Add a bus_dma_tag_t for DVMA usage, suitable for use for devices not sitting

behind a sun4m iommu.

Move the existing dvma routines from vm_machdep.c to this new dvma.c; this
allows for a few declarations to be removed from public headers.

Extend the device attachment arguments (struct confargs) to pass a
bus_dma_tag_t. mainbus receives the dvma bus_dma_tag_t, and devices pass the
tag unchanged to their children, except for iommu(4) which replaces it with
its own.

Change the few sun4m-only drivers to pick the bus_dma_tag_t from confargs
rather than assume iommu; this allows qlw(4) to attach and work on sun4c.

ok kettenis@

19 files changed, 812 insertions, 360 deletions

diff --git a/sys/arch/sparc/conf/SUN4C b/sys/arch/sparc/conf/SUN4C
index 6c48b8e4f14..7a788fba7bc 100644
--- a/sys/arch/sparc/conf/SUN4C
+++ b/sys/arch/sparc/conf/SUN4C
@@ -1,4 +1,4 @@
-#	$OpenBSD: SUN4C,v 1.68 2014/10/25 18:21:01 miod Exp $
+#	$OpenBSD: SUN4C,v 1.69 2015/03/30 20:30:20 miod Exp $
 #	$NetBSD: GENERIC,v 1.48 1997/08/23 19:19:01 mjacob Exp $
 
 # Machine architecture; required by config(8)
@@ -95,6 +95,8 @@ esp*	at dma? flags 0x00ff000f			# depending on model
 
 #sun4c, sun4m
 isp*	at sbus?
+qlw*	at sbus?
+#qla*	at sbus?
 
 # sun4/300 and sun4c Ethernet - an AMD 7990 LANCE
 le0	at sbus0					# sun4c on-board
diff --git a/sys/arch/sparc/conf/files.sparc b/sys/arch/sparc/conf/files.sparc
index c0566e0a9b0..e30debbc07c 100644
--- a/sys/arch/sparc/conf/files.sparc
+++ b/sys/arch/sparc/conf/files.sparc
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.sparc,v 1.98 2015/02/28 17:54:54 miod Exp $
+#	$OpenBSD: files.sparc,v 1.99 2015/03/30 20:30:20 miod Exp $
 #	$NetBSD: files.sparc,v 1.44 1997/08/31 21:29:16 pk Exp $
 
 # @(#)files.sparc	8.1 (Berkeley) 7/19/93
@@ -296,6 +296,7 @@ file	arch/sparc/sparc/cache.c
 file	arch/sparc/sparc/conf.c
 file	arch/sparc/sparc/consinit.c
 file	arch/sparc/sparc/disksubr.c
+file	arch/sparc/sparc/dvma.c
 file	arch/sparc/sparc/emul.c
 file	arch/sparc/sparc/in_cksum.c		inet
 file	arch/sparc/sparc/intr.c
diff --git a/sys/arch/sparc/dev/dma.c b/sys/arch/sparc/dev/dma.c
index 351b4b7a5d6..346b8368615 100644
--- a/sys/arch/sparc/dev/dma.c
+++ b/sys/arch/sparc/dev/dma.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: dma.c,v 1.30 2014/07/28 18:31:39 miod Exp $	*/
+/*	$OpenBSD: dma.c,v 1.31 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: dma.c,v 1.46 1997/08/27 11:24:16 bouyer Exp $ */
 
 /*
@@ -38,8 +38,8 @@
 #include <sys/buf.h>
 #include <sys/proc.h>
 
-#include <sparc/autoconf.h>
-#include <sparc/cpu.h>
+#include <machine/autoconf.h>
+#include <machine/cpu.h>
 
 #include <sparc/sparc/cpuvar.h>
 
@@ -236,15 +236,18 @@ dmaattach(parent, self, aux)
 
 	/* search through children */
 	node = firstchild(devnode);
-	if (node != 0) do {
-		name = getpropstring(node, "name");
-		if (!romprop(&oca.ca_ra, name, node))
-			continue;
-
-		sbus_translate(parent, &oca);
-		oca.ca_bustype = BUS_SBUS;
-		(void) config_found(&sc->sc_dev, (void *)&oca, dmaprint);
-	} while ((node = nextsibling(node)) != 0); else
+	if (node != 0) {
+		do {
+			name = getpropstring(node, "name");
+			if (!romprop(&oca.ca_ra, name, node))
+				continue;
+
+			sbus_translate(parent, &oca);
+			oca.ca_bustype = BUS_SBUS;
+			oca.ca_dmat = ca->ca_dmat;
+			config_found(&sc->sc_dev, (void *)&oca, dmaprint);
+		} while ((node = nextsibling(node)) != 0);
+	} else
 #endif /* SUN4C || SUN4D || SUN4E || SUN4M */
 
 	if (strcmp(ca->ca_ra.ra_name, "dma") == 0) {
diff --git a/sys/arch/sparc/dev/fga.c b/sys/arch/sparc/dev/fga.c
index 102a995725a..8c325bd0a11 100644
--- a/sys/arch/sparc/dev/fga.c
+++ b/sys/arch/sparc/dev/fga.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: fga.c,v 1.17 2010/11/11 17:46:58 miod Exp $	*/
+/*	$OpenBSD: fga.c,v 1.18 2015/03/30 20:30:22 miod Exp $	*/
 
 /*
  * Copyright (c) 1999 Jason L. Wright (jason@thought.net)
@@ -43,13 +43,11 @@
 #include <sys/device.h>
 #include <sys/malloc.h>
 #include <uvm/uvm_extern.h>
-#include <machine/pmap.h>
 
 #include <machine/autoconf.h>
+#include <machine/bus.h>
 #include <sparc/cpu.h>
 #include <sparc/sparc/cpuvar.h>
-#include <sparc/dev/sbusvar.h>
-#include <sparc/dev/dmareg.h>	/* for SBUS_BURST_* */
 
 #include <sparc/dev/fgareg.h>
 #include <sparc/dev/fgavar.h>
@@ -81,6 +79,7 @@ struct fga_softc {
 	int		sc_nrange;		/* number of sbus ranges */
 	struct		rom_range *sc_range;	/* sbus range data */
 	u_int8_t	sc_established;		/* which hw intrs installed */
+	bus_dma_tag_t	sc_dmat;
 };
 
 int	fgaopen(dev_t, int, int, struct proc *);
@@ -163,6 +162,7 @@ fgaattach(parent, self, aux)
 	    ca->ca_ra.ra_reg[0].rr_len);
 
 	sc->sc_node = ca->ca_ra.ra_node;
+	sc->sc_dmat = ca->ca_dmat;
 
 	i = opennode("/iommu/sbus");
 	if (i == 0) {
@@ -285,6 +285,7 @@ fga_vmerangemap(sc, vmebase, vmelen, vmecap, sbusslot, sbusoffset, oca)
 	oca->ca_ra.ra_reg[1].rr_iospace = vmecap;
 	oca->ca_ra.ra_reg[1].rr_paddr = (void *)vmebase;
 	oca->ca_ra.ra_reg[1].rr_len = vmelen;
+	oca->ca_dmat = sc->sc_dmat;
 
 	/* 1. Setup slot select register for this range. */
 	switch (sbusslot) {
@@ -329,7 +330,7 @@ fga_vmerangemap(sc, vmebase, vmelen, vmecap, sbusslot, sbusoffset, oca)
 	    ~(VME_MASTER_CAP_DATA | VME_MASTER_CAP_ADDR);
 	regs->vme_master_cap[range] |= vmecap;
 
-	(void)config_found(&sc->sc_dev, oca, fgaprint);
+	config_found(&sc->sc_dev, oca, fgaprint);
 
 	return (0);
 }
diff --git a/sys/arch/sparc/dev/lebuffer.c b/sys/arch/sparc/dev/lebuffer.c
index 65c16196158..702d3fd84f6 100644
--- a/sys/arch/sparc/dev/lebuffer.c
+++ b/sys/arch/sparc/dev/lebuffer.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: lebuffer.c,v 1.9 2010/07/10 19:32:24 miod Exp $	*/
+/*	$OpenBSD: lebuffer.c,v 1.10 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: lebuffer.c,v 1.3 1997/05/24 20:16:28 pk Exp $ */
 
 /*
@@ -41,8 +41,8 @@
 #include <sys/buf.h>
 #include <sys/proc.h>
 
-#include <sparc/autoconf.h>
-#include <sparc/cpu.h>
+#include <machine/autoconf.h>
+#include <machine/cpu.h>
 
 #include <sparc/dev/sbusvar.h>
 #include <sparc/dev/lebuffervar.h>
@@ -83,7 +83,7 @@ lebufmatch(parent, vcf, aux)
 {
 	struct cfdata *cf = vcf;
 	struct confargs *ca = aux;
-	register struct romaux *ra = &ca->ca_ra;
+	struct romaux *ra = &ca->ca_ra;
 
 	if (strcmp(cf->cf_driver->cd_name, ra->ra_name))
 		return(0);
@@ -155,7 +155,8 @@ lebufattach(parent, self, aux)
 
 		sbus_translate(parent, &oca);
 		oca.ca_bustype = BUS_SBUS;
-		(void) config_found(&sc->sc_dev, (void *)&oca, lebufprint);
+		oca.ca_dmat = ca->ca_dmat;
+		config_found(&sc->sc_dev, (void *)&oca, lebufprint);
 	}
 #endif /* SUN4C || SUN4D || SUN4E || SUN4M */
 }
diff --git a/sys/arch/sparc/dev/obio.c b/sys/arch/sparc/dev/obio.c
index 23ceb723622..ed7c4af6475 100644
--- a/sys/arch/sparc/dev/obio.c
+++ b/sys/arch/sparc/dev/obio.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: obio.c,v 1.24 2015/03/26 18:13:56 miod Exp $	*/
+/*	$OpenBSD: obio.c,v 1.25 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: obio.c,v 1.37 1997/07/29 09:58:11 fair Exp $	*/
 
 /*
@@ -273,7 +273,8 @@ obioattach(parent, self, args)
 
 		sbus_translate(self, &oca);
 		oca.ca_bustype = BUS_OBIO;
-		(void) config_found(self, (void *)&oca, busprint);
+		oca.ca_dmat = ca->ca_dmat;
+		config_found(self, (void *)&oca, busprint);
 	}
 
 	for (node = node0; node; node = nextsibling(node)) {
@@ -292,7 +293,8 @@ obioattach(parent, self, args)
 		/* Translate into parent address spaces */
 		sbus_translate(self, &oca);
 		oca.ca_bustype = BUS_OBIO;
-		(void) config_found(self, (void *)&oca, busprint);
+		oca.ca_dmat = ca->ca_dmat;
+		config_found(self, (void *)&oca, busprint);
 	}
 #endif
 }
@@ -388,11 +390,13 @@ vmeattach(parent, self, aux)
 
 	oca.ca_ra.ra_name = "vmes";
 	oca.ca_bustype = BUS_MAIN;
-	(void)config_found(self, (void *)&oca, vmeprint);
+	oca.ca_dmat = ca->ca_dmat;
+	config_found(self, (void *)&oca, vmeprint);
 
 	oca.ca_ra.ra_name = "vmel";
 	oca.ca_bustype = BUS_MAIN;
-	(void)config_found(self, (void *)&oca, vmeprint);
+	oca.ca_dmat = ca->ca_dmat;
+	config_found(self, (void *)&oca, vmeprint);
 }
 
 void
diff --git a/sys/arch/sparc/dev/qec.c b/sys/arch/sparc/dev/qec.c
index a887e51732d..65e9c41ba89 100644
--- a/sys/arch/sparc/dev/qec.c
+++ b/sys/arch/sparc/dev/qec.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: qec.c,v 1.23 2015/03/29 10:59:47 mpi Exp $	*/
+/*	$OpenBSD: qec.c,v 1.24 2015/03/30 20:30:22 miod Exp $	*/
 
 /*
  * Copyright (c) 1998 Theo de Raadt and Jason L. Wright.
@@ -47,8 +47,8 @@
 #include <netinet/in.h>
 #include <netinet/if_ether.h>
 
-#include <sparc/autoconf.h>
-#include <sparc/cpu.h>
+#include <machine/autoconf.h>
+#include <machine/cpu.h>
 
 #include <sparc/dev/sbusvar.h>
 #include <sparc/dev/dmareg.h>
@@ -190,7 +190,8 @@ qecattach(parent, self, aux)
 
 		qec_translate(sc, &oca);
 		oca.ca_bustype = BUS_SBUS;
-		(void) config_found(&sc->sc_dev, (void *)&oca, qecprint);
+		oca.ca_dmat = ca->ca_dmat;
+		config_found(&sc->sc_dev, (void *)&oca, qecprint);
 	}
 }
 
diff --git a/sys/arch/sparc/dev/qlw_sbus.c b/sys/arch/sparc/dev/qlw_sbus.c
index c0c92487737..3f13fc1274f 100644
--- a/sys/arch/sparc/dev/qlw_sbus.c
+++ b/sys/arch/sparc/dev/qlw_sbus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: qlw_sbus.c,v 1.2 2014/10/25 18:21:01 miod Exp $	*/
+/*	$OpenBSD: qlw_sbus.c,v 1.3 2015/03/30 20:30:22 miod Exp $	*/
 /*
  * Copyright (c) 2014 Mark Kettenis
  *
@@ -20,10 +20,9 @@
 #include <sys/malloc.h>
 #include <sys/systm.h>
 
+#include <machine/autoconf.h>
 #include <machine/bus.h>
 #include <machine/intr.h>
-#include <machine/autoconf.h>
-extern struct sparc_bus_dma_tag *iommu_dmatag;
 
 #include <sparc/dev/sbusvar.h>
 #include <sparc/dev/dmareg.h>
@@ -59,10 +58,6 @@ qlw_sbus_match(struct device *parent, void *cf, void *aux)
 	struct confargs *ca = aux;
 	struct romaux *ra = &ca->ca_ra;
 
-	/* XXX this assumes dma through sun4m's iommu */
-	if (!CPU_ISSUN4M)
-		return 0;
-
 	if (strcmp("ptisp", ra->ra_name) == 0 ||
 	    strcmp("PTI,ptisp", ra->ra_name) == 0 ||
 	    strcmp("SUNW,isp", ra->ra_name) == 0 ||
@@ -129,7 +124,7 @@ qlw_sbus_attach(struct device *parent, struct device *self, void *aux)
 	qsc->qsc_rr = ra->ra_reg[0];
 	sc->sc_iot = &qsc->qsc_rr;
 	sc->sc_ios = ra->ra_reg[0].rr_len;
-	sc->sc_dmat = iommu_dmatag;
+	sc->sc_dmat = ca->ca_dmat;
 
 	sc->sc_isp_gen = QLW_GEN_ISP1000;
 	sc->sc_isp_type = QLW_ISP1000;
diff --git a/sys/arch/sparc/dev/sbus.c b/sys/arch/sparc/dev/sbus.c
index d0b7bba97cc..6489bb27530 100644
--- a/sys/arch/sparc/dev/sbus.c
+++ b/sys/arch/sparc/dev/sbus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sbus.c,v 1.20 2015/03/21 19:55:31 miod Exp $	*/
+/*	$OpenBSD: sbus.c,v 1.21 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: sbus.c,v 1.17 1997/06/01 22:10:39 pk Exp $ */
 
 /*
@@ -75,42 +75,49 @@ struct cfdriver sbus_cd = {
 };
 
 /*
+ * SBus driver attach arguments.
+ * We carry a few more information than a mere struct confargs internally,
+ * for the sake of sbus_match() and sbus_print().
+ */
+struct sbus_attach_args {
+	struct confargs	sa_ca;
+	int		sa_slave_only;
+	int		sa_unsupported;
+};
+
+static /*const*/ char *sl = "slave-only";
+
+/*
  * Print the location of some sbus-attached device (called just
  * before attaching that device).  If `sbus' is not NULL, the
  * device was found but not configured; print the sbus as well.
  * Return UNCONF (config_find ignores this if the device was configured).
  */
 int
-sbus_print(args, sbus)
-	void *args;
-	const char *sbus;
+sbus_print(void *args, const char *sbus)
 {
-	struct confargs *ca = args;
+	struct sbus_attach_args *sa = args;
 	char *class;
-	static char *sl = "slave-only";
 
 	if (sbus != NULL) {
-		printf("\"%s\" at %s", ca->ca_ra.ra_name, sbus);
-		class = getpropstring(ca->ca_ra.ra_node, "device_type");
+		printf("\"%s\" at %s", sa->sa_ca.ca_ra.ra_name, sbus);
+		class = getpropstring(sa->sa_ca.ca_ra.ra_node, "device_type");
 		if (*class != '\0')
 			printf(" class %s", class);
 	}
-	/* Check root node for 'slave-only' property */
-	if (getpropint(0, sl, 0) & (1 << ca->ca_slot))
+	if (sa->sa_slave_only)
 		printf(" %s", sl);
-	printf(" slot %d offset 0x%x", ca->ca_slot, ca->ca_offset);
+	printf(" slot %d offset 0x%x", sa->sa_ca.ca_slot, sa->sa_ca.ca_offset);
 
-	return ca->ca_bustype < 0 ? UNSUPP : UNCONF;
+	return sa->sa_unsupported != 0 ? UNSUPP : UNCONF;
 }
 
 int
-sbus_match(parent, vcf, aux)
-	struct device *parent;
-	void *vcf, *aux;
+sbus_match(struct device *parent, void *vcf, void *aux)
 {
-	register struct cfdata *cf = vcf;
-	register struct confargs *ca = aux;
-	register struct romaux *ra = &ca->ca_ra;
+	struct cfdata *cf = vcf;
+	struct confargs *ca = aux;
+	struct romaux *ra = &ca->ca_ra;
 
 	if (CPU_ISSUN4)
 		return (0);
@@ -134,17 +141,14 @@ sbus_match(parent, vcf, aux)
  * Attach an SBus.
  */
 void
-sbus_attach(parent, self, aux)
-	struct device *parent;
-	struct device *self;
-	void *aux;
+sbus_attach(struct device *parent, struct device *self, void *aux)
 {
-	register struct sbus_softc *sc = (struct sbus_softc *)self;
+	struct sbus_softc *sc = (struct sbus_softc *)self;
 	struct confargs *ca = aux;
-	register struct romaux *ra = &ca->ca_ra;
-	register int node;
-	register char *name;
-	struct confargs oca;
+	struct romaux *ra = &ca->ca_ra;
+	int node;
+	char *name;
+	struct sbus_attach_args saa;
 	int rlen;
 
 	/*
@@ -180,23 +184,29 @@ sbus_attach(parent, self, aux)
 		sc->sc_nrange = xsc->sc_nrange;
 		sc->sc_range = xsc->sc_range;
 		xsc->sc_attached = 2;
+
+		sc->sc_slave_only = getpropint(node, sl, 0);
 	} else {
-		if (ra->ra_bp != NULL && strcmp(ra->ra_bp->name, "sbus") == 0)
-			oca.ca_ra.ra_bp = ra->ra_bp + 1;
-		else
-			oca.ca_ra.ra_bp = NULL;
+		/* Check root node for 'slave-only' property */
+		sc->sc_slave_only = getpropint(0, sl, 0);
 
 		rlen = getproplen(node, "ranges");
 		if (rlen > 0) {
 			sc->sc_nrange = rlen / sizeof(struct rom_range);
 			sc->sc_range =
-				(struct rom_range *)malloc(rlen, M_DEVBUF, M_NOWAIT);
-			if (sc->sc_range == 0)
+			    (struct rom_range *)malloc(rlen, M_DEVBUF, M_NOWAIT);
+			if (sc->sc_range == NULL)
 				panic("sbus: PROM ranges too large: %d", rlen);
 			(void)getprop(node, "ranges", sc->sc_range, rlen);
 		}
 	}
 
+	if (ca->ca_bustype != BUS_XBOX && ra->ra_bp != NULL &&
+	    strcmp(ra->ra_bp->name, "sbus") == 0)
+		saa.sa_ca.ca_ra.ra_bp = ra->ra_bp + 1;
+	else
+		saa.sa_ca.ca_ra.ra_bp = NULL;
+
 	/*
 	 * Loop through ROM children, fixing any relative addresses
 	 * and then configuring each device.
@@ -207,15 +217,16 @@ sbus_attach(parent, self, aux)
 		if (CPU_ISSUN4E && strcmp(name, "vm") == 0)
 			continue;
 #endif
-		if (!romprop(&oca.ca_ra, name, node))
+		if (!romprop(&saa.sa_ca.ca_ra, name, node))
 			continue;
 
-		if (sbus_translate(self, &oca) == 0)
-			oca.ca_bustype = BUS_SBUS;
-		else
-			oca.ca_bustype = -1;	/* force attach to fail */
+		saa.sa_ca.ca_bustype = BUS_SBUS;
+		saa.sa_ca.ca_dmat = ca->ca_dmat;
+		saa.sa_unsupported = sbus_translate(self, &saa.sa_ca) != 0;
+		saa.sa_slave_only =
+		    sc->sc_slave_only & (1 << saa.sa_ca.ca_slot);
 
-		config_found_sm(&sc->sc_dev, (void *)&oca, sbus_print,
+		config_found_sm(&sc->sc_dev, (void *)&saa, sbus_print,
 		    sbus_search);
 	}
 }
@@ -224,18 +235,16 @@ int
 sbus_search(struct device *parent, void *vcf, void *args)
 {
 	struct cfdata *cf = vcf;
-	struct confargs *oca = args;
+	struct sbus_attach_args *saa = args;
 
-	if (oca->ca_bustype < 0)
+	if (saa->sa_unsupported != 0)
 		return 0;
 
-	return (cf->cf_attach->ca_match)(parent, cf, oca);
+	return (cf->cf_attach->ca_match)(parent, cf, &saa->sa_ca);
 }
 
 int
-sbus_translate(dev, ca)
-	struct device *dev;
-	struct confargs *ca;
+sbus_translate(struct device *dev, struct confargs *ca)
 {
 	struct sbus_softc *sc = (struct sbus_softc *)dev;
 	int base, slot;
@@ -293,17 +302,11 @@ sbus_translate(dev, ca)
  * Returns true if this sbus slot is capable of dma
  */
 int
-sbus_testdma(sc, ca)
-	struct sbus_softc *sc;
-	struct confargs *ca;
+sbus_testdma(struct sbus_softc *sc, struct confargs *ca)
 {
         struct romaux *ra = &ca->ca_ra;
 
-	/*
-	 * XXX how to handle more than one sbus?
-	 */
-
-	if (getpropint(0, "slave-only", 0) & (1 << ca->ca_slot)) {
+	if (sc->sc_slave_only & (1 << ca->ca_slot)) {
 		printf("%s: dma card found in non-dma sbus slot %d"
 			": not supported\n", ra->ra_name, ca->ca_slot);
 		return (0);
diff --git a/sys/arch/sparc/dev/sbusvar.h b/sys/arch/sparc/dev/sbusvar.h
index f365feece5d..df91eb71ee7 100644
--- a/sys/arch/sparc/dev/sbusvar.h
+++ b/sys/arch/sparc/dev/sbusvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: sbusvar.h,v 1.10 2015/03/21 19:55:31 miod Exp $	*/
+/*	$OpenBSD: sbusvar.h,v 1.11 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: sbusvar.h,v 1.4 1996/04/22 02:35:05 abrown Exp $ */
 
 /*
@@ -42,25 +42,17 @@
  */
 
 /*
- * S-bus variables.
+ * SBus variables.
  */
 
-/*
- * SBus driver attach arguments.
- */
-struct sbus_attach_args {
-	struct	romaux sa_ra;		/* name, node, addr, etc */
-	int	sa_slot;		/* SBus slot number */
-	int	sa_offset;		/* offset within slot */
-};
-
 /* variables per SBus */
 struct sbus_softc {
-	struct	device sc_dev;		/* base device */
-	int	sc_clockfreq;		/* clock frequency (in Hz) */
+	struct	device	sc_dev;		/* base device */
+	int		sc_clockfreq;	/* clock frequency (in Hz) */
 	struct	rom_range *sc_range;
-	int	sc_nrange;
-	int	sc_burst;		/* burst transfer sizes supported */
+	int		sc_nrange;
+	int		sc_burst;	/* burst transfer sizes supported */
+	int		sc_slave_only;
 };
 
 int	sbus_translate(struct device *, struct confargs *);
diff --git a/sys/arch/sparc/include/autoconf.h b/sys/arch/sparc/include/autoconf.h
index f3b3088d138..ab1cc549ad8 100644
--- a/sys/arch/sparc/include/autoconf.h
+++ b/sys/arch/sparc/include/autoconf.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: autoconf.h,v 1.18 2010/06/29 21:28:08 miod Exp $	*/
+/*	$OpenBSD: autoconf.h,v 1.19 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: autoconf.h,v 1.20 1997/05/24 20:03:03 pk Exp $ */
 
 /*
@@ -95,12 +95,25 @@ struct rom_range {		/* Only used on v3 PROMs */
 	u_int32_t	size;		/* Size in bytes of this range */
 };
 
+/*
+ * mapiodev maps an I/O device to a virtual address, returning the address.
+ * mapdev does the real work: you can supply a special virtual address and
+ * it will use that instead of creating one, but you must only do this if
+ * you get it from ../sparc/vaddrs.h.
+ */
+void	*mapdev(struct rom_reg *pa, int va, int offset, int size);
+#define	mapiodev(pa, offset, size) \
+	mapdev(pa, 0, offset, size)
+
+#include <machine/bus.h>
 
 struct confargs {
-	int	ca_bustype;
-	struct	romaux ca_ra;
-	int	ca_slot;
-	int	ca_offset;
+	int		ca_bustype;
+	struct	romaux	ca_ra;
+	int		ca_slot;
+	int		ca_offset;
+
+	bus_dma_tag_t	ca_dmat;
 };
 #define BUS_MAIN	0
 #define BUS_OBIO	1
@@ -120,15 +133,6 @@ struct confargs {
 #define BUS_FGA_A32D32	15
 
 /*
- * mapiodev maps an I/O device to a virtual address, returning the address.
- * mapdev does the real work: you can supply a special virtual address and
- * it will use that instead of creating one, but you must only do this if
- * you get it from ../sparc/vaddrs.h.
- */
-void	*mapdev(struct rom_reg *pa, int va, int offset, int size);
-#define	mapiodev(pa, offset, size) \
-	mapdev(pa, 0, offset, size)
-/*
  * REG2PHYS is provided for drivers with a `d_mmap' function.
  */
 #define REG2PHYS(rr, offset) \
diff --git a/sys/arch/sparc/include/bus.h b/sys/arch/sparc/include/bus.h
index ec723213110..962be540393 100644
--- a/sys/arch/sparc/include/bus.h
+++ b/sys/arch/sparc/include/bus.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bus.h,v 1.13 2011/09/27 20:47:30 miod Exp $	*/
+/*	$OpenBSD: bus.h,v 1.14 2015/03/30 20:30:22 miod Exp $	*/
 /*
  * Copyright (c) 2003, Miodrag Vallat.
  *
@@ -39,19 +39,17 @@
 #ifndef	_MACHINE_BUS_H_
 #define	_MACHINE_BUS_H_
 
-#include <machine/autoconf.h>
-
-#include <uvm/uvm_extern.h>
-
-#include <machine/pmap.h>
-
-typedef	u_long	bus_space_handle_t;
-
+typedef	u_long				 bus_space_handle_t;
 /*
  * bus_space_tag_t are pointer to *modified* rom_reg structures.
  * rr_iospace is used to also carry bus endianness information.
  */
-typedef	struct rom_reg 	*bus_space_tag_t;
+typedef	struct rom_reg			*bus_space_tag_t;
+typedef struct sparc_bus_dma_tag	*bus_dma_tag_t;
+typedef struct sparc_bus_dmamap		*bus_dmamap_t;
+
+#include <machine/autoconf.h>
+#include <uvm/uvm_extern.h>
 
 #define	TAG_LITTLE_ENDIAN		0x80000000
 
@@ -526,9 +524,6 @@ struct uio;
 #define	BUS_DMASYNC_PREWRITE	0x04	/* pre-write synchronization */
 #define	BUS_DMASYNC_POSTWRITE	0x08	/* post-write synchronization */
 
-typedef struct sparc_bus_dma_tag	*bus_dma_tag_t;
-typedef struct sparc_bus_dmamap		*bus_dmamap_t;
-
 /*
  *	bus_dma_segment_t
  *
diff --git a/sys/arch/sparc/include/param.h b/sys/arch/sparc/include/param.h
index 7c8abf3a9ff..532961f5906 100644
--- a/sys/arch/sparc/include/param.h
+++ b/sys/arch/sparc/include/param.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: param.h,v 1.50 2015/03/18 20:56:40 miod Exp $	*/
+/*	$OpenBSD: param.h,v 1.51 2015/03/30 20:30:22 miod Exp $	*/
 
 /*
  * Copyright (c) 1992, 1993
@@ -112,10 +112,9 @@
 #define	NKMEMPAGES_MAX_DEFAULT		((4 * 1024 * 1024) >> PAGE_SHIFT)
 
 #ifndef _LOCORE
-extern vaddr_t		dvma_base;
-extern vaddr_t		dvma_end;
 extern struct extent	*dvmamap_extent;
 
+extern void	dvma_init(void);
 extern caddr_t	kdvma_mapin(caddr_t, int, int);
 extern caddr_t	dvma_malloc_space(size_t, void *, int, int);
 extern void	dvma_free(caddr_t, size_t, void *);
diff --git a/sys/arch/sparc/include/vmparam.h b/sys/arch/sparc/include/vmparam.h
index b7356412985..139e61a0dcc 100644
--- a/sys/arch/sparc/include/vmparam.h
+++ b/sys/arch/sparc/include/vmparam.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vmparam.h,v 1.44 2015/03/27 20:25:39 miod Exp $	*/
+/*	$OpenBSD: vmparam.h,v 1.45 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: vmparam.h,v 1.13 1997/07/12 16:20:03 perry Exp $	*/
 
 /*
@@ -129,11 +129,7 @@ extern vsize_t vm_kernel_space_size;
 #define VM_PHYSSEG_NOADD		/* can't add RAM after vm_mem_init */
 
 #if defined (_KERNEL)
-struct vm_map;
-#define		dvma_mapin(map,va,len,canwait)	dvma_mapin_space(map,va,len,canwait,0)
-vaddr_t		dvma_mapin_space(struct vm_map *, vaddr_t, int, int, int);
 void		dvma_mapout(vaddr_t, vaddr_t, int);
-
 #endif
 
 #endif /* _MACHINE_VMPARAM_H_ */
diff --git a/sys/arch/sparc/sparc/autoconf.c b/sys/arch/sparc/sparc/autoconf.c
index 7b33c80842d..a89f223e79c 100644
--- a/sys/arch/sparc/sparc/autoconf.c
+++ b/sys/arch/sparc/sparc/autoconf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: autoconf.c,v 1.99 2015/03/27 20:25:39 miod Exp $	*/
+/*	$OpenBSD: autoconf.c,v 1.100 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: autoconf.c,v 1.73 1997/07/29 09:41:53 fair Exp $ */
 
 /*
@@ -802,6 +802,7 @@ cpu_configure()
 	register char *cp;
 	int s;
 	extern struct user *proc0paddr;
+	extern struct sparc_bus_dma_tag dvma_dmatag;
 
 	/* build the bootpath */
 	bootpath_build();
@@ -859,6 +860,7 @@ cpu_configure()
 
 	oca.ca_ra.ra_node = node;
 	oca.ca_ra.ra_name = cp = "mainbus";
+	oca.ca_dmat = &dvma_dmatag;
 	if (config_rootfound(cp, (void *)&oca) == NULL)
 		panic("mainbus not configured");
 
@@ -1213,11 +1215,13 @@ mainbus_attach(parent, dev, aux)
 		/* Configure the CPU. */
 		bzero(&oca, sizeof(oca));
 		oca.ca_ra.ra_name = "cpu";
-		(void)config_found(dev, (void *)&oca, mbprint);
+		oca.ca_dmat = ca->ca_dmat;
+		config_found(dev, (void *)&oca, mbprint);
 
 		/* Start at the beginning of the bootpath */
 		bzero(&oca, sizeof(oca));
 		oca.ca_ra.ra_bp = bootpath;
+		oca.ca_dmat = ca->ca_dmat;
 
 		oca.ca_bustype = BUS_MAIN;
 		oca.ca_ra.ra_name = "obio";
@@ -1227,7 +1231,7 @@ mainbus_attach(parent, dev, aux)
 		for (ssp = oldmon_special; (sp = *ssp) != NULL; ssp++) {
 			oca.ca_bustype = BUS_MAIN;
 			oca.ca_ra.ra_name = sp;
-			(void)config_found(dev, (void *)&oca, mbprint);
+			config_found(dev, (void *)&oca, mbprint);
 		}
 		return;
 	}
@@ -1275,6 +1279,7 @@ mainbus_attach(parent, dev, aux)
 				oca.ca_ra.ra_name = "cpu";
 				oca.ca_ra.ra_paddr = 0;
 				oca.ca_ra.ra_nreg = 0;
+				oca.ca_dmat = ca->ca_dmat;
 				config_found(dev, (void *)&oca, mbprint);
 			}
 		}
@@ -1284,6 +1289,7 @@ mainbus_attach(parent, dev, aux)
 		oca.ca_ra.ra_name = "cpu";
 		oca.ca_ra.ra_paddr = 0;
 		oca.ca_ra.ra_nreg = 0;
+		oca.ca_dmat = ca->ca_dmat;
 		config_found(dev, (void *)&oca, mbprint);
 	}
 
@@ -1354,7 +1360,7 @@ mainbus_attach(parent, dev, aux)
 				splx(11 << 8);		/*XXX*/
 #endif
 			oca.ca_bustype = BUS_MAIN;
-			(void) config_found(dev, (void *)&oca, mbprint);
+			config_found(dev, (void *)&oca, mbprint);
 		}
 	}
 
@@ -1369,6 +1375,7 @@ mainbus_attach(parent, dev, aux)
 		bzero(&oca, sizeof(oca));
 		oca.ca_bustype = BUS_MAIN;
 		oca.ca_ra.ra_name = "led";
+		oca.ca_dmat = ca->ca_dmat;
 		config_found(dev, (void *)&oca, mbprint);
 	}
 #endif
diff --git a/sys/arch/sparc/sparc/dvma.c b/sys/arch/sparc/sparc/dvma.c
new file mode 100644
index 00000000000..c5ffdb91bdc
--- /dev/null
+++ b/sys/arch/sparc/sparc/dvma.c
@@ -0,0 +1,653 @@
+/*	$OpenBSD: dvma.c,v 1.1 2015/03/30 20:30:22 miod Exp $	*/
+
+/*
+ * Copyright (c) 1996
+ *	The President and Fellows of Harvard College. All rights reserved.
+ * Copyright (c) 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * This software was developed by the Computer Systems Engineering group
+ * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
+ * contributed to Berkeley.
+ *
+ * All advertising materials mentioning features or use of this software
+ * must display the following acknowledgement:
+ *	This product includes software developed by the University of
+ *	California, Lawrence Berkeley Laboratory.
+ *	This product includes software developed by Harvard University.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. 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.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *	This product includes software developed by Harvard University.
+ *	This product includes software developed by the University of
+ *	California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *	@(#)vm_machdep.c	8.2 (Berkeley) 9/23/93
+ */
+
+#include <sys/param.h>
+#include <sys/extent.h>
+#include <sys/malloc.h>
+#include <sys/mbuf.h>
+#include <sys/proc.h>
+
+#include <uvm/uvm_extern.h>
+
+#include <machine/bus.h>
+#include <machine/cpu.h>
+
+/*
+ * dvmamap_extent is used to manage DVMA memory.
+ */
+vaddr_t dvma_base, dvma_end;
+struct extent *dvmamap_extent;
+
+vaddr_t	dvma_mapin_space(struct vm_map *, vaddr_t, int, int, int);
+
+void
+dvma_init(void)
+{
+	dvma_base = CPU_ISSUN4M ? DVMA4M_BASE : DVMA_BASE;
+	dvma_end = CPU_ISSUN4M ? DVMA4M_END : DVMA_END;
+	dvmamap_extent = extent_create("dvmamap", dvma_base, dvma_end,
+				       M_DEVBUF, NULL, 0, EX_NOWAIT);
+	if (dvmamap_extent == NULL)
+		panic("unable to allocate extent for dvma");
+}
+
+/*
+ * Wrapper for dvma_mapin_space() in kernel space,
+ * so drivers need not include VM goo to get at kernel_map.
+ */
+caddr_t
+kdvma_mapin(va, len, canwait)
+	caddr_t	va;
+	int	len, canwait;
+{
+	return (caddr_t)dvma_mapin_space(kernel_map, (vaddr_t)va, len, canwait,
+	    0);
+}
+
+#if defined(SUN4M)	/* XXX iommu.c */
+extern int has_iocache;
+#endif
+
+caddr_t
+dvma_malloc_space(len, kaddr, flags, space)
+	size_t	len;
+	void	*kaddr;
+	int	flags;
+{
+	vaddr_t	kva;
+	vaddr_t	dva;
+
+	len = round_page(len);
+	kva = (vaddr_t)malloc(len, M_DEVBUF, flags);
+	if (kva == 0)
+		return (NULL);
+
+#if defined(SUN4M)
+	if (!has_iocache)
+#endif
+		kvm_uncache((caddr_t)kva, atop(len));
+
+	*(vaddr_t *)kaddr = kva;
+	dva = dvma_mapin_space(kernel_map, kva, len, (flags & M_NOWAIT) ? 0 : 1, space);
+	if (dva == 0) {
+		free((void *)kva, M_DEVBUF, 0);
+		return (NULL);
+	}
+	return (caddr_t)dva;
+}
+
+void
+dvma_free(dva, len, kaddr)
+	caddr_t	dva;
+	size_t	len;
+	void	*kaddr;
+{
+	vaddr_t	kva = *(vaddr_t *)kaddr;
+
+	len = round_page(len);
+
+	dvma_mapout((vaddr_t)dva, kva, len);
+	/*
+	 * Even if we're freeing memory here, we can't be sure that it will
+	 * be unmapped, so we must recache the memory range to avoid impact
+	 * on other kernel subsystems.
+	 */
+#if defined(SUN4M)
+	if (!has_iocache)
+#endif
+		kvm_recache(kaddr, atop(len));
+	free((void *)kva, M_DEVBUF, 0);
+}
+
+u_long dvma_cachealign = 0;
+
+/*
+ * Map a range [va, va+len] of wired virtual addresses in the given map
+ * to a valid DVMA address. On non-SRMMU systems, this establish a
+ * second mapping of that range.
+ */
+vaddr_t
+dvma_mapin_space(map, va, len, canwait, space)
+	struct vm_map	*map;
+	vaddr_t	va;
+	int		len, canwait, space;
+{
+	vaddr_t	kva, tva;
+	int npf, s;
+	paddr_t pa;
+	vaddr_t off;
+	vaddr_t ova;
+	int olen;
+	int error;
+
+	if (dvma_cachealign == 0)
+	        dvma_cachealign = PAGE_SIZE;
+
+	ova = va;
+	olen = len;
+
+	off = va & PAGE_MASK;
+	va &= ~PAGE_MASK;
+	len = round_page(len + off);
+	npf = atop(len);
+
+	s = splhigh();
+	if (space & M_SPACE_D24)
+		error = extent_alloc_subregion(dvmamap_extent,
+		    DVMA_D24_BASE, DVMA_D24_END, len, dvma_cachealign,
+		    va & (dvma_cachealign - 1), 0,
+		    canwait ? EX_WAITSPACE : EX_NOWAIT, &tva);
+	else
+		error = extent_alloc(dvmamap_extent, len, dvma_cachealign, 
+		    va & (dvma_cachealign - 1), 0,
+		    canwait ? EX_WAITSPACE : EX_NOWAIT, &tva);
+	splx(s);
+	if (error)
+		return 0;
+	kva = tva;
+
+	while (npf--) {
+		if (pmap_extract(vm_map_pmap(map), va, &pa) == FALSE)
+			panic("dvma_mapin: null page frame");
+		pa = trunc_page(pa);
+
+#if defined(SUN4M)
+		if (CPU_ISSUN4M) {
+			iommu_enter(tva, pa);
+		} else
+#endif
+		{
+			/*
+			 * pmap_enter distributes this mapping to all
+			 * contexts... maybe we should avoid this extra work
+			 */
+#ifdef notyet
+#if defined(SUN4)
+			if (have_iocache)
+				pa |= PG_IOC;
+#endif
+#endif
+			pmap_kenter_pa(tva, pa | PMAP_NC,
+			    PROT_READ | PROT_WRITE);
+		}
+
+		tva += PAGE_SIZE;
+		va += PAGE_SIZE;
+	}
+	pmap_update(pmap_kernel());
+
+	/*
+	 * XXX Only have to do this on write.
+	 */
+	if (CACHEINFO.c_vactype == VAC_WRITEBACK)		/* XXX */
+		cpuinfo.cache_flush((caddr_t)ova, olen);	/* XXX */
+
+	return kva + off;
+}
+
+/*
+ * Remove DVMA mapping of `va' in DVMA space at `dva'.
+ */
+void
+dvma_mapout(dva, va, len)
+	vaddr_t	dva, va;
+	int		len;
+{
+	int s, off;
+	int error;
+	int dlen;
+
+	off = (int)dva & PGOFSET;
+	dva -= off;
+	dlen = round_page(len + off);
+
+#if defined(SUN4M)
+	if (CPU_ISSUN4M)
+		iommu_remove(dva, dlen);
+	else
+#endif
+	{
+		pmap_kremove(dva, dlen);
+		pmap_update(pmap_kernel());
+	}
+
+	s = splhigh();
+	error = extent_free(dvmamap_extent, dva, dlen, EX_NOWAIT);
+	if (error)
+		printf("dvma_mapout: extent_free failed\n");
+	splx(s);
+
+	if (CACHEINFO.c_vactype != VAC_NONE)
+		cpuinfo.cache_flush((caddr_t)va, len);
+}
+
+int	dvma_dmamap_create(bus_dma_tag_t, bus_size_t, int, bus_size_t,
+	    bus_size_t, int, bus_dmamap_t *);
+int	dvma_dmamap_load(bus_dma_tag_t, bus_dmamap_t, void *, bus_size_t,
+	    struct proc *, int);
+int	dvma_dmamap_load_mbuf(bus_dma_tag_t, bus_dmamap_t, struct mbuf *, int);
+int	dvma_dmamap_load_raw(bus_dma_tag_t, bus_dmamap_t, bus_dma_segment_t *,
+	    int, bus_size_t, int);
+void	dvma_dmamap_unload(bus_dma_tag_t, bus_dmamap_t);
+void	dvma_dmamap_sync(bus_dma_tag_t, bus_dmamap_t, bus_addr_t, bus_size_t,
+	    int);
+
+int	dvma_dmamem_map(bus_dma_tag_t, bus_dma_segment_t *, int, size_t,
+	    caddr_t *, int);
+void	dvma_dmamem_unmap(bus_dma_tag_t, void *, size_t);
+
+int	dvma_dmamap_load_buffer(bus_dma_tag_t, bus_dmamap_t, void *, bus_size_t,
+	    struct proc *, int);
+
+int	dvma_alloc(bus_dmamap_t, vaddr_t, bus_size_t, int, bus_addr_t *,
+	    bus_size_t *);
+
+struct sparc_bus_dma_tag dvma_dmatag = {
+	._cookie = NULL,
+	._dmamap_create = dvma_dmamap_create,
+	._dmamap_destroy = _bus_dmamap_destroy,
+	._dmamap_load = dvma_dmamap_load,
+	._dmamap_load_mbuf = dvma_dmamap_load_mbuf,
+	._dmamap_load_uio = _bus_dmamap_load_uio,
+	._dmamap_load_raw = dvma_dmamap_load_raw,
+	._dmamap_unload = dvma_dmamap_unload,
+	._dmamap_sync = dvma_dmamap_sync,
+
+	._dmamem_alloc = _bus_dmamem_alloc,
+	._dmamem_free = _bus_dmamem_free,
+	._dmamem_map = dvma_dmamem_map,
+	._dmamem_unmap = _bus_dmamem_unmap,
+	._dmamem_mmap = _bus_dmamem_mmap,
+};
+
+/*
+ * DVMA DMA map functions.
+ */
+int
+dvma_dmamap_create(bus_dma_tag_t t, bus_size_t size, int nsegments,
+    bus_size_t maxsegsz, bus_size_t boundary, int flags, bus_dmamap_t *dmamp)
+{
+	bus_dmamap_t map;
+	int error;
+
+	if ((error = _bus_dmamap_create(t, size, nsegments, maxsegsz, boundary,
+	    flags, &map)) != 0)
+		return (error);
+
+	/* Enable allocations from the entire map */
+	map->_dm_ex_start = DVMA_BASE;
+	map->_dm_ex_end = DVMA_END;
+
+	*dmamp = map;
+	return (0);
+}
+
+/*
+ * Internal routine to allocate space in the DVMA extent.
+ */
+int
+dvma_alloc(bus_dmamap_t map, vaddr_t va, bus_size_t len, int flags,
+    bus_addr_t *dvap, bus_size_t *sgsizep)
+{
+	bus_size_t sgsize;
+	u_long align, voff, dvaddr;
+	int s, error;
+	int pagesz = PAGE_SIZE;
+
+	/*
+	 * Remember page offset, then truncate the buffer address to
+	 * a page boundary.
+	 */
+	voff = va & (pagesz - 1);
+	va &= -pagesz;
+
+	if (len > map->_dm_size)
+		return (EINVAL);
+
+	sgsize = (len + voff + pagesz - 1) & -pagesz;
+	align = dvma_cachealign ? dvma_cachealign : map->_dm_align;
+
+	s = splhigh();
+	error = extent_alloc_subregion(dvmamap_extent, map->_dm_ex_start,
+	    map->_dm_ex_end, sgsize, align, va & (align-1), map->_dm_boundary,
+	    (flags & BUS_DMA_NOWAIT) == 0 ? EX_WAITOK : EX_NOWAIT, &dvaddr);
+	splx(s);
+	*dvap = (bus_addr_t)dvaddr;
+	*sgsizep = sgsize;
+	return (error);
+}
+
+int
+dvma_dmamap_load_buffer(bus_dma_tag_t t, bus_dmamap_t map, void *buf,
+    bus_size_t buflen, struct proc *p, int flags)
+{
+	bus_size_t sgsize;
+	bus_addr_t dva;
+	vaddr_t va = (vaddr_t)buf;
+	int pagesz = PAGE_SIZE;
+	pmap_t pmap;
+	int error;
+
+	if (map->dm_nsegs >= map->_dm_segcnt)
+		return (EFBIG);
+
+	/* Allocate DVMA resources */
+	if ((error = dvma_alloc(map, va, buflen, flags, &dva, &sgsize)) != 0)
+		return (error);
+
+	if ((CACHEINFO.ec_totalsize == 0))
+		cpuinfo.cache_flush(buf, buflen); /* XXX - move to bus_dma_sync? */
+
+	/*
+	 * We always use just one segment.
+	 */
+	map->dm_segs[map->dm_nsegs].ds_addr = dva + (va & (pagesz - 1));
+	map->dm_segs[map->dm_nsegs].ds_len = buflen;
+	map->dm_segs[map->dm_nsegs]._ds_sgsize = sgsize;
+	map->dm_nsegs++;
+
+	if (p != NULL)
+		pmap = p->p_vmspace->vm_map.pmap;
+	else
+		pmap = pmap_kernel();
+
+	for (; buflen != 0; ) {
+		paddr_t pa;
+		/*
+		 * Get the physical address for this page.
+		 */
+		if (!pmap_extract(pmap, va, &pa)) {
+			pmap_update(pmap_kernel());
+			return (EFAULT);
+		}
+
+		/*
+		 * Compute the segment size, and adjust counts.
+		 */
+		sgsize = pagesz - (va & (pagesz - 1));
+		if (buflen < sgsize)
+			sgsize = buflen;
+
+#ifdef notyet
+#if defined(SUN4)
+		if (have_iocache)
+			pa |= PG_IOC;
+#endif
+#endif
+		pmap_kenter_pa(dva, (pa & -pagesz) | PMAP_NC,
+		    PROT_READ | PROT_WRITE);
+
+		dva += pagesz;
+		va += sgsize;
+		buflen -= sgsize;
+	}
+	pmap_update(pmap_kernel());
+
+	return (0);
+}
+
+/*
+ * Prepare buffer for DMA transfer.
+ */
+int
+dvma_dmamap_load(bus_dma_tag_t t, bus_dmamap_t map, void *buf,
+    bus_size_t buflen, struct proc *p, int flags)
+{
+	int error;
+
+	/*
+	 * Make sure that on error condition we return "no valid mappings".
+	 */
+	map->dm_mapsize = buflen;
+	map->dm_nsegs = 0;
+
+	error = dvma_dmamap_load_buffer(t, map, buf, buflen, p, flags);
+	if (error)
+		dvma_dmamap_unload(t, map);
+
+	return (error);
+}
+
+/*
+ * Like _bus_dmamap_load(), but for mbufs.
+ */
+int
+dvma_dmamap_load_mbuf(bus_dma_tag_t t, bus_dmamap_t map, struct mbuf *m0,
+    int flags)
+{
+	struct mbuf *m;
+	int error = 0;
+
+	/*
+	 * Make sure that on error condition we return "no valid mappings".
+	 */
+	map->dm_mapsize = m0->m_pkthdr.len;
+	map->dm_nsegs = 0;
+
+	for (m = m0; m != NULL && error == 0; m = m->m_next) {
+		if (m->m_len == 0)
+			continue;
+		error = dvma_dmamap_load_buffer(t, map, m->m_data, m->m_len,
+		    NULL, flags);
+	}
+
+	if (error)
+		dvma_dmamap_unload(t, map);
+
+	return (error);
+}
+
+/*
+ * Like _bus_dmamap_load(), but for raw memory allocated with
+ * bus_dmamem_alloc().
+ */
+int
+dvma_dmamap_load_raw(bus_dma_tag_t t, bus_dmamap_t map, bus_dma_segment_t *segs,
+    int nsegs, bus_size_t size, int flags)
+{
+	struct vm_page *m;
+	paddr_t pa;
+	bus_addr_t dva;
+	bus_size_t sgsize;
+	struct pglist *mlist;
+	int pagesz = PAGE_SIZE;
+	int error;
+
+	map->dm_nsegs = 0;
+
+	/* Allocate DVMA resources */
+	if ((error = dvma_alloc(map, segs[0]._ds_va, size, flags, &dva,
+	    &sgsize)) != 0)
+		return (error);
+
+	/*
+	 * Note DVMA address in case bus_dmamem_map() is called later.
+	 * It can then insure cache coherency by choosing a KVA that
+	 * is aligned to `ds_addr'.
+	 */
+	segs[0].ds_addr = dva;
+	segs[0].ds_len = size;
+
+	map->dm_segs[0].ds_addr = dva;
+	map->dm_segs[0].ds_len = size;
+	map->dm_segs[0]._ds_sgsize = sgsize;
+
+	/* Map physical pages into the DVMA address space */
+	mlist = segs[0]._ds_mlist;
+	for (m = TAILQ_FIRST(mlist); m != NULL; m = TAILQ_NEXT(m,pageq)) {
+		if (sgsize == 0)
+			panic("dvma_dmamap_load_raw: size botch");
+		pa = VM_PAGE_TO_PHYS(m);
+
+#ifdef notyet
+#if defined(SUN4)
+		if (have_iocache)
+			pa |= PG_IOC;
+#endif
+#endif
+		pmap_kenter_pa(dva, (pa & -pagesz) | PMAP_NC,
+		    PROT_READ | PROT_WRITE);
+
+		dva += pagesz;
+		sgsize -= pagesz;
+	}
+	pmap_update(pmap_kernel());
+
+	map->dm_nsegs = 1;
+	map->dm_mapsize = size;
+
+	return (0);
+}
+
+/*
+ * Unload a DVMA DMA map.
+ */
+void
+dvma_dmamap_unload(bus_dma_tag_t t, bus_dmamap_t map)
+{
+	bus_dma_segment_t *segs = map->dm_segs;
+	int nsegs = map->dm_nsegs;
+	bus_addr_t dva;
+	bus_size_t len;
+	int i, s, error;
+
+	for (i = 0; i < nsegs; i++) {
+		dva = segs[i].ds_addr & -PAGE_SIZE;
+		len = segs[i]._ds_sgsize;
+
+		pmap_kremove(dva, len);
+		s = splhigh();
+		error = extent_free(dvmamap_extent, dva, len, EX_NOWAIT);
+		splx(s);
+		if (error != 0)
+			printf("warning: %ld of DVMA space lost\n", (long)len);
+	}
+	pmap_update(pmap_kernel());
+
+	/* Mark the mappings as invalid. */
+	map->dm_mapsize = 0;
+	map->dm_nsegs = 0;
+}
+
+/*
+ * DMA map synchronization.
+ */
+void
+dvma_dmamap_sync(bus_dma_tag_t t, bus_dmamap_t map, bus_addr_t offset,
+    bus_size_t len, int ops)
+{
+
+	/*
+	 * XXX Should flush CPU write buffers.
+	 */
+}
+
+/*
+ * Map DMA-safe memory.
+ */
+int
+dvma_dmamem_map(bus_dma_tag_t t, bus_dma_segment_t *segs, int nsegs,
+    size_t size, caddr_t *kvap, int flags)
+{
+	struct vm_page *m;
+	vaddr_t va;
+	bus_addr_t addr;
+	struct pglist *mlist;
+	u_long align;
+	int pagesz = PAGE_SIZE;
+	const struct kmem_dyn_mode *kd;
+
+	if (nsegs != 1)
+		panic("dvma_dmamem_map: nsegs = %d", nsegs);
+
+	align = dvma_cachealign ? dvma_cachealign : pagesz;
+
+	size = round_page(size);
+
+	kd = flags & BUS_DMA_NOWAIT ? &kd_trylock : &kd_waitok;
+	va = (vaddr_t)km_alloc(size, &kv_any, &kp_none, kd);
+	if (va == 0)
+		return (ENOMEM);
+
+	segs[0]._ds_va = va;
+	*kvap = (void *)va;
+
+	/*
+	 * Map the pages allocated in _bus_dmamem_alloc() to the
+	 * kernel virtual address space.
+	 */
+	mlist = segs[0]._ds_mlist;
+	for (m = TAILQ_FIRST(mlist); m != NULL; m = TAILQ_NEXT(m,pageq)) {
+
+		if (size == 0)
+			panic("dvma_dmamem_map: size botch");
+
+		addr = VM_PAGE_TO_PHYS(m);
+		pmap_kenter_pa(va, addr | PMAP_NC, PROT_READ | PROT_WRITE);
+
+		va += pagesz;
+		size -= pagesz;
+	}
+	pmap_update(pmap_kernel());
+
+	return (0);
+}
+
+void
+dvma_dmamem_unmap(bus_dma_tag_t t, void *kva, size_t size)
+{
+
+#ifdef DIAGNOSTIC
+	if ((u_long)kva & PAGE_MASK)
+		panic("dvma_dmamem_unmap");
+#endif
+
+	km_free(kva, round_page(size), &kv_any, &kp_none);
+}
diff --git a/sys/arch/sparc/sparc/iommu.c b/sys/arch/sparc/sparc/iommu.c
index b93bcd48ab7..db721210be4 100644
--- a/sys/arch/sparc/sparc/iommu.c
+++ b/sys/arch/sparc/sparc/iommu.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: iommu.c,v 1.28 2014/11/16 12:30:58 deraadt Exp $	*/
+/*	$OpenBSD: iommu.c,v 1.29 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: iommu.c,v 1.13 1997/07/29 09:42:04 fair Exp $ */
 
 /*
@@ -46,8 +46,6 @@
 
 #include <uvm/uvm_extern.h>
 
-#include <machine/pmap.h>
-
 #include <machine/autoconf.h>
 #include <machine/bus.h>
 #include <machine/ctlreg.h>
@@ -335,12 +333,14 @@ iommu_attach(parent, self, aux)
 	/*
 	 * Loop through ROM children (expect SBus among them).
 	 */
+	oca.ca_dmat = dmat;
 	for (node = firstchild(node); node; node = nextsibling(node)) {
 		name = getpropstring(node, "name");
 		if (!romprop(&oca.ca_ra, name, node))
 			continue;
 		oca.ca_bustype = BUS_MAIN; /* ??? */
-		(void) config_found(&sc->sc_dev, (void *)&oca, iommu_print);
+		oca.ca_dmat = dmat;
+		config_found(&sc->sc_dev, (void *)&oca, iommu_print);
 	}
 #endif
 }
diff --git a/sys/arch/sparc/sparc/machdep.c b/sys/arch/sparc/sparc/machdep.c
index 45390ce5772..c66a7174613 100644
--- a/sys/arch/sparc/sparc/machdep.c
+++ b/sys/arch/sparc/sparc/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.170 2015/03/27 20:25:39 miod Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.171 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: machdep.c,v 1.85 1997/09/12 08:55:02 pk Exp $ */
 
 /*
@@ -116,12 +116,6 @@ int	sparc_led_blink = 1;
  */
 int	safepri = 0;
 
-/*
- * dvmamap_extent is used to manage DVMA memory.
- */
-vaddr_t dvma_base, dvma_end;
-struct extent *dvmamap_extent;
-
 void	dumpsys(void);
 void	stackdump(void);
 
@@ -194,17 +188,7 @@ cpu_startup()
 	exec_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
 				 16*NCARGS, VM_MAP_PAGEABLE, FALSE, NULL);
 
-	/*
-	 * Allocate a map for physio.  Others use a submap of the kernel
-	 * map, but we want one completely separate, even though it uses
-	 * the same pmap.
-	 */
-	dvma_base = CPU_ISSUN4M ? DVMA4M_BASE : DVMA_BASE;
-	dvma_end = CPU_ISSUN4M ? DVMA4M_END : DVMA_END;
-	dvmamap_extent = extent_create("dvmamap", dvma_base, dvma_end,
-				       M_DEVBUF, NULL, 0, EX_NOWAIT);
-	if (dvmamap_extent == NULL)
-		panic("unable to allocate extent for dvma");
+	dvma_init();
 
 #ifdef DEBUG
 	pmapdebug = opmapdebug;
diff --git a/sys/arch/sparc/sparc/vm_machdep.c b/sys/arch/sparc/sparc/vm_machdep.c
index afcab31befd..d0d611dce3c 100644
--- a/sys/arch/sparc/sparc/vm_machdep.c
+++ b/sys/arch/sparc/sparc/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.58 2014/11/16 12:30:59 deraadt Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.59 2015/03/30 20:30:22 miod Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.30 1997/03/10 23:55:40 pk Exp $ */
 
 /*
@@ -70,195 +70,6 @@
 #include <sparc/sparc/cpuvar.h>
 
 /*
- * Wrapper for dvma_mapin() in kernel space,
- * so drivers need not include VM goo to get at kernel_map.
- */
-caddr_t
-kdvma_mapin(va, len, canwait)
-	caddr_t	va;
-	int	len, canwait;
-{
-	return ((caddr_t)dvma_mapin(kernel_map, (vaddr_t)va, len, canwait));
-}
-
-#if defined(SUN4M)
-extern int has_iocache;
-#endif
-
-caddr_t
-dvma_malloc_space(len, kaddr, flags, space)
-	size_t	len;
-	void	*kaddr;
-	int	flags;
-{
-	vaddr_t	kva;
-	vaddr_t	dva;
-
-	len = round_page(len);
-	kva = (vaddr_t)malloc(len, M_DEVBUF, flags);
-	if (kva == 0)
-		return (NULL);
-
-#if defined(SUN4M)
-	if (!has_iocache)
-#endif
-		kvm_uncache((caddr_t)kva, atop(len));
-
-	*(vaddr_t *)kaddr = kva;
-	dva = dvma_mapin_space(kernel_map, kva, len, (flags & M_NOWAIT) ? 0 : 1, space);
-	if (dva == 0) {
-		free((void *)kva, M_DEVBUF, 0);
-		return (NULL);
-	}
-	return (caddr_t)dva;
-}
-
-void
-dvma_free(dva, len, kaddr)
-	caddr_t	dva;
-	size_t	len;
-	void	*kaddr;
-{
-	vaddr_t	kva = *(vaddr_t *)kaddr;
-
-	len = round_page(len);
-
-	dvma_mapout((vaddr_t)dva, kva, len);
-	/*
-	 * Even if we're freeing memory here, we can't be sure that it will
-	 * be unmapped, so we must recache the memory range to avoid impact
-	 * on other kernel subsystems.
-	 */
-#if defined(SUN4M)
-	if (!has_iocache)
-#endif
-		kvm_recache(kaddr, atop(len));
-	free((void *)kva, M_DEVBUF, 0);
-}
-
-u_long dvma_cachealign = 0;
-
-/*
- * Map a range [va, va+len] of wired virtual addresses in the given map
- * to a valid DVMA address. On non-SRMMU systems, this establish a
- * second mapping of that range.
- */
-vaddr_t
-dvma_mapin_space(map, va, len, canwait, space)
-	struct vm_map	*map;
-	vaddr_t	va;
-	int		len, canwait, space;
-{
-	vaddr_t	kva, tva;
-	int npf, s;
-	paddr_t pa;
-	vaddr_t off;
-	vaddr_t ova;
-	int olen;
-	int error;
-
-	if (dvma_cachealign == 0)
-	        dvma_cachealign = PAGE_SIZE;
-
-	ova = va;
-	olen = len;
-
-	off = va & PAGE_MASK;
-	va &= ~PAGE_MASK;
-	len = round_page(len + off);
-	npf = atop(len);
-
-	s = splhigh();
-	if (space & M_SPACE_D24)
-		error = extent_alloc_subregion(dvmamap_extent,
-		    DVMA_D24_BASE, DVMA_D24_END, len, dvma_cachealign,
-		    va & (dvma_cachealign - 1), 0,
-		    canwait ? EX_WAITSPACE : EX_NOWAIT, &tva);
-	else
-		error = extent_alloc(dvmamap_extent, len, dvma_cachealign, 
-		    va & (dvma_cachealign - 1), 0,
-		    canwait ? EX_WAITSPACE : EX_NOWAIT, &tva);
-	splx(s);
-	if (error)
-		return 0;
-	kva = tva;
-
-	while (npf--) {
-		if (pmap_extract(vm_map_pmap(map), va, &pa) == FALSE)
-			panic("dvma_mapin: null page frame");
-		pa = trunc_page(pa);
-
-#if defined(SUN4M)
-		if (CPU_ISSUN4M) {
-			iommu_enter(tva, pa);
-		} else
-#endif
-		{
-			/*
-			 * pmap_enter distributes this mapping to all
-			 * contexts... maybe we should avoid this extra work
-			 */
-#ifdef notyet
-#if defined(SUN4)
-			if (have_iocache)
-				pa |= PG_IOC;
-#endif
-#endif
-			pmap_kenter_pa(tva, pa | PMAP_NC,
-			    PROT_READ | PROT_WRITE);
-		}
-
-		tva += PAGE_SIZE;
-		va += PAGE_SIZE;
-	}
-	pmap_update(pmap_kernel());
-
-	/*
-	 * XXX Only have to do this on write.
-	 */
-	if (CACHEINFO.c_vactype == VAC_WRITEBACK)	/* XXX */
-		cpuinfo.cache_flush((caddr_t)ova, olen);	/* XXX */
-
-	return kva + off;
-}
-
-/*
- * Remove DVMA mapping of `va' in DVMA space at `dva'.
- */
-void
-dvma_mapout(dva, va, len)
-	vaddr_t	dva, va;
-	int		len;
-{
-	int s, off;
-	int error;
-	int dlen;
-
-	off = (int)dva & PGOFSET;
-	dva -= off;
-	dlen = round_page(len + off);
-
-#if defined(SUN4M)
-	if (CPU_ISSUN4M)
-		iommu_remove(dva, dlen);
-	else
-#endif
-	{
-		pmap_kremove(dva, dlen);
-		pmap_update(pmap_kernel());
-	}
-
-	s = splhigh();
-	error = extent_free(dvmamap_extent, dva, dlen, EX_NOWAIT);
-	if (error)
-		printf("dvma_mapout: extent_free failed\n");
-	splx(s);
-
-	if (CACHEINFO.c_vactype != VAC_NONE)
-		cpuinfo.cache_flush((caddr_t)va, len);
-}
-
-/*
  * Map an IO request into kernel virtual address space.
  */
 void