[PATCH] arch/parisc/mm

Update arch/parisc/mm

arch/parisc/mm/Makefile

@@ -2,6 +2,6 @@
 # Makefile for the linux parisc-specific parts of the memory manager.
 #
 
-objs-y	 := init.o fault.o kmap.o extable.o
+obj-y	 := init.o fault.o extable.o ioremap.o
 
 include $(TOPDIR)/Rules.make

arch/parisc/mm/extable.c

@@ -46,17 +46,17 @@ search_one_table (const struct exception_table_entry *first,
 const struct exception_table_entry *
 search_exception_table (unsigned long addr)
 {
-#ifndef CONFIG_MODULE
+#ifndef CONFIG_MODULES
 	/* There is only the kernel to search.  */
 	return search_one_table(__start___ex_table, 
                                 __stop___ex_table - 1, 
                                 addr);
 #else
-	struct exception_table_entry *ret;
 	/* The kernel is the last "module" -- no need to treat it special. */
 	struct module *mp;
 
 	for (mp = module_list; mp ; mp = mp->next) {
+		const struct exception_table_entry *ret;
 		if (!mp->ex_table_start)
 			continue;
 		ret = search_one_table(mp->ex_table_start, mp->ex_table_end - 1,

arch/parisc/mm/fault.c

@@ -17,6 +17,10 @@
 #include <linux/interrupt.h>
 
 #include <asm/uaccess.h>
+#include <asm/traps.h>
+
+#define PRINT_USER_FAULTS /* (turn this on if you want user faults to be */
+			 /*  dumped to the console via printk)          */
 
 
 /* Defines for parisc_acctyp()	*/
@@ -114,59 +118,31 @@ parisc_acctyp(unsigned long code, unsigned int inst)
 #undef isGraphicsFlushRead
 #undef BITSSET
 
-/* This is similar to expand_stack(), except that it is for stacks
- * that grow upwards.
- */
-
-static inline int expand_stackup(struct vm_area_struct * vma, unsigned long address)
-{
-	unsigned long grow;
-
-	address += 4 + PAGE_SIZE - 1;
-	address &= PAGE_MASK;
-	grow = (address - vma->vm_end) >> PAGE_SHIFT;
-	if (address - vma->vm_start > current->rlim[RLIMIT_STACK].rlim_cur ||
-	    ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) > current->rlim[RLIMIT_AS].rlim_cur)
-		return -ENOMEM;
-	vma->vm_end = address;
-	vma->vm_mm->total_vm += grow;
-	if (vma->vm_flags & VM_LOCKED)
-		vma->vm_mm->locked_vm += grow;
-	return 0;
-}
-
-
-/* This is similar to find_vma(), except that it understands that stacks
- * grow up rather than down.
- * XXX Optimise by making use of cache and avl tree as per find_vma().
- */
-
-struct vm_area_struct * pa_find_vma(struct mm_struct * mm, unsigned long addr)
-{
-	struct vm_area_struct *vma = NULL;
-
-	if (mm) {
-		vma = mm->mmap;
-		if (!vma || addr < vma->vm_start)
-			return NULL;
-		while (vma->vm_next && addr >= vma->vm_next->vm_start)
-			vma = vma->vm_next;
-	}
-	return vma;
-}
-
 
-/*
- * This routine handles page faults.  It determines the address,
- * and the problem, and then passes it off to one of the appropriate
- * routines.
+#if 0
+/* This is the treewalk to find a vma which is the highest that has
+ * a start < addr.  We're using find_vma_prev instead right now, but
+ * we might want to use this at some point in the future.  Probably
+ * not, but I want it committed to CVS so I don't lose it :-)
  */
-extern void parisc_terminate(char *, struct pt_regs *, int, unsigned long);
+			while (tree != vm_avl_empty) {
+				if (tree->vm_start > addr) {
+					tree = tree->vm_avl_left;
+				} else {
+					prev = tree;
+					if (prev->vm_next == NULL)
+						break;
+					if (prev->vm_next->vm_start > addr)
+						break;
+					tree = tree->vm_avl_right;
+				}
+			}
+#endif
 
 void do_page_fault(struct pt_regs *regs, unsigned long code,
 			      unsigned long address)
 {
-	struct vm_area_struct * vma;
+	struct vm_area_struct *vma, *prev_vma;
 	struct task_struct *tsk = current;
 	struct mm_struct *mm = tsk->mm;
 	const struct exception_table_entry *fix;
@@ -176,13 +152,9 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
 		goto no_context;
 
 	down_read(&mm->mmap_sem);
-	vma = pa_find_vma(mm, address);
-	if (!vma)
-		goto bad_area;
-	if (address < vma->vm_end)
-		goto good_area;
-	if (!(vma->vm_flags & VM_GROWSUP) || expand_stackup(vma, address))
-		goto bad_area;
+	vma = find_vma_prev(mm, address, &prev_vma);
+	if (!vma || address < vma->vm_start)
+		goto check_expansion;
 /*
  * Ok, we have a good vm_area for this memory access. We still need to
  * check the access permissions.
@@ -221,6 +193,11 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
 	up_read(&mm->mmap_sem);
 	return;
 
+check_expansion:
+	vma = prev_vma;
+	if (vma && (expand_stack(vma, address) == 0))
+		goto good_area;
+
 /*
  * Something tried to access memory that isn't in our memory map..
  */
@@ -230,9 +207,16 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
 	if (user_mode(regs)) {
 		struct siginfo si;
 
-		printk("\ndo_page_fault() pid=%d command='%s'\n",
-		    tsk->pid, tsk->comm);
+#ifdef PRINT_USER_FAULTS
+		printk(KERN_DEBUG "\n");
+		printk(KERN_DEBUG "do_page_fault() pid=%d command='%s' type=%lu address=0x%08lx\n",
+		    tsk->pid, tsk->comm, code, address);
+		if (vma) {
+			printk(KERN_DEBUG "vm_start = 0x%08lx, vm_end = 0x%08lx\n",
+					vma->vm_start, vma->vm_end);
+		}
 		show_regs(regs);
+#endif
 		/* FIXME: actually we need to get the signo and code correct */
 		si.si_signo = SIGSEGV;
 		si.si_errno = 0;
@@ -272,11 +256,11 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
 		}
 	}
 
-	parisc_terminate("Bad Address (null pointer deref?)",regs,code,address);
+	parisc_terminate("Bad Address (null pointer deref?)", regs, code, address);
 
   out_of_memory:
 	up_read(&mm->mmap_sem);
-	printk("VM: killing process %s\n", current->comm);
+	printk(KERN_CRIT "VM: killing process %s\n", current->comm);
 	if (user_mode(regs))
 		do_exit(SIGKILL);
 	goto no_context;

arch/parisc/mm/init.c

@@ -15,14 +15,406 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/pci.h>		/* for hppa_dma_ops and pcxl_dma_ops */
+#include <linux/blk.h>          /* for initrd_start and initrd_end */
 #include <linux/swap.h>
 #include <linux/unistd.h>
 
 #include <asm/pgalloc.h>
+#include <asm/tlb.h>
+#include <asm/pdc_chassis.h>
 
-extern unsigned long max_pfn, mem_max;
+mmu_gather_t mmu_gathers[NR_CPUS];
 
-void free_initmem(void)  {
+extern char _text;	/* start of kernel code, defined by linker */
+extern int  data_start;
+extern char _end;	/* end of BSS, defined by linker */
+extern char __init_begin, __init_end;
+
+#ifdef CONFIG_DISCONTIGMEM
+struct node_map_data node_data[MAX_PHYSMEM_RANGES];
+bootmem_data_t bmem_data[MAX_PHYSMEM_RANGES];
+unsigned char *chunkmap;
+unsigned int maxchunkmap;
+#endif
+
+static struct resource data_resource = {
+	name:	"Kernel data",
+	flags:	IORESOURCE_BUSY | IORESOURCE_MEM,
+};
+
+static struct resource code_resource = {
+	name:	"Kernel code",
+	flags:	IORESOURCE_BUSY | IORESOURCE_MEM,
+};
+
+static struct resource pdcdata_resource = {
+	name:	"PDC data (Page Zero)",
+	start:	0,
+	end:	0x9ff,
+	flags:	IORESOURCE_BUSY | IORESOURCE_MEM,
+};
+
+static struct resource sysram_resources[MAX_PHYSMEM_RANGES];
+
+static unsigned long max_pfn;
+
+/* The following array is initialized from the firmware specific
+ * information retrieved in kernel/inventory.c.
+ */
+
+physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES];
+int npmem_ranges;
+
+#ifdef __LP64__
+#define MAX_MEM         (~0UL)
+#else /* !__LP64__ */
+#define MAX_MEM         (3584U*1024U*1024U)
+#endif /* !__LP64__ */
+
+static unsigned long mem_limit = MAX_MEM;
+
+static void __init mem_limit_func(void)
+{
+	char *cp, *end;
+	unsigned long limit;
+	extern char saved_command_line[];
+
+	/* We need this before __setup() functions are called */
+
+	limit = MAX_MEM;
+	for (cp = saved_command_line; *cp; ) {
+		if (memcmp(cp, "mem=", 4) == 0) {
+			cp += 4;
+			limit = memparse(cp, &end);
+			if (end != cp)
+				break;
+			cp = end;
+		} else {
+			while (*cp != ' ' && *cp)
+				++cp;
+			while (*cp == ' ')
+				++cp;
+		}
+	}
+
+	if (limit < mem_limit)
+		mem_limit = limit;
+}
+
+#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
+
+static void __init setup_bootmem(void)
+{
+	unsigned long bootmap_size;
+	unsigned long mem_max;
+	unsigned long bootmap_pages;
+	unsigned long bootmap_start_pfn;
+	unsigned long bootmap_pfn;
+#ifndef CONFIG_DISCONTIGMEM
+	physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
+	int npmem_holes;
+#endif
+	int i, sysram_resource_count;
+
+	disable_sr_hashing(); /* Turn off space register hashing */
+
+#ifdef CONFIG_DISCONTIGMEM
+	/*
+	 * The below is still true as of 2.4.2. If this is ever fixed,
+	 * we can remove this warning!
+	 */
+
+	printk(KERN_WARNING "\n\n");
+	printk(KERN_WARNING "CONFIG_DISCONTIGMEM is enabled, which is probably a mistake. This\n");
+	printk(KERN_WARNING "option can lead to heavy swapping, even when there are gigabytes\n");
+	printk(KERN_WARNING "of free memory.\n\n");
+#endif
+
+#ifdef __LP64__
+
+#ifndef CONFIG_DISCONTIGMEM
+	/*
+	 * Sort the ranges. Since the number of ranges is typically
+	 * small, and performance is not an issue here, just do
+	 * a simple insertion sort.
+	 */
+
+	for (i = 1; i < npmem_ranges; i++) {
+		int j;
+
+		for (j = i; j > 0; j--) {
+			unsigned long tmp;
+
+			if (pmem_ranges[j-1].start_pfn <
+			    pmem_ranges[j].start_pfn) {
+
+				break;
+			}
+			tmp = pmem_ranges[j-1].start_pfn;
+			pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
+			pmem_ranges[j].start_pfn = tmp;
+			tmp = pmem_ranges[j-1].pages;
+			pmem_ranges[j-1].pages = pmem_ranges[j].pages;
+			pmem_ranges[j].pages = tmp;
+		}
+	}
+
+	/*
+	 * Throw out ranges that are too far apart (controlled by
+	 * MAX_GAP). If CONFIG_DISCONTIGMEM wasn't implemented so
+	 * poorly, we would recommend enabling that option, but,
+	 * until it is fixed, this is the best way to go.
+	 */
+
+	for (i = 1; i < npmem_ranges; i++) {
+		if (pmem_ranges[i].start_pfn -
+			(pmem_ranges[i-1].start_pfn +
+			 pmem_ranges[i-1].pages) > MAX_GAP) {
+			npmem_ranges = i;
+			break;
+		}
+	}
+#endif
+
+	if (npmem_ranges > 1) {
+
+		/* Print the memory ranges */
+
+		printk(KERN_INFO "Memory Ranges:\n");
+
+		for (i = 0; i < npmem_ranges; i++) {
+			unsigned long start;
+			unsigned long size;
+
+			size = (pmem_ranges[i].pages << PAGE_SHIFT);
+			start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
+			printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld Mb\n",
+				i,start, start + (size - 1), size >> 20);
+		}
+	}
+
+#endif /* __LP64__ */
+
+#if 1
+	/* KLUGE! this really belongs in kernel/resource.c! */
+	iomem_resource.end = ~0UL;
+#endif
+
+	sysram_resource_count = npmem_ranges;
+	for (i = 0; i < sysram_resource_count; i++) {
+		struct resource *res = &sysram_resources[i];
+		res->name = "System RAM";
+		res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
+		res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
+		res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
+		request_resource(&iomem_resource, res);
+	}
+
+	/*
+	 * For 32 bit kernels we limit the amount of memory we can
+	 * support, in order to preserve enough kernel address space
+	 * for other purposes. For 64 bit kernels we don't normally
+	 * limit the memory, but this mechanism can be used to
+	 * artificially limit the amount of memory (and it is written
+	 * to work with multiple memory ranges).
+	 */
+
+	mem_limit_func();       /* check for "mem=" argument */
+
+	mem_max = 0;
+	for (i = 0; i < npmem_ranges; i++) {
+		unsigned long rsize;
+
+		rsize = pmem_ranges[i].pages << PAGE_SHIFT;
+		if ((mem_max + rsize) > mem_limit) {
+			printk(KERN_WARNING "Memory truncated to %ld Mb\n", mem_limit >> 20);
+			if (mem_max == mem_limit)
+				npmem_ranges = i;
+			else {
+				pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
+						       - (mem_max >> PAGE_SHIFT);
+				npmem_ranges = i + 1;
+				mem_max = mem_limit;
+			}
+			break;
+		}
+		mem_max += rsize;
+	}
+
+	printk(KERN_INFO "Total Memory: %ld Mb\n",mem_max >> 20);
+
+#ifndef CONFIG_DISCONTIGMEM
+
+	/* Merge the ranges, keeping track of the holes */
+
+	{
+		unsigned long end_pfn;
+		unsigned long hole_pages;
+
+		npmem_holes = 0;
+		end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
+		for (i = 1; i < npmem_ranges; i++) {
+
+			hole_pages = pmem_ranges[i].start_pfn - end_pfn;
+			if (hole_pages) {
+				pmem_holes[npmem_holes].start_pfn = end_pfn;
+				pmem_holes[npmem_holes++].pages = hole_pages;
+				end_pfn += hole_pages;
+			}
+			end_pfn += pmem_ranges[i].pages;
+		}
+
+		pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
+		npmem_ranges = 1;
+	}
+#endif
+
+	bootmap_pages = 0;
+	for (i = 0; i < npmem_ranges; i++)
+		bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
+
+	bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
+
+#ifdef CONFIG_DISCONTIGMEM
+	for (i = 0; i < npmem_ranges; i++)
+		node_data[i].pg_data.bdata = &bmem_data[i];
+#endif
+	/*
+	 * Initialize and free the full range of memory in each range.
+	 * Note that the only writing these routines do are to the bootmap,
+	 * and we've made sure to locate the bootmap properly so that they
+	 * won't be writing over anything important.
+	 */
+
+	bootmap_pfn = bootmap_start_pfn;
+	max_pfn = 0;
+	for (i = 0; i < npmem_ranges; i++) {
+		unsigned long start_pfn;
+		unsigned long npages;
+
+		start_pfn = pmem_ranges[i].start_pfn;
+		npages = pmem_ranges[i].pages;
+
+		bootmap_size = init_bootmem_node(NODE_DATA(i),
+						bootmap_pfn,
+						start_pfn,
+						(start_pfn + npages) );
+		free_bootmem_node(NODE_DATA(i),
+				  (start_pfn << PAGE_SHIFT),
+				  (npages << PAGE_SHIFT) );
+		bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+		if ((start_pfn + npages) > max_pfn)
+			max_pfn = start_pfn + npages;
+	}
+
+	if ((bootmap_pfn - bootmap_start_pfn) != bootmap_pages) {
+		printk(KERN_WARNING "WARNING! bootmap sizing is messed up!\n");
+		BUG();
+	}
+
+	/* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
+
+#define PDC_CONSOLE_IO_IODC_SIZE 32768
+
+	reserve_bootmem_node(NODE_DATA(0), 0UL,
+			(unsigned long)(PAGE0->mem_free + PDC_CONSOLE_IO_IODC_SIZE));
+	reserve_bootmem_node(NODE_DATA(0),__pa((unsigned long)&_text),
+			(unsigned long)(&_end - &_text));
+	reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
+			((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT));
+
+#ifndef CONFIG_DISCONTIGMEM
+
+	/* reserve the holes */
+
+	for (i = 0; i < npmem_holes; i++) {
+		reserve_bootmem_node(NODE_DATA(0),
+				(pmem_holes[i].start_pfn << PAGE_SHIFT),
+				(pmem_holes[i].pages << PAGE_SHIFT));
+	}
+#endif
+
+#ifdef CONFIG_BLK_DEV_INITRD
+	if (initrd_start) {
+		printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
+		if (__pa(initrd_start) < mem_max) {
+			unsigned long initrd_reserve;
+
+			if (__pa(initrd_end) > mem_max) {
+				initrd_reserve = mem_max - __pa(initrd_start);
+			} else {
+				initrd_reserve = initrd_end - initrd_start;
+			}
+			initrd_below_start_ok = 1;
+			printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
+
+			reserve_bootmem_node(NODE_DATA(0),__pa(initrd_start), initrd_reserve);
+		}
+	}
+#endif
+
+	data_resource.start =  virt_to_phys(&data_start);
+	data_resource.end = virt_to_phys(&_end)-1;
+	code_resource.start = virt_to_phys(&_text);
+	code_resource.end = virt_to_phys(&data_start)-1;
+
+	/* We don't know which region the kernel will be in, so try
+	 * all of them.
+	 */
+	for (i = 0; i < sysram_resource_count; i++) {
+		struct resource *res = &sysram_resources[i];
+		request_resource(res, &code_resource);
+		request_resource(res, &data_resource);
+	}
+	request_resource(&sysram_resources[0], &pdcdata_resource);
+}
+
+void free_initmem(void)
+{
+	/* FIXME: */
+#if 0
+	printk(KERN_INFO "NOT FREEING INITMEM (%dk)\n",
+			(&__init_end - &__init_begin) >> 10);
+	return;
+#else
+	unsigned long addr;
+	
+	printk(KERN_INFO "Freeing unused kernel memory: ");
+
+#if 1
+	/* Attempt to catch anyone trying to execute code here
+	 * by filling the page with BRK insns.
+	 * 
+	 * If we disable interrupts for all CPUs, then IPI stops working.
+	 * Kinda breaks the global cache flushing.
+	 */
+	local_irq_disable();
+
+	memset(&__init_begin, 0x00, 
+		(unsigned long)&__init_end - (unsigned long)&__init_begin);
+
+	flush_data_cache();
+	asm volatile("sync" : : );
+	flush_icache_range((unsigned long)&__init_begin, (unsigned long)&__init_end);
+	asm volatile("sync" : : );
+
+	local_irq_enable();
+#endif
+	
+	addr = (unsigned long)(&__init_begin);
+	for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
+		ClearPageReserved(virt_to_page(addr));
+		set_page_count(virt_to_page(addr), 1);
+		free_page(addr);
+		num_physpages++;
+		totalram_pages++;
+	}
+
+	/* set up a new led state on systems shipped LED State panel */
+	pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
+	
+	printk("%luk freed\n", (unsigned long)(&__init_end - &__init_begin) >> 10);
+#endif
 }
 
 /*
@@ -36,67 +428,43 @@ void free_initmem(void)  {
  * a hole of 4kB between each vmalloced area for the same reason.
  */
 
+#define MAP_START 0x4000 /* Leave room for gateway page expansion */
 #define VM_MAP_OFFSET  (32*1024)
 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 				     & ~(VM_MAP_OFFSET-1)))
 
 void *vmalloc_start;
+#ifdef CONFIG_PA11
 unsigned long pcxl_dma_start;
+#endif
 
 void __init mem_init(void)
 {
-	max_mapnr = num_physpages = max_low_pfn;
-	high_memory = __va(max_low_pfn * PAGE_SIZE);
+	int i;
 
-	totalram_pages += free_all_bootmem();
-	printk("Memory: %luk available\n", totalram_pages << (PAGE_SHIFT-10));
+	high_memory = __va((max_pfn << PAGE_SHIFT));
+	max_mapnr = (virt_to_page(high_memory - 1) - mem_map) + 1;
 
-	if (hppa_dma_ops == &pcxl_dma_ops) {
-	    pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(high_memory);
-	    vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start + PCXL_DMA_MAP_SIZE);
-	}
-	else {
-	    pcxl_dma_start = 0;
-	    vmalloc_start = SET_MAP_OFFSET(high_memory);
-	}
-}
-
-void __bad_pgd(pgd_t *pgd)
-{
-	printk("Bad pgd in pmd_alloc: %08lx\n", pgd_val(*pgd));
-	pgd_val(*pgd) = _PAGE_TABLE + __pa(BAD_PAGETABLE);
-}
-
-void __bad_pmd(pmd_t *pmd)
-{
-	printk("Bad pmd in pte_alloc: %08lx\n", pmd_val(*pmd));
-	pmd_val(*pmd) = _PAGE_TABLE + __pa(BAD_PAGETABLE);
-}
-
-pte_t *get_pte_slow(pmd_t *pmd, unsigned long offset)
-{
-	pte_t *pte;
+	num_physpages = 0;
+	mem_map = zone_table[0]->zone_mem_map;
+	for (i = 0; i < npmem_ranges; i++)
+		num_physpages += free_all_bootmem_node(NODE_DATA(i));
+	totalram_pages = num_physpages;
 
-	pte = (pte_t *) __get_free_page(GFP_KERNEL);
+	printk(KERN_INFO "Memory: %luk available\n", num_physpages << (PAGE_SHIFT-10));
 
-	if (pmd_none(*pmd)) {
-		if (pte) {
-			clear_page(pte);
-			pmd_val(*pmd) = _PAGE_TABLE + __pa((unsigned long)pte);
-			return pte + offset;
-		}
-		pmd_val(*pmd) = _PAGE_TABLE + __pa(BAD_PAGETABLE);
-		return NULL;
-	}
-
-	free_page((unsigned long)pte);
-
-	if (pmd_bad(*pmd)) {
-		__bad_pmd(pmd);
-		return NULL;
+#ifdef CONFIG_PA11
+	if (hppa_dma_ops == &pcxl_dma_ops) {
+		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
+		vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start + PCXL_DMA_MAP_SIZE);
+	} else {
+		pcxl_dma_start = 0;
+		vmalloc_start = SET_MAP_OFFSET(MAP_START);
 	}
+#else
+	vmalloc_start = SET_MAP_OFFSET(MAP_START);
+#endif
 
-	return (pte_t *) pmd_page(*pmd) + offset;
 }
 
 int do_check_pgt_cache(int low, int high)
@@ -104,40 +472,16 @@ int do_check_pgt_cache(int low, int high)
 	return 0;
 }
 
-/*
- * BAD_PAGE is the page that is used for page faults when linux
- * is out-of-memory. Older versions of linux just did a
- * do_exit(), but using this instead means there is less risk
- * for a process dying in kernel mode, possibly leaving an inode
- * unused etc..
- *
- * BAD_PAGETABLE is the accompanying page-table: it is initialized
- * to point to BAD_PAGE entries.
- *
- * ZERO_PAGE is a special page that is used for zero-initialized
- * data and COW.
- */
-pte_t * __bad_pagetable(void)
-{
-	return (pte_t *) NULL;
-}
-
 unsigned long *empty_zero_page;
-unsigned long *empty_bad_page;
-
-pte_t __bad_page(void)
-{
-	return *(pte_t *)NULL;
-}
 
 void show_mem(void)
 {
 	int i,free = 0,total = 0,reserved = 0;
 	int shared = 0, cached = 0;
 
-	printk("Mem-info:\n");
+	printk(KERN_INFO "Mem-info:\n");
 	show_free_areas();
-	printk("Free swap:	 %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
+	printk(KERN_INFO "Free swap:	 %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
 	i = max_mapnr;
 	while (i-- > 0) {
 		total++;
@@ -150,53 +494,49 @@ void show_mem(void)
 		else
 			shared += atomic_read(&mem_map[i].count) - 1;
 	}
-	printk("%d pages of RAM\n",total);
-	printk("%d reserved pages\n",reserved);
-	printk("%d pages shared\n",shared);
-	printk("%d pages swap cached\n",cached);
-}
-
-void set_pte_phys (unsigned long vaddr, unsigned long phys)
-{
+	printk(KERN_INFO "%d pages of RAM\n", total);
+	printk(KERN_INFO "%d reserved pages\n", reserved);
+	printk(KERN_INFO "%d pages shared\n", shared);
+	printk(KERN_INFO "%d pages swap cached\n", cached);
 }
 
 
-/*
- * pagetable_init() sets up the page tables
- *
- * Note that gateway_init() places the Linux gateway page at page 0.
- * Since gateway pages cannot be dereferenced this has the desirable
- * side effect of trapping those pesky NULL-reference errors in the
- * kernel.
- */
-static void __init pagetable_init(void)
+static void __init map_pages(unsigned long start_vaddr, unsigned long start_paddr, unsigned long size, pgprot_t pgprot)
 {
 	pgd_t *pg_dir;
 	pmd_t *pmd;
 	pte_t *pg_table;
+	unsigned long end_paddr;
+	unsigned long start_pmd;
+	unsigned long start_pte;
 	unsigned long tmp1;
 	unsigned long tmp2;
 	unsigned long address;
 	unsigned long ro_start;
 	unsigned long ro_end;
 	unsigned long fv_addr;
-	extern  const int stext;
-	extern  int data_start;
-	extern  const unsigned long fault_vector_20;
+	unsigned long gw_addr;
+	extern const unsigned long fault_vector_20;
+	extern void * const linux_gateway_page;
 
-	ro_start = __pa((unsigned long)&stext);
+	ro_start = __pa((unsigned long)&_text);
 	ro_end   = __pa((unsigned long)&data_start);
 	fv_addr  = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
+	gw_addr  = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
 
-	printk("pagetable_init\n");
+	end_paddr = start_paddr + size;
 
-	/* Map whole memory from PAGE_OFFSET */
-	pg_dir = (pgd_t *)swapper_pg_dir + USER_PGD_PTRS;
+	pg_dir = pgd_offset_k(start_vaddr);
 
-	address = 0;
-	while (address < mem_max) {
-		/* XXX: BTLB should be done here */
+#if PTRS_PER_PMD == 1
+	start_pmd = 0;
+#else
+	start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
+#endif
+	start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 
+	address = start_paddr;
+	while (address < end_paddr) {
 #if PTRS_PER_PMD == 1
 		pmd = (pmd_t *)__pa(pg_dir);
 #else
@@ -207,7 +547,7 @@ static void __init pagetable_init(void)
 		 */
 
 		if (!pmd) {
-			pmd = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
+			pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
 			pmd = (pmd_t *) __pa(pmd);
 		}
 
@@ -217,8 +557,8 @@ static void __init pagetable_init(void)
 
 		/* now change pmd to kernel virtual addresses */
 
-		pmd = (pmd_t *) __va(pmd);
-		for (tmp1 = 0 ; tmp1 < PTRS_PER_PMD ; tmp1++,pmd++) {
+		pmd = (pmd_t *)__va(pmd) + start_pmd;
+		for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++,pmd++) {
 
 			/*
 			 * pg_table is physical at this point
@@ -227,7 +567,7 @@ static void __init pagetable_init(void)
 			pg_table = (pte_t *) (PAGE_MASK & pmd_val(*pmd));
 			if (!pg_table) {
 				pg_table = (pte_t *)
-					alloc_bootmem_low_pages(PAGE_SIZE);
+					alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
 				pg_table = (pte_t *) __pa(pg_table);
 			}
 
@@ -236,64 +576,90 @@ static void __init pagetable_init(void)
 
 			/* now change pg_table to kernel virtual addresses */
 
-			pg_table = (pte_t *) __va(pg_table);
-			for (tmp2=0; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
+			pg_table = (pte_t *) __va(pg_table) + start_pte;
+			for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
 				pte_t pte;
 
-#if !defined(CONFIG_KWDB) && !defined(CONFIG_STI_CONSOLE)
+#if !defined(CONFIG_STI_CONSOLE)
 #warning STI console should explicitly allocate executable pages but does not
-/* KWDB needs to write kernel text when setting break points.
-**
-** The right thing to do seems like KWDB modify only the pte which
-** has a break point on it...otherwise we might mask worse bugs.
-*/
+				/*
+				 * Map the fault vector writable so we can
+				 * write the HPMC checksum.
+				 */
 				if (address >= ro_start && address < ro_end
-							&& address != fv_addr)
+							&& address != fv_addr
+							&& address != gw_addr)
 				    pte = __mk_pte(address, PAGE_KERNEL_RO);
 				else
 #endif
-				    pte = __mk_pte(address, PAGE_KERNEL);
+				    pte = __mk_pte(address, pgprot);
 
-				if (address >= mem_max)
+				if (address >= end_paddr)
 					pte_val(pte) = 0;
 
 				set_pte(pg_table, pte);
 
 				address += PAGE_SIZE;
 			}
+			start_pte = 0;
 
-			if (address >= mem_max)
+			if (address >= end_paddr)
 			    break;
 		}
+		start_pmd = 0;
+	}
+}
+
+/*
+ * pagetable_init() sets up the page tables
+ *
+ * Note that gateway_init() places the Linux gateway page at page 0.
+ * Since gateway pages cannot be dereferenced this has the desirable
+ * side effect of trapping those pesky NULL-reference errors in the
+ * kernel.
+ */
+static void __init pagetable_init(void)
+{
+	int range;
+
+	printk("pagetable_init\n");
+
+	/* Map each physical memory range to its kernel vaddr */
+
+	for (range = 0; range < npmem_ranges; range++) {
+		unsigned long start_paddr;
+		unsigned long end_paddr;
+		unsigned long size;
+
+		start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
+		end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
+		size = pmem_ranges[range].pages << PAGE_SHIFT;
+
+		map_pages((unsigned long)__va(start_paddr), start_paddr,
+			size, PAGE_KERNEL);
+	}
+
+#ifdef CONFIG_BLK_DEV_INITRD
+	if (initrd_end && initrd_end > mem_limit) {
+		printk("initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
+		map_pages(initrd_start, __pa(initrd_start),
+			initrd_end - initrd_start, PAGE_KERNEL);
 	}
+#endif
 
 	empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
 	memset(empty_zero_page, 0, PAGE_SIZE);
 }
 
-unsigned long gateway_pgd_offset;
-unsigned long gateway_pgd_entry;
-
 static void __init gateway_init(void)
 {
-	unsigned long hpux_gateway_page_addr;
 	unsigned long linux_gateway_page_addr;
-	pgd_t *pg_dir;
-	pmd_t *pmd_base;
-	pmd_t *pmd;
-	pte_t *pg_table_base;
-	pte_t *pg_table;
-	/* FIXME: These are 'const' in order to trick the compiler
-           into not treating them as DP-relative data. */
-	extern void * const hpux_gateway_page;
+	/* FIXME: This is 'const' in order to trick the compiler
+	   into not treating it as DP-relative data. */
 	extern void * const linux_gateway_page;
-	pte_t pte;
 
-	hpux_gateway_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
 	linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 
-	gateway_pgd_offset = hpux_gateway_page_addr >> PGDIR_SHIFT;
-
 	/*
 	 * Setup Linux Gateway page.
 	 *
@@ -301,151 +667,345 @@ static void __init gateway_init(void)
 	 * page 0), so it doesn't need to be aliased into user space.
 	 */
 
-	pg_dir = (pgd_t *)swapper_pg_dir;
-
-#if PTRS_PER_PMD == 1
-	pmd_base = (pmd_t *)pg_dir;
-	pmd = pmd_base +
-		((linux_gateway_page_addr) >> PGDIR_SHIFT);
-
-#else
-	pmd_base = (pmd_t *) alloc_bootmem_pages(PAGE_SIZE);
-	pgd_val(*(pg_dir + (linux_gateway_page_addr >> PGDIR_SHIFT))) =
-		_PAGE_TABLE | __pa(pmd_base);
+	map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
+		PAGE_SIZE, PAGE_GATEWAY);
+}
 
-	pmd = pmd_base +
-		((linux_gateway_page_addr & (PMD_MASK) & (PGDIR_SIZE - 1)) >>
-								PMD_SHIFT);
-#endif
+void
+map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
+{
+	pgd_t *pg_dir;
+	pmd_t *pmd;
+	pte_t *pg_table;
+	unsigned long start_pmd;
+	unsigned long start_pte;
+	unsigned long address;
+	unsigned long hpux_gw_page_addr;
+	/* FIXME: This is 'const' in order to trick the compiler
+	   into not treating it as DP-relative data. */
+	extern void * const hpux_gateway_page;
 
-	pg_table_base = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
+	hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
 
-	pmd_val(*pmd) = _PAGE_TABLE | __pa(pg_table_base);
+	/*
+	 * Setup HP-UX Gateway page.
+	 *
+	 * The HP-UX gateway page resides in the user address space,
+	 * so it needs to be aliased into each process.
+	 */
 
-	pte = __mk_pte(__pa(&linux_gateway_page), PAGE_GATEWAY);
+	pg_dir = pgd_offset(mm,hpux_gw_page_addr);
 
-	pg_table = pg_table_base +
-		((linux_gateway_page_addr & (PAGE_MASK) & (PMD_SIZE - 1)) >>
-								PAGE_SHIFT);
+#if PTRS_PER_PMD == 1
+	start_pmd = 0;
+#else
+	start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
+#endif
+	start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 
-	set_pte(pg_table,pte);
+	address = __pa(&hpux_gateway_page);
+#if PTRS_PER_PMD == 1
+	pmd = (pmd_t *)__pa(pg_dir);
+#else
+	pmd = (pmd_t *) (PAGE_MASK & pgd_val(*pg_dir));
 
 	/*
-	 * Setup HP-UX gateway page.
-	 * This page will be aliased into each user address space.
+	 * pmd is physical at this point
 	 */
 
-	pg_table_base = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
-
-	pte = __mk_pte(__pa(&hpux_gateway_page), PAGE_GATEWAY);
-	pg_table = pg_table_base +
-		((hpux_gateway_page_addr & (PAGE_MASK) & (PMD_SIZE - 1)) >>
-								PAGE_SHIFT);
-
-	set_pte(pg_table,pte);
-
+	if (!pmd) {
+		pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
+		pmd = (pmd_t *) __pa(pmd);
+	}
 
-#if PTRS_PER_PMD == 1
-	pmd_base = (pmd_t *)pg_table_base;
-#else
-	pmd_base = (pmd_t *) alloc_bootmem_pages(PAGE_SIZE);
-	pmd = pmd_base +
-		((hpux_gateway_page_addr & (PMD_MASK) & (PGDIR_SIZE - 1)) >>
-								PMD_SHIFT);
-	pmd_val(*pmd) = _PAGE_TABLE | __pa(pg_table_base);
+	pgd_val(*pg_dir) = _PAGE_TABLE | (unsigned long) pmd;
 #endif
+	/* now change pmd to kernel virtual addresses */
 
-	gateway_pgd_entry = _PAGE_TABLE | __pa(pmd_base);
+	pmd = (pmd_t *)__va(pmd) + start_pmd;
 
 	/*
-	 * We will be aliasing the HP-UX gateway page into all HP-UX
-	 * user spaces at the same address (not counting the space register
-	 * value) that will be equivalently mapped as long as space register
-	 * hashing is disabled. It will be a problem if anyone touches
-	 * the gateway pages at its "kernel" address, since that is
-	 * NOT equivalently mapped. We'll flush the caches at this
-	 * point, just in case some code has touched those addresses
-	 * previous to this, but all bets are off if they get touched
-	 * after this point.
+	 * pg_table is physical at this point
 	 */
 
-	flush_all_caches();
+	pg_table = (pte_t *) (PAGE_MASK & pmd_val(*pmd));
+	if (!pg_table)
+		pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
 
-	return;
+	pmd_val(*pmd) = _PAGE_TABLE | (unsigned long) pg_table;
+
+	/* now change pg_table to kernel virtual addresses */
+
+	pg_table = (pte_t *) __va(pg_table) + start_pte;
+	set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
 }
 
+extern void flush_tlb_all_local(void);
+
 void __init paging_init(void)
 {
+	int i;
+
+	setup_bootmem();
 	pagetable_init();
 	gateway_init();
+	flush_cache_all_local(); /* start with known state */
+	flush_tlb_all_local();
 
-	{
-		unsigned long zones_size[MAX_NR_ZONES] = { max_pfn/2, max_pfn/2, };
+	for (i = 0; i < npmem_ranges; i++) {
+		unsigned long zones_size[MAX_NR_ZONES] = { 0, 0, 0, };
+
+		zones_size[ZONE_DMA] = pmem_ranges[i].pages;
+		free_area_init_node(i,NODE_DATA(i),NULL,zones_size,
+				(pmem_ranges[i].start_pfn << PAGE_SHIFT),0);
+	}
+
+#ifdef CONFIG_DISCONTIGMEM
+	/*
+	 * Initialize support for virt_to_page() macro.
+	 *
+	 * Note that MAX_ADDRESS is the largest virtual address that
+	 * we can map. However, since we map all physical memory into
+	 * the kernel address space, it also has an effect on the maximum
+	 * physical address we can map (MAX_ADDRESS - PAGE_OFFSET).
+	 */
 
-		free_area_init(zones_size);
+	maxchunkmap = MAX_ADDRESS >> CHUNKSHIFT;
+	chunkmap = (unsigned char *)alloc_bootmem(maxchunkmap);
+
+	for (i = 0; i < maxchunkmap; i++)
+	    chunkmap[i] = BADCHUNK;
+
+	for (i = 0; i < npmem_ranges; i++) {
+
+		ADJ_NODE_MEM_MAP(i) = NODE_MEM_MAP(i) - pmem_ranges[i].start_pfn;
+		{
+			unsigned long chunk_paddr;
+			unsigned long end_paddr;
+			int chunknum;
+
+			chunk_paddr = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
+			end_paddr = chunk_paddr + (pmem_ranges[i].pages << PAGE_SHIFT);
+			chunk_paddr &= CHUNKMASK;
+
+			chunknum = (int)CHUNKNUM(chunk_paddr);
+			while (chunk_paddr < end_paddr) {
+				if (chunknum >= maxchunkmap)
+					goto badchunkmap1;
+				if (chunkmap[chunknum] != BADCHUNK)
+					goto badchunkmap2;
+				chunkmap[chunknum] = (unsigned char)i;
+				chunk_paddr += CHUNKSZ;
+				chunknum++;
+			}
+		}
 	}
+
+	return;
+
+badchunkmap1:
+	panic("paging_init: Physical address exceeds maximum address space!\n");
+badchunkmap2:
+	panic("paging_init: Collision in chunk map array. CHUNKSZ needs to be smaller\n");
+#endif
 }
 
-#define NR_SPACE_IDS	8192
+#ifdef CONFIG_PA20
 
-static unsigned long space_id[NR_SPACE_IDS / (8 * sizeof(long))];
-static unsigned long space_id_index;
-static unsigned long free_space_ids = NR_SPACE_IDS;
+/*
+ * Currently, all PA20 chips have 18 bit protection id's, which is the
+ * limiting factor (space ids are 32 bits).
+ */
+
+#define NR_SPACE_IDS 262144
+
+#else
 
 /*
- * XXX: We should probably unfold the set_bit / test_bit / clear_bit
- * locking out of these two functions and have a single spinlock on the
- * space_id data structures.
- *
- * Don't bother. This is all going to be significantly changed in the
- * very near future.
+ * Currently we have a one-to-one relationship between space id's and
+ * protection id's. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
+ * support 15 bit protection id's, so that is the limiting factor.
+ * PCXT' has 18 bit protection id's, but only 16 bit spaceids, so it's
+ * probably not worth the effort for a special case here.
  */
 
-#define SPACEID_SHIFT (PAGE_SHIFT + (PT_NLEVELS)*(PAGE_SHIFT - PT_NLEVELS) - 32)
+#define NR_SPACE_IDS 32768
+
+#endif  /* !CONFIG_PA20 */
+
+#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
+#define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
+
+static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
+static unsigned long dirty_space_id[SID_ARRAY_SIZE];
+static unsigned long space_id_index;
+static unsigned long free_space_ids = NR_SPACE_IDS - 1;
+static unsigned long dirty_space_ids = 0;
+
+static spinlock_t sid_lock = SPIN_LOCK_UNLOCKED;
 
 unsigned long alloc_sid(void)
 {
 	unsigned long index;
 
-	if (free_space_ids == 0)
-		BUG();
+	spin_lock(&sid_lock);
 
-	free_space_ids--;
+	if (free_space_ids == 0) {
+		if (dirty_space_ids != 0) {
+			spin_unlock(&sid_lock);
+			flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
+			spin_lock(&sid_lock);
+		}
+		if (free_space_ids == 0)
+			BUG();
+	}
 
-	do {
-		index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
-	} while(test_and_set_bit(index, space_id));
+	free_space_ids--;
 
+	index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
+	space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 	space_id_index = index;
 
+	spin_unlock(&sid_lock);
+
 	return index << SPACEID_SHIFT;
 }
 
 void free_sid(unsigned long spaceid)
 {
 	unsigned long index = spaceid >> SPACEID_SHIFT;
-	if (index < 0)
-		BUG();
+	unsigned long *dirty_space_offset;
+
+	dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
+	index &= (BITS_PER_LONG - 1);
+
+	spin_lock(&sid_lock);
+
+	if (*dirty_space_offset & (1L << index))
+	    BUG(); /* attempt to free space id twice */
+
+	*dirty_space_offset |= (1L << index);
+	dirty_space_ids++;
+
+	spin_unlock(&sid_lock);
+}
 
-	clear_bit(index, space_id);
 
-	if (space_id_index > index) {
-		space_id_index = index;
+#ifdef CONFIG_SMP
+static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
+{
+	int i;
+
+	/* NOTE: sid_lock must be held upon entry */
+
+	*ndirtyptr = dirty_space_ids;
+	if (dirty_space_ids != 0) {
+	    for (i = 0; i < SID_ARRAY_SIZE; i++) {
+		dirty_array[i] = dirty_space_id[i];
+		dirty_space_id[i] = 0;
+	    }
+	    dirty_space_ids = 0;
 	}
-	free_space_ids++;
+
+	return;
 }
 
+static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
+{
+	int i;
+
+	/* NOTE: sid_lock must be held upon entry */
+
+	if (ndirty != 0) {
+		for (i = 0; i < SID_ARRAY_SIZE; i++) {
+			space_id[i] ^= dirty_array[i];
+		}
+
+		free_space_ids += ndirty;
+		space_id_index = 0;
+	}
+}
+
+#else /* CONFIG_SMP */
+
+static void recycle_sids(void)
+{
+	int i;
+
+	/* NOTE: sid_lock must be held upon entry */
+
+	if (dirty_space_ids != 0) {
+		for (i = 0; i < SID_ARRAY_SIZE; i++) {
+			space_id[i] ^= dirty_space_id[i];
+			dirty_space_id[i] = 0;
+		}
+
+		free_space_ids += dirty_space_ids;
+		dirty_space_ids = 0;
+		space_id_index = 0;
+	}
+}
+#endif
+
+/*
+ * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
+ * purged, we can safely reuse the space ids that were released but
+ * not flushed from the tlb.
+ */
+
+#ifdef CONFIG_SMP
+
+static unsigned long recycle_ndirty;
+static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
+static unsigned int recycle_inuse = 0;
+
+void flush_tlb_all(void)
+{
+	int do_recycle;
+
+	do_recycle = 0;
+	spin_lock(&sid_lock);
+	if (dirty_space_ids > RECYCLE_THRESHOLD) {
+	    if (recycle_inuse) {
+		BUG();  /* FIXME: Use a semaphore/wait queue here */
+	    }
+	    get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
+	    recycle_inuse++;
+	    do_recycle++;
+	}
+	spin_unlock(&sid_lock);
+	smp_call_function((void (*)(void *))flush_tlb_all_local, NULL, 1, 1);
+	flush_tlb_all_local();
+	if (do_recycle) {
+	    spin_lock(&sid_lock);
+	    recycle_sids(recycle_ndirty,recycle_dirty_array);
+	    recycle_inuse = 0;
+	    spin_unlock(&sid_lock);
+	}
+}
+#else
+void flush_tlb_all(void)
+{
+	spin_lock(&sid_lock);
+	flush_tlb_all_local();
+	recycle_sids();
+	spin_unlock(&sid_lock);
+}
+#endif
+
 #ifdef CONFIG_BLK_DEV_INITRD
 void free_initrd_mem(unsigned long start, unsigned long end)
 {
 #if 0
+	if (start < end)
+		printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 	for (; start < end; start += PAGE_SIZE) {
-		ClearPageReserved(mem_map + MAP_NR(start));
-		set_page_count(mem_map+MAP_NR(start), 1);
+		ClearPageReserved(virt_to_page(start));
+		set_page_count(virt_to_page(start), 1);
 		free_page(start);
+		num_physpages++;
 		totalram_pages++;
 	}
-	printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 #endif
 }
 #endif

arch/parisc/mm/ioremap.c

+/*
+ * arch/parisc/mm/ioremap.c
+ *
+ * Re-map IO memory to kernel address space so that we can access it.
+ * This is needed for high PCI addresses that aren't mapped in the
+ * 640k-1MB IO memory area on PC's
+ *
+ * (C) Copyright 1995 1996 Linus Torvalds
+ * (C) Copyright 2001 Helge Deller <deller@gmx.de>
+ */
+
+#include <linux/vmalloc.h>
+#include <linux/errno.h>
+#include <asm/io.h>
+#include <asm/pgalloc.h>
+
+static inline void remap_area_pte(pte_t * pte, unsigned long address, unsigned long size,
+	unsigned long phys_addr, unsigned long flags)
+{
+	unsigned long end;
+
+	address &= ~PMD_MASK;
+	end = address + size;
+	if (end > PMD_SIZE)
+		end = PMD_SIZE;
+	if (address >= end)
+		BUG();
+	do {
+		if (!pte_none(*pte)) {
+			printk(KERN_ERR "remap_area_pte: page already exists\n");
+			BUG();
+		}
+		set_pte(pte, mk_pte_phys(phys_addr, __pgprot(_PAGE_PRESENT | _PAGE_RW | 
+					_PAGE_DIRTY | _PAGE_ACCESSED | flags)));
+		address += PAGE_SIZE;
+		phys_addr += PAGE_SIZE;
+		pte++;
+	} while (address && (address < end));
+}
+
+static inline int remap_area_pmd(pmd_t * pmd, unsigned long address, unsigned long size,
+	unsigned long phys_addr, unsigned long flags)
+{
+	unsigned long end;
+
+	address &= ~PGDIR_MASK;
+	end = address + size;
+	if (end > PGDIR_SIZE)
+		end = PGDIR_SIZE;
+	phys_addr -= address;
+	if (address >= end)
+		BUG();
+	do {
+		pte_t * pte = pte_alloc_kernel(NULL, pmd, address);
+		if (!pte)
+			return -ENOMEM;
+		remap_area_pte(pte, address, end - address, address + phys_addr, flags);
+		address = (address + PMD_SIZE) & PMD_MASK;
+		pmd++;
+	} while (address && (address < end));
+	return 0;
+}
+
+#if (USE_HPPA_IOREMAP)
+static int remap_area_pages(unsigned long address, unsigned long phys_addr,
+				 unsigned long size, unsigned long flags)
+{
+	int error;
+	pgd_t * dir;
+	unsigned long end = address + size;
+
+	phys_addr -= address;
+	dir = pgd_offset(&init_mm, address);
+	flush_cache_all();
+	if (address >= end)
+		BUG();
+	spin_lock(&init_mm.page_table_lock);
+	do {
+		pmd_t *pmd;
+		pmd = pmd_alloc(dir, address);
+		error = -ENOMEM;
+		if (!pmd)
+			break;
+		if (remap_area_pmd(pmd, address, end - address,
+					 phys_addr + address, flags))
+			break;
+		error = 0;
+		address = (address + PGDIR_SIZE) & PGDIR_MASK;
+		dir++;
+	} while (address && (address < end));
+	spin_unlock(&init_mm.page_table_lock);
+	flush_tlb_all();
+	return error;
+}
+#endif /* USE_HPPA_IOREMAP */
+
+/*
+ * Generic mapping function (not visible outside):
+ */
+
+/*
+ * Remap an arbitrary physical address space into the kernel virtual
+ * address space. Needed when the kernel wants to access high addresses
+ * directly.
+ *
+ * NOTE! We need to allow non-page-aligned mappings too: we will obviously
+ * have to convert them into an offset in a page-aligned mapping, but the
+ * caller shouldn't need to know that small detail.
+ */
+void * __ioremap(unsigned long phys_addr, unsigned long size, unsigned long flags)
+{
+#if !(USE_HPPA_IOREMAP)
+
+	unsigned long end = phys_addr + size - 1;
+	/* Support EISA addresses */
+	if ((phys_addr >= 0x00080000 && end < 0x000fffff)
+			|| (phys_addr >= 0x00500000 && end < 0x03bfffff)) {
+		phys_addr |= 0xfc000000;
+	}
+
+	return (void *)phys_addr;
+
+#else
+	void * addr;
+	struct vm_struct * area;
+	unsigned long offset, last_addr;
+
+	/* Don't allow wraparound or zero size */
+	last_addr = phys_addr + size - 1;
+	if (!size || last_addr < phys_addr)
+		return NULL;
+
+	/*
+	 * Don't allow anybody to remap normal RAM that we're using..
+	 */
+	if (phys_addr < virt_to_phys(high_memory)) {
+		char *t_addr, *t_end;
+		struct page *page;
+
+		t_addr = __va(phys_addr);
+		t_end = t_addr + (size - 1);
+	   
+		for(page = virt_to_page(t_addr); page <= virt_to_page(t_end); page++)
+			if(!PageReserved(page))
+				return NULL;
+	}
+
+	/*
+	 * Mappings have to be page-aligned
+	 */
+	offset = phys_addr & ~PAGE_MASK;
+	phys_addr &= PAGE_MASK;
+	size = PAGE_ALIGN(last_addr) - phys_addr;
+
+	/*
+	 * Ok, go for it..
+	 */
+	area = get_vm_area(size, VM_IOREMAP);
+	if (!area)
+		return NULL;
+	addr = area->addr;
+	if (remap_area_pages(VMALLOC_VMADDR(addr), phys_addr, size, flags)) {
+		vfree(addr);
+		return NULL;
+	}
+	return (void *) (offset + (char *)addr);
+#endif
+}
+
+void iounmap(void *addr)
+{
+#if !(USE_HPPA_IOREMAP)
+	return;
+#else
+	if (addr > high_memory)
+		return vfree((void *) (PAGE_MASK & (unsigned long) addr));
+#endif
+}

arch/parisc/mm/pa11.c

-/* $Id: pa11.c,v 1.1 1999/03/17 01:05:41 pjlahaie Exp $
- *
- * pa11.c: PA 1.1 specific mmu/cache code.
- *
- */
-#include <linux/init.h>
-#include <linux/kernel.h>
-#include <linux/sched.h>
-#include <linux/mm.h>
-
-#include <asm/page.h>
-#include <asm/pgtable.h>
-#include <asm/system.h>
-#include <asm/sgialib.h>
-#include <asm/mmu_context.h>
-
-extern unsigned long mips_tlb_entries;
-
-/* page functions */
-void pa11_clear_page(unsigned long page)
-{
-}
-
-static void pa11_copy_page(unsigned long to, unsigned long from)
-{
-}
-
-/* Cache operations. */
-static inline void pa11_flush_cache_all(void) { }
-static void pa11_flush_cache_mm(struct mm_struct *mm) { }
-static void pa11_flush_cache_range(struct vm_area_struct *vma,
-				    unsigned long start,
-				    unsigned long end)
-{
-}
-
-static void pa11_flush_cache_page(struct vm_area_struct *vma,
-				   unsigned long page)
-{
-}
-
-static void pa11_flush_page_to_ram(unsigned long page)
-{
-}
-
-static void pa11_flush_cache_sigtramp(unsigned long page)
-{
-}
-
-/* TLB operations. */
-static inline void pa11_flush_tlb_all(void)
-{
-	unsigned long flags;
-	int entry;
-
-	save_and_cli(flags);
-/* Here we will need to flush all the TLBs */
-	restore_flags(flags);
-}
-
-static void pa11_flush_tlb_mm(struct mm_struct *mm)
-{
-/* This is what the MIPS does..  Is it the right thing for PA-RISC? */
-	if(mm == current->mm)
-		pa11_flush_tlb_all();
-}
-
-static void pa11_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
-				  unsigned long end)
-{
-	if (vma == NULL || vma->vm_mm == current->mm)
-		pa11_flush_tlb_all();
-}
-
-static void pa11_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
-{
-	if(vma->vm_mm == current->mm)
-		pa11_flush_tlb_all();
-}
-
-static void pa11_load_pgd(unsigned long pg_dir)
-{
-	unsigned long flags;
-    /* We need to do the right thing here */
-}
-
-/*
- * Initialize new page directory with pointers to invalid ptes
- */
-static void pa11_pgd_init(unsigned long page)
-{
-	unsigned long dummy1, dummy2;
-
-}
-
-static void pa11_update_mmu_cache(struct vm_area_struct * vma,
-				   unsigned long address, pte_t pte)
-{
-	pa11_flush_tlb_page(vma, address);
-}
-
-static void pa11_show_regs(struct pt_regs * regs)
-{
-	/*
-	 * Saved main processor registers
-	 */
-	printk("$0 : %08x %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       0, (unsigned long) regs->regs[1], (unsigned long) regs->regs[2],
-	       (unsigned long) regs->regs[3], (unsigned long) regs->regs[4],
-	       (unsigned long) regs->regs[5], (unsigned long) regs->regs[6],
-	       (unsigned long) regs->regs[7]);
-	printk("$8 : %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[8], (unsigned long) regs->regs[9],
-	       (unsigned long) regs->regs[10], (unsigned long) regs->regs[11],
-               (unsigned long) regs->regs[12], (unsigned long) regs->regs[13],
-	       (unsigned long) regs->regs[14], (unsigned long) regs->regs[15]);
-	printk("$16: %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[16], (unsigned long) regs->regs[17],
-	       (unsigned long) regs->regs[18], (unsigned long) regs->regs[19],
-               (unsigned long) regs->regs[20], (unsigned long) regs->regs[21],
-	       (unsigned long) regs->regs[22], (unsigned long) regs->regs[23]);
-	printk("$24: %08lx %08lx                   %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[24], (unsigned long) regs->regs[25],
-	       (unsigned long) regs->regs[28], (unsigned long) regs->regs[29],
-               (unsigned long) regs->regs[30], (unsigned long) regs->regs[31]);
-
-	/*
-	 * Saved cp0 registers
-	 */
-	printk("epc  : %08lx    %s\nStatus: %08x\nCause : %08x\n",
-	       (unsigned long) regs->cp0_epc, print_tainted(),
-	       (unsigned int) regs->cp0_status,
-	       (unsigned int) regs->cp0_cause);
-}
-
-static int pa11_user_mode(struct pt_regs *regs)
-{
-       /* Return user mode stuff?? */
-}
-
-__initfunc(void ld_mmu_pa11(void))
-{
-
-    /* Taken directly from the MIPS arch..  Lots of bad things here */
-	clear_page = pa11_clear_page;
-	copy_page = pa11_copy_page;
-
-	flush_cache_all = pa11_flush_cache_all;
-	flush_cache_mm = pa11_flush_cache_mm;
-	flush_cache_range = pa11_flush_cache_range;
-	flush_cache_page = pa11_flush_cache_page;
-	flush_cache_sigtramp = pa11_flush_cache_sigtramp;
-	flush_page_to_ram = pa11_flush_page_to_ram;
-
-	flush_tlb_all = pa11_flush_tlb_all;
-	flush_tlb_mm = pa11_flush_tlb_mm;
-	flush_tlb_range = pa11_flush_tlb_range;
-	flush_tlb_page = pa11_flush_tlb_page;
-	pa11_asid_setup();
-
-	load_pgd = pa11_load_pgd;
-	pgd_init = pa11_pgd_init;
-	update_mmu_cache = pa11_update_mmu_cache;
-
-	show_regs = pa11_show_regs;
-    
-        add_wired_entry = pa11_add_wired_entry;
-
-	user_mode = pa11_user_mode;
-	flush_tlb_all();
-}

arch/parisc/mm/pa20.c

-/* $Id: pa20.c,v 1.1 1999/03/17 01:05:41 pjlahaie Exp $
- *
- * pa20.c: PA 2.0 specific mmu/cache code.
- *
- */
-#include <linux/init.h>
-#include <linux/kernel.h>
-#include <linux/sched.h>
-#include <linux/mm.h>
-
-#include <asm/page.h>
-#include <asm/pgtable.h>
-#include <asm/system.h>
-#include <asm/sgialib.h>
-#include <asm/mmu_context.h>
-
-extern unsigned long mips_tlb_entries;
-
-/* page functions */
-void pa20_clear_page(unsigned long page)
-{
-}
-
-static void pa20_copy_page(unsigned long to, unsigned long from)
-{
-}
-
-/* Cache operations. */
-static inline void pa20_flush_cache_all(void) { }
-static void pa20_flush_cache_mm(struct mm_struct *mm) { }
-static void pa20_flush_cache_range(struct vm_area_struct *vma,
-				    unsigned long start,
-				    unsigned long end)
-{
-}
-
-static void pa20_flush_cache_page(struct vm_area_struct *vma,
-				   unsigned long page)
-{
-}
-
-static void pa20_flush_page_to_ram(unsigned long page)
-{
-}
-
-static void pa20_flush_cache_sigtramp(unsigned long page)
-{
-}
-
-/* TLB operations. */
-static inline void pa20_flush_tlb_all(void)
-{
-	unsigned long flags;
-	int entry;
-
-	save_and_cli(flags);
-/* Here we will need to flush all the TLBs */
-	restore_flags(flags);
-}
-
-static void pa20_flush_tlb_mm(struct mm_struct *mm)
-{
-/* This is what the MIPS does..  Is it the right thing for PA-RISC? */
-	if(mm == current->mm)
-		pa20_flush_tlb_all();
-}
-
-static void pa20_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
-				  unsigned long end)
-{
-	if (vma == NULL || vma->vm_mm == current->mm)
-		pa20_flush_tlb_all();
-}
-
-static void pa20_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
-{
-	if(vma->vm_mm == current->mm)
-		pa20_flush_tlb_all();
-}
-
-static void pa20_load_pgd(unsigned long pg_dir)
-{
-	unsigned long flags;
-    /* We need to do the right thing here */
-}
-
-/*
- * Initialize new page directory with pointers to invalid ptes
- */
-static void pa20_pgd_init(unsigned long page)
-{
-	unsigned long dummy1, dummy2;
-
-}
-
-static void pa20_update_mmu_cache(struct vm_area_struct * vma,
-				   unsigned long address, pte_t pte)
-{
-	pa20_flush_tlb_page(vma, address);
-}
-
-static void pa20_show_regs(struct pt_regs * regs)
-{
-	/*
-	 * Saved main processor registers
-	 */
-	printk("$0 : %08x %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       0, (unsigned long) regs->regs[1], (unsigned long) regs->regs[2],
-	       (unsigned long) regs->regs[3], (unsigned long) regs->regs[4],
-	       (unsigned long) regs->regs[5], (unsigned long) regs->regs[6],
-	       (unsigned long) regs->regs[7]);
-	printk("$8 : %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[8], (unsigned long) regs->regs[9],
-	       (unsigned long) regs->regs[10], (unsigned long) regs->regs[11],
-               (unsigned long) regs->regs[12], (unsigned long) regs->regs[13],
-	       (unsigned long) regs->regs[14], (unsigned long) regs->regs[15]);
-	printk("$16: %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[16], (unsigned long) regs->regs[17],
-	       (unsigned long) regs->regs[18], (unsigned long) regs->regs[19],
-               (unsigned long) regs->regs[20], (unsigned long) regs->regs[21],
-	       (unsigned long) regs->regs[22], (unsigned long) regs->regs[23]);
-	printk("$24: %08lx %08lx                   %08lx %08lx %08lx %08lx\n",
-	       (unsigned long) regs->regs[24], (unsigned long) regs->regs[25],
-	       (unsigned long) regs->regs[28], (unsigned long) regs->regs[29],
-               (unsigned long) regs->regs[30], (unsigned long) regs->regs[31]);
-
-	/*
-	 * Saved cp0 registers
-	 */
-	printk("epc  : %08lx    %s\nStatus: %08x\nCause : %08x\n",
-	       (unsigned long) regs->cp0_epc, print_tainted(),
-	       (unsigned int) regs->cp0_status,
-	       (unsigned int) regs->cp0_cause);
-}
-
-static int pa20_user_mode(struct pt_regs *regs)
-{
-       /* Return user mode stuff?? */
-}
-
-__initfunc(void ld_mmu_pa20(void))
-{
-
-    /* Taken directly from the MIPS arch..  Lots of bad things here */
-	clear_page = pa20_clear_page;
-	copy_page = pa20_copy_page;
-
-	flush_cache_all = pa20_flush_cache_all;
-	flush_cache_mm = pa20_flush_cache_mm;
-	flush_cache_range = pa20_flush_cache_range;
-	flush_cache_page = pa20_flush_cache_page;
-	flush_cache_sigtramp = pa20_flush_cache_sigtramp;
-	flush_page_to_ram = pa20_flush_page_to_ram;
-
-	flush_tlb_all = pa20_flush_tlb_all;
-	flush_tlb_mm = pa20_flush_tlb_mm;
-	flush_tlb_range = pa20_flush_tlb_range;
-	flush_tlb_page = pa20_flush_tlb_page;
-	pa20_asid_setup();
-
-	load_pgd = pa20_load_pgd;
-	pgd_init = pa20_pgd_init;
-	update_mmu_cache = pa20_update_mmu_cache;
-
-	show_regs = pa20_show_regs;
-    
-        add_wired_entry = pa20_add_wired_entry;
-
-	user_mode = pa20_user_mode;
-	flush_tlb_all();
-}