Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2018-03-26

This patch series adds the ice driver, which will support the Intel(R)
E800 Series of network devices.

This is the first phase in the release of this driver where we implement
basic transmit and receive. The idea behind the multi-phase release is to
aid in code review as well as testing. Subsequent phases will implement
advanced features (like SR-IOV, tunnelling, flow director, QoS, etc.) that
build upon the previous phase(s). Each phase will be submitted as a patch
series.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch implements multiple pieces of functionality:

1. Added ice_vsi_sync_filters, which is called through the service task
   to push filter updates to the hardware.

2. Add support to enable/disable promiscuous mode on an interface.
   Enabling/disabling promiscuous mode on an interface results in
   addition/removal of a promisc filter rule through ice_vsi_sync_filters.

3. Implement handlers for ndo_set_mac_address, ndo_change_mtu,
   ndo_poll_controller and ndo_set_rx_mode.

This patch also marks the end of the driver addition by bumping up the
driver version.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Link events are posted to a PF's admin receive queue (ARQ). This patch
adds the ability to detect and process link events.

This patch also adds the ability to process resets.

The driver can process the following resets:
    1) EMP Reset (EMPR)
    2) Global Reset (GLOBR)
    3) Core Reset (CORER)
    4) Physical Function Reset (PFR)

EMPR is the largest level of reset that the driver can handle. An EMPR
resets the manageability block and also the data path, including PHY and
link for all the PFs. The affected PFs are notified of this event through
a miscellaneous interrupt.

GLOBR is a subset of EMPR. It does everything EMPR does except that it
doesn't reset the manageability block.

CORER is a subset of GLOBR. It does everything GLOBR does but doesn't
reset PHY and link.

PFR is a subset of CORER and affects only the given physical function.
In other words, PFR can be thought of as a CORER for a single PF. Since
only the issuing PF is affected, a PFR doesn't result in the miscellaneous
interrupt being triggered.

All the resets have the following in common:
1) Tx/Rx is halted and all queues are stopped.
2) All the VSIs and filters programmed for the PF are lost and have to be
   reprogrammed.
3) Control queue interfaces are reset and have to be reprogrammed.

In the rebuild flow, control queues are reinitialized, VSIs are reallocated
and filters are restored.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds the ability for a VSI to use multiple Tx queues. More
specifically, the patch
    1) Provides the ability to update the Tx scheduler tree in the
       firmware. The driver can configure the Tx scheduler tree by
       adding/removing multiple Tx queues per TC per VSI.

    2) Allows a VSI to reconfigure its Tx queues during runtime.

    3) Synchronizes the Tx scheduler update operations using locks.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch implements a watchdog task to get packet statistics from
the device.

This patch also adds support for the following ethtool operations:

ethtool devname
ethtool -s devname [msglvl N] [msglevel type on|off]
ethtool -g|--show-ring devname
ethtool -G|--set-ring devname [rx N] [tx N]
ethtool -i|--driver devname
ethtool -d|--register-dump devname [raw on|off] [hex on|off] [file name]
ethtool -k|--show-features|--show-offload devname
ethtool -K|--features|--offload devname feature on|off
ethtool -P|--show-permaddr devname
ethtool -S|--statistics devname
ethtool -a|--show-pause devname
ethtool -A|--pause devname [autoneg on|off] [rx on|off] [tx on|off]
ethtool -r|--negotiate devname

CC: Andrew Lunn <andrew@lunn.ch>
CC: Jakub Kicinski <kubakici@wp.pl>
CC: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Acked-by: NStephen Hemminger <stephen@networkplumber.org>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds support for VLANs. When a VLAN is created a switch filter
is added to direct the VLAN traffic to the corresponding VSI. When a VLAN
is deleted, the filter is deleted as well.

This patch also adds support for the following hardware offloads.
    1) VLAN tag insertion/stripping
    2) Receive Side Scaling (RSS)
    3) Tx checksum and TCP segmentation
    4) Rx checksum
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch implements ice_start_xmit (the handler for ndo_start_xmit) and
related functions. ice_start_xmit ultimately calls ice_tx_map, where the
Tx descriptor is built and posted to the hardware by bumping the ring tail.

This patch also implements ice_napi_poll, which is invoked when there's an
interrupt on the VSI's queues. The interrupt can be due to either a
completed Tx or an Rx event. In case of a completed Tx/Rx event, resources
are reclaimed. Additionally, in case of an Rx event, the skb is fetched
and passed up to the network stack.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch configures the VSIs to be able to send and receive
packets by doing the following:

1) Initialize flexible parser to extract and include certain
   fields in the Rx descriptor.

2) Add Tx queues by programming the Tx queue context (implemented in
   ice_vsi_cfg_txqs). Note that adding the queues also enables (starts)
   the queues.

3) Add Rx queues by programming Rx queue context (implemented in
   ice_vsi_cfg_rxqs). Note that this only adds queues but doesn't start
   them. The rings will be started by calling ice_vsi_start_rx_rings on
   interface up.

4) Configure interrupts for VSI queues.

5) Implement ice_open and ice_stop.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

A VSI needs traffic directed towards it. This is done by programming
filter rules on the switch (embedded vSwitch) element in the hardware,
which connects the VSI to the ingress/egress port.

This patch introduces data structures and functions necessary to add
remove or update switch rules on the switch element. This is a pretty low
level function that is generic enough to add a whole range of filters.

This patch also introduces two top level functions ice_add_mac and
ice_remove mac which through a series of intermediate helper functions
eventually call ice_aq_sw_rules to add/delete simple MAC based filters.
It's worth noting that one invocation of ice_add_mac/ice_remove_mac
is capable of adding/deleting multiple MAC filters.

Also worth noting is the fact that the driver maintains a list of currently
active filters, so every filter addition/removal causes an update to this
list. This is done for a couple of reasons:

1) If two VSIs try to add the same filters, we need to detect it and do
   things a little differently (i.e. use VSI lists, described below) as
   the same filter can't be added more than once.

2) In the event of a hardware reset we can simply walk through this list
   and restore the filters.

VSI Lists:
In a multi-VSI situation, it's possible that multiple VSIs want to add the
same filter rule. For example, two VSIs that want to receive broadcast
traffic would both add a filter for destination MAC ff:ff:ff:ff:ff:ff.
This can become cumbersome to maintain and so this is handled using a
VSI list.

A VSI list is resource that can be allocated in the hardware using the
ice_aq_alloc_free_res admin queue command. Simply put, a VSI list can
be thought of as a subscription list containing a set of VSIs to which
the packet should be forwarded, should the filter match.

For example, if VSI-0 has already added a broadcast filter, and VSI-1
wants to do the same thing, the filter creation flow will detect this,
allocate a VSI list and update the switch rule so that broadcast traffic
will now be forwarded to the VSI list which contains VSI-0 and VSI-1.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch introduces data structures and functions to alloc/free
VSIs. The driver represents a VSI using the ice_vsi structure.

Some noteworthy points about VSI allocation:

1) A VSI is allocated in the firmware using the "add VSI" admin queue
   command (implemented as ice_aq_add_vsi). The firmware returns an
   identifier for the allocated VSI. The VSI context is used to program
   certain aspects (loopback, queue map, etc.) of the VSI's configuration.

2) A VSI is deleted using the "free VSI" admin queue command (implemented
   as ice_aq_free_vsi).

3) The driver represents a VSI using struct ice_vsi. This is allocated
   and initialized as part of the ice_vsi_alloc flow, and deallocated
   as part of the ice_vsi_delete flow.

4) Once the VSI is created, a netdev is allocated and associated with it.
   The VSI's ring and vector related data structures are also allocated
   and initialized.

5) A VSI's queues can either be contiguous or scattered. To do this, the
   driver maintains a bitmap (vsi->avail_txqs) which is kept in sync with
   the firmware's VSI queue allocation imap. If the VSI can't get a
   contiguous queue allocation, it will fallback to scatter. This is
   implemented in ice_vsi_get_qs which is called as part of the VSI setup
   flow. In the release flow, the VSI's queues are released and the bitmap
   is updated to reflect this by ice_vsi_put_qs.

CC: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Acked-by: NShannon Nelson <shannon.nelson@oracle.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch continues the initialization flow as follows:

1) Allocate and initialize necessary fields (like vsi, num_alloc_vsi,
   irq_tracker, etc) in the ice_pf instance.

2) Setup the miscellaneous interrupt handler. This also known as the
   "other interrupt causes" (OIC) handler and is used to handle non
   hotpath interrupts (like control queue events, link events,
   exceptions, etc.

3) Implement a background task to process admin queue receive (ARQ)
   events received by the driver.

CC: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Acked-by: NShannon Nelson <shannon.nelson@oracle.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds code to continue the initialization flow as follows:

1) Get PHY/link information and store it
2) Get default scheduler tree topology and store it
3) Get the MAC address associated with the port and store it
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch adds to the initialization flow by getting switch
configuration, scheduler configuration and device capabilities.

Switch configuration:
On boot, an L2 switch element is created in the firmware per physical
function. Each physical function is also mapped to a port, to which its
switch element is connected. In other words, this switch can be visualized
as an embedded vSwitch that can connect a physical function's virtual
station interfaces (VSIs) to the egress/ingress port. Egress/ingress
filters will be eventually created and applied on this switch element.
As part of the initialization flow, the driver gets configuration data
from this switch element and stores it.

Scheduler configuration:
The Tx scheduler is a subsystem responsible for setting and enforcing QoS.
As part of the initialization flow, the driver queries and stores the
default scheduler configuration for the given physical function.

Device capabilities:
As part of initialization, the driver has to determine what the device is
capable of (ex. max queues, VSIs, etc). This information is obtained from
the firmware and stored by the driver.

CC: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Acked-by: NShannon Nelson <shannon.nelson@oracle.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Ido Schimmel says:

====================
mlxsw: Offload IPv6 multicast routes

Yuval says:

The series is intended to allow offloading IPv6 multicast routes
and is split into two parts:

  - First half of the patches continue extending ip6mr [& refactor ipmr]
    with missing bits necessary for the offloading - fib-notifications,
    mfc refcounting and default rule identification.

  - Second half of the patches extend functionality inside mlxsw,
    beginning with extending lower-parts to support IPv6 mroutes
    to host and later extending the router/mr internal APIs within
    the driver to accommodate support in ipv6 configurations.
    Lastly it adds support in the RTNL_FAMILY_IP6MR notifications,
    allowing driver to react and offload related routes.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a new trap for PIMv6 packets. As PIM already has a designated trap
group [ & rate limiter], simply use the same for PIMv6 as well.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Following previous patches driver is ready to handle notifications
arriving from ip6mr - start processing those when they arrive following
the same manner ipmr currently goes through.

This should enable driver to start offloading ipv6 multicast routes.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Populate the various operation structures meant for IPv6 with logic
unique to that protocol suite.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

spectrum_router and spectrum_mr have several APIs that are used to
manipulate configurations originating from ipmr fib notifications.
Following previous patches all the protocol-specifics that are necessary
for the configuration are hidden within spectrum_mr. This allows us to
clean the API and make sure that other than choosing the mr_table based
on the fib notification family, spectrum_router wouldn't care about the
source of the notification when passing it onward to spectrum_mr.

This would later allow us to leverage the same code for fib
notifications originating from ip6mr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Current multicast routing logic in driver assumes it's always meant to
deal with IPv4 multicast routes, leaving several placeholders for
later IPv6 support [currently usually WARN()].

This patch changes the driver's internal multicast route struct into
holding a common mr_mfc instead of the IPv4 mfc_cache.
The various placeholders are grouped into 2:
  - Functions that require only the common bits; These remain and the
    restriction for IPv4-only is lifted.
  - Function that require IPv4-specifics - for handling these functions
    we add sets of operations that encapsulate the protocol differences
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A step toward offloading IPv6 routing, this adds an additional
multicast routing table meant for IPv6 [with its underlying TCAM
region] and populates the default rule for IPv6 multicast packets.

Following this, ingress IPv6 multicast packets would be trapped and
delivered to the host CPU.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since commit c011ec1b ("mlxsw: spectrum: Add the multicast routing
offloading logic") spectrum_mr did not populate the protocol portion of
the catcahall_route_params; mr-tcam logic worked correctly for ipv4
since the enum value for MLXSW_SP_L3_PROTO_IPV4 is '0'.

Explicitly fill the protocol as we'll soon need to differentiate between
ipv4 and ipv6.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add new fields for the rmft register necessary for setting the IPv6
multicast FIB table. Add a matching wrapper function for filling
the register in the IPv6 scenario.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Similarly to what was done in commit 4af5964e ("mlxsw: reg:
Configure RIF to forward IPv4 multicast packets by default") we now set
two additional bits to allow IPv6 multicast forwarding.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since ipmr and ip6mr are using the same mr_mfc struct at their core, we
can now refactor the ipmr_cache_{hold,put} logic and apply refcounting
to both ipmr and ip6mr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add the ability to discern whether a given FIB rule notification relates
to the default rule inserted when registering ip6mr or a different one.

Would later be used by drivers wishing to offload ipv6 multicast routes
but unable to offload rules other than the default one.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In similar fashion to ipmr, support fib notifications for ip6mr mfc and
vif related events. This would later allow drivers to react to said
notifications and offload the IPv6 mroutes.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since all the primitive elements used for the notification done by ipmr
are now common [mr_table, mr_mfc, vif_device] we can refactor the logic
for dumping them to a common file.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Like vif notifications, move the notifier struct for MFC as well as its
helpers into a common file; Currently they're only used by ipmr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The fib-notifiers are tightly coupled with the vif_device which is
already common. Move the notifier struct definition and helpers to the
common file; Currently they're only used by ipmr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Kirill Tkhai says:

====================
Converting pernet_operations (part #7.1)

this is a resending of the 4 patches from path #7.

Anna kindly reviewed them and suggested to take the patches
through net tree, since there is pernet_operations::async only
in net-next.git.

There is Anna's acks on every header, the rest of patch
has no changes.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These pernet_operations create and destroy per-net pipe
and dentry, and they seem safe to be marked as async.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These pernet_operations look similar to rpcsec_gss_net_ops,
they just create and destroy another cache. Also they create
and destroy directory. So, they also look safe to be async.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These pernet_operations look similar to rpcsec_gss_net_ops,
they just create and destroy another caches. So, they also
can be async.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These pernet_operations initialize and destroy sunrpc_net_id
refered per-net items. Only used global list is cache_list,
and accesses already serialized.

sunrpc_destroy_cache_detail() check for list_empty() without
cache_list_lock, but when it's called from unregister_pernet_subsys(),
there can't be callers in parallel, so we won't miss list_empty()
in this case.
Signed-off-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: NAnna Schumaker <Anna.Schumaker@netapp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Pieter Jansen van Vuuren says:

====================
nfp: flower: add ip fragmentation offloading support

This set allows offloading IP fragmentation classification. It Implements
ip fragmentation match offloading for both IPv4 and IPv6 and offloads
frag, nofrag, first and nofirstfrag classification.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Implement ip fragmentation match offloading for both IPv4 and IPv6. Allows
offloading frag, nofrag, first and nofirstfrag classification.
Signed-off-by: NPieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Refactored shared ip header code for IPv4 and IPv6 in match offload.
Signed-off-by: NPieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch implements multiple pieces of the initialization flow
as follows:

1) A reset is issued to ensure a clean device state, followed
   by initialization of admin queue interface.

2) Once the admin queue interface is up, clear the PF config
   and transition the device to non-PXE mode.

3) Get the NVM configuration stored in the device's non-volatile
   memory (NVM) using ice_init_nvm.

CC: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Acked-by: NShannon Nelson <shannon.nelson@oracle.com>
Tested-by: NTony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Sinan Kaya says:

====================
netdev: Eliminate duplicate barriers on weakly-ordered archs

Code includes wmb() followed by writel() in multiple places. writel()
already has a barrier on some architectures like arm64.

This ends up CPU observing two barriers back to back before executing the
register write.

Since code already has an explicit barrier call, changing writel() to
writel_relaxed().

I did a regex search for wmb() followed by writel() in each drivers
directory.
I scrubbed the ones I care about in this series.

I considered "ease of change", "popular usage" and "performance critical
path" as the determining criteria for my filtering.

We used relaxed API heavily on ARM for a long time but
it did not exist on other architectures. For this reason, relaxed
architectures have been paying double penalty in order to use the common
drivers.

Now that relaxed API is present on all architectures, we can go and scrub
all drivers to see what needs to change and what can remain.

We start with mostly used ones and hope to increase the coverage over time.
It will take a while to cover all drivers.

Feel free to apply patches individually.

Changes since v6:
- bring back amazon ena and add mmiowb, remove
  ena_com_write_sq_doorbell_rel().
- remove extra mmiowb in bnx2x
- correct spelling mistake in  bnx2x: Replace doorbell barrier() with wmb()
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Code includes barrier() followed by writel(). writel() already has a
barrier on some architectures like arm64.

This ends up CPU observing two barriers back to back before executing the
register write.

Create a new wrapper function with relaxed write operator. Use the new
wrapper when a write is following a barrier().

Since code already has an explicit barrier call, changing writel() to
writel_relaxed() and adding mmiowb() for ordering protection.
Signed-off-by: NSinan Kaya <okaya@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>