It never worked there. The ISR was never written for that kind of
stuff. So remove all that crap with a hash of linked lists and pass
the pointer directly to the ISR.

BTW this answers the question there:
  * I don't know exactly why they don't use the dev_id opaque data
  * pointer instead of this extra lookup table

-> Because they thought they will support more devices bound to a
single interrupt w/o IRQF_SHARED. They would need exactly the hash
there.

What I don't understand is rebinding of the interrupt in the shutdown
path. They perhaps meant to do just synchronize_irq? In any case, this
is all gone and free_irq there properly.

By removing the hash we save some bits (exactly NR_IRQS * 8 bytes of
.bss and over a kilo of .text):
before:
   text    data     bss     dec     hex filename
  19600     320    8227   28147    6df3 ../a/ia64/arch/ia64/hp/sim/simserial.o
after:
   text    data     bss     dec     hex filename
  18568     320      28   18916    49e4 ../a/ia64/arch/ia64/hp/sim/simserial.o

Note that a shared interrupt could not work too. request_irq requires
data parameter to be non-NULL. So the whole IRQ_T exercise was
pointless.

Finally, this helps us remove another two members of async_struct :).
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>