Where a PTP clock driver is associated with a net or PHY driver, it
should be enabled automatically whenever that driver is enabled.
Therefore:

- Make PTP clock drivers select rather than depending on PTP_1588_CLOCK
- Remove separate boolean options for PTP clock drivers that are built
  as part of net driver modules.  (This also fixes cases where the PTP
  subsystem is wrongly forced to be built-in.)
- Set 'default y' for PTP clock drivers that depend on specific net
  drivers but are built separately
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Calling the ixgbe_reset_pipeline_82599 function will ensure a full pipeline
reset on all 82599 devices.  This is necessary to avoid possible link issues.
Since this patch accomplishes this by modifying AUTOC.LMS we need to wrap
all AUTOC writes when LESM is enabled.

v2- fix LMS behaviour based on feedback by Martin Josefsson

CC: Martin Josefsson <gandalf@mjufs.se>
Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch modifies when and where PTP registers and data are set. Previously
a work-around was used inside cyclecounter_start in order to reset some of the
time registers. This patch creates a new ixgbe_ptp_reset specifically for this
purpose. The cyclecounter configuration has trimmed down to only modify what
is necessary. Due to hardware conditions after probe and before open, PTP init
has now moved into the ixgbe_open call. This allows the ptp device name in the
sysfs to be the ethernet device name instead of the MAC address.

The cyclecounter check flag is renamed to PTP_ENABLED and is used to prevent
PTP init from happening when PTP has not been enabled.

CC: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

It is necessary to track the default user priority in the PF so that we can
force it upon the VFs.  The motivation behind this is to keep the VFs from
getting access to user priorities meant for things like storage.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change allows us to add a mailbox versioning API.  This will allow us
to determine the features supported by the VFs from the PF.  For example we
will be implementing a version 1.1 API for the VF that will indicate that
it can support us enabling Jumbo frames as the VF will support buffer
chaining.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NRobert Garrett <RobertX.Garrett@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change limits the PF/VF driver to 9.5K max jumbo frame size in order
prevent a possible Tx hang in the adapter when sending frames between
pools.

All of the parts in ixgbe support a maximum frame of 15.5K for standard
traffic, however with SR-IOV or DCB enabled they should be limiting the
MTU size to 9.5K.  Instead of adding extra checks which would have to
change the MTU when we go into or out of these modes it is preferred to
just use a standard 9.5K MTU limit for all modes so that this extra
overhead can be avoided.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NSibai Li <sibai.li@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds debugfs support to the ixgbe driver to give
users the ability to access kernel information and to
simulate kernel events.

The filesystem is set up in the following driver/PCI-instance
hierarchy:
<debugfs>
   |-- ixgbe
	|-- PCI instance
	|	|-- attribute files
Signed-off-by: NCatherine Sullivan <catherine.sullivan@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change updates the code related to configuring the transmit frame
checksum.  Specifically I have updated the code so that we can only skip
inserting the checksum in the case that we are not performing some other
offload that will modify the frame data.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NPeter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>

This change makes it so that we do not use double buffering if the page
size is larger than 4K.  Instead we will simply walk through the page using
up to 3K per receive, and if we receive less than we only move the offset
by that amount.  We will free the page when there is no longer any space
left that we can use instead of checking the page count to see if we can
cycle back to the start.

The main motivation behind this is to avoid the unnecessary truesize cost
for using a half page when most packets are 2K or smaller. With this new
approach the largest possible truesize for a page fragment will be 3K when
PAGE_SIZE is larger than 4K.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NPeter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>

The recent changes to netdev_alloc_skb actually make it so that the size of
the buffer now actually has a more direct input on the truesize.  So in
order to make best use of the piece of a page we are allocated I am
reducing the IXGBE_RX_HDR_SIZE to 256 so that our truesize will be reduced
by 256 bytes as well.

This should result in performance improvements since the number of uses per
page should increase from 4 to 6 in the case of a 4K page.  In addition we
should see socket performance improvements due to the truesize dropping
to less than 1K for buffers less than 256 bytes.

Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch does two things.  First it drops the unnecessary work of
searching for enabled VFs when we first bring up the adapter and instead
just uses pci_num_vf to determine how many VFs are enabled on the adapter.

The second thing it does is drop the use of vfdev from the vf_data_storage
structure.  Instead we just search the entire system for a VF that has us
as it's PF, and then if that VF is assigned we indicate that the VFs are
assigned.  This allows us to still check for assigned VFs even if the
vfinfo allocation has failed, or vfinfo has been freed.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Acked-by: NGreg Rose <gregory.v.rose@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NSibai Li <sibai.li@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch changes the behavior of the FCoE configuration so that it is
much closer to how the main body of the ixgbe driver works for ring
allocation.

The first piece is the ixgbe_fcoe_ddp_enable/disable calls.  These allocate
the percpu values and if successful set the fcoe_ddp_xid value indicating
that we can support DDP.

The next piece is the ixgbe_setup/free_ddp_resources calls.  These are
called on open/close and will allocate and free the DMA pools.

Finally ixgbe_configure_fcoe is now just register configuration.  It can go
through and enable the registers for the FCoE redirection offload, and FIP
configuration without any interference from the DDP pool allocation.

The net result of all this is two fold.  First it adds a certain amount of
exception handling.  So for example if ixgbe_setup_fcoe_resources fails we
will actually generate an error in open and refuse to bring up the
interface.

Secondly it provides a much more graceful failure case than the previous
model which would skip setting up the registers for FCoE on failure to
allocate DDP resources leaving no Rx functionality enabled instead of just
disabling DDP.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change makes it so that we can use the VMDq ring feature offset value
to determine the default pool instead of using num_vfs.  The reason for
this change is to avoid issues should we fail to allocate vfinfo but have
pre-existing VFs.  What should happen in this case is that num_vfs will go
to 0, but the VMDq offset will contain the location of the first PF pool.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NSibai Li <Sibai.li@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change is just meant to defragment the flags as there are several hole
that have been introduced since several features, or the flags for them,
have been removed.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

All of our hardware supports RSS even if it is only for a single queue.  So
instead of toting around the RSS enable flag I am updating the code so that
all devices are enabled and if we want to disable RSS it is indicated via
the RSS mask.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change essentially makes it so that we can enable almost all of the
features all at once.  This patch allows for the combination of SR-IOV,
DCB, and FCoE in the case of the x540.  It also beefs up the SR-IOV by
adding support for RSS to the PF.

The testing matrix gets to be very complex for this patch as there are a
number of different features and subsets for queueing options.  I tried to
narrow these down a bit by restricting the PF to only supporting 4TC DCB
when it is enabled in addition to SR-IOV.

Cc: Greg Rose <gregory.v.rose@intel.com>
Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

In upcoming patches it will become increasingly common to need to determine
the FCoE traffic class in order to determine the correct queues for FCoE.
In order to make this easier I am adding a function for obtaining the FCoE
traffic class based on the user priority.

Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The mask value for ring features was overloaded for FCoE which can lead to
some confusion.  In order to avoid any confusion I am splitting the mask
value and adding an offset value.  This can be used for the start of the
FCoE rings, and in the future I hope to use it to store the start of the
registers for SR-IOV.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

We are currently using indices to indicate the upper limit on a ring
feature.  However since we can switch back and forth on features such as
DCB and that has effects on other features such as RSS it is preferable to
instead store the upper limit separate from the current value for the
number of rings related to the feature.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

It makes much more sense for us to count q_vectors instead of MSI-X
vectors.  We were using num_msix_vectors to find the number of q_vectors in
multiple places.  This was wasteful since we only had one place that
actually needs the number of MSI-X vectors and that is in slow path.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

FCoE target mode was experiencing issues due to the fact that we were
sending up data frames that were padded to 60 bytes after the DDP logic had
already stripped the frame down to 52 or 56 depending on the use of VLANs.
This was resulting in the FCoE DDP logic having issues since it thought the
frame still had data in it due to the padding.

To resolve this, adding code so that we do not pad FCoE frames prior to
handling them to the stack.

CC: <stable@vger.kernel.org>
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch fixes a potential Rx timestamp deadlock that causes the Rx
timestamping to stall indefinitely. The issue could occur when a PTP packet is
timestamped by hardware but never reaches the Rx queue. In order to prevent a
permanent loss of timestamping, the RXSTMP(L/H) registers have to be read to
unlock them. (This used to only occur when a packet that was timestamped
reached the software.) However the registers can't be read early otherwise
there is no way to correlate them to the packet.

This patch introduces a filter function which can be used to determine if a
packet should have been timestamped. Supplied with the filter setup by the
hwtstamp ioctl, check to make sure the PTP protocol and message type match the
expected values. If so, then read the timestamp registers (to free them.) At
this point check the descriptor bit, if the bit is set then we know this
packet correlates to the timestamp stored in the RXTSTAMP registers.
Otherwise, assume that packet was dropped by the hardware, and ignore this
timestamp value. However, we have at least unlocked the rxtstamp registers for
future timestamping.

Due to the way the driver handles skb data, it cannot be directly accessed. In
order to work around this, a copy of the skb data into a linear buffer is
made. From this buffer it becomes possible to read the data correctly
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Reviewed-by: NRichard Cochran <richardcochran@gmail.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

When the hwmon code was initially added it was with the assumption that a
sysfs patch would be also coming soon.  Since that isn't the case some
clean up needs to be done.  This patch does that.
Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch enables the PPS system in the PHC framework, by enabling
the clock-out feature on the X540 device. Causes the SDP0 to be set as
a 1Hz clock. Also configures the timesync interrupt cause in order to
report each pulse to the PPS via the PHC framework, which can be used
for general system clock synchronization. (This allows a stable method
for tuning the general system time via the on-board SYSTIM register
based clock.)
Signed-off-by: NJacob E Keller <jacob.e.keller@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch enables hardware timestamping for use with PTP software by
extracting a ns counter from an arbitrary fixed point cycles counter.
The hardware generates SYSTIME registers using the DMA tick which
changes based on the current link speed. These SYSTIME registers are
converted to ns using the cyclecounter and timecounter structures
provided by the kernel. Using the SO_TIMESTAMPING api, software can
enable and access timestamps for PTP packets.

The SO_TIMESTAMPING API has space for 3 different kinds of timestamps,
SYS, RAW, and SOF. SYS hardware timestamps are hardware ns values that
are then scaled to the software clock. RAW hardware timestamps are the
direct raw value of the ns counter. SOF software timestamps are the
software timestamp calculated as close as possible to the software
transmit, but are not offloaded to the hardware. This patch only
supports the RAW hardware timestamps due to inefficiency of the SYS
design.

This patch also enables the PHC subsystem features for atomically
adjusting the cycle register, and adjusting the clock frequency in
parts per billion. This frequency adjustment works by slightly
adjusting the value added to the cycle registers each DMA tick. This
causes the hardware registers to overflow rapidly (approximately once
every 34 seconds, when at 10gig link). To solve this, the timecounter
structure is used, along with a timer set for every 25 seconds. This
allows for detecting register overflow and converting the cycle
counter registers into ns values needed for providing useful
timestamps to the network stack.

Only the basic required clock functions are supported at this time,
although the hardware supports some ancillary features and these could
easily be enabled in the future.

Note that use of this hardware timestamping requires modifying daemon
software to use the SO_TIMESTAMPING API for timestamps, and the
ptp_clock PHC framework for accessing the clock. The timestamps have
no relation to the system time at all, so software must use the posix
clock generated by the PHC framework instead.
Signed-off-by: NJacob E Keller <jacob.e.keller@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The drop enable bit can be used to improve the performance of the adapter
in the case of multiple queues being present.  This performance gain is due
to the fact that some slower CPUs can cause the FIFO to backfill preventing
faster CPUs from receiving additional work.  By setting the drop enable bit
we prevent this and instead just drop the packets that would have been
bound for the slower CPU.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

After this commit:

commit aacc1bea
Author: Multanen, Eric W <eric.w.multanen@intel.com>
Date:   Wed Mar 28 07:49:09 2012 +0000

    ixgbe: driver fix for link flap

The BIT_APP_UPCHG bit is no longer set when ixgbe_dcbnl_set_all() is
called. This results in the FCoE app user priority never getting set
and the driver will not configure the tx_rings correctly for FCoE
packets which use the SAN MTU and FCoE offloads.

We resolve this regression by fixing ixgbe_copy_dcb_cfg() to also
check for FCoE application changes. Additionally, we can drop the
IEEE variants of get_dcb_app() because this path is never called
with the IEEE mode enabled.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Some of our adapters have thermal data available, this patch exports
this data via hwmon sysfs interface.
Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch consolidates the case logic for checking whether a device supports
WoL into a single place. Previously ethtool and probe used similar logic that
was copied and maintained separately. This patch encapsulates the core logic
into a function so that a user only has to update one place.
Signed-off-by: NJacob Keller <jacob.e.keller@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This was pointed out to me by Xiaojun Zhang on Source Forge.

CC: Xiaojun Zhang <zhangxiaojun@sourceforge.net>
Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Resolve namespace issues when FCoE or DCB is not enabled.
The issue is with certain configurations we end up with namespace
problems. A simple example:

ixgbe_main.c
 - defines func A()
 - uses func A()

ixgbe_fcoe.c
 - uses func A()

ixgbe.h
 - has prototype for func A()

For default (FCoE included) all is good.  But when it isn't the namespace
checker complains about how func A() could be static.

To resolve this, created a ixgbe_lib file to contain functions used
by DCB/FCoE and their helper functions so that they are always in
namespace whether or not DCB/FCoE is enabled.
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com>

This patch replaces the variable name data with the variable name features
for ixgbe_fix_features and ixgbe_set_features.  This helps to make some
issues more obvious such as the fact that we were disabling Rx VLAN tag
stripping when we should have been forcing it to be enabled when DCB is
enabled.

In addition there was deprecated code present that was disabling the LRO
flag if we had the itr value set too low.  I have updated this logic so
that we will now allow the LRO flag to be set, but will not enable RSC
until the rx-usecs value is high enough to allow enough time for Rx packet
coalescing.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds support for enabling or disabling UDP RSS via the
ethtool -N rx-flow-hash command.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change makes it so that only the 2nd cache line in the ring structure
should see frequent updates.  The advantage to this is that it should
reduce the amount of cross CPU cache bouncing since only the 2nd cache line
will be changing between most network transactions.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change makes it so that we store the tx_flags and protocol information
to the tx_buffer_info structure sooner. This allows us to avoid unnecessary
read/write transactions since we are placing the data in the final location
earlier.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This change makes it so that we always write the DMA address for the skb
itself on the same tx_buffer struct that the skb is written on.  This way
we don't need the MAPPED_AS_PAGE flag and we always know it will be the
first DMA value that we will have to unmap.

In addition I have found an issue in which we were leaking a DMA mapping if
the value happened to be 0 which is possible on some platforms.  In order
to resolve that I have updated the transmit path to use the length instead
of the DMA mapping in order to determine if a mapping is actually present.

One other tweak in this patch is that it only writes the olinfo information
on the first descriptor.  As it turns out it isn't necessary to write it
for anything but the first descriptor so there is no need to carry it
forward.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Instead of keeping a local copy of the skb on the stack for as long as long
as we do it makes sense to instead just place it on the first tx_buffer
structure so that we can save space on the stack and avoid unnecessary
read/write operations copying the pointer out of the stack and onto the
ring later.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

A separate value was added to track Tx completions in order to determine if
the Tx unit was hung.  However we can do the same thing using the number of
packets completed without having to add another stat to the Tx ring.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch replaces the existing Rx hot-path in the ixgbe driver with a new
implementation that is based on performing a double buffered receive.  The
ixgbe driver already had something similar in place for its' packet split
path, however in that case we were still receiving the header for the
packet into the sk_buff.  The big change here is the entire receive path
will receive into pages only, and then pull the header out of the page and
copy it into the sk_buff data.  There are several motivations behind this
approach.

First, this allows us to avoid several cache misses as we were taking a
set of cache misses for allocating the sk_buff and then another set for
receiving data into the sk_buff.  We are able to avoid these misses on
receive now as we allocate the sk_buff when data is available.

Second we are able to see a considerable performance gain when an IOMMU is
enabled because we are no longer unmapping every buffer on receive.
Instead we can delay the unmap until we are unable to use the page, and
instead we can simply call sync_single_range on the half of the page that
contains new data.

Finally we are able to drop a considerable amount of code from the driver
as we no longer have to support 2 different receive modes, packet split and
one buffer.  This allows us to optimize the Rx path further since less
branching is required.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

There isn't much point in using variables to store the values of eitr_low
and eitr_high since they are not user changeable.  As such I am replacing
them with the constants 10 and 20 in order to avoid any confusion on what
the values actually are.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Tested-by: NStephen Ko <stephen.s.ko@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>