Willem de Bruijn says:

====================
net: remove deprecated syststamp

The network stack can generate two kinds of hardware timestamps:
- hwtstamp stores a hw timestamp in device-specific raw format
- syststamp convers the raw format to system time

The second is deprecated and only implemented by a single device
driver. The suggested alternative is to communicate hwtstamp +
directly expose the NIC PTP clock device through ptp_clock_info.
The remaining driver (octeon) does not expose such a standard
interface as of now. It does have its own PTP library that depends
on its own shared memory PTP clock interface.

This patchset
1. reverts the syststamp code in the one driver (octeon)
2. reverts an unnecessary zero initialization in another (vxge)
3. modifies PF_PACKET to use syststamp is != 0 (because always == 0)
4. modifies SCM_TIMESTAMPING in the same way

For backwards compatibility, the interfaces are not removed.
Applications can still request SOF_TIMESTAMPING_SYS_HARDWARE. The
response field in scm_timestamping also remains. As was the case
for hardware/drivers that did not implement the feature, the
setsockopt succeeds, but the response field is always zero.
====================
Acked-by: NRichard Cochran <richardcochran@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The SO_TIMESTAMPING API defines three types of timestamps: software,
hardware in raw format (hwtstamp) and hardware converted to system
format (syststamp). The last has been deprecated in favor of combining
hwtstamp with a PTP clock driver. There are no active users in the
kernel.

The option was device driver dependent. If set, but without hardware
support, the correct behavior is to return zero in the relevant field
in the SCM_TIMESTAMPING ancillary message. Without device drivers
implementing the option, this field is effectively always zero.

Remove the internal plumbing to dissuage new drivers from implementing
the feature. Keep the SOF_TIMESTAMPING_SYS_HARDWARE flag, however, to
avoid breaking existing applications that request the timestamp.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

No device driver will ever return an skb_shared_info structure with
syststamp non-zero, so remove the branch that tests for this and
optionally marks the packet timestamp as TP_STATUS_TS_SYS_HARDWARE.

Do not remove the definition TP_STATUS_TS_SYS_HARDWARE, as processes
may refer to it.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This driver explicitly clears a field that is unused and about to be
removed. Remove the initialization.

All fields in skb_shared_info before dataref are cleared in
__alloc_skb, so the removal is safe even while syststamp exists.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Hardware timestamps can be exposed to userspace in raw hardware format
(hwtstamp) as well as converted to system time (syststamp). The second
variant is deprecated and only implemented by this driver.

The preferred method of hardware timestamp generation is to combine
hwtstamp with a device PTP clock. Octeon has its own PTP library
that relies on a shared memory interface to the PTP clock device.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

As per comment from David Miller, we try to make the buffer reassembly
function more resilient to user errors than it is today.

- We check that the "*buf" parameter always is set, since this is
  mandatory input.

- We ensure that *buf->next always is set to NULL before linking in
  the buffer, instead of relying of the caller to have done this.

- We ensure that the "tail" pointer in the head buffer's control
  block is initialized to NULL when the first fragment arrives.
Signed-off-by: NJon Maloy <jon.maloy@ericsson.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Merge tag 'master-2014-07-25' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-next

John W. Linville says:

====================
pull request: wireless-next 2014-07-25

Please pull this batch of updates intended for the 3.17 stream!

For the mac80211 bits, Johannes says:

"We have a lot of TDLS patches, among them a fix that should make hwsim
tests happy again. The rest, this time, is mostly small fixes."

For the Bluetooth bits, Gustavo says:

"Some more patches for 3.17. The most important change here is the move of
the 6lowpan code to net/6lowpan. It has been agreed with Davem that this
change will go through the bluetooth tree. The rest are mostly clean up and
fixes."

and,

"Here follows some more patches for 3.17. These are mostly fixes to what
we've sent to you before for next merge window."

For the iwlwifi bits, Emmanuel says:

"I have the usual amount of BT Coex stuff. Arik continues to work
on TDLS and Ariej contributes a few things for HS2.0. I added a few
more things to the firmware debugging infrastructure. Eran fixes a
small bug - pretty normal content."

And for the Atheros bits, Kalle says:

"For ath6kl me and Jessica added support for ar6004 hw3.0, our latest
version of ar6004.

For ath10k Janusz added a printout so that it's easier to check what
ath10k kconfig options are enabled. He also added a debugfs file to
configure maximum amsdu and ampdu values. Also we had few fixes as
usual."

On top of that is the usual large batch of various driver updates --
brcmfmac, mwifiex, the TI drivers, and wil6210 all get some action.
Rafał has also been very busy with b43 and related updates.

Also, I pulled the wireless tree into this in order to resolve a
merge conflict...

P.S.  The change to fs/compat_ioctl.c reflects a name change in a
Bluetooth header file...
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This test is reversed so the memory is always leaked.  It's better style
to remove the test anyway.

Fixes: 3e403a77 ('bonding: make it possible to have unlimited nested upper vlans')
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Acked-by: NVeaceslav Falico <vfalico@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Change formal parameter name to not shadow the global jiffies.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2014-07-25

This series contains updates to e1000e, ixgbe and ixgbevf.

Mark provides all the changes for ixgbe and ixgbevf.  Converts some udelay()
calls to the preferred usleep_range().  Fixes a spurious release of the
semaphore in several functions when there was a failure to acquire the
semaphore in the first place.  Fixes a X540 semaphore error where an
incorrect check was treating success as failure and vice-versa.  Fixed
ixgbe_write_mbx() error when it was being called and there was no
mbx->ops.write method defined, so no error code was returned.  The
corresponding read function would explicitly return an error in such a
case as do other functions.  Cleans up unused (dead) code by removing it.
Finally make return values more direct, eliminating some gotos and
otherwise unneeded conditionals, which allows the removal of some local
variables.

David provides all the changes for e1000e.  Fix CRC errors with jumbo
traffic for 82579, i217 and i218 client parts to increase the gap
between the read and write pointers in the transmit FIFO.  Added code
to check and respond to previously ignored return values from NVM
access functions.  Added support for EEE in Sx states and fixed EEE in
S5 with runtime PM enabled.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Nikolay Aleksandrov says:

====================
inet: frag: cleanup and update

The end goal of this patchset is to remove the LRU list and to move the
frag eviction to a work queue. It also does a couple of necessary cleanups
and fixes. Brief patch descriptions:
Patches 1 - 3 inclusive: necessary clean ups
Patch 4 moves the eviction from the softirqs to a workqueue.
Patch 5 removes the nqueues counter which was protected by the LRU lock
Patch 6 removes the, by now unused, lru list.
Patch 7 moves the rebuild timer to the workqueue and schedules the rebuilds
        only if we've hit the maximum queue length on some of the chains.
Patch 8 migrate the rwlock to a seqlock since the rehash is usually a rare
        operation.
Patch 9 introduces an artificial global memory limit based on the value of
        init_net's high_thresh which is used to cap the high_thresh of the
        other namespaces. Also introduces some sane limits on the other
        tunables, and makes it impossible to have low_thresh > high_thresh.

Here are some numbers from running netperf before and after the patchset:
Each test consists of the following setting: -I 95,5 -i 15,10

1. Bound test (-T 4,4)
1.1 Virtio before the patchset -
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.122.177 () port 0 AF_INET : +/-2.500% @ 95% conf.  : cpu bind
Socket  Message  Elapsed      Messages                   CPU      Service
Size    Size     Time         Okay Errors   Throughput   Util     Demand
bytes   bytes    secs            #      #   10^6bits/sec % SS     us/KB

212992   64000   30.00      722177      0    12325.1     34.55    2.025
212992           30.00      368020            6280.9     34.05    0.752

1.2 Virtio after the patchset -
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.122.177 () port 0 AF_INET : +/-2.500% @ 95% conf.  : cpu bind
Socket  Message  Elapsed      Messages                   CPU      Service
Size    Size     Time         Okay Errors   Throughput   Util     Demand
bytes   bytes    secs            #      #   10^6bits/sec % SS     us/KB

212992   64000   30.00      727030      0    12407.9     35.45    1.876
212992           30.00      505405            8625.5     34.92    0.693

2. Virtio unbound test
2.1 Before the patchset
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.122.177 () port 0 AF_INET : +/-2.500% @ 95% conf.
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   64000   30.00      730008      0    12458.77
212992           30.00      416721           7112.02

2.2 After the patchset
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.122.177 () port 0 AF_INET : +/-2.500% @ 95% conf.
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   64000   30.00      731129      0    12477.89
212992           30.00      487707           8323.50

3. 10 gig unbound tests
3.1 Before the patchset
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.133.1 () port 0 AF_INET : +/-2.500% @ 95% conf.
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   64000   30.00      417209      0    7120.33
212992           30.00      416740           7112.33

3.2 After the patchset
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.133.1 () port 0 AF_INET : +/-2.500% @ 95% conf.
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992   64000   30.00      438009      0    7475.33
212992           30.00      437630           7468.87

Given the options each netperf ran between 10 and 15 times for 30 seconds
to get the necessary confidence, also the tests themselves ran 3 times and
were consistent.
Another set of tests that I ran were parallel stress tests which consisted
of flooding the machine with fragmented packets from different sources with
frag timeout set to 0 (so there're lots of timeouts) and low_thresh set to
1 byte (so evictions are happening all the time) and on top of that running
a namespace create/destroy endless loop with network interfaces and
addresses that got flooded (for the brief periods they were up) in parallel.
This test ran for an hour without any issues.
====================

This patch makes init_net's high_thresh limit to be the maximum for all
namespaces, thus introducing a global memory limit threshold equal to the
sum of the individual high_thresh limits which are capped.
It also introduces some sane minimums for low_thresh as it shouldn't be
able to drop below 0 (or > high_thresh in the unsigned case), and
overall low_thresh should not ever be above high_thresh, so we make the
following relations for a namespace:
init_net:
 high_thresh - max(not capped), min(init_net low_thresh)
 low_thresh - max(init_net high_thresh), min (0)

all other namespaces:
 high_thresh = max(init_net high_thresh), min(namespace's low_thresh)
 low_thresh = max(namespace's high_thresh), min(0)

The major issue with having low_thresh > high_thresh is that we'll
schedule eviction but never evict anything and thus rely only on the
timers.
Signed-off-by: NNikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

rehash is rare operation, don't force readers to take
the read-side rwlock.

Instead, we only have to detect the (rare) case where
the secret was altered while we are trying to insert
a new inetfrag queue into the table.

If it was changed, drop the bucket lock and recompute
the hash to get the 'new' chain bucket that we have to
insert into.

Joint work with Nikolay Aleksandrov.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NNikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

merge functionality into the eviction workqueue.

Instead of rebuilding every n seconds, take advantage of the upper
hash chain length limit.

If we hit it, mark table for rebuild and schedule workqueue.
To prevent frequent rebuilds when we're completely overloaded,
don't rebuild more than once every 5 seconds.

ipfrag_secret_interval sysctl is now obsolete and has been marked as
deprecated, it still can be changed so scripts won't be broken but it
won't have any effect. A comment is left above each unused secret_timer
variable to avoid confusion.

Joint work with Nikolay Aleksandrov.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NNikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

no longer used.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The 'nqueues' counter is protected by the lru list lock,
once thats removed this needs to be converted to atomic
counter.  Given this isn't used for anything except for
reporting it to userspace via /proc, just remove it.

We still report the memory currently used by fragment
reassembly queues.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the high_thresh limit is reached we try to toss the 'oldest'
incomplete fragment queues until memory limits are below the low_thresh
value.  This happens in softirq/packet processing context.

This has two drawbacks:

1) processors might evict a queue that was about to be completed
by another cpu, because they will compete wrt. resource usage and
resource reclaim.

2) LRU list maintenance is expensive.

But when constantly overloaded, even the 'least recently used' element is
recent, so removing 'lru' queue first is not 'fairer' than removing any
other fragment queue.

This moves eviction out of the fast path:

When the low threshold is reached, a work queue is scheduled
which then iterates over the table and removes the queues that exceed
the memory limits of the namespace. It sets a new flag called
INET_FRAG_EVICTED on the evicted queues so the proper counters will get
incremented when the queue is forcefully expired.

When the high threshold is reached, no more fragment queues are
created until we're below the limit again.

The LRU list is now unused and will be removed in a followup patch.

Joint work with Nikolay Aleksandrov.
Suggested-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NNikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

First step to move eviction handling into a work queue.

We lose two spots that accounted evicted fragments in MIB counters.

Accounting will be restored since the upcoming work-queue evictor
invokes the frag queue timer callbacks instead.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

hide actual hash size from individual users: The _find
function will now fold the given hash value into the required range.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adding a function, and associated calls, to flush writes to (read) the LPIC
MAC register before entering the shutdown flow.  This fixes the problem
of the PHY never negotiating a 100M link (if both sides of the link support
EEE and 100M link) when Runtime PM is enabled.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Tested-by: NAaron Brown <aaron.f.brown@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The process of shutting down the system causes a call to the close PM
callback.  The reset in close causes a loss of link, and the resultant
LSC interrupt causes the Runtime PM idle callback to be called.  The
check for link (while link is down) in the idle callback is wiping the
information about the EEE ability of the link partner.  The information is
still gone when the PHY is powered back up in the shutdown flow.  This
causes EEE in S5 to fail when Runtime PM is active.

Save the link partner's EEE ability in the idle callback so that a Runtime
PM event will not cause a loss of this information.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Tested-by: NAaron Brown <aaron.f.brown@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

On I217 and newer hardware, EEE is enabled in the PHY by the software
when link is up and disabled by the hardware when link is lost.

To enable EEE in Sx (When both ends of the link support, and are enabled
for, EEE and 100Mbps), we need to disable LPLU and configure the PHY to
automatically enable EEE when link is up, since there will be no software
to complete the task.

To configure this in the PHY, the Auto Enable LPI bit in the Low Power
Idle GPIO Control register must be set.  For normal operation in S0, this
bit must be cleared.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Tested-by: NAaron Brown <aaron.f.brown@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Adding code to check and respond to previously ignored return values
from NVM access functions.

Issue discovered through static analysis.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Tested-by: NAaron Brown <aaron.f.brown@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Modifying the jumbo frame workaround for 82579, i217 and i218 client parts
to increase the gap between the read and write pointers in the Tx FIFO.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Tested-by: NAaron Brown <aaron.f.brown@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Make return values more direct, eliminating some gotos and
otherwise unneeded conditionals. This also eliminates some
local variables. Also a few minor cleanups in affected code
so checkpatch won't complain.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

All of the code involved with returning the supported physical
layer is actually unused, so delete it.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

If ixgbe_write_mbx is called and no mbx->ops.write method
exists, no error code is returned. The corresponding read
function explicitly returns an error in such a case as do
other functions, so this appears to be a minor bug. Fix
it for consistency, and generate return values directly
to make things clearer.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

In the function ixgbe_get_swfw_sync_semaphore, an incorrect
check was treating success as failure and vice-versa. This led
to manipulating the IXGBE_SWFW_SYNC register without holding
the software semaphore first, which is an error. In addition,
if getting the REGSMP bit in the IXGBE_SW_FW_SYNC register
timed out, no error code would be returned, making the caller
think that it had successfully acquired the lock. Fix both of
those issues and clean up the function a bit, such as make the
name in the comment match the function.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Failure to acquire the semaphore would lead to a spurious release
of the semaphore in several functions. Do not release a semaphore
that you did not get.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Convert some udelay calls to the preferred usleep_range.
Signed-off-by: NMark Rustad <mark.d.rustad@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Conflicts:
	net/mac80211/cfg.c
Signed-off-by: NJohn W. Linville <linville@tuxdriver.com>

After 11878b40[net-timestamp: SOCK_RAW and PING timestamping], this comment
becomes obsolete since the codes check not only UDP socket, but also RAW sock;
and the codes are clear, not need the comments
Signed-off-by: NLi RongQing <roy.qing.li@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Cyrille Pitchen says:

====================
net/macb: add HW features to macb driver

this series of patches adds new hardware features to macb driver. These
features can be enabled/disabled at runtime using ethtool. Depending on
hardware and design configuration, some are enabled by default whereas other
are disabled.

For instance, checksum offload features are enabled by default for gem designed
for packet buffer mode but disabled for fifo mode design or for old macb.

Besides, the scatter-gather feature is enabled and tested on macb but disabled
on sama5d3x gem. When testing this feature on sama5d3x gem, TX lockups occured
frequently.

Also, the RX checksum offload feature is enabled at GEM level only when both
IFF_PROMISC bit is clear in dev->flags and NETIF_F_RXCSUM bit is set in
dev->features.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

this new compatibility string prevents macb/gem driver from using the
scatter-gather and gso features on sama5d3x boards.
Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When RX checksum offload is enabled at GEM level (bit 24 set in the Network
Control Register), frames with invalid IP, TCP or UDP checksums are
discarted even if promiscuous mode is enabled (bit 4 set in the Network Control
Register).

This was verified with a simple userspace program, which corrupts UDP checksum
using libnetfilter_queue.

Then both IFF_PROMISC bit must be clear in dev->flags and NETIF_F_RXCSUM bit
must be set in dev->features to enable RX checksum offload at GEM level. This
way tcpdump is still able to capture corrupted frames.

Also skb->ip_summed is set to CHECKSUM_UNNECESSARY only when both TCP/IP or
UDP/IP checksums were verified by the GEM. Indeed the GEM may verify only IP
checksum but not the one for ICMP (or other protocol than TCP or UDP).
Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The scatter-gather feature will allow to enable the Generic Segmentation Offload.
Generic Segmentation Offload can be enabled/disabled using ethtool -K DEVNAME gso on|off.

e.g:
ethtool -K eth0 gso off

When enabled, the driver may be provided with socket buffers splitted into many fragments.
These fragments need to be queued into the TX ring in reverse order, starting from to the
last one down to the first one, to avoid a race condition with the MAC.
Especially the 'TX_USED' bit in word 1 of the transmit buffer descriptor of the
first fragment should be cleared at the very final step of the queueing algorithm.
This will tell the hardware that fragments are ready to be sent.

Also since the MAC only update the status word of the first buffer descriptor of the
ethernet frame, the queueing algorithm can no longer expect a 'TX_USED' bit to be set by
the MAC into the buffer descriptor following the one for last fragment of the skb.
This is why the driver sets the 'TX_USED' bit before queueing any fragment, so the end of
queue position is well defined for the MAC.
Signed-off-by: NCyrille Pitchen <cyrille.pitchen@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>