r8169: define the early size for 8111evl

For RTL8111EVL, the register of MaxTxPacketSize doesn't acctually
limit the tx size. It influnces the feature of early tx.

(cherry picked from commit 3090bd9a3102331ed981280ef05ee5433ce978d7)
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: convert to SKB paged frag API.

(cherry picked from commit 929f61896e194f45696ba52852ccbfd5edd04df1)
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Realtek linux nic maintainers <nic_swsd@realtek.com>
Cc: Francois Romieu <romieu@fr.zoreil.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

8139cp: convert to SKB paged frag API.

(cherry picked from commit deb8a069e39b272c0a37bfa7a5f52d49600963a4)
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

net: remove use of ndo_set_multicast_list in realtek drivers

replace it by ndo_set_rx_mode for realtek devices

(backported from commit afc4b13df143122f99a0eb10bfefb216c2806de0)
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169 : MAC address change fix for the 8168e-vl.

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=39252

Hayes suggested that the usual MAC{0, 4} register writes be completed
with writes to extended GigaMAC registers :
- 0xe0 .. 0xe5
- 0xf2 .. 0xf7

Registers 0xf0 and 0xf1 should be set to 0.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit c28aa38567101bad4e020f4392df41d0bf6c165c)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: use pci_dev->subsystem_{vendor|device}

The driver reads PCI subsystem IDs from the PCI configuration registers while
they are already stored by the PCI subsystem in the 'subsystem_{vendor|device}'
fields of 'struct pci_dev'...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit ccbae55e1c6dc18e95d72c044cf9345ea08abf7b)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: fix sticky accepts packet bits in RxConfig.

e542a2269f232d61270ceddd42b73a4348dee2bb (r8169: adjust the RxConfig settings)
broke the return from promiscuous mode to physical address match mode.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Signed-off-by: Hayes Wang <hayeswang@realtek.com>
(cherry picked from commit 1687b56679880a47164ae149530abee543f9d6b1)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: adjust the RxConfig settings.

Set the init value before reset in probe function. And then just
modify the relative bits and keep the init settings.

For 8110S, 8110SB, and 8110SC series, the initial value of RxConfig
needs to be set after the tx/rx is enabled.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit e542a2269f232d61270ceddd42b73a4348dee2bb)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: don't enable rx when shutdown.

Only 8111b needs to enable rx when shutdowning with WoL.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit aaa89c08d9ffa3739c93d65d98b73ec2aa2e93a5)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: fix wake on lan setting for non-8111E.

Only 8111E needs enable RxConfig bit 0 ~ 3 when suspending or
shutdowning for wake on lan.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit d4ed95d796e5126bba51466dc07e287cebc8bd19)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: support RTL8111E-VL.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit 70090424e59652c4b2e777b533cc23134b176b83)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: add ERI functions.

Add the ERI functions which would be used by the new chips.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit 133ac40ac26a6cba957ad5765892f0531ba6ccee)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: modify the flow of the hw reset.

- Disable tx and rx by resetting hw, so replace rtl8169_asic_down
  with rtl8169_hw_reset.
- RxConfig bits 0 ~ 5 have to be cleared before hw reset to avoid
  receiving spurious data.
- Certain chips need to do some checking before reset.
- Remove hw reset which is done before hw_start. It is done in close,
  down or device probe functions.
- Move rtl8169_init_ring_indexes function into rtl_hw_reset function.
  The indexes of tx and rx only need to be zero when the hw resets.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit 92fc43b4159b518f5baae57301f26d770b0834c9)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: adjust some registers.

Define new registers and modify some existing ones.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
(cherry picked from commit 4f6b00e5f139d7be3ca8371b769778f94fa549dd)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

r8169: remove unnecessary read of PCI_CAP_ID_EXP

The PCIE capability offset is saved during PCI bus walking.  Use the
value from pci_dev instead of checking in the driver and saving it off
the the driver specific structure.  Also, it will remove an unnecessary
search in the PCI configuration space if this value is referenced
instead of reacquiring it.

Signed-off-by: Jon Mason <jdmason@kudzu.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit e44daaded19baf8b7cf9ab3dedad8a01b8c49251)

Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: print MAC via printk format specifier

Print MAC/dev_addr via printk extended format specifier %pM
instead of custom code.

(cherry picked from commit f794e7efe40147ed0c34f889b75298fec10e66b8)
Signed-off-by: Danny Kukawka <danny.kukawka@bisect.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Update copyright notices

(cherry picked from commit 5c47a2b6211316ee9cd8740db453af7c5c01d0d6)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Fix mailbox interrupt ack bug

Remove jump to out label that was skipping over code to replace the
ack bit in the mailbox cache variable.  This was causing driver mailbox
routines that poll for the PF to ack a message to time out which would
in turn cause all other mailbox messages to fail.  Also replace the
mailbox cache variable message indication bit when a message is found
so that other functions that choose to poll for a mailbox message from
the PF won't miss it.

(cherry picked from commit 3a2c40336b5ee112f6a2bc784e79aa6124d7acf6)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: make operations tables const

The arrays of function pointers should be const to make life harder
for rootkits.

(cherry picked from commit 3d8fe98f8d5cb303d907d8f94ea6dc1f9a1d8b7a)
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: fix sparse warnings

Fixes sparse warnings:
drivers/net/ixgbevf/vf.c:418:21: warning: symbol 'ixgbevf_82599_vf_info' was not declared. Should it be static?
drivers/net/ixgbevf/vf.c:423:21: warning: symbol 'ixgbevf_X540_vf_info' was not declared. Should it be static?
drivers/net/ixgbevf/mbx.c:331:29: warning: symbol 'ixgbevf_mbx_ops' was not declared. Should it be static?

(cherry picked from commit b5417bf8e8952401bca69ded67c30ead484af823)
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: make ethtool ops and strings const

(cherry picked from commit b47aca135d6df5f676c768368baf3c099f054fcd)
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Prevent possible race condition by checking for message

The mailbox interrupt routine might cause a race condition sometimes
and cause a message to be missed.

(cherry picked from commit 375b27cf5db963e2bc2a34dc5643d0d7ceca1ee2)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Fix register defines to correctly handle complex expressions

This patch is meant to address possible issues with the IXGBEVF register
defines generating incorrect values when given a complex expression for the
register offset.

(cherry picked from commit f131a6c07ec0eb9d00f99af18be52a6da0458e82)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Update release version

(cherry picked from commit c1a7e1ebc17a9243d99ba0432d1138d74114dea7)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Fix broken trunk vlan

Changes to clean up the vlan rx path broke trunk vlan.  Trunk vlans in
a VF driver are those set using:

"ip link set <pfdev> vf <n> <vlanid>"

(cherry picked from commit dd1ed3b7bfed15f6162f63840941e9cf4f3611a1)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
CC: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: convert to ndo_fix_features

Private rx_csum flags are now duplicate of netdev->features & NETIF_F_RXCSUM.
Removing this needs deeper surgery.

Since ixgbevf doesn't change hardware state on RX csum enable/disable
its reset is avoided.

Things noticed:
 - HW VLAN acceleration probably can be toggled, but it's left as is
 - the resets on RX csum offload change can probably be avoided
 - there is A LOT of copy-and-pasted code here

(backported from commit 471a76ded87d3375a3449dfa3d1cec567edd0c50)
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbevf: Check if EOP has changed before using it

There is a chance that between the time EOP is read and the time it is
used another transmit on a different CPU could have run and completed,
thus leaving EOP in a bad state.

(cherry picked from commit 98b9e48fca11c8aa54b25c02d3329392b52db8ab)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Correct flag values set by ixgbe_fix_features

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.

(cherry picked from commit 567d2de291b5ddb83654c5e87c14b4c6fa7216ed)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix typo in enumeration name

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

CC: Xiaojun Zhang <zhangxiaojun@sourceforge.net>
(cherry picked from commit 70e5576cb0af5c1351432704a39319af119584bb)
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Add support for enabling UDP RSS via the ethtool rx-flow-hash command

This patch adds support for enabling or disabling UDP RSS via the
ethtool -N rx-flow-hash command.

(cherry picked from commit ef6afc0caceebb3b49cd384f8c634f89f5089997)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Whitespace cleanups

This patch contains several fixes for formatting in regards to whitespace.

(backported from commit 581330ba7b219fec0587452bd72fe5635c30c4ae)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Two minor fixes for RSS and FDIR set queues functions

This change fixes two minor issues. The first was the fact that we were
setting the return value to false twice in the set_rss_queues function.
The second is the fact that we should have been using "min_t(int," instead
of "min((int)" in set_fdir_queues.

(cherry picked from commit 6ca435078822d9728318dee49ba4f2ff8a2ead24)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: drop err_eeprom tag which is at same location as err_sw_init

The err_eeprom and err_sw_init tags both go to the same location.  So
instead of maintaining two tags this patch combines them so we only use
err_sw_init.

(cherry picked from commit 35937c055a691a125cf52343cd671f6190ae5b3b)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Move poll routine in order to improve readability

This change relocates the ixgbe_poll routine so it is right next to the
interrupt routine that schedules and calls it.

(cherry picked from commit eb01b9759e733c0ede11d35e3cde5b3c24ab32c9)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: cleanup logic for the service timer and VF hang detection

This change just cleans up some of the logic in the service_timer function
so that we can avoid unnecessary swapping of the ready value between true to
false and back to true.

(cherry picked from commit 6bb78cfb50bfbaf90646a80448ee8de2372e292a)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix issues with SR-IOV loopback when flow control is disabled

This patch allows us to avoid a Tx hang when SR-IOV is enabled.  This hang
can be triggered by sending small packets at a rate that was triggering Rx
missed errors from the adapter while the internal Tx switch and at least
one VF are enabled.

This was all due to the fact that under heavy stress the Rx FIFO never
drained below the flow control high water mark.  This resulted in the Tx
FIFO being head of line blocked due to the fact that it relies on the flow
control high water mark to determine when it is acceptable for the Tx to
place a packet in the Rx FIFO.

The resolution for this is to set the FCRTH value to the RXPBSIZE - 32 so
that even if the ring is almost completely full we can still place Tx
packets on the Rx ring and drop incoming Rx traffic if we do not have
sufficient space available in the Rx FIFO.

(cherry picked from commit c29f40ca19b8ca284a2d1415e26b802bc9b2fb67)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Place skb on first buffer_info structure to avoid using stack space

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.

(backported from commit fd0db0ed02a6abce5427e90d1e8522322107d62b)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Use packets to track Tx completions instead of a seperate value

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.

(cherry picked from commit 7d7ce682f8437ff91c964c478b8845ed98a3207d)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Modify setup of descriptor flags to avoid conditional jumps

This change makes it more likely that the descriptor flags setup will use
cmov instructions instead of conditional jumps when setting up the flags.
The advantage to this is that the code should just flow a bit more
smoothly.

To do this it is necessary to set the TX_FLAGS_CSUM bit in tx_flags when
doing TSO so that we also do the checksum in addition to the segmentation
offload.

(cherry picked from commit 93f5b3c1f148f2cca247a2c5afdd3ba7a123a6f1)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Make certain that all frames fit minimum size requirements

This change makes certain that any packet we attempt to transmit will meet
minimum size requirements for the hardware.

(cherry picked from commit a50c29dd09ed14f8489677813a4b1c8b62dcf19d)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: cleanup logic in ixgbe_change_mtu

This change is meant to just cleanup the logic in ixgbe_change_mtu since we
are making it unnecessarily complex due to a workaround required for 82599
when SR-IOV is enabled.

(cherry picked from commit 655309e944fd482e59850d55186571c1b2a91e55)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: dcb: use DCB config values for FCoE traffic class on open

Disabling and enabling DCB can cause FCoE hardware initialization to
occur on the incorrect traffic class when the up2tc mapping has not
yet been reconfigured.

Fix this by using the DCB configuration maps that are correct
and will be pushed at mqprio after DCB driver setup completes
successfully.

(cherry picked from commit cdf485be3a63d1f34293740fb726088c6840ceea)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Marcus Dennis <marcusx.e.dennis@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix race condition where RX buffer could become corrupted.

There was a race condition in the reset path where the RX buffer
could become corrupted during Fdir configuration.This is due to
a HW bug.The fix right now is to lock the buffer while we do the
fdir configuration.Since we were using similar workaround for another bug,
I moved the existing code to a function and reused it.HW team also recommended
that IXGBE_MAX_SECRX_POLL value be changed from 30 to 40.The erratum for this
bug will be published in the next release 82599 Spec Update

(cherry picked from commit d2f5e7f3af777d52ce110ae60e8cb788bc93141c)
Signed-off-by: Atita Shirwaikar <atita.shirwaikar@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: use typed min/max functions where possible

using the form min((int)var, ver)) is replaced by min_t(int, ...)

(cherry picked from commit 3ed69d7e318ce17c77d91fd09a963cf48ce3a8a9)
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix obvious return value bug.

This is clearly a typeo where we are not checking the return value from
get_link_capabilities but should.  This patch corrects that.

(cherry picked from commit 9cdcf098800d9549b0bf1e759f8b456baabfb939)
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Replace eitr_low and eitr_high with static values in ixgbe_update_itr

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.

(cherry picked from commit 621bd70eda47f33aa5f3a48da56649c0cb7e2329)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Do not disable read relaxed ordering when DCA is enabled

A previous fix had gone though and disabled relaxed ordering for Rx
descriptor read fetching.  This was not necessary as this functions
correctly and has no ill effects on the system.

In addition several of the defines used for the DCA control registers were
incorrect in that they indicated descriptor effects when they actually had
an impact on either data or header write back.  As such I have update these
to correctly reflect either DATA or HEAD.

(cherry picked from commit bdda1a61c1dbb1e3a4f76c944b7c1e3b3731c861)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Simplify logic for ethtool loopback frame creation and testing

This change makes it a bit easier to do the loopback frame creating and
testing.  Previously we were doing an and to drop the last bit, and then
dividing the frame_size by 2 in order to get locations for frame bytes and
testing.  Instead we can simplify it by just shifting the register one bit
to the right and using that for the frame offsets.

This change also replaces all instances of rx_buffer_info with just
rx_buffer since that is closer to the name of the actual structure being
used and can save a few extra characters.

In addition I have updated the logic for cleaning up a test frame so that
we pass an rx_buffer instead of the sk_buff.  The main motivation behind
this is changes that will replace the sk_buff with just a page in the
future.

(cherry picked from commit 3832b26e49ad9e585239b32f763c31679f9e41fe)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Add iterator for cycling through rings on a q_vector

Since there are multiple spots where we have to cycle through all of the
rings on a q_vector it makes sense to just add a function for iterating
through all of them.

(cherry picked from commit a557928e26b08496c8f4b6c04e3838ad8048ad85)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Allocate rings as part of the q_vector

This patch makes the rings a part of the q_vector directly instead of
indirectly.  Specifically on x86 systems this helps to avoid any cache
set conflicts between the q_vector, the tx_rings, and the rx_rings as the
critical stride is 4K and in order to cross that boundary you would need to
have over 15 rings on a single q_vector.

In addition this allows for smarter allocations when Flow Director is
enabled.  Previously Flow Director would set the irq_affinity hints based
on the CPU and was still using a node interleaving approach which on some
systems would end up with the two values mismatched.  With the new approach
we can set the affinity for the irq_vector and use the CPU for that
affinity to determine the node value for the node and the rings.

(cherry picked from commit de88eeeb16b164b615a5d71ad5fa0b7d51b14435)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Drop unnecessary napi_schedule_prep and spare blank line from ixgbe_intr

This patch is a minor cleanup to address the unnecessary use of
napi_schedule_prep in ixgbe_intr and to also remove a blank line that is
not needed since it is separating a comment from the line it is explaining.

(cherry picked from commit b9f6ed2bab0b087bdff870b7173f1d89efd96e54)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Default to queue pairs when number of queues is less than CPUs

The old code had several errors in how it was determining the vector
budget.  In order to simplify things this patch updates the code so that it
will attempt to always allocated paired Rx/Tx vectors instead of attempting
to allocate individual vectors when the number of queues is less than the
number of CPUs.

(cherry picked from commit 8f15486dd0f6e8423cb33fa8513ed9d074476471)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Correct Adaptive Interrupt Moderation so that it will change values

This change corrects an issue in which Adaptive Interrupt Moderation was
not changing values due to the fact that we were performing an and
operation on the resultant value that was causing the value to never change
from the default 20K interrupts per second.

(cherry picked from commit 5d967eb7f84d2761d624d033c2583187b37b56ce)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Address issues with Tx WHTRESH value not being set correctly

This change is meant to address the fact that the tx_itr_setting was
dropping to 0 when no separate Tx vectors were provided.  This had resulted
in the driver incorrectly configuring the Tx ring with a WTHRESH of 1 in
order to avoid Tx hangs even though that was not necessary. This change
makes it so that we instead take a look at the Tx ring's q_vector to
determine if the ring will have an ITR value less than 8us.

(cherry picked from commit e954b374b4e42c5335679ec6d79ca2c50ce5d422)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Reorder adapter contents for better cache utilization

This change moves several frequently accessed items together into one cache
line in order to reduce cache misses in the hot-path.

(backported from commit 46646e61ea068b11758e8ffed3f4db300323dfc6)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Do no clear Tx status bits since eop_desc provides enough info

There isn't any need to clear the status bits in the descriptors due to the
fact that the eop_desc provides enough information for us to know
that we have cleaned to the last packet that the software has put on the
ring.  The status bits are cleared as a part of putting the frame on the
ring so as long as we do not read the descriptor bit prior to reading the
value eop_desc we should be able to guarantee that we will not clean beyond
the end of the current data stream.

(cherry picked from commit 7f83a9e6292b49d1540f82f295ac8710ba562c5f)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: remove tie between NAPI work limits and interrupt moderation

As noted by Ben Hutchings and David Miller, work limits for NAPI
should not be tied to interrupt moderation parameters.  This
should be handled by NAPI, possibly through sysfs.

Neil Horman & Stephen Hemminger are working on a solution for
NAPI currently.  In the meantime, remove this tie between
work limits and interrupt moderation.

(cherry picked from commit 35551c47495a762fcfa5ca6217bf3eb36983d0df)
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: dcb: check setup_tc return codes

dcb netlink code calls setup_tc to init hardware traffic classes
to use for DCB. At some call sites the return values are not
checked for errors and in one case may return -EINVAL back to
the net/dcbnl.c caller which is expecting a u8.

This fixes some smatch hits and although failures are never
seen in practive its best to check return codes.

Reported-by: Dan Carenter <dan.carpenter@oracle.com>
(cherry picked from commit 89d27a3c0ad7d5f2ce9ece055b3c7c619dc42f6c)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix comments that are out of date or formatted incorrectly

This patch corrects several comments that are either incorrect or formatted
incorrectly for multiline comments.

(cherry picked from commit 24ddd967806a5f0a7915721b7d60cac6f2cf5db9)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix spelling errors

Correct spelling error caught with codespell.py.

(cherry picked from commit c466d7a73360ed02167598a94984ab7b4dadfa68)
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Minor formatting and comment corrections for ixgbe_xmit_frame_ring

This patch is meant to address several minor issues in
ixgbe_xmit_frame_ring.  Specifically it adds a comment explaining the TXSW
flag, and correctly wraps a line over 80 characters.

(cherry picked from commit 9e0c5648d973cfe6bb037df3fdd08903c81bdc3e)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Combine post-DMA processing of sk_buff fields into single function

This change combines a number of post-DMA Rx packet processing functions
into a single function.  The advantage of this is that it combines most of
the Rx descriptor processing into one spot so it should all be warm in the
cache.

(cherry picked from commit 8a0da21be84019d605af2c9a92b20a084db77de1)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Drop the _ADV of descriptor macros since all ixgbe descriptors are ADV

It doesn't make much sense to differentiate between advanced and legacy
descriptors when the only descriptors that ixgbe uses are advanced
descriptors.  As such we can drop the _ADV suffix since all ixgbe
descriptors are automatically advanced.

(cherry picked from commit e4f740287fbfdc7b68634e93c098c8ea8de691f1)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Add function for testing status bits in Rx descriptor

This change adds a small function for testing Rx status bits in the
descriptor.  The advantage to this is that we can avoid unnecessary
byte swaps on big endian systems.

(cherry picked from commit f56e0cb1fea6aa3caace1c1ddde3f847793dcf38)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Let the Rx buffer allocation clear status bits instead of cleanup

This change makes it so that we always clear the status/error bits in the
Rx descriptor in the allocation path instead of the cleanup path.  The
advantage to this is that we spend less time modifying data.  As such we
can modify the data once and then let it go cold in the cache instead of
writing it, reading it, and then writing it again.

(cherry picked from commit f990b79bc80ca7a23b8a6c33241c439072d0b85b)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Address fact that RSC was not setting GSO size for incoming frames

This patch is meant to address the fact that RSC has not been setting the
gso_size value on the skb.  As a result performance on lossy TCP
connections was negatively impacted.  This change resolves the issue by
setting gso_size to the average size for incoming packets.

(cherry picked from commit 1d2024f61ec14bdb0c57a97a3fe73685abc2d198)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Minor refactor of RSC

This change addresses several issue.

First I had left the use of the next and prev skb pointers floating around
in the code and they were overdue to be pulled since I had rewritten the
RSC code in the out-of-tree driver some time ago to address issues brought
up by David Miller in regards to this.

I am also now defaulting to always leaving the first buffer unmapped on any
packet and then unmapping it after we read the EOP descriptor.  This allows
a simplification of the path with less branching.

Instead of counting packets received the code was changed some time ago to
track the number of buffers received.  This leads to inaccurate counting
when you compare numbers of packets received by the hardware versus what is
tracked by the software.  To correct this I am revising things so that the
append_cnt value for RSC accurately tracks the number of frames received.

(cherry picked from commit 4c1975d77b73feed7161999aab4cc64c1ae7155c)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: ethtool: stats user buffer overrun

If the number of tx/rx queues changes the ethtool ioctl
ETHTOOL_GSTATS may overrun the userspace buffer. This
occurs because the general practice in user space to
query stats is to issue a ETHTOOL_GSSET cmd to learn the
buffer size needed, allocate the buffer, then call
ETHTOOL_GSTIRNGS and ETHTOOL_GSTATS. If the number of
real_num_queues is changed or flow control attributes
are changed after ETHTOOL_GSSET but before the
ETHTOOL_GSTRINGS/ETHTOOL_GSTATS a user space buffer
overrun occurs.

To fix the overrun always return the max buffer size
needed from get_sset_count() then return all strings
and stats from get_strings()/get_ethtool_stats().

This _will_ change the output from the ioctl() call
which could break applications and script parsing in
theory. I believe these changes should not break existing
tools because the only changes will be more {tx|rx}_queues
and the {tx|rx}_pb_* stats will always be returned.
Existing scripts already need to handle changing number
of queues because this occurs today depending on system
and current features. The {tx|rx}_pb_* stats are at the
end of the output and should be handled by scripts today
regardless.

Finally get_ethtool_stats and get_strings are free-form
outputs tools parsing these outputs should be defensive
anyways. In the end these updates are better then
having a tool segfault because of a buffer overrun.

(cherry picked from commit 9cc00b51a3cbb3b933065a55eaa9bfc306411b7a)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: dcb: up2tc mapping lost on disable/enable CEE DCB state

Users expect the up2tc mapping to be maintained across a DCB
enable/disable/enable transition. And since we maintain all
the other DCB attributes we should do this for up2tc mappings
as well just to be consistent. Also without this we break
user space applications that expect this to occur that
previously worked.

(cherry picked from commit 5facb8e0c4dc1c8c47b71fd7f376defe16185733)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: do not update real num queues when netdev is going away

If the netdev is already in NETREG_UNREGISTERING/_UNREGISTERED state, do not
update the real num tx queues. netdev_queue_update_kobjects() is already
called via remove_queue_kobjects() at NETREG_UNREGISTERING time. So, when
upper layer driver, e.g., FCoE protocol stack is monitoring the netdev
event of NETDEV_UNREGISTER and calls back to LLD ndo_fcoe_disable() to remove
extra queues allocated for FCoE, the associated txq sysfs kobjects are already
removed, and trying to update the real num queues would cause something like
below:

...
PID: 25138  TASK: ffff88021e64c440  CPU: 3   COMMAND: "kworker/3:3"
 #0 [ffff88021f007760] machine_kexec at ffffffff810226d9
 #1 [ffff88021f0077d0] crash_kexec at ffffffff81089d2d
 #2 [ffff88021f0078a0] oops_end at ffffffff813bca78
 #3 [ffff88021f0078d0] no_context at ffffffff81029e72
 #4 [ffff88021f007920] __bad_area_nosemaphore at ffffffff8102a155
 #5 [ffff88021f0079f0] bad_area_nosemaphore at ffffffff8102a23e
 #6 [ffff88021f007a00] do_page_fault at ffffffff813bf32e
 #7 [ffff88021f007b10] page_fault at ffffffff813bc045
    [exception RIP: sysfs_find_dirent+17]
    RIP: ffffffff81178611  RSP: ffff88021f007bc0  RFLAGS: 00010246
    RAX: ffff88021e64c440  RBX: ffffffff8156cc63  RCX: 0000000000000004
    RDX: ffffffff8156cc63  RSI: 0000000000000000  RDI: 0000000000000000
    RBP: ffff88021f007be0   R8: 0000000000000004   R9: 0000000000000008
    R10: ffffffff816fed00  R11: 0000000000000004  R12: 0000000000000000
    R13: ffffffff8156cc63  R14: 0000000000000000  R15: ffff8802222a0000
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
 #8 [ffff88021f007be8] sysfs_get_dirent at ffffffff81178c07
 #9 [ffff88021f007c18] sysfs_remove_group at ffffffff8117ac27
#10 [ffff88021f007c48] netdev_queue_update_kobjects at ffffffff813178f9
#11 [ffff88021f007c88] netif_set_real_num_tx_queues at ffffffff81303e38
#12 [ffff88021f007cc8] ixgbe_set_num_queues at ffffffffa0249763 [ixgbe]
#13 [ffff88021f007cf8] ixgbe_init_interrupt_scheme at ffffffffa024ea89 [ixgbe]
#14 [ffff88021f007d48] ixgbe_fcoe_disable at ffffffffa0267113 [ixgbe]
#15 [ffff88021f007d68] vlan_dev_fcoe_disable at ffffffffa014fef5 [8021q]
#16 [ffff88021f007d78] fcoe_interface_cleanup at ffffffffa02b7dfd [fcoe]
#17 [ffff88021f007df8] fcoe_destroy_work at ffffffffa02b7f08 [fcoe]
#18 [ffff88021f007e18] process_one_work at ffffffff8105d7ca
#19 [ffff88021f007e68] worker_thread at ffffffff81060513
#20 [ffff88021f007ee8] kthread at ffffffff810648b6
#21 [ffff88021f007f48] kernel_thread_helper at ffffffff813c40f4

(cherry picked from commit 9d837ea2b72321ffbdaf83c980923d23ae898baa)
Signed-off-by: Yi Zou <yi.zou@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix broken dependency on MAX_SKB_FRAGS being related to page size

This patch fixes an issue in which RSC will generate corrupted frames when
PAGE_SIZE is larger than 8K.  Specifically it looks like that in 2.6.39 a
change was made so that GRO would always have at least 16 frags available
for coalescing, but the ixgbe RSC logic was not updated.  As such the RSC
feature would generate a frame larger than 64K and then overflow the value
in the IP length field.

To correct that I am now basing things on the PAGE_SIZE.

(cherry picked from commit 642c680e9314fc102bc1f096f48ae3974931ef42)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix case of Tx Hang in PF with 32 VFs

A check for the number of VFs allocated should have used a greater than
equal operator instead of just greater than.  This caused allocation of
exactly 32 VFs to not enable the PF transmit and receive enables.

(cherry picked from commit 4cd6923d3481773f3fbcae5ca940c8823aa33475)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix vf lookup

Recent addition of code to find already allocated VFs failed to take
account that systems with 2 or more multi-port SR-IOV capable controllers
might have already enabled VFs.  Make sure that the VFs the function is
finding are actually subordinate to the particular instance of the adapter
that is looking for them and not subordinate to some device that has
previously enabled SR-IOV.

This bug exists in 3.2 stable as well as 3.3 release candidates.

CC: stable@vger.kernel.org
Reported-by: David Ahern <daahern@cisco.com>
(cherry picked from commit a4b08329c74985e5cc3a44b6d2b2c59444ed8079)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix typo in ixgbe_common.h

Correct spelling "packtetbuf_num" to "packetbuf_num" in
drivers/net/ixgbe/ixgbe_common.h

(cherry picked from commit 542a398080ff4a2fcf6a8f3ae642dad5fbc8877e)
Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: make ethtool strings table const

(cherry picked from commit 1bba2e81f5575825b59146f982004415d3751e6b)
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Add warning when no space left for more MAC filters

Send message to system log when the VF requests another MAC filter
but there is no space left on the device for it.

(backported from commit 68d6d4ab9a963cfaece88f9b4ea23be517ce46b0)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: update copyright to 2012

New year so bump the copyright date.

(cherry picked from commit 9497182051f261fe688bb2d672fdbc07ab3d5348)
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Add module parameter to allow untested and unsafe SFP+ modules

The X520 family of network devices, with the 82599 chip, support a
small number of Intel-verified SFP+ modules on their NICs.  To maintain
stability and quality, the current devices restrict untested 3rd party
SFP+ modules.

This patch introduces a module parameter for ixgbe to allow these untested
modules at the user's peril.  It also includes a warning to the syslog
alerting users that the modules aren't supported, and results may
vary.

CC: Jesper Dangaard Brouer <hawk@comx.dk>
(cherry picked from commit 8ef78adcb03b1fcb53c3bd62df4e96c1d2706c58)
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix register defines to correctly handle complex expressions

This patch is meant to address possible issues with the IXGBE register
defines generating incorrect values when given a complex expression for the
register offset.

(cherry picked from commit 795be954dcc4a72329c8591aafe5eadb03b264af)
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: add support for new 82599 device.

This device uses an already existing DevID but since it supports
WoL we need to add the Sub DevID.  It's support of WoL is limited
to the first port.

(cherry picked from commit 0e22d0437e6dea36c867b08ceb224c1cc98a45ab)
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: add support for new 82599 device id

Support for new 82599 based quad port adapter.

(cherry picked from commit 9e791e4a04c08868f02cd579a428a7268492e1b4)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: add write flush in ixgbe_clock_out_i2c_byte()

I2C access is timing critical. Always do a write flush after writing
to the I2CCTL register.

(cherry picked from commit 176f950d310c81b7fafd96aabe53704f778ce269)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix typo's

Saw typo in one message, so decided to run spell checker.

(cherry picked from commit 52f33af8ac479259f77abab6c535dac93ce654e8)
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix incorrect PHY register reads

Fix some register reads that had the opcode and register parameters swapped.
Also use define instead of a magic (0x3) number.

(cherry picked from commit c1085b10925a3271f90cb9619f22835e701e8657)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Remove function prototype for non-existent function

(cherry picked from commit 560f20dad7b46e6e8cd166d992679fc96fc50026)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: DCB: IEEE transitions may fail to reprogram hardware.

Transitioning through an IEEE DCBX version from a CEE DCBX
and back (CEE->IEEE->CEE) may leave IEEE attributes programmed
in the hardware. DCB uses a bit field in the set routines to
determine which attributes PG, PFC, APP need to be reprogrammed.
This is needed because user flow allows queueing a series
of changes and then reprogramming the hardware with the
entire set in one operation.

When transitioning from IEEE DCBX mode back into CEE DCBX
mode the PG and PFC bits need to be set so the possibly

different CEE attributes get programmed into the device.

This patch fixes broken logic that was evaluating to 0
and never setting any bits. Further this removes some
checks for num_tc in set routines. This logic only worked
when the number of traffic classes and user priorities
were equal. This is no longer the case for X540 devices.
Besides we can trust user input in this case if the
device is incorrectly configured the DCB bandwidths will
be incorrectly mapped but no OOPs, BUG, or hardware
failure will occur.

(cherry picked from commit 43497cc21889725a0fa0dcea1ce72ec567796f3a)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: DCBnl set_all, order of operations fix

The order of operations is important in DCBnl set_all(). When FCoE
is configured it uses the up2tc map to learn which queues to configure
the hardware offloads on. Therefore we need to setup the map before
configuring FCoE.

This is only seen when the both up2tc mappings and APP info are
configured simultaneously.

(backported from commit 4909fe979c4863bb748cd35e2cec9aab1e140a10)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix LED blink logic to check for link

Previously the driver would force link without checking whether the link was
already established. This caused some inconsistencies in the LED blink rate.

Do not force link if link is already up.

(cherry picked from commit 8d233633bab501567841aaa2935963fb908c9523)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix compile for kernel without CONFIG_PCI_IOV defined

Fix compiler errors and warnings with CONFIG_PCI_IOV defined and not
defined.

(cherry picked from commit 012641082b34433dac3cbb452e0a6ceccfd4643f)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: DCB, return max for IEEE traffic classes

Returning the max traffic classes on get requests simplifies
user space configurations because applications will know
explicitly how many traffic classes can be used.

Typical switch implementations use 2 or 3 traffic classes
so this not seen often today. And user space can learn
the number of traffic classes by return codes but this
allows user space to configure ixgbe correctly at the
start.

(cherry picked from commit 8599e251b3dd18c7bcb342d5b4acecc420f43606)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix reading of the buffer returned by the firmware

This patch fixes some issues found in the buffer read portion of
ixgbe_host_interface_command()

- use `bi` as the buffer index counter instead of `i`
- add conversion to native cpu byte ordering on register read
- fix conversion from bytes to dword
- use dword_len instead of buf_len when reading the register

(cherry picked from commit 331bcf45feb76d507a769d9d3b26ff5626804117)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix compiler warnings

Wrap SR-IOV specific functions in CONFIG_PCI_IOV to avoid compiler
warnings.

(cherry picked from commit 9487dc844054e1fc691fb82f4e19da337e2ca35e)
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Sibai Li <sibai.li@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix smatch splat due to missing NULL check

ixgbe_ieee_ets and ixgbe_ieee_pfc are intialized at
the same time. Do a check for both before configuring
IEEE802.1Qaz. Also max_frame was causing a sparse
warning resolved here as well.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
(cherry picked from commit b120818e652965669d3f1abaeaa5c3ccdfb28126)
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix disabling of Tx laser at probe

register_netdev() calls ndo_set_features() which may result in HW reset
which in turn will bring the laser back up.

This patch moves ixgbe_laser_tx_disable() below register_netdev()
in ixgbe_probe() to make sure laser is shut off on load.

(cherry picked from commit 93d3ce8fafb888702311fc8c5917faa4c25b8266)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Fix link issues caused by a reset while interface is down

Interface fails to obtain link on 82599 SFP in the following scenario:

1. Set advertised speed to GB:
ethtool -s eth0 advertise 0x20

2. Bring interface down
ip link set eth0 down

3. Issue any command that leads to a reset:
ethtool -t eth0

4. Bring link back up:
ip link set eth0 up

Following patch makes sure that the driver flaps the Tx laser every time
ixgbe_start_hw() is called, and not only when the speed is set.

(cherry picked from commit 232ef6bc451de2bc17c22fd116838cd89b94e1c1)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: change the eeprom version reported by ethtool

Use 32bit value starting at offset 0x2d for displaying the firmware
version in ethtool. This should work for all current ixgbe HW

(cherry picked from commit 15e5209f1c606e7c3e9b268f5c7b70b414a859cb)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: allow eeprom writes via ethtool

Implement support for ethtool -E

(cherry picked from commit 2fa5eef4d11383f1cfa99d31275e77dcf2d6a0a9)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix endianess when writing driver version to firmware

This patch makes sure that register writes are in little endian and
also converts the reads back to big-endian.

(cherry picked from commit 79488c58bb5aef58cfba1917617acf0db21c23a9)
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: fix skb truesize underestimation

ixgbe allocates half a page per skb fragment. We must account
PAGE_SIZE/2 increments on skb->truesize, not the actual frag length.

(cherry picked from commit 98130646770db42cd14c44ba0d7f2d0eb8078820)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

ixgbe: Correct check for change in FCoE priority

Correct a check for change in FCoE priority when IEEE mode DCB is in use.
In IEEE mode a different function has to be used to get the FCoE priority
mask. Also, the check for the mask assumed that only one priority was set.
In case there should be more than one, check just the bit.

These changes help avoid link flapping issues that can come up when IEEE
DCB is in use.

(cherry picked from commit 15d447ecaff457e6f89b459e70c0770b35b35533)
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>