Refresh the strong pointer assignment problem in the HP-UX system of Sanguan

        Upgrade the communication middleware in a province this week , stay RedHat Middleware of the platform Running normally , Business can be handled normally . But in HP-UX Server , 1. Carrying business , Middleware comes out Core restart .

        I immediately organized my colleagues in the Department to participate in the problem analysis .

        First of all, I worked in the company's laboratory HP Deploy the business of a province on the server . Then write in the laboratory votel Traffic test script , Conduct business test . The fault phenomenon appears ,HP The middleware under core restart . Where to analyze the code , It's in         int i=*(int*)(*begin+1); place .

        Then roll back the version , The original code assignment is

+   int i;

+   memcpy((char*)(&i),(*begin)+1,4);

    Change the code from strong pointer conversion to memcpy after , The business middleware runs normally .

    The problem is beyond all comprehension , So I arranged for Xiao Wang to write a simple test program , Only cast code is written in the code .

   

#include <stdio.h>
int main()
{
char  aa[10] = {'a','b','c','d','d','e','f','a','b','c'};
char *p = (aa+0);
char *p2 = (aa+0); 
char **begin = &p2;


printf("%p\n",*begin);

printf("%p\n",*begin+1);

printf("%p\n",**begin);

int functionID = *(int*)(*begin+1);
printf("functionID=%x\n",functionID);
}

  use aCC After compiling , Run the program , Sure enough core.

Then use the code

​
#include <stdio.h>
int main()
{
char  aa[10] = {'a','b','c','d','d','e','f','a','b','c'};
char *p = (aa+0);
char *p2 = (aa+0); 
char **begin = &p2;


printf("%p\n",*begin);

printf("%p\n",*begin+1);

printf("%p\n",**begin);

int functionID ;
memcpy((char*)&functionID,(*begin)+1,4);
printf("functionID=%x\n",functionID);
}

​

aCC It runs normally after compilation .

Xiao Wang told me that he saw a post , It can be used for reference , As follows ：

  Generally speaking, there are CPU There is a mechanism for memory parity , To cast an even number of memory addresses .

So I follow this article , The code has been modified and verified ：

#include <stdio.h>
int main()
{
char  aa[10] = {'a','b','c','d','d','e','f','a','b','c'};
char *p = (aa+0);
char *p2 = (aa+0); 
char **begin = &p2;


printf("%p\n",*begin);

printf("%p\n",*begin+1);

printf("%p\n",**begin);
int functionID = *(int*)(*begin+2);
printf("functionID=%x\n",functionID);
}

after aCC After compilation and verification , The program still works core.

Then I went on to make changes , As follows ：

​
#include <stdio.h>
int main()
{
char  aa[10] = {'a','b','c','d','d','e','f','a','b','c'};
char *p = (aa+0);
char *p2 = (aa+0); 
char **begin = &p2;


printf("%p\n",*begin);

printf("%p\n",*begin+1);

printf("%p\n",**begin);
int functionID = *(int*)(*begin+4);
printf("functionID=%x\n",functionID);
}

​

For pointer addresses +4 after , Make a cast , use aCC Run... After compilation , The program can run normally .

Then I told Xiao Wang about the situation , Xiao Wang did another test ：

​
#include <stdio.h>
int main()
{
char  aa[10] = {'a','b','c','d','d','e','f','a','b','c'};
char *p = (aa+0);
char *p2 = (aa+0); 
char **begin = &p2;


printf("%p\n",*begin);

printf("%p\n",*begin+1);

printf("%p\n",**begin);
int functionID = *(short*)(*begin+2);
printf("functionID=%x\n",functionID);
}

​

use aCC After compiling , The program can also run normally .

Sum up ：

It is necessary to be cautious when performing strong conversion assignment for values , yes , we have CPU There is memory alignment and parity of starting addresses . So try to use memcpy（ similar memcpy((char*)(&i),(*begin)+1,4);）, Less use of type forced conversion assignment （ similar int i=*(int*)(*begin+1);）.