Skip to content

Latest commit

 

History

History
1387 lines (1284 loc) · 55.8 KB

ftrace-design.md

File metadata and controls

1387 lines (1284 loc) · 55.8 KB

ftrace 的实现

https://www.ibm.com/developerworks/cn/linux/l-cn-ftrace/image001.jpg

Function tracer的实现

  • Ftrace 采用 GCC 的 profile 特性在所有内核函数的开始部分加入一段 stub 代码,ftrace 重载这段代码来实现 trace 功能。
  • gcc 的 -pg 选项将在每个函数入口处加入对 mcount 的调用代码。
  • 增加 -pg 选项后,gcc 在函数的入口处加入了对 mcount 的调用:call _mcount
  • 原本 mcount 由 libc 实现,但内核不会连接 libc 库,因此 ftrace 编写了自己的 mcount stub 函数,并借此实现 trace 功能。
  • 在每个内核函数入口加入 trace 代码,必然会影响内核的性能,为了减小对内核性能的影响,ftrace 支持动态 trace 功能。
  • 当未 CONFIG_DYNAMIC_FTRACE 选中时:
    • 每个函数调用都会根据不同的体系结构的实现调用mcount函数,
    • 如果 ftrace 使能了某些跟踪器,ftrace_trace_function指针不再指向ftrace_stub,而是指向具体的跟踪函数。
    • 否则就执行到体系结构相关的ftrace_stub从函数返回。
  • CONFIG_DYNAMIC_FTRACE 被选中后:
    • 内核编译时会调用一个 perl 脚本:recordmcount.pl 将每个函数的地址写入一个特殊的段:__mcount_loc
    • 在内核初始化的初期,ftrace 查询 __mcount_loc 段,得到每个函数的入口地址,并将 mcount 替换为 nop 指令。这样在默认情况下,ftrace 不会对内核性能产生影响。
    • 当用户打开 ftrace 功能时,ftrace 将这些 nop 指令动态替换为 ftrace_caller,该函数将调用用户注册的 trace 函数。

ftrace 依赖的-pg编译器选项

  • Makefile

     ifdef CONFIG_FUNCTION_TRACER
     ifndef CC_FLAGS_FTRACE
     CC_FLAGS_FTRACE := -pg
     endif
     export CC_FLAGS_FTRACE
     ...
     KBUILD_CFLAGS	+= $(CC_FLAGS_FTRACE) ...
  • 对于tracer自身而言,是不需要-pg选项的,因此我们可以看到在kernel/tracing/Makefile中将-pg从该模块的CFLAGS中剔除了。

  • kernel/tracing/Makefile

     # Do not instrument the tracer itself:
    
     ifdef CONFIG_FUNCTION_TRACER
     ORIG_CFLAGS := $(KBUILD_CFLAGS)
     KBUILD_CFLAGS = $(subst $(CC_FLAGS_FTRACE),,$(ORIG_CFLAGS))
    
     ifdef CONFIG_FTRACE_SELFTEST
     # selftest needs instrumentation 如自测功能开启单独给 trace_selftest_dynamic.c 文件插桩。
     CFLAGS_trace_selftest_dynamic.o = $(CC_FLAGS_FTRACE)
     obj-y += trace_selftest_dynamic.o
     endif
     endif
    
  • 对于不希望被 ftrace 跟踪的模块也可以如法炮制。

  • notrace标记的 函数 不会被跟踪。

  • CFLAGS_REMOVE_file.o = -pg禁止对 某个文件-pg编译

ftrace跟踪函数

  • ftrace_trace_function是个 函数指针call mcount时会调用该指针指向的函数。
    • 函数原型
    • include/linux/ftrace.h
       typedef void (*ftrace_func_t)(unsigned long ip, unsigned long parent_ip,
                                     struct ftrace_ops *op, struct pt_regs *regs);
    • 定义和初始值
    • kernel/trace/ftrace.c
       ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
  • ftrace_stub则是一个与体系结构相关 函数,通常是用汇编实现的一条简单的 从函数返回 的指令:
    • arch/powerpc/kernel/entry_64.S: blr
    • arch/powerpc/kernel/entry_32.S: blr
    • arch/arm/kernel/entry-ftrace.S: ret lr
    • arch/arm64/kernel/entry-ftrace.S: ret
    • arch/x86/entry/entry_32.S: ret
    • arch/x86/kernel/mcount_64.S: retq

ftrace函数跟踪器的 x86 64 实现

  • 在该文件里,会用function_hook来替换mcount
    • arch/x86/kernel/mcount_64.S
       #ifdef CC_USING_FENTRY
       # define function_hook  __fentry__
       EXPORT_SYMBOL(__fentry__)
       #else
       # define function_hook  mcount
       EXPORT_SYMBOL(mcount) /*记得为 modules 导出符号*/
       #endif

CONFIG_DYNAMIC_FTRACE未开启

  • 对于CONFIG_DYNAMIC_FTRACE没有开启的情况,显然,在关闭 ftrace 的时候也会有性能损失。

  • 具体表现在,无论是否开启 ftrace 调用函数时都会先调用 mcount,总是至少会多执行两条指令,即使ftrace_trace_function没有被指向某个跟踪函数。

  • arch/x86/kernel/mcount_64.S

ENTRY(function_hook)
        cmpq $ftrace_stub, ftrace_trace_function ; cmpq: compare quad words,比较 ftrace_stub 和指针 ftrace_trace_function
        jnz trace ; 如果上面比较结果不为 0,即不相等,跳转到标签 trace,开始跟踪;否则往下执行

fgraph_trace:
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
        cmpq $ftrace_stub, ftrace_graph_return ; 接着比较,可见如果开启了CONFIG_FUNCTION_GRAPH_TRACER,即使关闭 ftrace,也会有此比较
        jnz ftrace_graph_caller ; 如果上面比较结果不为 0,即不相等,跳转到标签 ftrace_graph_caller,开始跟踪;否则往下执行

        cmpq $ftrace_graph_entry_stub, ftrace_graph_entry ; 含义和之前类似
        jnz ftrace_graph_caller
#endif

GLOBAL(ftrace_stub)
        retq ; 从 mcount 函数返回

trace:
        /* save_mcount_regs fills in first two parameters */
        save_mcount_regs ; 一个汇编宏,目的是填入前两个参数:ip 和 parent_ip

        /*
         * When DYNAMIC_FTRACE is not defined, ARCH_SUPPORTS_FTRACE_OPS is not
         * set (see include/asm/ftrace.h and include/linux/ftrace.h).  Only the
         * ip and parent ip are used and the list function is called when
         * function tracing is enabled.
         */
        call   *ftrace_trace_function ; 调用 ftrace_trace_function 指向的 C 实现的函数

        restore_mcount_regs ; 另一个汇编宏,恢复一些寄存器的值

        jmp fgraph_trace ; 跳转到标签 fgraph_trace 处,并执行至 retq 返回调用者
END(function_hook)

开启CONFIG_DYNAMIC_FTRACE

  • 最初mcount的作用是直接从函数返回。
  • arch/x86/kernel/mcount_64.S
ENTRY(function_hook)
        retq
END(function_hook)

修改文本段

current_tracer

static const struct file_operations set_tracer_fops = {
       .open           = tracing_open_generic,
       .read           = tracing_set_trace_read,
       .write          = tracing_set_trace_write,
       .llseek         = generic_file_llseek,
};
fs_initcall(tracer_init_tracefs)
-> tracer_init_tracefs()
   -> trace_create_file("current_tracer", 0644, d_tracer, tr, &set_tracer_fops);

修改 current_tracer

static struct tracer function_trace __tracer_data =
{
        .name           = "function",
        .init           = function_trace_init,
        .reset          = function_trace_reset,
        .start          = function_trace_start,
        .flags          = &func_flags,
        .set_flag       = func_set_flag,
        .allow_instances = true,
#ifdef CONFIG_FTRACE_SELFTEST
        .selftest       = trace_selftest_startup_function,
#endif
};
tracing_set_trace_write()
-> tracing_set_tracer()
   -> tracer_init()
      -> tracing_reset_online_cpus()
      -> t->init(tr)
      => function_trace_init()
         -> tracing_start_cmdline_record()
         -> tracing_start_function_trace()
         -> register_ftrace_function()
            -> ftrace_startup()
               -> ftrace_startup_enable()
                  -> ftrace_run_update_code()
                     -> ftrace_arch_code_modify_prepare()
                        -> set_kernel_text_rw()
                        -> set_all_modules_text_rw()
                        arch/x86/kernel/ftrace.c
                     -> arch_ftrace_update_code()
                        -> ftrace_modify_all_code()
                     -> ftrace_arch_code_modify_post_process()
                        -> set_all_modules_text_ro()
                        -> set_kernel_text_ro()

function_graph tracer的实现

foo()
  |
  +-> bar()
        |
        +-> mcount()
        |     |
        |     +-> ftrace_trace_function() /*function hook*/
        |     |
        |     +-> if (ftrace_graph_return != ftrace_stub ||
        |             ftrace_graph_entry != ftrace_graph_entry_stub)
        |             |
        |             +-> ftrace_graph_caller()
        |                   |
        |                   +-> [ Save the return address of traced function ]
        |                   +-> prepare_ftrace_return()
        |                         |
        |                         +-> [ hijack the return address ]
        |                         +-> ftrace_graph_entry() /*function_graph hook 1*/
        |                         +-> ftrace_push_return_trace()
        |
        +-> [ bar() function body ]
        |   /*bar() will return to the hijack address, but not foo()*/
        +-> return_to_handler()
              |
              +-> ftrace_return_to_handler() /*return the original address*/
              |     +-> ftrace_pop_return_trace()
              |     +-> ftrace_graph_return() /*function_graph hook 2*/
              |   /*jump to the original call site*/
              +-> jmp original_return_point
                         |
  +<---------------------+
  | /*back to original flow*/
  v
;
  • function_graph 跟踪器要复杂一些,在 function 跟踪器调用完 ftrace_trace_function 指向的函数后执行,即jmp fgraph_trace
  • 这里会比较ftrace_graph_returnftrace_stubftrace_graph_entryftrace_graph_entry_stub
  • 上面两个比较一旦有不相等,即ftrace_graph_returnftrace_graph_entry被设置,都会跳到ftrace_graph_caller
    • ftrace_graph_entry是 function_graph 跟踪器的第一个 hook 点,在被调用函数体执行前被调用
    • ftrace_graph_return是 function_graph 跟踪器的第二个 hook 点,在被调用函数返回前会被调用
  • ftrace_stub之前列过了,通常为体系结构相关的汇编实现,而ftrace_graph_entry_stub 是通用的 C 实现
  • kernel/trace/ftrace.c
     int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
     {
             return 0;
     }

ftrace_graph_caller

  • 接下来看 ftrace_graph_caller
  • arch/x86/kernel/mcount_64.S
ENTRY(ftrace_graph_caller)
        /* Saves rbp into %rdx and fills first parameter  */
        save_mcount_regs

#ifdef CC_USING_FENTRY
        leaq MCOUNT_REG_SIZE+8(%rsp), %rsi
        movq $0, %rdx   /* No framepointers needed */
#else
        /* Save address of the return address of traced function */
        leaq 8(%rdx), %rsi     ;保存被跟踪的函数的返回地址
        /* ftrace does sanity checks against frame pointers */
        movq (%rdx), %rdx
#endif
        call    prepare_ftrace_return

        restore_mcount_regs

        retq
END(ftrace_graph_caller)
  • ftrace_trace_function指针稍有不同,传给prepare_ftrace_return()函数的parent_ip会是一个在栈上的指针,指向(foo()调用bar())调用点的地址。
  • 这么做的目的是让函数可以临时劫持返回地址,让它指向体系结构相关的return_to_handler函数。
    • 比如foo()调用bar(),那么劫持之后bar()函数运行完后不是返回到foo(),而是返回到return_to_handler
    • 如果像ftrace_trace_function那样,将调用点(foo()代码段上)的地址传给prepare_ftrace_return()函数,是无法修改的。
    • 如果传的是栈上某个指针的地址,那么prepare_ftrace_return()就可以通过该地址修改指针的指向,让它指向return_to_handler,这样就 劫持 了被调用函数的返回。

prepare_ftrace_return()

  • prepare_ftrace_return()是 function_graph 跟踪器需要实现的一个和体系结构相关的函数,要做的事情主要有
    • 将栈上指针原来保存的被调用者的返回地址替换成return_to_handler的地址,即劫持
    • 调用 function_graph 的 hook 函数 ftrace_graph_entry()
    • 调用 function_graph 的体系结构无关的 ftrace_push_return_trace()函数
  • arch/x86/kernel/ftrace.c
void prepare_ftrace_return(unsigned long self_addr, unsigned long *parent,
                           unsigned long frame_pointer)
{
        unsigned long old;
        int faulted;
        struct ftrace_graph_ent trace;
        unsigned long return_hooker = (unsigned long)
                                &return_to_handler;
        ...
        /*parent是栈上指针的地址,而不是指针的内容(虽然内容也是地址)。
          该指针的指向是被调用者的返回地址,应为代码段的某个地址,即原返回地址。*/
        asm volatile(
                "1: " _ASM_MOV " (%[parent]), %[old]\n" /*将原返回地址保存到old*/
                "2: " _ASM_MOV " %[return_hooker], (%[parent])\n" /*指针指向return_to_handler*/
                "   movl $0, %[faulted]\n"
                "3:\n"

                ".section .fixup, \"ax\"\n"
                "4: movl $1, %[faulted]\n"
                "   jmp 3b\n"
                ".previous\n"

                _ASM_EXTABLE(1b, 4b)
                _ASM_EXTABLE(2b, 4b)

                : [old] "=&r" (old), [faulted] "=r" (faulted)
                : [parent] "r" (parent), [return_hooker] "r" (return_hooker)
                : "memory"
        );
        if (unlikely(faulted)) {
                ftrace_graph_stop(); /*如果以上过程出现问题,停止ftrace_graph跟踪*/
                WARN_ON(1);
                return;
        }
        /*保存被调用者地址和当前调用栈的深度到 ftrace_graph 跟踪条目 trace 变量*/
        trace.func = self_addr;
        trace.depth = current->curr_ret_stack + 1;

        /* Only trace if the calling function expects to */
        if (!ftrace_graph_entry(&trace)) { /*ftrace_graph hook 点 1*/
                *parent = old; /*如果出错了,还原指针还需指回原返回地址*/
                return;
        }
        /*由ftrace_graph实现的体系结构无关函数*/
        if (ftrace_push_return_trace(old, self_addr, &trace.depth,
                                     frame_pointer, parent) == -EBUSY) {
                *parent = old; /*如果出错了,还原指针还需指回原返回地址*/
                return;
        }
}
..._```

return_to_handler

  • return_to_handler函数简单地调用通用的 C 函数ftrace_return_to_handler()返回最初的返回地址,这样我们就可以返回最初的调用地点了。
    • ftrace劫持完返回地址,完成在函数返回前所要做的事情后,最终还是要返回到foo()调用bar()的地方的,return_to_handler就是做这个事情的。
    • ftrace_return_to_handler()是 function_graph 实现的一个体系架构无关的通用函数,最主要的作用是要返回被调用函数(如bar())的原返回地址。
  • arch/x86/kernel/mcount_64.S
GLOBAL(return_to_handler)
        subq  $24, %rsp

        /* Save the return values */
        movq %rax, (%rsp)
        movq %rdx, 8(%rsp)
        movq %rbp, %rdi

        call ftrace_return_to_handler

        movq %rax, %rdi     ;函数返回值保存在%rax,把它放到%rdi
        movq 8(%rsp), %rdx
        movq (%rsp), %rax
        addq $24, %rsp
        jmp *%rdi           ;跳到原返回地址

irqsoff/preemptoff/preemptirqsoff tracer

  • Irqsoff tracer 当中断被禁止时,系统无法相应外部事件,比如键盘和鼠标,时钟也无法产生 tick 中断。这意味着系统响应延迟,irqsoff 这个 tracer 能够跟踪并记录内核中哪些函数禁止了中断,对于其中中断禁止时间最长的,irqsoff 将在 log 文件的第一行标示出来,从而使开发人员可以迅速定位造成响应延迟的罪魁祸首。
  • Preemptoff tracer 和前一个 tracer 类似,preemptoff tracer 跟踪并记录 禁止内核抢且关闭中断占期间 的函数,并清晰地显示出禁止抢占时间最长的内核函数。
  • Preemptirqsoff tracer 同上,跟踪和记录禁止中断或者禁止抢占的内核函数,以及禁止时间最长的函数。

preemptoffirqsoff跟踪器

  • preemptoffirqsoff跟踪器用的跟踪函数是相同的,都是irqsoff_tracer_call()
  • preemptoffirqsoff跟踪器的不同之处
    • irqsoff跟踪器的 start 点在开启或关闭中断的地方,如 local_irq_disable()
    • preemptoff跟踪器的 start 点在开启或关闭抢占的地方,如 preempt_disable()
    • 老版本的preemptoff跟踪器不会记录 中断开启期间的函数,尽管抢占计数不为 0;目前版本(4.9)的preemptoff跟踪器不会记录 中断开启且抢占计数为 0 时的函数(好像是废话)。
  • 注意:/sys/kernel/debug/tracing/events/preemptirq/preempt_{enable|disable}的跟踪事件的调用点在trace_preempt_off()->trace_preempt_disable_rcuidle(),结合stacktrace选项可以回溯调用路径

preemptoff跟踪器的开启和关闭

  • 跟踪器触发时函数调用简化流程
include/linux/preempt.h
preempt_disable()
  -> preempt_count_inc() => preempt_count_add(1)
     -> preempt_latency_start()
          kernel/trace/trace_irqsoff.c
          -> trace_preempt_off()
             -> trace_preempt_disable_rcuidle()
                -> trace_preempt_disable()
             -> tracer_preempt_off()
                -> start_critical_timing()
                   -> __trace_function()
// engine:dot
digraph G {
	rankdir="LR";
	node [shape="record"];
	preempt_disable[label="<f0> include/linux/preempt.h | <f1> preempt_disable"];
	preempt_count_add[label="preempt_count_add(1)"];
  preempt_latency_start[label="<f0> kernel/trace/trace_irqsoff.c | <f1> preempt_latency_start"];
	trace_preempt_off;
	tracer_preempt_off;
	start_critical_timing;
	__trace_function;

	preempt_disable:f1 -> preempt_count_inc -> preempt_count_add;
	preempt_count_inc -> preempt_latency_start:f1;

	subgraph cluster1 {
		node[shape="point", width=0, height=0];
		{
			edge[arrowhead="none", penwidth=0];
		  preempt_latency_start:f1 -> s01 -> s02 -> s03;
		}
		{
			edge[arrowhead="none", penwidth=0];
		  s10 -> trace_preempt_off -> s12 -> s13;
		}
		{
			edge[arrowhead="none", penwidth=0];
			s20 -> s21 -> start_critical_timing -> s23;
		}
		{
			edge[arrowhead="none", penwidth=0];
			s30 -> s31 -> s32 -> __trace_function;
		}
		preempt_latency_start:f1 -> trace_preempt_off -> tracer_preempt_off -> start_critical_timing -> __trace_function;
	}
}
  • kernel/sched/core.c
/*
 * If the value passed in is equal to the current preempt count
 * then we just disabled preemption. Start timing the latency.
 */
static inline void preempt_latency_start(int val)
{
        /*仅在抢占计数与要改变的递增值相等时才调用函数去动态改变插桩,不相等时是不需要去改变
          插桩的。我们需要记录的是抢占关闭到抢占开启期间的 trace,在此期间增加禁止抢占的深度
          并不改变不能被抢占的状况。*/
        if (preempt_count() == val) {
                unsigned long ip = get_lock_parent_ip();
...
                trace_preempt_off(CALLER_ADDR0, ip);
        }
}

void preempt_count_add(int val)
{
...
        __preempt_count_add(val);
...
        preempt_latency_start(val);
}
EXPORT_SYMBOL(preempt_count_add);
NOKPROBE_SYMBOL(preempt_count_add);

/*
 * If the value passed in equals to the current preempt count
 * then we just enabled preemption. Stop timing the latency.
 */
static inline void preempt_latency_stop(int val)
{
        /*仅在抢占计数与要改变的递减值相等时才调用函数去动态改变插桩,不相等时是不需要去改变
          插桩的。我们需要记录的是抢占关闭到抢占开启期间的 trace,在此期间减少禁止抢占的深度
          并不改变不能被抢占的状况。*/
        if (preempt_count() == val)
                trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
}

void preempt_count_sub(int val)
{
...
        preempt_latency_stop(val);
        __preempt_count_sub(val);
}
EXPORT_SYMBOL(preempt_count_sub);
NOKPROBE_SYMBOL(preempt_count_sub);
...
  • kernel/trace/trace_irqsoff.c
static DEFINE_PER_CPU(int, tracing_cpu);
...
static inline int
preempt_trace(void)
{       /*preemptoff跟踪器开启且抢占计数不为 0*/
        return ((trace_type & TRACER_PREEMPT_OFF) && preempt_count());
}
...
static inline int
irq_trace(void)
{       /*irqsoff跟踪器开启且在中断上下文*/
        return ((trace_type & TRACER_IRQS_OFF) &&
                irqs_disabled());
}
...
static inline void
start_critical_timing(unsigned long ip, unsigned long parent_ip)
{
        int cpu;
        struct trace_array *tr = irqsoff_trace;
        struct trace_array_cpu *data;
        unsigned long flags;
        /*如果当前跟踪器未使能或跟踪功能没开启,直接返回*/
        if (!tracer_enabled || !tracing_is_enabled())
                return;

        cpu = raw_smp_processor_id();
        /*tracing_cpu是个 Per-CPU 变量,它仅在该函数下面置 1,也就是说该函数下面的部分
          只在开启跟踪的时候运行一次。*/
        if (per_cpu(tracing_cpu, cpu))
                return;

        data = per_cpu_ptr(tr->trace_buffer.data, cpu);
        /*data->disabled 控制是否同时 trace_function*/
        if (unlikely(!data) || atomic_read(&data->disabled))
                return;
        /*不允许伪并发的 trace_function,注意,这是 Per-CPU 的*/
        atomic_inc(&data->disabled);

        data->critical_sequence = max_sequence;
        data->preempt_timestamp = ftrace_now(cpu);
        data->critical_start = parent_ip ? : ip;
        /*将 flags 寄存器的信息取出来*/
        local_save_flags(flags);
        /*写当前跟踪信息*/
        __trace_function(tr, ip, parent_ip, flags, preempt_count());
        /*此处将 tracing_cpu 置 1,不会因为伪并发而被运行两次start*/
        per_cpu(tracing_cpu, cpu) = 1;
        /*写完信息了,放开保护允许继续 trace*/
        atomic_dec(&data->disabled);
}
...
#ifdef CONFIG_TRACE_PREEMPT_TOGGLE

void trace_preempt_on(unsigned long a0, unsigned long a1)
{
    if (!in_nmi())
        trace_preempt_enable_rcuidle(a0, a1);
    tracer_preempt_on(a0, a1);
}

void trace_preempt_off(unsigned long a0, unsigned long a1)
{
    if (!in_nmi())
        trace_preempt_disable_rcuidle(a0, a1);
    tracer_preempt_off(a0, a1);
}
#endif
  • include/linux/ftrace.h
/* Archs may use other ways for ADDR1 and beyond */
#ifndef ftrace_return_address
# ifdef CONFIG_FRAME_POINTER
#  define ftrace_return_address(n) __builtin_return_address(n)
# else
#  define ftrace_return_address(n) 0UL
# endif
#endif

#define CALLER_ADDR0 ((unsigned long)ftrace_return_address0)
#define CALLER_ADDR1 ((unsigned long)ftrace_return_address(1))
#define CALLER_ADDR2 ((unsigned long)ftrace_return_address(2))
#define CALLER_ADDR3 ((unsigned long)ftrace_return_address(3))
#define CALLER_ADDR4 ((unsigned long)ftrace_return_address(4))
#define CALLER_ADDR5 ((unsigned long)ftrace_return_address(5))
#define CALLER_ADDR6 ((unsigned long)ftrace_return_address(6))

static inline unsigned long get_lock_parent_ip(void)
{
        unsigned long addr = CALLER_ADDR0;

        if (!in_lock_functions(addr)) /*如果地址不在锁函数的.text段范围内*/
                return addr;          /*返回该地址*/
        addr = CALLER_ADDR1;          /*否则向上走一层*/
        if (!in_lock_functions(addr))
                return addr;
        return CALLER_ADDR2;
}
...__```

irqsoff跟踪器的开启和关闭

  • 还是之前提到的,与preemptoff最大的不同之处就是起始点:
  • include/linux/irqflags.h
#define local_irq_enable() \
        do { trace_hardirqs_on(); raw_local_irq_enable(); } while (0)
#define local_irq_disable() \
        do { raw_local_irq_disable(); trace_hardirqs_off(); } while (0)
  • kernel/trace/trace_irqsoff.c
/*
 * We are only interested in hardirq on/off events:
 */
void trace_hardirqs_on(void)
{
        if (!preempt_trace() && irq_trace())
                stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
}
EXPORT_SYMBOL(trace_hardirqs_on);

void trace_hardirqs_off(void)
{
        if (!preempt_trace() && irq_trace())
                start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
}
EXPORT_SYMBOL(trace_hardirqs_off);

主框架

初始化

init/main.c
start_kernel()
  |   kernel/trace/trace.c
  +-> trace_init()
  |     -> tracer_alloc_buffers()
  |     -> trace_event_init()
  |   kernel/trace/ftrace.c
  +-> ftrace_init()
  |     |
  |     +-> ftrace_process_locs()
  |     |    -> ftrace_allocate_pages()
  |     |    -> ftrace_update_code()
  |     +-> set_ftrace_early_filters()
  +-> rest_init()
  • 关注两个全局变量__stop_mcount_loc__start_mcount_loc,以下是我的猜测:
    • 由编译器脚本创建并放到特定的 section 里
    • 系统启动时创建对应的全局变量
    • 内容为由-pg选项加上mcount调用的所有函数地址
    • 随着模块的增加会动态调整
    • include/asm-generic/vmlinux.lds.h
    • arch/x86/kernel/vmlinux.lds.S
  • 由于 ftrace plugins 的初始化函数多是用 initcalls 的宏进行声明,因此 plugins 的初始化在内核启动过程中是比较靠后的(见 initcalls 章节)
  • 通过 initcalls 把 tracer 提供的初始化函数放到 initcalls 所属的 section,这个初始化函数最重要的事情是调用register_tracer(tracer),注册struct tracer类型的 tracer plugin

tracefs的初始化

  • tracefs文件系统是借助fs_initcall()的方式来初始化的
  • 为保证向后兼容那些挂载debugfs来获得跟踪功能的工具,tracefs自动挂载到debugfs/tracing目录下,即/sys/kernel/debug/tracing目录
fs_initcall(tracer_init_tracefs)
  +-----------+
  +-> trace_access_lock_init()
  +-> tracing_init_dentry()
  |     -> debugfs_create_automount("tracing", NULL, trace_automount, NULL)
  +-> init_tracer_tracefs()
  |     +-> trace_create_file("available_tracers", ...)
  |     +-> trace_create_file("current_tracer", ...)
  |     +-> trace_create_file("trace_options", ...)
  |     +-> trace_create_file("trace", ...)
  |     +-> trace_create_file("trace_pipe", ...)
  |     +-> trace_create_file("tracing_on", ...)
  |     +-> trace_create_file ...
  |     +-> create_trace_options_dir()
  |     |    -> trace_options_init_dentry()
  |     |          -> tracefs_create_dir("options", ...)
  |     |    -> for (i = 0; trace_options[i]; i++) {
  |     |         ...
  |     |         create_trace_option_core_file(tr, trace_options[i], i)
  |     |       }
  |     |   kernel/trace/trace_functions.c
  |     +-> ftrace_create_function_files()
  |     |     -> allocate_ftrace_ops()
  |     |        kernel/trace/ftrace.c
  |     |     -> ftrace_create_filter_files()
  |     |          -> trace_create_file("set_ftrace_filter", ...)
  |     |          -> trace_create_file("set_ftrace_notrace", ...)
  |     +-> for_each_tracing_cpu(cpu)
  |     |      tracing_init_tracefs_percpu(tr, cpu);
  |     +-> ftrace_init_tracefs()
  |           -> trace_create_file("set_ftrace_pid", ...)
  +-> ftrace_init_tracefs_toplevel()
  +-> trace_create_file("tracing_thresh", ...)
  +-> trace_create_file("README", ...)
  +-> trace_create_file ...
  +-> trace_enum_init()
  +-> trace_create_enum_file()
  +-> create_trace_instances()
  +-> update_tracer_options(&global_trace)
  • 通过debugfs提供的fs/debugfs/inode.c::debugfs_create_automount()函数,在debugfs文件系统创建自动挂载点
    • 传入回调函数trace_automount()会在路径名解析的时候被调用,该函数将完成挂载tracefs这个动作
  • 通过create_trace_options_dir()会在初始化期间把tracing/options目录创建好,并根据trace_options[]数组在tracing/options目录下创建预定的 options 文件

TRACE_FLAGS

  • 通过不同的宏定义C(a, b)将同一个TRACE_FLAGS宏定义分别转成:
    • enum trace_iterator_bits 枚举 位名称 列表
    • enum trace_iterator_flags 枚举 位掩码 列表
    • const char *trace_options[] 给每个位提供对应的字符串定义,并且顺序与枚举位列表中的保持一致(因为是同一个TRACE_FLAGS转出来的)
  • kernel/trace/trace.h
     /*
      * trace_iterator_flags is an enumeration that defines bit
      * positions into trace_flags that controls the output.
      *
      * NOTE: These bits must match the trace_options array in
      *       trace.c (this macro guarantees it).
      */
     #define TRACE_FLAGS                                             \
                     C(PRINT_PARENT,         "print-parent"),        \
                     C(SYM_OFFSET,           "sym-offset"),          \
                     C(SYM_ADDR,             "sym-addr"),            \
                     C(VERBOSE,              "verbose"),             \
                     C(RAW,                  "raw"),                 \
                     C(HEX,                  "hex"),                 \
                     C(BIN,                  "bin"),                 \
                     C(BLOCK,                "block"),               \
                     C(PRINTK,               "trace_printk"),        \
                     C(ANNOTATE,             "annotate"),            \
                     C(USERSTACKTRACE,       "userstacktrace"),      \
                     C(SYM_USEROBJ,          "sym-userobj"),         \
                     C(PRINTK_MSGONLY,       "printk-msg-only"),     \
                     C(CONTEXT_INFO,         "context-info"),   /* Print pid/cpu/time */ \
                     C(LATENCY_FMT,          "latency-format"),      \
                     C(RECORD_CMD,           "record-cmd"),          \
                     C(OVERWRITE,            "overwrite"),           \
                     C(STOP_ON_FREE,         "disable_on_free"),     \
                     C(IRQ_INFO,             "irq-info"),            \
                     C(MARKERS,              "markers"),             \
                     C(EVENT_FORK,           "event-fork"),          \
                     FUNCTION_FLAGS                                  \
                     FGRAPH_FLAGS                                    \
                     STACK_FLAGS                                     \
                     BRANCH_FLAGS
    
     /*
      * By defining C, we can make TRACE_FLAGS a list of bit names
      * that will define the bits for the flag masks.
      */
     #undef C
     #define C(a, b) TRACE_ITER_##a##_BIT
    
     enum trace_iterator_bits {
             TRACE_FLAGS
             /* Make sure we don't go more than we have bits for */
             TRACE_ITER_LAST_BIT
     };
    
     /*
      * By redefining C, we can make TRACE_FLAGS a list of masks that
      * use the bits as defined above.
      */
     #undef C
     #define C(a, b) TRACE_ITER_##a = (1 << TRACE_ITER_##a##_BIT)
    
     enum trace_iterator_flags { TRACE_FLAGS };
  • kernel/trace/trace.c
     /*
      * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
      * It uses C(a, b) where 'a' is the enum name and 'b' is the string that
      * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
      * of strings in the order that the enums were defined.
      */
     #undef C
     #define C(a, b) b
    
     /* These must match the bit postions in trace_iterator_flags */
     static const char *trace_options[] = {
             TRACE_FLAGS
             NULL
     };
     ..._
  • 例如,对于TRACE_FLAGS C(VERBOSE, "verbose")会有TRACE_ITER_VERBOSE_BITTRACE_ITER_VERBOSE两个枚举成员和一个/sys/kernel/debug/tracing/option/verbose文件被创建出来

struct tracer结构

  • struct tracer结构 用于定义一个跟踪器以及它与 tracefs 交互的回调函数
  • kernel/trace/trace.h
     /**
      * struct tracer - a specific tracer and its callbacks to interact with tracefs
      * @name: the name chosen to select it on the available_tracers file
      * @init: called when one switches to this tracer (echo name > current_tracer)
      * @reset: called when one switches to another tracer
      * @start: called when tracing is unpaused (echo 1 > tracing_on)
      * @stop: called when tracing is paused (echo 0 > tracing_on)
      * @update_thresh: called when tracing_thresh is updated
      * @open: called when the trace file is opened
      * @pipe_open: called when the trace_pipe file is opened
      * @close: called when the trace file is released
      * @pipe_close: called when the trace_pipe file is released
      * @read: override the default read callback on trace_pipe
      * @splice_read: override the default splice_read callback on trace_pipe
      * @selftest: selftest to run on boot (see trace_selftest.c)
      * @print_headers: override the first lines that describe your columns
      * @print_line: callback that prints a trace
      * @set_flag: signals one of your private flags changed (trace_options file)
      * @flags: your private flags
      */
     struct tracer {
             const char              *name;
             int                     (*init)(struct trace_array *tr);
             void                    (*reset)(struct trace_array *tr);
             void                    (*start)(struct trace_array *tr);
             void                    (*stop)(struct trace_array *tr);
             int                     (*update_thresh)(struct trace_array *tr);
             void                    (*open)(struct trace_iterator *iter);
             void                    (*pipe_open)(struct trace_iterator *iter);
             void                    (*close)(struct trace_iterator *iter);
             void                    (*pipe_close)(struct trace_iterator *iter);
             ssize_t                 (*read)(struct trace_iterator *iter,
                                             struct file *filp, char __user *ubuf,
                                             size_t cnt, loff_t *ppos);
             ssize_t                 (*splice_read)(struct trace_iterator *iter,
                                                    struct file *filp,
                                                    loff_t *ppos,
                                                    struct pipe_inode_info *pipe,
                                                    size_t len,
                                                    unsigned int flags);
     #ifdef CONFIG_FTRACE_STARTUP_TEST
             int                     (*selftest)(struct tracer *trace,
                                                 struct trace_array *tr);
     #endif
             void                    (*print_header)(struct seq_file *m);
             enum print_line_t       (*print_line)(struct trace_iterator *iter);
             /* If you handled the flag setting, return 0 */
             int                     (*set_flag)(struct trace_array *tr,
                                                 u32 old_flags, u32 bit, int set);
             /* Return 0 if OK with change, else return non-zero */
             int                     (*flag_changed)(struct trace_array *tr,
                                                     u32 mask, int set);
             struct tracer           *next;
             struct tracer_flags     *flags;
             int                     enabled;
             int                     ref;
             bool                    print_max;
             bool                    allow_instances;
     #ifdef CONFIG_TRACER_MAX_TRACE
             bool                    use_max_tr;
     #endif
     };
     ...*```
  • name 用于在available_tracers中显示的跟踪器名字,选择跟踪器时用的也是这个名字
  • flags 跟踪器的私有 flags,用于设定特定于某个跟踪器的选项的集合,定义如下
  • kernel/trace/trace.h
     /*
      * An option specific to a tracer. This is a boolean value.
      * The bit is the bit index that sets its value on the
      * flags value in struct tracer_flags.
      */
     struct tracer_opt {
             const char      *name; /* Will appear on the trace_options file */
             u32             bit; /* Mask assigned in val field in tracer_flags */
     };
    
     /*
      * The set of specific options for a tracer. Your tracer
      * have to set the initial value of the flags val.
      */
     struct tracer_flags {
             u32                     val;
             struct tracer_opt       *opts;
             struct tracer           *trace;
     };
    
     /* Makes more easy to define a tracer opt */
     #define TRACER_OPT(s, b)        .name = #s, .bit = b
     ...*```
  • val tracer 私有 flags 的初始值
  • opts 指向 tracer 私有的 options
  • trace 回指向所属的 tracer plugin 实例

tracer 的方法

方法名称 作用
init 切换到这个 tracer 时调用 (echo name > current_tracer)
reset 换到其他 tracer 时调用
start 跟踪开启时调用 (echo 1 > tracing_on)
stop 跟踪暂停时调用 (echo 0 > tracing_on)
update_thresh tracing_thresh 文件更新时调用,当延迟大于该文件指定的 微秒,延迟跟踪器记录一条trace
open trace 文件打开时调用
pipe_open trace_pipe 文件被打开时调用
close trace 文件释放时调用
pipe_close trace_pipe 被释放时调用
read 让 tracer 可以覆写 trace_pipe 缺省的 read 回调
splice_read 让 tracer 可以覆写 trace_pipe 缺省的 splice_read 回调
selftest 在启动时运行 selftest (见 trace_selftest.c)
print_headers 覆写首行的列描述
print_line 打印一条 trace 的回调
set_flag 当 tracer 的私有 flags 发生改变时调用 (trace_options 文件)
flag_changed 当通用的 flag 发生改变时调用,让 tracer 有机会决定是否接受某个 option 的改变

trace_options

  • 例如一个 function_graph 跟踪器的例子
  • kernel/trace/trace_functions_graph.c
     static struct tracer_opt trace_opts[] = {
             /* Display overruns? (for self-debug purpose) */
             { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
             /* Display CPU ? */
             { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
             /* Display Overhead ? */
             { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
             /* Display proc name/pid */
             { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
             /* Display duration of execution */
             { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
             /* Display absolute time of an entry */
             { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
             /* Display interrupts */
             { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
             /* Display function name after trailing } */
             { TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
             /* Include sleep time (scheduled out) between entry and return */
             { TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
             /* Include time within nested functions */
             { TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
             { } /* Empty entry */
     };
    
     static struct tracer_flags tracer_flags = {
             /* Don't display overruns, proc, or tail by default */
             .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
                    TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
                    TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
             .opts = trace_opts
     };
  • struct tracer_opt trace_opts[]提供了一个跟踪器选项集合
  • struct tracer_flags tracer_flagsval 则是 function_graph 的初始选项,通过 opts 把变量tracer_flags能支持的选项与trace_opts[]集合联系起来
  • kernel/trace/trace.c
     static const struct file_operations tracing_iter_fops = {
             .open           = tracing_trace_options_open,
             .read           = seq_read,
             .llseek         = seq_lseek,
             .release        = tracing_single_release_tr,
             .write          = tracing_trace_options_write,
     };
     ...
     static void
     init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
     {
         ...
         /*/sys/kernel/debug/tracing/trace_options*/
         trace_create_file("trace_options", 0644, d_tracer,
             tr, &tracing_iter_fops);
         ...
     };
  • 打开/sys/kernel/debug/tracing/trace_options文件时会通过tracing_trace_options_open()设置seq_read时的回调为tracing_trace_options_show()
  • /sys/kernel/debug/tracing/trace_options文件时最终会调用 tracer 的set_flag
    tracing_trace_options_write()
      -> trace_set_options()
            |
            +-> set_tracer_flag()
            |     -> tr->current_trace->flag_changed(tr, mask, !!enabled)
            +-> set_tracer_option()
                  -> set_tracer_option()
                        -> __set_tracer_option()
                            -> trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
  • kernel/trace/trace.c
     /* These must match the bit postions in trace_iterator_flags */
     static const char *trace_options[] = {
             TRACE_FLAGS /*定义在 kernel/trace/trace.h 中的一列由 option 名字组成的字符串*/
             NULL
     };
    
     static int trace_set_options(struct trace_array *tr, char *option)
     {
             char *cmp;
             int neg = 0;
             int ret = -ENODEV;
             int i;
             size_t orig_len = strlen(option);
             /*去除多余的空格*/
             cmp = strstrip(option);
             /*关闭某个 option 只需在其之前加上“no”*/
             if (strncmp(cmp, "no", 2) == 0) {
                     neg = 1;
                     cmp += 2;
             }
    
             mutex_lock(&trace_types_lock);
             /*设置 tracer 通用的 option,设置完一个 option 后结束循环*/
             for (i = 0; trace_options[i]; i++) {
                     if (strcmp(cmp, trace_options[i]) == 0) {
                             ret = set_tracer_flag(tr, 1 << i, !neg);
                             break;
                     }
             }
             /*如果设置的不是通用 option,测试设置的是不是 tracer 私有的 options*/
             /* If no option could be set, test the specific tracer options */
             if (!trace_options[i])
                     ret = set_tracer_option(tr, cmp, neg);
    
             mutex_unlock(&trace_types_lock);
    
             /*
              * If the first trailing whitespace is replaced with '\0' by strstrip,
              * turn it back into a space.
              */
             if (orig_len > strlen(option))
                     option[strlen(option)] = ' ';
    
             return ret;
     }

注册跟踪器

  • 全局变量trace_types指向所有可用 tracer 组成的链表
  • trace_types_lock是保护trace_types链表的锁
  • kernel/trace/trace.c
/* trace_types holds a link list of available tracers. */
static struct tracer            *trace_types __read_mostly;

/*
 * trace_types_lock is used to protect the trace_types list.
 */
DEFINE_MUTEX(trace_types_lock);
...
#define MAX_TRACER_SIZE         100
static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
static char *default_bootup_tracer;
...
static int __init set_cmdline_ftrace(char *str)
{
        strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
        default_bootup_tracer = bootup_tracer_buf;
        /* We are using ftrace early, expand it */
        ring_buffer_expanded = true;
        return 1;
}
/*可以通过内核命令行定制缺省启动 tracer*/
__setup("ftrace=", set_cmdline_ftrace);
...
/*
 * The global_trace is the descriptor that holds the top-level tracing
 * buffers for the live tracing.
 */
static struct trace_array global_trace = {
        .trace_flags = TRACE_DEFAULT_FLAGS,
};
...
static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;

static int __init set_trace_boot_options(char *str)
{
        strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
        return 0;
}
/*可以通过内核命令行定制缺省启动 tracer 的option*/
__setup("trace_options=", set_trace_boot_options);
...
static void
create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
{
        struct trace_option_dentry *topts;
        struct trace_options *tr_topts;
        struct tracer_flags *flags;
        struct tracer_opt *opts;
        int cnt;
        int i;

        if (!tracer)
                return;
        /*跟踪器的私有 flags*/
        flags = tracer->flags;

        if (!flags || !flags->opts)
                return;

        /*
         * If this is an instance, only create flags for tracers
         * the instance may have.
         */
        if (!trace_ok_for_array(tracer, tr))
                return;
        /*新的 flags 与原有的 flags 不能重复,比较 struct trace_flags类型的实例地址*/
        for (i = 0; i < tr->nr_topts; i++) {
                /* Make sure there's no duplicate flags. */
                if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
                        return;
        }
        /*该 struct trace_flags 类型实例包含的 options*/
        opts = flags->opts;
        /*计算新 options 的数目*/
        for (cnt = 0; opts[cnt].name; cnt++)
                ;
        /*给新 options 分配空间*/
        topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
        if (!topts)
                return;
        /*原 options 数组扩容*/
        tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
                            GFP_KERNEL);
        if (!tr_topts) {
                kfree(topts);
                return;
        }
        /*指向扩容后的 options 数组地址*/
        tr->topts = tr_topts;
        tr->topts[tr->nr_topts].tracer = tracer; /*放入新 options 元素*/
        tr->topts[tr->nr_topts].topts = topts;
        tr->nr_topts++; /*options 数组计数增加*/
        /*在 /sys/kernel/debug/tracing/options 下创建新的 option 文件*/
        for (cnt = 0; opts[cnt].name; cnt++) {
                create_trace_option_file(tr, &topts[cnt], flags,
                                         &opts[cnt]);
                WARN_ONCE(topts[cnt].entry == NULL,
                          "Failed to create trace option: %s",
                          opts[cnt].name);
        }
}
...
static void add_tracer_options(struct trace_array *tr, struct tracer *t)
{
        /* Only enable if the directory has been created already. */
        if (!tr->dir)
                return;

        create_trace_option_files(tr, t);
}
...
/**
 * register_tracer - register a tracer with the ftrace system.
 * @type - the plugin for the tracer
 *
 * Register a new plugin tracer.
 */
int __init register_tracer(struct tracer *type)
{
        struct tracer *t;
        int ret = 0;
        /*tracer 必须得有名字*/
        if (!type->name) {
                pr_info("Tracer must have a name\n");
                return -1;
        }
        /*tracer 名字最大长度是 100*/
        if (strlen(type->name) >= MAX_TRACER_SIZE) {
                pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
                return -1;
        }

        mutex_lock(&trace_types_lock);

        tracing_selftest_running = true;
        /*新 tracer 名字不能和已有的重名*/
        for (t = trace_types; t; t = t->next) {
                if (strcmp(type->name, t->name) == 0) {
                        /* already found */
                        pr_info("Tracer %s already registered\n",
                                type->name);
                        ret = -1;
                        goto out;
                }
        }
        /*新 tracer 没有定义自己的 set_flag 回调则给它一个空的 dummy_set_flag() 函数*/
        if (!type->set_flag)
                type->set_flag = &dummy_set_flag;
        /*新 tracer 没有定义自己的私有 flag 则给它一个空的 tracer_flags 实例*/
        if (!type->flags) {
                /*allocate a dummy tracer_flags*/
                type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
                if (!type->flags) {
                        ret = -ENOMEM;
                        goto out;
                }
                type->flags->val = 0;
                type->flags->opts = dummy_tracer_opt;
        } else
                if (!type->flags->opts)
                        type->flags->opts = dummy_tracer_opt;
        /*设置私有 option 时通过这个回指的指针找到提供该私有 flag 的 tracer,并调用它提供
          的 set_flag 回调函数*/
        /* store the tracer for __set_tracer_option */
        type->flags->trace = type;
        /*如果开启了 tracer 自测,会在这里先测试一下*/
        ret = run_tracer_selftest(type);
        if (ret < 0)
                goto out;
        /*前插的方式插入 trace_types 链表*/
        type->next = trace_types;
        trace_types = type;
        add_tracer_options(&global_trace, type);/*新跟踪器的 flags 合并到全局 options*/

 out:
        tracing_selftest_running = false;
        mutex_unlock(&trace_types_lock);
        /*如果之前都没出错,且没通过内核命令行参数设置缺省启动 tracer,则注册过程结束*/
        if (ret || !default_bootup_tracer)
                goto out_unlock;
        /*通过内核命令行参数设置缺省启动 tracer,则看看是不是缺省启动 tracer*/
        if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
                goto out_unlock;

        printk(KERN_INFO "Starting tracer '%s'\n", type->name);
        /* Do we want this tracer to start on bootup? */
        tracing_set_tracer(&global_trace, type->name); /*启动 tracer*/
        default_bootup_tracer = NULL; /*tracer 已启动,该值可以清空*/
        /*应用内核命令行"trace_options="设置的 options*/
        apply_trace_boot_options();

        /* disable other selftests, since this will break it. */
        tracing_selftest_disabled = true;
#ifdef CONFIG_FTRACE_STARTUP_TEST
        printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
               type->name);
#endif

 out_unlock:
        return ret;
}
...__```

function tracer的初始化

  • function tracer 用core_initcall()进行注册
  • kernel/trace/trace_functions.c
static struct tracer function_trace __tracer_data =
{
        .name           = "function",
        .init           = function_trace_init,
        .reset          = function_trace_reset,
        .start          = function_trace_start,
        .flags          = &func_flags,
        .set_flag       = func_set_flag,
        .allow_instances = true,
#ifdef CONFIG_FTRACE_SELFTEST
        .selftest       = trace_selftest_startup_function,
#endif
};
...
static __init int init_function_trace(void)
{
        init_func_cmd_traceon();
        return register_tracer(&function_trace);
}
core_initcall(init_function_trace);

栈跟踪

  • 核心函数是__ftrace_trace_stack()__trace_stack()对其有一个简单的封装

func_stack_trace (function tracer)

kernel/trace/trace_functions.c
function_stack_trace_call()
   kernel/trace/trace.c
-> trace_function()
-> __trace_stack()
   -> __ftrace_trace_stack()
      kernel/stacktrace.c
      -> stack_trace_save()
        stack_trace_consume_fn consume_entry = stack_trace_consume_entry;
        arch/x86/kernel/stacktrace.c
        -> arch_stack_walk(consume_entry, &c, current, NULL)
           -> unwind_start()
           -> unwind_done()
           -> unwind_next_frame()
           -> unwind_get_return_address()
           -> consume_entry() => stack_trace_consume_entry()

stacktrace (tracepoint)

kernel/trace/trace_events_trigger.c
stacktrace_trigger()
   kernel/trace/trace.c
-> trace_dump_stack(STACK_SKIP)
   -> __ftrace_trace_stack()

function-trace (latency tracer)

kernel/trace/trace_irqsoff.c
tracer_hardirqs_on()/tracer_preempt_on()
-> stop_critical_timings()
   -> check_critical_timing()
      -> __trace_function()
      -> __trace_stack()

ftrace_stacks[4] 和 ftrace_stack[1024] 数组

/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
#define FTRACE_KSTACK_NESTING   4

#define FTRACE_KSTACK_ENTRIES   (PAGE_SIZE / FTRACE_KSTACK_NESTING)
/*每一级 1024 个条目。对于 64 位系统,占 8192 Byte (2 pages)*/
struct ftrace_stack {
    unsigned long       calls[FTRACE_KSTACK_ENTRIES];
};
/*每个条目存一个调用函数的地址,故而用的类型是 unsigned long*/
/*允许 4 级嵌套:normal, softirq, irq, NMI,2*4=8个pages*/
struct ftrace_stacks {
    struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
};
/*Per-CPU 变量 ftrace_stacks 占 CPU x 8 个 pages*/
static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
static DEFINE_PER_CPU(int, ftrace_stack_reserve);

Ring Buffer

Ring Buffer 的内存空间分配

  • Ring buffer 的分配接口为ring_buffer_alloc()
  • Trace buffers 的分配接口为allocate_trace_buffers(),它会调用ring_buffer_alloc()来分配 ring buffer
//init/main.c
start_kernel()
   //kernel/trace/trace.c
-> early_trace_init()
   -> tracer_alloc_buffers()
      +-> cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE, "trace/RB:preapre", trace_rb_cpu_prepare, NULL);
      |      //kernel/cpu.c
      |   -> __cpuhp_setup_state(state, name, false, startup, teardown, true);
      |      -> __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, teardown, multi_instance)
      |         -> cpuhp_store_callbacks()
      |            sp->startup.single = startup; //startup 即 trace_rb_cpu_prepare()
      |         if (ret || !invoke || !startup) goto out; //invoke 是 false,所以初始化时以下调用并不会发生
      |            //for_each_present_cpu(cpu)
      |            //-> cpuhp_issue_call()
      +-> temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
      +-> allocate_trace_buffers()
         -> allocate_trace_buffer()
            +-> buf->buffer = ring_buffer_alloc(size, rb_flags);
            |   //kernel/trace/ring_buffer.c
            |   __ring_buffer_alloc()
            |   -> buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), GFP_KERNEL);
            |   -> buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
            |      -> rb_allocate_pages()
            |         -> __rb_allocate_pages()
            |   -> cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node)
            |         //kernel/cpu.c
            |      -> __cpuhp_state_add_instance(state, node, true)
            |         -> __cpuhp_state_add_instance_cpuslocked(state, node, invoke)
            |             for_each_present_cpu(cpu)
            |             -> cpuhp_issue_call()
            |                -> cpuhp_invoke_ap_callback()
            |                  -> cpuhp_invoke_callback()
            +-> buf->data = alloc_percpu(struct trace_array_cpu);
  • 注意,这个路径上先初始化了一个 CPU 使用的 buffer,即buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu)
  • 其他 CPU 使用的 buffer 会根据当前 online CPU 的情况动态分配
cpuhp_invoke_callback()
-> cb = bringup ? step->startup.single : step->teardown.single;
-> cb(cpu)
-> cbm = bringup ? step->startup.multi : step->teardown.multi;
-> cbm(cpu, node)
=> trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
      buffer = container_of(node, struct ring_buffer, node);
   -> buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
  • singlemulti是同一个联合体的成员

  • tracer_alloc_buffers()已经初始化了cpuhp_step->startup.singletrace_rb_cpu_prepare()

  • 因为cpuhp_setup_state_multi()设定好了回调函数trace_rb_cpu_prepare(),随后有 CPU 热插的时候仍然是通过rb_allocate_cpu_buffer()接口初始化其他 CPU 的 ring buffer

  • ring_buffer_alloc()这个宏很关键,能让每个调用它的地方有一个与之绑定的struct lock_class_key __key

    • include/linux/ring_buffer.h
     /*
      * Because the ring buffer is generic, if other users of the ring buffer get
      * traced by ftrace, it can produce lockdep warnings. We need to keep each
      * ring buffer's lock class separate.
      */
     #define ring_buffer_alloc(size, flags)          \
     ({                          \
         static struct lock_class_key __key;     \
         __ring_buffer_alloc((size), (flags), &__key);   \
     })
     ...__
    • 比如说,tracer 的 ring buffer 和 instances 的 ring buffer 都认为是同一个类

Instances

  • Instances 功能的 ring buffer 也是通过allocate_trace_buffers()来分配的(见create_trace_instances()instance_mkdir()等函数),所以 instance 的 ring buffer 是独立的
  • 因为cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE, "trace/RB:preapre", trace_rb_cpu_prepare, NULL)这里将CPUHP_TRACE_RB_PREPARE类型的回调函数设定为trace_rb_cpu_prepare(),所以当instance_mkdir()调用allocate_trace_buffers()去分配 ring buffers 的时候,最终也会通过trace_rb_cpu_prepare()这个接口来分配其他 CPU 的 instances ring buffer,只需要传递不同的node就可以了

Reference