GDB笔记

概览

基础

用户态会话

调试模式

• 转储文件调试
  
  瞬间静态分析，进程死状态。
  gdb –core=file

• 交互调试

  1. 调试新进程
     gdb exe
     gdb –args exe [args]

  2. 调试已经运行的进程
     gdb –pid= 进程号

  3. 内核调试

3种调试已有进程

gdb exe

.

gdb –args exe [args]

.

gdb
file exe
run [args]

.

gdb –args gcc a.c -o a

命令行参数

show args

set args 多次运行设置命令行参数

环境变量和执行路径

path directory

.

show paths

.

show environment [varname]

.

set environment varname[=value] 清除或者设置环境变量

工作目录

继承进入gdb工作目录

改变工作目录

cd dirctory

显示路径

pwd

输入输出

info terminal
run > a.txt
tty /dev/ttyb

远程调试可用这些辅助。

inferior 下层 多个进程调试

inferior gdb维护的一系列对象，每个inf对应一个调试目标进程。

info inferior 显示下层信息
NULL 程序没有跑或者已经终止
clone-inferior -copies 2 复制当前下层2份

(gdb) info inferiors
Num Description Executable
* 1 process 10087 /home/gao/code/a
(gdb) clone-inferior -copies 2
Added inferior 2.
Added inferior 3.
(gdb) info inferiors
Num Description Executable
3 程序没有跑或者已经终止 /home/gao/code/a
2 /home/gao/code/a
* 1 process 10087 /home/gao/code/a
(gdb)

切换下程

inferior 2 切换2这个下程。
进程号是0，没开始运行。
run 运行起来。

.

增加一个运行下层

add-inferior -exec executeable 增加一个运行下层
比如说调试一个服务端程序，一个客户端程序。

remove-inferior n 删掉一个下层
detach inferior 继续运行 quit
kill inferior 调试进程退了，但是inferior纪录还在。

Tab 帮助

(gdb) remove-
remove-inferiors remove-symbol-file
(gdb) remove-
remove-inferiors remove-symbol-file
(gdb) remove-

file 命令

file a.exe 可自行文件和符号文件是一个文件

可自行文件和符号文件分开

exec-file 指定目标文件

.

symbol-file 指定符号文件

run 开始运行
可以支持 run > >> < 重定向

set args 清理命令行参数

附加到进程

gdb –pid= pid

attach pid

终止调试进程

detach pid 分离进程继续运行

.

quit 进程退出

.

q
ctrl + D

执行控制

断点

软件断点

break 普通
tbreak 一次性
rbreak 正则表达式一批断点

• 基于cpu断点指令，x86 int3机器码0xcc。
• 替换断点位置的指令
• CPU自执行这里触发断点异常。
• 没有断点数量限制。

硬件断点

• 基于cpu调试寄存器，dr0～dr7，数量限制。x86可以设定4个断点。数量限制。
• 不修改代码，在只读内存上设置断点。EEPROM上的代码设置。
• 有数量限制。

location

• linespec

1. 行号
2. －／＋ 偏移
3. 文件名 ：行号
4. 函数名
5. 函数:标号
6. 文件名：函数

7. 标号

• explicit

1. -source linename
2. -function function
3. -label label

4. -line number

• address location break * address

实践

file banner
b main 中断在main函数
info funciton useage 显示useage函数地址
info *0x88888e4 直接写地址设定断点
info b 显示所有断点
list usage 显示useage函数
b line.c:11
b +2 当前显示到14行＋2行所以设定在17行

虚拟机设定硬件断点会失败.
(gdb)hbreak hd_ioctl
(gdb) info b
物理机可以设定.

(gdb) hbreak v
Hardware assisted breakpoint 2 at 0x40053a: v. (2 locations)
(gdb) info b
Num Type Disp Enb Address What
1 breakpoint keep y
breakpoint already hit 1 time
1.1 y 0x000000000040054f in main at a.c:10 inf 1
1.2 y 0x000000000040054f in main at a.c:10 inf 2
2 hw breakpoint keep y
2.1 y 0x000000000040053a in v at a.c:4 inf 1
2.2 y 0x000000000040053a in v at a.c:4 inf 2
(gdb)

管理断点

info b 显示断点
delete 1 删除
disable 1
enable 1
delete 删除所有的断点
clear sum 删除sum函数入口的所有断点

扩展断点

watchpoint 监视一个表达式，值变化中断。

watch a＊b＋c／d
watch *( int * )

watch fd 监控局部变量fd
c 继续运行
info b //
注意，x86硬件调试，写操作，执行过这一行，gdb显示下一行，要看上一行。hw watchpoint，在x86平台wachtpoint基于硬件实现，其他平台可能是基于软件实现。 vc6纯软件执行，执行目标速度低。
执行位置超过当前区域，无效监视点会被删除。

访问监视点
rwatch 读停下来。
awatch 读或写停下来。 awatch fd
watch -l

b hd_ioctl thread 1

info threads 带＊ 当前线程

繁忙函数解决方案

b hd_ioctl thread 1 if fd > 0

当断点，断了后执行命令
()command 12
()silent
()print “fd is %d\n”,fd
()continue
()end

动态ping不修改代码。

tracepoint 远程主机通讯调试，前端stub立刻恢复执行，但会记录下来。

catchpoint

执行控制

进入子函数内部。单步 step

汇编 stepi

stepi 4
类似nexti

不要进入子函数

next

继续执行

continue

跑起来直到3号断点命中

until 3
跑起来直到3号断点命中，快捷禁止其他断点。

恢复执行直到函数返回。

finish

调用函数！

gdb 杜撰代码调用函数。
call sum（1，2） …

强制返回

强制main函数 return。
return 1

触发中断

异常或断点进入调试器。
调试器发起中断，让程序中断下来。ctrl＋C，app收到中断信号，进入调试器。

符号

调试器读区 调试符号。

二进制－调试符号－源码

linux dwarf 存储调试符号信息。gcc

readelf -h filename
里面如果有line location debug标示

readelf -w 导出调试文件

gcc -g 才能输出符号

ubuntu 符号服务器
< ddebs.ubuntu.com/pool/main/>
分离操作
strip

安装ubunte的linux 内核符号
https://askubuntu.com/questions/197016/how-to-install-a-package-that-contains-ubuntu-kernel-debug-symbols

安装libc符号

dpkg -s /lib/x86_64-linux-gun/libc-2.15.so
dpkg -s libc.so.6
sudo apt-get -c aptproxy.conf install libc6-dbg

libc 调试符号

sudo apt-get install libc6-dbg

符号路径

gdb 使用file 或 symbol－file 加载符号文件
自动搜索 path 路径

（gdb）i share
* 共享库没调试信息

搜索符号
info vaiables regex 类名／函数名／变量名

内存地址与符号互换

info addriess 函数名

info symbol 地址

(gdb) info address main
Symbol “main” is a function at address 0x400547.
(gdb) info symbol 0x400547
main in section .text of /home/gao/code/a

.

info os

查看加载的文件内存位置

info files

列出全局变量

info variable
info va

显示源码

list
list -
dir 源码路径
show dir

常用命令源码

安装系统工具源码和调试

apt-get source coreutils
sudo apt-get install coreutils-dbgsym
gdb /bin/ls
list main
dir ~/src/coreutils-7.4/src
list main

libc

sudo apt-get source libc6-dev
/home/ge/eglibc-2.15

dir 搜索路径 ：分割
$cdir 编译路径
cwd 当前工作路径

查看调试目标

观察寄存器
info reg

栈
子函数返回地址
函数参数
局部变量

bt n 观察函数返回地址
frame n 切换栈帧
up n
down n
info frame ［address］
info args
info locals

注意，切换栈帧之后可能会发生，本地变量值不准确，因为值存在寄存器中需要小心。

观察内存
print

p /f 表达式 表达式要打印位置
xduotcf

x

x /Nuf
N 打印几个单元
u 每个单元大小 b-1byte w-2byte h-4byte g-8byte

f s字符串i指令格式

x/s 0xfffff81946000 打印字符串

x /32bx arg bit 16禁制

(gdb) x /32bx &i
0x7fffffffc76c: 0x01 0x00 0x00 0x00 0x70 0x05 0x40 0x00
0x7fffffffc774: 0x00 0x00 0x00 0x00 0x40 0xfa 0xa2 0xf7
0x7fffffffc77c: 0xff 0x7f 0x00 0x00 0x58 0xc8 0xff 0xff
0x7fffffffc784: 0xff 0x7f 0x00 0x00 0x58 0xc8 0xff 0xff
(gdb)

p arg[0]
p arg[i]

p *&a[0]@10 a0数组开始的10个元素

反汇编

disas main main反汇编代码
x/5i schedule 这个地址开始的5条汇编指令。

gdb mov 从左往右赋值at&t汇编。

高级技巧

信号

• info signals 异常／同步／中断

stop 要不要中断下来看
printf 打印信息
pass 要不要传递给应用程序。

• handle 修改规则 handle signal act print noprint stop nostop pass nopass

handle SIGPIPE 不要中断下来，打印一个信息，网络程序常用

(gdb) handle SIGPIPE nostop
Signal Stop Print Pass to program Description
SIGPIPE No Yes Yes Broken pipe

Thread

info threads

LWP－light weight process 线程编号。
* gdb当前线程

切换当前线程

thread 2

打印所有线程

thread apply all bt 针对一群线程的命令避免切换来看。

线程改名字

thread name ［name］

我自己经验 LWP 可以很好的观察线程负载情况。

C面向对象编程汇集

参考资料：
Object-oriented Programming with ANSI-C
1993年，第一份c如何编写OO的资料,free.
OOC.PDF
https://www.cs.rit.edu/~ats/books/ooc.pdf

中文翻译：
https://code.google.com/p/ooc/downloads/detail?name=ooc-translate-preview-r26.pdf&can=2&q=

轻量级的C语言面向对象编程框架
http://sinojelly.blog.51cto.com/479153/281184

UML—OOPC嵌入式C语言开发精讲
里面有一套框架可以 c写OO.
http://pan.baidu.com/share/link?shareid=3402978666&uk=3188261067&adapt=pc&fr=ftw#path=%252FC%25E8%25AF%25AD%25E8%25A8%2580

你试过这样写C程序吗 －－函数式编程
< >

我所偏爱的 C 语言面向对象编程范式－－云风
http://blog.codingnow.com/2010/03/object_oriented_programming_in_c.html

C语言面向对象编程 -- 6篇专栏
http://blog.csdn.net/column/details/object-orient-c.html

测试代码的locality。
数组的读区方式不同，按照行读，被cache也是按行连续加载的。
如果按照列读区，那么效率很低，除非cache足够大，而且也要遍历所有的数据，并且cache hash算法也好，实现的硬件还是多路组相联的cache硬件实现。

** valgrind --tool=cachegrind ./test2**

code1:

#include <stdio.h>
#define MAXROW 8000
#define MAXCOL 8000
int main () {
int i,j;
 static int x[MAXROW][MAXCOL];
 printf ("Starting!\n");
       for (i=0;i<MAXROW;i++)
       for (j=0;j<MAXCOL;j++)
              x[i][j] = i*j;
             printf("Completed!\n");
return 0;                                                    
 }

code2:

#include <stdio.h>                                                         
 #define MAXROW 8000
 #define MAXCOL 8000
 int main () {
 int i,j;
 static int x[MAXROW][MAXCOL];
 printf ("Starting!\n");
          for (j=0;j<MAXCOL;j++)
                         for (i=0;i<MAXROW;i++)
                 x[i][j] = i*j;
 printf("Completed!\n");
 return 0;
 }
 ```

##结果

Command: ./test1
Starting!
Completed!

I refs: 905,721,688
I1 misses: 4,177
LLi misses: 2,808
I1 miss rate: 0.00%
LLi miss rate: 0.00%

D refs: 514,830,867 (386,118,735 rd + 128,712,132 wr)
D1 misses: 4,025,828 ( 23,565 rd + 4,002,263 wr)
LLd misses: 4,008,456 ( 6,997 rd + 4,001,459 wr)

D1 miss rate: 0.8% ( 0.0% + 3.1% )
LLd miss rate: 0.8% ( 0.0% + 3.1% )

LL refs: 4,030,005 ( 27,742 rd + 4,002,263 wr)
LL misses: 4,011,264 ( 9,805 rd + 4,001,459 wr)
LL miss rate: 0.3% ( 0.0% + 3.1% )

gcc -o test2 test2.c
** valgrind --tool=cachegrind ./test2**

I refs: 905,720,801
I1 misses: 4,113
LLi misses: 2,811
I1 miss rate: 0.00%
LLi miss rate: 0.00%

D refs: 514,830,348 (386,118,427 rd + 128,711,921 wr)
D1 misses: 64,025,705 ( 23,462 rd + 64,002,243 wr)
LLd misses: 4,016,427 ( 6,977 rd + 4,009,450 wr)
D1 miss rate: 12.4% ( 0.0% + 49.7% )
LLd miss rate: 0.8% ( 0.0% + 3.1% )

LL refs: 64,029,818 ( 27,575 rd + 64,002,243 wr)
LL misses: 4,019,238 ( 9,788 rd + 4,009,450 wr)
LL miss rate: 0.3% ( 0.0% + 3.1% )

Starting!
Completed!
```

参考：

valgrind调试CPU缓存命中率和内存泄漏
http://laoxu.blog.51cto.com/4120547/1395236

1.vc  编译

头文件

#include <omp.h

omp_set_num_threads(threadNumber);
#pragma omp parallel default(shared) private(i,riskId) 
    {
#pragma  omp for   

}

}

GCC配置

1. 编译

-fopenmp

设定线程数量 export OMP_NUM_THREADS=6

参考

通过 GCC 学习 OpenMP 框架

http://www.ibm.com/developerworks/cn/aix/library/au-aix-openmp-framework/

Intellij IDEA和Android Studio默认使用JDK6，所以你有如下选择：

1，安装JDK6

2，安装其他版本的JDK，然后修改IDE的指向。

如果选择2，那么修改方式为：

打开应用的.app包，然后在Contents这个目录下，有个文件info.plist，修改JVMVersion这个key对应的value，从1.6*修改为你安装的版本即可。

1 用文本编辑器打开
/Applications/IntelliJ IDEA 13.app/Contents/Info.plist

2 搜索JVMVersion，将其值改为1.7*

3 再次运行应用即可看到应用成功运行

sudo vim /etc/resolv.conf

wget http://www.alidns.com/static/soft/SetAliDNS.sh -O SetAliDNS.sh && chmod +x SetAliDNS.sh && sudo ./SetAliDNS.sh

卸载 

sudo ./SetAliDNS.sh restore

API对设计流程的影响——Joshua Bloch访谈

http://blog.sina.com.cn/s/blog_6310e0b20100o7cv.html 

工程师Joshua Bloch谈如何设计一款优秀的API【附PPT】

http://www.csdn.net/article/2014-02-18/2818441-How-to-design-a-good-API

《Effective Java》: Joshua Bloch访谈

http://eastsun.iteye.com/blog/195861

The Java™ Tutorials

http://docs.oracle.com/javase/tutorial/java/javaOO/lambdaexpressions.html

1.5

New Features and Enhancements J2SE 5.0

http://docs.oracle.com/javase/1.5.0/docs/relnotes/features.html

Experiences with the New Java 5 Language Features

Pages: 1, 2, 3, 4

1.6

Highlights of Technology Changes in Java SE 6

http://www.oracle.com/technetwork/java/javase/features-141434.html

1.7

Highlights of Technology Changes in Java SE 7

http://www.oracle.com/technetwork/java/javase/jdk7-relnotes-418459.html

http://www.javacodegeeks.com/2014/05/java-8-features-tutorial.html

学习步骤
1.c 
2.c object 
3. stl
4. linux os api
5. socket  
6. 分布式
7. web html5 
8. mac UI  
9. obj c
10.算法。


一些想法：
微博上关注的人很多淘宝系的很多都不搞技术了，原因是阿里上市了，足够的钱退休了，还有些人离职去创业了，
可能氛围比较好，看来创业的都有滋有味。大规模的系统搞过了，心里有底气了，知道方向，估计成功率也会比较高。
一些人技术人靠知乎刷存在感，靠名气说话 。
一些2流技术角色投稿文章csdn，写专题blog。很多毫无营养。
国内的技术书，还是一如既往类似官方教程的翻译版。

 

blog

http://preshing.com/ 

Jeff Atwood 大stackoverflo

http://blog.codinghorror.com/

https://github.com/coding-horror

陈皓

http://coolshell.cn/ 

云风

http://blog.codingnow.com/ 

何_登成

http://weibo.com/2216172320/BcUmvbyvW?type=repost#_rnd1418722416088

http://ifeve.com/

http://www.importnew.com/14105.html

网络教学:

http://www.codecademy.com/

PLSQL

http://plsql-tutorial.com/plsql-functions.htm

先确保执行这个. win7 以上才支持 WER windows error report 功能 .

windows 弹出提示框, 然后进程管理器,右键手工导出也行.

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps]
"DumpFolder"="c:\\"
"DumpCount"=dword:00000030
"DumpType"=dword:00000002
"CustomDumpFlags"=dword:00000000

c++ debug版本 + pdb文件

另外需要注意不是所有情况都生成core.

windows 改了注册表不生成core的原因:

Requirements and limitations

· Debugging dump files of optimized code can be confusing. For example, compiler inlining of functions can result in unexpected call stacks and other optimizations might change the lifetime of variables.

· Dump files from 64-bit machines must be debugged on an instance of Visual Studio that is running on a 64-bit computer.

· In versions of Visual Studio before VS 2013, dumps of 32-bit apps that were run on 64-bit machines that were collected by some tools (such as Task Manager and 64-bit WinDbg) could not be opened in Visual Studio. This limitation has been removed in VS 2013.

· Visual Studio can debug dump files of native apps from ARM devices. Visual Studio can also debug apps dump files of managed apps from ARM devices, but only in the native debugger.

· To debug kernel-mode dump files in Visual Studio 2013, download the Windows 8.1 Version of Debugging Tools for Windows. See Kernel Debugging in Visual Studio.

· Visual Studio can't debug dump files saved in the older dump format known as a full user-mode dump. Note that a full user-mode dump is not the same a dump with heap.

· To debug with the SOS.dll (SOS Debugging Extension) in Visual Studio, you must install the Debugging Tools for Windows that is part of the Windows Driver Kit (WDK). See Windows 8.1 Preview: Download kits, bits, and tools.

参考:

Collecting User-Mode Dumps   --注册表

http://msdn.microsoft.com/en-us/library/windows/desktop/bb787181(v=vs.85).aspx

Crash Dump Analysis  --- c++ 自己生成dump

http://msdn.microsoft.com/en-us/library/windows/desktop/ee416349(v=vs.85).aspx 

Use Dump Files to Debug App Crashes and Hangs in Visual Studio

http://msdn.microsoft.com/en-us/library/d5zhxt22.aspx#bkmk_requirements_and_limitations

Attach to Running Processes with the Visual Studio Debugger

http://msdn.microsoft.com/en-us/library/3s68z0b3.aspx

Just-In-Time Debugging in Visual Studio  ------- 还没看

http://msdn.microsoft.com/en-us/library/5hs4b7a6.aspx

The Discipline and Method Architecture for Reusable Libraries (2000)

Citations

relocation error: multiget: symbol _Z23wxHandleFatalExceptionsb, version WXU_2.8 not defined in file libwx_baseu-2.8.so.0

下载地址:

http://www.codeblocks.org/downloads/26 

http://sourceforge.net/projects/codeblocks/files/Binaries/13.12-RC1/Linux%20(64bit)/

codeblocks-13.12-1.el5.x86_64.tar.bz2

安装的关键

Note: On RedHat/CentOS 5 and older revisions of 6 (up to 6.2 as far as I know) you need to add repoforge (former rpmforge) to your repolist, to be able to install the needed wxGTK-package. Seehttp://repoforge.org/use for an instruction.

更新yum的源 很重要

http://repoforge.org/use/  下载对应的rpm

cat /etc/redhat-release

uname –a

wget http://apt.sw.be/redhat/el6/en/x86_64/rpmforge/RPMS/rpmforge-release-0.5.3-1.el6.rf.x86_64.rpm

rpm -ivh package-filename

# yum install gcc
# yum install gcc-c++

如果之前失败过  yum erase wxGTK

# yum install wxGTK

然后

 ]# ls
codeblocks-12.11-1.el6.i686.rpm
codeblocks-contrib-12.11-1.el6.i686.rpm
codeblocks-contrib-devel-12.11-1.el6.i686.rpm
codeblocks-devel-12.11-1.el6.i686.rpm

# rpm -ivh codeblocks*

1. 下载repo文件
    下载地址：http://mirrors.163.com/.help/CentOS6-Base-163.repo

2. 备份并替换系统的repo文件
[root@localhost ~]# cd /etc/yum.repos.d/
[root@localhost ~]# mv CentOS-Base.repo CentOS-Base.repo.bak
[root@localhost ~]# mv /root/CentOS6-Base-163.repo CentOS-Base.repo

3. 执行yum源更新
[root@localhost ~]# yum clean all
[root@localhost ~]# yum makecache
[root@localhost ~]# yum update

yum install gcc gcc-g++ autoconfig automake 

参考:

http://blog.csdn.net/dupei/article/details/6428346

http://www.cnblogs.com/magialmoon/archive/2013/05/05/3061108.html

sqlplus 外面执行正常.

tomcat调用老有问题.多调用几次久正常.

可能是包中的全局变量引起的.

If so, that's because your package is stateful:

The values of the variables, constants, and cursors that a package declares (in either its specification or body) comprise its package state. If a PL/SQL package declares at least one variable, constant, or cursor, then the package is stateful; otherwise, it is stateless.

When you recompile the state is lost:

If the body of an instantiated, stateful package is recompiled (either explicitly, with the "ALTER PACKAGE Statement", or implicitly), the next invocation of a subprogram in the package causes Oracle Database to discard the existing package state and raise the exception ORA-04068.

After PL/SQL raises the exception, a reference to the package causes Oracle Database to re-instantiate the package, which re-initializes it...

You can't avoid this if your package has state. I think it's fairly rare to really need a package to be stateful though, so you should revisit anything you have declared in the package, but outside a function or procedure, to see if it's really needed at that level. Since you're on 10g though, that includes constants, not just variables and cursors.

But the last paragraph from the quoted documentation means that the next time you reference the package in the same session, you won't get the error and it will work as normal (until you recompile again).

existing state of packages has been discarded means, that your Package had some sort of state.

This is caused by a global variable (or constant) stored in your Package Body.

Since the package has already been used in your session, Oracle assumes that this state is relevant for you. Some of these variables might have different values now, and when you recompile the Body, the values are reset.

This exception is thrown, so that your clients know that they can't rely on those variables any more.

You could try to remove all global variables and constants from the Package Body, or close your session and reconnect before calling the package again.

参考:

http://stackoverflow.com/questions/2502722/pl-sql-package-invalidated

http://stackoverflow.com/questions/19376440/ora-06508-pl-sql-could-not-find-program-unit-being-called