Lock busdma operations and serialize detach with open/close

[marcel]

Summary

admbugs: 782

Test Plan

% sudo python2 /usr/src/tools/bus_space/examples/am79c900_diag.py
DEBUG:root:pcicfg=0 (pci0:2:1:0)
DEBUG:root:io=0 (pci0:2:1:0)
INFO:root:ethernet address = 00:0c:29:d9:ef:0d
DEBUG:root:DMA memory: size = 0x8000 (TX buffers: 4, RX buffers: 16)
DEBUG:root:DMA memory: CPU address: 0x8013cc000, device address: 0x133f6000
DEBUG:root:Waiting for initialization to complete
INFO:root:DHCP discovery packet sent
DEBUG:root:Waiting 2 seconds for an offer to be received
DEBUG:root:RX #0: broadcast packet: length 66
DEBUG:root:RX #1: broadcast packet: length 64
DEBUG:root:RX 2: packet from 00:50:56:e8:00:8e!
INFO:root:Our IP address = 172.16.188.134
DEBUG:root:RX #3: broadcast packet: length 64
DEBUG:root:RX #4: broadcast packet: length 64

[markj]

sys/dev/proto/proto.h

40

The two values of a signed 1-bit bitfield are 0 and -1. I don't think that causes any problems in this diff, but it's probably better to use unsigned fields just in case.

[markj]

This looks ok to me modulo the one lingering WAITOK allocation under the busdma mutex. I don't see any reason not to use an sx lock for that though, in which case you could avoid the refactoring needed to do preallocation.

sys/dev/proto/proto_busdma.c

190

This is a WAITOK allocation unless you explicitly specify BUS_DMA_NOWAIT.

[marcel]

Use sx instead of mtx to allow M_WAITOK allocations

[marcel]

sys/dev/proto/proto_busdma.c

190

Good catch, thanks!

[markj]

Committed in r349714.