高端内存(续)--临时内存映射

一.为什么引入临时内存映射(temporary kernel mappings)

在永久内存映射中我们看到，如果 pkmap_page_table页表里面没有空的entry，那么就会导致这次映射被阻塞，所以我们说不能在一些原子的上下文情况下调用kmap()函数。而在临时内存映射中，不会去判断该pte是否已经被用掉了，它采用的是覆盖的策略，所以把总是能成功的建立映射。 会不会被阻塞就是临时内存映射和永久内存映射一个最明显的区别 。

二. 临时内存映射

内核地址空间布局

FIXADDR_TOP = 0xfffff000:因为最后一个页面（0xfffff000--0xffffffff）被内核所保留。

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,         VSYSCALL_HPET, #endif         FIX_DBGP_BASE,         FIX_EARLYCON_MEM_BASE, #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 #ifdef CONFIG_X86_F00F_BUG         FIX_F00F_IDT, /* Virtual mapping for IDT */ #endif #ifdef CONFIG_X86_CYCLONE_TIMER         FIX_CYCLONE_TIMER, /*cyclone timer register*/ #endif #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 */         __end_of_permanent_fixed_addresses,         /*          * 256 temporary boot-time mappings, used by early_ioremap(),          * before ioremap() is functional.          *          * 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         FIX_BTMAP_END = __end_of_permanent_fixed_addresses + 256 -                         (__end_of_permanent_fixed_addresses & 255),         FIX_BTMAP_BEGIN = FIX_BTMAP_END + NR_FIX_BTMAPS*FIX_BTMAPS_SLOTS - 1, #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT         FIX_OHCI1394_BASE, #endif #ifdef CONFIG_X86_32         FIX_WP_TEST, #endif #ifdef CONFIG_INTEL_TXT         FIX_TBOOT_BASE, #endif         __end_of_fixed_addresses };

#define FIXADDR_SIZE (__end_of_permanent_fixed_addresses << PAGE_SHIFT) #define FIXADDR_START (FIXADDR_TOP - FIXADDR_SIZE)

从FIXADDR_START到FIXADDR_TOP这段地址空间被称为固定映射线性空间，这段地址空间主要用来映射一些设备的内存和寄存器，像FIX_APIC_BASE是用来映射local apic的寄存器的。其中的FIX_KMAP_BEGIN到FIX_KMAP_END是用来供临时内存映射的使用。

这段空间的大小是KM_TYPE_NR*NR_CPUS，也就是说，每个CPU都有独立的KM_TYPE_NR大小的空间，来看看看KM_TYPE_NR是什么

#ifdef __WITH_KM_FENCE # define KMAP_D(n) __KM_FENCE_##n , #else # define KMAP_D(n) #endif enum km_type { KMAP_D(0) KM_BOUNCE_READ, KMAP_D(1) KM_SKB_SUNRPC_DATA, KMAP_D(2) KM_SKB_DATA_SOFTIRQ, KMAP_D(3) KM_USER0, KMAP_D(4) KM_USER1, KMAP_D(5) KM_BIO_SRC_IRQ, KMAP_D(6) KM_BIO_DST_IRQ, KMAP_D(7) KM_PTE0, KMAP_D(8) KM_PTE1, KMAP_D(9) KM_IRQ0, KMAP_D(10) KM_IRQ1, KMAP_D(11) KM_SOFTIRQ0, KMAP_D(12) KM_SOFTIRQ1, KMAP_D(13) KM_SYNC_ICACHE, KMAP_D(14) KM_SYNC_DCACHE, /* UML specific, for copy_*_user - used in do_op_one_page */ KMAP_D(15) KM_UML_USERCOPY, KMAP_D(16) KM_IRQ_PTE, KMAP_D(17) KM_NMI, KMAP_D(18) KM_NMI_PTE, KMAP_D(19) KM_TYPE_NR };

每个枚举项都代表了一个pte，每个CPU都有20个这样的pte供临时内核映射使用。我们先来看看内核如何来获得这些对应的枚举项所对应的pte地址：

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

这个宏返回的是固定线性映射里面枚举项所对应的线性地址，FIXADDR_TOP=0xfffff000,以FIX_APIC_BASE为例，__fix_to_virt(FIX_APIC_BASE),FIX_APIC_BASE等于4，那么他的线性地址就是0xfffff000-(4<<PAGE_SHIFT),这里我们看到，线性地址和枚举项的位置关系是成反比的，越靠前的枚举项对应的线性地址越靠后。

下面的代码是完整的如何通过枚举项，获得对应的pte项所对应的线性地址。

static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) {         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), vaddr), vaddr); } static void __init kmap_init(void) {         unsigned long kmap_vstart;         /*          * Cache the first kmap pte:          */         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);         kmap_prot = PAGE_KERNEL; }

注意：kmap_pte是临时内存映射起始的页表项，接下来的其他页表项根据他们的相对位置差就可以求得。比如idx对应的pte=kmap_pte - idx，为什么是减不是加，看上面的说明（越靠前的枚举项对应的线性地址越靠后）。

接下来我们看实际的建立映射的函数（kmap_atomic）

/*  * kmap_atomic/kunmap_atomic is significantly faster than kmap/kunmap because  * no global lock is needed and because the kmap code must perform a global TLB  * invalidation when the kmap pool wraps.  *  * However when holding an atomic kmap it is not legal to sleep, so atomic  * kmaps are appropriate for short, tight code paths only.  */ void *kmap_atomic_prot(struct page *page, enum km_type type, pgprot_t prot) {         enum fixed_addresses idx;         unsigned long vaddr;         /* even !CONFIG_PREEMPT needs this, for in_atomic in do_page_fault */         pagefault_disable();         //判断要建立映射的页在高端内存中，否则其线性地址已经存在         if (!PageHighMem(page))                 return page_address(page);         debug_kmap_atomic(type);         //求得对应的idx         idx = type + KM_TYPE_NR*smp_processor_id();         //内核可以访问该页的线性地址         vaddr = __fix_to_virt(FIX_KMAP_BEGIN + idx);         BUG_ON(!pte_none(*(kmap_pte-idx)));         //根据kmap_pte，建立对应的页表项映射，之后就可以通过vaddr访问该页了         set_pte(kmap_pte-idx, mk_pte(page, prot));         return (void *)vaddr; } void *kmap_atomic(struct page *page, enum km_type type) {         return kmap_atomic_prot(page, type, kmap_prot); }

解除映射也很简单,没有做什么特殊处理：

void kunmap_atomic(void *kvaddr, enum km_type type) {         unsigned long vaddr = (unsigned long) kvaddr & PAGE_MASK;         enum fixed_addresses idx = type + KM_TYPE_NR*smp_processor_id();         /*          * Force other mappings to Oops if they'll try to access this pte          * without first remap it. Keeping stale mappings around is a bad idea          * also, in case the page changes cacheability attributes or becomes          * a protected page in a hypervisor.          */         if (vaddr == __fix_to_virt(FIX_KMAP_BEGIN+idx))                 kpte_clear_flush(kmap_pte-idx, vaddr);         else { #ifdef CONFIG_DEBUG_HIGHMEM                 BUG_ON(vaddr < PAGE_OFFSET);                 BUG_ON(vaddr >= (unsigned long)high_memory); #endif         }         pagefault_enable(); }

三. 总结

其实，不管是那样的方式，原理都是一样的，都是在固定映射区外选定一个地址，然后再修改PTE项，使其指向相应的page。特别值得我们注意的是，因为kmap()会引起睡眠，所以它不能用于中断处理。但每一种映射方式都有自己的优点和缺点，这需要我们在写代码的时候仔细考虑了。

转自： http://blog.chinaunix.net/space.php?uid=21752164&do=blog&id=2689673