Linux-3.14.12内存管理笔记【建立内核页表（3）】

前面已经分析了内核页表的准备工作以及内核低端内存页表的建立，接着回到init_mem_mapping()中，低端内存页表建立后紧随着还有一个函数early_ioremap_page_table_range_init()：

【file：/arch/x86/mm/init.c】
/*
 * Build a proper pagetable for the kernel mappings. Up until this
 * point, we've been running on some set of pagetables constructed by
 * the boot process.
 *
 * If we're booting on native hardware, this will be a pagetable
 * constructed in arch/x86/kernel/head_32.S. The root of the
 * pagetable will be swapper_pg_dir.
 *
 * If we're booting paravirtualized under a hypervisor, then there are
 * more options: we may already be running PAE, and the pagetable may
 * or may not be based in swapper_pg_dir. In any case,
 * paravirt_pagetable_init() will set up swapper_pg_dir
 * appropriately for the rest of the initialization to work.
 *
 * In general, pagetable_init() assumes that the pagetable may already
 * be partially populated, and so it avoids stomping on any existing
 * mappings.
 */
void __init early_ioremap_page_table_range_init(void)
{
    pgd_t *pgd_base = swapper_pg_dir;
    unsigned long vaddr, end;

    /*
     * Fixed mappings, only the page table structure has to be
     * created - mappings will be set by set_fixmap():
     */
    vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
    end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
    page_table_range_init(vaddr, end, pgd_base);
    early_ioremap_reset();
}

该函数主要是用于建立固定内存映射区的。固定内存映射区是指FIXADDR_START到FIXADDR_TOP的地址空间，而该地址空间因功能特性不同通过索引来定义区分，其中索引以枚举类型的形式定义在enum fixed_addresses里面。

【file：/arch/x86/include/asm/fixmap.h】
/*
 * Here we define all the compile-time 'special' virtual
 * addresses. The point is to have a constant address at
 * compile time, but to set the physical address only
 * in the boot process.
 * for x86_32: We allocate these special addresses
 * from the end of virtual memory (0xfffff000) backwards.
 * Also this lets us do fail-safe vmalloc(), we
 * can guarantee that these special addresses and
 * vmalloc()-ed addresses never overlap.
 *
 * These 'compile-time allocated' memory buffers are
 * fixed-size 4k pages (or larger if used with an increment
 * higher than 1). Use set_fixmap(idx,phys) to associate
 * physical memory with fixmap indices.
 *
 * TLB entries of such buffers will not be flushed across
 * task switches.
 */
enum fixed_addresses {
#ifdef CONFIG_X86_32
    FIX_HOLE,
    FIX_VDSO,
#else
    VSYSCALL_LAST_PAGE,
    VSYSCALL_FIRST_PAGE = VSYSCALL_LAST_PAGE
                + ((VSYSCALL_END-VSYSCALL_START) >> PAGE_SHIFT) - 1,
    VVAR_PAGE,
    VSYSCALL_HPET,
#ifdef CONFIG_PARAVIRT_CLOCK
    PVCLOCK_FIXMAP_BEGIN,
    PVCLOCK_FIXMAP_END = PVCLOCK_FIXMAP_BEGIN+PVCLOCK_VSYSCALL_NR_PAGES-1,
#endif
#endif
    FIX_DBGP_BASE,
    FIX_EARLYCON_MEM_BASE,
#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
    FIX_OHCI1394_BASE,
#endif
#ifdef CONFIG_X86_LOCAL_APIC
    FIX_APIC_BASE, /* local (CPU) APIC) -- required for SMP or not */
#endif
#ifdef CONFIG_X86_IO_APIC
    FIX_IO_APIC_BASE_0,
    FIX_IO_APIC_BASE_END = FIX_IO_APIC_BASE_0 + MAX_IO_APICS - 1,
#endif
#ifdef CONFIG_X86_VISWS_APIC
    FIX_CO_CPU, /* Cobalt timer */
    FIX_CO_APIC, /* Cobalt APIC Redirection Table */
    FIX_LI_PCIA, /* Lithium PCI Bridge A */
    FIX_LI_PCIB, /* Lithium PCI Bridge B */
#endif
    FIX_RO_IDT, /* Virtual mapping for read-only IDT */
#ifdef CONFIG_X86_32
    FIX_KMAP_BEGIN, /* reserved pte's for temporary kernel mappings */
    FIX_KMAP_END = FIX_KMAP_BEGIN+(KM_TYPE_NR*NR_CPUS)-1,
#ifdef CONFIG_PCI_MMCONFIG
    FIX_PCIE_MCFG,
#endif
#endif
#ifdef CONFIG_PARAVIRT
    FIX_PARAVIRT_BOOTMAP,
#endif
    FIX_TEXT_POKE1, /* reserve 2 pages for text_poke() */
    FIX_TEXT_POKE0, /* first page is last, because allocation is backward */
#ifdef  CONFIG_X86_INTEL_MID
    FIX_LNW_VRTC,
#endif
    __end_of_permanent_fixed_addresses,

    /*
     * 256 temporary boot-time mappings, used by early_ioremap(),
     * before ioremap() is functional.
     *
     * If necessary we round it up to the next 256 pages boundary so
     * that we can have a single pgd entry and a single pte table:
     */
#define NR_FIX_BTMAPS       64
#define FIX_BTMAPS_SLOTS    4
#define TOTAL_FIX_BTMAPS (NR_FIX_BTMAPS * FIX_BTMAPS_SLOTS)
    FIX_BTMAP_END =
     (__end_of_permanent_fixed_addresses ^
      (__end_of_permanent_fixed_addresses + TOTAL_FIX_BTMAPS - 1)) &
     -PTRS_PER_PTE
     ? __end_of_permanent_fixed_addresses + TOTAL_FIX_BTMAPS -
       (__end_of_permanent_fixed_addresses & (TOTAL_FIX_BTMAPS - 1))
     : __end_of_permanent_fixed_addresses,
    FIX_BTMAP_BEGIN = FIX_BTMAP_END + TOTAL_FIX_BTMAPS - 1,
#ifdef CONFIG_X86_32
    FIX_WP_TEST,
#endif
#ifdef CONFIG_INTEL_TXT
    FIX_TBOOT_BASE,
#endif
    __end_of_fixed_addresses
};

但是各枚举标识的分区并不是从低地址往高地址分布，而是自高地址往低地址分布。其中__fix_to_virt宏定义就是用来通过索引来计算相应的固定映射区域的线性地址。

#define __fix_to_virt(x)         (FIXADDR_TOP - ((x) << PAGE_SHIFT))

对应的有虚拟地址转索引的宏：

#define __virt_to_fix(x)         ((FIXADDR_TOP - ((x)&PAGE_MASK)) >> PAGE_SHIFT)

接着回到early_ioremap_page_table_range_init()的第一个函数page_table_range_init()：

【file：/arch/x86/mm/init_32.c】
/*
 * This function initializes a certain range of kernel virtual memory
 * with new bootmem page tables, everywhere page tables are missing in
 * the given range.
 *
 * NOTE: The pagetables are allocated contiguous on the physical space
 * so we can cache the place of the first one and move around without
 * checking the pgd every time.
 */
static void __init
page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
{
    int pgd_idx, pmd_idx;
    unsigned long vaddr;
    pgd_t *pgd;
    pmd_t *pmd;
    pte_t *pte = NULL;
    unsigned long count = page_table_range_init_count(start, end);
    void *adr = NULL;

    if (count)
        adr = alloc_low_pages(count);

    vaddr = start;
    pgd_idx = pgd_index(vaddr);
    pmd_idx = pmd_index(vaddr);
    pgd = pgd_base + pgd_idx;

    for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
        pmd = one_md_table_init(pgd);
        pmd = pmd + pmd_index(vaddr);
        for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
                            pmd++, pmd_idx++) {
            pte = page_table_kmap_check(one_page_table_init(pmd),
                            pmd, vaddr, pte, &adr);

            vaddr += PMD_SIZE;
        }
        pmd_idx = 0;
    }
}

该函数里面其中调用的page_table_range_init_count()：

【file：/arch/x86/mm/init_32.c】
static unsigned long __init
page_table_range_init_count(unsigned long start, unsigned long end)
{
    unsigned long count = 0;
#ifdef CONFIG_HIGHMEM
    int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
    int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
    int pgd_idx, pmd_idx;
    unsigned long vaddr;

    if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
        return 0;

    vaddr = start;
    pgd_idx = pgd_index(vaddr);

    for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
        for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
                            pmd_idx++) {
            if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
                (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
                count++;
            vaddr += PMD_SIZE;
        }
        pmd_idx = 0;
    }
#endif
    return count;
}

page_table_range_init_count()用来计算指临时内核映射区间的页表数量。前面提到FIXADDR_START到FIXADDR_TOP是固定映射区，其间有多个索引标识不同功能的映射区间，其中的一个区间FIX_KMAP_BEGIN到FIX_KMAP_END是临时内核映射区。顺便可以看一下两者的定义：

    FIX_KMAP_BEGIN, /* reserved pte's for temporary kernel mappings */

    FIX_KMAP_END = FIX_KMAP_BEGIN+(KM_TYPE_NR*NR_CPUS)-1,

其中KM_TYPE_NR表示“窗口”数量，在高端内存的任意一个页框都可以通过一个“窗口”映射到内核地址空间，调用kmap_atomic可以搭建起“窗口”到高端内存的关系，即建立临时内核映射。而NR_CPUS则表示CPU数量。总的来说就是该临时内核映射区间是为了给各个CPU准备一个指定的窗口空间。由于kmap_atomic()对该区间的使用，所以该区间必须保证其页表连续性。

如果页全局目录数不为0的时候，紧接着page_table_range_init_count()的是alloc_low_pages()：

【file：/arch/x86/mm/init.c】
/*
 * Pages returned are already directly mapped.
 *
 * Changing that is likely to break Xen, see commit:
 *
 * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve
 *
 * for detailed information.
 */
__ref void *alloc_low_pages(unsigned int num)
{
    unsigned long pfn;
    int i;

    if (after_bootmem) {
        unsigned int order;

        order = get_order((unsigned long)num << PAGE_SHIFT);
        return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK |
                        __GFP_ZERO, order);
    }

    if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) {
        unsigned long ret;
        if (min_pfn_mapped >= max_pfn_mapped)
            panic("alloc_low_pages: ran out of memory");
        ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
                    max_pfn_mapped << PAGE_SHIFT,
                    PAGE_SIZE * num , PAGE_SIZE);
        if (!ret)
            panic("alloc_low_pages: can not alloc memory");
        memblock_reserve(ret, PAGE_SIZE * num);
        pfn = ret >> PAGE_SHIFT;
    } else {
        pfn = pgt_buf_end;
        pgt_buf_end += num;
        printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n",
            pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1);
    }

    for (i = 0; i < num; i++) {
        void *adr;

        adr = __va((pfn + i) << PAGE_SHIFT);
        clear_page(adr);
    }

    return __va(pfn << PAGE_SHIFT);
}

则是根据前面early_alloc_pgt_buf()申请保留的页表缓冲空间使用情况来判断，是从页表缓冲空间中申请还是通过memblock算法申请页表内存。

回到page_table_range_init()，其中one_md_table_init()是用于当pgd入参为空时，申请新物理页作为页中间目录的，但是此次仅分析x86非PAE环境的情况，不存在页中间目录，故实际上返回的仍是入参。附代码：

【file：/arch/x86/mm/init_32.c】
/*
 * Creates a middle page table and puts a pointer to it in the
 * given global directory entry. This only returns the gd entry
 * in non-PAE compilation mode, since the middle layer is folded.
 */
static pmd_t * __init one_md_table_init(pgd_t *pgd)
{
    pud_t *pud;
    pmd_t *pmd_table;

#ifdef CONFIG_X86_PAE
    if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
        pmd_table = (pmd_t *)alloc_low_page();
        paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
        set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
        pud = pud_offset(pgd, 0);
        BUG_ON(pmd_table != pmd_offset(pud, 0));

        return pmd_table;
    }
#endif
    pud = pud_offset(pgd, 0);
    pmd_table = pmd_offset(pud, 0);

    return pmd_table;
}

接着的是page_table_kmap_check()，其入参调用的one_page_table_init()是用于当入参pmd没有页表指向时，创建页表并使其指向被创建的页表。page_table_kmap_check()实现：

【file：/arch/x86/mm/init_32.c】
static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
                       unsigned long vaddr, pte_t *lastpte,
                       void **adr)
{
#ifdef CONFIG_HIGHMEM
    /*
     * Something (early fixmap) may already have put a pte
     * page here, which causes the page table allocation
     * to become nonlinear. Attempt to fix it, and if it
     * is still nonlinear then we have to bug.
     */
    int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
    int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;

    if (pmd_idx_kmap_begin != pmd_idx_kmap_end
        && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
        && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
        pte_t *newpte;
        int i;

        BUG_ON(after_bootmem);
        newpte = *adr;
        for (i = 0; i < PTRS_PER_PTE; i++)
            set_pte(newpte + i, pte[i]);
        *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);

        paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
        set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
        BUG_ON(newpte != pte_offset_kernel(pmd, 0));
        __flush_tlb_all();

        paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
        pte = newpte;
    }
    BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
           && vaddr > fix_to_virt(FIX_KMAP_END)
           && lastpte && lastpte + PTRS_PER_PTE != pte);
#endif
    return pte;
}

可以看到这里在此出现临时内核映射区间的标识（FIX_KMAP_END和FIX_KMAP_BEGIN），检查当前页表初始化的地址是否处于该区间范围，如果是，则把其pte页表的内容拷贝到page_table_range_init()申请的页表空间中，并将newpte新页表的地址设置到pmd中（32bit系统实际上就是页全局目录），然后调用__flush_tlb_all()刷新TLB缓存；如果不是该区间，则仅是由入参中调用的one_page_table_init()被分配到了页表空间。

由此，可以知道page_table_range_init()主要是做了什么了。这是由于kmap_atomic()对该区间的使用，该区间必须保证其页表连续性。为了避免前期可能对固定映射区已经分配了页表项，基于临时内核映射区间要求页表连续性的保证，所以在此重新申请连续的页表空间将原页表内容拷贝至此。值得注意的是，与低端内存的页表初始化不同的是，这里的页表只是被分配，相应的PTE项并未初始化，这个工作将会交由以后各个固定映射区部分的相关代码调用set_fixmap()来将相关的固定映射区页表与物理内存关联。

early_ioremap_page_table_range_init()函数再往下的early_ioremap_reset()仅是对after_paging_init全局变量赋值。

最后退出early_ioremap_page_table_range_init()后，init_mem_mapping()调用load_cr3()刷新CR3寄存器，__flush_tlb_all()则用于刷新TLB，由此启用新的内存分页映射。

     至此，内核页表建立完毕。