Merge commit 'v2.6.34-rc1' into perf/urgent

Conflicts:
	tools/perf/util/probe-event.c

Merge reason: Pick up -rc1 and resolve the conflict as well.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 646 insertions, 194 deletions

diff --git a/arch/sh/mm/pmb.c b/arch/sh/mm/pmb.c
index 280f6a166035..a4662e2782c3 100644
--- a/arch/sh/mm/pmb.c
+++ b/arch/sh/mm/pmb.c
@@ -3,11 +3,8 @@
  *
  * Privileged Space Mapping Buffer (PMB) Support.
  *
- * Copyright (C) 2005, 2006, 2007 Paul Mundt
- *
- * P1/P2 Section mapping definitions from map32.h, which was:
- *
- *	Copyright 2003 (c) Lineo Solutions,Inc.
+ * Copyright (C) 2005 - 2010  Paul Mundt
+ * Copyright (C) 2010  Matt Fleming
  *
  * This file is subject to the terms and conditions of the GNU General Public
  * License.  See the file "COPYING" in the main directory of this archive
@@ -24,47 +21,216 @@
 #include <linux/fs.h>
 #include <linux/seq_file.h>
 #include <linux/err.h>
+#include <linux/io.h>
+#include <linux/spinlock.h>
+#include <linux/vmalloc.h>
+#include <asm/cacheflush.h>
+#include <asm/sizes.h>
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
+#include <asm/page.h>
 #include <asm/mmu.h>
-#include <asm/io.h>
 #include <asm/mmu_context.h>
 
-#define NR_PMB_ENTRIES	16
+struct pmb_entry;
+
+struct pmb_entry {
+	unsigned long vpn;
+	unsigned long ppn;
+	unsigned long flags;
+	unsigned long size;
+
+	spinlock_t lock;
 
-static void __pmb_unmap(struct pmb_entry *);
+	/*
+	 * 0 .. NR_PMB_ENTRIES for specific entry selection, or
+	 * PMB_NO_ENTRY to search for a free one
+	 */
+	int entry;
 
+	/* Adjacent entry link for contiguous multi-entry mappings */
+	struct pmb_entry *link;
+};
+
+static struct {
+	unsigned long size;
+	int flag;
+} pmb_sizes[] = {
+	{ .size	= SZ_512M, .flag = PMB_SZ_512M, },
+	{ .size = SZ_128M, .flag = PMB_SZ_128M, },
+	{ .size = SZ_64M,  .flag = PMB_SZ_64M,  },
+	{ .size = SZ_16M,  .flag = PMB_SZ_16M,  },
+};
+
+static void pmb_unmap_entry(struct pmb_entry *, int depth);
+
+static DEFINE_RWLOCK(pmb_rwlock);
 static struct pmb_entry pmb_entry_list[NR_PMB_ENTRIES];
-static unsigned long pmb_map;
+static DECLARE_BITMAP(pmb_map, NR_PMB_ENTRIES);
+
+static unsigned int pmb_iomapping_enabled;
 
-static inline unsigned long mk_pmb_entry(unsigned int entry)
+static __always_inline unsigned long mk_pmb_entry(unsigned int entry)
 {
 	return (entry & PMB_E_MASK) << PMB_E_SHIFT;
 }
 
-static inline unsigned long mk_pmb_addr(unsigned int entry)
+static __always_inline unsigned long mk_pmb_addr(unsigned int entry)
 {
 	return mk_pmb_entry(entry) | PMB_ADDR;
 }
 
-static inline unsigned long mk_pmb_data(unsigned int entry)
+static __always_inline unsigned long mk_pmb_data(unsigned int entry)
 {
 	return mk_pmb_entry(entry) | PMB_DATA;
 }
 
-static int pmb_alloc_entry(void)
+static __always_inline unsigned int pmb_ppn_in_range(unsigned long ppn)
+{
+	return ppn >= __pa(memory_start) && ppn < __pa(memory_end);
+}
+
+/*
+ * Ensure that the PMB entries match our cache configuration.
+ *
+ * When we are in 32-bit address extended mode, CCR.CB becomes
+ * invalid, so care must be taken to manually adjust cacheable
+ * translations.
+ */
+static __always_inline unsigned long pmb_cache_flags(void)
+{
+	unsigned long flags = 0;
+
+#if defined(CONFIG_CACHE_OFF)
+	flags |= PMB_WT | PMB_UB;
+#elif defined(CONFIG_CACHE_WRITETHROUGH)
+	flags |= PMB_C | PMB_WT | PMB_UB;
+#elif defined(CONFIG_CACHE_WRITEBACK)
+	flags |= PMB_C;
+#endif
+
+	return flags;
+}
+
+/*
+ * Convert typical pgprot value to the PMB equivalent
+ */
+static inline unsigned long pgprot_to_pmb_flags(pgprot_t prot)
+{
+	unsigned long pmb_flags = 0;
+	u64 flags = pgprot_val(prot);
+
+	if (flags & _PAGE_CACHABLE)
+		pmb_flags |= PMB_C;
+	if (flags & _PAGE_WT)
+		pmb_flags |= PMB_WT | PMB_UB;
+
+	return pmb_flags;
+}
+
+static inline bool pmb_can_merge(struct pmb_entry *a, struct pmb_entry *b)
 {
-	unsigned int pos;
+	return (b->vpn == (a->vpn + a->size)) &&
+	       (b->ppn == (a->ppn + a->size)) &&
+	       (b->flags == a->flags);
+}
+
+static bool pmb_mapping_exists(unsigned long vaddr, phys_addr_t phys,
+			       unsigned long size)
+{
+	int i;
+
+	read_lock(&pmb_rwlock);
+
+	for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
+		struct pmb_entry *pmbe, *iter;
+		unsigned long span;
+
+		if (!test_bit(i, pmb_map))
+			continue;
+
+		pmbe = &pmb_entry_list[i];
+
+		/*
+		 * See if VPN and PPN are bounded by an existing mapping.
+		 */
+		if ((vaddr < pmbe->vpn) || (vaddr >= (pmbe->vpn + pmbe->size)))
+			continue;
+		if ((phys < pmbe->ppn) || (phys >= (pmbe->ppn + pmbe->size)))
+			continue;
+
+		/*
+		 * Now see if we're in range of a simple mapping.
+		 */
+		if (size <= pmbe->size) {
+			read_unlock(&pmb_rwlock);
+			return true;
+		}
 
-repeat:
-	pos = find_first_zero_bit(&pmb_map, NR_PMB_ENTRIES);
+		span = pmbe->size;
 
-	if (unlikely(pos > NR_PMB_ENTRIES))
-		return -ENOSPC;
+		/*
+		 * Finally for sizes that involve compound mappings, walk
+		 * the chain.
+		 */
+		for (iter = pmbe->link; iter; iter = iter->link)
+			span += iter->size;
+
+		/*
+		 * Nothing else to do if the range requirements are met.
+		 */
+		if (size <= span) {
+			read_unlock(&pmb_rwlock);
+			return true;
+		}
+	}
 
-	if (test_and_set_bit(pos, &pmb_map))
-		goto repeat;
+	read_unlock(&pmb_rwlock);
+	return false;
+}
+
+static bool pmb_size_valid(unsigned long size)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(pmb_sizes); i++)
+		if (pmb_sizes[i].size == size)
+			return true;
+
+	return false;
+}
+
+static inline bool pmb_addr_valid(unsigned long addr, unsigned long size)
+{
+	return (addr >= P1SEG && (addr + size - 1) < P3SEG);
+}
+
+static inline bool pmb_prot_valid(pgprot_t prot)
+{
+	return (pgprot_val(prot) & _PAGE_USER) == 0;
+}
+
+static int pmb_size_to_flags(unsigned long size)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(pmb_sizes); i++)
+		if (pmb_sizes[i].size == size)
+			return pmb_sizes[i].flag;
+
+	return 0;
+}
+
+static int pmb_alloc_entry(void)
+{
+	int pos;
+
+	pos = find_first_zero_bit(pmb_map, NR_PMB_ENTRIES);
+	if (pos >= 0 && pos < NR_PMB_ENTRIES)
+		__set_bit(pos, pmb_map);
+	else
+		pos = -ENOSPC;
 
 	return pos;
 }
@@ -73,21 +239,34 @@ static struct pmb_entry *pmb_alloc(unsigned long vpn, unsigned long ppn,
 				   unsigned long flags, int entry)
 {
 	struct pmb_entry *pmbe;
+	unsigned long irqflags;
+	void *ret = NULL;
 	int pos;
 
+	write_lock_irqsave(&pmb_rwlock, irqflags);
+
 	if (entry == PMB_NO_ENTRY) {
 		pos = pmb_alloc_entry();
-		if (pos < 0)
-			return ERR_PTR(pos);
+		if (unlikely(pos < 0)) {
+			ret = ERR_PTR(pos);
+			goto out;
+		}
 	} else {
-		if (test_bit(entry, &pmb_map))
-			return ERR_PTR(-ENOSPC);
+		if (__test_and_set_bit(entry, pmb_map)) {
+			ret = ERR_PTR(-ENOSPC);
+			goto out;
+		}
+
 		pos = entry;
 	}
 
+	write_unlock_irqrestore(&pmb_rwlock, irqflags);
+
 	pmbe = &pmb_entry_list[pos];
-	if (!pmbe)
-		return ERR_PTR(-ENOMEM);
+
+	memset(pmbe, 0, sizeof(struct pmb_entry));
+
+	spin_lock_init(&pmbe->lock);
 
 	pmbe->vpn	= vpn;
 	pmbe->ppn	= ppn;
@@ -95,169 +274,217 @@ static struct pmb_entry *pmb_alloc(unsigned long vpn, unsigned long ppn,
 	pmbe->entry	= pos;
 
 	return pmbe;
+
+out:
+	write_unlock_irqrestore(&pmb_rwlock, irqflags);
+	return ret;
 }
 
 static void pmb_free(struct pmb_entry *pmbe)
 {
-	int pos = pmbe->entry;
+	__clear_bit(pmbe->entry, pmb_map);
 
-	pmbe->vpn	= 0;
-	pmbe->ppn	= 0;
-	pmbe->flags	= 0;
-	pmbe->entry	= 0;
-
-	clear_bit(pos, &pmb_map);
+	pmbe->entry	= PMB_NO_ENTRY;
+	pmbe->link	= NULL;
 }
 
 /*
- * Must be in P2 for __set_pmb_entry()
+ * Must be run uncached.
  */
-static void __set_pmb_entry(unsigned long vpn, unsigned long ppn,
-			    unsigned long flags, int pos)
+static void __set_pmb_entry(struct pmb_entry *pmbe)
 {
-	ctrl_outl(vpn | PMB_V, mk_pmb_addr(pos));
-
-#ifdef CONFIG_CACHE_WRITETHROUGH
-	/*
-	 * When we are in 32-bit address extended mode, CCR.CB becomes
-	 * invalid, so care must be taken to manually adjust cacheable
-	 * translations.
-	 */
-	if (likely(flags & PMB_C))
-		flags |= PMB_WT;
-#endif
+	unsigned long addr, data;
 
-	ctrl_outl(ppn | flags | PMB_V, mk_pmb_data(pos));
-}
+	addr = mk_pmb_addr(pmbe->entry);
+	data = mk_pmb_data(pmbe->entry);
 
-static void __uses_jump_to_uncached set_pmb_entry(struct pmb_entry *pmbe)
-{
 	jump_to_uncached();
-	__set_pmb_entry(pmbe->vpn, pmbe->ppn, pmbe->flags, pmbe->entry);
+
+	/* Set V-bit */
+	__raw_writel(pmbe->vpn | PMB_V, addr);
+	__raw_writel(pmbe->ppn | pmbe->flags | PMB_V, data);
+
 	back_to_cached();
 }
 
-static void __uses_jump_to_uncached clear_pmb_entry(struct pmb_entry *pmbe)
+static void __clear_pmb_entry(struct pmb_entry *pmbe)
 {
-	unsigned int entry = pmbe->entry;
-	unsigned long addr;
+	unsigned long addr, data;
+	unsigned long addr_val, data_val;
 
-	if (unlikely(entry >= NR_PMB_ENTRIES))
-		return;
+	addr = mk_pmb_addr(pmbe->entry);
+	data = mk_pmb_data(pmbe->entry);
 
-	jump_to_uncached();
+	addr_val = __raw_readl(addr);
+	data_val = __raw_readl(data);
 
 	/* Clear V-bit */
-	addr = mk_pmb_addr(entry);
-	ctrl_outl(ctrl_inl(addr) & ~PMB_V, addr);
-
-	addr = mk_pmb_data(entry);
-	ctrl_outl(ctrl_inl(addr) & ~PMB_V, addr);
-
-	back_to_cached();
+	writel_uncached(addr_val & ~PMB_V, addr);
+	writel_uncached(data_val & ~PMB_V, data);
 }
 
+static void set_pmb_entry(struct pmb_entry *pmbe)
+{
+	unsigned long flags;
 
-static struct {
-	unsigned long size;
-	int flag;
-} pmb_sizes[] = {
-	{ .size	= 0x20000000, .flag = PMB_SZ_512M, },
-	{ .size = 0x08000000, .flag = PMB_SZ_128M, },
-	{ .size = 0x04000000, .flag = PMB_SZ_64M,  },
-	{ .size = 0x01000000, .flag = PMB_SZ_16M,  },
-};
+	spin_lock_irqsave(&pmbe->lock, flags);
+	__set_pmb_entry(pmbe);
+	spin_unlock_irqrestore(&pmbe->lock, flags);
+}
 
-long pmb_remap(unsigned long vaddr, unsigned long phys,
-	       unsigned long size, unsigned long flags)
+int pmb_bolt_mapping(unsigned long vaddr, phys_addr_t phys,
+		     unsigned long size, pgprot_t prot)
 {
 	struct pmb_entry *pmbp, *pmbe;
-	unsigned long wanted;
-	int pmb_flags, i;
-	long err;
-
-	/* Convert typical pgprot value to the PMB equivalent */
-	if (flags & _PAGE_CACHABLE) {
-		if (flags & _PAGE_WT)
-			pmb_flags = PMB_WT;
-		else
-			pmb_flags = PMB_C;
-	} else
-		pmb_flags = PMB_WT | PMB_UB;
+	unsigned long orig_addr, orig_size;
+	unsigned long flags, pmb_flags;
+	int i, mapped;
+
+	if (!pmb_addr_valid(vaddr, size))
+		return -EFAULT;
+	if (pmb_mapping_exists(vaddr, phys, size))
+		return 0;
 
+	orig_addr = vaddr;
+	orig_size = size;
+
+	flush_tlb_kernel_range(vaddr, vaddr + size);
+
+	pmb_flags = pgprot_to_pmb_flags(prot);
 	pmbp = NULL;
-	wanted = size;
 
-again:
-	for (i = 0; i < ARRAY_SIZE(pmb_sizes); i++) {
-		if (size < pmb_sizes[i].size)
-			continue;
+	do {
+		for (i = mapped = 0; i < ARRAY_SIZE(pmb_sizes); i++) {
+			if (size < pmb_sizes[i].size)
+				continue;
+
+			pmbe = pmb_alloc(vaddr, phys, pmb_flags |
+					 pmb_sizes[i].flag, PMB_NO_ENTRY);
+			if (IS_ERR(pmbe)) {
+				pmb_unmap_entry(pmbp, mapped);
+				return PTR_ERR(pmbe);
+			}
 
-		pmbe = pmb_alloc(vaddr, phys, pmb_flags | pmb_sizes[i].flag,
-				 PMB_NO_ENTRY);
-		if (IS_ERR(pmbe)) {
-			err = PTR_ERR(pmbe);
-			goto out;
+			spin_lock_irqsave(&pmbe->lock, flags);
+
+			pmbe->size = pmb_sizes[i].size;
+
+			__set_pmb_entry(pmbe);
+
+			phys	+= pmbe->size;
+			vaddr	+= pmbe->size;
+			size	-= pmbe->size;
+
+			/*
+			 * Link adjacent entries that span multiple PMB
+			 * entries for easier tear-down.
+			 */
+			if (likely(pmbp)) {
+				spin_lock(&pmbp->lock);
+				pmbp->link = pmbe;
+				spin_unlock(&pmbp->lock);
+			}
+
+			pmbp = pmbe;
+
+			/*
+			 * Instead of trying smaller sizes on every
+			 * iteration (even if we succeed in allocating
+			 * space), try using pmb_sizes[i].size again.
+			 */
+			i--;
+			mapped++;
+
+			spin_unlock_irqrestore(&pmbe->lock, flags);
 		}
+	} while (size >= SZ_16M);
 
-		set_pmb_entry(pmbe);
+	flush_cache_vmap(orig_addr, orig_addr + orig_size);
 
-		phys	+= pmb_sizes[i].size;
-		vaddr	+= pmb_sizes[i].size;
-		size	-= pmb_sizes[i].size;
+	return 0;
+}
 
-		/*
-		 * Link adjacent entries that span multiple PMB entries
-		 * for easier tear-down.
-		 */
-		if (likely(pmbp))
-			pmbp->link = pmbe;
+void __iomem *pmb_remap_caller(phys_addr_t phys, unsigned long size,
+			       pgprot_t prot, void *caller)
+{
+	unsigned long vaddr;
+	phys_addr_t offset, last_addr;
+	phys_addr_t align_mask;
+	unsigned long aligned;
+	struct vm_struct *area;
+	int i, ret;
 
-		pmbp = pmbe;
+	if (!pmb_iomapping_enabled)
+		return NULL;
 
-		/*
-		 * Instead of trying smaller sizes on every iteration
-		 * (even if we succeed in allocating space), try using
-		 * pmb_sizes[i].size again.
-		 */
-		i--;
-	}
+	/*
+	 * Small mappings need to go through the TLB.
+	 */
+	if (size < SZ_16M)
+		return ERR_PTR(-EINVAL);
+	if (!pmb_prot_valid(prot))
+		return ERR_PTR(-EINVAL);
 
-	if (size >= 0x1000000)
-		goto again;
+	for (i = 0; i < ARRAY_SIZE(pmb_sizes); i++)
+		if (size >= pmb_sizes[i].size)
+			break;
 
-	return wanted - size;
+	last_addr = phys + size;
+	align_mask = ~(pmb_sizes[i].size - 1);
+	offset = phys & ~align_mask;
+	phys &= align_mask;
+	aligned = ALIGN(last_addr, pmb_sizes[i].size) - phys;
 
-out:
-	if (pmbp)
-		__pmb_unmap(pmbp);
+	/*
+	 * XXX: This should really start from uncached_end, but this
+	 * causes the MMU to reset, so for now we restrict it to the
+	 * 0xb000...0xc000 range.
+	 */
+	area = __get_vm_area_caller(aligned, VM_IOREMAP, 0xb0000000,
+				    P3SEG, caller);
+	if (!area)
+		return NULL;
+
+	area->phys_addr = phys;
+	vaddr = (unsigned long)area->addr;
 
-	return err;
+	ret = pmb_bolt_mapping(vaddr, phys, size, prot);
+	if (unlikely(ret != 0))
+		return ERR_PTR(ret);
+
+	return (void __iomem *)(offset + (char *)vaddr);
 }
 
-void pmb_unmap(unsigned long addr)
+int pmb_unmap(void __iomem *addr)
 {
 	struct pmb_entry *pmbe = NULL;
-	int i;
+	unsigned long vaddr = (unsigned long __force)addr;
+	int i, found = 0;
+
+	read_lock(&pmb_rwlock);
 
 	for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
-		if (test_bit(i, &pmb_map)) {
+		if (test_bit(i, pmb_map)) {
 			pmbe = &pmb_entry_list[i];
-			if (pmbe->vpn == addr)
+			if (pmbe->vpn == vaddr) {
+				found = 1;
 				break;
+			}
 		}
 	}
 
-	if (unlikely(!pmbe))
-		return;
+	read_unlock(&pmb_rwlock);
 
-	__pmb_unmap(pmbe);
+	if (found) {
+		pmb_unmap_entry(pmbe, NR_PMB_ENTRIES);
+		return 0;
+	}
+
+	return -EINVAL;
 }
 
-static void __pmb_unmap(struct pmb_entry *pmbe)
+static void __pmb_unmap_entry(struct pmb_entry *pmbe, int depth)
 {
-	BUG_ON(!test_bit(pmbe->entry, &pmb_map));
-
 	do {
 		struct pmb_entry *pmblink = pmbe;
 
@@ -268,102 +495,321 @@ static void __pmb_unmap(struct pmb_entry *pmbe)
 		 * this entry in pmb_alloc() (even if we haven't filled
 		 * it yet).
 		 *
-		 * Therefore, calling clear_pmb_entry() is safe as no
+		 * Therefore, calling __clear_pmb_entry() is safe as no
 		 * other mapping can be using that slot.
 		 */
-		clear_pmb_entry(pmbe);
+		__clear_pmb_entry(pmbe);
+
+		flush_cache_vunmap(pmbe->vpn, pmbe->vpn + pmbe->size);
 
 		pmbe = pmblink->link;
 
 		pmb_free(pmblink);
-	} while (pmbe);
+	} while (pmbe && --depth);
 }
 
-#ifdef CONFIG_PMB
-int __uses_jump_to_uncached pmb_init(void)
+static void pmb_unmap_entry(struct pmb_entry *pmbe, int depth)
 {
-	unsigned int i;
-	long size, ret;
+	unsigned long flags;
 
-	jump_to_uncached();
+	if (unlikely(!pmbe))
+		return;
+
+	write_lock_irqsave(&pmb_rwlock, flags);
+	__pmb_unmap_entry(pmbe, depth);
+	write_unlock_irqrestore(&pmb_rwlock, flags);
+}
+
+static void __init pmb_notify(void)
+{
+	int i;
+
+	pr_info("PMB: boot mappings:\n");
+
+	read_lock(&pmb_rwlock);
+
+	for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
+		struct pmb_entry *pmbe;
+
+		if (!test_bit(i, pmb_map))
+			continue;
+
+		pmbe = &pmb_entry_list[i];
+
+		pr_info("       0x%08lx -> 0x%08lx [ %4ldMB %2scached ]\n",
+			pmbe->vpn >> PAGE_SHIFT, pmbe->ppn >> PAGE_SHIFT,
+			pmbe->size >> 20, (pmbe->flags & PMB_C) ? "" : "un");
+	}
+
+	read_unlock(&pmb_rwlock);
+}
+
+/*
+ * Sync our software copy of the PMB mappings with those in hardware. The
+ * mappings in the hardware PMB were either set up by the bootloader or
+ * very early on by the kernel.
+ */
+static void __init pmb_synchronize(void)
+{
+	struct pmb_entry *pmbp = NULL;
+	int i, j;
 
 	/*
-	 * Insert PMB entries for the P1 and P2 areas so that, after
-	 * we've switched the MMU to 32-bit mode, the semantics of P1
-	 * and P2 are the same as in 29-bit mode, e.g.
+	 * Run through the initial boot mappings, log the established
+	 * ones, and blow away anything that falls outside of the valid
+	 * PPN range. Specifically, we only care about existing mappings
+	 * that impact the cached/uncached sections.
+	 *
+	 * Note that touching these can be a bit of a minefield; the boot
+	 * loader can establish multi-page mappings with the same caching
+	 * attributes, so we need to ensure that we aren't modifying a
+	 * mapping that we're presently executing from, or may execute
+	 * from in the case of straddling page boundaries.
 	 *
-	 *	P1 - provides a cached window onto physical memory
-	 *	P2 - provides an uncached window onto physical memory
+	 * In the future we will have to tidy up after the boot loader by
+	 * jumping between the cached and uncached mappings and tearing
+	 * down alternating mappings while executing from the other.
 	 */
-	size = __MEMORY_START + __MEMORY_SIZE;
+	for (i = 0; i < NR_PMB_ENTRIES; i++) {
+		unsigned long addr, data;
+		unsigned long addr_val, data_val;
+		unsigned long ppn, vpn, flags;
+		unsigned long irqflags;
+		unsigned int size;
+		struct pmb_entry *pmbe;
 
-	ret = pmb_remap(P1SEG, 0x00000000, size, PMB_C);
-	BUG_ON(ret != size);
+		addr = mk_pmb_addr(i);
+		data = mk_pmb_data(i);
 
-	ret = pmb_remap(P2SEG, 0x00000000, size, PMB_WT | PMB_UB);
-	BUG_ON(ret != size);
+		addr_val = __raw_readl(addr);
+		data_val = __raw_readl(data);
 
-	ctrl_outl(0, PMB_IRMCR);
+		/*
+		 * Skip over any bogus entries
+		 */
+		if (!(data_val & PMB_V) || !(addr_val & PMB_V))
+			continue;
 
-	/* PMB.SE and UB[7] */
-	ctrl_outl(PASCR_SE | (1 << 7), PMB_PASCR);
+		ppn = data_val & PMB_PFN_MASK;
+		vpn = addr_val & PMB_PFN_MASK;
 
-	/* Flush out the TLB */
-	i =  ctrl_inl(MMUCR);
-	i |= MMUCR_TI;
-	ctrl_outl(i, MMUCR);
+		/*
+		 * Only preserve in-range mappings.
+		 */
+		if (!pmb_ppn_in_range(ppn)) {
+			/*
+			 * Invalidate anything out of bounds.
+			 */
+			writel_uncached(addr_val & ~PMB_V, addr);
+			writel_uncached(data_val & ~PMB_V, data);
+			continue;
+		}
 
-	back_to_cached();
+		/*
+		 * Update the caching attributes if necessary
+		 */
+		if (data_val & PMB_C) {
+			data_val &= ~PMB_CACHE_MASK;
+			data_val |= pmb_cache_flags();
 
-	return 0;
+			writel_uncached(data_val, data);
+		}
+
+		size = data_val & PMB_SZ_MASK;
+		flags = size | (data_val & PMB_CACHE_MASK);
+
+		pmbe = pmb_alloc(vpn, ppn, flags, i);
+		if (IS_ERR(pmbe)) {
+			WARN_ON_ONCE(1);
+			continue;
+		}
+
+		spin_lock_irqsave(&pmbe->lock, irqflags);
+
+		for (j = 0; j < ARRAY_SIZE(pmb_sizes); j++)
+			if (pmb_sizes[j].flag == size)
+				pmbe->size = pmb_sizes[j].size;
+
+		if (pmbp) {
+			spin_lock(&pmbp->lock);
+
+			/*
+			 * Compare the previous entry against the current one to
+			 * see if the entries span a contiguous mapping. If so,
+			 * setup the entry links accordingly. Compound mappings
+			 * are later coalesced.
+			 */
+			if (pmb_can_merge(pmbp, pmbe))
+				pmbp->link = pmbe;
+
+			spin_unlock(&pmbp->lock);
+		}
+
+		pmbp = pmbe;
+
+		spin_unlock_irqrestore(&pmbe->lock, irqflags);
+	}
 }
-#else
-int __uses_jump_to_uncached pmb_init(void)
+
+static void __init pmb_merge(struct pmb_entry *head)
 {
+	unsigned long span, newsize;
+	struct pmb_entry *tail;
+	int i = 1, depth = 0;
+
+	span = newsize = head->size;
+
+	tail = head->link;
+	while (tail) {
+		span += tail->size;
+
+		if (pmb_size_valid(span)) {
+			newsize = span;
+			depth = i;
+		}
+
+		/* This is the end of the line.. */
+		if (!tail->link)
+			break;
+
+		tail = tail->link;
+		i++;
+	}
+
+	/*
+	 * The merged page size must be valid.
+	 */
+	if (!pmb_size_valid(newsize))
+		return;
+
+	head->flags &= ~PMB_SZ_MASK;
+	head->flags |= pmb_size_to_flags(newsize);
+
+	head->size = newsize;
+
+	__pmb_unmap_entry(head->link, depth);
+	__set_pmb_entry(head);
+}
+
+static void __init pmb_coalesce(void)
+{
+	unsigned long flags;
 	int i;
-	unsigned long addr, data;
 
-	jump_to_uncached();
+	write_lock_irqsave(&pmb_rwlock, flags);
 
-	for (i = 0; i < PMB_ENTRY_MAX; i++) {
+	for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
 		struct pmb_entry *pmbe;
-		unsigned long vpn, ppn, flags;
 
-		addr = PMB_DATA + (i << PMB_E_SHIFT);
-		data = ctrl_inl(addr);
-		if (!(data & PMB_V))
+		if (!test_bit(i, pmb_map))
 			continue;
 
-		if (data & PMB_C) {
-#if defined(CONFIG_CACHE_WRITETHROUGH)
-			data |= PMB_WT;
-#elif defined(CONFIG_CACHE_WRITEBACK)
-			data &= ~PMB_WT;
-#else
-			data &= ~(PMB_C | PMB_WT);
-#endif
-		}
-		ctrl_outl(data, addr);
+		pmbe = &pmb_entry_list[i];
 
-		ppn = data & PMB_PFN_MASK;
+		/*
+		 * We're only interested in compound mappings
+		 */
+		if (!pmbe->link)
+			continue;
 
-		flags = data & (PMB_C | PMB_WT | PMB_UB);
-		flags |= data & PMB_SZ_MASK;
+		/*
+		 * Nothing to do if it already uses the largest possible
+		 * page size.
+		 */
+		if (pmbe->size == SZ_512M)
+			continue;
 
-		addr = PMB_ADDR + (i << PMB_E_SHIFT);
-		data = ctrl_inl(addr);
+		pmb_merge(pmbe);
+	}
 
-		vpn = data & PMB_PFN_MASK;
+	write_unlock_irqrestore(&pmb_rwlock, flags);
+}
 
-		pmbe = pmb_alloc(vpn, ppn, flags, i);
-		WARN_ON(IS_ERR(pmbe));
+#ifdef CONFIG_UNCACHED_MAPPING
+static void __init pmb_resize(void)
+{
+	int i;
+
+	/*
+	 * If the uncached mapping was constructed by the kernel, it will
+	 * already be a reasonable size.
+	 */
+	if (uncached_size == SZ_16M)
+		return;
+
+	read_lock(&pmb_rwlock);
+
+	for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
+		struct pmb_entry *pmbe;
+		unsigned long flags;
+
+		if (!test_bit(i, pmb_map))
+			continue;
+
+		pmbe = &pmb_entry_list[i];
+
+		if (pmbe->vpn != uncached_start)
+			continue;
+
+		/*
+		 * Found it, now resize it.
+		 */
+		spin_lock_irqsave(&pmbe->lock, flags);
+
+		pmbe->size = SZ_16M;
+		pmbe->flags &= ~PMB_SZ_MASK;
+		pmbe->flags |= pmb_size_to_flags(pmbe->size);
+
+		uncached_resize(pmbe->size);
+
+		__set_pmb_entry(pmbe);
+
+		spin_unlock_irqrestore(&pmbe->lock, flags);
 	}
 
-	back_to_cached();
+	read_lock(&pmb_rwlock);
+}
+#endif
+
+static int __init early_pmb(char *p)
+{
+	if (!p)
+		return 0;
+
+	if (strstr(p, "iomap"))
+		pmb_iomapping_enabled = 1;
 
 	return 0;
 }
-#endif /* CONFIG_PMB */
+early_param("pmb", early_pmb);
+
+void __init pmb_init(void)
+{
+	/* Synchronize software state */
+	pmb_synchronize();
+
+	/* Attempt to combine compound mappings */
+	pmb_coalesce();
+
+#ifdef CONFIG_UNCACHED_MAPPING
+	/* Resize initial mappings, if necessary */
+	pmb_resize();
+#endif
+
+	/* Log them */
+	pmb_notify();
+
+	writel_uncached(0, PMB_IRMCR);
+
+	/* Flush out the TLB */
+	__raw_writel(__raw_readl(MMUCR) | MMUCR_TI, MMUCR);
+	ctrl_barrier();
+}
+
+bool __in_29bit_mode(void)
+{
+        return (__raw_readl(PMB_PASCR) & PASCR_SE) == 0;
+}
 
 static int pmb_seq_show(struct seq_file *file, void *iter)
 {
@@ -378,8 +824,8 @@ static int pmb_seq_show(struct seq_file *file, void *iter)
 		unsigned int size;
 		char *sz_str = NULL;
 
-		addr = ctrl_inl(mk_pmb_addr(i));
-		data = ctrl_inl(mk_pmb_data(i));
+		addr = __raw_readl(mk_pmb_addr(i));
+		data = __raw_readl(mk_pmb_data(i));
 
 		size = data & PMB_SZ_MASK;
 		sz_str = (size == PMB_SZ_16M)  ? " 16MB":
@@ -425,7 +871,7 @@ static int __init pmb_debugfs_init(void)
 
 	return 0;
 }
-postcore_initcall(pmb_debugfs_init);
+subsys_initcall(pmb_debugfs_init);
 
 #ifdef CONFIG_PM
 static int pmb_sysdev_suspend(struct sys_device *dev, pm_message_t state)
@@ -437,14 +883,21 @@ static int pmb_sysdev_suspend(struct sys_device *dev, pm_message_t state)
 	if (state.event == PM_EVENT_ON &&
 	    prev_state.event == PM_EVENT_FREEZE) {
 		struct pmb_entry *pmbe;
+
+		read_lock(&pmb_rwlock);
+
 		for (i = 0; i < ARRAY_SIZE(pmb_entry_list); i++) {
-			if (test_bit(i, &pmb_map)) {
+			if (test_bit(i, pmb_map)) {
 				pmbe = &pmb_entry_list[i];
 				set_pmb_entry(pmbe);
 			}
 		}
+
+		read_unlock(&pmb_rwlock);
 	}
+
 	prev_state = state;
+
 	return 0;
 }
 
@@ -462,6 +915,5 @@ static int __init pmb_sysdev_init(void)
 {
 	return sysdev_driver_register(&cpu_sysdev_class, &pmb_sysdev_driver);
 }
-
 subsys_initcall(pmb_sysdev_init);
 #endif