Android Kernel - Boot Loader

 

Android Boot loader 的 code 在 bootable/bootloader/lk 底下, LK 是 Little Kernel 的缩写, 是 andriod bootloader 的核心精神.

 

入口函数在 kernel/main.c 中的 kmain(), 以下就来读读这一段 code.

 

void kmain(void) { // get us into some sort of thread context thread_init_early(); // early arch stuff arch_early_init(); // do any super early platform initialization platform_early_init(); // do any super early target initialization target_early_init(); dprintf(INFO, "welcome to lk/n/n"); // deal with any static constructors dprintf(SPEW, "calling constructors/n"); call_constructors(); // bring up the kernel heap dprintf(SPEW, "initializing heap/n"); heap_init(); // initialize the threading system dprintf(SPEW, "initializing threads/n"); thread_init(); // initialize the dpc system dprintf(SPEW, "initializing dpc/n"); dpc_init(); // initialize kernel timers dprintf(SPEW, "initializing timers/n"); timer_init(); #if (!ENABLE_NANDWRITE) // create a thread to complete system initialization dprintf(SPEW, "creating bootstrap completion thread/n"); thread_resume(thread_create("bootstrap2", &bootstrap2, NULL, DEFAULT_PRIORITY, DEFAULT_STACK_SIZE)); // enable interrupts exit_critical_section(); // become the idle thread thread_become_idle(); #else bootstrap_nandwrite(); #endif } 

 

In include/debug.h: 我们可以看到 dprintf 的第一个参数是代表 debug level.

 

/* debug levels */ #define CRITICAL 0 #define ALWAYS 0 #define INFO 1 #define SPEW 2  

 

In include/debug.h: 

 

#define dprintf(level, x...) do { if ((level) <= DEBUGLEVEL) { _dprintf(x); } } while (0) 

 

所以 dprintf 会依 DEBUGLEVEL 来判断是否输出信息.

 

来看第一个 call 的函数: thread_init_early, define in thread.c

 

void thread_init_early(void) { int i; /* initialize the run queues */ for (i=0; i < NUM_PRIORITIES; i++) list_initialize(&run_queue[i]); /* initialize the thread list */ list_initialize(&thread_list); /* create a thread to cover the current running state */ thread_t *t = &bootstrap_thread; init_thread_struct(t, "bootstrap"); /* half construct this thread, since we're already running */ t->priority = HIGHEST_PRIORITY; t->state = THREAD_RUNNING; t->saved_critical_section_count = 1; list_add_head(&thread_list, &t->thread_list_node); current_thread = t; }  

 

#define NUM_PRIORITIES 32 in include/kernel/thread.h

 

list_initialize() defined in include/list.h: initialized a list

 

static inline void list_initialize(struct list_node *list) { list->prev = list->next = list; }  

 

run_queue 是 static struct list_node run_queue[NUM_PRIORITIES]

 

thread_list 是 static struct list_node thread_list

 

再来要 call  的函数是: arch_early_init() defined in arch/arm/arch.c

 

void arch_early_init(void) { /* turn off the cache */ arch_disable_cache(UCACHE); /* set the vector base to our exception vectors so we dont need to double map at 0 */ #if ARM_CPU_CORTEX_A8 set_vector_base(MEMBASE); #endif #if ARM_WITH_MMU arm_mmu_init(); platform_init_mmu_mappings(); #endif /* turn the cache back on */ arch_enable_cache(UCACHE); #if ARM_WITH_NEON /* enable cp10 and cp11 */ uint32_t val; __asm__ volatile("mrc p15, 0, %0, c1, c0, 2" : "=r" (val)); val |= (3<<22)|(3<<20); __asm__ volatile("mcr p15, 0, %0, c1, c0, 2" :: "r" (val)); /* set enable bit in fpexc */ val = (1<<30); __asm__ volatile("mcr p10, 7, %0, c8, c0, 0" :: "r" (val)); #endif } 

 

现代操作系统普遍采用虚拟内存管理（Virtual Memory Management）机制，这需要处理器中的MMU（Memory Management Unit，

 

内存管理单元）提供支持。

 

CPU执行单元发出的内存地址将被MMU截获，从CPU到MMU的地址称为虚拟地址（Virtual Address，以下简称VA），而MMU将这个地

 

址翻译成另一个地址发到CPU芯片的外部地址引脚上，也就是将VA映射成PA

 

MMU将VA映射到PA是以页（Page）为单位的，32位处理器的页尺寸通常是4KB。例如，MMU可以通过一个映射项将VA的一页

 

0xb7001000~0xb7001fff映射到PA的一页0x2000~0x2fff，如果CPU执行单元要访问虚拟地址0xb7001008，则实际访问到的物理地

 

址是0x2008。物理内存中的页称为物理页面或者页帧（Page Frame）。虚拟内存的哪个页面映射到物理内存的哪个页帧是通过页

 

表（Page Table）来描述的，页表保存在物理内存中，MMU会查找页表来确定一个VA应该映射到什么PA。

 

 

 

操作系统和MMU是这样配合的：

 

1. 操作系统在初始化或分配、释放内存时会执行一些指令在物理内存中填写页表，然后用指令设置MMU，告诉MMU页表在物理内存中

 

   的什么位置。

 

2. 设置好之后，CPU每次执行访问内存的指令都会自动引发MMU做查表和地址转换操作，地址转换操作由硬件自动完成，不需要用指令

 

    控制MMU去做。

 

 

MMU除了做地址转换之外，还提供内存保护机制。各种体系结构都有用户模式（User Mode）和特权模式（Privileged Mode）之分，

 

操作系统可以在页表中设置每个内存页面的访问权限，有些页面不允许访问，有些页面只有在CPU处于特权模式时才允许访问，有些页面

 

在用户模式和特权模式都可以访问，访问权限又分为可读、可写和可执行三种。这样设定好之后，当CPU要访问一个VA时，MMU会检查

 

CPU当前处于用户模式还是特权模式，访问内存的目的是读数据、写数据还是取指令，如果和操作系统设定的页面权限相符，就允许访

 

问，把它转换成PA，否则不允许访问，产生一个异常（Exception）

 

常见的 segmentation fault 产生的原因:

 

用户程序要访问一段 VA, 经 MMU 检查后无权访问, MMU 会产生异常, CPU 从用户模式切换到特权模式, 跳转到内核代码中执行异常服务程序.

 

内核就会把这个异常解释为 segmentation fault, 将引发异常的程序终止.

 

简单的讲一下 NEON: NEON technology can accelerate multimedia and signal processing algorithms such as video encode/decode,

 

2D/3D graphics, gaming, audio and speech processing, image processing, telephony, and sound synthesis.

 

platform_early_init() defined in platform/<your-platform>/platform.c

 

void platform_early_init(void) { uart_init(); platform_init_interrupts(); platform_init_timer(); }  

 

uart_init.c defined in platform/<your-platform>/uart.c 所有用到的变数,也都定义在 uart.c

 

void uart_init(void) { uwr(0x0A, UART_CR); /* disable TX and RX */ uwr(0x30, UART_CR); /* reset error status */ uwr(0x10, UART_CR); /* reset receiver */ uwr(0x20, UART_CR); /* reset transmitter */ #if PLATFORM_QSD8K /* TCXO */ uwr(0x06, UART_MREG); uwr(0xF1, UART_NREG); uwr(0x0F, UART_DREG); uwr(0x1A, UART_MNDREG); #else /* TCXO/4 */ uwr(0xC0, UART_MREG); uwr(0xAF, UART_NREG); uwr(0x80, UART_DREG); uwr(0x19, UART_MNDREG); #endif uwr(0x10, UART_CR); /* reset RX */ uwr(0x20, UART_CR); /* reset TX */ uwr(0x30, UART_CR); /* reset error status */ uwr(0x40, UART_CR); /* reset RX break */ uwr(0x70, UART_CR); /* rest? */ uwr(0xD0, UART_CR); /* reset */ uwr(0x7BF, UART_IPR); /* stale timeout = 630 * bitrate */ uwr(0, UART_IMR); uwr(115, UART_RFWR); /* RX watermark = 58 * 2 - 1 */ uwr(10, UART_TFWR); /* TX watermark */ uwr(0, UART_RFWR); uwr(UART_CSR_115200, UART_CSR); uwr(0, UART_IRDA); uwr(0x1E, UART_HCR); // uwr(0x7F4, UART_MR1); /* RFS/ CTS/ 500chr RFR */ uwr(16, UART_MR1); uwr(0x34, UART_MR2); /* 8N1 */ uwr(0x05, UART_CR); /* enable TX & RX */ uart_ready = 1; }  

 

platform_init_interrupts: defined in platform/msm8x60/interrupts.c

 

void platform_init_interrupts(void) { platform_gic_dist_init(); platform_gic_cpu_init(); } 

 

GIC 指的是 Generic Interrupt Controller. The gic-cpu and gic-dist are two subcomponents of GIC.

 

Devices are wired to the git-dist which is in charge of distributing interrupts to the gic-cpu (per cpu IRQ IF).

 

platform_init_timer(): defined in platform/<your-platform>/timer.c

 

void platform_init_timer(void) { writel(0, DGT_ENABLE); } 

 

DGT: Digital Game Timer: presents the countdowns of two players, it is also called chess timer or chess clock.

 

target_early_init(): defined in target/init.c

 

/* * default implementations of these routines, if the target code * chooses not to implement. */ __WEAK void target_early_init(void) { } __WEAK void target_init(void) { }  

 

call_constructors() is defined in kernel/main.c:

 

static void call_constructors(void) { void **ctor; ctor = &__ctor_list; while(ctor != &__ctor_end) { void (*func)(void); func = (void (*)())*ctor; func(); ctor++; } } 

 

 

heap_init is defined in lib/heap/heap.c:

 

void heap_init(void) { LTRACE_ENTRY; // set the heap range theheap.base = (void *)HEAP_START; theheap.len = HEAP_LEN; LTRACEF("base %p size %zd bytes/n", theheap.base, theheap.len); // initialize the free list list_initialize(&theheap.free_list); // create an initial free chunk heap_insert_free_chunk(heap_create_free_chunk(theheap.base, theheap.len)); // dump heap info // heap_dump(); // dprintf(INFO, "running heap tests/n"); // heap_test(); }  

 

 

thread_init is defined in kernel/thread.c but nothing coded, 先记下.

 

void thread_init(void) { }  

 

dpc_init() is defined in kernel/dpc.c:

 

 

void dpc_init(void) { event_init(&dpc_event, false, 0); thread_resume(thread_create("dpc", &dpc_thread_routine, NULL, DPC_PRIORITY, DEFAULT_STACK_SIZE)); } 

 

dpc 为 Delayed Procedure Call 延迟过程调用的缩写.

 

timer_init() is defined in kernel/timer.c:

 

void timer_init(void) { list_initialize(&timer_queue); /* register for a periodic timer tick */ platform_set_periodic_timer(timer_tick, NULL, 10); /* 10ms */ } 

 

 

 

执行 thread_resume 或是 bootstrap_nandwrite

 

 

#if (!ENABLE_NANDWRITE) // create a thread to complete system initialization dprintf(SPEW, "creating bootstrap completion thread/n"); thread_resume(thread_create("bootstrap2", &bootstrap2, NULL, DEFAULT_PRIORITY, DEFAULT_STACK_SIZE)); // enable interrupts exit_critical_section(); // become the idle thread thread_become_idle(); #else bootstrap_nandwrite(); #endif 

 

In kermel/main.c:

 

static int bootstrap2(void *arg) { dprintf(SPEW, "top of bootstrap2()/n"); arch_init(); // initialize the rest of the platform dprintf(SPEW, "initializing platform/n"); platform_init(); // initialize the target dprintf(SPEW, "initializing target/n"); target_init(); dprintf(SPEW, "calling apps_init()/n"); apps_init(); return 0; } #if (ENABLE_NANDWRITE) void bootstrap_nandwrite(void) { dprintf(SPEW, "top of bootstrap2()/n"); arch_init(); // initialize the rest of the platform dprintf(SPEW, "initializing platform/n"); platform_init(); // initialize the target dprintf(SPEW, "initializing target/n"); target_init(); dprintf(SPEW, "calling nandwrite_init()/n"); nandwrite_init(); return 0; } #endif 

 

continue to see apps_init(): app/app.c:void apps_init(void)

 

#include <app.h> #include <kernel/thread.h> extern const struct app_descriptor __apps_start; extern const struct app_descriptor __apps_end; static void start_app(const struct app_descriptor *app); /* one time setup */ void apps_init(void) { const struct app_descriptor *app; /* call all the init routines */ for (app = &__apps_start; app != &__apps_end; app++) { if (app->init) app->init(app); } /* start any that want to start on boot */ for (app = &__apps_start; app != &__apps_end; app++) { if (app->entry && (app->flags & APP_FLAG_DONT_START_ON_BOOT) == 0) { start_app(app); } } } static int app_thread_entry(void *arg) { const struct app_descriptor *app = (const struct app_descriptor *)arg; app->entry(app, NULL); return 0; } static void start_app(const struct app_descriptor *app) { printf("starting app %s/n", app->name); thread_resume(thread_create(app->name, &app_thread_entry, (void *)app, DEFAULT_PRIORITY, DEFAULT_STACK_SIZE)); }  

 

至于会有那些 app 被放入 boot thread section, 则定义在 include/app.h 中的 APP_START(appname)

 

#define APP_START(appname) struct app_descriptor _app_##appname __SECTION(".apps") = { .name = #appname, #define APP_END };  

 

在 app 中只要像 app/aboot/aboot.c 指定就会在 bootloader bootup 时放入 thread section 中被执行.

 

APP_START(aboot) .init = aboot_init, APP_END  

 

在我的 bootloader 中有 app/aboot/aboot.c, app/tests/tests.c, app/shell/shell.c, 及 app/stringtests/string_tests.c 皆有此声明.

 

接下来关注: aboot.c 中的 aboot_init()

 

void aboot_init(const struct app_descriptor *app) { unsigned reboot_mode = 0; unsigned disp_init = 0; unsigned usb_init = 0; //test_ram(); /* Setup page size information for nand/emmc reads */ if (target_is_emmc_boot()) { page_size = 2048; page_mask = page_size - 1; } else { page_size = flash_page_size(); page_mask = page_size - 1; } /* Display splash screen if enabled */ #if DISPLAY_SPLASH_SCREEN display_init(); dprintf(INFO, "Diplay initialized/n"); disp_init = 1; diplay_image_on_screen(); #endif /* Check if we should do something other than booting up */ if (keys_get_state(KEY_HOME) != 0) boot_into_recovery = 1; if (keys_get_state(KEY_BACK) != 0) goto fastboot; if (keys_get_state(KEY_CLEAR) != 0) goto fastboot; #if NO_KEYPAD_DRIVER /* With no keypad implementation, check the status of USB connection. */ /* If USB is connected then go into fastboot mode. */ usb_init = 1; udc_init(&surf_udc_device); if (usb_cable_status()) goto fastboot; #endif init_vol_key(); if(voldown_press()) goto fastboot; reboot_mode = check_reboot_mode(); if (reboot_mode == RECOVERY_MODE) { boot_into_recovery = 1; } else if(reboot_mode == FASTBOOT_MODE) { goto fastboot; } if (target_is_emmc_boot()) { boot_linux_from_mmc(); } else { recovery_init(); boot_linux_from_flash(); } dprintf(CRITICAL, "ERROR: Could not do normal boot. Reverting " "to fastboot mode./n"); fastboot: if(!usb_init) udc_init(&surf_udc_device); fastboot_register("boot", cmd_boot); if (target_is_emmc_boot()) { fastboot_register("flash:", cmd_flash_mmc); fastboot_register("erase:", cmd_erase_mmc); } else { fastboot_register("flash:", cmd_flash); fastboot_register("erase:", cmd_erase); } fastboot_register("continue", cmd_continue); fastboot_register("reboot", cmd_reboot); fastboot_register("reboot-bootloader", cmd_reboot_bootloader); fastboot_publish("product", TARGET(BOARD)); fastboot_publish("kernel", "lk"); fastboot_init(target_get_scratch_address(), 120 * 1024 * 1024); udc_start(); target_battery_charging_enable(1, 0); }  

 

 

target_is_emmc_boot() is defined in target/init.c: _EMMC_BOOT 是 compiler 时的 flags

 

__WEAK int target_is_emmc_boot(void) { #if _EMMC_BOOT return 1; #else return 0; #endif } 

 

check_reboot_mode is defined in target/<your-platform>/init.c:

 

unsigned check_reboot_mode(void) { unsigned restart_reason = 0; void *restart_reason_addr = 0x401FFFFC; /* Read reboot reason and scrub it */ restart_reason = readl(restart_reason_addr); writel(0x00, restart_reason_addr); return restart_reason; }  

 

reboot mode in bootloader:

 

#define RECOVERY_MODE 0x77665502 #define FASTBOOT_MODE 0x77665500 

 

 

再来就会执行 boot_linux_from_mmc():

 

int boot_linux_from_mmc(void) { struct boot_img_hdr *hdr = (void*) buf; struct boot_img_hdr *uhdr; unsigned offset = 0; unsigned long long ptn = 0; unsigned n = 0; const char *cmdline; uhdr = (struct boot_img_hdr *)EMMC_BOOT_IMG_HEADER_ADDR; if (!memcmp(uhdr->magic, BOOT_MAGIC, BOOT_MAGIC_SIZE)) { dprintf(INFO, "Unified boot method!/n"); hdr = uhdr; goto unified_boot; } if(!boot_into_recovery) { ptn = mmc_ptn_offset("boot"); if(ptn == 0) { dprintf(CRITICAL, "ERROR: No boot partition found/n"); return -1; } } else { ptn = mmc_ptn_offset("recovery"); if(ptn == 0) { dprintf(CRITICAL, "ERROR: No recovery partition found/n"); return -1; } } if (mmc_read(ptn + offset, (unsigned int *)buf, page_size)) { dprintf(CRITICAL, "ERROR: Cannot read boot image header/n"); return -1; } if (memcmp(hdr->magic, BOOT_MAGIC, BOOT_MAGIC_SIZE)) { dprintf(CRITICAL, "ERROR: Invaled boot image header/n"); return -1; } if (hdr->page_size && (hdr->page_size != page_size)) { page_size = hdr->page_size; page_mask = page_size - 1; } offset += page_size; n = ROUND_TO_PAGE(hdr->kernel_size, page_mask); if (mmc_read(ptn + offset, (void *)hdr->kernel_addr, n)) { dprintf(CRITICAL, "ERROR: Cannot read kernel image/n"); return -1; } offset += n; n = ROUND_TO_PAGE(hdr->ramdisk_size, page_mask); if (mmc_read(ptn + offset, (void *)hdr->ramdisk_addr, n)) { dprintf(CRITICAL, "ERROR: Cannot read ramdisk image/n"); return -1; } offset += n; unified_boot: dprintf(INFO, "/nkernel @ %x (%d bytes)/n", hdr->kernel_addr, hdr->kernel_size); dprintf(INFO, "ramdisk @ %x (%d bytes)/n", hdr->ramdisk_addr, hdr->ramdisk_size); if(hdr->cmdline[0]) { cmdline = (char*) hdr->cmdline; } else { cmdline = DEFAULT_CMDLINE; } dprintf(INFO, "cmdline = '%s'/n", cmdline); dprintf(INFO, "/nBooting Linux/n"); boot_linux((void *)hdr->kernel_addr, (void *)TAGS_ADDR, (const char *)cmdline, board_machtype(), (void *)hdr->ramdisk_addr, hdr->ramdisk_size); return 0; }  

 

mmc_read() is defined in platform/<your-platform>/mmc.c:

 

unsigned int mmc_read (unsigned long long data_addr, unsigned int* out, unsigned int data_len) { int val = 0; val = mmc_boot_read_from_card( &mmc_host, &mmc_card, data_addr, data_len, out); return val; }  

 

boot_linux(): 启动 Linux, 看看函数定义

 

void boot_linux(void *kernel, unsigned *tags, const char *cmdline, unsigned machtype, void *ramdisk, unsigned ramdisk_size) { unsigned *ptr = tags; unsigned pcount = 0; void (*entry)(unsigned,unsigned,unsigned*) = kernel; struct ptable *ptable; int cmdline_len = 0; int have_cmdline = 0; int pause_at_bootup = 0; /* CORE */ *ptr++ = 2; *ptr++ = 0x54410001; if (ramdisk_size) { *ptr++ = 4; *ptr++ = 0x54420005; *ptr++ = (unsigned)ramdisk; *ptr++ = ramdisk_size; } ptr = target_atag_mem(ptr); if (!target_is_emmc_boot()) { /* Skip NAND partition ATAGS for eMMC boot */ if ((ptable = flash_get_ptable()) && (ptable->count != 0)) { int i; for(i=0; i < ptable->count; i++) { struct ptentry *ptn; ptn = ptable_get(ptable, i); if (ptn->type == TYPE_APPS_PARTITION) pcount++; } *ptr++ = 2 + (pcount * (sizeof(struct atag_ptbl_entry) / sizeof(unsigned))); *ptr++ = 0x4d534d70; for (i = 0; i < ptable->count; ++i) ptentry_to_tag(&ptr, ptable_get(ptable, i)); } } if (cmdline && cmdline[0]) { cmdline_len = strlen(cmdline); have_cmdline = 1; } if (target_is_emmc_boot()) { cmdline_len += strlen(emmc_cmdline); } if (target_pause_for_battery_charge()) { pause_at_bootup = 1; cmdline_len += strlen(battchg_pause); } if (cmdline_len > 0) { const char *src; char *dst; unsigned n; /* include terminating 0 and round up to a word multiple */ n = (cmdline_len + 4) & (~3); *ptr++ = (n / 4) + 2; *ptr++ = 0x54410009; dst = (char *)ptr; if (have_cmdline) { src = cmdline; while ((*dst++ = *src++)); } if (target_is_emmc_boot()) { src = emmc_cmdline; if (have_cmdline) --dst; have_cmdline = 1; while ((*dst++ = *src++)); } if (pause_at_bootup) { src = battchg_pause; if (have_cmdline) --dst; while ((*dst++ = *src++)); } ptr += (n / 4); } /* END */ *ptr++ = 0; *ptr++ = 0; dprintf(INFO, "booting linux @ %p, ramdisk @ %p (%d)/n", kernel, ramdisk, ramdisk_size); if (cmdline) dprintf(INFO, "cmdline: %s/n", cmdline); enter_critical_section(); platform_uninit_timer(); arch_disable_cache(UCACHE); arch_disable_mmu(); #if DISPLAY_SPLASH_SCREEN display_shutdown(); #endif entry(0, machtype, tags); }  

 

配合 boot.img 来看会比较好理解.

 

 

 

由此可知 boot_img_hdr 中各成员值为：

 

 

 

TAGS_ADDR 如上 target/<your-platform>/rules.mk 所定义的 : 0x40200100, 所以 boot_linux(), 就是传入TAGS_ADDR,

 

然后将资料写入 tag, tag 的结构如下所示.

 

 

然后进入到 kernel 的入口函数: entry(0, machtype, tags)