duplicated host_no

Cool bug. Two SCSI devs with the same host_id.

This is with fcoe and a bunch of patches in development.

I ran a loop including creating and deleting a bunch of fcoe instances on an MP system. After 140 some times, I see this:

[ 7250.276749] scsi148 : FCoE Driver
[ 7250.276752] device eth0 entered promiscuous mode
[ 7250.276784] scsi148 : FCoE Driver
[ 7250.277526] host148: lport ffffffff: Entered RESET state from none state
[ 7250.277530] libfc: Link up on port (     0)
[ 7250.277532] host148: lport      0: Entered FLOGI state from reset state
[ 7250.305961] ------------[ cut here ]------------
[ 7250.310653] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xc7/0xde()
[ 7250.317096] Hardware name: X7DB8
[ 7250.320372] sysfs: cannot create duplicate filename '/bus/scsi/devices/host148'

This is unlikely to show up in real life, but it can happen in stress tests like this, and the system doesn't work so well afterwards.

I'll submit a patch that uses atomic_inc_return for the host numbers.

Joe