struct ifnet or the layer 2 common structure it was embedded in have
been replaced with a struct ifnet pointer to be filled by a call to the
new function, if_alloc(). The layer 2 common structure is also allocated
via if_alloc() based on the interface type. It is hung off the new
struct ifnet member, if_l2com.
This change removes the size of these structures from the kernel ABI and
will allow us to better manage them as interfaces come and go.
Other changes of note:
- Struct arpcom is no longer referenced in normal interface code.
Instead the Ethernet address is accessed via the IFP2ENADDR() macro.
To enforce this ac_enaddr has been renamed to _ac_enaddr.
- The second argument to ether_ifattach is now always the mac address
from driver private storage rather than sometimes being ac_enaddr.
Reviewed by: sobomax, sam
status registers for error conditions and updating statistics
when there are cycles left (inspired by the nge(4) driver).
- Removed the TX list counter and the producer/consumer gap; it's
enough to just ensure we don't reuse the last (free) descriptor,
as the chip may not have read its next pointer yet. If we reuse
it, the TX may stall under a heavy TX load with polling enabled.
- Dropped code to recharge the watchdog timer, it's pointless; the
watchdog routine will re-init the chip and both RX and TX lists.
Moved the RX ring resyncing code to ste_rxeoc(), and only run it
if we were asked to POLL_AND_CHECK_STATUS, under DEVICE_POLLING.
(This significantly reduces the CPU load.)
Improved the RX ring resyncing code by re-checking if the head
is still empty before doing resyncing. This mostly affects the
DEVICE_POLLING mode, where we run this code periodically. We
could start checking with an empty head (well, an empty ring
even), and after doing a few iterations, the chip might write
a few entries, including the head, and we would bogusly consider
this case as requiring resyncing. On a test box, this reduced
the number of resyncs done by a factor of 10.
In ste_txeof(sc), only reset the watchdog timer to zero when
the TX list is completely empty.
Converted ste_tx_prev_idx to a pointer -- faster.
Removed some bitrot.
the packets are immediately returned for sending (e.g. when bridging
or packet forwarding). There are more efficient ways to do this
but for now use the least intrusive approach.
Reviewed by: imp, rwatson
This is pretty much fixes any issue I can find:
- Watchdog timeouts were due to starting the TX DMA engine
before we had a packet ready for it. So the first packet
sent never got out only if we sent more then one packet
at a time did the others make it out and not blow up.
Of course reseting the chip then caused us not to transmit
the first packet again ie. catch-22. This required logic changes.
- Combine interrupts on TX packets being queued up.
- Don't keep running around the RX ring since we might get
out of sync so only go around once per receive
- Let the RX engine recover via the poll interface which is
similar to the TX interface. This way the chip wakes
up with no effort when we read enough packets.
- Do better hand-shaking on RX & TX packets so they don't
start of to soon.
- Force a duplex setting when the link comes up after
an ste_init or it will default to half-duplex and be
really slow. This only happens on subsequent ste_init.
The first one worked.
- Don't call stat_update for every overflow. We only monitor
the collisions so the tick interval is good enough for that.
Just read in the collision stats to minimize bus reads.
- Don't read the miibus every tick since it uses delays and
delays are not good for performance.
- Tie link events directly to the miibus code so the port
gets set correctly if someone changes the port settings.
- Reduce the extreme number of {R,T}FD's. They would consume
130K of kernel memory for each NIC.
- Set the TX_THRESH to wait for the DMA engine to complete
before running the TX FIFO. This hurts peak TX performance
but under bi-directional load the DMA engine can't keep up
with the FIFO. Testing shows that we end up in the case
anyways (a la dc(4) issues but worse since the RX engine hogs
everything).
- When stopping the card do a reset since the reset verifies the
card has stopped. Otherwise on heavy RX load the RX DMA engine
is still stuffing packets into memory. If that happens after
we free the DMA area memory bits get scribled in memory and
bad things happen.
This card still has seemingly unfixable issues under heavy RX load in
which the card takes over the PCI bus.
Sponsored by: Vernier Networks
MFC after: 1 week
mtx_enter(lock, type) becomes:
mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks)
mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized)
similarily, for releasing a lock, we now have:
mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN.
We change the caller interface for the two different types of locks
because the semantics are entirely different for each case, and this
makes it explicitly clear and, at the same time, it rids us of the
extra `type' argument.
The enter->lock and exit->unlock change has been made with the idea
that we're "locking data" and not "entering locked code" in mind.
Further, remove all additional "flags" previously passed to the
lock acquire/release routines with the exception of two:
MTX_QUIET and MTX_NOSWITCH
The functionality of these flags is preserved and they can be passed
to the lock/unlock routines by calling the corresponding wrappers:
mtx_{lock, unlock}_flags(lock, flag(s)) and
mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN
locks, respectively.
Re-inline some lock acq/rel code; in the sleep lock case, we only
inline the _obtain_lock()s in order to ensure that the inlined code
fits into a cache line. In the spin lock case, we inline recursion and
actually only perform a function call if we need to spin. This change
has been made with the idea that we generally tend to avoid spin locks
and that also the spin locks that we do have and are heavily used
(i.e. sched_lock) do recurse, and therefore in an effort to reduce
function call overhead for some architectures (such as alpha), we
inline recursion for this case.
Create a new malloc type for the witness code and retire from using
the M_DEV type. The new type is called M_WITNESS and is only declared
if WITNESS is enabled.
Begin cleaning up some machdep/mutex.h code - specifically updated the
"optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN
and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently
need those.
Finally, caught up to the interface changes in all sys code.
Contributors: jake, jhb, jasone (in no particular order)
- Convert to using TX descritor polling similar to the xl driver (the
ST201 is a clone of the 3c90xB chipset and offers the same transmit
polling scheme). This should reduce TX overhad a little.
- Make sure to reset PHY when switching mode, as in the starfire driver.
- Fix instances of free() that should be contigfree().
- Remove dead code.
PCI fast ethernet controller. Currently, the only card I know that uses
this chip is the D-Link DFE-550TX. (Don't ask me where to buy these: the
only cards I have are samples sent to me by D-Link.)
This driver is the first to make use of the miibus code once I'm sure
it all works together nicely, I'll start converting the other drivers.
The Sundance chip is a clone of the 3Com 3c90x Etherlink XL design
only with its own register layout. Support is provided for ifmedia,
hardware multicast filtering, bridging and promiscuous mode.
