Index: include/bus_private.h
===================================================================
--- include/bus_private.h	(revision 182964)
+++ include/bus_private.h	(working copy)
@@ -66,9 +66,10 @@
 	int			dm_flags;		/* (p) */
 };
 
-/* Flag values. */
-#define	DMF_LOADED	1	/* Map is loaded */
-#define	DMF_COHERENT	2	/* Coherent mapping requested */
+/* Flag values */
+#define	DMF_LOADED	(1 << 0)	/* Map is loaded. */
+#define	DMF_COHERENT	(1 << 1)	/* Coherent mapping requested. */
+#define	DMF_STREAMED	(1 << 2)	/* Streaming cache used. */
 
 int sparc64_dma_alloc_map(bus_dma_tag_t dmat, bus_dmamap_t *mapp);
 void sparc64_dma_free_map(bus_dma_tag_t dmat, bus_dmamap_t map);
Index: pci/schizo.c
===================================================================
--- pci/schizo.c	(revision 186290)
+++ pci/schizo.c	(working copy)
@@ -449,14 +449,18 @@
 		tc_init(tc);
 	}
 
-	/* Set up the IOMMU.  Both Schizo and Tomatillo have one per PBM. */
+	/*
+	 * Set up the IOMMU.  Schizo, Tomatillo and XMITS all have
+	 * one per PBM.  Schizo and XMITS also have a streaming buffer,
+	 * in Schizo version < 5 (i.e. revision < 2.3) it's affected
+	 * by several errata and basically unusable though.
+	 */
 	sc->sc_is.is_pmaxaddr = IOMMU_MAXADDR(STX_IOMMU_BITS);
 	sc->sc_is.is_sb[0] = 0;
 	sc->sc_is.is_sb[1] = 0;
-#ifdef notyet
-	if (OF_getproplen(node, "no-streaming-cache") < 0)
+	if (OF_getproplen(node, "no-streaming-cache") < 0 &&
+	    !(sc->sc_mode == SCHIZO_MODE_SCZ && sc->sc_ver < 5))
 		sc->sc_is.is_sb[0] = STX_PCI_STRBUF;
-#endif
 
 #define	TSBCASE(x)							\
 	case (IOTSB_BASESZ << (x)) << (IO_PAGE_SHIFT - IOTTE_SHIFT):	\
Index: sparc64/iommu.c
===================================================================
--- sparc64/iommu.c	(revision 185008)
+++ sparc64/iommu.c	(working copy)
@@ -119,7 +119,6 @@
  * - When running out of DVMA space, return EINPROGRESS in the non-
  *   BUS_DMA_NOWAIT case and delay the callback until sufficient space
  *   becomes available.
- * - Use the streaming cache unless BUS_DMA_COHERENT is specified.
  */
 
 #include "opt_iommu.h"
@@ -527,17 +526,8 @@
 iommu_use_streaming(struct iommu_state *is, bus_dmamap_t map, bus_size_t size)
 {
 
-	/*
-	 * This cannot be enabled yet, as many driver are still missing
-	 * bus_dmamap_sync() calls.  As soon as there is a BUS_DMA_STREAMING
-	 * flag, this should be reenabled conditionally on it.
-	 */
-#ifdef notyet
 	return (size >= IOMMU_STREAM_THRESH && IOMMU_HAS_SB(is) &&
 	    (map->dm_flags & DMF_COHERENT) == 0);
-#else
-	return (0);
-#endif
 }
 
 /*
@@ -833,8 +823,8 @@
 	bus_size_t sgsize, esize;
 	vm_offset_t vaddr, voffs;
 	vm_paddr_t curaddr;
-	int error, sgcnt, firstpg, stream;
 	pmap_t pmap = NULL;
+	int error, firstpg, sgcnt;
 
 	KASSERT(buflen != 0, ("%s: buflen == 0!", __func__));
 	if (buflen > dt->dt_maxsize)
@@ -855,7 +845,9 @@
 
 	sgcnt = *segp;
 	firstpg = 1;
-	stream = iommu_use_streaming(is, map, buflen);
+	map->dm_flags &= ~DMF_STREAMED;
+	map->dm_flags |= iommu_use_streaming(is, map, buflen) != 0 ?
+	    DMF_STREAMED : 0;
 	for (; buflen > 0; ) {
 		/*
 		 * Get the physical address for this page.
@@ -876,7 +868,7 @@
 		vaddr += sgsize;
 
 		iommu_enter(is, trunc_io_page(dvmaddr), trunc_io_page(curaddr),
-		    stream, flags);
+		    (map->dm_flags & DMF_STREAMED) != 0, flags);
 
 		/*
 		 * Chop the chunk up into segments of at most maxsegsz, but try
@@ -1141,7 +1133,7 @@
 	/* XXX This is probably bogus. */
 	if ((op & BUS_DMASYNC_PREREAD) != 0)
 		membar(Sync);
-	if ((map->dm_flags & DMF_COHERENT) == 0 && IOMMU_HAS_SB(is) &&
+	if ((map->dm_flags & DMF_STREAMED) != 0 &&
 	    ((op & BUS_DMASYNC_POSTREAD) != 0 ||
 	    (op & BUS_DMASYNC_PREWRITE) != 0)) {
 		IS_LOCK(is);