#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <fcntl.h>
//#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdint.h>

typedef unsigned long long uint64;
typedef unsigned int uint32;


int main(int argc,char **argv)
{
  int i,j;
  //int ncpus=sysconf(_SC_NPROCESSORS_ONLN);
  char filename[1024];
  int fd[4][1024];
  int msrfd[1024];
  uint32 value;
  uint64_t value64;
  int bus[2]={0x7f,0xff};
  ssize_t rs;
  off_t addr;


  for(i=0;i<2;i++)
  {
    sprintf(filename,"/proc/bus/pci/%x/14.0",bus[i]);
    fd[0][i]=open(filename,O_RDWR);
    sprintf(filename,"/proc/bus/pci/%x/14.1",bus[i]);
    fd[1][i]=open(filename,O_RDWR);
    sprintf(filename,"/proc/bus/pci/%x/17.0",bus[i]);
    fd[2][i]=open(filename,O_RDWR);
    sprintf(filename,"/proc/bus/pci/%x/17.1",bus[i]);
    fd[3][i]=open(filename,O_RDWR);
  }


  sprintf(filename,"/dev/cpu/0/msr");
  msrfd[0]=open(filename,O_RDWR);



  /*=== START TO SET BITS ===*/
  // a) step
  value64=0x1<<31ULL&0x0FFFFFFFF;
  rs=pwrite(msrfd[0],(void*)&value64,sizeof(value64),0x0700); // U_MSR_PMON_GLOBAL_CTL

  for(i=0;i<4;i++)
  {
  for(j=0;j<2;j++)
  {
    // or a) step
    value64=0x1<<8ULL|0x1<<16|0x1<<17;
    rs=pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xF4);

    // b) step
    value64=0x1<<22ULL;
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xE0);
    // c) step
    value64=0x04|0x300|0x1<<22ULL; // RD
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xE0);
    
    // b) step
    value64=0x1<<22ULL;
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xE4);
    // c) step
    value64=0x04|0xC00|0x1<<22ULL; // WR
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xE4);

    // d) step
    value64=0x1|0x1<<1ULL|0x1<<8ULL|0x1<<16|0x1<<17;
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xF4);
  }
  }

#if 0
  // e) step
  value64=0x03FFFF;
  rs=pwrite(msrfd[0],(void*)&value64,sizeof(value64),0x0700); // U_MSR_PMON_GLOBAL_CTL
#endif

  // f) step
  value64=(0x1<<29ULL|0x03FFFF)&0x0FFFFFFFF;
  rs=pwrite(msrfd[0],(void*)&value64,sizeof(value64),0x0700); // U_MSR_PMON_GLOBAL_CTL

  // or f) step
  for(i=0;i<4;i++)
  {
  for(j=0;j<2;j++)
  {
    value64=0x1<<1ULL;
    pwrite(fd[i][j],(void*)&value64,sizeof(value64),0xF4);
  }
  }
  /*=== END TO SET BITS ===*/


  uint64 val1,val2,val3,val4;
  while(1)
  {

  for(i=0;i<4;i++)
  {
  for(j=0;j<2;j++)
  {
    pread(fd[i][j],(void*)&val1,sizeof(uint64),0xB4);
    pread(fd[i][j],(void*)&val2,sizeof(uint64),0xB0);
    pread(fd[i][j],(void*)&val3,sizeof(uint64),0xBC);
    pread(fd[i][j],(void*)&val4,sizeof(uint64),0xB8);
    printf("[%d][%d] B4=%ld B0=%ld BC=%ld B8=%ld\n",i,j,val1,val2,val3,val4);
  }
  }

  sleep(1);
  }
  for(i=0;i<4;i++)
  {
  for(j=0;j<2;j++)
  {
    close(fd[i][j]);
  }
  }

  return 0;
}