Linux 1.0 memory.c 学习日记

/*
 *  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) {
        if (!(PAGE_PRESENT & *dir)) {
            /* clearing page here, needed?  SRB. */
            if (!(page_table = (unsigned long*) get_free_page(GFP_KERNEL))) {
                invalidate();
                return -1;
            }
            *dir++ = ((unsigned long) page_table) | PAGE_TABLE;
        }
        else
            page_table = (unsigned long *)(PAGE_MASK & *dir++);
        if (poff) {
            page_table += poff;
            poff = 0;
        }

        for (size -= pcnt; pcnt-- ;) {
            if ((page = *page_table) != 0) {
                *page_table = 0;
                if (PAGE_PRESENT & 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);
            }

            /*
             * the first condition should return an invalid access
             * when the page is referenced. current assumptions
             * cause it to be treated as demand allocation in some
             * cases.
             */
            if (!mask)
                *page_table++ = 0;    /* not present */
            else if (to >= high_memory)
                *page_table++ = (to | mask);
            else if (!mem_map[MAP_NR(to)])
                *page_table++ = 0;    /* not present */
            else {
                *page_table++ = (to | mask);
                if (!(mem_map[MAP_NR(to)] & MAP_PAGE_RESERVED)) {
                    ++current->rss;
                    mem_map[MAP_NR(to)]++;
                }
            }
            to += PAGE_SIZE;
        }
        pcnt = (size > PTRS_PER_PAGE ? PTRS_PER_PAGE : size);
    }
    invalidate();
    return 0;
}

/*
 * This function puts a page in memory at the wanted address.
 * It returns the physical address of the page gotten, 0 if
 * out of memory (either when trying to access page-table or
 * page.)
 */
unsigned long put_page(struct task_struct * tsk,unsigned long page,
    unsigned long address,int prot)
{
    unsigned long *page_table;

    if ((prot & (PAGE_MASK|PAGE_PRESENT)) != PAGE_PRESENT)
        printk("put_page: prot = %08x\n",prot);
    if (page >= high_memory) {
        printk("put_page: trying to put page %08lx at %08lx\n",page,address);
        return 0;
    }
    page_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    if ((*page_table) & PAGE_PRESENT)
        page_table = (unsigned long *) (PAGE_MASK & *page_table);
    else {
        printk("put_page: bad page directory entry\n");
        oom(tsk);
        *page_table = BAD_PAGETABLE | PAGE_TABLE;
        return 0;
    }
    page_table += (address >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
    if (*page_table) {
        printk("put_page: page already exists\n");
        *page_table = 0;
        invalidate();
    }
    *page_table = page | prot;
/* no need for invalidate */
    return page;
}

/*
 * The previous function doesn't work very well if you also want to mark
 * the page dirty: exec.c wants this, as it has earlier changed the page,
 * and we want the dirty-status to be correct (for VM). Thus the same
 * routine, but this time we mark it dirty too.
 */
unsigned long put_dirty_page(struct task_struct * tsk, unsigned long page, unsigned long address)
{
    unsigned long tmp, *page_table;

    if (page >= high_memory)
        printk("put_dirty_page: trying to put page %08lx at %08lx\n",page,address);
    if (mem_map[MAP_NR(page)] != 1)
        printk("mem_map disagrees with %08lx at %08lx\n",page,address);
    page_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    if (PAGE_PRESENT & *page_table)
        page_table = (unsigned long *) (PAGE_MASK & *page_table);
    else {
        if (!(tmp = get_free_page(GFP_KERNEL)))
            return 0;
        if (PAGE_PRESENT & *page_table) {
            free_page(tmp);
            page_table = (unsigned long *) (PAGE_MASK & *page_table);
        } else {
            *page_table = tmp | PAGE_TABLE;
            page_table = (unsigned long *) tmp;
        }
    }
    page_table += (address >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
    if (*page_table) {
        printk("put_dirty_page: page already exists\n");
        *page_table = 0;
        invalidate();
    }
    *page_table = page | (PAGE_DIRTY | PAGE_PRIVATE);
/* no need for invalidate */
    return page;
}

/*
 * This routine handles present pages, when users try to write
 * to a shared page. It is done by copying the page to a new address
 * and decrementing the shared-page counter for the old page.
 *
 * Note that we do many checks twice (look at do_wp_page()), as
 * we have to be careful about race-conditions.
 *
 * Goto-purists beware: the only reason for goto's here is that it results
 * in better assembly code.. The "default" path will see no jumps at all.
 */
static void __do_wp_page(unsigned long error_code, unsigned long address,
    struct task_struct * tsk, unsigned long user_esp)
{
    unsigned long *pde, pte, old_page, prot;
    unsigned long new_page;

    new_page = __get_free_page(GFP_KERNEL);
    pde = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    pte = *pde;
    if (!(pte & PAGE_PRESENT))
        goto end_wp_page;
    if ((pte & PAGE_TABLE) != PAGE_TABLE || pte >= high_memory)
        goto bad_wp_pagetable;
    pte &= PAGE_MASK;
    pte += PAGE_PTR(address);
    old_page = *(unsigned long *) pte;
    if (!(old_page & PAGE_PRESENT))
        goto end_wp_page;
    if (old_page >= high_memory)
        goto bad_wp_page;
    if (old_page & PAGE_RW)
        goto end_wp_page;
    tsk->min_flt++;
    prot = (old_page & ~PAGE_MASK) | PAGE_RW;
    old_page &= PAGE_MASK;
    if (mem_map[MAP_NR(old_page)] != 1) {
        if (new_page) {
            if (mem_map[MAP_NR(old_page)] & MAP_PAGE_RESERVED)
                ++tsk->rss;
            copy_page(old_page,new_page);
            *(unsigned long *) pte = new_page | prot;
            free_page(old_page);
            invalidate();
            return;
        }
        free_page(old_page);
        oom(tsk);
        *(unsigned long *) pte = BAD_PAGE | prot;
        invalidate();
        return;
    }
    *(unsigned long *) pte |= PAGE_RW;
    invalidate();
    if (new_page)
        free_page(new_page);
    return;
bad_wp_page:
    printk("do_wp_page: bogus page at address %08lx (%08lx)\n",address,old_page);
    *(unsigned long *) pte = BAD_PAGE | PAGE_SHARED;
    send_sig(SIGKILL, tsk, 1);
    goto end_wp_page;
bad_wp_pagetable:
    printk("do_wp_page: bogus page-table at address %08lx (%08lx)\n",address,pte);
    *pde = BAD_PAGETABLE | PAGE_TABLE;
    send_sig(SIGKILL, tsk, 1);
end_wp_page:
    if (new_page)
        free_page(new_page);
    return;
}

/*
 * check that a page table change is actually needed, and call
 * the low-level function only in that case..
 */
void do_wp_page(unsigned long error_code, unsigned long address,
    struct task_struct * tsk, unsigned long user_esp)
{
    unsigned long page;
    unsigned long * pg_table;

    pg_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    page = *pg_table;
    if (!page)
        return;
    if ((page & PAGE_PRESENT) && page < high_memory) {
        pg_table = (unsigned long *) ((page & PAGE_MASK) + PAGE_PTR(address));
        page = *pg_table;
        if (!(page & PAGE_PRESENT))
            return;
        if (page & PAGE_RW)
            return;
        if (!(page & PAGE_COW)) {
            if (user_esp && tsk == current) {
                current->tss.cr2 = address;
                current->tss.error_code = error_code;
                current->tss.trap_no = 14;
                send_sig(SIGSEGV, tsk, 1);
                return;
            }
        }
        if (mem_map[MAP_NR(page)] == 1) {
            *pg_table |= PAGE_RW | PAGE_DIRTY;
            invalidate();
            return;
        }
        __do_wp_page(error_code, address, tsk, user_esp);
        return;
    }
    printk("bad page directory entry %08lx\n",page);
    *pg_table = 0;
}

int __verify_write(unsigned long start, unsigned long size)
{
    size--;
    size += start & ~PAGE_MASK;
    size >>= PAGE_SHIFT;
    start &= PAGE_MASK;
    do {
        do_wp_page(1,start,current,0);
        start += PAGE_SIZE;
    } while (size--);
    return 0;
}

static inline void get_empty_page(struct task_struct * tsk, unsigned long address)
{
    unsigned long tmp;

    if (!(tmp = get_free_page(GFP_KERNEL))) {
        oom(tsk);
        tmp = BAD_PAGE;
    }
    if (!put_page(tsk,tmp,address,PAGE_PRIVATE))
        free_page(tmp);
}

/*
 * try_to_share() checks the page at address "address" in the task "p",
 * to see if it exists, and if it is clean. If so, share it with the current
 * task.
 *
 * NOTE! This assumes we have checked that p != current, and that they
 * share the same executable or library.
 *
 * We may want to fix this to allow page sharing for PIC pages at different
 * addresses so that ELF will really perform properly. As long as the vast
 * majority of sharable libraries load at fixed addresses this is not a
 * big concern. Any sharing of pages between the buffer cache and the
 * code space reduces the need for this as well.  - ERY
 */
static int try_to_share(unsigned long address, struct task_struct * tsk,
    struct task_struct * p, unsigned long error_code, unsigned long newpage)
{
    unsigned long from;
    unsigned long to;
    unsigned long from_page;
    unsigned long to_page;

    from_page = (unsigned long)PAGE_DIR_OFFSET(p->tss.cr3,address);
    to_page = (unsigned long)PAGE_DIR_OFFSET(tsk->tss.cr3,address);
/* is there a page-directory at from? */
    from = *(unsigned long *) from_page;
    if (!(from & PAGE_PRESENT))
        return 0;
    from &= PAGE_MASK;
    from_page = from + PAGE_PTR(address);
    from = *(unsigned long *) from_page;
/* is the page clean and present? */
    if ((from & (PAGE_PRESENT | PAGE_DIRTY)) != PAGE_PRESENT)
        return 0;
    if (from >= high_memory)
        return 0;
    if (mem_map[MAP_NR(from)] & MAP_PAGE_RESERVED)
        return 0;
/* is the destination ok? */
    to = *(unsigned long *) to_page;
    if (!(to & PAGE_PRESENT))
        return 0;
    to &= PAGE_MASK;
    to_page = to + PAGE_PTR(address);
    if (*(unsigned long *) to_page)
        return 0;
/* share them if read - do COW immediately otherwise */
    if (error_code & PAGE_RW) {
        if(!newpage)    /* did the page exist?  SRB. */
            return 0;
        copy_page((from & PAGE_MASK),newpage);
        to = newpage | PAGE_PRIVATE;
    } else {
        mem_map[MAP_NR(from)]++;
        from &= ~PAGE_RW;
        to = from;
        if(newpage)    /* only if it existed. SRB. */
            free_page(newpage);
    }
    *(unsigned long *) from_page = from;
    *(unsigned long *) to_page = to;
    invalidate();
    return 1;
}

/*
 * share_page() tries to find a process that could share a page with
 * the current one. Address is the address of the wanted page relative
 * to the current data space.
 *
 * We first check if it is at all feasible by checking executable->i_count.
 * It should be >1 if there are other tasks sharing this inode.
 */
int share_page(struct vm_area_struct * area, struct task_struct * tsk,
    struct inode * inode,
    unsigned long address, unsigned long error_code, unsigned long newpage)
{
    struct task_struct ** p;

    if (!inode || inode->i_count < 2 || !area->vm_ops)
        return 0;
    for (p = &LAST_TASK ; p > &FIRST_TASK ; --p) {
        if (!*p)
            continue;
        if (tsk == *p)
            continue;
        if (inode != (*p)->executable) {
              if(!area) continue;
            /* Now see if there is something in the VMM that
               we can share pages with */
            if(area){
              struct vm_area_struct * mpnt;
              for (mpnt = (*p)->mmap; mpnt; mpnt = mpnt->vm_next) {
                if (mpnt->vm_ops == area->vm_ops &&
                   mpnt->vm_inode->i_ino == area->vm_inode->i_ino&&
                   mpnt->vm_inode->i_dev == area->vm_inode->i_dev){
                  if (mpnt->vm_ops->share(mpnt, area, address))
                break;
                };
              };
              if (!mpnt) continue;  /* Nope.  Nuthin here */
            };
        }
        if (try_to_share(address,tsk,*p,error_code,newpage))
            return 1;
    }
    return 0;
}

/*
 * fill in an empty page-table if none exists.
 */
static inline unsigned long get_empty_pgtable(struct task_struct * tsk,unsigned long address)
{
    unsigned long page;
    unsigned long *p;

    p = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    if (PAGE_PRESENT & *p)
        return *p;
    if (*p) {
        printk("get_empty_pgtable: bad page-directory entry \n");
        *p = 0;
    }
    page = get_free_page(GFP_KERNEL);
    p = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
    if (PAGE_PRESENT & *p) {
        free_page(page);
        return *p;
    }
    if (*p) {
        printk("get_empty_pgtable: bad page-directory entry \n");
        *p = 0;
    }
    if (page) {
        *p = page | PAGE_TABLE;
        return *p;
    }
    oom(current);
    *p = BAD_PAGETABLE | PAGE_TABLE;
    return 0;
}

void do_no_page(unsigned long error_code, unsigned long address,
    struct task_struct *tsk, unsigned long user_esp)
{
    unsigned long tmp;
    unsigned long page;
    struct vm_area_struct * mpnt;

    page = get_empty_pgtable(tsk,address);
    if (!page)
        return;
    page &= PAGE_MASK;
    page += PAGE_PTR(address);
    tmp = *(unsigned long *) page;
    if (tmp & PAGE_PRESENT)
        return;
    ++tsk->rss;
    if (tmp) {
        ++tsk->maj_flt;
        swap_in((unsigned long *) page);
        return;
    }
    address &= 0xfffff000;
    tmp = 0;
    for (mpnt = tsk->mmap; mpnt != NULL; mpnt = mpnt->vm_next) {
        if (address < mpnt->vm_start)
            break;
        if (address >= mpnt->vm_end) {
            tmp = mpnt->vm_end;
            continue;
        }
        if (!mpnt->vm_ops || !mpnt->vm_ops->nopage) {
            ++tsk->min_flt;
            get_empty_page(tsk,address);
            return;
        }
        mpnt->vm_ops->nopage(error_code, mpnt, address);
        return;
    }
    if (tsk != current)
        goto ok_no_page;
    if (address >= tsk->end_data && address < tsk->brk)
        goto ok_no_page;
    if (mpnt && mpnt == tsk->stk_vma &&
        address - tmp > mpnt->vm_start - address &&
        tsk->rlim[RLIMIT_STACK].rlim_cur > mpnt->vm_end - address) {
        mpnt->vm_start = address;
        goto ok_no_page;
    }
    tsk->tss.cr2 = address;
    current->tss.error_code = error_code;
    current->tss.trap_no = 14;
    send_sig(SIGSEGV,tsk,1);
    if (error_code & 4)    /* user level access? */
        return;
ok_no_page:
    ++tsk->min_flt;
    get_empty_page(tsk,address);
}

/*
 * This routine handles page faults.  It determines the address,
 * and the problem, and then passes it off to one of the appropriate
 * routines.
 */
asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code)
{
    unsigned long address;
    unsigned long user_esp = 0;
    unsigned int bit;

    /* get the address */
    __asm__("movl %%cr2,%0":"=r" (address));
    if (address < TASK_SIZE) {
        if (error_code & 4) {    /* user mode access? */
            if (regs->eflags & VM_MASK) {
                bit = (address - 0xA0000) >> PAGE_SHIFT;
                if (bit < 32)
                    current->screen_bitmap |= 1 << bit;
            } else
                user_esp = regs->esp;
        }
        if (error_code & 1)
            do_wp_page(error_code, address, current, user_esp);
        else
            do_no_page(error_code, address, current, user_esp);
        return;
    }
    address -= TASK_SIZE;
    if (wp_works_ok < 0 && address == 0 && (error_code & PAGE_PRESENT)) {
        wp_works_ok = 1;
        pg0[0] = PAGE_SHARED;
        printk("This processor honours the WP bit even when in supervisor mode. Good.\n");
        return;
    }
    if (address < PAGE_SIZE) {
        printk("Unable to handle kernel NULL pointer dereference");
        pg0[0] = PAGE_SHARED;
    } else
        printk("Unable to handle kernel paging request");
    printk(" at address %08lx\n",address);
    die_if_kernel("Oops", regs, error_code);
    do_exit(SIGKILL);
}

/*
 * 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 a 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.
 */
unsigned long __bad_pagetable(void)
{
    extern char empty_bad_page_table[PAGE_SIZE];

    __asm__ __volatile__("cld ; rep ; stosl":
        :"a" (BAD_PAGE + PAGE_TABLE),
         "D" ((long) empty_bad_page_table),
         "c" (PTRS_PER_PAGE)
        :"di","cx");
    return (unsigned long) empty_bad_page_table;
}

unsigned long __bad_page(void)
{
    extern char empty_bad_page[PAGE_SIZE];

    __asm__ __volatile__("cld ; rep ; stosl":
        :"a" (0),
         "D" ((long) empty_bad_page),
         "c" (PTRS_PER_PAGE)
        :"di","cx");
    return (unsigned long) empty_bad_page;
}

unsigned long __zero_page(void)
{
    extern char empty_zero_page[PAGE_SIZE];

    __asm__ __volatile__("cld ; rep ; stosl":
        :"a" (0),
         "D" ((long) empty_zero_page),
         "c" (PTRS_PER_PAGE)
        :"di","cx");
    return (unsigned long) empty_zero_page;
}

void show_mem(void)
{
    int i,free = 0,total = 0,reserved = 0;
    int shared = 0;

    printk("Mem-info:\n");
    printk("Free pages:      %6dkB\n",nr_free_pages<<(PAGE_SHIFT-10));
    printk("Secondary pages: %6dkB\n",nr_secondary_pages<<(PAGE_SHIFT-10));
    printk("Free swap:       %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
    i = high_memory >> PAGE_SHIFT;
    while (i-- > 0) {
        total++;
        if (mem_map[i] & MAP_PAGE_RESERVED)
            reserved++;
        else if (!mem_map[i])
            free++;
        else
            shared += mem_map[i]-1;
    }
    printk("%d pages of RAM\n",total);
    printk("%d free pages\n",free);
    printk("%d reserved pages\n",reserved);
    printk("%d pages shared\n",shared);
    show_buffers();
}

/*
 * paging_init() sets up the page tables - note that the first 4MB are
 * already mapped by head.S.
 *
 * This routines also unmaps the page at virtual kernel address 0, so
 * that we can trap those pesky NULL-reference errors in the kernel.
 */
//从0xC0000000~0x3FFFFFFF的1GB虚拟空间为内核态空间。所以，如果物理内存足够大，也最多只能映射1GB的物理内存。
//内核从0xc0000000(即3G)开始的那一段虚拟地址都是是连续映射到从0开始的物理地址的。
//在此版本的kernel中，物理内存的最大值假定为16M。所以，只能从0xC0000000到0xC0000000+16M的虚拟地址，映射到0到16M的物理地址。

//从内核虚拟地址0xc0000000开始的16M虚拟地址都是连续映射到从0~16M的的物理地址的。
//所以显然可知，内核物理地址加上0xC0000000即得到内核虚拟地址；内核虚拟地址减去0xC0000000得到内核物理地址。
unsigned long paging_init(unsigned long start_mem, unsigned long end_mem)
{
    unsigned long * pg_dir;
    unsigned long * pg_table;
    unsigned long tmp;
    unsigned long address;

/*
 * Physical page 0 is special; it's not touched by Linux since BIOS
 * and SMM (for laptops with [34]86/SL chips) may need it.  It is read
 * and write protected to detect null pointer references in the
 * kernel.
 */
#if 0
    memset((void *) 0, 0, PAGE_SIZE);
#endif
    start_mem = PAGE_ALIGN(start_mem); //start_mem为内核源代码的end物理地址的next物理地址
    address = 0; //映射的物理地址是从0开始的
    pg_dir = swapper_pg_dir;   //swapper_pg_dir为head.S中建立的页目录表，它的0项和768项都是页_pg0的地址
    while (address < end_mem) {
        tmp = *(pg_dir + 768);        /* at virtual addr 0xC0000000 */
        if (!tmp) {
            tmp = start_mem | PAGE_TABLE;
            *(pg_dir + 768) = tmp;//即从内核源代码的end物理地址的next物理地址开始分配二级页表
            start_mem += PAGE_SIZE;//二级页表的大小为4096
        }
        //页目录表中，第0项和第768项都指向页_pg0(对应0～4M的物理地址);第1项和769项都指向新分配的二级页表(对应4~8M的物理地址);后面依次类推，直到16M内存map完了。
        *pg_dir = tmp;            /* also map it in at 0x0000000 for init */
        pg_dir++;
        pg_table = (unsigned long *) (tmp & PAGE_MASK);//tmp为二级页表的地址和页表属性相加的结果，所以用PAGE_MASK去掉页表属性即得到二级页表的地址。

        //PTRS_PER_PAGE为1024，所以为新的二级页表pg_table的1024个项分配一个物理页的首地址+页属性。
        //对于第768项，它指向_pg0，所对应的0～4M的物理地址在head.S中已经map过了，这里又重新map了下，不过没关系，还是与head.S中一样映射到0～4M的物理地址。这里paging_init中真正map的就是4~16M的物理内存。

        for (tmp = 0 ; tmp < PTRS_PER_PAGE ; tmp++,pg_table++) {
            if (address < end_mem)    //如果物理地址16M还没有分配完
                *pg_table = address | PAGE_SHARED;
            else
                *pg_table = 0;
            address += PAGE_SIZE; //每分配一项，物理地址就减少4096
        }
    }
    invalidate();
    return start_mem;

}

//mem_init()告诉我们一共有多少内存，并且标示出其中的空闲可用内存区域等信息。

//start_low_mem为地址1M之前的可用内存的起始地址(即内核代码段之前的内存);start_mem位内核代码段之后的可用内存的起始地址；
//end_mem为内核代码段之后的可用内存的最大地址。

void mem_init(unsigned long start_low_mem,
          unsigned long start_mem, unsigned long end_mem)
{
    int codepages = 0;
    int reservedpages = 0;
    int datapages = 0;
    unsigned long tmp;
    unsigned short * p;
    extern int etext;

    cli();
    end_mem &= PAGE_MASK;
    high_memory = end_mem;
    start_mem +=  0x0000000f;
    start_mem &= ~0x0000000f;
    tmp = MAP_NR(end_mem); //tmp为根据最大内存地址算出来的内存页数
    mem_map = (unsigned short *) start_mem;//mem_map数组的开始地址为当前可用内存的开始地址，即这里分配当前可用内存的开头一部分作为mem_map数组
    p = mem_map + tmp; //start_mem开始的tmp*4个字节分配为mem_map数组
    start_mem = (unsigned long) p;//start_mem的新起始地址就是start_mem加上tmp*4
    while (p > mem_map)  //将数组mem_map的tmp个元素一一赋值为MAP_PAGE_RESERVED
        *--p = MAP_PAGE_RESERVED;
    start_low_mem = PAGE_ALIGN(start_low_mem);
    start_mem = PAGE_ALIGN(start_mem);
    while (start_low_mem < 0xA0000) { //在0x1000~0xA0000内存是可用的
        mem_map[MAP_NR(start_low_mem)] = 0; //0x1000/PAGE_SIZE刚好得到0x1000物理地址所对应的mem_map里的index为1，即刚好是4~8M内存内存。mem_map[0]即是对应0~4k内存。
        start_low_mem += PAGE_SIZE; //映射一页后低的可用内存减少PAGE_SIZE大小
    }
    while (start_mem < end_mem) {  //当前可用内存的开始地址为内核代码后除去二级页表和mem_map数组等占用的内存后的起始地址
        mem_map[MAP_NR(start_mem)] = 0;//start_mem/PAGE_SIZE得到该地址所对应的第几个物理页，及时mem_map中的index
        start_mem += PAGE_SIZE;//映射一页后可用内存减少PAGE_SIZE大小
    }
#ifdef CONFIG_SOUND
    sound_mem_init();
#endif
    free_page_list = 0;
    nr_free_pages = 0;
    for (tmp = 0 ; tmp < end_mem ; tmp += PAGE_SIZE) {
        if (mem_map[MAP_NR(tmp)]) {//0~16M内存中，除去0x1000~0xA0000和start_mem ~ end_mem之间的物理页可用外，其它都为MAP_PAGE_RESERVED，不可用
            if (tmp >= 0xA0000 && tmp < 0x100000) //0xA0000~0x100000为显示内存区和BIOS中断处理程序区，是保留的区域
                reservedpages++;
            else if (tmp < (unsigned long) &etext) //etext应该是内核代码的end地址，所以0x100000~etext为内核代码页
                codepages++;
            else    //否则的话都算作是数据代码页，像0x0~0x1000为中断向量表和BIOS数据区,内核代码段后面的二级页表数组和mem_map数组等内存都是数据页了。
                datapages++;
            continue;
        }
        *(unsigned long *) tmp = free_page_list;
        free_page_list = tmp;
        nr_free_pages++;
    }
    tmp = nr_free_pages << PAGE_SHIFT;
    printk("Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data)\n",
        tmp >> 10,
        end_mem >> 10,
        codepages << (PAGE_SHIFT-10),
        reservedpages << (PAGE_SHIFT-10),
        datapages << (PAGE_SHIFT-10));
/* test if the WP bit is honoured in supervisor mode */
    wp_works_ok = -1;
    pg0[0] = PAGE_READONLY;
    invalidate();
    __asm__ __volatile__("movb 0,%%al ; movb %%al,0": : :"ax", "memory");
    pg0[0] = 0;
    invalidate();
    if (wp_works_ok < 0)
        wp_works_ok = 0;
    return;
}

void si_meminfo(struct sysinfo *val)
{
    int i;

    i = high_memory >> PAGE_SHIFT;
    val->totalram = 0;
    val->freeram = 0;
    val->sharedram = 0;
    val->bufferram = buffermem;
    while (i-- > 0)  {
        if (mem_map[i] & MAP_PAGE_RESERVED)
            continue;
        val->totalram++;
        if (!mem_map[i]) {
            val->freeram++;
            continue;
        }
        val->sharedram += mem_map[i]-1;
    }
    val->totalram <<= PAGE_SHIFT;
    val->freeram <<= PAGE_SHIFT;
    val->sharedram <<= PAGE_SHIFT;
    return;
}


/* This handles a generic mmap of a disk file */
void file_mmap_nopage(int error_code, struct vm_area_struct * area, unsigned long address)
{
    struct inode * inode = area->vm_inode;
    unsigned int block;
    unsigned long page;
    int nr[8];
    int i, j;
    int prot = area->vm_page_prot;

    address &= PAGE_MASK;
    block = address - area->vm_start + area->vm_offset;
    block >>= inode->i_sb->s_blocksize_bits;

    page = get_free_page(GFP_KERNEL);
    if (share_page(area, area->vm_task, inode, address, error_code, page)) {
        ++area->vm_task->min_flt;
        return;
    }

    ++area->vm_task->maj_flt;
    if (!page) {
        oom(current);
        put_page(area->vm_task, BAD_PAGE, address, PAGE_PRIVATE);
        return;
    }
    for (i=0, j=0; i< PAGE_SIZE ; j++, block++, i += inode->i_sb->s_blocksize)
        nr[j] = bmap(inode,block);
    if (error_code & PAGE_RW)
        prot |= PAGE_RW | PAGE_DIRTY;
    page = bread_page(page, inode->i_dev, nr, inode->i_sb->s_blocksize, prot);

    if (!(prot & PAGE_RW)) {
        if (share_page(area, area->vm_task, inode, address, error_code, page))
            return;
    }
    if (put_page(area->vm_task,page,address,prot))
        return;
    free_page(page);
    oom(current);
}

void file_mmap_free(struct vm_area_struct * area)
{
    if (area->vm_inode)
        iput(area->vm_inode);
#if 0
    if (area->vm_inode)
        printk("Free inode %x:%d (%d)\n",area->vm_inode->i_dev,
                 area->vm_inode->i_ino, area->vm_inode->i_count);
#endif
}

/*
 * Compare the contents of the mmap entries, and decide if we are allowed to
 * share the pages
 */
int file_mmap_share(struct vm_area_struct * area1,
            struct vm_area_struct * area2,
            unsigned long address)
{
    if (area1->vm_inode != area2->vm_inode)
        return 0;
    if (area1->vm_start != area2->vm_start)
        return 0;
    if (area1->vm_end != area2->vm_end)
        return 0;
    if (area1->vm_offset != area2->vm_offset)
        return 0;
    if (area1->vm_page_prot != area2->vm_page_prot)
        return 0;
    return 1;
}

struct vm_operations_struct file_mmap = {
    NULL,            /* open */
    file_mmap_free,        /* close */
    file_mmap_nopage,    /* nopage */
    NULL,            /* wppage */
    file_mmap_share,    /* share */
    NULL,            /* unmap */
};