Update to 3.4-final.
[linux-flexiantxendom0-3.2.10.git] / arch / arm / mm / mmu.c
index e534801..aa78de8 100644 (file)
 #include <linux/kernel.h>
 #include <linux/errno.h>
 #include <linux/init.h>
-#include <linux/bootmem.h>
 #include <linux/mman.h>
 #include <linux/nodemask.h>
-#include <linux/sort.h>
+#include <linux/memblock.h>
+#include <linux/fs.h>
+#include <linux/vmalloc.h>
 
+#include <asm/cp15.h>
 #include <asm/cputype.h>
-#include <asm/mach-types.h>
 #include <asm/sections.h>
 #include <asm/cachetype.h>
 #include <asm/setup.h>
 #include <asm/smp_plat.h>
 #include <asm/tlb.h>
 #include <asm/highmem.h>
+#include <asm/system_info.h>
+#include <asm/traps.h>
 
 #include <asm/mach/arch.h>
 #include <asm/mach/map.h>
 
 #include "mm.h"
 
-DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
-
 /*
  * empty_zero_page is a special page that is used for
  * zero-initialized data and COW.
@@ -62,8 +63,8 @@ EXPORT_SYMBOL(pgprot_kernel);
 struct cachepolicy {
        const char      policy[16];
        unsigned int    cr_mask;
-       unsigned int    pmd;
-       unsigned int    pte;
+       pmdval_t        pmd;
+       pteval_t        pte;
 };
 
 static struct cachepolicy cache_policies[] __initdata = {
@@ -152,6 +153,7 @@ static int __init early_nowrite(char *__unused)
 }
 early_param("nowb", early_nowrite);
 
+#ifndef CONFIG_ARM_LPAE
 static int __init early_ecc(char *p)
 {
        if (memcmp(p, "on", 2) == 0)
@@ -161,6 +163,7 @@ static int __init early_ecc(char *p)
        return 0;
 }
 early_param("ecc", early_ecc);
+#endif
 
 static int __init noalign_setup(char *__unused)
 {
@@ -191,7 +194,7 @@ void adjust_cr(unsigned long mask, unsigned long set)
 }
 #endif
 
-#define PROT_PTE_DEVICE                L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_WRITE
+#define PROT_PTE_DEVICE                L_PTE_PRESENT|L_PTE_YOUNG|L_PTE_DIRTY|L_PTE_XN
 #define PROT_SECT_DEVICE       PMD_TYPE_SECT|PMD_SECT_AP_WRITE
 
 static struct mem_type mem_types[] = {
@@ -230,23 +233,27 @@ static struct mem_type mem_types[] = {
                .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
                .domain    = DOMAIN_KERNEL,
        },
+#ifndef CONFIG_ARM_LPAE
        [MT_MINICLEAN] = {
                .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN | PMD_SECT_MINICACHE,
                .domain    = DOMAIN_KERNEL,
        },
+#endif
        [MT_LOW_VECTORS] = {
                .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
-                               L_PTE_EXEC,
+                               L_PTE_RDONLY,
                .prot_l1   = PMD_TYPE_TABLE,
                .domain    = DOMAIN_USER,
        },
        [MT_HIGH_VECTORS] = {
                .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
-                               L_PTE_USER | L_PTE_EXEC,
+                               L_PTE_USER | L_PTE_RDONLY,
                .prot_l1   = PMD_TYPE_TABLE,
                .domain    = DOMAIN_USER,
        },
        [MT_MEMORY] = {
+               .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY,
+               .prot_l1   = PMD_TYPE_TABLE,
                .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
                .domain    = DOMAIN_KERNEL,
        },
@@ -255,21 +262,31 @@ static struct mem_type mem_types[] = {
                .domain    = DOMAIN_KERNEL,
        },
        [MT_MEMORY_NONCACHED] = {
+               .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
+                               L_PTE_MT_BUFFERABLE,
+               .prot_l1   = PMD_TYPE_TABLE,
                .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE,
                .domain    = DOMAIN_KERNEL,
        },
        [MT_MEMORY_DTCM] = {
-               .prot_pte       = L_PTE_PRESENT | L_PTE_YOUNG |
-                                 L_PTE_DIRTY | L_PTE_WRITE,
-               .prot_l1        = PMD_TYPE_TABLE,
-               .prot_sect      = PMD_TYPE_SECT | PMD_SECT_XN,
-               .domain         = DOMAIN_KERNEL,
+               .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
+                               L_PTE_XN,
+               .prot_l1   = PMD_TYPE_TABLE,
+               .prot_sect = PMD_TYPE_SECT | PMD_SECT_XN,
+               .domain    = DOMAIN_KERNEL,
        },
        [MT_MEMORY_ITCM] = {
+               .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY,
+               .prot_l1   = PMD_TYPE_TABLE,
+               .domain    = DOMAIN_KERNEL,
+       },
+       [MT_MEMORY_SO] = {
                .prot_pte  = L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_DIRTY |
-                               L_PTE_USER | L_PTE_EXEC,
+                               L_PTE_MT_UNCACHED,
                .prot_l1   = PMD_TYPE_TABLE,
-               .domain    = DOMAIN_IO,
+               .prot_sect = PMD_TYPE_SECT | PMD_SECT_AP_WRITE | PMD_SECT_S |
+                               PMD_SECT_UNCACHED | PMD_SECT_XN,
+               .domain    = DOMAIN_KERNEL,
        },
 };
 
@@ -286,7 +303,7 @@ static void __init build_mem_type_table(void)
 {
        struct cachepolicy *cp;
        unsigned int cr = get_cr();
-       unsigned int user_pgprot, kern_pgprot, vecs_pgprot;
+       pteval_t user_pgprot, kern_pgprot, vecs_pgprot;
        int cpu_arch = cpu_architecture();
        int i;
 
@@ -304,9 +321,8 @@ static void __init build_mem_type_table(void)
                        cachepolicy = CPOLICY_WRITEBACK;
                ecc_mask = 0;
        }
-#ifdef CONFIG_SMP
-       cachepolicy = CPOLICY_WRITEALLOC;
-#endif
+       if (is_smp())
+               cachepolicy = CPOLICY_WRITEALLOC;
 
        /*
         * Strip out features not present on earlier architectures.
@@ -400,25 +416,27 @@ static void __init build_mem_type_table(void)
        cp = &cache_policies[cachepolicy];
        vecs_pgprot = kern_pgprot = user_pgprot = cp->pte;
 
-#ifndef CONFIG_SMP
        /*
         * Only use write-through for non-SMP systems
         */
-       if (cpu_arch >= CPU_ARCH_ARMv5 && cachepolicy > CPOLICY_WRITETHROUGH)
+       if (!is_smp() && cpu_arch >= CPU_ARCH_ARMv5 && cachepolicy > CPOLICY_WRITETHROUGH)
                vecs_pgprot = cache_policies[CPOLICY_WRITETHROUGH].pte;
-#endif
 
        /*
         * Enable CPU-specific coherency if supported.
         * (Only available on XSC3 at the moment.)
         */
-       if (arch_is_coherent() && cpu_is_xsc3())
+       if (arch_is_coherent() && cpu_is_xsc3()) {
                mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
-
+               mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
+               mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_S;
+               mem_types[MT_MEMORY_NONCACHED].prot_pte |= L_PTE_SHARED;
+       }
        /*
         * ARMv6 and above have extended page tables.
         */
        if (cpu_arch >= CPU_ARCH_ARMv6 && (cr & CR_XP)) {
+#ifndef CONFIG_ARM_LPAE
                /*
                 * Mark cache clean areas and XIP ROM read only
                 * from SVC mode and no access from userspace.
@@ -426,21 +444,25 @@ static void __init build_mem_type_table(void)
                mem_types[MT_ROM].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
                mem_types[MT_MINICLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
                mem_types[MT_CACHECLEAN].prot_sect |= PMD_SECT_APX|PMD_SECT_AP_WRITE;
-
-#ifdef CONFIG_SMP
-               /*
-                * Mark memory with the "shared" attribute for SMP systems
-                */
-               user_pgprot |= L_PTE_SHARED;
-               kern_pgprot |= L_PTE_SHARED;
-               vecs_pgprot |= L_PTE_SHARED;
-               mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_S;
-               mem_types[MT_DEVICE_WC].prot_pte |= L_PTE_SHARED;
-               mem_types[MT_DEVICE_CACHED].prot_sect |= PMD_SECT_S;
-               mem_types[MT_DEVICE_CACHED].prot_pte |= L_PTE_SHARED;
-               mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
-               mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_S;
 #endif
+
+               if (is_smp()) {
+                       /*
+                        * Mark memory with the "shared" attribute
+                        * for SMP systems
+                        */
+                       user_pgprot |= L_PTE_SHARED;
+                       kern_pgprot |= L_PTE_SHARED;
+                       vecs_pgprot |= L_PTE_SHARED;
+                       mem_types[MT_DEVICE_WC].prot_sect |= PMD_SECT_S;
+                       mem_types[MT_DEVICE_WC].prot_pte |= L_PTE_SHARED;
+                       mem_types[MT_DEVICE_CACHED].prot_sect |= PMD_SECT_S;
+                       mem_types[MT_DEVICE_CACHED].prot_pte |= L_PTE_SHARED;
+                       mem_types[MT_MEMORY].prot_sect |= PMD_SECT_S;
+                       mem_types[MT_MEMORY].prot_pte |= L_PTE_SHARED;
+                       mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_S;
+                       mem_types[MT_MEMORY_NONCACHED].prot_pte |= L_PTE_SHARED;
+               }
        }
 
        /*
@@ -461,6 +483,19 @@ static void __init build_mem_type_table(void)
                mem_types[MT_MEMORY_NONCACHED].prot_sect |= PMD_SECT_BUFFERABLE;
        }
 
+#ifdef CONFIG_ARM_LPAE
+       /*
+        * Do not generate access flag faults for the kernel mappings.
+        */
+       for (i = 0; i < ARRAY_SIZE(mem_types); i++) {
+               mem_types[i].prot_pte |= PTE_EXT_AF;
+               if (mem_types[i].prot_sect)
+                       mem_types[i].prot_sect |= PMD_SECT_AF;
+       }
+       kern_pgprot |= PTE_EXT_AF;
+       vecs_pgprot |= PTE_EXT_AF;
+#endif
+
        for (i = 0; i < 16; i++) {
                unsigned long v = pgprot_val(protection_map[i]);
                protection_map[i] = __pgprot(v | user_pgprot);
@@ -471,11 +506,13 @@ static void __init build_mem_type_table(void)
 
        pgprot_user   = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG | user_pgprot);
        pgprot_kernel = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
-                                L_PTE_DIRTY | L_PTE_WRITE | kern_pgprot);
+                                L_PTE_DIRTY | kern_pgprot);
 
        mem_types[MT_LOW_VECTORS].prot_l1 |= ecc_mask;
        mem_types[MT_HIGH_VECTORS].prot_l1 |= ecc_mask;
        mem_types[MT_MEMORY].prot_sect |= ecc_mask | cp->pmd;
+       mem_types[MT_MEMORY].prot_pte |= kern_pgprot;
+       mem_types[MT_MEMORY_NONCACHED].prot_sect |= ecc_mask;
        mem_types[MT_ROM].prot_sect |= cp->pmd;
 
        switch (cp->pmd) {
@@ -499,31 +536,59 @@ static void __init build_mem_type_table(void)
        }
 }
 
+#ifdef CONFIG_ARM_DMA_MEM_BUFFERABLE
+pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
+                             unsigned long size, pgprot_t vma_prot)
+{
+       if (!pfn_valid(pfn))
+               return pgprot_noncached(vma_prot);
+       else if (file->f_flags & O_SYNC)
+               return pgprot_writecombine(vma_prot);
+       return vma_prot;
+}
+EXPORT_SYMBOL(phys_mem_access_prot);
+#endif
+
 #define vectors_base() (vectors_high() ? 0xffff0000 : 0)
 
-static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
-                                 unsigned long end, unsigned long pfn,
-                                 const struct mem_type *type)
+static void __init *early_alloc_aligned(unsigned long sz, unsigned long align)
+{
+       void *ptr = __va(memblock_alloc(sz, align));
+       memset(ptr, 0, sz);
+       return ptr;
+}
+
+static void __init *early_alloc(unsigned long sz)
 {
-       pte_t *pte;
+       return early_alloc_aligned(sz, sz);
+}
 
+static pte_t * __init early_pte_alloc(pmd_t *pmd, unsigned long addr, unsigned long prot)
+{
        if (pmd_none(*pmd)) {
-               pte = alloc_bootmem_low_pages(2 * PTRS_PER_PTE * sizeof(pte_t));
-               __pmd_populate(pmd, __pa(pte) | type->prot_l1);
+               pte_t *pte = early_alloc(PTE_HWTABLE_OFF + PTE_HWTABLE_SIZE);
+               __pmd_populate(pmd, __pa(pte), prot);
        }
+       BUG_ON(pmd_bad(*pmd));
+       return pte_offset_kernel(pmd, addr);
+}
 
-       pte = pte_offset_kernel(pmd, addr);
+static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr,
+                                 unsigned long end, unsigned long pfn,
+                                 const struct mem_type *type)
+{
+       pte_t *pte = early_pte_alloc(pmd, addr, type->prot_l1);
        do {
                set_pte_ext(pte, pfn_pte(pfn, __pgprot(type->prot_pte)), 0);
                pfn++;
        } while (pte++, addr += PAGE_SIZE, addr != end);
 }
 
-static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
-                                     unsigned long end, unsigned long phys,
+static void __init alloc_init_section(pud_t *pud, unsigned long addr,
+                                     unsigned long end, phys_addr_t phys,
                                      const struct mem_type *type)
 {
-       pmd_t *pmd = pmd_offset(pgd, addr);
+       pmd_t *pmd = pmd_offset(pud, addr);
 
        /*
         * Try a section mapping - end, addr and phys must all be aligned
@@ -534,8 +599,10 @@ static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
        if (((addr | end | phys) & ~SECTION_MASK) == 0) {
                pmd_t *p = pmd;
 
+#ifndef CONFIG_ARM_LPAE
                if (addr & SECTION_SIZE)
                        pmd++;
+#endif
 
                do {
                        *pmd = __pmd(phys | type->prot_sect);
@@ -552,20 +619,35 @@ static void __init alloc_init_section(pgd_t *pgd, unsigned long addr,
        }
 }
 
+static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
+       unsigned long end, unsigned long phys, const struct mem_type *type)
+{
+       pud_t *pud = pud_offset(pgd, addr);
+       unsigned long next;
+
+       do {
+               next = pud_addr_end(addr, end);
+               alloc_init_section(pud, addr, next, phys, type);
+               phys += next - addr;
+       } while (pud++, addr = next, addr != end);
+}
+
+#ifndef CONFIG_ARM_LPAE
 static void __init create_36bit_mapping(struct map_desc *md,
                                        const struct mem_type *type)
 {
-       unsigned long phys, addr, length, end;
+       unsigned long addr, length, end;
+       phys_addr_t phys;
        pgd_t *pgd;
 
        addr = md->virtual;
-       phys = (unsigned long)__pfn_to_phys(md->pfn);
+       phys = __pfn_to_phys(md->pfn);
        length = PAGE_ALIGN(md->length);
 
        if (!(cpu_architecture() >= CPU_ARCH_ARMv6 || cpu_is_xsc3())) {
                printk(KERN_ERR "MM: CPU does not support supersection "
                       "mapping for 0x%08llx at 0x%08lx\n",
-                      __pfn_to_phys((u64)md->pfn), addr);
+                      (long long)__pfn_to_phys((u64)md->pfn), addr);
                return;
        }
 
@@ -578,14 +660,14 @@ static void __init create_36bit_mapping(struct map_desc *md,
        if (type->domain) {
                printk(KERN_ERR "MM: invalid domain in supersection "
                       "mapping for 0x%08llx at 0x%08lx\n",
-                      __pfn_to_phys((u64)md->pfn), addr);
+                      (long long)__pfn_to_phys((u64)md->pfn), addr);
                return;
        }
 
        if ((addr | length | __pfn_to_phys(md->pfn)) & ~SUPERSECTION_MASK) {
-               printk(KERN_ERR "MM: cannot create mapping for "
-                      "0x%08llx at 0x%08lx invalid alignment\n",
-                      __pfn_to_phys((u64)md->pfn), addr);
+               printk(KERN_ERR "MM: cannot create mapping for 0x%08llx"
+                      " at 0x%08lx invalid alignment\n",
+                      (long long)__pfn_to_phys((u64)md->pfn), addr);
                return;
        }
 
@@ -598,7 +680,8 @@ static void __init create_36bit_mapping(struct map_desc *md,
        pgd = pgd_offset_k(addr);
        end = addr + length;
        do {
-               pmd_t *pmd = pmd_offset(pgd, addr);
+               pud_t *pud = pud_offset(pgd, addr);
+               pmd_t *pmd = pmd_offset(pud, addr);
                int i;
 
                for (i = 0; i < 16; i++)
@@ -609,6 +692,7 @@ static void __init create_36bit_mapping(struct map_desc *md,
                pgd += SUPERSECTION_SIZE >> PGDIR_SHIFT;
        } while (addr != end);
 }
+#endif /* !CONFIG_ARM_LPAE */
 
 /*
  * Create the page directory entries and any necessary
@@ -619,26 +703,29 @@ static void __init create_36bit_mapping(struct map_desc *md,
  */
 static void __init create_mapping(struct map_desc *md)
 {
-       unsigned long phys, addr, length, end;
+       unsigned long addr, length, end;
+       phys_addr_t phys;
        const struct mem_type *type;
        pgd_t *pgd;
 
        if (md->virtual != vectors_base() && md->virtual < TASK_SIZE) {
-               printk(KERN_WARNING "BUG: not creating mapping for "
-                      "0x%08llx at 0x%08lx in user region\n",
-                      __pfn_to_phys((u64)md->pfn), md->virtual);
+               printk(KERN_WARNING "BUG: not creating mapping for 0x%08llx"
+                      " at 0x%08lx in user region\n",
+                      (long long)__pfn_to_phys((u64)md->pfn), md->virtual);
                return;
        }
 
        if ((md->type == MT_DEVICE || md->type == MT_ROM) &&
-           md->virtual >= PAGE_OFFSET && md->virtual < VMALLOC_END) {
-               printk(KERN_WARNING "BUG: mapping for 0x%08llx at 0x%08lx "
-                      "overlaps vmalloc space\n",
-                      __pfn_to_phys((u64)md->pfn), md->virtual);
+           md->virtual >= PAGE_OFFSET &&
+           (md->virtual < VMALLOC_START || md->virtual >= VMALLOC_END)) {
+               printk(KERN_WARNING "BUG: mapping for 0x%08llx"
+                      " at 0x%08lx out of vmalloc space\n",
+                      (long long)__pfn_to_phys((u64)md->pfn), md->virtual);
        }
 
        type = &mem_types[md->type];
 
+#ifndef CONFIG_ARM_LPAE
        /*
         * Catch 36-bit addresses
         */
@@ -646,15 +733,16 @@ static void __init create_mapping(struct map_desc *md)
                create_36bit_mapping(md, type);
                return;
        }
+#endif
 
        addr = md->virtual & PAGE_MASK;
-       phys = (unsigned long)__pfn_to_phys(md->pfn);
+       phys = __pfn_to_phys(md->pfn);
        length = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
 
        if (type->prot_l1 == 0 && ((addr | phys | length) & ~SECTION_MASK)) {
-               printk(KERN_WARNING "BUG: map for 0x%08lx at 0x%08lx can not "
+               printk(KERN_WARNING "BUG: map for 0x%08llx at 0x%08lx can not "
                       "be mapped using pages, ignoring.\n",
-                      __pfn_to_phys(md->pfn), addr);
+                      (long long)__pfn_to_phys(md->pfn), addr);
                return;
        }
 
@@ -663,7 +751,7 @@ static void __init create_mapping(struct map_desc *md)
        do {
                unsigned long next = pgd_addr_end(addr, end);
 
-               alloc_init_section(pgd, addr, next, phys, type);
+               alloc_init_pud(pgd, addr, next, phys, type);
 
                phys += next - addr;
                addr = next;
@@ -675,22 +763,37 @@ static void __init create_mapping(struct map_desc *md)
  */
 void __init iotable_init(struct map_desc *io_desc, int nr)
 {
-       int i;
+       struct map_desc *md;
+       struct vm_struct *vm;
+
+       if (!nr)
+               return;
 
-       for (i = 0; i < nr; i++)
-               create_mapping(io_desc + i);
+       vm = early_alloc_aligned(sizeof(*vm) * nr, __alignof__(*vm));
+
+       for (md = io_desc; nr; md++, nr--) {
+               create_mapping(md);
+               vm->addr = (void *)(md->virtual & PAGE_MASK);
+               vm->size = PAGE_ALIGN(md->length + (md->virtual & ~PAGE_MASK));
+               vm->phys_addr = __pfn_to_phys(md->pfn); 
+               vm->flags = VM_IOREMAP | VM_ARM_STATIC_MAPPING; 
+               vm->flags |= VM_ARM_MTYPE(md->type);
+               vm->caller = iotable_init;
+               vm_area_add_early(vm++);
+       }
 }
 
-static unsigned long __initdata vmalloc_reserve = SZ_128M;
+static void * __initdata vmalloc_min =
+       (void *)(VMALLOC_END - (240 << 20) - VMALLOC_OFFSET);
 
 /*
  * vmalloc=size forces the vmalloc area to be exactly 'size'
  * bytes. This can be used to increase (or decrease) the vmalloc
- * area - the default is 128m.
+ * area - the default is 240m.
  */
 static int __init early_vmalloc(char *arg)
 {
-       vmalloc_reserve = memparse(arg, NULL);
+       unsigned long vmalloc_reserve = memparse(arg, NULL);
 
        if (vmalloc_reserve < SZ_16M) {
                vmalloc_reserve = SZ_16M;
@@ -705,13 +808,15 @@ static int __init early_vmalloc(char *arg)
                        "vmalloc area is too big, limiting to %luMB\n",
                        vmalloc_reserve >> 20);
        }
+
+       vmalloc_min = (void *)(VMALLOC_END - vmalloc_reserve);
        return 0;
 }
 early_param("vmalloc", early_vmalloc);
 
-#define VMALLOC_MIN    (void *)(VMALLOC_END - vmalloc_reserve)
+static phys_addr_t lowmem_limit __initdata = 0;
 
-static void __init sanity_check_meminfo(void)
+void __init sanity_check_meminfo(void)
 {
        int i, j, highmem = 0;
 
@@ -719,8 +824,11 @@ static void __init sanity_check_meminfo(void)
                struct membank *bank = &meminfo.bank[j];
                *bank = meminfo.bank[i];
 
+               if (bank->start > ULONG_MAX)
+                       highmem = 1;
+
 #ifdef CONFIG_HIGHMEM
-               if (__va(bank->start) > VMALLOC_MIN ||
+               if (__va(bank->start) >= vmalloc_min ||
                    __va(bank->start) < (void *)PAGE_OFFSET)
                        highmem = 1;
 
@@ -730,8 +838,8 @@ static void __init sanity_check_meminfo(void)
                 * Split those memory banks which are partially overlapping
                 * the vmalloc area greatly simplifying things later.
                 */
-               if (__va(bank->start) < VMALLOC_MIN &&
-                   bank->size > VMALLOC_MIN - __va(bank->start)) {
+               if (!highmem && __va(bank->start) < vmalloc_min &&
+                   bank->size > vmalloc_min - __va(bank->start)) {
                        if (meminfo.nr_banks >= NR_BANKS) {
                                printk(KERN_CRIT "NR_BANKS too low, "
                                                 "ignoring high memory\n");
@@ -740,25 +848,37 @@ static void __init sanity_check_meminfo(void)
                                        (meminfo.nr_banks - i) * sizeof(*bank));
                                meminfo.nr_banks++;
                                i++;
-                               bank[1].size -= VMALLOC_MIN - __va(bank->start);
-                               bank[1].start = __pa(VMALLOC_MIN - 1) + 1;
+                               bank[1].size -= vmalloc_min - __va(bank->start);
+                               bank[1].start = __pa(vmalloc_min - 1) + 1;
                                bank[1].highmem = highmem = 1;
                                j++;
                        }
-                       bank->size = VMALLOC_MIN - __va(bank->start);
+                       bank->size = vmalloc_min - __va(bank->start);
                }
 #else
                bank->highmem = highmem;
 
                /*
+                * Highmem banks not allowed with !CONFIG_HIGHMEM.
+                */
+               if (highmem) {
+                       printk(KERN_NOTICE "Ignoring RAM at %.8llx-%.8llx "
+                              "(!CONFIG_HIGHMEM).\n",
+                              (unsigned long long)bank->start,
+                              (unsigned long long)bank->start + bank->size - 1);
+                       continue;
+               }
+
+               /*
                 * Check whether this memory bank would entirely overlap
                 * the vmalloc area.
                 */
-               if (__va(bank->start) >= VMALLOC_MIN ||
+               if (__va(bank->start) >= vmalloc_min ||
                    __va(bank->start) < (void *)PAGE_OFFSET) {
-                       printk(KERN_NOTICE "Ignoring RAM at %.8lx-%.8lx "
+                       printk(KERN_NOTICE "Ignoring RAM at %.8llx-%.8llx "
                               "(vmalloc region overlap).\n",
-                              bank->start, bank->start + bank->size - 1);
+                              (unsigned long long)bank->start,
+                              (unsigned long long)bank->start + bank->size - 1);
                        continue;
                }
 
@@ -766,16 +886,20 @@ static void __init sanity_check_meminfo(void)
                 * Check whether this memory bank would partially overlap
                 * the vmalloc area.
                 */
-               if (__va(bank->start + bank->size) > VMALLOC_MIN ||
+               if (__va(bank->start + bank->size) > vmalloc_min ||
                    __va(bank->start + bank->size) < __va(bank->start)) {
-                       unsigned long newsize = VMALLOC_MIN - __va(bank->start);
-                       printk(KERN_NOTICE "Truncating RAM at %.8lx-%.8lx "
-                              "to -%.8lx (vmalloc region overlap).\n",
-                              bank->start, bank->start + bank->size - 1,
-                              bank->start + newsize - 1);
+                       unsigned long newsize = vmalloc_min - __va(bank->start);
+                       printk(KERN_NOTICE "Truncating RAM at %.8llx-%.8llx "
+                              "to -%.8llx (vmalloc region overlap).\n",
+                              (unsigned long long)bank->start,
+                              (unsigned long long)bank->start + bank->size - 1,
+                              (unsigned long long)bank->start + newsize - 1);
                        bank->size = newsize;
                }
 #endif
+               if (!bank->highmem && bank->start + bank->size > lowmem_limit)
+                       lowmem_limit = bank->start + bank->size;
+
                j++;
        }
 #ifdef CONFIG_HIGHMEM
@@ -789,18 +913,6 @@ static void __init sanity_check_meminfo(void)
                         * rather difficult.
                         */
                        reason = "with VIPT aliasing cache";
-#ifdef CONFIG_SMP
-               } else if (tlb_ops_need_broadcast()) {
-                       /*
-                        * kmap_high needs to occasionally flush TLB entries,
-                        * however, if the TLB entries need to be broadcast
-                        * we may deadlock:
-                        *  kmap_high(irqs off)->flush_all_zero_pkmaps->
-                        *  flush_tlb_kernel_range->smp_call_function_many
-                        *   (must not be called with irqs off)
-                        */
-                       reason = "without hardware TLB ops broadcasting";
-#endif
                }
                if (reason) {
                        printk(KERN_CRIT "HIGHMEM is not supported %s, ignoring high memory\n",
@@ -811,135 +923,75 @@ static void __init sanity_check_meminfo(void)
        }
 #endif
        meminfo.nr_banks = j;
+       high_memory = __va(lowmem_limit - 1) + 1;
+       memblock_set_current_limit(lowmem_limit);
 }
 
 static inline void prepare_page_table(void)
 {
        unsigned long addr;
+       phys_addr_t end;
 
        /*
         * Clear out all the mappings below the kernel image.
         */
-       for (addr = 0; addr < MODULES_VADDR; addr += PGDIR_SIZE)
+       for (addr = 0; addr < MODULES_VADDR; addr += PMD_SIZE)
                pmd_clear(pmd_off_k(addr));
 
 #ifdef CONFIG_XIP_KERNEL
        /* The XIP kernel is mapped in the module area -- skip over it */
-       addr = ((unsigned long)_etext + PGDIR_SIZE - 1) & PGDIR_MASK;
+       addr = ((unsigned long)_etext + PMD_SIZE - 1) & PMD_MASK;
 #endif
-       for ( ; addr < PAGE_OFFSET; addr += PGDIR_SIZE)
+       for ( ; addr < PAGE_OFFSET; addr += PMD_SIZE)
                pmd_clear(pmd_off_k(addr));
 
        /*
+        * Find the end of the first block of lowmem.
+        */
+       end = memblock.memory.regions[0].base + memblock.memory.regions[0].size;
+       if (end >= lowmem_limit)
+               end = lowmem_limit;
+
+       /*
         * Clear out all the kernel space mappings, except for the first
-        * memory bank, up to the end of the vmalloc region.
+        * memory bank, up to the vmalloc region.
         */
-       for (addr = __phys_to_virt(bank_phys_end(&meminfo.bank[0]));
-            addr < VMALLOC_END; addr += PGDIR_SIZE)
+       for (addr = __phys_to_virt(end);
+            addr < VMALLOC_START; addr += PMD_SIZE)
                pmd_clear(pmd_off_k(addr));
 }
 
-/*
- * Reserve the various regions of node 0
- */
-void __init reserve_node_zero(pg_data_t *pgdat)
-{
-       unsigned long res_size = 0;
-
-       /*
-        * Register the kernel text and data with bootmem.
-        * Note that this can only be in node 0.
-        */
-#ifdef CONFIG_XIP_KERNEL
-       reserve_bootmem_node(pgdat, __pa(_data), _end - _data,
-                       BOOTMEM_DEFAULT);
+#ifdef CONFIG_ARM_LPAE
+/* the first page is reserved for pgd */
+#define SWAPPER_PG_DIR_SIZE    (PAGE_SIZE + \
+                                PTRS_PER_PGD * PTRS_PER_PMD * sizeof(pmd_t))
 #else
-       reserve_bootmem_node(pgdat, __pa(_stext), _end - _stext,
-                       BOOTMEM_DEFAULT);
+#define SWAPPER_PG_DIR_SIZE    (PTRS_PER_PGD * sizeof(pgd_t))
 #endif
 
+/*
+ * Reserve the special regions of memory
+ */
+void __init arm_mm_memblock_reserve(void)
+{
        /*
         * Reserve the page tables.  These are already in use,
         * and can only be in node 0.
         */
-       reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
-                            PTRS_PER_PGD * sizeof(pgd_t), BOOTMEM_DEFAULT);
-
-       /*
-        * Hmm... This should go elsewhere, but we really really need to
-        * stop things allocating the low memory; ideally we need a better
-        * implementation of GFP_DMA which does not assume that DMA-able
-        * memory starts at zero.
-        */
-       if (machine_is_integrator() || machine_is_cintegrator())
-               res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
-
-       /*
-        * These should likewise go elsewhere.  They pre-reserve the
-        * screen memory region at the start of main system memory.
-        */
-       if (machine_is_edb7211())
-               res_size = 0x00020000;
-       if (machine_is_p720t())
-               res_size = 0x00014000;
-
-       /* H1940, RX3715 and RX1950 need to reserve this for suspend */
-
-       if (machine_is_h1940() || machine_is_rx3715()
-               || machine_is_rx1950()) {
-               reserve_bootmem_node(pgdat, 0x30003000, 0x1000,
-                               BOOTMEM_DEFAULT);
-               reserve_bootmem_node(pgdat, 0x30081000, 0x1000,
-                               BOOTMEM_DEFAULT);
-       }
-
-       if (machine_is_palmld() || machine_is_palmtx()) {
-               reserve_bootmem_node(pgdat, 0xa0000000, 0x1000,
-                               BOOTMEM_EXCLUSIVE);
-               reserve_bootmem_node(pgdat, 0xa0200000, 0x1000,
-                               BOOTMEM_EXCLUSIVE);
-       }
-
-       if (machine_is_treo680() || machine_is_centro()) {
-               reserve_bootmem_node(pgdat, 0xa0000000, 0x1000,
-                               BOOTMEM_EXCLUSIVE);
-               reserve_bootmem_node(pgdat, 0xa2000000, 0x1000,
-                               BOOTMEM_EXCLUSIVE);
-       }
-
-       if (machine_is_palmt5())
-               reserve_bootmem_node(pgdat, 0xa0200000, 0x1000,
-                               BOOTMEM_EXCLUSIVE);
-
-       /*
-        * U300 - This platform family can share physical memory
-        * between two ARM cpus, one running Linux and the other
-        * running another OS.
-        */
-       if (machine_is_u300()) {
-#ifdef CONFIG_MACH_U300_SINGLE_RAM
-#if ((CONFIG_MACH_U300_ACCESS_MEM_SIZE & 1) == 1) &&   \
-       CONFIG_MACH_U300_2MB_ALIGNMENT_FIX
-               res_size = 0x00100000;
-#endif
-#endif
-       }
+       memblock_reserve(__pa(swapper_pg_dir), SWAPPER_PG_DIR_SIZE);
 
 #ifdef CONFIG_SA1111
        /*
         * Because of the SA1111 DMA bug, we want to preserve our
         * precious DMA-able memory...
         */
-       res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
+       memblock_reserve(PHYS_OFFSET, __pa(swapper_pg_dir) - PHYS_OFFSET);
 #endif
-       if (res_size)
-               reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size,
-                               BOOTMEM_DEFAULT);
 }
 
 /*
- * Set up device the mappings.  Since we clear out the page tables for all
- * mappings above VMALLOC_END, we will remove any debug device mappings.
+ * Set up the device mappings.  Since we clear out the page tables for all
+ * mappings above VMALLOC_START, we will remove any debug device mappings.
  * This means you have to be careful how you debug this function, or any
  * called function.  This means you can't use any function or debugging
  * method which may touch any device, otherwise the kernel _will_ crash.
@@ -953,9 +1005,11 @@ static void __init devicemaps_init(struct machine_desc *mdesc)
        /*
         * Allocate the vector page early.
         */
-       vectors = alloc_bootmem_low_pages(PAGE_SIZE);
+       vectors = early_alloc(PAGE_SIZE);
+
+       early_trap_init(vectors);
 
-       for (addr = VMALLOC_END; addr; addr += PGDIR_SIZE)
+       for (addr = VMALLOC_START; addr; addr += PMD_SIZE)
                pmd_clear(pmd_off_k(addr));
 
        /*
@@ -1024,45 +1078,33 @@ static void __init devicemaps_init(struct machine_desc *mdesc)
 static void __init kmap_init(void)
 {
 #ifdef CONFIG_HIGHMEM
-       pmd_t *pmd = pmd_off_k(PKMAP_BASE);
-       pte_t *pte = alloc_bootmem_low_pages(2 * PTRS_PER_PTE * sizeof(pte_t));
-       BUG_ON(!pmd_none(*pmd) || !pte);
-       __pmd_populate(pmd, __pa(pte) | _PAGE_KERNEL_TABLE);
-       pkmap_page_table = pte + PTRS_PER_PTE;
+       pkmap_page_table = early_pte_alloc(pmd_off_k(PKMAP_BASE),
+               PKMAP_BASE, _PAGE_KERNEL_TABLE);
 #endif
 }
 
-static inline void map_memory_bank(struct membank *bank)
-{
-       struct map_desc map;
-
-       map.pfn = bank_pfn_start(bank);
-       map.virtual = __phys_to_virt(bank_phys_start(bank));
-       map.length = bank_phys_size(bank);
-       map.type = MT_MEMORY;
-
-       create_mapping(&map);
-}
-
 static void __init map_lowmem(void)
 {
-       struct meminfo *mi = &meminfo;
-       int i;
+       struct memblock_region *reg;
 
        /* Map all the lowmem memory banks. */
-       for (i = 0; i < mi->nr_banks; i++) {
-               struct membank *bank = &mi->bank[i];
+       for_each_memblock(memory, reg) {
+               phys_addr_t start = reg->base;
+               phys_addr_t end = start + reg->size;
+               struct map_desc map;
+
+               if (end > lowmem_limit)
+                       end = lowmem_limit;
+               if (start >= end)
+                       break;
 
-               if (!bank->highmem)
-                       map_memory_bank(bank);
-       }
-}
+               map.pfn = __phys_to_pfn(start);
+               map.virtual = __phys_to_virt(start);
+               map.length = end - start;
+               map.type = MT_MEMORY;
 
-static int __init meminfo_cmp(const void *_a, const void *_b)
-{
-       const struct membank *a = _a, *b = _b;
-       long cmp = bank_pfn_start(a) - bank_pfn_start(b);
-       return cmp < 0 ? -1 : cmp > 0 ? 1 : 0;
+               create_mapping(&map);
+       }
 }
 
 /*
@@ -1073,58 +1115,21 @@ void __init paging_init(struct machine_desc *mdesc)
 {
        void *zero_page;
 
-       sort(&meminfo.bank, meminfo.nr_banks, sizeof(meminfo.bank[0]), meminfo_cmp, NULL);
+       memblock_set_current_limit(lowmem_limit);
 
        build_mem_type_table();
-       sanity_check_meminfo();
        prepare_page_table();
        map_lowmem();
-       bootmem_init();
        devicemaps_init(mdesc);
        kmap_init();
 
        top_pmd = pmd_off_k(0xffff0000);
 
-       /*
-        * allocate the zero page.  Note that this always succeeds and
-        * returns a zeroed result.
-        */
-       zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
-       empty_zero_page = virt_to_page(zero_page);
-       __flush_dcache_page(NULL, empty_zero_page);
-}
-
-/*
- * In order to soft-boot, we need to insert a 1:1 mapping in place of
- * the user-mode pages.  This will then ensure that we have predictable
- * results when turning the mmu off
- */
-void setup_mm_for_reboot(char mode)
-{
-       unsigned long base_pmdval;
-       pgd_t *pgd;
-       int i;
-
-       /*
-        * We need to access to user-mode page tables here. For kernel threads
-        * we don't have any user-mode mappings so we use the context that we
-        * "borrowed".
-        */
-       pgd = current->active_mm->pgd;
-
-       base_pmdval = PMD_SECT_AP_WRITE | PMD_SECT_AP_READ | PMD_TYPE_SECT;
-       if (cpu_architecture() <= CPU_ARCH_ARMv5TEJ && !cpu_is_xscale())
-               base_pmdval |= PMD_BIT4;
+       /* allocate the zero page. */
+       zero_page = early_alloc(PAGE_SIZE);
 
-       for (i = 0; i < FIRST_USER_PGD_NR + USER_PTRS_PER_PGD; i++, pgd++) {
-               unsigned long pmdval = (i << PGDIR_SHIFT) | base_pmdval;
-               pmd_t *pmd;
-
-               pmd = pmd_off(pgd, i << PGDIR_SHIFT);
-               pmd[0] = __pmd(pmdval);
-               pmd[1] = __pmd(pmdval + (1 << (PGDIR_SHIFT - 1)));
-               flush_pmd_entry(pmd);
-       }
+       bootmem_init();
 
-       local_flush_tlb_all();
+       empty_zero_page = virt_to_page(zero_page);
+       __flush_dcache_page(NULL, empty_zero_page);
 }