memory.c

LINUX1.0源代码,代码条理清晰

/*
 *  linux/mm/memory.c
 *
 *  Copyright (C) 1991, 1992  Linus Torvalds
 */

/*
 * demand-loading started 01.12.91 - seems it is high on the list of
 * things wanted, and it should be easy to implement. - Linus
 */

/*
 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
 * pages started 02.12.91, seems to work. - Linus.
 *
 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
 * would have taken more than the 6M I have free, but it worked well as
 * far as I could see.
 *
 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
 */

/*
 * Real VM (paging to/from disk) started 18.12.91. Much more work and
 * thought has to go into this. Oh, well..
 * 19.12.91  -  works, somewhat. Sometimes I get faults, don't know why.
 *		Found it. Everything seems to work now.
 * 20.12.91  -  Ok, making the swap-device changeable like the root.
 */

#include <asm/system.h>
#include <linux/config.h>

#include <linux/signal.h>
#include <linux/sched.h>
#include <linux/head.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>

unsigned long high_memory = 0;

extern unsigned long pg0[1024];		/* page table for 0-4MB for everybody */

extern void sound_mem_init(void);
extern void die_if_kernel(char *,struct pt_regs *,long);

int nr_swap_pages = 0;
int nr_free_pages = 0;
unsigned long free_page_list = 0;
/*
 * The secondary free_page_list is used for malloc() etc things that
 * may need pages during interrupts etc. Normal get_free_page() operations
 * don't touch it, so it stays as a kind of "panic-list", that can be
 * accessed when all other mm tricks have failed.
 */
int nr_secondary_pages = 0;
unsigned long secondary_page_list = 0;

#define copy_page(from,to) \
__asm__("cld ; rep ; movsl": :"S" (from),"D" (to),"c" (1024):"cx","di","si")

unsigned short * mem_map = NULL;

#define CODE_SPACE(addr,p) ((addr) < (p)->end_code)

/*
 * oom() prints a message (so that the user knows why the process died),
 * and gives the process an untrappable SIGSEGV.
 */
void oom(struct task_struct * task)
{
	printk("\nout of memory\n");
	task->sigaction[SIGKILL-1].sa_handler = NULL;
	task->blocked &= ~(1<<(SIGKILL-1));
	send_sig(SIGKILL,task,1);
}

static void free_one_table(unsigned long * page_dir)
{
	int j;
	unsigned long pg_table = *page_dir;
	unsigned long * page_table;

	if (!pg_table)
		return;
	*page_dir = 0;
	if (pg_table >= high_memory || !(pg_table & PAGE_PRESENT)) {
		printk("Bad page table: [%p]=%08lx\n",page_dir,pg_table);
		return;
	}
	if (mem_map[MAP_NR(pg_table)] & MAP_PAGE_RESERVED)
		return;
	page_table = (unsigned long *) (pg_table & PAGE_MASK);
	for (j = 0 ; j < PTRS_PER_PAGE ; j++,page_table++) {
		unsigned long pg = *page_table;
		
		if (!pg)
			continue;
		*page_table = 0;
		if (pg & PAGE_PRESENT)
			free_page(PAGE_MASK & pg);
		else
			swap_free(pg);
	}
	free_page(PAGE_MASK & pg_table);
}

/*
 * This function clears all user-level page tables of a process - this
 * is needed by execve(), so that old pages aren't in the way. Note that
 * unlike 'free_page_tables()', this function still leaves a valid
 * page-table-tree in memory: it just removes the user pages. The two
 * functions are similar, but there is a fundamental difference.
 */
void clear_page_tables(struct task_struct * tsk)
{
	int i;
	unsigned long pg_dir;
	unsigned long * page_dir;

	if (!tsk)
		return;
	if (tsk == task[0])
		panic("task[0] (swapper) doesn't support exec()\n");
	pg_dir = tsk->tss.cr3;
	page_dir = (unsigned long *) pg_dir;
	if (!page_dir || page_dir == swapper_pg_dir) {
		printk("Trying to clear kernel page-directory: not good\n");
		return;
	}
	if (mem_map[MAP_NR(pg_dir)] > 1) {
		unsigned long * new_pg;

		if (!(new_pg = (unsigned long*) get_free_page(GFP_KERNEL))) {
			oom(tsk);
			return;
		}
		for (i = 768 ; i < 1024 ; i++)
			new_pg[i] = page_dir[i];
		free_page(pg_dir);
		tsk->tss.cr3 = (unsigned long) new_pg;
		return;
	}
	for (i = 0 ; i < 768 ; i++,page_dir++)
		free_one_table(page_dir);
	invalidate();
	return;
}

/*
 * This function frees up all page tables of a process when it exits.
 */
void free_page_tables(struct task_struct * tsk)
{
	int i;
	unsigned long pg_dir;
	unsigned long * page_dir;

	if (!tsk)
		return;
	if (tsk == task[0]) {
		printk("task[0] (swapper) killed: unable to recover\n");
		panic("Trying to free up swapper memory space");
	}
	pg_dir = tsk->tss.cr3;
	if (!pg_dir || pg_dir == (unsigned long) swapper_pg_dir) {
		printk("Trying to free kernel page-directory: not good\n");
		return;
	}
	tsk->tss.cr3 = (unsigned long) swapper_pg_dir;
	if (tsk == current)
		__asm__ __volatile__("movl %0,%%cr3": :"a" (tsk->tss.cr3));
	if (mem_map[MAP_NR(pg_dir)] > 1) {
		free_page(pg_dir);
		return;
	}
	page_dir = (unsigned long *) pg_dir;
	for (i = 0 ; i < PTRS_PER_PAGE ; i++,page_dir++)
		free_one_table(page_dir);
	free_page(pg_dir);
	invalidate();
}

/*
 * clone_page_tables() clones the page table for a process - both
 * processes will have the exact same pages in memory. There are
 * probably races in the memory management with cloning, but we'll
 * see..
 */
int clone_page_tables(struct task_struct * tsk)
{
	unsigned long pg_dir;

	pg_dir = current->tss.cr3;
	mem_map[MAP_NR(pg_dir)]++;
	tsk->tss.cr3 = pg_dir;
	return 0;
}

/*
 * copy_page_tables() just copies the whole process memory range:
 * note the special handling of RESERVED (ie kernel) pages, which
 * means that they are always shared by all processes.
 */
int copy_page_tables(struct task_struct * tsk)
{
	int i;
	unsigned long old_pg_dir, *old_page_dir;
	unsigned long new_pg_dir, *new_page_dir;

	if (!(new_pg_dir = get_free_page(GFP_KERNEL)))
		return -ENOMEM;
	old_pg_dir = current->tss.cr3;
	tsk->tss.cr3 = new_pg_dir;
	old_page_dir = (unsigned long *) old_pg_dir;
	new_page_dir = (unsigned long *) new_pg_dir;
	for (i = 0 ; i < PTRS_PER_PAGE ; i++,old_page_dir++,new_page_dir++) {
		int j;
		unsigned long old_pg_table, *old_page_table;
		unsigned long new_pg_table, *new_page_table;

		old_pg_table = *old_page_dir;
		if (!old_pg_table)
			continue;
		if (old_pg_table >= high_memory || !(old_pg_table & PAGE_PRESENT)) {
			printk("copy_page_tables: bad page table: "
				"probable memory corruption");
			*old_page_dir = 0;
			continue;
		}
		if (mem_map[MAP_NR(old_pg_table)] & MAP_PAGE_RESERVED) {
			*new_page_dir = old_pg_table;
			continue;
		}
		if (!(new_pg_table = get_free_page(GFP_KERNEL))) {
			free_page_tables(tsk);
			return -ENOMEM;
		}
		old_page_table = (unsigned long *) (PAGE_MASK & old_pg_table);
		new_page_table = (unsigned long *) (PAGE_MASK & new_pg_table);
		for (j = 0 ; j < PTRS_PER_PAGE ; j++,old_page_table++,new_page_table++) {
			unsigned long pg;
			pg = *old_page_table;
			if (!pg)
				continue;
			if (!(pg & PAGE_PRESENT)) {
				*new_page_table = swap_duplicate(pg);
				continue;
			}
			if ((pg & (PAGE_RW | PAGE_COW)) == (PAGE_RW | PAGE_COW))
				pg &= ~PAGE_RW;
			*new_page_table = pg;
			if (mem_map[MAP_NR(pg)] & MAP_PAGE_RESERVED)
				continue;
			*old_page_table = pg;
			mem_map[MAP_NR(pg)]++;
		}
		*new_page_dir = new_pg_table | PAGE_TABLE;
	}
	invalidate();
	return 0;
}

/*
 * a more complete version of free_page_tables which performs with page
 * granularity.
 */
int unmap_page_range(unsigned long from, unsigned long size)
{
	unsigned long page, page_dir;
	unsigned long *page_table, *dir;
	unsigned long poff, pcnt, pc;

	if (from & ~PAGE_MASK) {
		printk("unmap_page_range called with wrong alignment\n");
		return -EINVAL;
	}
	size = (size + ~PAGE_MASK) >> PAGE_SHIFT;
	dir = PAGE_DIR_OFFSET(current->tss.cr3,from);
	poff = (from >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
	if ((pcnt = PTRS_PER_PAGE - poff) > size)
		pcnt = size;

	for ( ; size > 0; ++dir, size -= pcnt,
	     pcnt = (size > PTRS_PER_PAGE ? PTRS_PER_PAGE : size)) {
		if (!(page_dir = *dir))	{
			poff = 0;
			continue;
		}
		if (!(page_dir & PAGE_PRESENT)) {
			printk("unmap_page_range: bad page directory.");
			continue;
		}
		page_table = (unsigned long *)(PAGE_MASK & page_dir);
		if (poff) {
			page_table += poff;
			poff = 0;
		}
		for (pc = pcnt; pc--; page_table++) {
			if ((page = *page_table) != 0) {
				*page_table = 0;
				if (1 & page) {
					if (!(mem_map[MAP_NR(page)] & MAP_PAGE_RESERVED))
						if (current->rss > 0)
							--current->rss;
					free_page(PAGE_MASK & page);
				} else
					swap_free(page);
			}
		}
		if (pcnt == PTRS_PER_PAGE) {
			*dir = 0;
			free_page(PAGE_MASK & page_dir);
		}
	}
	invalidate();
	return 0;
}

int zeromap_page_range(unsigned long from, unsigned long size, int mask)
{
	unsigned long *page_table, *dir;
	unsigned long poff, pcnt;
	unsigned long page;

	if (mask) {
		if ((mask & (PAGE_MASK|PAGE_PRESENT)) != PAGE_PRESENT) {
			printk("zeromap_page_range: mask = %08x\n",mask);
			return -EINVAL;
		}
		mask |= ZERO_PAGE;
	}
	if (from & ~PAGE_MASK) {
		printk("zeromap_page_range: from = %08lx\n",from);
		return -EINVAL;
	}
	dir = PAGE_DIR_OFFSET(current->tss.cr3,from);
	size = (size + ~PAGE_MASK) >> PAGE_SHIFT;
	poff = (from >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
	if ((pcnt = PTRS_PER_PAGE - poff) > size)
		pcnt = size;

	while (size > 0) {
		if (!(PAGE_PRESENT & *dir)) {
				/* clear page needed here?  SRB. */
			if (!(page_table = (unsigned long*) get_free_page(GFP_KERNEL))) {
				invalidate();
				return -ENOMEM;
			}
			if (PAGE_PRESENT & *dir) {
				free_page((unsigned long) page_table);
				page_table = (unsigned long *)(PAGE_MASK & *dir++);
			} else
				*dir++ = ((unsigned long) page_table) | PAGE_TABLE;
		} else
			page_table = (unsigned long *)(PAGE_MASK & *dir++);
		page_table += poff;
		poff = 0;
		for (size -= pcnt; pcnt-- ;) {
			if ((page = *page_table) != 0) {
				*page_table = 0;
				if (page & PAGE_PRESENT) {
					if (!(mem_map[MAP_NR(page)] & MAP_PAGE_RESERVED))
						if (current->rss > 0)
							--current->rss;
					free_page(PAGE_MASK & page);
				} else
					swap_free(page);
			}
			*page_table++ = mask;
		}
		pcnt = (size > PTRS_PER_PAGE ? PTRS_PER_PAGE : size);
	}
	invalidate();
	return 0;
}

/*
 * maps a range of physical memory into the requested pages. the old
 * mappings are removed. any references to nonexistent pages results
 * in null mappings (currently treated as "copy-on-access")
 */
int remap_page_range(unsigned long from, unsigned long to, unsigned long size, int mask)
{
	unsigned long *page_table, *dir;
	unsigned long poff, pcnt;
	unsigned long page;

	if (mask) {
		if ((mask & (PAGE_MASK|PAGE_PRESENT)) != PAGE_PRESENT) {
			printk("remap_page_range: mask = %08x\n",mask);
			return -EINVAL;
		}
	}
	if ((from & ~PAGE_MASK) || (to & ~PAGE_MASK)) {
		printk("remap_page_range: from = %08lx, to=%08lx\n",from,to);
		return -EINVAL;
	}
	dir = PAGE_DIR_OFFSET(current->tss.cr3,from);
	size = (size + ~PAGE_MASK) >> PAGE_SHIFT;
	poff = (from >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
	if ((pcnt = PTRS_PER_PAGE - poff) > size)
		pcnt = size;

	while (size > 0) {