reproductible hang with if_em

[Pierre Beyssac]

Hello,

I'm having a problem (under 5.3-BETA, but I don't think it matters
much) with the if_em driver on bursty outbound traffic: the card
is stuck in a state where it only returns ENOBUFS.  No buffers ever
become available again, but ifconfig down+up fixes the problem.

>From a quick analysis it looks like the TX descriptor cleanup routine
em_clean_transmit_interrupts() as called by em_encap() in such cases
doesn't manage to find anything free with E1000_TXD_STAT_DD set.
The card seems stuck in this state. A simple albeit dirty fix would
be to cancel some or all untransmitted packets in such cases.

Perhaps this is related to my using the card on a 100Mbps switch.
I have a small program to exercise the hang in case someone would
like to have a look at it.

See below for the result of "sysctl hw.em0.debug_info=1" (notice
the not avail1 = 12 and not avail2 = 0 hinting at
em_clean_transmit_interrupts).

em0: [MPSAFE]
em0: bpf attached
em0: Ethernet address: 00:0d:56:d0:5c:15
em0:  Speed:N/A  Duplex:N/A
em0: Link is up 100 Mbps Full Duplex
em0: Adapter hardware address = 0xc3570b34 
em0:tx_int_delay = 66, tx_abs_int_delay = 66
em0:rx_int_delay = 0, rx_abs_int_delay = 66
em0: fifo workaround = 0, fifo_reset = 0
em0: hw tdh = 238, hw tdt = 238
em0: Num Tx descriptors avail = 256
em0: Tx Descriptors not avail1 = 12
em0: Tx Descriptors not avail2 = 0
em0: Std mbuf failed = 0
em0: Std mbuf cluster failed = 0
em0: Driver dropped packets = 0

-- 
Pierre Beyssac						pb at enst.fr