We correct this by handling the requeue directly on
the ath_rx_tasklet() and trying to allocate an skb
*prior* to sending up the last hardware processed
skb. If we run out of memory this gauranteees we have
skbs to work with while it simply drops new received
frames.
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com> Signed-off-by: John W. Linville <linville@tuxdriver.com>