[SPARC64]: IOMMU allocations using iommu-helper layer.

Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/arch/sparc64/Kconfig b/arch/sparc64/Kconfig
index b810f2b..4ac22f4 100644
--- a/arch/sparc64/Kconfig
+++ b/arch/sparc64/Kconfig
@@ -40,6 +40,10 @@
 	bool
 	default y
 
+config IOMMU_HELPER
+	bool
+	default y
+
 config QUICKLIST
 	bool
 	default y
diff --git a/arch/sparc64/kernel/iommu.c b/arch/sparc64/kernel/iommu.c
index 5623a4d..90a5907 100644
--- a/arch/sparc64/kernel/iommu.c
+++ b/arch/sparc64/kernel/iommu.c
@@ -1,6 +1,6 @@
 /* iommu.c: Generic sparc64 IOMMU support.
  *
- * Copyright (C) 1999, 2007 David S. Miller (davem@davemloft.net)
+ * Copyright (C) 1999, 2007, 2008 David S. Miller (davem@davemloft.net)
  * Copyright (C) 1999, 2000 Jakub Jelinek (jakub@redhat.com)
  */
 
@@ -10,6 +10,7 @@
 #include <linux/device.h>
 #include <linux/dma-mapping.h>
 #include <linux/errno.h>
+#include <linux/iommu-helper.h>
 
 #ifdef CONFIG_PCI
 #include <linux/pci.h>
@@ -41,7 +42,7 @@
 			       "i" (ASI_PHYS_BYPASS_EC_E))
 
 /* Must be invoked under the IOMMU lock. */
-static void __iommu_flushall(struct iommu *iommu)
+static void iommu_flushall(struct iommu *iommu)
 {
 	if (iommu->iommu_flushinv) {
 		iommu_write(iommu->iommu_flushinv, ~(u64)0);
@@ -83,54 +84,91 @@
 	iopte_val(*iopte) = val;
 }
 
-/* Based largely upon the ppc64 iommu allocator.  */
-static long arena_alloc(struct iommu *iommu, unsigned long npages)
+/* Based almost entirely upon the ppc64 iommu allocator.  If you use the 'handle'
+ * facility it must all be done in one pass while under the iommu lock.
+ *
+ * On sun4u platforms, we only flush the IOMMU once every time we've passed
+ * over the entire page table doing allocations.  Therefore we only ever advance
+ * the hint and cannot backtrack it.
+ */
+unsigned long iommu_range_alloc(struct device *dev,
+				struct iommu *iommu,
+				unsigned long npages,
+				unsigned long *handle)
 {
+	unsigned long n, end, start, limit, boundary_size;
 	struct iommu_arena *arena = &iommu->arena;
-	unsigned long n, i, start, end, limit;
-	int pass;
+	int pass = 0;
+
+	/* This allocator was derived from x86_64's bit string search */
+
+	/* Sanity check */
+	if (unlikely(npages == 0)) {
+		if (printk_ratelimit())
+			WARN_ON(1);
+		return DMA_ERROR_CODE;
+	}
+
+	if (handle && *handle)
+		start = *handle;
+	else
+		start = arena->hint;
 
 	limit = arena->limit;
-	start = arena->hint;
-	pass = 0;
 
-again:
-	n = find_next_zero_bit(arena->map, limit, start);
-	end = n + npages;
-	if (unlikely(end >= limit)) {
+	/* The case below can happen if we have a small segment appended
+	 * to a large, or when the previous alloc was at the very end of
+	 * the available space. If so, go back to the beginning and flush.
+	 */
+	if (start >= limit) {
+		start = 0;
+		if (iommu->flush_all)
+			iommu->flush_all(iommu);
+	}
+
+ again:
+
+	if (dev)
+		boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
+				      1 << IO_PAGE_SHIFT);
+	else
+		boundary_size = ALIGN(1UL << 32, 1 << IO_PAGE_SHIFT);
+
+	n = iommu_area_alloc(arena->map, limit, start, npages, 0,
+			     boundary_size >> IO_PAGE_SHIFT, 0);
+	if (n == -1) {
 		if (likely(pass < 1)) {
-			limit = start;
+			/* First failure, rescan from the beginning.  */
 			start = 0;
-			__iommu_flushall(iommu);
+			if (iommu->flush_all)
+				iommu->flush_all(iommu);
 			pass++;
 			goto again;
 		} else {
-			/* Scanned the whole thing, give up. */
-			return -1;
+			/* Second failure, give up */
+			return DMA_ERROR_CODE;
 		}
 	}
 
-	for (i = n; i < end; i++) {
-		if (test_bit(i, arena->map)) {
-			start = i + 1;
-			goto again;
-		}
-	}
-
-	for (i = n; i < end; i++)
-		__set_bit(i, arena->map);
+	end = n + npages;
 
 	arena->hint = end;
 
+	/* Update handle for SG allocations */
+	if (handle)
+		*handle = end;
+
 	return n;
 }
 
-static void arena_free(struct iommu_arena *arena, unsigned long base, unsigned long npages)
+void iommu_range_free(struct iommu *iommu, dma_addr_t dma_addr, unsigned long npages)
 {
-	unsigned long i;
+	struct iommu_arena *arena = &iommu->arena;
+	unsigned long entry;
 
-	for (i = base; i < (base + npages); i++)
-		__clear_bit(i, arena->map);
+	entry = (dma_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
+
+	iommu_area_free(arena->map, entry, npages);
 }
 
 int iommu_table_init(struct iommu *iommu, int tsbsize,
@@ -156,6 +194,9 @@
 	}
 	iommu->arena.limit = num_tsb_entries;
 
+	if (tlb_type != hypervisor)
+		iommu->flush_all = iommu_flushall;
+
 	/* Allocate and initialize the dummy page which we
 	 * set inactive IO PTEs to point to.
 	 */
@@ -192,22 +233,18 @@
 	return -ENOMEM;
 }
 
-static inline iopte_t *alloc_npages(struct iommu *iommu, unsigned long npages)
+static inline iopte_t *alloc_npages(struct device *dev, struct iommu *iommu,
+				    unsigned long npages)
 {
-	long entry;
+	unsigned long entry;
 
-	entry = arena_alloc(iommu, npages);
-	if (unlikely(entry < 0))
+	entry = iommu_range_alloc(dev, iommu, npages, NULL);
+	if (unlikely(entry == DMA_ERROR_CODE))
 		return NULL;
 
 	return iommu->page_table + entry;
 }
 
-static inline void free_npages(struct iommu *iommu, dma_addr_t base, unsigned long npages)
-{
-	arena_free(&iommu->arena, base >> IO_PAGE_SHIFT, npages);
-}
-
 static int iommu_alloc_ctx(struct iommu *iommu)
 {
 	int lowest = iommu->ctx_lowest_free;
@@ -258,7 +295,7 @@
 	iommu = dev->archdata.iommu;
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	iopte = alloc_npages(iommu, size >> IO_PAGE_SHIFT);
+	iopte = alloc_npages(dev, iommu, size >> IO_PAGE_SHIFT);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
 	if (unlikely(iopte == NULL)) {
@@ -296,7 +333,7 @@
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	free_npages(iommu, dvma - iommu->page_table_map_base, npages);
+	iommu_range_free(iommu, dvma, npages);
 
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
@@ -327,7 +364,7 @@
 	npages >>= IO_PAGE_SHIFT;
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	base = alloc_npages(iommu, npages);
+	base = alloc_npages(dev, iommu, npages);
 	ctx = 0;
 	if (iommu->iommu_ctxflush)
 		ctx = iommu_alloc_ctx(iommu);
@@ -465,7 +502,7 @@
 	for (i = 0; i < npages; i++)
 		iopte_make_dummy(iommu, base + i);
 
-	free_npages(iommu, bus_addr - iommu->page_table_map_base, npages);
+	iommu_range_free(iommu, bus_addr, npages);
 
 	iommu_free_ctx(iommu, ctx);
 
@@ -503,7 +540,7 @@
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	base = alloc_npages(iommu, npages);
+	base = alloc_npages(dev, iommu, npages);
 	ctx = 0;
 	if (iommu->iommu_ctxflush)
 		ctx = iommu_alloc_ctx(iommu);
@@ -592,7 +629,7 @@
 	for (i = 0; i < npages; i++)
 		iopte_make_dummy(iommu, base + i);
 
-	free_npages(iommu, bus_addr - iommu->page_table_map_base, npages);
+	iommu_range_free(iommu, bus_addr, npages);
 
 	iommu_free_ctx(iommu, ctx);
 
diff --git a/arch/sparc64/kernel/iommu_common.h b/arch/sparc64/kernel/iommu_common.h
index 8390f04..0713bd5 100644
--- a/arch/sparc64/kernel/iommu_common.h
+++ b/arch/sparc64/kernel/iommu_common.h
@@ -58,4 +58,12 @@
 	return npages;
 }
 
+extern unsigned long iommu_range_alloc(struct device *dev,
+				       struct iommu *iommu,
+				       unsigned long npages,
+				       unsigned long *handle);
+extern void iommu_range_free(struct iommu *iommu,
+			     dma_addr_t dma_addr,
+			     unsigned long npages);
+
 #endif /* _IOMMU_COMMON_H */
diff --git a/arch/sparc64/kernel/pci_sun4v.c b/arch/sparc64/kernel/pci_sun4v.c
index 61baf8d..c8b6199 100644
--- a/arch/sparc64/kernel/pci_sun4v.c
+++ b/arch/sparc64/kernel/pci_sun4v.c
@@ -1,6 +1,6 @@
 /* pci_sun4v.c: SUN4V specific PCI controller support.
  *
- * Copyright (C) 2006, 2007 David S. Miller (davem@davemloft.net)
+ * Copyright (C) 2006, 2007, 2008 David S. Miller (davem@davemloft.net)
  */
 
 #include <linux/kernel.h>
@@ -113,54 +113,6 @@
 	return iommu_batch_flush(p);
 }
 
-static long arena_alloc(struct iommu_arena *arena, unsigned long npages)
-{
-	unsigned long n, i, start, end, limit;
-	int pass;
-
-	limit = arena->limit;
-	start = arena->hint;
-	pass = 0;
-
-again:
-	n = find_next_zero_bit(arena->map, limit, start);
-	end = n + npages;
-	if (unlikely(end >= limit)) {
-		if (likely(pass < 1)) {
-			limit = start;
-			start = 0;
-			pass++;
-			goto again;
-		} else {
-			/* Scanned the whole thing, give up. */
-			return -1;
-		}
-	}
-
-	for (i = n; i < end; i++) {
-		if (test_bit(i, arena->map)) {
-			start = i + 1;
-			goto again;
-		}
-	}
-
-	for (i = n; i < end; i++)
-		__set_bit(i, arena->map);
-
-	arena->hint = end;
-
-	return n;
-}
-
-static void arena_free(struct iommu_arena *arena, unsigned long base,
-		       unsigned long npages)
-{
-	unsigned long i;
-
-	for (i = base; i < (base + npages); i++)
-		__clear_bit(i, arena->map);
-}
-
 static void *dma_4v_alloc_coherent(struct device *dev, size_t size,
 				   dma_addr_t *dma_addrp, gfp_t gfp)
 {
@@ -185,11 +137,11 @@
 	iommu = dev->archdata.iommu;
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	entry = arena_alloc(&iommu->arena, npages);
+	entry = iommu_range_alloc(dev, iommu, npages, NULL);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
-	if (unlikely(entry < 0L))
-		goto arena_alloc_fail;
+	if (unlikely(entry == DMA_ERROR_CODE))
+		goto range_alloc_fail;
 
 	*dma_addrp = (iommu->page_table_map_base +
 		      (entry << IO_PAGE_SHIFT));
@@ -219,10 +171,10 @@
 iommu_map_fail:
 	/* Interrupts are disabled.  */
 	spin_lock(&iommu->lock);
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, *dma_addrp, npages);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
-arena_alloc_fail:
+range_alloc_fail:
 	free_pages(first_page, order);
 	return NULL;
 }
@@ -243,7 +195,7 @@
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, dvma, npages);
 
 	do {
 		unsigned long num;
@@ -281,10 +233,10 @@
 	npages >>= IO_PAGE_SHIFT;
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	entry = arena_alloc(&iommu->arena, npages);
+	entry = iommu_range_alloc(dev, iommu, npages, NULL);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
-	if (unlikely(entry < 0L))
+	if (unlikely(entry == DMA_ERROR_CODE))
 		goto bad;
 
 	bus_addr = (iommu->page_table_map_base +
@@ -319,7 +271,7 @@
 iommu_map_fail:
 	/* Interrupts are disabled.  */
 	spin_lock(&iommu->lock);
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, bus_addr, npages);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
 	return DMA_ERROR_CODE;
@@ -350,9 +302,9 @@
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, bus_addr, npages);
 
+	entry = (bus_addr - iommu->page_table_map_base) >> IO_PAGE_SHIFT;
 	do {
 		unsigned long num;
 
@@ -369,10 +321,10 @@
 			 int nelems, enum dma_data_direction direction)
 {
 	unsigned long flags, npages, i, prot;
+	u32 dma_base, orig_dma_base;
 	struct scatterlist *sg;
 	struct iommu *iommu;
 	long entry, err;
-	u32 dma_base;
 
 	/* Fast path single entry scatterlists. */
 	if (nelems == 1) {
@@ -393,13 +345,13 @@
 	npages = calc_npages(sglist, nelems);
 
 	spin_lock_irqsave(&iommu->lock, flags);
-	entry = arena_alloc(&iommu->arena, npages);
+	entry = iommu_range_alloc(dev, iommu, npages, NULL);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
-	if (unlikely(entry < 0L))
+	if (unlikely(entry == DMA_ERROR_CODE))
 		goto bad;
 
-	dma_base = iommu->page_table_map_base +
+	orig_dma_base = dma_base = iommu->page_table_map_base +
 		(entry << IO_PAGE_SHIFT);
 
 	prot = HV_PCI_MAP_ATTR_READ;
@@ -449,7 +401,7 @@
 
 iommu_map_failed:
 	spin_lock_irqsave(&iommu->lock, flags);
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, orig_dma_base, npages);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
 	return 0;
@@ -481,7 +433,7 @@
 
 	spin_lock_irqsave(&iommu->lock, flags);
 
-	arena_free(&iommu->arena, entry, npages);
+	iommu_range_free(iommu, bus_addr, npages);
 
 	do {
 		unsigned long num;
diff --git a/include/asm-sparc64/iommu.h b/include/asm-sparc64/iommu.h
index 9eac667..46325dd 100644
--- a/include/asm-sparc64/iommu.h
+++ b/include/asm-sparc64/iommu.h
@@ -26,6 +26,7 @@
 struct iommu {
 	spinlock_t		lock;
 	struct iommu_arena	arena;
+	void			(*flush_all)(struct iommu *);
 	iopte_t			*page_table;
 	u32			page_table_map_base;
 	unsigned long		iommu_control;