MIT 6.S081 Lecture 2 Notes

5 Jul 2022

The creation of a thousand forests is one acorn.

6.S081第二课，主要是TA教授一些C的知识与GDB使用。在这里仅记录一些有意思的点，

Intor to C

课程里这个介绍C知识的slides比较短小精悍，基本覆盖了大部分常见且关键的知识点。美中不足的是没有关于 spiral rule 的知识。这个处理规则在写一些复杂的 typedef 与函数指针声明时十分有用，比如 signal handler 的类型声明。

void* pointer

You can’t define a variable of type void, because… what would that mean?

But you can define a variable of type void *. You just can’t dereference it.

(You aren’t allowed to use arithmetic on void * pointers officially, but we use GCC, which supports it as an extension to C.)

关于void* 指针可以指但无法解引用的特性，这点常常在底层库中应用为各式指针类型的转换。关于void* 的指针算术标准是不提供的，具体可以看标准用数组的类比，由于void数组不可能，所以 void* 算术也不可能。但GCC开了一个口，进行扩展，将 sizeof void 当作一字节，以支持 void* 算术计算。

local static variable initialization

关于局部静态对象何时初始化的问题。C与C++明显有差异。slides的例子里，

int add_cumulative_numbers(int increase) {
  static int total_sum = 0;
  total_sum += increase;
  return total_sum;
}

total_sum will be initialized to zero at program start, and it will keep its value across calls to add_cumulative_numbers! It won’t be reinitialized.

但是在C++里，情况稍复杂一点。Effective C++系列（忘了是不是Modern）提到过，局部静态对象可以实现为仅在对应函数第一次被调用时初始化，做到类似 lazy initialization 的实现。例如learncpp中这段C++代码，

#include <iostream>

void incrementAndPrint()
{
    static int s_value{ 1 }; // static duration via static keyword.  This initializer is only executed once.
    ++s_value;
    std::cout << s_value << '\n';
} // s_value is not destroyed here, but becomes inaccessible because it goes out of scope

int main()
{
    incrementAndPrint();
    incrementAndPrint();
    incrementAndPrint();

    return 0;
}

In this program, because s_value has been declared as static, it is created at the program start.

Static local variables that are zero initialized or have a constexpr initializer can be initialized at program start. Static local variables with non-constexpr initializers are initialized the first time the variable definition is encountered (the definition is skipped on subsequent calls, so no reinitialization happens). Because s_value has constexpr initializer 1, s_value will be initialized at program start.

简单说就是虽然在程序开始时创建，但初始化的时间根据 initializer 是否为 constexpr 有区别。对于 non-constexpr 可以做到lazy initialization。这个由于比较确定，就不找标准的原文验证了。

typedef in header file

将一些类型定义在头文件中，有利编写跨平台代码。只更换头文件即可作出平台对不同类型的适应。如在32位平台，可以将xv6 的 typedef unsigned int uint32; 替换为 typedef unsigned long uint32;等。编写一些跨平台代码时，最好也用这些明确定义比特位数的类型。关于include guard就不展开说了，xv6的头文件暂时都没用到这个特性（可能头文件太少，内核太简单，没有include重复展开的问题）。

// from kernel/types.h:
typedef unsigned char uint8;
typedef unsigned short uint16;
typedef unsigned int uint32;
typedef unsigned long uint64;

Miscellaneous

一点有趣的杂项。通过字段名字而不是顺序初始化一个结构体，

struct xy_point my_point = {
  .y = -6.2,
  .x = 12.5,
};

判断一个整数是否为2的幂次，

// Check if integer is a power of two
if (my_int && !(my_int & (my_int - 1))) { /* ... */ }

对于一个大于0的整数其实 !(my_int & (my_int - 1)) 已经能够判断。因为如果二进制表示有1（大于0二进制表示必有1），my & (my_int - 1) 相当于把最低位置的1去掉, 如果为2幂次，则二进制表示只有一个1，结果为0；如果不为2幂次，则二进制表示多个1，结果为非0。考虑 my_int 为0的特殊情况，所以还要在前面加 my_int && ... 形式。

A pointer challenge

最后 slides 提供了一个来自The Ksplice Pointer Challenge的问题，检验对于C指针的理解，问题要求基于64位小端机器，int为32的假设，写出四个 printf 的内容。

int main() {
 int x[5]; // x is at 0x7fffdfbf7f00
 printf("%p\n", x);
 printf("%p\n", x+1);
 printf("%p\n", &x);
 printf("%p\n", &x+1);
 return 0;
}

答案与解释博客中均有，比较有意思的是最后 &x+1 的答案。因为 &x is a int *[5] 所以 offset 的单位是 5 * sizeof(int)，即20个字节。最终答案为 0x7fffdfbf7f14。

GDB

课程slides教授了GDB的基本使用。最开始讲了执行代码时栈的变化，即函数调用时如何压栈保存寄存器(caller-saved registers)，以及为函数局部变量分配的栈空间大小，这类问题属于本科课程与CSAPP都讲过，就不再赘述。我主要把这个slides的GDB使用做成 cheat sheet。

command	effect
`help <command-name>`	show usage about command
Stepping
`step`	runs one line of code at a time. When there is a function call, it steps into the called function
`next`	does the same thing, except that it steps over function calls
`stepi` and `nexti`	do the same thing for assembly instructions rather than lines of code
All take a numerical argument to specify repetition. Pressing the enter key repeats the previous command.
Running
`continue`	runs code until a breakpoint is encountered or you interrupt it with Control-C
`finish`	runs code until the current function returns
`advance <location>`	runs code until the instruction pointer gets to the specified location
Breakpoints
`break <location>`	sets a breakpoint at the specified
location Locations can be memory addresses (“*0x7c00”) or names (“mon backtrace”, “monitor.c:71”). Modify breakpoints using `delete`, `disable`, `enable`.
Conditional breakpoints
`break <location>`	if `<condition>` sets a breakpoint at the specified location, but only breaks if the condition is satisfied.
`cond <number> <condition>`	adds a condition on an existing breakpoint.
Watchpoints
`watch <expression>`	will stop execution whenever the expression’s value changes.
`watch -l <address>`	will stop execution whenever the contents of the specified memory address change.
`rwatch [-l] <expression>`	will stop execution whenever the value of the expression is read.
Examining
`x`	prints the raw contents of memory in whatever format you specify (x/x for hexadecimal, x/i for assembly, etc).
`print`	evaluates a C expression and prints the result as its proper type. It is often more useful than x.
The output from p ((struct elfhdr ) 0x10000) is much nicer than the output from x/13x 0x10000.
More examining
`info registers`	prints the value of every register.
`info frame`	prints the current stack frame.
`list <location>`	prints the source code of the function at the specified location.
`backtrace`	display the current call stack (can be used after a runtime error, eg. segfault).
Layouts
`layout <name>`	switches to the given layout. GDB has a text user interface that shows useful information like code listing, disassembly, and register contents in a curses UI.
Miscellaneous
`set var var=expr`	stores value of expression expr into program variable var during execution. e.g. `set var g=4` store 4 into program variable g
`symbol-file`	switches symbol files. e.g. when debugging JOS, `symbol-file obj/user/<name>` and `symbol-file obj/kern/kernel` will switch function and variable names between environments and the kernel.

P.S.

之前发布的公众号文章超链接无法跳转，以后公众号文章尝试添加博客原文链接与引用链接。

Reference

system-programming