/* Interrupts are enabled now so all GFP allocations are safe. */
gfp_allowed_mask = __GFP_BITS_MASK;
// gfp_allowed_mask 默认为 GFP_BOOT_MASK = (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS));
// 显然原来在调用page allocator时是不能sleep的,现在中断处理什么的都ready了,可以去掉这些限制了
kmem_cache_init_late(); 空函数,什么也没做.
console_init(); @see drivers/tty/tty_io.c#3212
tty_ldisc_begin();
tty_register_ldisc(N_TTY = 0, &tty_ldisc_N_TTY);
tty_ldiscs[NR_LDISCS = 30] = {
0 => tty_ldisc_N_TTY // tty_ldisc_N_TTY 定义了各项操作 open, close, read, write 对应的函数
}
接下来一个while循环把__con_initcall_start到end的函数挨个执行一遍.
// 查System.map可知 start = 0x1b88428, end = 0x1b88448, 所以应该有0x448-0x428 = 0x20 / 8 = 4个函数要执行,它们分别是:
// con_init, hvc_console_init, xen_cons_init, serial8250_console_init
con_init();
// 在setup_arch里conswitchp = &vga_con;
display_desc = conswitchp->con_startup(); // vgacon_startup()
/// screen_info是在setup_arch里直接从boot_params.screen_info copy的, boot_params.screen_info是grub加载内核时填充的
// bochs: writemem "/tmp/boot_params.memdump" 0xffffffff81b28380 4096
/* ./print_boot_params /tmp/boot_params.memdump
screen_info.orig_x = 0
screen_info.orig_y = 0
screen_info.ext_mem_k = 0x8000
screen_info.orig_video_page = 0
screen_info.orig_video_mode = 3
screen_info.orig_video_cols = 80
screen_info.flags = 0
screen_info.orig_video_ega_bx = 0
screen_info.orig_video_lines = 25
screen_info.orig_video_isVGA = 1
screen_info.orig_video_points = 16
*/
vga_video_num_lines = 25;
vga_video_num_columns = 80;
vga_can_do_color = 1;
vga_vram_base = 0xb8000;
vga_video_port_reg = VGA_CRT_IC = 0x3d4;
vga_video_port_val = VGA_CRT_DC = 0x3d5;
vga_vram_size = 0x8000;
vga_console_resource = {.name = "vga+", .start = 0x3c0, .end = 0x3df}
vga_video_type = VIDEO_TYPE_VGAC;
display_desc = "VGA+";
request_resource(&ioport_resource, &vga_console_resource);
// @see http://www.jamesmolloy.co.uk/tutorial_html/3.-The%20Screen.html
...
vga_vram_base = 0xffff8800000b8000
vga_vram_end = 0xffff8800000c0000
vga_hardscroll_enabled = 1;
vga_default_font_height = 16;
vga_video_font_height = 16;
vga_scan_lines = 16 * 25 = 400
vga_xres = 80 * 8 = 640
vga_yres = 400
// .config里没有定义 CONFIG_VGACON_SOFT_SCROLLBACK, 所以vgacon_scrollback_startup();不产生任何实际指令
vga_init_done = 1;
registered_con_driver[MAX_NR_OCN_DRIVER] 当前是空的,接下来 registered_con_dirver[0] = {conswitchp}
con_driver_map当前也是空的,把它全都填充成conswitchp // MAX_NR_CONSOLES = 63
blankinterval = 600
blank_state = blank_normal_wait = 1
mod_timer(&console_timer, jiffies + (blankinterval * HZ)); // 把console_timer设为10分钟以后
MIN_NR_CONSOLES = 1
接下来一个for循环,初始化了一个vc(virtual console).
printable = 1
register_console(&vt_console_driver);
// 理一理console之间的关系
struct vc {
struct vc_data *d;
struct work_struct SAK_work;
}
// kernel定义了63个vc, struct vc vc_cons [MAX_NR_CONSOLES = 63]
// 当一个vc初始化时,给它分配一块内存用作vc_data,然后INIT SAK_work,其实就是把函数vc_SAK加到workqueue里,那么什么是Secure Attention Key(SAK)呢?
// 比如有一些木马程序伪装成login program,让用户输入用户名密码,从而窃取帐号,SAK的作用就是允许用户按下键盘上的一个组合键,然后触发已经加到workqueue里的vc_SAK,
// vc_SAK()把当前终端关联的所有程序全都kill掉,然后启动真的login program,从而让钓鱼程序起不上作用.
struct vc_data {
struct tty_port port;
unsigned short vc_num;
unsigned int vc_cols;
unsigned int vc_rows;
...
const struct consw *vc_sw;
unsigned short *vc_screenbuf;
unsigned int vc_screenbuf_size;
...
}
// 在初始化vc的过程中,会设 vc_data.vc_sw = conswitchp; 对应我们的情况,就是 vga_con, 这个是真正干活的那个,在vc初始化之前, vgacon_startup()就已经运行过了,
// vgacon_startup()根据GRUB传递过来的screen_info参数,得知vgacon的情况,比如是否支持颜色,console有多少行多少列等.
// vc初始化过程中,会把vgacon的相关参数拿到,设到自身的数据结构中(visual_init -> con_init -> vgacon_init).
// 然后根据console有多少行,每行的大小计算出screenbuf_size,再分配内存给vc->screenbuf.
vc初始化好之后,选一个当作fg_console(foreground),我们目前只有一个vc,那就是当前这个了.
如果有多个的话,显然多个vc之间的切换其实就是选择不同的vc做fg_console,然后把vc->screenbuf的内容打印出屏幕上.
多个vc最终归属到struct console上,在con_init的结尾,register_console(&vt_console_driver)把vt_console_driver加到了console_drivers里,
vt_console_driver的write方法,也就是vt_console_print,找到fg_console,把字符打印到屏幕上.
打印字符其实很简单了,关键是要先把pos给计算好,然后*pos = char就可以了.
printk先把msg保存到log_buf里,log_buf默认大小为 1 << 18 = 256K,之后调用console_unlock时会遍历console_drivers,调用对应的console_driver的write方法把msg打印到屏幕上.
// 接下来的三个init(hvc_console_init, xen_cons_init, serial8250_console_init), 由于con_init register_console时已经把preferred_console设为vt_console_driver了,
// 所以hvc_console_init,serial8250_console_init尝试register_console(&hvc_console),register_console(&serial8250_console);失败,而xen_cons_init由于xen_pv_domain()返回0,早早的就返回了
if (panic_later) panic(...); panic_later = 0
lockdep_info();locking_selftest(); 由于.config里没有定义CONFIG_LOCKDEP和CONFIG_DEBUG_LOCKING_API_SELFTESTS,这两个函数不产生任何实际指令.
if (initrd_start && !initrd_below_start_ok && page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n", page_to_pfn(virt_to_page((void *)initrd_start)), min_low_pfn);
initrd_start = 0;
}
// initrd_start 在setup_arch -> reserve_initrd时赋值为 0xFFFF88000796E000
// initrd_below_start_ok = 0
// min_low_pfn = 0
由于if的最后一个条件不满足,这段代码啥也没做
page_cgroup_init();先跳过cgroup相关的代码.
enable_debug_pagealloc(); kmemleak_init(); debug_objects_mem_init();
.config里没有定义CONFIG_DEBUG_PAGEALLOC,CONFIG_DEBUG_KMEMLEAK,CONFIG_DEBUG_OBJECTS, 所以这三个函数不产生任何实际指令.
setup_per_cpu_pageset();
原来 zone->pageset=&boot_pageset, 这里重新分配了percpu pageset,不清楚做什么用的.
numa_policy_init();我们只有一个dummy_node,这个policy应该没啥作用,先跳过去.
if (late_time_init) late_time_init(); late_time_init = x86_late_time_init;
x86_init.timers.timer_init(); // => hpet_time_init();
// 在setup_arch -> acpi_boot_init -> acpi_table_parse(ACPI_SIG_HPET, acpi_parse_hpet) 时, 我们就知道没有hpet,这是因为bochs没有提供, 所以:
setup_pit_timer();
/* @see http://wiki.osdev.org/PIT
The Programmable Interval Timer (PIT) chip (also called an 8253/8254 chip) basically consists of an oscillator, a prescaler and 3 independent frequency dividers.
The oscillator used by the PIT chip runs at (roughly) 1.193182 MHz.
The basic principle of a frequency divider is to divide one frequency to obtain a slower frequency. This is typically done by using a counter.
Each "pulse" from the input frequency causes the counter to be decreased, and when that counter has reached zero a pulse is generated on the output and the counter is reset.
The PIT has only 16 bits that are used as frequency divider, which can represent the values from 0 to 65535.
Since the frequency can't be divided by 0 in a sane way, many implementations use 0 to represent the value 65536.
The PIT chip has three separate frequency dividers (or 3 separate channels) that are programmable, in that the value of the "reset counter" is set by software (the OS).
Software also specifies an action to be taken when the counter reaches zero on each individual channel.
In this way, each channel can be used in one of several "modes" - for example, as a frequency divider (where the count is automatically reset) or
as a "one shot" timer (where the count isn't automatically reset).
Channel 0 is connected directly to IRQ0, so it is best to use it only for purposes that should generate interrupts.
Channel 1 is unusable, and may not even exist.
Channel 2 is connected to the PC speaker, but can be used for other purposes without producing audible speaker tones.
The PIT chip uses the following I/O ports:
I/O port Usage
0x40 Channel 0 data port (read/write)
0x41 Channel 1 data port (read/write)
0x42 Channel 2 data port (read/write)
0x43 Mode/Command register (write only, a read is ignored)
*/
/*
看了OSDEV上对PIT的介绍我们知道,PIT oscillator runs at (roughly) 1.193182 MHz, 要想获取一个低频率的, 需要用到frequency divider,也就是个counter.
counter值的范围是0-65535,也就是说,最慢我们能得到一个1193182/65535=18.2Hz=55.6ms的频率,最快就是1000000000ns/1193182=838ns
这个静态的值称为reload value, 而实际运行过程中动态的counter值称为current counter,如果得到一个current counter值,怎么能计算出它相对上次reload的时间点过去多长时间呢?
每次tick是838ns,乘以current counter就得到结果了
kernel代码里不是这样做的,它先根据定义好的shift值,计算出一个multiplication factor,来做tick和ns之间的换算. 数学不好,不明白这是怎么个道理.
*/
// pit_ce 0x1a0e600
pit_ce.name = "pit";
pit_ce.mult = 0x4e3245;
pit_ce.max_delta_ns = 0x1a308ea;
pit_ce.min_delta_ns = 0x311b;
pit_ce.shift = 0x20;
// max_delta_ns = 0x7FFF ticks, min_delta_ns = 0xF ticks, 也就是说kernel配置的的reload value范围在0xF-0x7FFF之间???
pit_ce.features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT;
// 这两个feature好理解,PERIODIC就是counter减为0的话,PIT会把counter重设为reload value,继续
// ONESHOT就是减为0后就不变了,等着程序去控制和处理接下来的事
pit_ce.set_mode = init_pit_timer;
// 当mode是PERIODIC时,先向port 0x43(Mode/Command register)写入0x34=0x0011 0100,表示:
/* 00 - select channel 0
11 - Access mode: lobyte/hibyte
010 - Operating mode: Mode 2 (rate generator)
0 - BCD/Binary mode: 16-bit binary
*/
// 接着向port 0x40(channel 0 data port)写入LATCH = (CLOCK_TICK_RATE + HZ/2) / HZ = 0x12a5 = 4773 (4ms的频率)
// 由于一次只能写入8bit,先写lowbyte,再写highbyte, LATCH & 0xff 得到 lowbyte, LATCH >> 8 得到highbyte
// 当mode是SHUTDOWN或者是UNUSED时,向port 0x43(Mode/Command register)写入0x30=0x0011 0000,表示select channel 0, access mode lo/hi, ,operating mode: mode 0
// 然后向data port写入0,也就是把reload value调到最大,其实就相当于电脑刚启动时PIT的状态
// Typically during boot the BIOS sets channel 0 with a count of 65535 or 0 (which translates to 65536),
// which gives an output frequency of 18.2065 Hz (or an IRQ every 54.9254 ms)
// 当mode是ONESHOT时,向mode/command register写入0x38=0x0011 1000,表示operate mode: Mode 4 (software triggered strobe)
pit_ce.set_next_event = pit_next_event;
// pit_next_event()做的事就是修改reload value的值,配合ONESHOT,可以实现sleep功能了
// 总结一下,pit_ce给出了把PIT ticks和ns之间的换算办法,调用set_mode可以配置pit的工作模式和频率,set_next_event配合ONESHOT可以实现程序控制的timer
clockevents_register_device(&pit_ce);
static LIST_HEAD(clockevent_devices);
static RAW_NOTIFIER_HEAD(clockevents_chain);
list_add(&dev->list, &clockevent_devices);
clockevents_do_notify(CLOCK_EVT_NOTIFY_ADD, dev);
/* 在 start_kernel 的刚开始,有一个tick_init(),把tick_notifier加到了clockents_chain里, 当前clockevents_chain只有tick_notifier这一个
clockevents_do_notify调用了 tick_notifer 的handler, 对ADD事件, 处理方法是 tick_check_new_device
看下kernel/time/tick-common.c和include/linux/tick.h,看起来是每个cpu都会有一个tick_device.
struct tick_device {
struct clock_event_device *evtdev;
enum tick_device_mode mode; // 要么是PERIODIC,要么是ONESHOT
}
每个cpu的tick_device使用per_cpu变量tick_cpu_device保存,当前是空的. tick_cpu_device = 0x101a0 + 0x7c00000 = 0x7c101a0
tick_check_new_device先将pit_ce shutdown // clockevents_exchange_device(curdev = NULL, newdev = pit_ce);
然后把pit_ce设到当前cpu的tick_device里,并设mode为PERIODIC.
接着把pit_ce.event_handler设为tick_handle_periodic,并启用pit_ce的PERIODIC mode
OK, 到此,pit算是设好了
*/
clockevents_notify_released(); // 这个函数把上边do_notify过程中release的device重新加到clockevent_devices list里去
global_clock_event = &pit_ce;
setup_default_timer_irq();
setup_irq(0, &irq0);
// 前边在init_IRQ时,irq 0做为legacy irq,分配了irq_desc,并将irq_desc.handle_irq设为handle_level_irq
// irq_settings_can_autoenable, irq_setup(desc), 看来init_IRQ里虽然初始化了PIC,但并没有让irq 0起作用,irq_setup unmask了 irq 0
// 然后将irq_desc.action设为irq0, irq_count = 0, irqs_unhandled = 0
register_irq_proc(irq, desc); // 由于root_irq_dir = NULL, 这个函数早早的返回了
new->dir = NULL;
register_handler_proc(irq, new); // 也什么都没做就返回了
tsc_init(); // @see UnderstandingKernel Time Stamp Counter (TSC)
x86_init.timers.tsc_pre_init(); // => x86_init_noop
tsc_khz = x86_platform.calibrate_tsc(); // => native_calibrate_tsc, 在bochs上使用PIT计算出的结果是 0xf9f khz
syslog打印出 Fast TSC calibration using PIT. Detected 3.999 MHz processor. // 现在console能用了,就是好呀
// 接下来注释里写的很明白了,先假设所有的cpu都是这个速度
lpj_fine = 3999 * 1000 / 250 = 0x3e7c // lpj的意思也就是loops per jiffy, 一个tick是一个jiffy,在这一个tick的时间里,cpu能执行多少个nop指令
use_tsc_delay();
tsc_clocksource_reliable = 0;
sched_clock_init(); .config里配置了 CONFIG_HAVE_UNSTABLE_SCHED_CLOCK=y, 所以这个函数在 kernel/sched_clock.c#0105, 还不太明白具体有什么作用
这个函数做的事就是用当前时间(wall time)来初始化每个cpu的sched_clock_data. 然后标记 sched_clock_running = 1.
calibrate_delay();
打印出 Calibrating delay loop (skipped), value calculated using timer frequency.. 7.99 BogoMIPS (lpj=15996=0x3e7c)
@see Linux Kernel Development Page 227