Update ifp->if_opackets correctly

- Fix one typo (instead of two disable_rx, one disable_rx and one disable_tx)
- Instead of setting OACTIVE on the interface when encap fails and never
clearing it, set a 2 second timer instead an

- Fix one typo (instead of two disable_rx, one disable_rx and one disable_tx)
- Instead of setting OACTIVE on the interface when encap fails and never
clearing it, set a 2 second timer instead and hope that the memory
situation will improve. This stops the lockups of the interface when
memory is running low.
jason@ ok.

show more ...

shorten address printout

spelling; from Brian Poole <raj@cerias.purdue.edu>

make gem compile, arpcom is sc_arpcom.

a step towards consistancy; in general:
'struct arpcom foo' -> 'struct arpcom sc_arpcom'
ok itojun@

Multicast fixes, Gig-E support, mostly from netbsd. ok jason@

Simplify multicast handling based on powerpc's gm.c and remove ether_cmp()

- read the rx completion register exactly once.
- don't complain about still owned descriptors
tested by myself and Joey Coleman <joeycoleman@acm.org>

- Move reading collision counters out of splimp()
- all non-debug prints should print the interface name
- quite whining about RX counter expiring, who cares.

First round of __P removal in sys

Call ether_ioctl() the way [insert favorite diety] intended.

Revert to previous (non hacked =) version now that the iommu is handled
correctly and it doesn't crash

Re-port the NetBSD driver, but rewrite tx logic to use static buffers (and
copies). This makes the Blade100 gem stable for me.

In encap, sync mbuf before writing descriptors

changes to work with 32bit platforms and get ethernet address on powerpc.
Inspired by NetBSD.

Replace inclusion of <vm/foo.h> with the correct <uvm/bar.h> when necessary.
(Look ma, I might have broken the tree)

Rewrite transmit handling (looks something more like hme), also takes advantage
of some register shadowing provided by the asic.
[This still isn't optimal, but it's getting closer...]
Move all of the

Rewrite transmit handling (looks something more like hme), also takes advantage
of some register shadowing provided by the asic.
[This still isn't optimal, but it's getting closer...]
Move all of the stat stuff out of the interrupt context and into the timeout.

show more ...

increase rx descriptors, remove some debugging code, deal with different endian; NetBSD
general cleanup; me.

Some KNF.

Fix a DIAGNOSTIC check.

Use splimp instead of splnet in some places.
I don't understand why, but it helps stability a lot.

solve mtu problem a little more elegantly (ramdisk works on blade100 and
u5 now).

configured mtu in BOTH places (grumble)
Also, make it a bit higher than normal (this is an ugly hack and will be
removed when I figure out what this thing wants).

openbsd style %b handling