diff --git a/arch/sparc64/kernel/iommu.c b/arch/sparc64/kernel/iommu.c
index 90a5907080a122938202a35154b777968714c046..d3276ebcfb4715a118268a8cb6c92640a33afcca 100644
--- a/arch/sparc64/kernel/iommu.c
+++ b/arch/sparc64/kernel/iommu.c
@@ -512,124 +512,209 @@ static void dma_4u_unmap_single(struct device *dev, dma_addr_t bus_addr,
 static int dma_4u_map_sg(struct device *dev, struct scatterlist *sglist,
 			 int nelems, enum dma_data_direction direction)
 {
-	unsigned long flags, ctx, i, npages, iopte_protection;
-	struct scatterlist *sg;
+	struct scatterlist *s, *outs, *segstart;
+	unsigned long flags, handle, prot, ctx;
+	dma_addr_t dma_next = 0, dma_addr;
+	unsigned int max_seg_size;
+	int outcount, incount, i;
 	struct strbuf *strbuf;
 	struct iommu *iommu;
-	iopte_t *base;
-	u32 dma_base;
-
-	/* Fast path single entry scatterlists. */
-	if (nelems == 1) {
-		sglist->dma_address =
-			dma_4u_map_single(dev, sg_virt(sglist),
-					  sglist->length, direction);
-		if (unlikely(sglist->dma_address == DMA_ERROR_CODE))
-			return 0;
-		sglist->dma_length = sglist->length;
-		return 1;
-	}
+
+	BUG_ON(direction == DMA_NONE);
 
 	iommu = dev->archdata.iommu;
 	strbuf = dev->archdata.stc;
-
-	if (unlikely(direction == DMA_NONE))
-		goto bad_no_ctx;
-
-	npages = calc_npages(sglist, nelems);
+	if (nelems == 0 || !iommu)
+		return 0;
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	base = alloc_npages(dev, iommu, npages);
 	ctx = 0;
 	if (iommu->iommu_ctxflush)
 		ctx = iommu_alloc_ctx(iommu);
 
-	spin_unlock_irqrestore(&iommu->lock, flags);
-
-	if (base == NULL)
-		goto bad;
-
-	dma_base = iommu->page_table_map_base +
-		((base - iommu->page_table) << IO_PAGE_SHIFT);
-
 	if (strbuf->strbuf_enabled)
-		iopte_protection = IOPTE_STREAMING(ctx);
+		prot = IOPTE_STREAMING(ctx);
 	else
-		iopte_protection = IOPTE_CONSISTENT(ctx);
+		prot = IOPTE_CONSISTENT(ctx);
 	if (direction != DMA_TO_DEVICE)
-		iopte_protection |= IOPTE_WRITE;
-
-	for_each_sg(sglist, sg, nelems, i) {
-		unsigned long paddr = SG_ENT_PHYS_ADDRESS(sg);
-		unsigned long slen = sg->length;
-		unsigned long this_npages;
+		prot |= IOPTE_WRITE;
+
+	outs = s = segstart = &sglist[0];
+	outcount = 1;
+	incount = nelems;
+	handle = 0;
+
+	/* Init first segment length for backout at failure */
+	outs->dma_length = 0;
+
+	max_seg_size = dma_get_max_seg_size(dev);
+	for_each_sg(sglist, s, nelems, i) {
+		unsigned long paddr, npages, entry, slen;
+		iopte_t *base;
+
+		slen = s->length;
+		/* Sanity check */
+		if (slen == 0) {
+			dma_next = 0;
+			continue;
+		}
+		/* Allocate iommu entries for that segment */
+		paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
+		npages = iommu_num_pages(paddr, slen);
+		entry = iommu_range_alloc(dev, iommu, npages, &handle);
+
+		/* Handle failure */
+		if (unlikely(entry == DMA_ERROR_CODE)) {
+			if (printk_ratelimit())
+				printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
+				       " npages %lx\n", iommu, paddr, npages);
+			goto iommu_map_failed;
+		}
 
-		this_npages = iommu_num_pages(paddr, slen);
+		base = iommu->page_table + entry;
 
-		sg->dma_address = dma_base | (paddr & ~IO_PAGE_MASK);
-		sg->dma_length = slen;
+		/* Convert entry to a dma_addr_t */
+		dma_addr = iommu->page_table_map_base +
+			(entry << IO_PAGE_SHIFT);
+		dma_addr |= (s->offset & ~IO_PAGE_MASK);
 
+		/* Insert into HW table */
 		paddr &= IO_PAGE_MASK;
-		while (this_npages--) {
-			iopte_val(*base) = iopte_protection | paddr;
-
+		while (npages--) {
+			iopte_val(*base) = prot | paddr;
 			base++;
 			paddr += IO_PAGE_SIZE;
-			dma_base += IO_PAGE_SIZE;
 		}
+
+		/* If we are in an open segment, try merging */
+		if (segstart != s) {
+			/* We cannot merge if:
+			 * - allocated dma_addr isn't contiguous to previous allocation
+			 */
+			if ((dma_addr != dma_next) ||
+			    (outs->dma_length + s->length > max_seg_size)) {
+				/* Can't merge: create a new segment */
+				segstart = s;
+				outcount++;
+				outs = sg_next(outs);
+			} else {
+				outs->dma_length += s->length;
+			}
+		}
+
+		if (segstart == s) {
+			/* This is a new segment, fill entries */
+			outs->dma_address = dma_addr;
+			outs->dma_length = slen;
+		}
+
+		/* Calculate next page pointer for contiguous check */
+		dma_next = dma_addr + slen;
 	}
 
-	return nelems;
+	spin_unlock_irqrestore(&iommu->lock, flags);
+
+	if (outcount < incount) {
+		outs = sg_next(outs);
+		outs->dma_address = DMA_ERROR_CODE;
+		outs->dma_length = 0;
+	}
+
+	return outcount;
+
+iommu_map_failed:
+	for_each_sg(sglist, s, nelems, i) {
+		if (s->dma_length != 0) {
+			unsigned long vaddr, npages, entry, i;
+			iopte_t *base;
+
+			vaddr = s->dma_address & IO_PAGE_MASK;
+			npages = iommu_num_pages(s->dma_address, s->dma_length);
+			iommu_range_free(iommu, vaddr, npages);
+
+			entry = (vaddr - iommu->page_table_map_base)
+				>> IO_PAGE_SHIFT;
+			base = iommu->page_table + entry;
+
+			for (i = 0; i < npages; i++)
+				iopte_make_dummy(iommu, base + i);
+
+			s->dma_address = DMA_ERROR_CODE;
+			s->dma_length = 0;
+		}
+		if (s == outs)
+			break;
+	}
+	spin_unlock_irqrestore(&iommu->lock, flags);
 
-bad:
-	iommu_free_ctx(iommu, ctx);
-bad_no_ctx:
-	if (printk_ratelimit())
-		WARN_ON(1);
 	return 0;
 }
 
+/* If contexts are being used, they are the same in all of the mappings
+ * we make for a particular SG.
+ */
+static unsigned long fetch_sg_ctx(struct iommu *iommu, struct scatterlist *sg)
+{
+	unsigned long ctx = 0;
+
+	if (iommu->iommu_ctxflush) {
+		iopte_t *base;
+		u32 bus_addr;
+
+		bus_addr = sg->dma_address & IO_PAGE_MASK;
+		base = iommu->page_table +
+			((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
+
+		ctx = (iopte_val(*base) & IOPTE_CONTEXT) >> 47UL;
+	}
+	return ctx;
+}
+
 static void dma_4u_unmap_sg(struct device *dev, struct scatterlist *sglist,
 			    int nelems, enum dma_data_direction direction)
 {
-	unsigned long flags, ctx, i, npages;
+	unsigned long flags, ctx;
+	struct scatterlist *sg;
 	struct strbuf *strbuf;
 	struct iommu *iommu;
-	iopte_t *base;
-	u32 bus_addr;
 
-	if (unlikely(direction == DMA_NONE)) {
-		if (printk_ratelimit())
-			WARN_ON(1);
-	}
+	BUG_ON(direction == DMA_NONE);
 
 	iommu = dev->archdata.iommu;
 	strbuf = dev->archdata.stc;
 
-	bus_addr = sglist->dma_address & IO_PAGE_MASK;
+	ctx = fetch_sg_ctx(iommu, sglist);
 
-	npages = calc_npages(sglist, nelems);
+	spin_lock_irqsave(&iommu->lock, flags);
 
-	base = iommu->page_table +
-		((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
+	sg = sglist;
+	while (nelems--) {
+		dma_addr_t dma_handle = sg->dma_address;
+		unsigned int len = sg->dma_length;
+		unsigned long npages, entry;
+		iopte_t *base;
+		int i;
 
-	spin_lock_irqsave(&iommu->lock, flags);
+		if (!len)
+			break;
+		npages = iommu_num_pages(dma_handle, len);
+		iommu_range_free(iommu, dma_handle, npages);
 
-	/* Record the context, if any. */
-	ctx = 0;
-	if (iommu->iommu_ctxflush)
-		ctx = (iopte_val(*base) & IOPTE_CONTEXT) >> 47UL;
+		entry = ((dma_handle - iommu->page_table_map_base)
+			 >> IO_PAGE_SHIFT);
+		base = iommu->page_table + entry;
 
-	/* Step 1: Kick data out of streaming buffers if necessary. */
-	if (strbuf->strbuf_enabled)
-		strbuf_flush(strbuf, iommu, bus_addr, ctx, npages, direction);
+		dma_handle &= IO_PAGE_MASK;
+		if (strbuf->strbuf_enabled)
+			strbuf_flush(strbuf, iommu, dma_handle, ctx,
+				     npages, direction);
 
-	/* Step 2: Clear out the TSB entries. */
-	for (i = 0; i < npages; i++)
-		iopte_make_dummy(iommu, base + i);
+		for (i = 0; i < npages; i++)
+			iopte_make_dummy(iommu, base + i);
 
-	iommu_range_free(iommu, bus_addr, npages);
+		sg = sg_next(sg);
+	}
 
 	iommu_free_ctx(iommu, ctx);
 
diff --git a/arch/sparc64/kernel/pci_sun4v.c b/arch/sparc64/kernel/pci_sun4v.c
index c8b6199a5dc4b405b4153eb0b084842266391b71..ddca6c6c0b492709193e1a3c8a86dcb30f884fbf 100644
--- a/arch/sparc64/kernel/pci_sun4v.c
+++ b/arch/sparc64/kernel/pci_sun4v.c
@@ -89,6 +89,17 @@ static long iommu_batch_flush(struct iommu_batch *p)
 	return 0;
 }
 
+static inline void iommu_batch_new_entry(unsigned long entry)
+{
+	struct iommu_batch *p = &__get_cpu_var(iommu_batch);
+
+	if (p->entry + p->npages == entry)
+		return;
+	if (p->entry != ~0UL)
+		iommu_batch_flush(p);
+	p->entry = entry;
+}
+
 /* Interrupts must be disabled.  */
 static inline long iommu_batch_add(u64 phys_page)
 {
@@ -320,88 +331,131 @@ static void dma_4v_unmap_single(struct device *dev, dma_addr_t bus_addr,
 static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
 			 int nelems, enum dma_data_direction direction)
 {
-	unsigned long flags, npages, i, prot;
-	u32 dma_base, orig_dma_base;
-	struct scatterlist *sg;
+	struct scatterlist *s, *outs, *segstart;
+	unsigned long flags, handle, prot;
+	dma_addr_t dma_next = 0, dma_addr;
+	unsigned int max_seg_size;
+	int outcount, incount, i;
 	struct iommu *iommu;
-	long entry, err;
-
-	/* Fast path single entry scatterlists. */
-	if (nelems == 1) {
-		sglist->dma_address =
-			dma_4v_map_single(dev, sg_virt(sglist),
-					  sglist->length, direction);
-		if (unlikely(sglist->dma_address == DMA_ERROR_CODE))
-			return 0;
-		sglist->dma_length = sglist->length;
-		return 1;
-	}
+	long err;
+
+	BUG_ON(direction == DMA_NONE);
 
 	iommu = dev->archdata.iommu;
+	if (nelems == 0 || !iommu)
+		return 0;
 	
-	if (unlikely(direction == DMA_NONE))
-		goto bad;
-
-	npages = calc_npages(sglist, nelems);
+	prot = HV_PCI_MAP_ATTR_READ;
+	if (direction != DMA_TO_DEVICE)
+		prot |= HV_PCI_MAP_ATTR_WRITE;
 
-	spin_lock_irqsave(&iommu->lock, flags);
-	entry = iommu_range_alloc(dev, iommu, npages, NULL);
-	spin_unlock_irqrestore(&iommu->lock, flags);
+	outs = s = segstart = &sglist[0];
+	outcount = 1;
+	incount = nelems;
+	handle = 0;
 
-	if (unlikely(entry == DMA_ERROR_CODE))
-		goto bad;
+	/* Init first segment length for backout at failure */
+	outs->dma_length = 0;
 
-	orig_dma_base = dma_base = iommu->page_table_map_base +
-		(entry << IO_PAGE_SHIFT);
+	spin_lock_irqsave(&iommu->lock, flags);
 
-	prot = HV_PCI_MAP_ATTR_READ;
-	if (direction != DMA_TO_DEVICE)
-		prot |= HV_PCI_MAP_ATTR_WRITE;
+	iommu_batch_start(dev, prot, ~0UL);
 
-	local_irq_save(flags);
+	max_seg_size = dma_get_max_seg_size(dev);
+	for_each_sg(sglist, s, nelems, i) {
+		unsigned long paddr, npages, entry, slen;
 
-	iommu_batch_start(dev, prot, entry);
+		slen = s->length;
+		/* Sanity check */
+		if (slen == 0) {
+			dma_next = 0;
+			continue;
+		}
+		/* Allocate iommu entries for that segment */
+		paddr = (unsigned long) SG_ENT_PHYS_ADDRESS(s);
+		npages = iommu_num_pages(paddr, slen);
+		entry = iommu_range_alloc(dev, iommu, npages, &handle);
 
-	for_each_sg(sglist, sg, nelems, i) {
-		unsigned long paddr = SG_ENT_PHYS_ADDRESS(sg);
-		unsigned long slen = sg->length;
-		unsigned long this_npages;
+		/* Handle failure */
+		if (unlikely(entry == DMA_ERROR_CODE)) {
+			if (printk_ratelimit())
+				printk(KERN_INFO "iommu_alloc failed, iommu %p paddr %lx"
+				       " npages %lx\n", iommu, paddr, npages);
+			goto iommu_map_failed;
+		}
 
-		this_npages = iommu_num_pages(paddr, slen);
+		iommu_batch_new_entry(entry);
 
-		sg->dma_address = dma_base | (paddr & ~IO_PAGE_MASK);
-		sg->dma_length = slen;
+		/* Convert entry to a dma_addr_t */
+		dma_addr = iommu->page_table_map_base +
+			(entry << IO_PAGE_SHIFT);
+		dma_addr |= (s->offset & ~IO_PAGE_MASK);
 
+		/* Insert into HW table */
 		paddr &= IO_PAGE_MASK;
-		while (this_npages--) {
+		while (npages--) {
 			err = iommu_batch_add(paddr);
-			if (unlikely(err < 0L)) {
-				local_irq_restore(flags);
+			if (unlikely(err < 0L))
 				goto iommu_map_failed;
+			paddr += IO_PAGE_SIZE;
+		}
+
+		/* If we are in an open segment, try merging */
+		if (segstart != s) {
+			/* We cannot merge if:
+			 * - allocated dma_addr isn't contiguous to previous allocation
+			 */
+			if ((dma_addr != dma_next) ||
+			    (outs->dma_length + s->length > max_seg_size)) {
+				/* Can't merge: create a new segment */
+				segstart = s;
+				outcount++;
+				outs = sg_next(outs);
+			} else {
+				outs->dma_length += s->length;
 			}
+		}
 
-			paddr += IO_PAGE_SIZE;
-			dma_base += IO_PAGE_SIZE;
+		if (segstart == s) {
+			/* This is a new segment, fill entries */
+			outs->dma_address = dma_addr;
+			outs->dma_length = slen;
 		}
+
+		/* Calculate next page pointer for contiguous check */
+		dma_next = dma_addr + slen;
 	}
 
 	err = iommu_batch_end();
 
-	local_irq_restore(flags);
-
 	if (unlikely(err < 0L))
 		goto iommu_map_failed;
 
-	return nelems;
+	spin_unlock_irqrestore(&iommu->lock, flags);
 
-bad:
-	if (printk_ratelimit())
-		WARN_ON(1);
-	return 0;
+	if (outcount < incount) {
+		outs = sg_next(outs);
+		outs->dma_address = DMA_ERROR_CODE;
+		outs->dma_length = 0;
+	}
+
+	return outcount;
 
 iommu_map_failed:
-	spin_lock_irqsave(&iommu->lock, flags);
-	iommu_range_free(iommu, orig_dma_base, npages);
+	for_each_sg(sglist, s, nelems, i) {
+		if (s->dma_length != 0) {
+			unsigned long vaddr, npages;
+
+			vaddr = s->dma_address & IO_PAGE_MASK;
+			npages = iommu_num_pages(s->dma_address, s->dma_length);
+			iommu_range_free(iommu, vaddr, npages);
+			/* XXX demap? XXX */
+			s->dma_address = DMA_ERROR_CODE;
+			s->dma_length = 0;
+		}
+		if (s == outs)
+			break;
+	}
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
 	return 0;
@@ -410,39 +464,43 @@ static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,
 static void dma_4v_unmap_sg(struct device *dev, struct scatterlist *sglist,
 			    int nelems, enum dma_data_direction direction)
 {
-	unsigned long flags, npages;
 	struct pci_pbm_info *pbm;
-	u32 devhandle, bus_addr;
+	struct scatterlist *sg;
 	struct iommu *iommu;
-	long entry;
+	unsigned long flags;
+	u32 devhandle;
 
-	if (unlikely(direction == DMA_NONE)) {
-		if (printk_ratelimit())
-			WARN_ON(1);
-	}
+	BUG_ON(direction == DMA_NONE);
 
 	iommu = dev->archdata.iommu;
 	pbm = dev->archdata.host_controller;
 	devhandle = pbm->devhandle;
 	
-	bus_addr = sglist->dma_address & IO_PAGE_MASK;
-
-	npages = calc_npages(sglist, nelems);
-
-	entry = ((bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
-
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	iommu_range_free(iommu, bus_addr, npages);
-
-	do {
-		unsigned long num;
+	sg = sglist;
+	while (nelems--) {
+		dma_addr_t dma_handle = sg->dma_address;
+		unsigned int len = sg->dma_length;
+		unsigned long npages, entry;
+
+		if (!len)
+			break;
+		npages = iommu_num_pages(dma_handle, len);
+		iommu_range_free(iommu, dma_handle, npages);
+
+		entry = ((dma_handle - iommu->page_table_map_base) >> IO_PAGE_SHIFT);
+		while (npages) {
+			unsigned long num;
+
+			num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
+						    npages);
+			entry += num;
+			npages -= num;
+		}
 
-		num = pci_sun4v_iommu_demap(devhandle, HV_PCI_TSBID(0, entry),
-					    npages);
-		entry += num;
-		npages -= num;
-	} while (npages != 0);
+		sg = sg_next(sg);
+	}
 
 	spin_unlock_irqrestore(&iommu->lock, flags);
 }
diff --git a/include/asm-sparc64/io.h b/include/asm-sparc64/io.h
index b6ece223562de86719dc403595eca551140be570..c299b853b5babae9520be2238620319cf490c149 100644
--- a/include/asm-sparc64/io.h
+++ b/include/asm-sparc64/io.h
@@ -16,7 +16,7 @@
 /* BIO layer definitions. */
 extern unsigned long kern_base, kern_size;
 #define page_to_phys(page)	(page_to_pfn(page) << PAGE_SHIFT)
-#define BIO_VMERGE_BOUNDARY	0
+#define BIO_VMERGE_BOUNDARY	8192
 
 static inline u8 _inb(unsigned long addr)
 {