metag: Memory management

Add memory management files for metag.

Meta's 32bit virtual address space is split into two halves:
 - local (0x08000000-0x7fffffff): traditionally local to a hardware
   thread and incoherent between hardware threads. Each hardware thread
   has it's own local MMU table. On Meta2 the local space can be
   globally coherent (GCOn) if the cache partitions coincide.
 - global (0x88000000-0xffff0000): coherent and traditionally global
   between hardware threads. On Meta2, each hardware thread has it's own
   global MMU table.

The low 128MiB of each half is non-MMUable and maps directly to the
physical address space:
 - 0x00010000-0x07ffffff: contains Meta core registers and maps SoC bus
 - 0x80000000-0x87ffffff: contains low latency global core memories

Linux usually further splits the local virtual address space like this:
 - 0x08000000-0x3fffffff: user mappings
 - 0x40000000-0x7fffffff: kernel mappings

Signed-off-by: James Hogan <james.hogan@imgtec.com>

5 files changed, 1066 insertions, 0 deletions

diff --git a/arch/metag/mm/extable.c b/arch/metag/mm/extable.c
new file mode 100644
index 000000000000..2a21eaebe84d
--- /dev/null
+++ b/arch/metag/mm/extable.c
@@ -0,0 +1,15 @@
+
+#include <linux/module.h>
+#include <linux/uaccess.h>
+
+int fixup_exception(struct pt_regs *regs)
+{
+	const struct exception_table_entry *fixup;
+	unsigned long pc = instruction_pointer(regs);
+
+	fixup = search_exception_tables(pc);
+	if (fixup)
+		regs->ctx.CurrPC = fixup->fixup;
+
+	return fixup != NULL;
+}
diff --git a/arch/metag/mm/fault.c b/arch/metag/mm/fault.c
new file mode 100644
index 000000000000..2c75bf7357c5
--- /dev/null
+++ b/arch/metag/mm/fault.c
@@ -0,0 +1,239 @@
+/*
+ *  Meta page fault handling.
+ *
+ *  Copyright (C) 2005-2012 Imagination Technologies Ltd.
+ */
+
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/kernel.h>
+#include <linux/ptrace.h>
+#include <linux/interrupt.h>
+#include <linux/uaccess.h>
+
+#include <asm/tlbflush.h>
+#include <asm/mmu.h>
+#include <asm/traps.h>
+
+/* Clear any pending catch buffer state. */
+static void clear_cbuf_entry(struct pt_regs *regs, unsigned long addr,
+			     unsigned int trapno)
+{
+	PTBICTXEXTCB0 cbuf = regs->extcb0;
+
+	switch (trapno) {
+		/* Instruction fetch faults leave no catch buffer state. */
+	case TBIXXF_SIGNUM_IGF:
+	case TBIXXF_SIGNUM_IPF:
+		return;
+	default:
+		if (cbuf[0].CBAddr == addr) {
+			cbuf[0].CBAddr = 0;
+			cbuf[0].CBFlags &= ~TXCATCH0_FAULT_BITS;
+
+			/* And, as this is the ONLY catch entry, we
+			 * need to clear the cbuf bit from the context!
+			 */
+			regs->ctx.SaveMask &= ~(TBICTX_CBUF_BIT |
+						TBICTX_XCBF_BIT);
+
+			return;
+		}
+		pr_err("Failed to clear cbuf entry!\n");
+	}
+}
+
+int show_unhandled_signals = 1;
+
+int do_page_fault(struct pt_regs *regs, unsigned long address,
+		  unsigned int write_access, unsigned int trapno)
+{
+	struct task_struct *tsk;
+	struct mm_struct *mm;
+	struct vm_area_struct *vma, *prev_vma;
+	siginfo_t info;
+	int fault;
+	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
+				(write_access ? FAULT_FLAG_WRITE : 0);
+
+	tsk = current;
+
+	if ((address >= VMALLOC_START) && (address < VMALLOC_END)) {
+		/*
+		 * Synchronize this task's top level page-table
+		 * with the 'reference' page table.
+		 *
+		 * Do _not_ use "tsk" here. We might be inside
+		 * an interrupt in the middle of a task switch..
+		 */
+		int offset = pgd_index(address);
+		pgd_t *pgd, *pgd_k;
+		pud_t *pud, *pud_k;
+		pmd_t *pmd, *pmd_k;
+		pte_t *pte_k;
+
+		pgd = ((pgd_t *)mmu_get_base()) + offset;
+		pgd_k = swapper_pg_dir + offset;
+
+		/* This will never happen with the folded page table. */
+		if (!pgd_present(*pgd)) {
+			if (!pgd_present(*pgd_k))
+				goto bad_area_nosemaphore;
+			set_pgd(pgd, *pgd_k);
+			return 0;
+		}
+
+		pud = pud_offset(pgd, address);
+		pud_k = pud_offset(pgd_k, address);
+		if (!pud_present(*pud_k))
+			goto bad_area_nosemaphore;
+		set_pud(pud, *pud_k);
+
+		pmd = pmd_offset(pud, address);
+		pmd_k = pmd_offset(pud_k, address);
+		if (!pmd_present(*pmd_k))
+			goto bad_area_nosemaphore;
+		set_pmd(pmd, *pmd_k);
+
+		pte_k = pte_offset_kernel(pmd_k, address);
+		if (!pte_present(*pte_k))
+			goto bad_area_nosemaphore;
+
+		/* May only be needed on Chorus2 */
+		flush_tlb_all();
+		return 0;
+	}
+
+	mm = tsk->mm;
+
+	if (in_atomic() || !mm)
+		goto no_context;
+
+retry:
+	down_read(&mm->mmap_sem);
+
+	vma = find_vma_prev(mm, address, &prev_vma);
+
+	if (!vma || address < vma->vm_start)
+		goto check_expansion;
+
+good_area:
+	if (write_access) {
+		if (!(vma->vm_flags & VM_WRITE))
+			goto bad_area;
+	} else {
+		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
+			goto bad_area;
+	}
+
+	/*
+	 * If for any reason at all we couldn't handle the fault,
+	 * make sure we exit gracefully rather than endlessly redo
+	 * the fault.
+	 */
+	fault = handle_mm_fault(mm, vma, address, flags);
+
+	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
+		return 0;
+
+	if (unlikely(fault & VM_FAULT_ERROR)) {
+		if (fault & VM_FAULT_OOM)
+			goto out_of_memory;
+		else if (fault & VM_FAULT_SIGBUS)
+			goto do_sigbus;
+		BUG();
+	}
+	if (flags & FAULT_FLAG_ALLOW_RETRY) {
+		if (fault & VM_FAULT_MAJOR)
+			tsk->maj_flt++;
+		else
+			tsk->min_flt++;
+		if (fault & VM_FAULT_RETRY) {
+			flags &= ~FAULT_FLAG_ALLOW_RETRY;
+			flags |= FAULT_FLAG_TRIED;
+
+			/*
+			 * No need to up_read(&mm->mmap_sem) as we would
+			 * have already released it in __lock_page_or_retry
+			 * in mm/filemap.c.
+			 */
+
+			goto retry;
+		}
+	}
+
+	up_read(&mm->mmap_sem);
+	return 0;
+
+check_expansion:
+	vma = prev_vma;
+	if (vma && (expand_stack(vma, address) == 0))
+		goto good_area;
+
+bad_area:
+	up_read(&mm->mmap_sem);
+
+bad_area_nosemaphore:
+	if (user_mode(regs)) {
+		info.si_signo = SIGSEGV;
+		info.si_errno = 0;
+		info.si_code = SEGV_MAPERR;
+		info.si_addr = (__force void __user *)address;
+		info.si_trapno = trapno;
+
+		if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
+		    printk_ratelimit()) {
+			pr_info("%s%s[%d]: segfault at %lx pc %08x sp %08x write %d trap %#x (%s)",
+			       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
+			       tsk->comm, task_pid_nr(tsk), address,
+			       regs->ctx.CurrPC, regs->ctx.AX[0].U0,
+			       write_access, trapno, trap_name(trapno));
+			print_vma_addr(" in ", regs->ctx.CurrPC);
+			print_vma_addr(" rtp in ", regs->ctx.DX[4].U1);
+			printk("\n");
+			show_regs(regs);
+		}
+		force_sig_info(SIGSEGV, &info, tsk);
+		return 1;
+	}
+	goto no_context;
+
+do_sigbus:
+	up_read(&mm->mmap_sem);
+
+	/*
+	 * Send a sigbus, regardless of whether we were in kernel
+	 * or user mode.
+	 */
+	info.si_signo = SIGBUS;
+	info.si_errno = 0;
+	info.si_code = BUS_ADRERR;
+	info.si_addr = (__force void __user *)address;
+	info.si_trapno = trapno;
+	force_sig_info(SIGBUS, &info, tsk);
+
+	/* Kernel mode? Handle exceptions or die */
+	if (!user_mode(regs))
+		goto no_context;
+
+	return 1;
+
+	/*
+	 * We ran out of memory, or some other thing happened to us that made
+	 * us unable to handle the page fault gracefully.
+	 */
+out_of_memory:
+	up_read(&mm->mmap_sem);
+	if (user_mode(regs))
+		do_group_exit(SIGKILL);
+
+no_context:
+	/* Are we prepared to handle this kernel fault?  */
+	if (fixup_exception(regs)) {
+		clear_cbuf_entry(regs, address, trapno);
+		return 1;
+	}
+
+	die("Oops", regs, (write_access << 15) | trapno, address);
+	do_exit(SIGKILL);
+}
diff --git a/arch/metag/mm/init.c b/arch/metag/mm/init.c
new file mode 100644
index 000000000000..514376d90db4
--- /dev/null
+++ b/arch/metag/mm/init.c
@@ -0,0 +1,448 @@
+/*
+ *  Copyright (C) 2005,2006,2007,2008,2009,2010 Imagination Technologies
+ *
+ */
+
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/pagemap.h>
+#include <linux/percpu.h>
+#include <linux/memblock.h>
+#include <linux/initrd.h>
+#include <linux/of_fdt.h>
+
+#include <asm/setup.h>
+#include <asm/page.h>
+#include <asm/pgalloc.h>
+#include <asm/mmu.h>
+#include <asm/mmu_context.h>
+#include <asm/sections.h>
+#include <asm/tlb.h>
+#include <asm/user_gateway.h>
+#include <asm/mmzone.h>
+#include <asm/fixmap.h>
+
+unsigned long pfn_base;
+
+pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_data;
+
+unsigned long empty_zero_page;
+
+extern char __user_gateway_start;
+extern char __user_gateway_end;
+
+void *gateway_page;
+
+/*
+ * Insert the gateway page into a set of page tables, creating the
+ * page tables if necessary.
+ */
+static void insert_gateway_page(pgd_t *pgd, unsigned long address)
+{
+	pud_t *pud;
+	pmd_t *pmd;
+	pte_t *pte;
+
+	BUG_ON(!pgd_present(*pgd));
+
+	pud = pud_offset(pgd, address);
+	BUG_ON(!pud_present(*pud));
+
+	pmd = pmd_offset(pud, address);
+	if (!pmd_present(*pmd)) {
+		pte = alloc_bootmem_pages(PAGE_SIZE);
+		set_pmd(pmd, __pmd(_PAGE_TABLE | __pa(pte)));
+	}
+
+	pte = pte_offset_kernel(pmd, address);
+	set_pte(pte, pfn_pte(__pa(gateway_page) >> PAGE_SHIFT, PAGE_READONLY));
+}
+
+/* Alloc and map a page in a known location accessible to userspace. */
+static void __init user_gateway_init(void)
+{
+	unsigned long address = USER_GATEWAY_PAGE;
+	int offset = pgd_index(address);
+	pgd_t *pgd;
+
+	gateway_page = alloc_bootmem_pages(PAGE_SIZE);
+
+	pgd = swapper_pg_dir + offset;
+	insert_gateway_page(pgd, address);
+
+#ifdef CONFIG_METAG_META12
+	/*
+	 * Insert the gateway page into our current page tables even
+	 * though we've already inserted it into our reference page
+	 * table (swapper_pg_dir). This is because with a META1 mmu we
+	 * copy just the user address range and not the gateway page
+	 * entry on context switch, see switch_mmu().
+	 */
+	pgd = (pgd_t *)mmu_get_base() + offset;
+	insert_gateway_page(pgd, address);
+#endif /* CONFIG_METAG_META12 */
+
+	BUG_ON((&__user_gateway_end - &__user_gateway_start) > PAGE_SIZE);
+
+	gateway_page += (address & ~PAGE_MASK);
+
+	memcpy(gateway_page, &__user_gateway_start,
+	       &__user_gateway_end - &__user_gateway_start);
+
+	/*
+	 * We don't need to flush the TLB here, there should be no mapping
+	 * present at boot for this address and only valid mappings are in
+	 * the TLB (apart from on Meta 1.x, but those cached invalid
+	 * mappings should be impossible to hit here).
+	 *
+	 * We don't flush the code cache here even though we have written
+	 * code through the data cache and they may not be coherent. At
+	 * this point we assume there is no stale data in the code cache
+	 * for this address so there is no need to flush.
+	 */
+}
+
+static void __init allocate_pgdat(unsigned int nid)
+{
+	unsigned long start_pfn, end_pfn;
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+	unsigned long phys;
+#endif
+
+	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
+
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+	phys = __memblock_alloc_base(sizeof(struct pglist_data),
+				SMP_CACHE_BYTES, end_pfn << PAGE_SHIFT);
+	/* Retry with all of system memory */
+	if (!phys)
+		phys = __memblock_alloc_base(sizeof(struct pglist_data),
+					     SMP_CACHE_BYTES,
+					     memblock_end_of_DRAM());
+	if (!phys)
+		panic("Can't allocate pgdat for node %d\n", nid);
+
+	NODE_DATA(nid) = __va(phys);
+	memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
+
+	NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
+#endif
+
+	NODE_DATA(nid)->node_start_pfn = start_pfn;
+	NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
+}
+
+static void __init bootmem_init_one_node(unsigned int nid)
+{
+	unsigned long total_pages, paddr;
+	unsigned long end_pfn;
+	struct pglist_data *p;
+
+	p = NODE_DATA(nid);
+
+	/* Nothing to do.. */
+	if (!p->node_spanned_pages)
+		return;
+
+	end_pfn = p->node_start_pfn + p->node_spanned_pages;
+#ifdef CONFIG_HIGHMEM
+	if (end_pfn > max_low_pfn)
+		end_pfn = max_low_pfn;
+#endif
+
+	total_pages = bootmem_bootmap_pages(end_pfn - p->node_start_pfn);
+
+	paddr = memblock_alloc(total_pages << PAGE_SHIFT, PAGE_SIZE);
+	if (!paddr)
+		panic("Can't allocate bootmap for nid[%d]\n", nid);
+
+	init_bootmem_node(p, paddr >> PAGE_SHIFT, p->node_start_pfn, end_pfn);
+
+	free_bootmem_with_active_regions(nid, end_pfn);
+
+	/*
+	 * XXX Handle initial reservations for the system memory node
+	 * only for the moment, we'll refactor this later for handling
+	 * reservations in other nodes.
+	 */
+	if (nid == 0) {
+		struct memblock_region *reg;
+
+		/* Reserve the sections we're already using. */
+		for_each_memblock(reserved, reg) {
+			unsigned long size = reg->size;
+
+#ifdef CONFIG_HIGHMEM
+			/* ...but not highmem */
+			if (PFN_DOWN(reg->base) >= highstart_pfn)
+				continue;
+
+			if (PFN_UP(reg->base + size) > highstart_pfn)
+				size = (highstart_pfn - PFN_DOWN(reg->base))
+				       << PAGE_SHIFT;
+#endif
+
+			reserve_bootmem(reg->base, size, BOOTMEM_DEFAULT);
+		}
+	}
+
+	sparse_memory_present_with_active_regions(nid);
+}
+
+static void __init do_init_bootmem(void)
+{
+	struct memblock_region *reg;
+	int i;
+
+	/* Add active regions with valid PFNs. */
+	for_each_memblock(memory, reg) {
+		unsigned long start_pfn, end_pfn;
+		start_pfn = memblock_region_memory_base_pfn(reg);
+		end_pfn = memblock_region_memory_end_pfn(reg);
+		memblock_set_node(PFN_PHYS(start_pfn),
+				  PFN_PHYS(end_pfn - start_pfn), 0);
+	}
+
+	/* All of system RAM sits in node 0 for the non-NUMA case */
+	allocate_pgdat(0);
+	node_set_online(0);
+
+	soc_mem_setup();
+
+	for_each_online_node(i)
+		bootmem_init_one_node(i);
+
+	sparse_init();
+}
+
+extern char _heap_start[];
+
+static void __init init_and_reserve_mem(void)
+{
+	unsigned long start_pfn, heap_start;
+	u64 base = min_low_pfn << PAGE_SHIFT;
+	u64 size = (max_low_pfn << PAGE_SHIFT) - base;
+
+	heap_start = (unsigned long) &_heap_start;
+
+	memblock_add(base, size);
+
+	/*
+	 * Partially used pages are not usable - thus
+	 * we are rounding upwards:
+	 */
+	start_pfn = PFN_UP(__pa(heap_start));
+
+	/*
+	 * Reserve the kernel text.
+	 */
+	memblock_reserve(base, (PFN_PHYS(start_pfn) + PAGE_SIZE - 1) - base);
+
+#ifdef CONFIG_HIGHMEM
+	/*
+	 * Add & reserve highmem, so page structures are initialised.
+	 */
+	base = highstart_pfn << PAGE_SHIFT;
+	size = (highend_pfn << PAGE_SHIFT) - base;
+	if (size) {
+		memblock_add(base, size);
+		memblock_reserve(base, size);
+	}
+#endif
+}
+
+#ifdef CONFIG_HIGHMEM
+/*
+ * Ensure we have allocated page tables in swapper_pg_dir for the
+ * fixed mappings range from 'start' to 'end'.
+ */
+static void __init allocate_pgtables(unsigned long start, unsigned long end)
+{
+	pgd_t *pgd;
+	pmd_t *pmd;
+	pte_t *pte;
+	int i, j;
+	unsigned long vaddr;
+
+	vaddr = start;
+	i = pgd_index(vaddr);
+	j = pmd_index(vaddr);
+	pgd = swapper_pg_dir + i;
+
+	for ( ; (i < PTRS_PER_PGD) && (vaddr != end); pgd++, i++) {
+		pmd = (pmd_t *)pgd;
+		for (; (j < PTRS_PER_PMD) && (vaddr != end); pmd++, j++) {
+			vaddr += PMD_SIZE;
+
+			if (!pmd_none(*pmd))
+				continue;
+
+			pte = (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
+			pmd_populate_kernel(&init_mm, pmd, pte);
+		}
+		j = 0;
+	}
+}
+
+static void __init fixedrange_init(void)
+{
+	unsigned long vaddr, end;
+	pgd_t *pgd;
+	pud_t *pud;
+	pmd_t *pmd;
+	pte_t *pte;
+
+	/*
+	 * Fixed mappings:
+	 */
+	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
+	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
+	allocate_pgtables(vaddr, end);
+
+	/*
+	 * Permanent kmaps:
+	 */
+	vaddr = PKMAP_BASE;
+	allocate_pgtables(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP);
+
+	pgd = swapper_pg_dir + pgd_index(vaddr);
+	pud = pud_offset(pgd, vaddr);
+	pmd = pmd_offset(pud, vaddr);
+	pte = pte_offset_kernel(pmd, vaddr);
+	pkmap_page_table = pte;
+}
+#endif /* CONFIG_HIGHMEM */
+
+/*
+ * paging_init() continues the virtual memory environment setup which
+ * was begun by the code in arch/metag/kernel/setup.c.
+ */
+void __init paging_init(unsigned long mem_end)
+{
+	unsigned long max_zone_pfns[MAX_NR_ZONES];
+	int nid;
+
+	init_and_reserve_mem();
+
+	memblock_allow_resize();
+
+	memblock_dump_all();
+
+	nodes_clear(node_online_map);
+
+	init_new_context(&init_task, &init_mm);
+
+	memset(swapper_pg_dir, 0, sizeof(swapper_pg_dir));
+
+	do_init_bootmem();
+	mmu_init(mem_end);
+
+#ifdef CONFIG_HIGHMEM
+	fixedrange_init();
+	kmap_init();
+#endif
+
+	/* Initialize the zero page to a bootmem page, already zeroed. */
+	empty_zero_page = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
+
+	user_gateway_init();
+
+	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
+
+	for_each_online_node(nid) {
+		pg_data_t *pgdat = NODE_DATA(nid);
+		unsigned long low, start_pfn;
+
+		start_pfn = pgdat->bdata->node_min_pfn;
+		low = pgdat->bdata->node_low_pfn;
+
+		if (max_zone_pfns[ZONE_NORMAL] < low)
+			max_zone_pfns[ZONE_NORMAL] = low;
+
+#ifdef CONFIG_HIGHMEM
+		max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
+#endif
+		pr_info("Node %u: start_pfn = 0x%lx, low = 0x%lx\n",
+			nid, start_pfn, low);
+	}
+
+	free_area_init_nodes(max_zone_pfns);
+}
+
+void __init mem_init(void)
+{
+	int nid;
+
+#ifdef CONFIG_HIGHMEM
+	unsigned long tmp;
+	for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
+		struct page *page = pfn_to_page(tmp);
+		ClearPageReserved(page);
+		init_page_count(page);
+		__free_page(page);
+		totalhigh_pages++;
+	}
+	totalram_pages += totalhigh_pages;
+	num_physpages += totalhigh_pages;
+#endif /* CONFIG_HIGHMEM */
+
+	for_each_online_node(nid) {
+		pg_data_t *pgdat = NODE_DATA(nid);
+		unsigned long node_pages = 0;
+
+		num_physpages += pgdat->node_present_pages;
+
+		if (pgdat->node_spanned_pages)
+			node_pages = free_all_bootmem_node(pgdat);
+
+		totalram_pages += node_pages;
+	}
+
+	pr_info("Memory: %luk/%luk available\n",
+		(unsigned long)nr_free_pages() << (PAGE_SHIFT - 10),
+		num_physpages << (PAGE_SHIFT - 10));
+
+	show_mem(0);
+
+	return;
+}
+
+static void free_init_pages(char *what, unsigned long begin, unsigned long end)
+{
+	unsigned long addr;
+
+	for (addr = begin; addr < end; addr += PAGE_SIZE) {
+		ClearPageReserved(virt_to_page(addr));
+		init_page_count(virt_to_page(addr));
+		memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
+		free_page(addr);
+		totalram_pages++;
+	}
+	pr_info("Freeing %s: %luk freed\n", what, (end - begin) >> 10);
+}
+
+void free_initmem(void)
+{
+	free_init_pages("unused kernel memory",
+			(unsigned long)(&__init_begin),
+			(unsigned long)(&__init_end));
+}
+
+#ifdef CONFIG_BLK_DEV_INITRD
+void free_initrd_mem(unsigned long start, unsigned long end)
+{
+	end = end & PAGE_MASK;
+	free_init_pages("initrd memory", start, end);
+}
+#endif
+
+#ifdef CONFIG_OF_FLATTREE
+void __init early_init_dt_setup_initrd_arch(unsigned long start,
+					    unsigned long end)
+{
+	pr_err("%s(%lx, %lx)\n",
+	       __func__, start, end);
+}
+#endif /* CONFIG_OF_FLATTREE */
diff --git a/arch/metag/mm/mmu-meta1.c b/arch/metag/mm/mmu-meta1.c
new file mode 100644
index 000000000000..91f4255bcb5c
--- /dev/null
+++ b/arch/metag/mm/mmu-meta1.c
@@ -0,0 +1,157 @@
+/*
+ *  Copyright (C) 2005,2006,2007,2008,2009 Imagination Technologies
+ *
+ * Meta 1 MMU handling code.
+ *
+ */
+
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/io.h>
+
+#include <asm/mmu.h>
+
+#define DM3_BASE (LINSYSDIRECT_BASE + (MMCU_DIRECTMAPn_ADDR_SCALE * 3))
+
+/*
+ * This contains the physical address of the top level 2k pgd table.
+ */
+static unsigned long mmu_base_phys;
+
+/*
+ * Given a physical address, return a mapped virtual address that can be used
+ * to access that location.
+ * In practice, we use the DirectMap region to make this happen.
+ */
+static unsigned long map_addr(unsigned long phys)
+{
+	static unsigned long dm_base = 0xFFFFFFFF;
+	int offset;
+
+	offset = phys - dm_base;
+
+	/* Are we in the current map range ? */
+	if ((offset < 0) || (offset >= MMCU_DIRECTMAPn_ADDR_SCALE)) {
+		/* Calculate new DM area */
+		dm_base = phys & ~(MMCU_DIRECTMAPn_ADDR_SCALE - 1);
+
+		/* Actually map it in! */
+		metag_out32(dm_base, MMCU_DIRECTMAP3_ADDR);
+
+		/* And calculate how far into that area our reference is */
+		offset = phys - dm_base;
+	}
+
+	return DM3_BASE + offset;
+}
+
+/*
+ * Return the physical address of the base of our pgd table.
+ */
+static inline unsigned long __get_mmu_base(void)
+{
+	unsigned long base_phys;
+	unsigned int stride;
+
+	if (is_global_space(PAGE_OFFSET))
+		stride = 4;
+	else
+		stride = hard_processor_id();	/* [0..3] */
+
+	base_phys = metag_in32(MMCU_TABLE_PHYS_ADDR);
+	base_phys += (0x800 * stride);
+
+	return base_phys;
+}
+
+/* Given a virtual address, return the virtual address of the relevant pgd */
+static unsigned long pgd_entry_addr(unsigned long virt)
+{
+	unsigned long pgd_phys;
+	unsigned long pgd_virt;
+
+	if (!mmu_base_phys)
+		mmu_base_phys = __get_mmu_base();
+
+	/*
+	 * Are we trying to map a global address.  If so, then index
+	 * the global pgd table instead of our local one.
+	 */
+	if (is_global_space(virt)) {
+		/* Scale into 2gig map */
+		virt &= ~0x80000000;
+	}
+
+	/* Base of the pgd table plus our 4Meg entry, 4bytes each */
+	pgd_phys = mmu_base_phys + ((virt >> PGDIR_SHIFT) * 4);
+
+	pgd_virt = map_addr(pgd_phys);
+
+	return pgd_virt;
+}
+
+/* Given a virtual address, return the virtual address of the relevant pte */
+static unsigned long pgtable_entry_addr(unsigned long virt)
+{
+	unsigned long pgtable_phys;
+	unsigned long pgtable_virt, pte_virt;
+
+	/* Find the physical address of the 4MB page table*/
+	pgtable_phys = metag_in32(pgd_entry_addr(virt)) & MMCU_ENTRY_ADDR_BITS;
+
+	/* Map it to a virtual address */
+	pgtable_virt = map_addr(pgtable_phys);
+
+	/* And index into it for our pte */
+	pte_virt = pgtable_virt + ((virt >> PAGE_SHIFT) & 0x3FF) * 4;
+
+	return pte_virt;
+}
+
+unsigned long mmu_read_first_level_page(unsigned long vaddr)
+{
+	return metag_in32(pgd_entry_addr(vaddr));
+}
+
+unsigned long mmu_read_second_level_page(unsigned long vaddr)
+{
+	return metag_in32(pgtable_entry_addr(vaddr));
+}
+
+unsigned long mmu_get_base(void)
+{
+	static unsigned long __base;
+
+	/* Find the base of our MMU pgd table */
+	if (!__base)
+		__base = pgd_entry_addr(0);
+
+	return __base;
+}
+
+void __init mmu_init(unsigned long mem_end)
+{
+	unsigned long entry, addr;
+	pgd_t *p_swapper_pg_dir;
+
+	/*
+	 * Now copy over any MMU pgd entries already in the mmu page tables
+	 * over to our root init process (swapper_pg_dir) map.  This map is
+	 * then inherited by all other processes, which means all processes
+	 * inherit a map of the kernel space.
+	 */
+	addr = PAGE_OFFSET;
+	entry = pgd_index(PAGE_OFFSET);
+	p_swapper_pg_dir = pgd_offset_k(0) + entry;
+
+	while (addr <= META_MEMORY_LIMIT) {
+		unsigned long pgd_entry;
+		/* copy over the current MMU value */
+		pgd_entry = mmu_read_first_level_page(addr);
+		pgd_val(*p_swapper_pg_dir) = pgd_entry;
+
+		p_swapper_pg_dir++;
+		addr += PGDIR_SIZE;
+		entry++;
+	}
+}
diff --git a/arch/metag/mm/mmu-meta2.c b/arch/metag/mm/mmu-meta2.c
new file mode 100644
index 000000000000..81dcbb0bba34
--- /dev/null
+++ b/arch/metag/mm/mmu-meta2.c
@@ -0,0 +1,207 @@
+/*
+ * Copyright (C) 2008,2009,2010,2011 Imagination Technologies Ltd.
+ *
+ * Meta 2 enhanced mode MMU handling code.
+ *
+ */
+
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/io.h>
+#include <linux/bootmem.h>
+#include <linux/syscore_ops.h>
+
+#include <asm/mmu.h>
+#include <asm/mmu_context.h>
+
+unsigned long mmu_read_first_level_page(unsigned long vaddr)
+{
+	unsigned int cpu = hard_processor_id();
+	unsigned long offset, linear_base, linear_limit;
+	unsigned int phys0;
+	pgd_t *pgd, entry;
+
+	if (is_global_space(vaddr))
+		vaddr &= ~0x80000000;
+
+	offset = vaddr >> PGDIR_SHIFT;
+
+	phys0 = metag_in32(mmu_phys0_addr(cpu));
+
+	/* Top bit of linear base is always zero. */
+	linear_base = (phys0 >> PGDIR_SHIFT) & 0x1ff;
+
+	/* Limit in the range 0 (4MB) to 9 (2GB). */
+	linear_limit = 1 << ((phys0 >> 8) & 0xf);
+	linear_limit += linear_base;
+
+	/*
+	 * If offset is below linear base or above the limit then no
+	 * mapping exists.
+	 */
+	if (offset < linear_base || offset > linear_limit)
+		return 0;
+
+	offset -= linear_base;
+	pgd = (pgd_t *)mmu_get_base();
+	entry = pgd[offset];
+
+	return pgd_val(entry);
+}
+
+unsigned long mmu_read_second_level_page(unsigned long vaddr)
+{
+	return __builtin_meta2_cacherd((void *)(vaddr & PAGE_MASK));
+}
+
+unsigned long mmu_get_base(void)
+{
+	unsigned int cpu = hard_processor_id();
+	unsigned long stride;
+
+	stride = cpu * LINSYSMEMTnX_STRIDE;
+
+	/*
+	 * Bits 18:2 of the MMCU_TnLocal_TABLE_PHYS1 register should be
+	 * used as an offset to the start of the top-level pgd table.
+	 */
+	stride += (metag_in32(mmu_phys1_addr(cpu)) & 0x7fffc);
+
+	if (is_global_space(PAGE_OFFSET))
+		stride += LINSYSMEMTXG_OFFSET;
+
+	return LINSYSMEMT0L_BASE + stride;
+}
+
+#define FIRST_LEVEL_MASK	0xffffffc0
+#define SECOND_LEVEL_MASK	0xfffff000
+#define SECOND_LEVEL_ALIGN	64
+
+static void repriv_mmu_tables(void)
+{
+	unsigned long phys0_addr;
+	unsigned int g;
+
+	/*
+	 * Check that all the mmu table regions are priv protected, and if not
+	 * fix them and emit a warning. If we left them without priv protection
+	 * then userland processes would have access to a 2M window into
+	 * physical memory near where the page tables are.
+	 */
+	phys0_addr = MMCU_T0LOCAL_TABLE_PHYS0;
+	for (g = 0; g < 2; ++g) {
+		unsigned int t, phys0;
+		unsigned long flags;
+		for (t = 0; t < 4; ++t) {
+			__global_lock2(flags);
+			phys0 = metag_in32(phys0_addr);
+			if ((phys0 & _PAGE_PRESENT) && !(phys0 & _PAGE_PRIV)) {
+				pr_warn("Fixing priv protection on T%d %s MMU table region\n",
+					t,
+					g ? "global" : "local");
+				phys0 |= _PAGE_PRIV;
+				metag_out32(phys0, phys0_addr);
+			}
+			__global_unlock2(flags);
+
+			phys0_addr += MMCU_TnX_TABLE_PHYSX_STRIDE;
+		}
+
+		phys0_addr += MMCU_TXG_TABLE_PHYSX_OFFSET
+			    - 4*MMCU_TnX_TABLE_PHYSX_STRIDE;
+	}
+}
+
+#ifdef CONFIG_METAG_SUSPEND_MEM
+static void mmu_resume(void)
+{
+	/*
+	 * If a full suspend to RAM has happened then the original bad MMU table
+	 * priv may have been restored, so repriv them again.
+	 */
+	repriv_mmu_tables();
+}
+#else
+#define mmu_resume NULL
+#endif	/* CONFIG_METAG_SUSPEND_MEM */
+
+static struct syscore_ops mmu_syscore_ops = {
+	.resume  = mmu_resume,
+};
+
+void __init mmu_init(unsigned long mem_end)
+{
+	unsigned long entry, addr;
+	pgd_t *p_swapper_pg_dir;
+#ifdef CONFIG_KERNEL_4M_PAGES
+	unsigned long mem_size = mem_end - PAGE_OFFSET;
+	unsigned int pages = DIV_ROUND_UP(mem_size, 1 << 22);
+	unsigned int second_level_entry = 0;
+	unsigned long *second_level_table;
+#endif
+
+	/*
+	 * Now copy over any MMU pgd entries already in the mmu page tables
+	 * over to our root init process (swapper_pg_dir) map.  This map is
+	 * then inherited by all other processes, which means all processes
+	 * inherit a map of the kernel space.
+	 */
+	addr = META_MEMORY_BASE;
+	entry = pgd_index(META_MEMORY_BASE);
+	p_swapper_pg_dir = pgd_offset_k(0) + entry;
+
+	while (entry < (PTRS_PER_PGD - pgd_index(META_MEMORY_BASE))) {
+		unsigned long pgd_entry;
+		/* copy over the current MMU value */
+		pgd_entry = mmu_read_first_level_page(addr);
+		pgd_val(*p_swapper_pg_dir) = pgd_entry;
+
+		p_swapper_pg_dir++;
+		addr += PGDIR_SIZE;
+		entry++;
+	}
+
+#ifdef CONFIG_KERNEL_4M_PAGES
+	/*
+	 * At this point we can also map the kernel with 4MB pages to
+	 * reduce TLB pressure.
+	 */
+	second_level_table = alloc_bootmem_pages(SECOND_LEVEL_ALIGN * pages);
+
+	addr = PAGE_OFFSET;
+	entry = pgd_index(PAGE_OFFSET);
+	p_swapper_pg_dir = pgd_offset_k(0) + entry;
+
+	while (pages > 0) {
+		unsigned long phys_addr, second_level_phys;
+		pte_t *pte = (pte_t *)&second_level_table[second_level_entry];
+
+		phys_addr = __pa(addr);
+
+		second_level_phys = __pa(pte);
+
+		pgd_val(*p_swapper_pg_dir) = ((second_level_phys &
+					       FIRST_LEVEL_MASK) |
+					      _PAGE_SZ_4M |
+					      _PAGE_PRESENT);
+
+		pte_val(*pte) = ((phys_addr & SECOND_LEVEL_MASK) |
+				 _PAGE_PRESENT | _PAGE_DIRTY |
+				 _PAGE_ACCESSED | _PAGE_WRITE |
+				 _PAGE_CACHEABLE | _PAGE_KERNEL);
+
+		p_swapper_pg_dir++;
+		addr += PGDIR_SIZE;
+		/* Second level pages must be 64byte aligned. */
+		second_level_entry += (SECOND_LEVEL_ALIGN /
+				       sizeof(unsigned long));
+		pages--;
+	}
+	load_pgd(swapper_pg_dir, hard_processor_id());
+	flush_tlb_all();
+#endif
+
+	repriv_mmu_tables();
+	register_syscore_ops(&mmu_syscore_ops);
+}