[2.4.22-rc2] x86-64 register_ioctl32_conversion() breakage

[2.4.22-rc2] x86-64 register_ioctl32_conversion() breakage

[Mikael Pettersson]

include/asm-x86_64/ioctl32.h states:

/* 
 * Register an 32bit ioctl translation handler for ioctl cmd.
 *
 * handler == NULL: use 64bit ioctl handler.
 * arguments to handler:  fd: file descriptor
 *                        cmd: ioctl command.
 *                        arg: ioctl argument
 *                        struct file *file: file descriptor pointer.
 */ 

extern int register_ioctl32_conversion(unsigned int cmd, int (*handler)(unsigned int, unsigned int, unsigned long, struct file *));

The "handler == NULL" comment is true in 2.6 kernels (sys_ioctl()
get called, which is what we want for compatible ioctls), but not
in 2.4.22-rc2, where instead sys32_ioctl() oopses because it calls
a NULL function pointer. (I found that out when testing ia32
compatibility in the latest version of the perfctr driver.)

Either ia32_ioctl.c or the comment in ioctl32.h is wrong and should
be fixed. I'd prefer the code to work as in 2.6 since that avoids
#if LINUX_VERSION_CODE crap and dummy ioctl handlers.

/Mikael

Re: [2.4.22-rc2] x86-64 register_ioctl32_conversion() breakage

[Andi Kleen]

Mikael Pettersson <mikpe@csd.uu.se> writes:

> Either ia32_ioctl.c or the comment in ioctl32.h is wrong and should
> be fixed. I'd prefer the code to work as in 2.6 since that avoids
> #if LINUX_VERSION_CODE crap and dummy ioctl handlers.

I fixed it in x86-64.org CVS for 2.4, but it will take some time
to propagate into the various trees. I doubt it will be in 
2.4.22-final.

-Andi

P.S.: The better list would be discuss@x86-64.org, not l-k