pasemi_mac: RX/TX ring management cleanup

Prepare a bit for supporting multiple TX queues by cleaning up some
of the ring management and shuffle things around a bit.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Turns out we're freeing the skb when we detect CRC error, but we're
not clearing out info->skb. We could either clear it and have the stack
reallocate it, or just leave it and the rx ring refill code will reuse
the one that was allocated.

Reusing a freed skb obviously caused some nasty crashes of various kind,
as reported by Brent Baude and David Woodhouse.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Signed-off-by: NDavid Woodhouse <dwmw2@infradead.org>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Make sure we don't feed packets with bad CRC up the network stack,
and discount the packet length as reported from the MAC for the CRC
field.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Don't use the "replace source address with local MAC address" bits, since
it causes problems on some variations of the hardware due to an erratum.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Add missing &:

drivers/net/pasemi_mac.c: In function 'pasemi_mac_clean_rx':
drivers/net/pasemi_mac.c:553: warning: passing argument 1 of 'prefetch'
makes pointer from integer without a cast
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: enable iommu support

Enable IOMMU support for pasemi_mac, but avoid using it on non-partitioned
systems for performance reasons.

The user can override this by selecting the PPC_PASEMI_IOMMU_DMA_FORCE
configuration option.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: use buffer index pointer in clean_rx()

Use the new features in B0 for buffer ring index on the receive side. This
means we no longer have to search in the ring for where the buffer
came from.

Also cleanup the RX cleaning side a little, while I was at it.

Note: Pre-B0 hardware is no longer supported, and needs a pile of other
workarounds that are not being submitted for mainline inclusion. So the
fact that this breaks old hardware is not a problem at this time.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: clear out old errors on interface open

Clear out any pending errors when an interface is brought up. Since the bits
are sticky, they might be from interface shutdown time after firmware has
used it, etc.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: update todo list

Remove some stale todo items that have been taken care of. Add a couple
of upcoming ones.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: further performance tweaks

Misc driver tweaks for pasemi_mac:
	* Increase ring size (really needed mostly on 10G)
	* Take out an unneeded barrier
	* Move around a few prefetches and reorder a few calls
	* Don't try to clean on full tx buffer, just let things
	  take their course and stop the queue directly
	* Avoid filling on the same line as the interface is
	  working on to reduce cache line bouncing
	* Avoid unneeded clearing of software state (and make the
	  interface shutdown code handle it)
	* Fix up some of the tx ring wrap logic.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: add local skb alignment

Add local SKB alignment to pasemi_mac, since ppc64 in general has it at 0
because of design flaws in some of the IBM server bridge chips. However,
for PWRficient doing the unaligned copies is more expensive than doing
unaligned DMA so make sure the data is aligned instead.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: workaround for erratum 5971

Implement workarounds for erratum 5971, where L2 hints aren't considered
properly unless the way hint is enabled on the interface. Since L2 isn't
setup to dedicate a way to headers, we need to reset the packet count
by hand so it won't run out of credits.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: implement sg support

Implement SG support for pasemi_mac
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: rework ring management

Rework ring management, switching to an opaque ring format instead of
the struct-based descriptor+pointer setup, since it will be needed for
SG support.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: fix bug in receive buffer dma mapping

skb->len isn't actually set to the size of the allocated skb, so don't
try to use it when figuring out how much to map.

(This hasn't surfaced as a real bug because we effectively disable
translation for the interface, but it still needs fixing for the future)
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: basic error checking

Add some rudimentary error checking to pasemi_mac.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: pass in count of buffers to replenish rx ring with

Refactor replenish_rx_ring to take an argument for how many entries to
fill. Since it's normally available from where it's called anyway, this
is just simpler. It also removes the awkward logic to try to figure out
if we're filling for the first time or not.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: don't enable rx before there are buffers on the ring

Reorder initialization of the DMA channels and the interface. Before there
was a time window when the interface was enabled before DMA was enabled.
Also, now there will always be RX buffers available at the time the
MAC interface is enabled, to avoid temporary out-of-buffer errors for the
very first packets (on busy networks).
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: flags as passed to spin_*_irqsave() should be unsigned long.
Signed-off-by: NTony Breeds <tony@bakeyournoodle.com>
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

pasemi_mac: set interface speed correctly on XAUI ports

Set interface speed for XAUI to 10G per default, not 1G.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

drivers/net/pasemi_mac.c: At top level:
drivers/net/pasemi_mac.c:89: warning: 'read_iob_reg' defined but not used
Signed-off-by: NSatyam Sharma <satyam@infradead.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is nicer than the MAC_FMT stuff.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We now have struct net_device_stats embedded in struct net_device,
and the default ->get_stats() hook does the obvious thing for us.

Run through drivers/net/* and remove the driver-local storage of
statistics, and driver-local ->get_stats() hook where applicable.

This was just the low-hanging fruit in drivers/net; plenty more drivers
remain to be updated.

[ Resolved conflicts with napi_struct changes and fix sunqe build
  regression... -DaveM ]
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It's been a useless no-op for long enough in 2.6 so I figured it's time to
remove it.  The number of people that could object because they're
maintaining unified 2.4 and 2.6 drivers is probably rather small.

[ Handled drivers added by netdev tree and some missed IRDA cases... -DaveM ]
Signed-off-by: NRalf Baechle <ralf@linux-mips.org>
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Unfortunately there's no timeout for how long a packet can sit on
the TX ring after completion before an interrupt is generated, and
we want to have a threshold that's larger than one packet per interrupt.

So we have to have a timer that occasionally cleans the TX ring even
though there hasn't been an interrupt. Instead of setting up a dedicated
timer for this, just clean it in the NAPI poll routine instead.

[ Resolved conflicts with napi_struct changes... -DaveM ]
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Enable LLTX on pasemi_mac: we're already doing sufficient locking
in the driver to enable it.

[ Resolved merge conflicts with napi_struct changes... -DaveM ]
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

RX side flag to use is CHECKSUM_UNNECESSARY, not CHECKSUM_COMPLETE.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

The old logic didn't detect full (tx) ring cases properly, causing
overruns and general badness. Clean it up a bit and abstract out the
ring size checks, always making sure to leave 1 slot open.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Postpone pci unmap and skb free of the transmitted buffers to outside of
the tx ring lock, batching them up 32 at a time.

Also increase the count threshold to 128.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Various RX performance tweaks, do some explicit prefetching of packet
data, etc.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Fix up memcpy for short receives.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Enable settings to target l2 for the first few cachelines of the packet, since
we'll access them to get to the various headers.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Move away from using the pci config access functions for simple register
access.  Our device has all of the registers in the config space (hey,
from the hardware point of view it looks reasonable :-), so we need to
somehow get to it. Newer firmwares have it in the device tree such that
we can just get it and ioremap it there (in case it ever moves in future
products). For now, provide a hardcoded fallback for older firmwares.

[ Resolved napi_struct conflicts... -DaveM ]
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Abstract out the PCI config read/write accesses into reg read/write ones,
still calling the pci accessors on the back end.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.

In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.

The signature of the ->poll() call back goes from:

	int foo_poll(struct net_device *dev, int *budget)

to

	int foo_poll(struct napi_struct *napi, int budget)

The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract).  The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.

The napi_struct is to be embedded in the device driver private data
structures.

Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler.  Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.

With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.

Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.

[ Ported to current tree and all drivers converted.  Integrated
  Stephen's follow-on kerneldoc additions, and restored poll_list
  handling to the old style to fix mutual exclusion issues.  -DaveM ]
Signed-off-by: NStephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It was mistakenly set to interrupt on the second packet instead of first, causing
some interesting latency behaviour.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Turns out we have an old version of firmware that stores the mac address
in 'mac-address' as a string instead of a byte array. All versions that
use local-mac-address should have it as byte array, so no need to do
string parsing for that case.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

This caused some very interesting behaviour depending on what happened to
be built at the same time. Add terminating empty entry to the list of IDs.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Interrupt ack fixes

Fix the packet count resets at interrupt time, using the cacheable
packet count status to set number of processed/received packets, since
the ack count is the cumulative number of packets processed, and not
incremental.
Signed-off-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NJeff Garzik <jeff@garzik.org>