Hey all,

I'm working on a driver write now that makes use of some ioctl calls and I'm having some strange problems. In the kernel I have a header file for the driver that defines the ioctl commands similar to this:
#define DRIVER_IOCTL_MAGIC 0xFA
#define DRIVER_IOCTL_REG_PROC _IOW(DRIVER_IOCTL_MAGIC, 1, struct sigproc_info)
a few more commands...

sigproc_info is a pretty simple struct made up mostly of ints that I have already defined in the header file.

I go ahead and include this header file in the relevant user space applications and send the DRIVER_IOCTL_REG_PROC command as such:
const int fd = open ( "/dev/driver_file", O_RDWR );
...
struct sigproc_info info;
...
rc = ( fd, DRIVER_IOCTL_REG_PROC, &info);
...

So here's the problem-the value that the kernel assigns to the DRIVER_IOCTL_REG_PROC command differs from the value that the user space application uses. Further, it is called from a couple user space applications which all seem to agree on the value of DRIVER_IOCTL_REG_PROC. So in the actual driver implementation of ioctl handling the switch statement hits the default case due to the mismatch and I can't get it to recognize the command correctly.

Any ideas?

Problem was in the struct I was passing. Guess it wasn't as simple and benign as I made it sound. Really stupid mistake just too involved with the code to notice it. Here's the original with the mistake, can you figure it out?
struct sigproc_info
{
#if defined(__KERNEL__)
int dirty;
#endif
int pid;
int signum;
int irq;
} __attribute__ ((__packed__));

...
...
...
The call to _IOW takes the last argument, struct sigproc_info, and calls sizeof on it then performs some bit-shifting based on this value. So the kernel-space version was 4 larger than the user-space version since it had an extra int, dirty