Linux-2.6.12-rc2v2.6.12-rc2

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!

8 files changed, 1521 insertions, 0 deletions

diff --git a/arch/arm26/mm/Makefile b/arch/arm26/mm/Makefile
new file mode 100644
index 000000000000..a8fb166d5c6d
--- /dev/null
+++ b/arch/arm26/mm/Makefile
@@ -0,0 +1,6 @@
+#
+# Makefile for the linux arm26-specific parts of the memory manager.
+#
+
+obj-y		:= init.o extable.o proc-funcs.o memc.o fault.o \
+		   small_page.o
diff --git a/arch/arm26/mm/extable.c b/arch/arm26/mm/extable.c
new file mode 100644
index 000000000000..2d9f5b5a78d6
--- /dev/null
+++ b/arch/arm26/mm/extable.c
@@ -0,0 +1,25 @@
+/*
+ *  linux/arch/arm26/mm/extable.c
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <asm/uaccess.h>
+
+int fixup_exception(struct pt_regs *regs)
+{
+        const struct exception_table_entry *fixup;
+
+        fixup = search_exception_tables(instruction_pointer(regs));
+
+	/*
+	 * The kernel runs in SVC mode - make sure we keep running in SVC mode
+	 * by frobbing the PSR appropriately (PSR and PC are in the same reg.
+	 * on ARM26)
+	 */
+        if (fixup)
+                regs->ARM_pc = fixup->fixup | PSR_I_BIT | MODE_SVC26;
+
+        return fixup != NULL;
+}
+
diff --git a/arch/arm26/mm/fault.c b/arch/arm26/mm/fault.c
new file mode 100644
index 000000000000..dacca8bb7744
--- /dev/null
+++ b/arch/arm26/mm/fault.c
@@ -0,0 +1,318 @@
+/*
+ *  linux/arch/arm26/mm/fault.c
+ *
+ *  Copyright (C) 1995  Linus Torvalds
+ *  Modifications for ARM processor (c) 1995-2001 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/interrupt.h>
+#include <linux/proc_fs.h>
+#include <linux/init.h>
+
+#include <asm/system.h>
+#include <asm/pgtable.h>
+#include <asm/uaccess.h> //FIXME this header may be bogusly included
+
+#include "fault.h"
+
+#define FAULT_CODE_LDRSTRPOST   0x80
+#define FAULT_CODE_LDRSTRPRE    0x40
+#define FAULT_CODE_LDRSTRREG    0x20
+#define FAULT_CODE_LDMSTM       0x10
+#define FAULT_CODE_LDCSTC       0x08
+#define FAULT_CODE_PREFETCH     0x04
+#define FAULT_CODE_WRITE        0x02
+#define FAULT_CODE_FORCECOW     0x01
+
+#define DO_COW(m)               ((m) & (FAULT_CODE_WRITE|FAULT_CODE_FORCECOW))
+#define READ_FAULT(m)           (!((m) & FAULT_CODE_WRITE))
+#define DEBUG
+/*
+ * This is useful to dump out the page tables associated with
+ * 'addr' in mm 'mm'.
+ */
+void show_pte(struct mm_struct *mm, unsigned long addr)
+{
+	pgd_t *pgd;
+
+	if (!mm)
+		mm = &init_mm;
+
+	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
+	pgd = pgd_offset(mm, addr);
+	printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
+
+	do {
+		pmd_t *pmd;
+		pte_t *pte;
+
+		pmd = pmd_offset(pgd, addr);
+
+		if (pmd_none(*pmd))
+			break;
+
+		if (pmd_bad(*pmd)) {
+			printk("(bad)");
+			break;
+		}
+
+		/* We must not map this if we have highmem enabled */
+		/* FIXME */
+		pte = pte_offset_map(pmd, addr);
+		printk(", *pte=%08lx", pte_val(*pte));
+		pte_unmap(pte);
+	} while(0);
+
+	printk("\n");
+}
+
+/*
+ * Oops.  The kernel tried to access some page that wasn't present.
+ */
+static void
+__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
+		  struct pt_regs *regs)
+{
+	/*
+         * Are we prepared to handle this kernel fault?
+         */
+        if (fixup_exception(regs))
+                return;
+
+	/*
+	 * No handler, we'll have to terminate things with extreme prejudice.
+	 */
+	bust_spinlocks(1);
+	printk(KERN_ALERT
+		"Unable to handle kernel %s at virtual address %08lx\n",
+		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
+		"paging request", addr);
+
+	show_pte(mm, addr);
+	die("Oops", regs, fsr);
+	bust_spinlocks(0);
+	do_exit(SIGKILL);
+}
+
+/*
+ * Something tried to access memory that isn't in our memory map..
+ * User mode accesses just cause a SIGSEGV
+ */
+static void
+__do_user_fault(struct task_struct *tsk, unsigned long addr,
+		unsigned int fsr, int code, struct pt_regs *regs)
+{
+	struct siginfo si;
+
+#ifdef CONFIG_DEBUG_USER
+	printk("%s: unhandled page fault at 0x%08lx, code 0x%03x\n",
+	       tsk->comm, addr, fsr);
+	show_pte(tsk->mm, addr);
+	show_regs(regs);
+	//dump_backtrace(regs, tsk); // FIXME ARM32 dropped this - why?
+	while(1); //FIXME - hack to stop debug going nutso
+#endif
+
+	tsk->thread.address = addr;
+	tsk->thread.error_code = fsr;
+	tsk->thread.trap_no = 14;
+	si.si_signo = SIGSEGV;
+	si.si_errno = 0;
+	si.si_code = code;
+	si.si_addr = (void *)addr;
+	force_sig_info(SIGSEGV, &si, tsk);
+}
+
+static int
+__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
+		struct task_struct *tsk)
+{
+	struct vm_area_struct *vma;
+	int fault, mask;
+
+	vma = find_vma(mm, addr);
+	fault = -2; /* bad map area */
+	if (!vma)
+		goto out;
+	if (vma->vm_start > addr)
+		goto check_stack;
+
+	/*
+	 * Ok, we have a good vm_area for this
+	 * memory access, so we can handle it.
+	 */
+good_area:
+	if (READ_FAULT(fsr)) /* read? */
+		mask = VM_READ|VM_EXEC;
+	else
+		mask = VM_WRITE;
+
+	fault = -1; /* bad access type */
+	if (!(vma->vm_flags & mask))
+		goto out;
+
+	/*
+	 * If for any reason at all we couldn't handle
+	 * the fault, make sure we exit gracefully rather
+	 * than endlessly redo the fault.
+	 */
+survive:
+	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, DO_COW(fsr));
+
+	/*
+	 * Handle the "normal" cases first - successful and sigbus
+	 */
+	switch (fault) {
+	case 2:
+		tsk->maj_flt++;
+		return fault;
+	case 1:
+		tsk->min_flt++;
+	case 0:
+		return fault;
+	}
+
+	fault = -3; /* out of memory */
+	if (tsk->pid != 1)
+		goto out;
+
+	/*
+	 * If we are out of memory for pid1,
+	 * sleep for a while and retry
+	 */
+	yield();
+	goto survive;
+
+check_stack:
+	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
+		goto good_area;
+out:
+	return fault;
+}
+
+int do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
+{
+	struct task_struct *tsk;
+	struct mm_struct *mm;
+	int fault;
+
+	tsk = current;
+	mm  = tsk->mm;
+
+	/*
+	 * If we're in an interrupt or have no user
+	 * context, we must not take the fault..
+	 */
+	if (in_interrupt() || !mm)
+		goto no_context;
+
+	down_read(&mm->mmap_sem);
+	fault = __do_page_fault(mm, addr, fsr, tsk);
+	up_read(&mm->mmap_sem);
+
+	/*
+	 * Handle the "normal" case first
+	 */
+	if (fault > 0)
+		return 0;
+
+	/*
+	 * We had some memory, but were unable to
+	 * successfully fix up this page fault.
+	 */
+	if (fault == 0){
+		goto do_sigbus;
+	}
+
+	/*
+	 * If we are in kernel mode at this point, we
+	 * have no context to handle this fault with.
+         * FIXME - is this test right?
+	 */
+	if (!user_mode(regs)){
+		goto no_context;
+	}
+
+	if (fault == -3) {
+		/*
+		 * We ran out of memory, or some other thing happened to
+		 * us that made us unable to handle the page fault gracefully.
+		 */
+		printk("VM: killing process %s\n", tsk->comm);
+		do_exit(SIGKILL);
+	}
+	else{
+		__do_user_fault(tsk, addr, fsr, fault == -1 ? SEGV_ACCERR : SEGV_MAPERR, regs);
+	}
+
+	return 0;
+
+
+/*
+ * We ran out of memory, or some other thing happened to us that made
+ * us unable to handle the page fault gracefully.
+ */
+do_sigbus:
+	/*
+	 * Send a sigbus, regardless of whether we were in kernel
+	 * or user mode.
+	 */
+	tsk->thread.address = addr;  //FIXME - need other bits setting?
+	tsk->thread.error_code = fsr;
+	tsk->thread.trap_no = 14;
+	force_sig(SIGBUS, tsk);
+#ifdef CONFIG_DEBUG_USER
+	printk(KERN_DEBUG "%s: sigbus at 0x%08lx, pc=0x%08lx\n",
+		current->comm, addr, instruction_pointer(regs));
+#endif
+
+	/* Kernel mode? Handle exceptions or die */
+	if (user_mode(regs))
+		return 0;
+
+no_context:
+	__do_kernel_fault(mm, addr, fsr, regs);
+	return 0;
+}
+
+/*
+ * Handle a data abort.  Note that we have to handle a range of addresses
+ * on ARM2/3 for ldm.  If both pages are zero-mapped, then we have to force
+ * a copy-on-write.  However, on the second page, we always force COW.
+ */
+asmlinkage void
+do_DataAbort(unsigned long min_addr, unsigned long max_addr, int mode, struct pt_regs *regs)
+{
+        do_page_fault(min_addr, mode, regs);
+
+        if ((min_addr ^ max_addr) >> PAGE_SHIFT){
+               do_page_fault(max_addr, mode | FAULT_CODE_FORCECOW, regs);
+	}
+}
+
+asmlinkage int
+do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
+{
+#if 0
+        if (the memc mapping for this page exists) {
+                printk ("Page in, but got abort (undefined instruction?)\n");
+                return 0;
+        }
+#endif
+        do_page_fault(addr, FAULT_CODE_PREFETCH, regs);
+        return 1;
+}
+
diff --git a/arch/arm26/mm/fault.h b/arch/arm26/mm/fault.h
new file mode 100644
index 000000000000..4442d00d86ac
--- /dev/null
+++ b/arch/arm26/mm/fault.h
@@ -0,0 +1,5 @@
+void show_pte(struct mm_struct *mm, unsigned long addr);
+
+int do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
+
+unsigned long search_extable(unsigned long addr); //FIXME - is it right?
diff --git a/arch/arm26/mm/init.c b/arch/arm26/mm/init.c
new file mode 100644
index 000000000000..1f09a9d0fb83
--- /dev/null
+++ b/arch/arm26/mm/init.c
@@ -0,0 +1,412 @@
+/*
+ *  linux/arch/arm26/mm/init.c
+ *
+ *  Copyright (C) 1995-2002 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/smp.h>
+#include <linux/init.h>
+#include <linux/initrd.h>
+#include <linux/bootmem.h>
+#include <linux/blkdev.h>
+
+#include <asm/segment.h>
+#include <asm/mach-types.h>
+#include <asm/dma.h>
+#include <asm/hardware.h>
+#include <asm/setup.h>
+#include <asm/tlb.h>
+
+#include <asm/map.h>
+
+
+#define TABLE_SIZE	PTRS_PER_PTE * sizeof(pte_t))
+
+struct mmu_gather mmu_gathers[NR_CPUS];
+
+extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
+extern char _stext, _text, _etext, _end, __init_begin, __init_end;
+#ifdef CONFIG_XIP_KERNEL
+extern char _endtext, _sdata;
+#endif
+extern unsigned long phys_initrd_start;
+extern unsigned long phys_initrd_size;
+
+/*
+ * The sole use of this is to pass memory configuration
+ * data from paging_init to mem_init.
+ */
+static struct meminfo meminfo __initdata = { 0, };
+
+/*
+ * empty_zero_page is a special page that is used for
+ * zero-initialized data and COW.
+ */
+struct page *empty_zero_page;
+
+void show_mem(void)
+{
+	int free = 0, total = 0, reserved = 0;
+	int shared = 0, cached = 0, slab = 0;
+	struct page *page, *end;
+
+	printk("Mem-info:\n");
+	show_free_areas();
+	printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
+
+
+	page = NODE_MEM_MAP(0);
+	end  = page + NODE_DATA(0)->node_spanned_pages;
+
+	do {
+		total++;
+		if (PageReserved(page))
+			reserved++;
+		else if (PageSwapCache(page))
+			cached++;
+		else if (PageSlab(page))
+			slab++;
+		else if (!page_count(page))
+			free++;
+		else
+			shared += page_count(page) - 1;
+		page++;
+	} while (page < end);
+
+	printk("%d pages of RAM\n", total);
+	printk("%d free pages\n", free);
+	printk("%d reserved pages\n", reserved);
+	printk("%d slab pages\n", slab);
+	printk("%d pages shared\n", shared);
+	printk("%d pages swap cached\n", cached);
+}
+
+struct node_info {
+	unsigned int start;
+	unsigned int end;
+	int bootmap_pages;
+};
+
+#define PFN_DOWN(x)	((x) >> PAGE_SHIFT)
+#define PFN_UP(x)	(PAGE_ALIGN(x) >> PAGE_SHIFT)
+#define PFN_SIZE(x)	((x) >> PAGE_SHIFT)
+#define PFN_RANGE(s,e)	PFN_SIZE(PAGE_ALIGN((unsigned long)(e)) - \
+				(((unsigned long)(s)) & PAGE_MASK))
+
+/*
+ * FIXME: We really want to avoid allocating the bootmap bitmap
+ * over the top of the initrd.  Hopefully, this is located towards
+ * the start of a bank, so if we allocate the bootmap bitmap at
+ * the end, we won't clash.
+ */
+static unsigned int __init
+find_bootmap_pfn(struct meminfo *mi, unsigned int bootmap_pages)
+{
+	unsigned int start_pfn, bootmap_pfn;
+	unsigned int start, end;
+
+	start_pfn   = PFN_UP((unsigned long)&_end);
+	bootmap_pfn = 0;
+
+	/* ARM26 machines only have one node */
+	if (mi->bank->node != 0)
+		BUG();
+
+	start = PFN_UP(mi->bank->start);
+	end   = PFN_DOWN(mi->bank->size + mi->bank->start);
+
+	if (start < start_pfn)
+		start = start_pfn;
+
+	if (end <= start)
+		BUG();
+
+	if (end - start >= bootmap_pages) 
+		bootmap_pfn = start;
+	else
+		BUG();
+
+	return bootmap_pfn;
+}
+
+/*
+ * Scan the memory info structure and pull out:
+ *  - the end of memory
+ *  - the number of nodes
+ *  - the pfn range of each node
+ *  - the number of bootmem bitmap pages
+ */
+static void __init
+find_memend_and_nodes(struct meminfo *mi, struct node_info *np)
+{
+	unsigned int memend_pfn = 0;
+
+	nodes_clear(node_online_map);
+	node_set_online(0);
+
+	np->bootmap_pages = 0;
+
+	if (mi->bank->size == 0) {
+		BUG();
+	}
+
+	/*
+	 * Get the start and end pfns for this bank
+	 */
+	np->start = PFN_UP(mi->bank->start);
+	np->end   = PFN_DOWN(mi->bank->start + mi->bank->size);
+
+	if (memend_pfn < np->end)
+		memend_pfn = np->end;
+
+	/*
+	 * Calculate the number of pages we require to
+	 * store the bootmem bitmaps.
+	 */
+	np->bootmap_pages = bootmem_bootmap_pages(np->end - np->start);
+
+	/*
+	 * This doesn't seem to be used by the Linux memory
+	 * manager any more.  If we can get rid of it, we
+	 * also get rid of some of the stuff above as well.
+	 */
+	max_low_pfn = memend_pfn - PFN_DOWN(PHYS_OFFSET);
+	max_pfn = memend_pfn - PFN_DOWN(PHYS_OFFSET);
+	mi->end = memend_pfn << PAGE_SHIFT;
+
+}
+
+/*
+ * Initialise the bootmem allocator for all nodes.  This is called
+ * early during the architecture specific initialisation.
+ */
+void __init bootmem_init(struct meminfo *mi)
+{
+	struct node_info node_info;
+	unsigned int bootmap_pfn;
+	pg_data_t *pgdat = NODE_DATA(0);
+
+	find_memend_and_nodes(mi, &node_info);
+
+	bootmap_pfn   = find_bootmap_pfn(mi, node_info.bootmap_pages);
+
+	/*
+	 * Note that node 0 must always have some pages.
+	 */
+	if (node_info.end == 0)
+		BUG();
+
+	/*
+	 * Initialise the bootmem allocator.
+	 */
+	init_bootmem_node(pgdat, bootmap_pfn, node_info.start, node_info.end);
+
+ 	/*
+	 * Register all available RAM in this node with the bootmem allocator. 
+	 */
+	free_bootmem_node(pgdat, mi->bank->start, mi->bank->size);
+
+        /*
+         * Register the kernel text and data with bootmem.
+         * Note: with XIP we dont register .text since
+         * its in ROM.
+         */
+#ifdef CONFIG_XIP_KERNEL
+        reserve_bootmem_node(pgdat, __pa(&_sdata), &_end - &_sdata);
+#else
+        reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
+#endif
+
+        /*
+         * And don't forget to reserve the allocator bitmap,
+         * which will be freed later.
+         */
+        reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT,
+                             node_info.bootmap_pages << PAGE_SHIFT);
+
+        /*
+         * These should likewise go elsewhere.  They pre-reserve
+         * the screen memory region at the start of main system
+         * memory. FIXME - screen RAM is not 512K!
+         */
+        reserve_bootmem_node(pgdat, 0x02000000, 0x00080000);
+
+#ifdef CONFIG_BLK_DEV_INITRD
+        initrd_start = phys_initrd_start;
+        initrd_end = initrd_start + phys_initrd_size;
+
+        /* Achimedes machines only have one node, so initrd is in node 0 */
+#ifdef CONFIG_XIP_KERNEL
+	/* Only reserve initrd space if it is in RAM */
+        if(initrd_start && initrd_start < 0x03000000){
+#else
+        if(initrd_start){
+#endif
+                reserve_bootmem_node(pgdat, __pa(initrd_start),
+                                             initrd_end - initrd_start);
+	}
+#endif   /* CONFIG_BLK_DEV_INITRD */
+
+
+}
+
+/*
+ * paging_init() sets up the page tables, initialises the zone memory
+ * maps, and sets up the zero page, bad page and bad page tables.
+ */
+void __init paging_init(struct meminfo *mi)
+{
+	void *zero_page;
+	unsigned long zone_size[MAX_NR_ZONES];
+        unsigned long zhole_size[MAX_NR_ZONES];
+        struct bootmem_data *bdata;
+        pg_data_t *pgdat;
+	int i;
+
+	memcpy(&meminfo, mi, sizeof(meminfo));
+
+	/*
+	 * allocate the zero page.  Note that we count on this going ok.
+	 */
+	zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
+
+	/*
+	 * initialise the page tables.
+	 */
+	memtable_init(mi);
+	flush_tlb_all();
+
+	/*
+	 * initialise the zones in node 0 (archimedes have only 1 node)
+	 */
+
+	for (i = 0; i < MAX_NR_ZONES; i++) {
+		zone_size[i]  = 0;
+		zhole_size[i] = 0;
+	}
+
+	pgdat = NODE_DATA(0);
+	bdata = pgdat->bdata;
+	zone_size[0] = bdata->node_low_pfn -
+			(bdata->node_boot_start >> PAGE_SHIFT);
+	if (!zone_size[0])
+		BUG();
+	pgdat->node_mem_map = NULL;
+	free_area_init_node(0, pgdat, zone_size,
+			bdata->node_boot_start >> PAGE_SHIFT, zhole_size);
+
+	/*
+	 * finish off the bad pages once
+	 * the mem_map is initialised
+	 */
+	memzero(zero_page, PAGE_SIZE);
+	empty_zero_page = virt_to_page(zero_page);
+}
+
+static inline void free_area(unsigned long addr, unsigned long end, char *s)
+{
+	unsigned int size = (end - addr) >> 10;
+
+	for (; addr < end; addr += PAGE_SIZE) {
+		struct page *page = virt_to_page(addr);
+		ClearPageReserved(page);
+		set_page_count(page, 1);
+		free_page(addr);
+		totalram_pages++;
+	}
+
+	if (size && s)
+		printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
+}
+
+/*
+ * mem_init() marks the free areas in the mem_map and tells us how much
+ * memory is free.  This is done after various parts of the system have
+ * claimed their memory after the kernel image.
+ */
+void __init mem_init(void)
+{
+	unsigned int codepages, datapages, initpages;
+	pg_data_t *pgdat = NODE_DATA(0);
+	extern int sysctl_overcommit_memory;
+
+
+	/* Note: data pages includes BSS */
+#ifdef CONFIG_XIP_KERNEL
+	codepages = &_endtext - &_text;
+	datapages = &_end - &_sdata;
+#else
+	codepages = &_etext - &_text;
+	datapages = &_end - &_etext;
+#endif
+	initpages = &__init_end - &__init_begin;
+
+	high_memory = (void *)__va(meminfo.end);
+	max_mapnr   = virt_to_page(high_memory) - mem_map;
+
+	/* this will put all unused low memory onto the freelists */
+	if (pgdat->node_spanned_pages != 0)
+		totalram_pages += free_all_bootmem_node(pgdat);
+
+	num_physpages = meminfo.bank[0].size >> PAGE_SHIFT;
+
+	printk(KERN_INFO "Memory: %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
+	printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
+		"%dK data, %dK init)\n",
+		(unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
+		codepages >> 10, datapages >> 10, initpages >> 10);
+
+	/*
+	 * Turn on overcommit on tiny machines
+	 */
+	if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
+		sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
+		printk("Turning on overcommit\n");
+	}
+}
+
+void free_initmem(void){
+#ifndef CONFIG_XIP_KERNEL
+	free_area((unsigned long)(&__init_begin),
+		  (unsigned long)(&__init_end),
+		  "init");
+#endif
+}
+
+#ifdef CONFIG_BLK_DEV_INITRD
+
+static int keep_initrd;
+
+void free_initrd_mem(unsigned long start, unsigned long end)
+{
+#ifdef CONFIG_XIP_KERNEL
+	/* Only bin initrd if it is in RAM... */
+	if(!keep_initrd && start < 0x03000000)
+#else
+	if (!keep_initrd)
+#endif
+		free_area(start, end, "initrd");
+}
+
+static int __init keepinitrd_setup(char *__unused)
+{
+	keep_initrd = 1;
+	return 1;
+}
+
+__setup("keepinitrd", keepinitrd_setup);
+#endif
diff --git a/arch/arm26/mm/memc.c b/arch/arm26/mm/memc.c
new file mode 100644
index 000000000000..8e8a2bb2487d
--- /dev/null
+++ b/arch/arm26/mm/memc.c
@@ -0,0 +1,202 @@
+/*
+ *  linux/arch/arm26/mm/memc.c
+ *
+ *  Copyright (C) 1998-2000 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ *  Page table sludge for older ARM processor architectures.
+ */
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/page.h>
+#include <asm/memory.h>
+#include <asm/hardware.h>
+
+#include <asm/map.h>
+
+#define MEMC_TABLE_SIZE (256*sizeof(unsigned long))
+
+kmem_cache_t *pte_cache, *pgd_cache;
+int page_nr;
+
+/*
+ * Allocate space for a page table and a MEMC table.
+ * Note that we place the MEMC
+ * table before the page directory.  This means we can
+ * easily get to both tightly-associated data structures
+ * with a single pointer.
+ */
+static inline pgd_t *alloc_pgd_table(void)
+{
+	void *pg2k = kmem_cache_alloc(pgd_cache, GFP_KERNEL);
+
+	if (pg2k)
+		pg2k += MEMC_TABLE_SIZE;
+
+	return (pgd_t *)pg2k;
+}
+
+/*
+ * Free a page table. this function is the counterpart to get_pgd_slow
+ * below, not alloc_pgd_table above.
+ */
+void free_pgd_slow(pgd_t *pgd)
+{
+	unsigned long tbl = (unsigned long)pgd;
+
+	tbl -= MEMC_TABLE_SIZE;
+
+	kmem_cache_free(pgd_cache, (void *)tbl);
+}
+
+/*
+ * Allocate a new pgd and fill it in ready for use
+ *
+ * A new tasks pgd is completely empty (all pages !present) except for:
+ *
+ * o The machine vectors at virtual address 0x0
+ * o The vmalloc region at the top of address space
+ *
+ */
+#define FIRST_KERNEL_PGD_NR     (FIRST_USER_PGD_NR + USER_PTRS_PER_PGD)
+
+pgd_t *get_pgd_slow(struct mm_struct *mm)
+{
+	pgd_t *new_pgd, *init_pgd;
+	pmd_t *new_pmd, *init_pmd;
+	pte_t *new_pte, *init_pte;
+
+	new_pgd = alloc_pgd_table();
+	if (!new_pgd)
+		goto no_pgd;
+
+	/*
+	 * This lock is here just to satisfy pmd_alloc and pte_lock
+         * FIXME: I bet we could avoid taking it pretty much altogether
+	 */
+	spin_lock(&mm->page_table_lock);
+
+	/*
+	 * On ARM, first page must always be allocated since it contains
+	 * the machine vectors.
+	 */
+	new_pmd = pmd_alloc(mm, new_pgd, 0);
+	if (!new_pmd)
+		goto no_pmd;
+
+	new_pte = pte_alloc_kernel(mm, new_pmd, 0);
+	if (!new_pte)
+		goto no_pte;
+
+	init_pgd = pgd_offset(&init_mm, 0);
+	init_pmd = pmd_offset(init_pgd, 0);
+	init_pte = pte_offset(init_pmd, 0);
+
+	set_pte(new_pte, *init_pte);
+
+	/*
+	 * the page table entries are zeroed
+	 * when the table is created. (see the cache_ctor functions below)
+	 * Now we need to plonk the kernel (vmalloc) area at the end of
+	 * the address space. We copy this from the init thread, just like
+	 * the init_pte we copied above...
+	 */
+	memcpy(new_pgd + FIRST_KERNEL_PGD_NR, init_pgd + FIRST_KERNEL_PGD_NR,
+		(PTRS_PER_PGD - FIRST_KERNEL_PGD_NR) * sizeof(pgd_t));
+
+	spin_unlock(&mm->page_table_lock);
+
+	/* update MEMC tables */
+	cpu_memc_update_all(new_pgd);
+	return new_pgd;
+
+no_pte:
+	spin_unlock(&mm->page_table_lock);
+	pmd_free(new_pmd);
+	free_pgd_slow(new_pgd);
+	return NULL;
+
+no_pmd:
+	spin_unlock(&mm->page_table_lock);
+	free_pgd_slow(new_pgd);
+	return NULL;
+
+no_pgd:
+	return NULL;
+}
+
+/*
+ * No special code is required here.
+ */
+void setup_mm_for_reboot(char mode)
+{
+}
+
+/*
+ * This contains the code to setup the memory map on an ARM2/ARM250/ARM3
+ *  o swapper_pg_dir = 0x0207d000
+ *  o kernel proper starts at 0x0208000
+ *  o create (allocate) a pte to contain the machine vectors
+ *  o populate the pte (points to 0x02078000) (FIXME - is it zeroed?)
+ *  o populate the init tasks page directory (pgd) with the new pte
+ *  o zero the rest of the init tasks pgdir (FIXME - what about vmalloc?!)
+ */
+void __init memtable_init(struct meminfo *mi)
+{
+	pte_t *pte;
+	int i;
+
+	page_nr = max_low_pfn;
+
+	pte = alloc_bootmem_low_pages(PTRS_PER_PTE * sizeof(pte_t));
+	pte[0] = mk_pte_phys(PAGE_OFFSET + SCREEN_SIZE, PAGE_READONLY);
+	pmd_populate(&init_mm, pmd_offset(swapper_pg_dir, 0), pte);
+
+	for (i = 1; i < PTRS_PER_PGD; i++)
+		pgd_val(swapper_pg_dir[i]) = 0;
+}
+
+void __init iotable_init(struct map_desc *io_desc)
+{
+	/* nothing to do */
+}
+
+/*
+ * We never have holes in the memmap
+ */
+void __init create_memmap_holes(struct meminfo *mi)
+{
+}
+
+static void pte_cache_ctor(void *pte, kmem_cache_t *cache, unsigned long flags)
+{
+	memzero(pte, sizeof(pte_t) * PTRS_PER_PTE);
+}
+
+static void pgd_cache_ctor(void *pgd, kmem_cache_t *cache, unsigned long flags)
+{
+	memzero(pgd + MEMC_TABLE_SIZE, USER_PTRS_PER_PGD * sizeof(pgd_t));
+}
+
+void __init pgtable_cache_init(void)
+{
+	pte_cache = kmem_cache_create("pte-cache",
+				sizeof(pte_t) * PTRS_PER_PTE,
+				0, 0, pte_cache_ctor, NULL);
+	if (!pte_cache)
+		BUG();
+
+	pgd_cache = kmem_cache_create("pgd-cache", MEMC_TABLE_SIZE +
+				sizeof(pgd_t) * PTRS_PER_PGD,
+				0, 0, pgd_cache_ctor, NULL);
+	if (!pgd_cache)
+		BUG();
+}
diff --git a/arch/arm26/mm/proc-funcs.S b/arch/arm26/mm/proc-funcs.S
new file mode 100644
index 000000000000..c3d4cd3f457e
--- /dev/null
+++ b/arch/arm26/mm/proc-funcs.S
@@ -0,0 +1,359 @@
+/*
+ *  linux/arch/arm26/mm/proc-arm2,3.S
+ *
+ *  Copyright (C) 1997-1999 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ *  MMU functions for ARM2,3
+ *
+ *  These are the low level assembler for performing cache
+ *  and memory functions on ARM2, ARM250 and ARM3 processors.
+ */
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+#include <asm/asm_offsets.h>
+#include <asm/procinfo.h>
+#include <asm/ptrace.h>
+
+/*
+ * MEMC workhorse code.  It's both a horse which things it's a pig.
+ */
+/*
+ * Function: cpu_memc_update_entry(pgd_t *pgd, unsigned long phys_pte, unsigned long addr)
+ * Params  : pgd	Page tables/MEMC mapping
+ *         : phys_pte	physical address, or PTE
+ *         : addr	virtual address
+ */
+ENTRY(cpu_memc_update_entry)
+		tst	r1, #PAGE_PRESENT		@ is the page present
+		orreq	r1, r1, #PAGE_OLD | PAGE_CLEAN
+		moveq	r2, #0x01f00000
+		mov	r3, r1, lsr #13			@ convert to physical page nr
+		and	r3, r3, #0x3fc
+		adr	ip, memc_phys_table_32
+		ldr	r3, [ip, r3]
+		tst	r1, #PAGE_OLD | PAGE_NOT_USER
+		biceq	r3, r3, #0x200
+		tsteq	r1, #PAGE_READONLY | PAGE_CLEAN
+		biceq	r3, r3, #0x300
+		mov	r2, r2, lsr #15			@ virtual -> nr
+		orr	r3, r3, r2, lsl #15
+		and	r2, r2, #0x300
+		orr	r3, r3, r2, lsl #2
+		and	r2, r3, #255
+		sub	r0, r0, #256 * 4
+		str	r3, [r0, r2, lsl #2]
+		strb	r3, [r3]
+		movs	pc, lr
+/*
+ * Params  : r0 = preserved
+ *         : r1 = memc table base (preserved)
+ *         : r2 = page table entry
+ *         : r3 = preserved
+ *         : r4 = unused
+ *         : r5 = memc physical address translation table
+ *         : ip = virtual address (preserved)
+ */
+update_pte:
+		mov	r4, r2, lsr #13
+		and	r4, r4, #0x3fc
+		ldr	r4, [r5, r4]			@ covert to MEMC page
+
+		tst	r2, #PAGE_OLD | PAGE_NOT_USER	@ check for MEMC read
+		biceq	r4, r4, #0x200
+		tsteq	r2, #PAGE_READONLY | PAGE_CLEAN	@ check for MEMC write
+		biceq	r4, r4, #0x300
+
+		orr	r4, r4, ip
+		and	r2, ip, #0x01800000
+		orr	r4, r4, r2, lsr #13
+
+		and	r2, r4, #255
+		str	r4, [r1, r2, lsl #2]
+		movs	pc, lr
+
+/*
+ * Params  : r0 = preserved
+ *         : r1 = memc table base (preserved)
+ *         : r2 = page table base
+ *         : r3 = preserved
+ *         : r4 = unused
+ *         : r5 = memc physical address translation table
+ *         : ip = virtual address (updated)
+ */
+update_pte_table:
+		stmfd	sp!, {r0, lr}
+		bic	r0, r2, #3
+1:		ldr	r2, [r0], #4			@ get entry
+		tst	r2, #PAGE_PRESENT		@ page present
+		blne	update_pte			@ process pte
+		add	ip, ip, #32768			@ increment virt addr
+		ldr	r2, [r0], #4			@ get entry
+		tst	r2, #PAGE_PRESENT		@ page present
+		blne	update_pte			@ process pte
+		add	ip, ip, #32768			@ increment virt addr
+		ldr	r2, [r0], #4			@ get entry
+		tst	r2, #PAGE_PRESENT		@ page present
+		blne	update_pte			@ process pte
+		add	ip, ip, #32768			@ increment virt addr
+		ldr	r2, [r0], #4			@ get entry
+		tst	r2, #PAGE_PRESENT		@ page present
+		blne	update_pte			@ process pte
+		add	ip, ip, #32768			@ increment virt addr
+		tst	ip, #32768 * 31			@ finished?
+		bne	1b
+		ldmfd	sp!, {r0, pc}^
+
+/*
+ * Function: cpu_memc_update_all(pgd_t *pgd)
+ * Params  : pgd	Page tables/MEMC mapping
+ * Notes   : this is optimised for 32k pages
+ */
+ENTRY(cpu_memc_update_all)
+		stmfd	sp!, {r4, r5, lr}
+		bl	clear_tables
+		sub	r1, r0, #256 * 4		@ start of MEMC tables
+		adr	r5, memc_phys_table_32		@ Convert to logical page number
+		mov	ip, #0				@ virtual address
+1:		ldmia	r0!, {r2, r3}                   @ load two pgd entries
+		tst	r2, #PAGE_PRESENT               @ is pgd entry present?
+		addeq	ip, ip, #1048576        @FIXME - PAGE_PRESENT is for PTEs technically...
+		blne	update_pte_table
+		mov	r2, r3
+		tst	r2, #PAGE_PRESENT		@ is pgd entry present?
+		addeq	ip, ip, #1048576
+		blne	update_pte_table
+		teq	ip, #32 * 1048576
+		bne	1b
+		ldmfd	sp!, {r4, r5, pc}^
+
+/*
+ * Build the table to map from physical page number to memc page number
+ */
+		.type	memc_phys_table_32, #object
+memc_phys_table_32:
+		.irp	b7, 0x00, 0x80
+		.irp	b6, 0x00, 0x02
+		.irp	b5, 0x00, 0x04
+		.irp	b4, 0x00, 0x01
+
+		.irp	b3, 0x00, 0x40
+		.irp	b2, 0x00, 0x20
+		.irp	b1, 0x00, 0x10
+		.irp	b0, 0x00, 0x08
+		.long	0x03800300 + \b7 + \b6 + \b5 + \b4 + \b3 + \b2 + \b1 + \b0
+		.endr
+		.endr
+		.endr
+		.endr
+
+		.endr
+		.endr
+		.endr
+		.endr
+		.size	memc_phys_table_32, . - memc_phys_table_32
+
+/*
+ * helper for cpu_memc_update_all, this clears out all
+ * mappings, setting them close to the top of memory,
+ * and inaccessible (0x01f00000).
+ * Params  : r0 = page table pointer
+ */
+clear_tables:	ldr	r1, _arm3_set_pgd - 4
+		ldr	r2, [r1]
+		sub	r1, r0, #256 * 4		@ start of MEMC tables
+		add	r2, r1, r2, lsl #2		@ end of tables
+		mov	r3, #0x03f00000			@ Default mapping (null mapping)
+		orr	r3, r3, #0x00000f00
+		orr	r4, r3, #1
+		orr	r5, r3, #2
+		orr	ip, r3, #3
+1:		stmia	r1!, {r3, r4, r5, ip}
+		add	r3, r3, #4
+		add	r4, r4, #4
+		add	r5, r5, #4
+		add	ip, ip, #4
+		stmia	r1!, {r3, r4, r5, ip}
+		add	r3, r3, #4
+		add	r4, r4, #4
+		add	r5, r5, #4
+		add	ip, ip, #4
+		teq	r1, r2
+		bne	1b
+		mov	pc, lr
+
+/*
+ * Function: *_set_pgd(pgd_t *pgd)
+ * Params  : pgd	New page tables/MEMC mapping
+ * Purpose : update MEMC hardware with new mapping
+ */
+		.word	page_nr   @ extern - declared in mm-memc.c
+_arm3_set_pgd:	mcr	p15, 0, r1, c1, c0, 0		@ flush cache
+_arm2_set_pgd:	stmfd	sp!, {lr}
+		ldr	r1, _arm3_set_pgd - 4
+		ldr	r2, [r1]
+		sub	r0, r0, #256 * 4		@ start of MEMC tables
+		add	r1, r0, r2, lsl #2		@ end of tables
+1:		ldmia	r0!, {r2, r3, ip, lr}
+		strb	r2, [r2]
+		strb	r3, [r3]
+		strb	ip, [ip]
+		strb	lr, [lr]
+		ldmia	r0!, {r2, r3, ip, lr}
+		strb	r2, [r2]
+		strb	r3, [r3]
+		strb	ip, [ip]
+		strb	lr, [lr]
+		teq	r0, r1
+		bne	1b
+		ldmfd	sp!, {pc}^
+
+/*
+ * Function: *_proc_init (void)
+ * Purpose : Initialise the cache control registers
+ */
+_arm3_proc_init:
+		mov	r0, #0x001f0000
+		orr	r0, r0, #0x0000ff00
+		orr	r0, r0, #0x000000ff
+		mcr	p15, 0, r0, c3, c0		@ ARM3 Cacheable
+		mcr     p15, 0, r0, c4, c0		@ ARM3 Updateable
+		mov	r0, #0
+		mcr     p15, 0, r0, c5, c0		@ ARM3 Disruptive
+		mcr     p15, 0, r0, c1, c0		@ ARM3 Flush
+		mov	r0, #3
+		mcr     p15, 0, r0, c2, c0		@ ARM3 Control
+_arm2_proc_init:
+		movs	pc, lr
+
+/*
+ * Function: *_proc_fin (void)
+ * Purpose : Finalise processor (disable caches)
+ */
+_arm3_proc_fin:	mov	r0, #2
+		mcr	p15, 0, r0, c2, c0
+_arm2_proc_fin:	orrs	pc, lr, #PSR_I_BIT|PSR_F_BIT
+
+/*
+ * Function: *_xchg_1 (int new, volatile void *ptr)
+ * Params  : new	New value to store at...
+ *	   : ptr	pointer to byte-wide location
+ * Purpose : Performs an exchange operation
+ * Returns : Original byte data at 'ptr'
+ */
+_arm2_xchg_1:	mov	r2, pc
+		orr	r2, r2, #PSR_I_BIT
+		teqp	r2, #0
+		ldrb	r2, [r1]
+		strb	r0, [r1]
+		mov	r0, r2
+		movs	pc, lr
+
+_arm3_xchg_1:	swpb	r0, r0, [r1]
+		movs	pc, lr
+
+/*
+ * Function: *_xchg_4 (int new, volatile void *ptr)
+ * Params  : new	New value to store at...
+ *	   : ptr	pointer to word-wide location
+ * Purpose : Performs an exchange operation
+ * Returns : Original word data at 'ptr'
+ */
+_arm2_xchg_4:	mov	r2, pc
+		orr	r2, r2, #PSR_I_BIT
+		teqp	r2, #0
+		ldr	r2, [r1]
+		str	r0, [r1]
+		mov	r0, r2
+		movs	pc, lr
+
+_arm3_xchg_4:	swp	r0, r0, [r1]
+		movs	pc, lr
+
+_arm2_3_check_bugs:
+		bics	pc, lr, #PSR_F_BIT		@ Clear FIQ disable bit
+
+armvlsi_name:	.asciz	"ARM/VLSI"
+_arm2_name:	.asciz	"ARM 2"
+_arm250_name:	.asciz	"ARM 250"
+_arm3_name:	.asciz	"ARM 3"
+
+		.section ".init.text", #alloc, #execinstr
+/*
+ * Purpose : Function pointers used to access above functions - all calls
+ *	     come through these
+ */
+		.globl	arm2_processor_functions
+arm2_processor_functions:
+		.word	_arm2_3_check_bugs
+		.word	_arm2_proc_init
+		.word	_arm2_proc_fin
+		.word	_arm2_set_pgd
+		.word	_arm2_xchg_1
+		.word	_arm2_xchg_4
+
+cpu_arm2_info:
+		.long	armvlsi_name
+		.long	_arm2_name
+
+		.globl	arm250_processor_functions
+arm250_processor_functions:
+		.word	_arm2_3_check_bugs
+		.word	_arm2_proc_init
+		.word	_arm2_proc_fin
+		.word	_arm2_set_pgd
+		.word	_arm3_xchg_1
+		.word	_arm3_xchg_4
+
+cpu_arm250_info:
+		.long	armvlsi_name
+		.long	_arm250_name
+
+		.globl	arm3_processor_functions
+arm3_processor_functions:
+		.word	_arm2_3_check_bugs
+		.word	_arm3_proc_init
+		.word	_arm3_proc_fin
+		.word	_arm3_set_pgd
+		.word	_arm3_xchg_1
+		.word	_arm3_xchg_4
+
+cpu_arm3_info:
+		.long	armvlsi_name
+		.long	_arm3_name
+
+arm2_arch_name:	.asciz	"armv1"
+arm3_arch_name:	.asciz	"armv2"
+arm2_elf_name:	.asciz	"v1"
+arm3_elf_name:	.asciz	"v2"
+		.align
+
+		.section ".proc.info", #alloc, #execinstr
+
+		.long	0x41560200
+		.long	0xfffffff0
+		.long	arm2_arch_name
+		.long	arm2_elf_name
+		.long   0
+		.long	cpu_arm2_info
+		.long	arm2_processor_functions
+
+		.long	0x41560250
+		.long	0xfffffff0
+		.long	arm3_arch_name
+		.long	arm3_elf_name
+		.long   0
+		.long	cpu_arm250_info
+		.long	arm250_processor_functions
+
+		.long	0x41560300
+		.long	0xfffffff0
+		.long	arm3_arch_name
+		.long	arm3_elf_name
+		.long   0
+		.long	cpu_arm3_info
+		.long	arm3_processor_functions
+
diff --git a/arch/arm26/mm/small_page.c b/arch/arm26/mm/small_page.c
new file mode 100644
index 000000000000..77be86cca789
--- /dev/null
+++ b/arch/arm26/mm/small_page.c
@@ -0,0 +1,194 @@
+/*
+ *  linux/arch/arm26/mm/small_page.c
+ *
+ *  Copyright (C) 1996  Russell King
+ *  Copyright (C) 2003  Ian Molton
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ *  Changelog:
+ *   26/01/1996	RMK	Cleaned up various areas to make little more generic
+ *   07/02/1999	RMK	Support added for 16K and 32K page sizes
+ *			containing 8K blocks
+ *   23/05/2004 IM	Fixed to use struct page->lru (thanks wli)
+ *
+ */
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/smp.h>
+#include <linux/bitops.h>
+
+#include <asm/pgtable.h>
+
+#define PEDANTIC
+
+/*
+ * Requirement:
+ *  We need to be able to allocate naturally aligned memory of finer
+ *  granularity than the page size.  This is typically used for the
+ *  second level page tables on 32-bit ARMs.
+ *
+ * FIXME - this comment is *out of date*
+ * Theory:
+ *  We "misuse" the Linux memory management system.  We use alloc_page
+ *  to allocate a page and then mark it as reserved.  The Linux memory
+ *  management system will then ignore the "offset", "next_hash" and
+ *  "pprev_hash" entries in the mem_map for this page.
+ *
+ *  We then use a bitstring in the "offset" field to mark which segments
+ *  of the page are in use, and manipulate this as required during the
+ *  allocation and freeing of these small pages.
+ *
+ *  We also maintain a queue of pages being used for this purpose using
+ *  the "next_hash" and "pprev_hash" entries of mem_map;
+ */
+
+struct order {
+	struct list_head queue;
+	unsigned int mask;		/* (1 << shift) - 1		*/
+	unsigned int shift;		/* (1 << shift) size of page	*/
+	unsigned int block_mask;	/* nr_blocks - 1		*/
+	unsigned int all_used;		/* (1 << nr_blocks) - 1		*/
+};
+
+
+static struct order orders[] = {
+#if PAGE_SIZE == 32768
+	{ LIST_HEAD_INIT(orders[0].queue), 2047, 11, 15, 0x0000ffff },
+	{ LIST_HEAD_INIT(orders[1].queue), 8191, 13,  3, 0x0000000f }
+#else
+#error unsupported page size (ARGH!)
+#endif
+};
+
+#define USED_MAP(pg)			((pg)->index)
+#define TEST_AND_CLEAR_USED(pg,off)	(test_and_clear_bit(off, &USED_MAP(pg)))
+#define SET_USED(pg,off)		(set_bit(off, &USED_MAP(pg)))
+
+static DEFINE_SPINLOCK(small_page_lock);
+
+static unsigned long __get_small_page(int priority, struct order *order)
+{
+	unsigned long flags;
+	struct page *page;
+	int offset;
+
+	do {
+		spin_lock_irqsave(&small_page_lock, flags);
+
+		if (list_empty(&order->queue))
+			goto need_new_page;
+
+		page = list_entry(order->queue.next, struct page, lru);
+again:
+#ifdef PEDANTIC
+		if (USED_MAP(page) & ~order->all_used)
+			PAGE_BUG(page);
+#endif
+		offset = ffz(USED_MAP(page));
+		SET_USED(page, offset);
+		if (USED_MAP(page) == order->all_used)
+			list_del_init(&page->lru);
+		spin_unlock_irqrestore(&small_page_lock, flags);
+
+		return (unsigned long) page_address(page) + (offset << order->shift);
+
+need_new_page:
+		spin_unlock_irqrestore(&small_page_lock, flags);
+		page = alloc_page(priority);
+		spin_lock_irqsave(&small_page_lock, flags);
+
+		if (list_empty(&order->queue)) {
+			if (!page)
+				goto no_page;
+			SetPageReserved(page);
+			USED_MAP(page) = 0;
+			list_add(&page->lru, &order->queue);
+			goto again;
+		}
+
+		spin_unlock_irqrestore(&small_page_lock, flags);
+		__free_page(page);
+	} while (1);
+
+no_page:
+	spin_unlock_irqrestore(&small_page_lock, flags);
+	return 0;
+}
+
+static void __free_small_page(unsigned long spage, struct order *order)
+{
+	unsigned long flags;
+	struct page *page;
+
+	if (virt_addr_valid(spage)) {
+		page = virt_to_page(spage);
+
+		/*
+		 * The container-page must be marked Reserved
+		 */
+		if (!PageReserved(page) || spage & order->mask)
+			goto non_small;
+
+#ifdef PEDANTIC
+		if (USED_MAP(page) & ~order->all_used)
+			PAGE_BUG(page);
+#endif
+
+		spage = spage >> order->shift;
+		spage &= order->block_mask;
+
+		/*
+		 * the following must be atomic wrt get_page
+		 */
+		spin_lock_irqsave(&small_page_lock, flags);
+
+		if (USED_MAP(page) == order->all_used)
+			list_add(&page->lru, &order->queue);
+
+		if (!TEST_AND_CLEAR_USED(page, spage))
+			goto already_free;
+
+		if (USED_MAP(page) == 0)
+			goto free_page;
+
+		spin_unlock_irqrestore(&small_page_lock, flags);
+	}
+	return;
+
+free_page:
+	/*
+	 * unlink the page from the small page queue and free it
+	 */
+	list_del_init(&page->lru);
+	spin_unlock_irqrestore(&small_page_lock, flags);
+	ClearPageReserved(page);
+	__free_page(page);
+	return;
+
+non_small:
+	printk("Trying to free non-small page from %p\n", __builtin_return_address(0));
+	return;
+already_free:
+	printk("Trying to free free small page from %p\n", __builtin_return_address(0));
+}
+
+unsigned long get_page_8k(int priority)
+{
+	return __get_small_page(priority, orders+1);
+}
+
+void free_page_8k(unsigned long spage)
+{
+	__free_small_page(spage, orders+1);
+}