Patch just for comment, do not apply.

In network land, there exists struct ifreq, which passed contains 
per-ioctl data passed to the net driver.  Recently, I marked the user 
pointer component with '__user',

struct ifreq
{
...
         union {
                 struct  sockaddr ifru_addr;
                 struct  sockaddr ifru_dstaddr;
                 struct  sockaddr ifru_broadaddr;
                 struct  sockaddr ifru_netmask;
                 struct  sockaddr ifru_hwaddr;
                 short   ifru_flags;
                 int     ifru_ivalue;
                 int     ifru_mtu;
                 struct  ifmap ifru_map;
                 char    ifru_slave[IFNAMSIZ];   /* Just fits the size */
                 char    ifru_newname[IFNAMSIZ];
                 char __user *   ifru_data;
                 struct  if_settings ifru_settings;
         } ifr_ifru;
};

However, Al noticed that network drivers were doing

	struct mii_ioctl_data *data =
		(struct mii_ioctl_data *) &ifr->ifr_data;

So, we have some code that will copy_to_user from the pointer stored in 
->ifr_data [aka ifr_ifru.ifru_data], and other code that uses the 
address of irf_data to indicate an anonymous BLOB of data stored in the 
union.

I'm a bit curious why sparse complained about taking the _address_ of 
pointer, but nonetheless it's an ugliness that should be contained 
before it spreads :)  Attached is a proposed cleanup patch for comment. 
  The ugliness is confined to include/linux/mii.h, and I avoid use the 
ifru_data field completely.

Untested, but compiles.

	Jeff