (ysize < xsize)
+		abd_free(yabd);
+
+	return ((1 << CODE_P) | (1 << CODE_Q));
+}
+
+
+/*
+ * Generate Q and R syndromes
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @tsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_qr_abd(void **c, const void *dc, const size_t tsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)c[TARGET_X];
+	v_t *y = (v_t *)c[TARGET_Y];
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+
+	SYN_QR_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend; d += SYN_STRIDE, x += SYN_STRIDE, y += SYN_STRIDE) {
+		LOAD(d, SYN_PQ_D);
+		Q_D_SYNDROME(SYN_QR_D, SYN_QR_X, x);
+		R_D_SYNDROME(SYN_QR_D, SYN_QR_X, y);
+	}
+	for (; x < xend; x += SYN_STRIDE, y += SYN_STRIDE) {
+		Q_SYNDROME(SYN_QR_X, x);
+		R_SYNDROME(SYN_QR_X, y);
+	}
+}
+
+
+/*
+ * Reconstruct data using QR parity and QR syndromes
+ *
+ * @tc		syndrome/result columns
+ * @tsize	size of syndrome/result columns
+ * @c		parity columns
+ * @mul		array of multiplication constants
+ */
+static void
+raidz_rec_qr_abd(void **t, const size_t tsize, void **c,
+    const unsigned *mul)
+{
+	v_t *x = (v_t *)t[TARGET_X];
+	v_t *y = (v_t *)t[TARGET_Y];
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+	const v_t *p = (v_t *)c[CODE_P];
+	const v_t *q = (v_t *)c[CODE_Q];
+
+	REC_QR_DEFINE();
+
+	for (; x < xend; x += REC_QR_STRIDE, y += REC_QR_STRIDE,
+	    p += REC_QR_STRIDE, q += REC_QR_STRIDE) {
+		LOAD(x, REC_QR_X);
+		LOAD(y, REC_QR_Y);
+
+		XOR_ACC(p, REC_QR_X);
+		XOR_ACC(q, REC_QR_Y);
+
+		/* Save Pxy */
+		COPY(REC_QR_X,  REC_QR_T);
+
+		/* Calc X */
+		MUL(mul[MUL_QR_XQ], REC_QR_X);	/* X = Q * xqm */
+		XOR(REC_QR_Y, REC_QR_X);	/* X = R ^ X   */
+		MUL(mul[MUL_QR_X], REC_QR_X);	/* X = X * xm  */
+		STORE(x, REC_QR_X);
+
+		/* Calc Y */
+		MUL(mul[MUL_QR_YQ], REC_QR_T);	/* X = Q * xqm */
+		XOR(REC_QR_Y, REC_QR_T);	/* X = R ^ X   */
+		MUL(mul[MUL_QR_Y], REC_QR_T);	/* X = X * xm  */
+		STORE(y, REC_QR_T);
+	}
+}
+
+
+/*
+ * Reconstruct two data columns using QR parity
+ *
+ * @syn_method	raidz_syn_qr_abd()
+ * @rec_method	raidz_rec_qr_abd()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_qr_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	const size_t y = tgtidx[TARGET_Y];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	const size_t ysize = rm->rm_col[y].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	abd_t *yabd = rm->rm_col[y].rc_abd;
+	abd_t *tabds[2] = { xabd, yabd };
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_Q].rc_abd,
+		rm->rm_col[CODE_R].rc_abd
+	};
+	unsigned coeff[MUL_CNT];
+	raidz_rec_qr_coeff(rm, tgtidx, coeff);
+
+	/*
+	 * Check if some of targets is shorter then others
+	 * In this case, shorter target needs to be replaced with
+	 * new buffer so that syndrome can be calculated.
+	 */
+	if (ysize < xsize) {
+		yabd = abd_alloc(xsize, B_FALSE);
+		tabds[1] = yabd;
+	}
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+		raidz_copy(yabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+		raidz_zero(yabd, xsize);
+	}
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x || c == y) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 2,
+		    raidz_syn_qr_abd);
+	}
+
+	abd_raidz_rec_iterate(cabds, tabds, xsize, 2, raidz_rec_qr_abd, coeff);
+
+	/*
+	 * Copy shorter targets back to the original abd buffer
+	 */
+	if (ysize < xsize)
+		raidz_copy(rm->rm_col[y].rc_abd, yabd, ysize);
+
+	raidz_math_end();
+
+	if (ysize < xsize)
+		abd_free(yabd);
+
+
+	return ((1 << CODE_Q) | (1 << CODE_R));
+}
+
+
+/*
+ * Generate P, Q, and R syndromes
+ *
+ * @xc		array of pointers to syndrome columns
+ * @dc		data column (NULL if missing)
+ * @tsize	size of syndrome columns
+ * @dsize	size of data column (0 if missing)
+ */
+static void
+raidz_syn_pqr_abd(void **c, const void *dc, const size_t tsize,
+    const size_t dsize)
+{
+	v_t *x = (v_t *)c[TARGET_X];
+	v_t *y = (v_t *)c[TARGET_Y];
+	v_t *z = (v_t *)c[TARGET_Z];
+	const v_t * const yend = y + (tsize / sizeof (v_t));
+	const v_t *d = (const v_t *)dc;
+	const v_t * const dend = d + (dsize / sizeof (v_t));
+
+	SYN_PQR_DEFINE();
+
+	MUL2_SETUP();
+
+	for (; d < dend;  d += SYN_STRIDE, x += SYN_STRIDE, y += SYN_STRIDE,
+	    z += SYN_STRIDE) {
+		LOAD(d, SYN_PQR_D);
+		P_D_SYNDROME(SYN_PQR_D, SYN_PQR_X, x)
+		Q_D_SYNDROME(SYN_PQR_D, SYN_PQR_X, y);
+		R_D_SYNDROME(SYN_PQR_D, SYN_PQR_X, z);
+	}
+	for (; y < yend; y += SYN_STRIDE, z += SYN_STRIDE) {
+		Q_SYNDROME(SYN_PQR_X, y);
+		R_SYNDROME(SYN_PQR_X, z);
+	}
+}
+
+
+/*
+ * Reconstruct data using PRQ parity and PQR syndromes
+ *
+ * @tc		syndrome/result columns
+ * @tsize	size of syndrome/result columns
+ * @c		parity columns
+ * @mul		array of multiplication constants
+ */
+static void
+raidz_rec_pqr_abd(void **t, const size_t tsize, void **c,
+    const unsigned * const mul)
+{
+	v_t *x = (v_t *)t[TARGET_X];
+	v_t *y = (v_t *)t[TARGET_Y];
+	v_t *z = (v_t *)t[TARGET_Z];
+	const v_t * const xend = x + (tsize / sizeof (v_t));
+	const v_t *p = (v_t *)c[CODE_P];
+	const v_t *q = (v_t *)c[CODE_Q];
+	const v_t *r = (v_t *)c[CODE_R];
+
+	REC_PQR_DEFINE();
+
+	for (; x < xend; x += REC_PQR_STRIDE, y += REC_PQR_STRIDE,
+	    z += REC_PQR_STRIDE, p += REC_PQR_STRIDE, q += REC_PQR_STRIDE,
+	    r += REC_PQR_STRIDE) {
+		LOAD(x, REC_PQR_X);
+		LOAD(y, REC_PQR_Y);
+		LOAD(z, REC_PQR_Z);
+
+		XOR_ACC(p, REC_PQR_X);
+		XOR_ACC(q, REC_PQR_Y);
+		XOR_ACC(r, REC_PQR_Z);
+
+		/* Save Pxyz and Qxyz */
+		COPY(REC_PQR_X, REC_PQR_XS);
+		COPY(REC_PQR_Y, REC_PQR_YS);
+
+		/* Calc X */
+		MUL(mul[MUL_PQR_XP], REC_PQR_X);	/* Xp = Pxyz * xp   */
+		MUL(mul[MUL_PQR_XQ], REC_PQR_Y);	/* Xq = Qxyz * xq   */
+		XOR(REC_PQR_Y, REC_PQR_X);
+		MUL(mul[MUL_PQR_XR], REC_PQR_Z);	/* Xr = Rxyz * xr   */
+		XOR(REC_PQR_Z, REC_PQR_X);		/* X = Xp + Xq + Xr */
+		STORE(x, REC_PQR_X);
+
+		/* Calc Y */
+		XOR(REC_PQR_X, REC_PQR_XS); 		/* Pyz = Pxyz + X */
+		MUL(mul[MUL_PQR_YU], REC_PQR_X);  	/* Xq = X * upd_q */
+		XOR(REC_PQR_X, REC_PQR_YS); 		/* Qyz = Qxyz + Xq */
+		COPY(REC_PQR_XS, REC_PQR_X);		/* restore Pyz */
+		MUL(mul[MUL_PQR_YP], REC_PQR_X);	/* Yp = Pyz * yp */
+		MUL(mul[MUL_PQR_YQ], REC_PQR_YS);	/* Yq = Qyz * yq */
+		XOR(REC_PQR_X, REC_PQR_YS); 		/* Y = Yp + Yq */
+		STORE(y, REC_PQR_YS);
+
+		/* Calc Z */
+		XOR(REC_PQR_XS, REC_PQR_YS);		/* Z = Pz = Pyz + Y */
+		STORE(z, REC_PQR_YS);
+	}
+}
+
+
+/*
+ * Reconstruct three data columns using PQR parity
+ *
+ * @syn_method	raidz_syn_pqr_abd()
+ * @rec_method	raidz_rec_pqr_abd()
+ *
+ * @rm		RAIDZ map
+ * @tgtidx	array of missing data indexes
+ */
+static raidz_inline int
+raidz_reconstruct_pqr_impl(raidz_map_t *rm, const int *tgtidx)
+{
+	size_t c;
+	size_t dsize;
+	abd_t *dabd;
+	const size_t firstdc = raidz_parity(rm);
+	const size_t ncols = raidz_ncols(rm);
+	const size_t x = tgtidx[TARGET_X];
+	const size_t y = tgtidx[TARGET_Y];
+	const size_t z = tgtidx[TARGET_Z];
+	const size_t xsize = rm->rm_col[x].rc_size;
+	const size_t ysize = rm->rm_col[y].rc_size;
+	const size_t zsize = rm->rm_col[z].rc_size;
+	abd_t *xabd = rm->rm_col[x].rc_abd;
+	abd_t *yabd = rm->rm_col[y].rc_abd;
+	abd_t *zabd = rm->rm_col[z].rc_abd;
+	abd_t *tabds[] = { xabd, yabd, zabd };
+	abd_t *cabds[] = {
+		rm->rm_col[CODE_P].rc_abd,
+		rm->rm_col[CODE_Q].rc_abd,
+		rm->rm_col[CODE_R].rc_abd
+	};
+	unsigned coeff[MUL_CNT];
+	raidz_rec_pqr_coeff(rm, tgtidx, coeff);
+
+	/*
+	 * Check if some of targets is shorter then others
+	 * In this case, shorter target needs to be replaced with
+	 * new buffer so that syndrome can be calculated.
+	 */
+	if (ysize < xsize) {
+		yabd = abd_alloc(xsize, B_FALSE);
+		tabds[1] = yabd;
+	}
+	if (zsize < xsize) {
+		zabd = abd_alloc(xsize, B_FALSE);
+		tabds[2] = zabd;
+	}
+
+	raidz_math_begin();
+
+	/* Start with first data column if present */
+	if (firstdc != x) {
+		raidz_copy(xabd, rm->rm_col[firstdc].rc_abd, xsize);
+		raidz_copy(yabd, rm->rm_col[firstdc].rc_abd, xsize);
+		raidz_copy(zabd, rm->rm_col[firstdc].rc_abd, xsize);
+	} else {
+		raidz_zero(xabd, xsize);
+		raidz_zero(yabd, xsize);
+		raidz_zero(zabd, xsize);
+	}
+
+	/* generate q_syndrome */
+	for (c = firstdc+1; c < ncols; c++) {
+		if (c == x || c == y || c == z) {
+			dabd = NULL;
+			dsize = 0;
+		} else {
+			dabd = rm->rm_col[c].rc_abd;
+			dsize = rm->rm_col[c].rc_size;
+		}
+
+		abd_raidz_gen_iterate(tabds, dabd, xsize, dsize, 3,
+		    raidz_syn_pqr_abd);
+	}
+
+	abd_raidz_rec_iterate(cabds, tabds, xsize, 3, raidz_rec_pqr_abd, coeff);
+
+	/*
+	 * Copy shorter targets back to the original abd buffer
+	 */
+	if (ysize < xsize)
+		raidz_copy(rm->rm_col[y].rc_abd, yabd, ysize);
+	if (zsize < xsize)
+		raidz_copy(rm->rm_col[z].rc_abd, zabd, zsize);
+
+	raidz_math_end();
+
+	if (ysize < xsize)
+		abd_free(yabd);
+	if (zsize < xsize)
+		abd_free(zabd);
+
+	return ((1 << CODE_P) | (1 << CODE_Q) | (1 << CODE_R));
+}
+
+#endif /* _VDEV_RAIDZ_MATH_IMPL_H */
diff --git a/usr/src/uts/common/fs/zfs/vdev_raidz_math_scalar.c b/usr/src/uts/common/fs/zfs/vdev_raidz_math_scalar.c
new file mode 100644
index 0000000000..cd742e146c
--- /dev/null
+++ b/usr/src/uts/common/fs/zfs/vdev_raidz_math_scalar.c
@@ -0,0 +1,337 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
+ */
+
+#include <sys/vdev_raidz_impl.h>
+
+/*
+ * Provide native CPU scalar routines.
+ * Support 32bit and 64bit CPUs.
+ */
+#if ((~(0x0ULL)) >> 24) == 0xffULL
+#define	ELEM_SIZE	4
+typedef uint32_t iv_t;
+#elif ((~(0x0ULL)) >> 56) == 0xffULL
+#define	ELEM_SIZE	8
+typedef uint64_t iv_t;
+#endif
+
+/*
+ * Vector type used in scalar implementation
+ *
+ * The union is expected to be of native CPU register size. Since addition
+ * uses XOR operation, it can be performed an all byte elements at once.
+ * Multiplication requires per byte access.
+ */
+typedef union {
+	iv_t e;
+	uint8_t b[ELEM_SIZE];
+} v_t;
+
+/*
+ * Precomputed lookup tables for multiplication by a constant
+ *
+ * Reconstruction path requires multiplication by a constant factors. Instead of
+ * performing two step lookup (log & exp tables), a direct lookup can be used
+ * instead. Multiplication of element 'a' by a constant 'c' is obtained as:
+ *
+ * 	r = vdev_raidz_mul_lt[c_log][a];
+ *
+ * where c_log = vdev_raidz_log2[c]. Log of coefficient factors is used because
+ * they are faster to obtain while solving the syndrome equations.
+ *
+ * PERFORMANCE NOTE:
+ * Even though the complete lookup table uses 64kiB, only relatively small
+ * portion of it is used at the same time. Following shows number of accessed
+ * bytes for different cases:
+ * 	- 1 failed disk: 256B (1 mul. coefficient)
+ * 	- 2 failed disks: 512B (2 mul. coefficients)
+ * 	- 3 failed disks: 1536B (6 mul. coefficients)
+ *
+ * Size of actually accessed lookup table regions is only larger for
+ * reconstruction of 3 failed disks, when compared to traditional log/exp
+ * method. But since the result is obtained in one lookup step performance is
+ * doubled.
+ */
+static uint8_t vdev_raidz_mul_lt[256][256] __attribute__((aligned(256)));
+
+static void
+raidz_init_scalar(void)
+{
+	int c, i;
+	for (c = 0; c < 256; c++)
+		for (i = 0; i < 256; i++)
+			vdev_raidz_mul_lt[c][i] = gf_mul(c, i);
+
+}
+
+#define	PREFETCHNTA(ptr, offset)	{}
+#define	PREFETCH(ptr, offset) 		{}
+
+#define	XOR_ACC(src, acc)	acc.e ^= ((v_t *)src)[0].e
+#define	XOR(src, acc)		acc.e ^= src.e
+#define	ZERO(acc)		acc.e = 0
+#define	COPY(src, dst)		dst = src
+#define	LOAD(src, val) 		val = ((v_t *)src)[0]
+#define	STORE(dst, val)		((v_t *)dst)[0] = val
+
+/*
+ * Constants used for optimized multiplication by 2.
+ */
+static const struct {
+	iv_t mod;
+	iv_t mask;
+	iv_t msb;
+} scalar_mul2_consts = {
+#if ELEM_SIZE == 8
+	.mod	= 0x1d1d1d1d1d1d1d1dULL,
+	.mask	= 0xfefefefefefefefeULL,
+	.msb	= 0x8080808080808080ULL,
+#else
+	.mod	= 0x1d1d1d1dULL,
+	.mask	= 0xfefefefeULL,
+	.msb	= 0x80808080ULL,
+#endif
+};
+
+#define	MUL2_SETUP() {}
+
+#define	MUL2(a)								\
+{									\
+	iv_t _mask;							\
+									\
+	_mask = (a).e & scalar_mul2_consts.msb;				\
+	_mask = (_mask << 1) - (_mask >> 7);				\
+	(a).e = ((a).e << 1) & scalar_mul2_consts.mask;			\
+	(a).e = (a).e ^ (_mask & scalar_mul2_consts.mod);		\
+}
+
+#define	MUL4(a) 							\
+{									\
+	MUL2(a);							\
+	MUL2(a);							\
+}
+
+#define	MUL(c, a)							\
+{									\
+	const uint8_t *mul_lt = vdev_raidz_mul_lt[c];			\
+	switch (ELEM_SIZE) {						\
+	case 8:								\
+		a.b[7] = mul_lt[a.b[7]];				\
+		a.b[6] = mul_lt[a.b[6]];				\
+		a.b[5] = mul_lt[a.b[5]];				\
+		a.b[4] = mul_lt[a.b[4]];				\
+		/* falls through */					\
+	case 4:								\
+		a.b[3] = mul_lt[a.b[3]];				\
+		a.b[2] = mul_lt[a.b[2]];				\
+		a.b[1] = mul_lt[a.b[1]];				\
+		a.b[0] = mul_lt[a.b[0]];				\
+		break;							\
+	}								\
+}
+
+#define	raidz_math_begin()	{}
+#define	raidz_math_end()	{}
+
+#define	SYN_STRIDE		1
+
+#define	ZERO_DEFINE()		v_t d0
+#define	ZERO_STRIDE		1
+#define	ZERO_D			d0
+
+#define	COPY_DEFINE()		v_t d0
+#define	COPY_STRIDE		1
+#define	COPY_D			d0
+
+#define	ADD_DEFINE()		v_t d0
+#define	ADD_STRIDE		1
+#define	ADD_D			d0
+
+#define	MUL_DEFINE()		v_t d0
+#define	MUL_STRIDE		1
+#define	MUL_D			d0
+
+#define	GEN_P_STRIDE		1
+#define	GEN_P_DEFINE()		v_t p0
+#define	GEN_P_P			p0
+
+#define	GEN_PQ_STRIDE		1
+#define	GEN_PQ_DEFINE()		v_t d0, c0
+#define	GEN_PQ_D		d0
+#define	GEN_PQ_C		c0
+
+#define	GEN_PQR_STRIDE		1
+#define	GEN_PQR_DEFINE()	v_t d0, c0
+#define	GEN_PQR_D		d0
+#define	GEN_PQR_C		c0
+
+#define	SYN_Q_DEFINE()		v_t d0, x0
+#define	SYN_Q_D			d0
+#define	SYN_Q_X			x0
+
+
+#define	SYN_R_DEFINE()		v_t d0, x0
+#define	SYN_R_D			d0
+#define	SYN_R_X			x0
+
+
+#define	SYN_PQ_DEFINE()		v_t d0, x0
+#define	SYN_PQ_D		d0
+#define	SYN_PQ_X		x0
+
+
+#define	REC_PQ_STRIDE		1
+#define	REC_PQ_DEFINE()		v_t x0, y0, t0
+#define	REC_PQ_X		x0
+#define	REC_PQ_Y		y0
+#define	REC_PQ_T		t0
+
+
+#define	SYN_PR_DEFINE()		v_t d0, x0
+#define	SYN_PR_D		d0
+#define	SYN_PR_X		x0
+
+#define	REC_PR_STRIDE		1
+#define	REC_PR_DEFINE()		v_t x0, y0, t0
+#define	REC_PR_X		x0
+#define	REC_PR_Y		y0
+#define	REC_PR_T		t0
+
+
+#define	SYN_QR_DEFINE()		v_t d0, x0
+#define	SYN_QR_D		d0
+#define	SYN_QR_X		x0
+
+
+#define	REC_QR_STRIDE		1
+#define	REC_QR_DEFINE()		v_t x0, y0, t0
+#define	REC_QR_X		x0
+#define	REC_QR_Y		y0
+#define	REC_QR_T		t0
+
+
+#define	SYN_PQR_DEFINE()	v_t d0, x0
+#define	SYN_PQR_D		d0
+#define	SYN_PQR_X		x0
+
+#define	REC_PQR_STRIDE		1
+#define	REC_PQR_DEFINE()	v_t x0, y0, z0, xs0, ys0
+#define	REC_PQR_X		x0
+#define	REC_PQR_Y		y0
+#define	REC_PQR_Z		z0
+#define	REC_PQR_XS		xs0
+#define	REC_PQR_YS		ys0
+
+#include "vdev_raidz_math_impl.h"
+
+DEFINE_GEN_METHODS(scalar);
+DEFINE_REC_METHODS(scalar);
+
+boolean_t
+raidz_will_scalar_work(void)
+{
+	return (B_TRUE); /* always */
+}
+
+const raidz_impl_ops_t vdev_raidz_scalar_impl = {
+	.init = raidz_init_scalar,
+	.fini = NULL,
+	.gen = RAIDZ_GEN_METHODS(scalar),
+	.rec = RAIDZ_REC_METHODS(scalar),
+	.is_supported = &raidz_will_scalar_work,
+	.name = "scalar"
+};
+
+/* Powers of 2 in the RAID-Z Galois field. */
+const uint8_t vdev_raidz_pow2[256] __attribute__((aligned(256))) = {
+	0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
+	0x1d, 0x3a, 0x74, 0xe8, 0xcd, 0x87, 0x13, 0x26,
+	0x4c, 0x98, 0x2d, 0x5a, 0xb4, 0x75, 0xea, 0xc9,
+	0x8f, 0x03, 0x06, 0x0c, 0x18, 0x30, 0x60, 0xc0,
+	0x9d, 0x27, 0x4e, 0x9c, 0x25, 0x4a, 0x94, 0x35,
+	0x6a, 0xd4, 0xb5, 0x77, 0xee, 0xc1, 0x9f, 0x23,
+	0x46, 0x8c, 0x05, 0x0a, 0x14, 0x28, 0x50, 0xa0,
+	0x5d, 0xba, 0x69, 0xd2, 0xb9, 0x6f, 0xde, 0xa1,
+	0x5f, 0xbe, 0x61, 0xc2, 0x99, 0x2f, 0x5e, 0xbc,
+	0x65, 0xca, 0x89, 0x0f, 0x1e, 0x3c, 0x78, 0xf0,
+	0xfd, 0xe7, 0xd3, 0xbb, 0x6b, 0xd6, 0xb1, 0x7f,
+	0xfe, 0xe1, 0xdf, 0xa3, 0x5b, 0xb6, 0x71, 0xe2,
+	0xd9, 0xaf, 0x43, 0x86, 0x11, 0x22, 0x44, 0x88,
+	0x0d, 0x1a, 0x34, 0x68, 0xd0, 0xbd, 0x67, 0xce,
+	0x81, 0x1f, 0x3e, 0x7c, 0xf8, 0xed, 0xc7, 0x93,
+	0x3b, 0x76, 0xec, 0xc5, 0x97, 0x33, 0x66, 0xcc,
+	0x85, 0x17, 0x2e, 0x5c, 0xb8, 0x6d, 0xda, 0xa9,
+	0x4f, 0x9e, 0x21, 0x42, 0x84, 0x15, 0x2a, 0x54,
+	0xa8, 0x4d, 0x9a, 0x29, 0x52, 0xa4, 0x55, 0xaa,
+	0x49, 0x92, 0x39, 0x72, 0xe4, 0xd5, 0xb7, 0x73,
+	0xe6, 0xd1, 0xbf, 0x63, 0xc6, 0x91, 0x3f, 0x7e,
+	0xfc, 0xe5, 0xd7, 0xb3, 0x7b, 0xf6, 0xf1, 0xff,
+	0xe3, 0xdb, 0xab, 0x4b, 0x96, 0x31, 0x62, 0xc4,
+	0x95, 0x37, 0x6e, 0xdc, 0xa5, 0x57, 0xae, 0x41,
+	0x82, 0x19, 0x32, 0x64, 0xc8, 0x8d, 0x07, 0x0e,
+	0x1c, 0x38, 0x70, 0xe0, 0xdd, 0xa7, 0x53, 0xa6,
+	0x51, 0xa2, 0x59, 0xb2, 0x79, 0xf2, 0xf9, 0xef,
+	0xc3, 0x9b, 0x2b, 0x56, 0xac, 0x45, 0x8a, 0x09,
+	0x12, 0x24, 0x48, 0x90, 0x3d, 0x7a, 0xf4, 0xf5,
+	0xf7, 0xf3, 0xfb, 0xeb, 0xcb, 0x8b, 0x0b, 0x16,
+	0x2c, 0x58, 0xb0, 0x7d, 0xfa, 0xe9, 0xcf, 0x83,
+	0x1b, 0x36, 0x6c, 0xd8, 0xad, 0x47, 0x8e, 0x01
+};
+
+/* Logs of 2 in the RAID-Z Galois field. */
+const uint8_t vdev_raidz_log2[256] __attribute__((aligned(256))) = {
+	0x00, 0x00, 0x01, 0x19, 0x02, 0x32, 0x1a, 0xc6,
+	0x03, 0xdf, 0x33, 0xee, 0x1b, 0x68, 0xc7, 0x4b,
+	0x04, 0x64, 0xe0, 0x0e, 0x34, 0x8d, 0xef, 0x81,
+	0x1c, 0xc1, 0x69, 0xf8, 0xc8, 0x08, 0x4c, 0x71,
+	0x05, 0x8a, 0x65, 0x2f, 0xe1, 0x24, 0x0f, 0x21,
+	0x35, 0x93, 0x8e, 0xda, 0xf0, 0x12, 0x82, 0x45,
+	0x1d, 0xb5, 0xc2, 0x7d, 0x6a, 0x27, 0xf9, 0xb9,
+	0xc9, 0x9a, 0x09, 0x78, 0x4d, 0xe4, 0x72, 0xa6,
+	0x06, 0xbf, 0x8b, 0x62, 0x66, 0xdd, 0x30, 0xfd,
+	0xe2, 0x98, 0x25, 0xb3, 0x10, 0x91, 0x22, 0x88,
+	0x36, 0xd0, 0x94, 0xce, 0x8f, 0x96, 0xdb, 0xbd,
+	0xf1, 0xd2, 0x13, 0x5c, 0x83, 0x38, 0x46, 0x40,
+	0x1e, 0x42, 0xb6, 0xa3, 0xc3, 0x48, 0x7e, 0x6e,
+	0x6b, 0x3a, 0x28, 0x54, 0xfa, 0x85, 0xba, 0x3d,
+	0xca, 0x5e, 0x9b, 0x9f, 0x0a, 0x15, 0x79, 0x2b,
+	0x4e, 0xd4, 0xe5, 0xac, 0x73, 0xf3, 0xa7, 0x57,
+	0x07, 0x70, 0xc0, 0xf7, 0x8c, 0x80, 0x63, 0x0d,
+	0x67, 0x4a, 0xde, 0xed, 0x31, 0xc5, 0xfe, 0x18,
+	0xe3, 0xa5, 0x99, 0x77, 0x26, 0xb8, 0xb4, 0x7c,
+	0x11, 0x44, 0x92, 0xd9, 0x23, 0x20, 0x89, 0x2e,
+	0x37, 0x3f, 0xd1, 0x5b, 0x95, 0xbc, 0xcf, 0xcd,
+	0x90, 0x87, 0x97, 0xb2, 0xdc, 0xfc, 0xbe, 0x61,
+	0xf2, 0x56, 0xd3, 0xab, 0x14, 0x2a, 0x5d, 0x9e,
+	0x84, 0x3c, 0x39, 0x53, 0x47, 0x6d, 0x41, 0xa2,
+	0x1f, 0x2d, 0x43, 0xd8, 0xb7, 0x7b, 0xa4, 0x76,
+	0xc4, 0x17, 0x49, 0xec, 0x7f, 0x0c, 0x6f, 0xf6,
+	0x6c, 0xa1, 0x3b, 0x52, 0x29, 0x9d, 0x55, 0xaa,
+	0xfb, 0x60, 0x86, 0xb1, 0xbb, 0xcc, 0x3e, 0x5a,
+	0xcb, 0x59, 0x5f, 0xb0, 0x9c, 0xa9, 0xa0, 0x51,
+	0x0b, 0xf5, 0x16, 0xeb, 0x7a, 0x75, 0x2c, 0xd7,
+	0x4f, 0xae, 0xd5, 0xe9, 0xe6, 0xe7, 0xad, 0xe8,
+	0x74, 0xd6, 0xf4, 0xea, 0xa8, 0x50, 0x58, 0xaf,
+};
diff --git a/usr/src/uts/common/fs/zfs/zfs_vfsops.c b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
index 47c4a190cf..d06cb875cf 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vfsops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
@@ -26,6 +26,7 @@
  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
  * Copyright 2019 Joyent, Inc.
  * Copyright 2020 Joshua M. Clulow <josh@sysmgr.org>
+ * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -2229,18 +2230,34 @@ zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
 		 * Our count is maintained in the vfs structure, but the
 		 * number is off by 1 to indicate a hold on the vfs
 		 * structure itself.
-		 *
-		 * The '.zfs' directory maintains a reference of its
-		 * own, and any active references underneath are
-		 * reflected in the vnode count.
 		 */
-		if (zfsvfs->z_ctldir == NULL) {
-			if (vfsp->vfs_count > 1)
-				return (SET_ERROR(EBUSY));
-		} else {
-			if (vfsp->vfs_count > 2 ||
-			    zfsvfs->z_ctldir->v_count > 1)
-				return (SET_ERROR(EBUSY));
+		boolean_t draining;
+		uint_t thresh = 1;
+
+		/*
+		 * The '.zfs' directory maintains a reference of its own, and
+		 * any active references underneath are reflected in the vnode
+		 * count. Allow one additional reference for it.
+		 */
+		if (zfsvfs->z_ctldir != NULL)
+			thresh++;
+
+		/*
+		 * If it's running, the asynchronous unlinked drain task needs
+		 * to be stopped before the number of active vnodes can be
+		 * reliably checked.
+		 */
+		draining = zfsvfs->z_draining;
+		if (draining)
+			zfs_unlinked_drain_stop_wait(zfsvfs);
+
+		if (vfsp->vfs_count > thresh || (zfsvfs->z_ctldir != NULL &&
+		    zfsvfs->z_ctldir->v_count > 1)) {
+			if (draining) {
+				/* If it was draining, restart the task */
+				zfs_unlinked_drain(zfsvfs);
+			}
+			return (SET_ERROR(EBUSY));
 		}
 	}
 
diff --git a/usr/src/uts/common/io/ixgbe/ixgbe_osdep.h b/usr/src/uts/common/io/ixgbe/ixgbe_osdep.h
index a8df3df932..ce0cc20879 100644
--- a/usr/src/uts/common/io/ixgbe/ixgbe_osdep.h
+++ b/usr/src/uts/common/io/ixgbe/ixgbe_osdep.h
@@ -113,7 +113,7 @@ boolean_t ixgbe_removed(struct ixgbe_hw *);
 #ifdef _BIG_ENDIAN
 #define	IXGBE_CPU_TO_LE16	BSWAP_16
 #define	IXGBE_CPU_TO_LE32	BSWAP_32
-#define	IXGBE_LE32_TO_CPUS	BSWAP_32
+#define	IXGBE_LE32_TO_CPUS(x)	*(x) = BSWAP_32(*(x))
 #define	IXGBE_CPU_TO_BE16(x)	(x)
 #define	IXGBE_CPU_TO_BE32(x)	(x)
 #define	IXGBE_BE32_TO_CPU(x)	(x)
diff --git a/usr/src/uts/common/sys/sysevent.h b/usr/src/uts/common/sys/sysevent.h
index 392b426977..c2be00ad27 100644
--- a/usr/src/uts/common/sys/sysevent.h
+++ b/usr/src/uts/common/sys/sysevent.h
@@ -73,7 +73,7 @@ extern "C" {
 #define	SE_KERN_PUB	"kern:"
 #define	SUNW_KERN_PUB	SUNW_VENDOR ":" SE_KERN_PUB
 #define	SUNW_USR_PUB	SUNW_VENDOR ":" SE_USR_PUB
-#define	ILLUMOS_KERN_PUB	ILLUMOS_VENDOR":"SE_KERN_PUB
+#define	ILLUMOS_KERN_PUB	ILLUMOS_VENDOR ":" SE_KERN_PUB
 
 /*
  * Event header and attribute value limits
diff --git a/usr/src/uts/i86pc/io/vmm/vmm_sol_dev.c b/usr/src/uts/i86pc/io/vmm/vmm_sol_dev.c
index 634575427e..ec338b1cdd 100644
--- a/usr/src/uts/i86pc/io/vmm/vmm_sol_dev.c
+++ b/usr/src/uts/i86pc/io/vmm/vmm_sol_dev.c
@@ -12,6 +12,7 @@
 /*
  * Copyright 2015 Pluribus Networks Inc.
  * Copyright 2019 Joyent, Inc.
+ * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
  */
 
 #include <sys/types.h>
@@ -65,6 +66,7 @@
 static kmutex_t		vmmdev_mtx;
 static dev_info_t	*vmmdev_dip;
 static hma_reg_t	*vmmdev_hma_reg;
+static uint_t		vmmdev_hma_ref;
 static sdev_plugin_hdl_t vmmdev_sdev_hdl;
 
 static kmutex_t		vmm_mtx;
@@ -1322,6 +1324,56 @@ vmm_lookup(const char *name)
 	return (sc);
 }
 
+/*
+ * Acquire an HMA registration if not already held.
+ */
+static boolean_t
+vmm_hma_acquire(void)
+{
+	ASSERT(MUTEX_NOT_HELD(&vmm_mtx));
+
+	mutex_enter(&vmmdev_mtx);
+
+	if (vmmdev_hma_reg == NULL) {
+		VERIFY3U(vmmdev_hma_ref, ==, 0);
+		vmmdev_hma_reg = hma_register(vmmdev_hvm_name);
+		if (vmmdev_hma_reg == NULL) {
+			cmn_err(CE_WARN, "%s HMA registration failed.",
+			    vmmdev_hvm_name);
+			mutex_exit(&vmmdev_mtx);
+			return (B_FALSE);
+		}
+	}
+
+	vmmdev_hma_ref++;
+
+	mutex_exit(&vmmdev_mtx);
+
+	return (B_TRUE);
+}
+
+/*
+ * Release the HMA registration if held and there are no remaining VMs.
+ */
+static void
+vmm_hma_release(void)
+{
+	ASSERT(MUTEX_NOT_HELD(&vmm_mtx));
+
+	mutex_enter(&vmmdev_mtx);
+
+	VERIFY3U(vmmdev_hma_ref, !=, 0);
+
+	vmmdev_hma_ref--;
+
+	if (vmmdev_hma_ref == 0) {
+		VERIFY(vmmdev_hma_reg != NULL);
+		hma_unregister(vmmdev_hma_reg);
+		vmmdev_hma_reg = NULL;
+	}
+	mutex_exit(&vmmdev_mtx);
+}
+
 static int
 vmmdev_do_vm_create(char *name, cred_t *cr)
 {
@@ -1333,11 +1385,15 @@ vmmdev_do_vm_create(char *name, cred_t *cr)
 		return (EINVAL);
 	}
 
+	if (!vmm_hma_acquire())
+		return (ENXIO);
+
 	mutex_enter(&vmm_mtx);
 
-	/* Look for duplicates names */
+	/* Look for duplicate names */
 	if (vmm_lookup(name) != NULL) {
 		mutex_exit(&vmm_mtx);
+		vmm_hma_release();
 		return (EEXIST);
 	}
 
@@ -1347,6 +1403,7 @@ vmmdev_do_vm_create(char *name, cred_t *cr)
 		    sc = list_next(&vmm_list, sc)) {
 			if (sc->vmm_zone == curzone) {
 				mutex_exit(&vmm_mtx);
+				vmm_hma_release();
 				return (EINVAL);
 			}
 		}
@@ -1397,6 +1454,7 @@ fail:
 		ddi_soft_state_free(vmm_statep, minor);
 	}
 	mutex_exit(&vmm_mtx);
+	vmm_hma_release();
 
 	return (error);
 }
@@ -1685,13 +1743,16 @@ done:
 }
 
 static int
-vmm_do_vm_destroy_locked(vmm_softc_t *sc, boolean_t clean_zsd)
+vmm_do_vm_destroy_locked(vmm_softc_t *sc, boolean_t clean_zsd,
+    boolean_t *hma_release)
 {
 	dev_info_t	*pdip = ddi_get_parent(vmmdev_dip);
 	minor_t		minor;
 
 	ASSERT(MUTEX_HELD(&vmm_mtx));
 
+	*hma_release = B_FALSE;
+
 	if (clean_zsd) {
 		vmm_zsd_rem_vm(sc);
 	}
@@ -1714,6 +1775,7 @@ vmm_do_vm_destroy_locked(vmm_softc_t *sc, boolean_t clean_zsd)
 		vm_destroy(sc->vmm_vm);
 		ddi_soft_state_free(vmm_statep, minor);
 		id_free(vmm_minors, minor);
+		*hma_release = B_TRUE;
 	}
 	(void) devfs_clean(pdip, NULL, DV_CLEAN_FORCE);
 
@@ -1723,12 +1785,16 @@ vmm_do_vm_destroy_locked(vmm_softc_t *sc, boolean_t clean_zsd)
 int
 vmm_do_vm_destroy(vmm_softc_t *sc, boolean_t clean_zsd)
 {
+	boolean_t	hma_release = B_FALSE;
 	int		err;
 
 	mutex_enter(&vmm_mtx);
-	err = vmm_do_vm_destroy_locked(sc, clean_zsd);
+	err = vmm_do_vm_destroy_locked(sc, clean_zsd, &hma_release);
 	mutex_exit(&vmm_mtx);
 
+	if (hma_release)
+		vmm_hma_release();
+
 	return (err);
 }
 
@@ -1736,6 +1802,7 @@ vmm_do_vm_destroy(vmm_softc_t *sc, boolean_t clean_zsd)
 static int
 vmmdev_do_vm_destroy(const char *name, cred_t *cr)
 {
+	boolean_t	hma_release = B_FALSE;
 	vmm_softc_t	*sc;
 	int		err;
 
@@ -1756,13 +1823,16 @@ vmmdev_do_vm_destroy(const char *name, cred_t *cr)
 		mutex_exit(&vmm_mtx);
 		return (EPERM);
 	}
-	err = vmm_do_vm_destroy_locked(sc, B_TRUE);
+	err = vmm_do_vm_destroy_locked(sc, B_TRUE, &hma_release);
+
 	mutex_exit(&vmm_mtx);
 
+	if (hma_release)
+		vmm_hma_release();
+
 	return (err);
 }
 
-
 static int
 vmm_open(dev_t *devp, int flag, int otyp, cred_t *credp)
 {
@@ -1799,6 +1869,7 @@ vmm_close(dev_t dev, int flag, int otyp, cred_t *credp)
 {
 	minor_t		minor;
 	vmm_softc_t	*sc;
+	boolean_t	hma_release = B_FALSE;
 
 	minor = getminor(dev);
 	if (minor == VMM_CTL_MINOR)
@@ -1814,14 +1885,22 @@ vmm_close(dev_t dev, int flag, int otyp, cred_t *credp)
 	VERIFY(sc->vmm_is_open);
 	sc->vmm_is_open = B_FALSE;
 
+	/*
+	 * If this VM was destroyed while the vmm device was open, then
+	 * clean it up now that it is closed.
+	 */
 	if (sc->vmm_flags & VMM_DESTROY) {
 		list_remove(&vmm_destroy_list, sc);
 		vm_destroy(sc->vmm_vm);
 		ddi_soft_state_free(vmm_statep, minor);
 		id_free(vmm_minors, minor);
+		hma_release = B_TRUE;
 	}
 	mutex_exit(&vmm_mtx);
 
+	if (hma_release)
+		vmm_hma_release();
+
 	return (0);
 }
 
@@ -2076,12 +2155,18 @@ vmm_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
 	vmm_sol_glue_init();
 	vmm_arena_init();
 
+	/*
+	 * Perform temporary HMA registration to determine if the system
+	 * is capable.
+	 */
 	if ((reg = hma_register(vmmdev_hvm_name)) == NULL) {
 		goto fail;
 	} else if (vmm_mod_load() != 0) {
 		goto fail;
 	}
 	vmm_loaded = B_TRUE;
+	hma_unregister(reg);
+	reg = NULL;
 
 	/* Create control node.  Other nodes will be created on demand. */
 	if (ddi_create_minor_node(dip, "ctl", S_IFCHR,
@@ -2096,7 +2181,6 @@ vmm_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
 	}
 
 	ddi_report_dev(dip);
-	vmmdev_hma_reg = reg;
 	vmmdev_sdev_hdl = sph;
 	vmmdev_dip = dip;
 	mutex_exit(&vmmdev_mtx);
@@ -2153,8 +2237,7 @@ vmm_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
 	vmmdev_dip = NULL;
 
 	VERIFY0(vmm_mod_unload());
-	hma_unregister(vmmdev_hma_reg);
-	vmmdev_hma_reg = NULL;
+	VERIFY3U(vmmdev_hma_reg, ==, NULL);
 	vmm_arena_fini();
 	vmm_sol_glue_cleanup();
 
diff --git a/usr/src/uts/i86pc/ppt/Makefile b/usr/src/uts/i86pc/ppt/Makefile
index 7c41368efd..7c10482e98 100644
--- a/usr/src/uts/i86pc/ppt/Makefile
+++ b/usr/src/uts/i86pc/ppt/Makefile
@@ -44,8 +44,8 @@ INSTALL_TARGET	= $(BINARY) $(ROOTMODULE) $(ROOT_CONFFILE)
 #
 ALL_BUILDS	= $(ALL_BUILDSONLY64)
 DEF_BUILDS	= $(DEF_BUILDSONLY64)
-PRE_INC_PATH	= -I$(COMPAT)/freebsd -I$(COMPAT)/freebsd/amd64 \
-	-I$(CONTRIB)/freebsd -I$(CONTRIB)/freebsd/amd64
+PRE_INC_PATH	= -I$(COMPAT)/bhyve -I$(COMPAT)/bhyve/amd64 \
+	-I$(CONTRIB)/bhyve -I$(CONTRIB)/bhyve/amd64
 INC_PATH	+= -I$(UTSBASE)/i86pc/io/vmm -I$(UTSBASE)/i86pc/io/vmm/io
 AS_INC_PATH	+= -I$(UTSBASE)/i86pc/io/vmm -I$(OBJS_DIR)
 
diff --git a/usr/src/uts/i86pc/vmm/Makefile b/usr/src/uts/i86pc/vmm/Makefile
index c55abf6090..e7f07c4c4e 100644
--- a/usr/src/uts/i86pc/vmm/Makefile
+++ b/usr/src/uts/i86pc/vmm/Makefile
@@ -56,8 +56,8 @@ $(OBJS_DIR)/vmx.o := SMOFF += deref_check
 
 ALL_BUILDS	= $(ALL_BUILDSONLY64)
 DEF_BUILDS	= $(DEF_BUILDSONLY64)
-PRE_INC_PATH	= -I$(COMPAT)/freebsd -I$(COMPAT)/freebsd/amd64 \
-	-I$(CONTRIB)/freebsd -I$(CONTRIB)/freebsd/amd64
+PRE_INC_PATH	= -I$(COMPAT)/bhyve -I$(COMPAT)/bhyve/amd64 \
+	-I$(CONTRIB)/bhyve -I$(CONTRIB)/bhyve/amd64
 INC_PATH	+= -I$(UTSBASE)/i86pc/io/vmm -I$(UTSBASE)/i86pc/io/vmm/io
 AS_INC_PATH	+= -I$(UTSBASE)/i86pc/io/vmm -I$(OBJS_DIR)
 
diff --git a/usr/src/uts/req.flg b/usr/src/uts/req.flg
index 15085a486d..76b8de8999 100644
--- a/usr/src/uts/req.flg
+++ b/usr/src/uts/req.flg
@@ -38,7 +38,6 @@ echo_file usr/src/Makefile
 
 # For full builds (open and closed), we want both etc/certs and
 # etc/keys.  For an open source build, there's no etc/keys directory.
-find_files "s.*" usr/contrib/freebsd
 find_files "s.*" usr/src/cmd/cmd-crypto/etc
 find_files "s.*" usr/src/common/acl
 find_files "s.*" usr/src/common/atomic
@@ -58,5 +57,6 @@ find_files "s.*" usr/src/common/smbios
 find_files "s.*" usr/src/common/tsol
 find_files "s.*" usr/src/common/util
 find_files "s.*" usr/src/common/zfs
-find_files "s.*" usr/src/compat/freebsd
+find_files "s.*" usr/src/compat/bhyve
+find_files "s.*" usr/src/contrib/bhyve
 find_files "s.*" usr/src/psm/promif
diff --git a/usr/src/uts/sun/io/zs_async.c b/usr/src/uts/sun/io/zs_async.c
index 6470434f6c..afdf563c4f 100644
--- a/usr/src/uts/sun/io/zs_async.c
+++ b/usr/src/uts/sun/io/zs_async.c
@@ -225,7 +225,7 @@ ushort_t zshiwat[NSPEED] = {
  */
 #define	ZSA_GETBLOCK(zs, allocbcount) \
 { \
-	register int n = zsa_rstandby; \
+	int n = zsa_rstandby; \
 	while (--n >= 0 && allocbcount > 0) { \
 		if (!za->za_rstandby[n]) { \
 			if ((za->za_rstandby[n] = allocb(ZSA_RCV_SIZE, \
@@ -243,7 +243,7 @@ ushort_t zshiwat[NSPEED] = {
 	if (za->za_ttycommon.t_cflag & CRTSXOFF) { \
 		mutex_enter(zs->zs_excl_hi); \
 		if (!(zs->zs_wreg[5] & ZSWR5_RTS)) { \
-			register int usedcnt = 0; \
+			int usedcnt = 0; \
 			for (n = 0; n < zsa_rstandby; n++) \
 				if (!za->za_rstandby[n]) \
 					usedcnt++; \
@@ -260,7 +260,7 @@ ushort_t zshiwat[NSPEED] = {
  */
 #define	ZSA_ALLOCB(mp) \
 { \
-	register int n = zsa_rstandby; \
+	int n = zsa_rstandby; \
 	while (--n >= 0) { \
 		if ((mp = za->za_rstandby[n]) != NULL) { \
 			za->za_rstandby[n] = NULL; \
@@ -274,7 +274,7 @@ ushort_t zshiwat[NSPEED] = {
 			cmn_err(CE_WARN, "zs%d: message lost\n", \
 				UNIT(za->za_dev)); \
 		} else if (zs->zs_wreg[5] & ZSWR5_RTS) { \
-			register int usedcnt = 0; \
+			int usedcnt = 0; \
 			for (n = 0; n < zsa_rstandby; n++) \
 				if (!za->za_rstandby[n]) \
 					usedcnt++; \
@@ -301,7 +301,7 @@ ushort_t zshiwat[NSPEED] = {
  */
 #define	ZSA_PUTQ(mp) \
 { \
-	register int wptr, rptr; \
+	int wptr, rptr; \
 	wptr = za->za_rdone_wptr; \
 	rptr = za->za_rdone_rptr; \
 	za->za_rdone[wptr] = mp; \
@@ -321,7 +321,7 @@ ushort_t zshiwat[NSPEED] = {
  */
 #define	ZSA_KICK_RCV \
 { \
-	register mblk_t *mp = za->za_rcvblk; \
+	mblk_t *mp = za->za_rcvblk; \
 	if (mp) { \
 		if (zs->zs_rd_cur) {	/* M_DATA */ \
 			mp->b_wptr = zs->zs_rd_cur; \
@@ -364,7 +364,7 @@ ushort_t zshiwat[NSPEED] = {
  */
 #define	ZSA_FLUSHQ \
 { \
-	register mblk_t *tmp; \
+	mblk_t *tmp; \
 	for (;;) { \
 		ZSA_GETQ(tmp); \
 		if (!(tmp)) \
@@ -412,7 +412,7 @@ int zsa_h_log_n[40];
 
 #define	zsa_h_log_clear \
 { \
-	register char *p; \
+	char *p; \
 	for (p = &zsa_h_log[zs->zs_unit][ZSA_H_LOG_MAX]; \
 		p >= &zsa_h_log[zs->zs_unit][0]; /* null */) \
 		*p-- = '\0'; \
@@ -498,8 +498,7 @@ static int	zsa_resume(struct zscom *zs);
 static void
 zsa_null(struct zscom *zs)
 {
-	/* LINTED */
-	register short	c;
+	short	c;
 
 	SCC_WRITE0(ZSWR0_RESET_TXINT);
 	SCC_WRITE0(ZSWR0_RESET_STATUS);
@@ -542,8 +541,8 @@ static void	zsa_reioctl(void *);
 static void	zsa_ioctl(struct asyncline *za, queue_t *q, mblk_t *mp);
 static void	zsa_program(struct asyncline *za, int setibaud);
 static void	zsa_start(struct zscom *zs);
-static void 	zsa_kick_rcv(void *);
-static void 	zsa_callback(void *);
+static void	zsa_kick_rcv(void *);
+static void	zsa_callback(void *);
 static void	zsa_set_za_rcv_flags_mask(struct asyncline *za);
 int		zsgetspeed(dev_t dev);
 
@@ -554,9 +553,9 @@ int
 zsc_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
     void **result)
 {
-	register dev_t dev = (dev_t)arg;
-	register int unit, error;
-	register struct zscom *zs;
+	dev_t dev = (dev_t)arg;
+	int unit, error;
+	struct zscom *zs;
 
 	if ((unit = UNIT(dev)) >= nzs)
 		return (DDI_FAILURE);
@@ -716,12 +715,12 @@ zsa_init(struct zscom *zs)
 static int
 zsa_open(queue_t *rq, dev_t *dev, int flag, int sflag, cred_t *cr)
 {
-	register struct zscom *zs;
-	register struct asyncline *za;
-	register int	speed, unit;
+	struct zscom *zs;
+	struct asyncline *za;
+	int	speed, unit;
 	struct termios *termiosp;
 	int len;
-	register int allocbcount = zsa_rstandby;
+	int allocbcount = zsa_rstandby;
 	boolean_t set_zsoptinit = B_FALSE;
 
 	unit = UNIT(*dev);
@@ -1205,10 +1204,10 @@ out:
 static void
 zsa_wput(queue_t *q, mblk_t *mp)
 {
-	register struct asyncline	*za;
-	register struct zscom		*zs;
-	register struct copyresp	*resp;
-	register mblk_t			*bp = NULL;
+	struct asyncline	*za;
+	struct zscom		*zs;
+	struct copyresp	*resp;
+	mblk_t			*bp = NULL;
 	int				error;
 	struct iocblk			*iocp;
 
@@ -1321,7 +1320,7 @@ zsa_wput(queue_t *q, mblk_t *mp)
 				freemsg(mp->b_cont);
 				mp->b_cont = NULL;
 			}
-			if (za->za_pps == NULL) {
+			if (za->za_pps == 0) {
 				mp->b_datap->db_type = M_IOCNAK;
 				iocp->ioc_error = ENXIO;
 				ZSA_QREPLY(q, mp);
@@ -1640,9 +1639,9 @@ zsa_rsrv(queue_t *q)
 static void
 zsa_txint(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register uchar_t *wr_cur;
-	register uchar_t s0;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	uchar_t *wr_cur;
+	uchar_t s0;
 
 	s0 = SCC_READ0();
 
@@ -1701,8 +1700,8 @@ zsa_txint(struct zscom *zs)
 static void
 zsa_xsint(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register uchar_t s0, x0;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	uchar_t s0, x0;
 
 	s0 = SCC_READ0();
 	ZSA_R0_LOG(s0);
@@ -1833,12 +1832,12 @@ zsa_xsint(struct zscom *zs)
 static void
 zsa_rxint(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register uchar_t c;
-	register uchar_t *rd_cur = zs->zs_rd_cur;
-	register uchar_t *rd_lim = zs->zs_rd_lim;
-	register mblk_t	*bp;
-	register uint_t	fm = za->za_rcv_flags_mask;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	uchar_t c;
+	uchar_t *rd_cur = zs->zs_rd_cur;
+	uchar_t *rd_lim = zs->zs_rd_lim;
+	mblk_t	*bp;
+	uint_t	fm = za->za_rcv_flags_mask;
 
 
 #ifdef ZSA_DEBUG
@@ -1956,12 +1955,12 @@ zsa_rxint(struct zscom *zs)
 static void
 zsa_srint(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register short s1;
-	register uchar_t c;
-	register uchar_t c1;
-	register mblk_t *bp = za->za_rcvblk;
-	register uchar_t *rd_cur = zs->zs_rd_cur;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	short s1;
+	uchar_t c;
+	uchar_t c1;
+	mblk_t *bp = za->za_rcvblk;
+	uchar_t *rd_cur = zs->zs_rd_cur;
 
 	SCC_READ(1, s1);
 	if (s1 & (ZSRR1_FE | ZSRR1_PE | ZSRR1_DO)) {
@@ -2080,17 +2079,17 @@ zsa_srint(struct zscom *zs)
 static int
 zsa_softint(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register uchar_t r0;
-	register uchar_t za_kick_active;
-	register int	m_error;
-	register int	allocbcount = 0;
-	register int 	do_ttycommon_qfull = 0;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	uchar_t r0;
+	uchar_t za_kick_active;
+	int	m_error;
+	int	allocbcount = 0;
+	int	do_ttycommon_qfull = 0;
 	boolean_t	hangup = B_FALSE, unhangup = B_FALSE;
 	boolean_t	m_break = B_FALSE, wakeup = B_FALSE;
-	register queue_t *q;
-	register mblk_t	*bp;
-	register mblk_t *head = NULL, *tail = NULL;
+	queue_t *q;
+	mblk_t	*bp;
+	mblk_t *head = NULL, *tail = NULL;
 
 	mutex_enter(zs->zs_excl);
 	if (zs->zs_suspended || (zs->zs_flags & ZS_CLOSED)) {
@@ -2351,10 +2350,10 @@ zsa_softint(struct zscom *zs)
 static void
 zsa_start(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register int cc;
-	register queue_t *q;
-	register mblk_t *bp;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	int cc;
+	queue_t *q;
+	mblk_t *bp;
 	uchar_t *rptr, *wptr;
 
 	/*
@@ -2501,8 +2500,8 @@ zsa_start_transmit:
 	 * so we need to explicitly mask before transmitting
 	 */
 	if ((za->za_ttycommon.t_cflag & CSIZE) == CS5) {
-		register unsigned char *p = rptr;
-		register int cnt = cc;
+		unsigned char *p = rptr;
+		int cnt = cc;
 
 		while (cnt--)
 			*p++ &= (unsigned char) 0x1f;
@@ -2741,11 +2740,11 @@ zsa_reioctl(void *arg)
 static void
 zsa_ioctl(struct asyncline *za, queue_t *wq, mblk_t *mp)
 {
-	register struct zscom *zs = za->za_common;
-	register struct iocblk *iocp;
-	register unsigned datasize;
+	struct zscom *zs = za->za_common;
+	struct iocblk *iocp;
+	unsigned datasize;
 	int error;
-	register mblk_t *tmp;
+	mblk_t *tmp;
 
 	if (za->za_ttycommon.t_iocpending != NULL) {
 		/*
@@ -2934,7 +2933,7 @@ zsa_ioctl(struct asyncline *za, queue_t *wq, mblk_t *mp)
 static int
 dmtozs(int bits)
 {
-	register int b = 0;
+	int b = 0;
 
 	if (bits & TIOCM_CAR)
 		b |= ZSRR0_CD;
@@ -2950,7 +2949,7 @@ dmtozs(int bits)
 static int
 zstodm(int bits)
 {
-	register int b;
+	int b;
 
 	b = 0;
 	if (bits & ZSRR0_CD)
@@ -2972,9 +2971,9 @@ zstodm(int bits)
 static void
 zsa_program(struct asyncline *za, int setibaud)
 {
-	register struct zscom *zs = za->za_common;
-	register struct zs_prog *zspp;
-	register int wr3, wr4, wr5, wr15, speed, baudrate, flags = 0;
+	struct zscom *zs = za->za_common;
+	struct zs_prog *zspp;
+	int wr3, wr4, wr5, wr15, speed, baudrate, flags = 0;
 
 	if ((baudrate = SPEED(za->za_ttycommon.t_cflag)) == 0) {
 		/*
@@ -3129,9 +3128,9 @@ zsa_program(struct asyncline *za, int setibaud)
 int
 zsgetspeed(dev_t dev)
 {
-	register struct zscom *zs;
-	register int uspeed, zspeed;
-	register uchar_t rr;
+	struct zscom *zs;
+	int uspeed, zspeed;
+	uchar_t rr;
 
 	zs = &zscom[UNIT(dev)];
 	SCC_READ(12, zspeed);
@@ -3170,7 +3169,7 @@ zsa_callback(void *arg)
 static void
 zsa_set_za_rcv_flags_mask(struct asyncline *za)
 {
-	register uint_t mask;
+	uint_t mask;
 
 	za->za_rcv_flags_mask &= ~0xFF;
 	switch (za->za_ttycommon.t_cflag & CSIZE) {
@@ -3303,7 +3302,7 @@ zsa_suspend(struct zscom *zs)
 static int
 zsa_resume(struct zscom *zs)
 {
-	register struct asyncline *za;
+	struct asyncline *za;
 	struct zs_prog	*zspp;
 
 	za = (struct asyncline *)&zs->zs_priv_str;
@@ -3343,8 +3342,8 @@ zsa_resume(struct zscom *zs)
 static void
 zsa_print_info(struct zscom *zs)
 {
-	register struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
-	register queue_t *q = za->za_ttycommon.t_writeq;
+	struct asyncline *za = (struct asyncline *)&zs->zs_priv_str;
+	queue_t *q = za->za_ttycommon.t_writeq;
 
 	printf(" next q=%s\n", (RD(q))->q_next->q_qinfo->qi_minfo->mi_idname);
 	printf("unit=%d\n", zs->zs_unit);
diff --git a/usr/src/uts/sun/io/zs_common.c b/usr/src/uts/sun/io/zs_common.c
index 7c39106941..43e5a2b4ec 100644
--- a/usr/src/uts/sun/io/zs_common.c
+++ b/usr/src/uts/sun/io/zs_common.c
@@ -595,14 +595,14 @@ zsattach(dev_info_t *dev, ddi_attach_cmd_t cmd)
 			(void) strcpy(name, "a");
 			if (ddi_create_minor_node(dev, name, S_IFCHR,
 			    ddi_get_instance(dev) * 2,
-			    obp_type, NULL) == DDI_FAILURE) {
+			    obp_type, 0) == DDI_FAILURE) {
 				ddi_remove_minor_node(dev, NULL);
 				return (DDI_FAILURE);
 			}
 			(void) strcpy(name, "b");
 			if (ddi_create_minor_node(dev, name, S_IFCHR,
 			    (ddi_get_instance(dev) * 2) + 1,
-			    obp_type, NULL) == DDI_FAILURE) {
+			    obp_type, 0) == DDI_FAILURE) {
 				ddi_remove_minor_node(dev, NULL);
 				return (DDI_FAILURE);
 			}
@@ -614,28 +614,28 @@ zsattach(dev_info_t *dev, ddi_attach_cmd_t cmd)
 		(void) sprintf(name, "%c", (ddi_get_instance(dev) + 'a'));
 		if (ddi_create_minor_node(dev, name, S_IFCHR,
 		    ddi_get_instance(dev) * 2,
-		    serial_line, NULL) == DDI_FAILURE) {
+		    serial_line, 0) == DDI_FAILURE) {
 			ddi_remove_minor_node(dev, NULL);
 			return (DDI_FAILURE);
 		}
 		(void) sprintf(name, "%c", (ddi_get_instance(dev) + 'b'));
 		if (ddi_create_minor_node(dev, name, S_IFCHR,
 		    (ddi_get_instance(dev) * 2) + 1,
-		    serial_line, NULL) == DDI_FAILURE) {
+		    serial_line, 0) == DDI_FAILURE) {
 			ddi_remove_minor_node(dev, NULL);
 			return (DDI_FAILURE);
 		}
 		(void) sprintf(name, "%c,cu", (ddi_get_instance(dev) + 'a'));
 		if (ddi_create_minor_node(dev, name, S_IFCHR,
 		    (ddi_get_instance(dev) * 2) | OUTLINE,
-		    dial_out, NULL) == DDI_FAILURE) {
+		    dial_out, 0) == DDI_FAILURE) {
 			ddi_remove_minor_node(dev, NULL);
 			return (DDI_FAILURE);
 		}
 		(void) sprintf(name, "%c,cu", (ddi_get_instance(dev) + 'b'));
 		if (ddi_create_minor_node(dev, name, S_IFCHR,
 		    ((ddi_get_instance(dev)  * 2) + 1) | OUTLINE,
-		    dial_out, NULL) == DDI_FAILURE) {
+		    dial_out, 0) == DDI_FAILURE) {
 			ddi_remove_minor_node(dev, NULL);
 			return (DDI_FAILURE);
 		}
@@ -871,7 +871,7 @@ zsintr(caddr_t intarg)
 		for (zs = &zscom[0]; zs <= zslast; zs++) {
 			if (zs->zs_flags & ZS_NEEDSOFT) {
 				zs->zs_flags &= ~ZS_NEEDSOFT;
-				(*zs->zs_ops->zsop_softint)(zs);
+				(void) (*zs->zs_ops->zsop_softint)(zs);
 				if (zs->intrstats) {
 					KIOIP->intrs[KSTAT_INTR_SOFT]++;
 				}
diff --git a/usr/src/uts/sun4u/zs/Makefile b/usr/src/uts/sun4u/zs/Makefile
index 198362da01..0af5aa7379 100644
--- a/usr/src/uts/sun4u/zs/Makefile
+++ b/usr/src/uts/sun4u/zs/Makefile
@@ -40,7 +40,6 @@ UTSBASE	= ../..
 #
 MODULE		= zs
 OBJECTS		= $(ZS_OBJS:%=$(OBJS_DIR)/%)
-LINTS		= $(ZS_OBJS:%.o=$(LINTS_DIR)/%.ln)
 ROOTMODULE	= $(ROOT_PSM_DRV_DIR)/$(MODULE)
 
 #
@@ -60,23 +59,8 @@ CLEANFILES	+= $(MODSTUBS_O)
 #	Define targets
 #
 ALL_TARGET	= $(BINARY)
-LINT_TARGET	= $(MODULE).lint
 INSTALL_TARGET	= $(BINARY) $(ROOTMODULE)
 
-#
-# lint pass one enforcement
-#
-CFLAGS += $(CCVERBOSE)
-
-#
-# For now, disable these lint checks; maintainers should endeavor
-# to investigate and remove these for maximum lint coverage.
-# Please do not carry these forward to new Makefiles.
-#
-LINTTAGS	+= -erroff=E_BAD_PTR_CAST_ALIGN
-LINTTAGS	+= -erroff=E_ASSIGN_NARROW_CONV
-LINTTAGS	+= -erroff=E_PTRDIFF_OVERFLOW
-
 CERRWARN	+= -_gcc=-Wno-parentheses
 CERRWARN	+= $(CNOWARN_UNINIT)
 
@@ -93,12 +77,6 @@ clean:		$(CLEAN_DEPS)
 
 clobber:	$(CLOBBER_DEPS)
 
-lint:		$(LINT_DEPS)
-
-modlintlib:	$(MODLINTLIB_DEPS)
-
-clean.lint:	$(CLEAN_LINT_DEPS)
-
 install:	$(INSTALL_DEPS)
 
 #