[系统运维]linux-4.19 内存

目录

加载解析fdt 存储信息

kaslr和页表映射

虚拟地址到物理地址转换

?参考

加载解析fdt 存储信息

start_kernel（）-》setup_arch（）-》setup_machine_fdt（）-》 early_init_dt_scan_memory（）

通过fdt 获取存储信息，base_address和size，再通过early_init_dt_add_memory_arch(base, size);?加入memblock 子系统。

kaslr和页表映射

kernel-4.19/arch/arm64/mm/init.c

659#define MLK(b, t) b, t, ((t) - (b)) >> 10
660#define MLM(b, t) b, t, ((t) - (b)) >> 20
661#define MLG(b, t) b, t, ((t) - (b)) >> 30
662#define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
663
664?? ?pr_notice("Virtual kernel memory layout:\n");
665#ifdef CONFIG_KASAN
666?? ?pr_notice(" ? ?kasan ? : 0x%16lx - 0x%16lx ? (%6ld GB)\n",
667?? ??? ?MLG(KASAN_SHADOW_START, KASAN_SHADOW_END));
668#endif
669?? ?pr_notice(" ? ?modules : 0x%16lx - 0x%16lx ? (%6ld MB)\n",
670?? ??? ?MLM(MODULES_VADDR, MODULES_END));
671?? ?pr_notice(" ? ?vmalloc : 0x%16lx - 0x%16lx ? (%6ld GB)\n",
672?? ??? ?MLG(VMALLOC_START, VMALLOC_END));
673?? ?pr_notice(" ? ? ?.text : 0x%p" " - 0x%p" " ? (%6ld KB)\n",
674?? ??? ?MLK_ROUNDUP(_text, _etext));
675?? ?pr_notice(" ? ?.rodata : 0x%p" " - 0x%p" " ? (%6ld KB)\n",
676?? ??? ?MLK_ROUNDUP(__start_rodata, __init_begin));
677?? ?pr_notice(" ? ? ?.init : 0x%p" " - 0x%p" " ? (%6ld KB)\n",
678?? ??? ?MLK_ROUNDUP(__init_begin, __init_end));
679?? ?pr_notice(" ? ? ?.data : 0x%p" " - 0x%p" " ? (%6ld KB)\n",
680?? ??? ?MLK_ROUNDUP(_sdata, _edata));
681?? ?pr_notice(" ? ? ? .bss : 0x%p" " - 0x%p" " ? (%6ld KB)\n",
682?? ??? ?MLK_ROUNDUP(__bss_start, __bss_stop));
683?? ?pr_notice(" ? ?fixed ? : 0x%16lx - 0x%16lx ? (%6ld KB)\n",
684?? ??? ?MLK(FIXADDR_START, FIXADDR_TOP));
685?? ?pr_notice(" ? ?PCI I/O : 0x%16lx - 0x%16lx ? (%6ld MB)\n",
686?? ??? ?MLM(PCI_IO_START, PCI_IO_END));
687#ifdef CONFIG_SPARSEMEM_VMEMMAP
688?? ?pr_notice(" ? ?vmemmap : 0x%16lx - 0x%16lx ? (%6ld GB maximum)\n",
689?? ??? ?MLG(VMEMMAP_START, VMEMMAP_START + VMEMMAP_SIZE));
690?? ?pr_notice(" ? ? ? ? ? ? ?0x%16lx - 0x%16lx ? (%6ld MB actual)\n",
691?? ??? ?MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
692?? ??? ? ? ?(unsigned long)virt_to_page(high_memory)));
693#endif
694?? ?pr_notice(" ? ?memory ?: 0x%16lx - 0x%16lx ? (%6ld MB)\n",
695?? ??? ?MLM(__phys_to_virt(memblock_start_of_DRAM()),
696?? ??? ? ? ?(unsigned long)high_memory));

kernel-4.19/arch/arm64/include/asm/pgtable.h
/*
27 * VMALLOC range.
28 *
29 * VMALLOC_START: beginning of the kernel vmalloc space
30 * VMALLOC_END: extends to the available space below vmmemmap, PCI I/O space
31 *?? ?and fixed mappings
32 */
33#define VMALLOC_START?? ??? ?(MODULES_END)
34#define VMALLOC_END?? ??? ?(PAGE_OFFSET - PUD_SIZE - VMEMMAP_SIZE - SZ_64K)
35
36#define vmemmap?? ??? ??? ?((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
37
38#define FIRST_USER_ADDRESS?? ?0UL

这里VMALLOC_START 跟 KIMAGE_VADDR 一样

kernel-4.19/arch/arm64/kernel/head.S
?? ?/*
377?? ? * Map the kernel image (starting with PHYS_OFFSET).
378?? ? */
379?? ?adrp?? ?x0, swapper_pg_dir
380?? ?mov_q?? ?x5, KIMAGE_VADDR + TEXT_OFFSET?? ?// compile time __va(_text)
381?? ?add?? ?x5, x5, x23?? ??? ??? ?// add KASLR displacement
382?? ?mov?? ?x4, PTRS_PER_PGD
383?? ?adrp?? ?x6, _end?? ??? ??? ?// runtime __pa(_end)
384?? ?adrp?? ?x3, _text?? ??? ??? ?// runtime __pa(_text)
385?? ?sub?? ?x6, x6, x3?? ??? ??? ?// _end - _text
386?? ?add?? ?x6, x6, x5?? ??? ??? ?// runtime __va(_end)

map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14

map_memory 是将虚拟地址x5 ，长度x6 映射到物理地址x3 开始的位置；其中x0 页表地址，x1 第一个页表项，通常x1=x0+PAGE_SIZE;x4 表示对应页表等级由多少项。这里是将内核代码段.text 从虚拟地址KIMAGE_VADDR + TEXT_OFFSET+KASLR 偏移? 对应虚拟地址映射到物理地址_text 对应连续物理地址。

vmlinux 对应的是编译链接地址；coredump 及内核堆栈对应的是运行时地址；

支持kaslr之前，kernel加载到system RAM的某个位置，它之前的内存kernel是无法管理的，所以一般将kernel加载到system RAM的 起始位置+TEXT_OFFSET（0x080000）处，因为kaslr修改成可以随意加载到system RAM的任何位置，只要满足对齐要求就可以；
支持kaslr之前，kernel image是映射到线性映射区域的(4.15 之前)，因为kaslr才修改成映射到vmalloc区域；
为了支持kaslr，内核要编译成PIE(Position Independent Execuable)，才能重映射

这样.text? 其实位置跟 VMALLOC区其实地址有一个偏移

add_link?= addr_run - (VAMLLOC_START - .text_start) +?TEXT_OFFSET? ?？？？

add_link? 是addr2line 使用，addr_run 是虚拟地址，运行时堆栈地址。 .text_start 是load物理地址？？？

MTK 平台：

static inline void show_kaslr(void)
{
	u64 const kaslr_offset = aee_get_kimage_vaddr() - KIMAGE_VADDR;

	pr_notice("Kernel Offset: 0x%llx from 0x%lx\n",
			kaslr_offset, KIMAGE_VADDR);
	pr_notice("PHYS_OFFSET: 0x%llx\n", PHYS_OFFSET);
	aee_rr_rec_kaslr_offset(kaslr_offset);
}

aee_get_kimage_vaddr? 从coredump 里面读取kimage_vaddr 对应地址

9#if defined(KIMAGE_VADDR)
90?? ?machdesc_p->kimage_vaddr = KIMAGE_VADDR;
91#endif
92#if defined(TEXT_OFFSET)
93?? ?machdesc_p->kimage_vaddr += TEXT_OFFSET;
94#endif

kernel-4.19/arch/arm64/include/asm/memory.h

35#define PCI_IO_SIZE?? ??? ?SZ_16M
36
37/*
38 * Log2 of the upper bound of the size of a struct page. Used for sizing
39 * the vmemmap region only, does not affect actual memory footprint.
40 * We don't use sizeof(struct page) directly since taking its size here
41 * requires its definition to be available at this point in the inclusion
42 * chain, and it may not be a power of 2 in the first place.
43 */
44#define STRUCT_PAGE_MAX_SHIFT?? ?6
45
46/*
47 * VMEMMAP_SIZE - allows the whole linear region to be covered by
48 * ? ? ? ? ? ? ? ?a struct page array
49 */
50#define VMEMMAP_SIZE (UL(1) << (VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT))
51
52/*
53 * PAGE_OFFSET - the virtual address of the start of the linear map (top
54 *?? ??? ? (VA_BITS - 1))
55 * KIMAGE_VADDR - the virtual address of the start of the kernel image
56 * VA_BITS - the maximum number of bits for virtual addresses.
57 * VA_START - the first kernel virtual address.
58 */
59#define VA_BITS?? ??? ??? ?(CONFIG_ARM64_VA_BITS)
60#define VA_START?? ??? ?(UL(0xffffffffffffffff) - \
61?? ?(UL(1) << VA_BITS) + 1)
62#define PAGE_OFFSET?? ??? ?(UL(0xffffffffffffffff) - \
63?? ?(UL(1) << (VA_BITS - 1)) + 1)
64#define KIMAGE_VADDR?? ??? ?(MODULES_END)
65#define MODULES_END?? ??? ?(MODULES_VADDR + MODULES_VSIZE)
66#define MODULES_VADDR?? ??? ?(VA_START + KASAN_SHADOW_SIZE)
67#define MODULES_VSIZE?? ??? ?(SZ_128M)
68#define VMEMMAP_START?? ??? ?(PAGE_OFFSET - VMEMMAP_SIZE)
69#define PCI_IO_END?? ??? ?(VMEMMAP_START - SZ_2M)
70#define PCI_IO_START?? ??? ?(PCI_IO_END - PCI_IO_SIZE)
71#define FIXADDR_TOP?? ??? ?(PCI_IO_START - SZ_2M)
72
73#define KERNEL_START ? ? ?_text
74#define KERNEL_END ? ? ? ?_end

内核虚拟地址起点：VA_START = 0xffff_0000_0000_0000

PAGE_OFFSET? =0xffff_1000_0000_0000

?PAGE_OFFSET - the virtual address of the start of the linear map (top(VA_BITS - 1))

KIMAGE_VADDR - the virtual address of the start of the kernel image

这里MODULES_VSIZE = 128M = 0x8000000

?kernel-4.19/arch/arm64/Makefile
90# The byte offset of the kernel image in RAM from the start of RAM.
91ifeq ($(CONFIG_ARM64_RANDOMIZE_TEXT_OFFSET), y)
92TEXT_OFFSET := $(shell awk "BEGIN {srand(); printf \"0x%06x\n\", \
93?? ??? ? int(2 * 1024 * 1024 / (2 ^ $(CONFIG_ARM64_PAGE_SHIFT)) * \
94?? ??? ? rand()) * (2 ^ $(CONFIG_ARM64_PAGE_SHIFT))}")
95else
96TEXT_OFFSET := 0x00080000

这里CONFIG_ARM64_PAGE_SHIFT 页大小 12 位

User space 地址mmu转换示例：
基本概念介绍：

task_struct->mm
如果是用户进程，指向当前的进程地址空间。

如果是内核线程，为空(内核线程没有进程地址空间)。

task_struct->active_mm
如果是用户进程，mm与active_mm相同，都指向进程的地址空间。

如果是内核线程，指向被借用的用户进程的地址空间(mm)。

user space各个process 保存自己独立的pgd，存放在task__struct->mm->pgd里面，每次做context ?switch时，会把next_task的pgd存放到TTBR0_EL0里面，从而实现不同process不同的地址空间。

TTBR0_EL1 对应内核pgd

cr3寄存器的加载

cr3寄存器的加载是在进程调度的时候更新的，具体如下

schedule()->context_switch()->switch_mm()->load_cr3(next->pgd)

load_cr3加载的是mm_struct->pgd，即线性地址，而实际上加裁到cr3寄存器的是实际的物理地址write_cr3(__pa(pgdir));在装载cr3寄存器时将线性地址通过__pa转换成了物理地址了，所以cr3寄存器是装的是实实在在的物理地址。正在使用的页目录的物理地址存在cr3控制寄存器中

ARM64 Kernel Image Mapping的变化

假设页表映射层级是4，即配置CONFIG_ARM64_PGTABLE_LEVELS=4。地址宽度是48，即配置CONFIG_ARM64_VA_BITS=48，页大小4K，每个页表项占 8字节

PGD? [47,39]? ? ? ? ? 512*512G=256T

PUD? [38,30]? ? ? ? ? 512G

PMD [29,21]? ? ? ? ? 512*2M = 1G

PTE? [20,12]? ? ? ? ? 4K/8=512 项，512*4K = 2M

PAGE_SHIFT? ?[11~0]

kernel-4.19/arch/arm64/include/asm/pgtable-hwdef.h

 */
#ifndef __ASM_PGTABLE_HWDEF_H
#define __ASM_PGTABLE_HWDEF_H

#include <asm/memory.h>

/*
 * Number of page-table levels required to address 'va_bits' wide
 * address, without section mapping. We resolve the top (va_bits - PAGE_SHIFT)
 * bits with (PAGE_SHIFT - 3) bits at each page table level. Hence:
 *
 *  levels = DIV_ROUND_UP((va_bits - PAGE_SHIFT), (PAGE_SHIFT - 3))
 *
 * where DIV_ROUND_UP(n, d) => (((n) + (d) - 1) / (d))
 *
 * We cannot include linux/kernel.h which defines DIV_ROUND_UP here
 * due to build issues. So we open code DIV_ROUND_UP here:
 *
 *	((((va_bits) - PAGE_SHIFT) + (PAGE_SHIFT - 3) - 1) / (PAGE_SHIFT - 3))
 *
 * which gets simplified as :
 */
#define ARM64_HW_PGTABLE_LEVELS(va_bits) (((va_bits) - 4) / (PAGE_SHIFT - 3))

/*
 * Size mapped by an entry at level n ( 0 <= n <= 3)
 * We map (PAGE_SHIFT - 3) at all translation levels and PAGE_SHIFT bits
 * in the final page. The maximum number of translation levels supported by
 * the architecture is 4. Hence, starting at at level n, we have further
 * ((4 - n) - 1) levels of translation excluding the offset within the page.
 * So, the total number of bits mapped by an entry at level n is :
 *
 *  ((4 - n) - 1) * (PAGE_SHIFT - 3) + PAGE_SHIFT
 *
 * Rearranging it a bit we get :
 *   (4 - n) * (PAGE_SHIFT - 3) + 3
 */
#define ARM64_HW_PGTABLE_LEVEL_SHIFT(n)	((PAGE_SHIFT - 3) * (4 - (n)) + 3)

#define PTRS_PER_PTE		(1 << (PAGE_SHIFT - 3))
55
56/*
57 * PMD_SHIFT determines the size a level 2 page table entry can map.
58 */
59#if CONFIG_PGTABLE_LEVELS > 2
60#define PMD_SHIFT		ARM64_HW_PGTABLE_LEVEL_SHIFT(2)
61#define PMD_SIZE		(_AC(1, UL) << PMD_SHIFT)
62#define PMD_MASK		(~(PMD_SIZE-1))
#define PTRS_PER_PMD		PTRS_PER_PTE
#endif

/*
 * PUD_SHIFT determines the size a level 1 page table entry can map.
 */
#if CONFIG_PGTABLE_LEVELS > 3
#define PUD_SHIFT		ARM64_HW_PGTABLE_LEVEL_SHIFT(1)
#define PUD_SIZE		(_AC(1, UL) << PUD_SHIFT)
#define PUD_MASK		(~(PUD_SIZE-1))
#define PTRS_PER_PUD		PTRS_PER_PTE
#endif

/*
 * PGDIR_SHIFT determines the size a top-level page table entry can map
 * (depending on the configuration, this level can be 0, 1 or 2).
 */
#define PGDIR_SHIFT		ARM64_HW_PGTABLE_LEVEL_SHIFT(4 - CONFIG_PGTABLE_LEVELS)
#define PGDIR_SIZE		(_AC(1, UL) << PGDIR_SHIFT)
#define PGDIR_MASK		(~(PGDIR_SIZE-1))
#define PTRS_PER_PGD		(1 << (VA_BITS - PGDIR_SHIFT))

/*
 * Section address mask and size definitions.
 */
#define SECTION_SHIFT		PMD_SHIFT
#define SECTION_SIZE		(_AC(1, UL) << SECTION_SHIFT)
#define SECTION_MASK		(~(SECTION_SIZE-1))

?当配置CONFIG_PGTABLE_LEVELS为4 ，则为4级页表

PGDIR_SHIFT = ARM64_HW_PGTABLE_LEVEL_SHIFT（0）? ?= 39 ，表示VA 中除了本级页表地址，还有39 位表示其它级地址

?PTRS_PER_PGD? = (1 << (VA_BITS - PGDIR_SHIFT))? = 1<<9?

虚拟地址到物理地址转换

virt_to_phys和phys_to_virt

内核虚拟地址起点：VA_START = 0xffff_0000_0000_0000

PAGE_OFFSET? =0xffff_1000_0000_0000

?PAGE_OFFSET - the virtual address of the start of the linear map (top(VA_BITS - 1))

对于48位虚拟地址，从PAGE_OFFSET 开始的往大地址的区域是线性区域，跟物理地址就是一个PHYS_OFFSET 偏差；如果不是线性区域，这个时候是?kimage_voffset? 偏移；

1. #define PHYS_OFFSET ({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
3. #define __is_lm_address(addr) (!!((addr) & BIT(VA_BITS - 1)))
4. #define __lm_to_phys(addr) (((addr) & ~PAGE_OFFSET) + PHYS_OFFSET)
5. #define __kimg_to_phys(addr) ((addr) - kimage_voffset)

1. #define __virt_to_phys_nodebug(x) ({ \
2. phys_addr_t __x = (phys_addr_t)(x); \
3. __is_lm_address(__x) ? __lm_to_phys(__x) : \__kimg_to_phys(__x); \
4. #define __virt_to_phys(x) __virt_to_phys_nodebug(x)
6. static inline phys_addr_t virt_to_phys(const volatile void *x)
7. {
8. return __virt_to_phys((unsigned long)(x));
9. }

kimage_voffset 的获取：

1. #define KERNEL_START _text
2. #define __PHYS_OFFSET (KERNEL_START - TEXT_OFFSET)
3. ENTRY(kimage_vaddr)
4. .quad _text - TEXT_OFFSET
5. /*
6. * The following fragment of code is executed with the MMU enabled.
7. *
8. * x0 = __PHYS_OFFSET
9. */
10. __primary_switched:
11. ldr_l x4, kimage_vaddr // Save the offset between /* 2 */
12. sub x4, x4, x0 // the kernel virtual and /* 3 */
13. str_l x4, kimage_voffset, x5 // physical mappings /* 4 */
15. b start_kernel
17. __primary_switch:
18. bl __enable_mmu
19. ldr x8, =__primary_switched
20. adrp x0, __PHYS_OFFSET /* 1 */
21. br x8

__primary_switch 这里获取的是MMU 没有打开时的_text 链接的地址（相对VMALLOC有一个偏移），加载地址跟链接地址一样；

__primary_switched? x4 获取是运行时_text 运行虚拟地址，这个时候运行地址跟加载地址不一样，从而x4-x0 就是运行虚拟地址跟加载地址的一个偏移存入kimage_voffset。

?参考

device tree 简介 - 灰信网（软件开发博客聚合）

ARM64 Kernel Image Mapping的变化