platform/kernel/linux-rpi3.git
7 years agoe1000e: add statistic indicating number of skipped Tx timestamps
Jacob Keller [Wed, 3 May 2017 17:28:57 +0000 (10:28 -0700)]
e1000e: add statistic indicating number of skipped Tx timestamps

The e1000e driver can only handle one Tx timestamp request at a time.
This means it is possible for an application timestamp request to be
ignored.

There is no easy way for an administrator to determine if this occurred.
Add a new statistic which tracks this, tx_hwtstamp_skipped.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: avoid permanent lock of *_PTP_TX_IN_PROGRESS
Jacob Keller [Wed, 3 May 2017 17:28:55 +0000 (10:28 -0700)]
igb: avoid permanent lock of *_PTP_TX_IN_PROGRESS

The igb driver uses a state bit lock to avoid handling more than one Tx
timestamp request at once. This is required because hardware is limited
to a single set of registers for Tx timestamps.

The state bit lock is not properly cleaned up during
igb_xmit_frame_ring() if the transmit fails such as due to DMA or TSO
failure. In some hardware this results in blocking timestamps until the
service task times out. In other hardware this results in a permanent
lock of the timestamp bit because we never receive an interrupt
indicating the timestamp occurred, since indeed the packet was never
transmitted.

Fix this by checking for DMA and TSO errors in igb_xmit_frame_ring() and
properly cleaning up after ourselves when these occur.

Reported-by: Reported-by: David Mirabito <davidm@metamako.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: fix race condition with PTP_TX_IN_PROGRESS bits
Jacob Keller [Wed, 3 May 2017 17:28:52 +0000 (10:28 -0700)]
igb: fix race condition with PTP_TX_IN_PROGRESS bits

Hardware related to the igb driver has a limitation of only handling one
Tx timestamp at a time. Thus, the driver uses a state bit lock to
enforce that only one timestamp request is honored at a time.

Unfortunately this suffers from a simple race condition. The bit lock is
not cleared until after skb_tstamp_tx() is called notifying the stack of
a new Tx timestamp. Even a well behaved application which sends only one
timestamp request at once and waits for a response might wake up and
send a new packet before the bit lock is cleared. This results in
needlessly dropping some Tx timestamp requests.

We can fix this by unlocking the state bit as soon as we read the
Timestamp register, as this is the first point at which it is safe to
unlock.

To avoid issues with the skb pointer, we'll use a copy of the pointer
and set the global variable in the driver structure to NULL first. This
ensures that the next timestamp request does not modify our local copy
of the skb pointer.

This ensures that well behaved applications do not accidentally race
with the unlock bit. Obviously an application which sends multiple Tx
timestamp requests at once will still only timestamp one packet at
a time. Unfortunately there is nothing we can do about this.

Reported-by: David Mirabito <davidm@metamako.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000e: fix race condition around skb_tstamp_tx()
Jacob Keller [Wed, 3 May 2017 17:28:50 +0000 (10:28 -0700)]
e1000e: fix race condition around skb_tstamp_tx()

The e1000e driver and related hardware has a limitation on Tx PTP
packets which requires we limit to timestamping a single packet at once.
We do this by verifying that we never request a new Tx timestamp while
we still have a tx_hwtstamp_skb pointer.

Unfortunately the driver suffers from a race condition around this. The
tx_hwtstamp_skb pointer is not set to NULL until after skb_tstamp_tx()
is called. This function notifies the stack and applications of a new
timestamp. Even a well behaved application that only sends a new request
when the first one is finished might be woken up and possibly send
a packet before we can free the timestamp in the driver again. The
result is that we needlessly ignore some Tx timestamp requests in this
corner case.

Fix this by assigning the tx_hwtstamp_skb pointer prior to calling
skb_tstamp_tx() and use a temporary pointer to hold the timestamped skb
until that function finishes. This ensures that the application is not
woken up until the driver is ready to begin timestamping a new packet.

This ensures that well behaved applications do not accidentally race
with condition to skip Tx timestamps. Obviously an application which
sends multiple Tx timestamp requests at once will still only timestamp
one packet at a time. Unfortunately there is nothing we can do about
this.

Reported-by: David Mirabito <davidm@metamako.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: mark PM functions as __maybe_unused
Arnd Bergmann [Thu, 27 Apr 2017 19:09:52 +0000 (21:09 +0200)]
igb: mark PM functions as __maybe_unused

The new wake function is only used by the suspend/resume handlers that
are defined in inside of an #ifdef, which can cause this harmless
warning:

drivers/net/ethernet/intel/igb/igb_main.c:7988:13: warning: 'igb_deliver_wake_packet' defined but not used [-Wunused-function]

Removing the #ifdef, instead using a __maybe_unused annotation
simplifies the code and avoids the warning.

Fixes: b90fa8763560 ("igb: Enable reading of wake up packet")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Explicitly select page 0 at initialization
Matwey V Kornilov [Thu, 24 Nov 2016 10:32:48 +0000 (13:32 +0300)]
igb: Explicitly select page 0 at initialization

The functions igb_read_phy_reg_gs40g/igb_write_phy_reg_gs40g (which were
removed in 2a3cdea) explicitly selected the required page at every phy_reg
access. Currently, igb_get_phy_id_82575 relays on the fact that page 0 is
already selected. The assumption is not fulfilled for my Lex 3I380CW
motherboard with integrated dual i211 based gigabit ethernet. This leads to igb
initialization failure and network interfaces are not working:

    igb: Intel(R) Gigabit Ethernet Network Driver - version 5.4.0-k
    igb: Copyright (c) 2007-2014 Intel Corporation.
    igb: probe of 0000:01:00.0 failed with error -2
    igb: probe of 0000:02:00.0 failed with error -2

In order to fix it, we explicitly select page 0 before first access to phy
registers.

See also: https://bugzilla.suse.com/show_bug.cgi?id=1009911
See also: http://www.lex.com.tw/products/pdf/3I380A&3I380CW.pdf

Fixes: 2a3cdea ("igb: Remove GS40G specific defines/functions")
Cc: <stable@vger.kernel.org> # 4.5+
Signed-off-by: Matwey V Kornilov <matwey@sai.msu.ru>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agomdio: mux: fix an incorrect less than zero error check using a u32
Colin Ian King [Mon, 5 Jun 2017 18:08:04 +0000 (19:08 +0100)]
mdio: mux: fix an incorrect less than zero error check using a u32

The u32 variable v is being checked to see if an error return is
less than zero and this check has no effect because it is unsigned.
Fix this by making v and int (this also matches the type of
cb->bus_number which is assigned to the value in v).

Detected by CoverityScan, CID#1440454 ("Unsigned compared against zero")

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet-next: stmmac: dwmac-sun8i: ensure the EPHY is properly reseted
Icenowy Zheng [Sun, 4 Jun 2017 17:53:23 +0000 (01:53 +0800)]
net-next: stmmac: dwmac-sun8i: ensure the EPHY is properly reseted

The EPHY may be already enabled by bootloaders which have Ethernet
capability (e.g. current U-Boot). Thus it should be reseted properly
before doing the enabling sequence in the dwmac-sun8i driver, otherwise
the EMAC reset process may fail if no cable is plugged, and then fail
the dwmac-sun8i probing.

Tested on Orange Pi PC, One and Zero. All the boards fail to have
dwmac-sun8i probed with "EMAC reset timeout" without cable plugged
before, and with this fix they're now all able to successfully probe the
EMAC without cable plugged and then use the connection after a cable is
hot-plugged in.

Fixes: 9f93ac8d408 ("net-next: stmmac: Add dwmac-sun8i")
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Tested-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Reviewed-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Acked-by: is not as formal as Signed-off-by:. It is a record that the acker
Reviewed-by: is similar.
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/3com: Make el3_netdev_get_ecmd return void
yuval.shaia@oracle.com [Sun, 4 Jun 2017 17:24:46 +0000 (20:24 +0300)]
net/3com: Make el3_netdev_get_ecmd return void

Make return value void since function never returns meaningfull value.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/{mii, smsc}: Make mii_ethtool_get_link_ksettings and smc_netdev_get_ecmd return...
yuval.shaia@oracle.com [Sun, 4 Jun 2017 17:22:00 +0000 (20:22 +0300)]
net/{mii, smsc}: Make mii_ethtool_get_link_ksettings and smc_netdev_get_ecmd return void

Make return value void since functions never returns meaningfull value.

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/dec: Make __de_get_link_ksettings return void
yuval.shaia@oracle.com [Sun, 4 Jun 2017 17:08:51 +0000 (20:08 +0300)]
net/dec: Make __de_get_link_ksettings return void

Make return value void since function never return meaningfull value

Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: sched: select cls when cls_act is enabled
Jiri Pirko [Sun, 4 Jun 2017 16:49:28 +0000 (18:49 +0200)]
net: sched: select cls when cls_act is enabled

It really makes no sense to have cls_act enabled without cls. In that
case, the cls_act code is dead. So select it.

This also fixes an issue recently reported by kbuild robot:
[linux-next:master 1326/4151] net/sched/act_api.c:37:18: error: implicit declaration of function 'tcf_chain_get'

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Fixes: db50514f9a9c ("net: sched: add termination action to allow goto chain")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agogenetlink: remove ops_list from genetlink header.
Rosen, Rami [Sun, 4 Jun 2017 12:20:01 +0000 (15:20 +0300)]
genetlink: remove ops_list from genetlink header.

commit d91824c08fbc ("genetlink: register family ops as array") removed the
ops_list member from both genl_family and genl_ops; while the
documentation of genl_family was updated accordingly by this patch,
ops_list remained in the documentation of the genl_ops object.
This patch fixes it by removing ops_list from genl_ops documentation.

Signed-off-by: Rami Rosen <rami.rosen@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mlxsw-Minor-cleanup'
David S. Miller [Mon, 5 Jun 2017 03:49:49 +0000 (23:49 -0400)]
Merge branch 'mlxsw-Minor-cleanup'

Jiri Pirko says:

====================
mlxsw: Minor cleanup

Fix small issues I noticed during the refactoring.

First patch adds file name comments in the header file to make it clear
what goes where. Second patch fixes a typo and third patch simply aligns
RIF index allocation with similar allocations in the driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Align RIF index allocation with existing code
Ido Schimmel [Sun, 4 Jun 2017 14:53:40 +0000 (16:53 +0200)]
mlxsw: spectrum_router: Align RIF index allocation with existing code

The way we usually allocate an index is by letting the allocation
function return an error instead of an invalid index.

Do the same for RIF index.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: Fix typo inside enumeration
Ido Schimmel [Sun, 4 Jun 2017 14:53:39 +0000 (16:53 +0200)]
mlxsw: Fix typo inside enumeration

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Tidy up header file
Ido Schimmel [Sun, 4 Jun 2017 14:53:38 +0000 (16:53 +0200)]
mlxsw: spectrum: Tidy up header file

Make it clear where functions are defined and move misplaced declaration
to their correct place.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum: Rename the firmware file
Yotam Gigi [Sun, 4 Jun 2017 14:49:58 +0000 (16:49 +0200)]
mlxsw: spectrum: Rename the firmware file

Change the firmware file name to be in "mellanox" directory.

This commit is a followup to the linux-firmware commit a4c72696f5f4
("Mellanox: Add firmware for mlxsw_spectrum")

Signed-off-by: Yotam Gigi <yotamg@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'qed-vf-xdp'
David S. Miller [Mon, 5 Jun 2017 03:08:32 +0000 (23:08 -0400)]
Merge branch 'qed-vf-xdp'

Yuval Mintz says:

qed*: Support VF XDP attachment

====================
Each driver queue [Rx, Tx, XDP-forwarding] requires an allocated HW/FW
connection + configured queue-zone.

VF handling by the PF has several limitations that prevented adding the
capability to perform XDP at driver-level:

 - The VF assumes there's 1-to-1 correspondance between the VF queue and
   the used connection, meaning q<x> is always going to use cid<x>,
   whereas for its own queues the PF is acquiring a new cid per each new
   queue.

 - There's a 1-to-1 correspondate between the VF-queues and the HW queue
   zones. While this is necessary for Rx-queues [as the queue-zone
   contains the producer], transmission queues can share the underlaying
   queue-zone [only shared configuration is coalescing].
   But all VF<->PF communication mechanisms assume there's a single
   identifier that identify a queue [as queue-zone == queue], while
   sharing queue-zones requires passing additional information.

 - VFs currently don't try mapping a doorbell bar - there's a small
   doorbell window in the regview allowing VFs to doorbell up to 16
   connections; but this window isn's wide enough for the added XDP
   forwarding queues.

This series is going to add the necessary infrastrucutre to finally let
our VFs support XDP assuming both the PF and VF drivers are sufficiently
new [Legacy support would be retained both for older VFs and older PFs,
but both will be needed for this new support to work].
Basically, the various database driver maintains for its queue-cids
would be revised, and queue-cids would be identified using the
(queue-zone, unique index) pair. The TLV mechanism would then be
extended to allow VFs to communicate that unique-index as well as the
already provided queue-zone. Finally, the VFs would try to map their
doorbell bar and inform their PF that they're using it.

Almost all the changes are in qed, with exception of #3 [which does some
cleanup in qede as well] and #11 that actually enables the feature.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqede: VF XDP support
Mintz, Yuval [Sun, 4 Jun 2017 10:31:09 +0000 (13:31 +0300)]
qede: VF XDP support

This introduces 2 changes needed for XDP to be supported for VFs:

 a. On VF-side, publish the NDO based on qed outputs

 b. On PF-side, request qed to allocate sufficient cids per-VF
    to allow the child vfs to support it

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: VF XDP support
Mintz, Yuval [Sun, 4 Jun 2017 10:31:08 +0000 (13:31 +0300)]
qed: VF XDP support

The final addition on the qed front -
 - VFs would now require their PFs to provide multiple CIDs
 - Based on the availability of connections from PF, determine whether
   XDP is feasible and share it with qede via dev_info.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: VFs to try utilizing the doorbell bar
Mintz, Yuval [Sun, 4 Jun 2017 10:31:07 +0000 (13:31 +0300)]
qed: VFs to try utilizing the doorbell bar

VFs are currently not mapping their doorbell bar, instead relying
on the small doorbell window they have in their limited regview bar.

In order to increase the number of possible Tx connections [queues]
employeed by VF past 16, we need to start using the doorbell bar if
one such is exposed - VF would communicate this fact to PF which would
return the size-bar internally configured into chip, according to
which the VF would decide whether to actually utilize the doorbell
bar.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Multiple qzone queues for VFs
Mintz, Yuval [Sun, 4 Jun 2017 10:31:06 +0000 (13:31 +0300)]
qed: Multiple qzone queues for VFs

This adds the infrastructure for supporting VFs that want to open
multiple transmission queues on the same queue-zone.
At this point, there are no VFs that actually request this functionality,
but later patches would remedy that.

 a. VF and PF would communicate the capability during ACQUIRE;
    Legacy VFs would continue on behaving as they do today

 b. PF would communicate number of supported CIDs to the VF
    and would enforce said limitation

 c. Whenever VF passes a request for a given queue configuration
    it would also pass an associated index within said queue-zone

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: IOV db support multiple queues per qzone
Mintz, Yuval [Sun, 4 Jun 2017 10:31:05 +0000 (13:31 +0300)]
qed: IOV db support multiple queues per qzone

Allow the infrastructure a PF maintains for each one of its VFs
to support multiple queue-cids on a single queue-zone.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Make VF legacy a bitfield
Mintz, Yuval [Sun, 4 Jun 2017 10:31:04 +0000 (13:31 +0300)]
qed: Make VF legacy a bitfield

Until now we used to have a single VF legacy compatibility mode,
one that affected the place of the Rx producers of those VFs [mostly].

As PF would soon support allocating CIDs for VFs instead of having
a static CID<->queue configuration for them, we'll need to have
an additional legacy mode since existing VFs would need to continue
on using the older mode of operation.

Change the infrastrucutre so that the legacy would be able to indicate
which of the legacy behaviors is needed for a given VF.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Assign a unique per-queue index to queue-cid
Mintz, Yuval [Sun, 4 Jun 2017 10:31:03 +0000 (13:31 +0300)]
qed: Assign a unique per-queue index to queue-cid

When a queue-cid is allocated, assign an index inside that's
CID's queue-zone.

For PFs and VFS, this number is going to be unique and derive
from a per-queue-zone bitmap, while for PF's VFs queues the
number is currently going to constant; Later, we'd add the
capability of a VF to communicate such an index to its PF.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Pass vf_params when creating a queue-cid
Mintz, Yuval [Sun, 4 Jun 2017 10:31:02 +0000 (13:31 +0300)]
qed: Pass vf_params when creating a queue-cid

We're going to need additional information for queue-cids
that a PF creates for its VFs, so start by refactoring existing
logic used for initializing said struct into receiving a structure
encapsulating the VF-specific information that needs to be provided.

This also introduces QED_QUEUE_CID_SELF - each queue-cid would hold
an indication to whether it belongs to the hw-function holding it
[whether that's a PF or a VF], or else what's the VF id it belongs
to.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed*: L2 interface to use the SB structures directly
Mintz, Yuval [Sun, 4 Jun 2017 10:31:01 +0000 (13:31 +0300)]
qed*: L2 interface to use the SB structures directly

Part of an effort of a cleaner seperation between qed and the protocol
drivers, the L2 interface is to use the SB structure for initialization
purposes opaquely.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Create L2 queue database
Mintz, Yuval [Sun, 4 Jun 2017 10:31:00 +0000 (13:31 +0300)]
qed: Create L2 queue database

First step in allowing a single PF/VF to open multiple queues on
the same queue zone is to add per-hwfn database of queue-cids
as a two-dimensional array where entry would be according to
[queue zone][internal index].

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Add bitmaps for VF CIDs
Mintz, Yuval [Sun, 4 Jun 2017 10:30:59 +0000 (13:30 +0300)]
qed: Add bitmaps for VF CIDs

Each PF has a bitmap for its own ranges of CIDs, to allow easy grabbing
of an available CID when such is needed. But VFs are not using the same
mechanism, instead relying on hard-coded CIDs [ queue-index == cid ].

As an infrastructure step toward increasing number of CIDs of VFs,
the PF is going to maintain bitmaps for the VF CIDs as well -
the bitmaps would be per-VF and the ranges would be the same [in HW all
VFs of a given PF have the same mapping of CIDs, and the HW is capable
of distinguishing between those according to the VF index]

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'skb-sgvec-overflow'
David S. Miller [Mon, 5 Jun 2017 03:01:48 +0000 (23:01 -0400)]
Merge branch 'skb-sgvec-overflow'

Jason A. Donenfeld says:

====================
net: Avoiding stack overflow in skb_to_sgvec

The recent bug with macsec and historical one with virtio have
indicated that letting skb_to_sgvec trounce all over an sglist
without checking the length is probably a bad idea. And it's not
necessary either: an sglist already explicitly marks its last
item, and the initialization functions are diligent in doing so.
Thus there's a clear way of avoiding future overflows.

So, this patchset, from a high level, makes skb_to_sgvec return
a potential error code, and then adjusts all callers to check
for the error code. There are two situations in which skb_to_sgvec
might return such an error:

   1) When the passed in sglist is too small; and
   2) When the passed in skbuff is too deeply nested.

So, the first patch in this series handles the issues with
skb_to_sgvec directly, and the remaining ones then handle the call
sites.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovirtio_net: check return value of skb_to_sgvec always
Jason A. Donenfeld [Sun, 4 Jun 2017 02:16:26 +0000 (04:16 +0200)]
virtio_net: check return value of skb_to_sgvec always

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Reviewed-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomacsec: check return value of skb_to_sgvec always
Jason A. Donenfeld [Sun, 4 Jun 2017 02:16:25 +0000 (04:16 +0200)]
macsec: check return value of skb_to_sgvec always

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agorxrpc: check return value of skb_to_sgvec always
Jason A. Donenfeld [Sun, 4 Jun 2017 02:16:24 +0000 (04:16 +0200)]
rxrpc: check return value of skb_to_sgvec always

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipsec: check return value of skb_to_sgvec always
Jason A. Donenfeld [Sun, 4 Jun 2017 02:16:23 +0000 (04:16 +0200)]
ipsec: check return value of skb_to_sgvec always

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoskbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
Jason A. Donenfeld [Sun, 4 Jun 2017 02:16:22 +0000 (04:16 +0200)]
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow

This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.

As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.

While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-Add-BPF-support-to-all-perf_event'
David S. Miller [Mon, 5 Jun 2017 01:58:24 +0000 (21:58 -0400)]
Merge branch 'bpf-Add-BPF-support-to-all-perf_event'

Merge branch 'bpf-Add-BPF-support-to-all-perf_event'

Alexei Starovoitov says:

====================
bpf: Add BPF support to all perf_event

v3->v4: one more tweak to reject unsupported events at map
update time as Peter suggested

v2->v3: more refactoring to address Peter's feedback.
Now all perf_events are attachable and readable

v1->v2: address Peter's feedback. Refactor patch 1 to allow attaching
bpf programs to all event types and reading counters from all of them as well
patch 2 - more tests
patch 3 - address Dave's feedback and document bpf_perf_event_read()
and bpf_perf_event_output() properly
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: update perf event helper functions documentation
Teng Qin [Sat, 3 Jun 2017 04:03:54 +0000 (21:03 -0700)]
bpf: update perf event helper functions documentation

This commit updates documentation of the bpf_perf_event_output and
bpf_perf_event_read helpers to match their implementation.

Signed-off-by: Teng Qin <qinteng@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: add tests for more perf event types
Teng Qin [Sat, 3 Jun 2017 04:03:53 +0000 (21:03 -0700)]
samples/bpf: add tests for more perf event types

$ trace_event

tests attaching BPF program to HW_CPU_CYCLES, SW_CPU_CLOCK, HW_CACHE_L1D and other events.
It runs 'dd' in the background while bpf program collects user and kernel
stack trace on counter overflow.
User space expects to see sys_read and sys_write in the kernel stack.

$ tracex6

tests reading of various perf counters from BPF program.

Both tests were refactored to increase coverage and be more accurate.

Signed-off-by: Teng Qin <qinteng@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoperf, bpf: Add BPF support to all perf_event types
Alexei Starovoitov [Sat, 3 Jun 2017 04:03:52 +0000 (21:03 -0700)]
perf, bpf: Add BPF support to all perf_event types

Allow BPF_PROG_TYPE_PERF_EVENT program types to attach to all
perf_event types, including HW_CACHE, RAW, and dynamic pmu events.
Only tracepoint/kprobe events are treated differently which require
BPF_PROG_TYPE_TRACEPOINT/BPF_PROG_TYPE_KPROBE program types accordingly.

Also add support for reading all event counters using
bpf_perf_event_read() helper.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoneigh: Really delete an arp/neigh entry on "ip neigh delete" or "arp -d"
Sowmini Varadhan [Fri, 2 Jun 2017 16:01:49 +0000 (09:01 -0700)]
neigh: Really delete an arp/neigh entry on "ip neigh delete" or "arp -d"

The command
  # arp -s 62.2.0.1 a:b:c:d:e:f dev eth2
adds an entry like the following (listed by "arp -an")
  ? (62.2.0.1) at 0a:0b:0c:0d:0e:0f [ether] PERM on eth2
but the symmetric deletion command
  # arp -i eth2 -d 62.2.0.1
does not remove the PERM entry from the table, and instead leaves behind
  ? (62.2.0.1) at <incomplete> on eth2

The reason is that there is a refcnt of 1 for the arp_tbl itself
(neigh_alloc starts off the entry with a refcnt of 1), thus
the neigh_release() call from arp_invalidate() will (at best) just
decrement the ref to 1, but will never actually free it from the
table.

To fix this, we need to do something like neigh_forced_gc: if
the refcnt is 1 (i.e., on the table's ref), remove the entry from
the table and free it. This patch refactors and shares common code
between neigh_forced_gc and the newly added neigh_remove_one.

A similar issue exists for IPv6 Neighbor Cache entries, and is fixed
in a similar manner by this patch.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Reviewed-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: smsc: Implement PHY statistics
Andrew Lunn [Sat, 3 Jun 2017 18:00:36 +0000 (20:00 +0200)]
net: phy: smsc: Implement PHY statistics

Most of the PHYs supported by the SMSC driver have a counter of symbol
errors. This is 16 bit wide and wraps around when it reaches its
maximum value.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-By: Woojung Huh <Woojung.Huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-Fixes-for-mv88e6161'
David S. Miller [Mon, 5 Jun 2017 01:34:17 +0000 (21:34 -0400)]
Merge branch 'dsa-Fixes-for-mv88e6161'

Andrew Lunn says:

====================
dsa: Fixes for mv88e6161

Testing a board with an mv88e6161 turned up two issues. The PHYs were
not found, because the wrong method to access them was used. The
statistics did not work, because the wrong snapshot method was used
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: mv88e6161 uses mv88e6320 stats snapshot
Andrew Lunn [Fri, 2 Jun 2017 21:22:46 +0000 (23:22 +0200)]
net: dsa: mv88e6xxx: mv88e6161 uses mv88e6320 stats snapshot

The mv88e6161 was using the wrong method to perform statistics
snapshot.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: 6161 uses global 2 for PHY access
Andrew Lunn [Fri, 2 Jun 2017 21:22:45 +0000 (23:22 +0200)]
net: dsa: mv88e6xxx: 6161 uses global 2 for PHY access

Access to the internal PHYs of the 6161 and 6123 go through global 2
SMI registers. Fix the ops structure.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-mv88e6xxx-move-registers-macros'
David S. Miller [Mon, 5 Jun 2017 00:07:41 +0000 (20:07 -0400)]
Merge branch 'dsa-mv88e6xxx-move-registers-macros'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: move registers macros

This patchset brings no functional changes.

It is the first step of a cleanup renaming the chip header file and
moving the Register definitions _as is_ in their proper header files.

A following patchset will prefix them with the appropriate model
(MV88E6XXX_ or e.g. MV88E6390_) to respect an implicit namespace and
easily identify model subtleties in registers layout, as correctly done
in the newly added serdes.h header.
====================

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: move the Global 2 macros
Vivien Didelot [Fri, 2 Jun 2017 21:06:19 +0000 (17:06 -0400)]
net: dsa: mv88e6xxx: move the Global 2 macros

Move the GLOBAL2_* macros where they belong, in the related global2.h
header.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: move the Global 1 macros
Vivien Didelot [Fri, 2 Jun 2017 21:06:18 +0000 (17:06 -0400)]
net: dsa: mv88e6xxx: move the Global 1 macros

Move the GLOBAL_* macros where they belong, in the related global1.h
header. Include it in global2.c which uses GLOBAL_STATUS_IRQ_DEVICE.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: move the Port macros
Vivien Didelot [Fri, 2 Jun 2017 21:06:17 +0000 (17:06 -0400)]
net: dsa: mv88e6xxx: move the Port macros

Move the PORT_* macros where they belong, in the related port.h header.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: move PHY macros
Vivien Didelot [Fri, 2 Jun 2017 21:06:16 +0000 (17:06 -0400)]
net: dsa: mv88e6xxx: move PHY macros

Move the PHY_* macros where they belong, in the related phy.h header.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: rename chip header
Vivien Didelot [Fri, 2 Jun 2017 21:06:15 +0000 (17:06 -0400)]
net: dsa: mv88e6xxx: rename chip header

The mv88e6xxx.h is meant to contains the chip structures and data.
Rename it to chip.h, as for other source/header pairs of the driver.

At the same time, ensure that relative header inclusions are separated
by a newline and sorted alphabetically.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-cleanups-before-multi-CPU-port'
David S. Miller [Mon, 5 Jun 2017 00:05:15 +0000 (20:05 -0400)]
Merge branch 'dsa-cleanups-before-multi-CPU-port'

Florian Fainelli says:

====================
net: dsa: Cleanups before multi-CPU port

This patch series does a bunch of cleanups before we start adding support
for multiple CPU ports.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Initialize all CPU and enabled ports masks in dsa_ds_parse()
Florian Fainelli [Fri, 2 Jun 2017 19:31:23 +0000 (12:31 -0700)]
net: dsa: Initialize all CPU and enabled ports masks in dsa_ds_parse()

There was no reason for duplicating the code that initializes
ds->enabled_port_mask in both dsa_parse_ports_dn() and
dsa_parse_ports(), instead move this to dsa_ds_parse() which is early
enough before ops->setup() has run.

While at it, we can now make dsa_is_cpu_port() check ds->cpu_port_mask
which is a step towards being multi-CPU port capable.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Consistently use dsa_port for dsa_*_port_{apply, unapply}
Florian Fainelli [Fri, 2 Jun 2017 19:31:22 +0000 (12:31 -0700)]
net: dsa: Consistently use dsa_port for dsa_*_port_{apply, unapply}

We have all the information we need in dsa_port, so use it instead of
repeating the same arguments over and over again.

Suggested-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Pass dsa_port reference to ethtool setup/restore
Florian Fainelli [Fri, 2 Jun 2017 19:31:21 +0000 (12:31 -0700)]
net: dsa: Pass dsa_port reference to ethtool setup/restore

We do not need to have a reference to a dsa_switch, instead we should
pass a reference to a CPU dsa_port, change that. This is a preliminary
change to better support multiple CPU ports.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agorxrpc: remove redundant proc_remove call
Colin Ian King [Fri, 2 Jun 2017 15:22:27 +0000 (16:22 +0100)]
rxrpc: remove redundant proc_remove call

The proc_remove call is dead code as it occurs after a return and
hence can never be called. Remove it.

Detected by CoverityScan, CID#1437743 ("Logically dead code")

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodccp: consistently use dccp_write_space()
Eric Dumazet [Fri, 2 Jun 2017 15:02:26 +0000 (08:02 -0700)]
dccp: consistently use dccp_write_space()

DCCP uses dccp_write_space() for sk->sk_write_space method.

Unfortunately a passive connection (as provided by accept())
is using the generic sk_stream_write_space() function.

Lets simply inherit sk->sk_write_space from the parent
instead of forcing the generic one.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: marvell: make some functions static
Colin Ian King [Fri, 2 Jun 2017 14:13:34 +0000 (15:13 +0100)]
net: phy: marvell: make some functions static

functions m88e1510_get_temp_critical, m88e1510_set_temp_critical and
m88e1510_get_temp_alarm can be made static as they not need to be
in global scope.

Cleans up sparse warnings:
 "symbol 'm88e1510_get_temp_alarm' was not declared. Should it be static?"
 "symbol 'm88e1510_get_temp_critical' was not declared. Should it be
  static?"

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetxen: remove writeq/readq function definitions
LABBE Corentin [Fri, 2 Jun 2017 11:36:27 +0000 (13:36 +0200)]
netxen: remove writeq/readq function definitions

Instead of rewriting write/readq, use linux/io-64-nonatomic-lo-hi.h which
already have them.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet-procfs: Use vsnprintf extension %phN
Joe Perches [Fri, 2 Jun 2017 09:40:44 +0000 (02:40 -0700)]
net-procfs: Use vsnprintf extension %phN

Save a bit of code by using the kernel extension.

$ size net/core/net-procfs.o*
   text    data     bss     dec     hex filename
   3701     120       0    3821     eed net/core/net-procfs.o.new
   3764     120       0    3884     f2c net/core/net-procfs.o.old

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomdio: mux: make child bus walking more permissive and errors more verbose
Jon Mason [Wed, 31 May 2017 19:44:50 +0000 (15:44 -0400)]
mdio: mux: make child bus walking more permissive and errors more verbose

If any errors are encountered while walking the device tree structure of
the MDIO bus for children, the code may silently continue, silently
exit, or throw an error and exit.  This make it difficult for device
tree writers to know there is an error.  Also, it makes any error in a
child entry of the MDIO bus be fatal for all entries.  Instead, we
should provide verbose errors describing the error and then attempt to
continue if it all possible.  Also, use of_mdio_parse_addr()

Signed-off-by: Jon Mason <jon.mason@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'net-dissection-and-matching-on-tos-and-ttl'
David S. Miller [Sun, 4 Jun 2017 22:12:24 +0000 (18:12 -0400)]
Merge branch 'net-dissection-and-matching-on-tos-and-ttl'

Or Gerlitz says:

====================
net: add support for dissection and matching on ip tos and ttl

The 1st two patches enable matching/classifying on ip tos and ttl by
the flow dissector and flower. The other two patches offload matching
on tcp flags and ip tos in mlx5.

The mlx5 patches touch single file/function and not interfere with
other inflight mlx5 submissions.

V2:  repost as asked by Dave.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx5e: Offload TC matching on ip tos / traffic-class
Or Gerlitz [Thu, 1 Jun 2017 18:37:40 +0000 (21:37 +0300)]
net/mlx5e: Offload TC matching on ip tos / traffic-class

Enable offloading of TC matching on ipv4 tos or ipv6 traffic-class.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/mlx5e: Offload TC matching on tcp flags
Or Gerlitz [Thu, 1 Jun 2017 18:37:39 +0000 (21:37 +0300)]
net/mlx5e: Offload TC matching on tcp flags

Enable offloading of TC matching on tcp flags.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: cls_flower: add support for matching on ip tos and ttl
Or Gerlitz [Thu, 1 Jun 2017 18:37:38 +0000 (21:37 +0300)]
net/sched: cls_flower: add support for matching on ip tos and ttl

Benefit from the support of ip header fields dissection and
allow users to set rules matching on ipv4 tos and ttl or
ipv6 traffic-class and hoplimit.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/flow_dissector: add support for dissection of misc ip header fields
Or Gerlitz [Thu, 1 Jun 2017 18:37:37 +0000 (21:37 +0300)]
net/flow_dissector: add support for dissection of misc ip header fields

Add support for dissection of ip tos and ttl and ipv6 traffic-class
and hoplimit. Both are dissected into the same struct.

Uses similar call to ip dissection function as with tcp, arp and others.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: remove unnecessary skb_reset_tail_pointer()
Eric Dumazet [Thu, 1 Jun 2017 21:18:36 +0000 (14:18 -0700)]
tcp: remove unnecessary skb_reset_tail_pointer()

__pskb_trim_head() does not need to reset skb tail pointer.

Also change the comments, __pskb_pull_head() does not exist.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: Remove the capability check for cgroup skb eBPF program
Chenbo Feng [Thu, 1 Jun 2017 01:16:00 +0000 (18:16 -0700)]
bpf: Remove the capability check for cgroup skb eBPF program

Currently loading a cgroup skb eBPF program require a CAP_SYS_ADMIN
capability while attaching the program to a cgroup only requires the
user have CAP_NET_ADMIN privilege. We can escape the capability
check when load the program just like socket filter program to make
the capability requirement consistent.

Change since v1:
Change the code style in order to be compliant with checkpatch.pl
preference

Signed-off-by: Chenbo Feng <fengc@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: Allow CGROUP_SKB eBPF program to access sk_buff
Chenbo Feng [Thu, 1 Jun 2017 01:15:59 +0000 (18:15 -0700)]
bpf: Allow CGROUP_SKB eBPF program to access sk_buff

This allows cgroup eBPF program to classify packet based on their
protocol or other detail information. Currently program need
CAP_NET_ADMIN privilege to attach a cgroup eBPF program, and A
process with CAP_NET_ADMIN can already see all packets on the system,
for example, by creating an iptables rules that causes the packet to
be passed to userspace via NFLOG.

Signed-off-by: Chenbo Feng <fengc@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: use of_mdio_parse_addr
Jon Mason [Wed, 31 May 2017 19:43:30 +0000 (15:43 -0400)]
net: phy: use of_mdio_parse_addr

use of_mdio_parse_addr() in place of an OF read of reg and a bounds
check (which is litterally the exact same thing that
of_mdio_parse_addr() does)

Signed-off-by: Jon Mason <jon.mason@broadcom.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: use TS opt on RTTs for congestion control
Yuchung Cheng [Wed, 31 May 2017 18:30:53 +0000 (11:30 -0700)]
tcp: use TS opt on RTTs for congestion control

Currently when a data packet is retransmitted, we do not compute an
RTT sample for congestion control due to Kern's check. Therefore the
congestion control that uses RTT signals may not receive any update
during loss recovery which could last many round trips. For example,
BBR and Vegas may not be able to update its min RTT estimation if the
network path has shortened until it recovers from losses. This patch
mitigates that by using TCP timestamp options for RTT measurement
for congestion control. Note that we already use timestamps for
RTT estimation.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: Fix checkpatch errors with references crossing single line
Satanand Burla [Wed, 31 May 2017 17:45:15 +0000 (10:45 -0700)]
liquidio: Fix checkpatch errors with references crossing single line

Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Derek Chickles <derek.chickles@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotools: hv: set hotplug for VF on Suse
sixiao@microsoft.com [Wed, 31 May 2017 17:28:27 +0000 (10:28 -0700)]
tools: hv: set hotplug for VF on Suse

On HyperV, the VF interface can be offered by a host at any time.
Mark the VF interface as hotplug, to make sure it will be brought up
automatically when it is registered.

Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotools: hv: set allow-hotplug for VF on Ubuntu
sixiao@microsoft.com [Wed, 31 May 2017 17:28:13 +0000 (10:28 -0700)]
tools: hv: set allow-hotplug for VF on Ubuntu

On HyperV, the VF interface can be offered by a host at any time.
Mark the VF interface as hotplug, to make sure it will be brought up
automatically when it is registered.

Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: VF interrupt initialization cleanup
Rick Farrington [Wed, 31 May 2017 16:48:09 +0000 (09:48 -0700)]
liquidio: VF interrupt initialization cleanup

Set initialization state variable to (reflect interrupt initialization) at
correct time (immediately after having configured interrupts).  This fixes
problem of inconsistent IRQ allocation in case of [obscure] failure when
negotiating with PF driver during init.

Clean-up of interrupt enablement during initialization & avoid potential
race condition with chip-specific code (i.e. perform interrupt control in
main driver module).  Added explanatory comments regarding interrupt
enablement.

Signed-off-by: Rick Farrington <ricardo.farrington@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agocxgb4: fix incorrect cim_la output for T6
Ganesh Goudar [Wed, 31 May 2017 13:40:21 +0000 (19:10 +0530)]
cxgb4: fix incorrect cim_la output for T6

take care of UpDbgLaRdPtr[0-3] restriction for T6.

Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'sctp-improve-asoc-streams-management'
David S. Miller [Fri, 2 Jun 2017 17:56:27 +0000 (13:56 -0400)]
Merge branch 'sctp-improve-asoc-streams-management'

Xin Long says:

====================
sctp: improve asoc streams management

This patchset changes to define asoc->stream as an object, and also
improve some codes to make it more clearly.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: merge sctp_stream_new and sctp_stream_init
Xin Long [Wed, 31 May 2017 08:36:32 +0000 (16:36 +0800)]
sctp: merge sctp_stream_new and sctp_stream_init

Since last patch, sctp doesn't need to alloc memory for asoc->stream any
more. sctp_stream_new and sctp_stream_init both are used to alloc memory
for stream.in or stream.out, and their names are also confusing.

This patch is to merge them into sctp_stream_init, and only pass stream
and streamcnt parameters into it, instead of the whole asoc.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosctp: define the member stream as an object instead of pointer in asoc
Xin Long [Wed, 31 May 2017 08:36:31 +0000 (16:36 +0800)]
sctp: define the member stream as an object instead of pointer in asoc

As Marcelo's suggestion, stream is a fixed size member of asoc and would
not grow with more streams. To avoid an allocation for it, this patch is
to define it as an object instead of pointer and update the places using
it, also create sctp_stream_update() called in sctp_assoc_update() to
migrate the stream info from one stream to another.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Fri, 2 Jun 2017 17:47:29 +0000 (13:47 -0400)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-06-01

This series contains updates to i40e, i40evf and the "new" AVF virtchnl.

This is the introduction of the Intel(R) Ethernet Adaptive Virtual
Function driver code and device ID, as presented at the NetDEV 1.2
conference in 2016.
http://netdevconf.org/1.2/session.html?anjali-singhai

The idea is to convert the interface between the i40evf driver
and the parent i40e PF driver to be generic, as the i40evf driver
should in the future be able to run on top of other Intel PF
drivers, and negotiate any features beyond a "base expected" set.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'qed-Enhance-storage-APIs'
David S. Miller [Fri, 2 Jun 2017 14:33:04 +0000 (10:33 -0400)]
Merge branch 'qed-Enhance-storage-APIs'

Yuval Mintz says:

====================
qed: Enhance storage APIs

This series is intended to add additional information and features
to the API between qed and its storage protocol drivers [qedi, qedf].

Patch #2 adds some information stored on device such as wwpn & wwnn
to allow qedf utilize it; #1 fixes an issue with the reading of those
values [which were unused until now].

Patch #3 would allow the protocol drivers access to images on persistent
storage which is a prerequirement for adding boot from SAN support.

Patch #4 adds infrastrucutre to a future feature for qedi.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Add support for changing iSCSI mac
Mintz, Yuval [Fri, 2 Jun 2017 05:58:33 +0000 (08:58 +0300)]
qed: Add support for changing iSCSI mac

Enhance API between qedi and qed, allowing qedi to inform device's
firmware when the iSCSI mac is to be changed.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Support NVM-image reading API
Mintz, Yuval [Fri, 2 Jun 2017 05:58:32 +0000 (08:58 +0300)]
qed: Support NVM-image reading API

Storage drivers require images from the nvram in boot-from-SAN
scenarios. This provides the necessary API between qed and the
protocol drivers to perform such reads.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Share additional information with qedf
Mintz, Yuval [Fri, 2 Jun 2017 05:58:31 +0000 (08:58 +0300)]
qed: Share additional information with qedf

Share several new tidbits with qedf:
 - wwpn & wwnn
 - Absolute pf-id [this one is actually meant for qedi as well]
 - Number of available CQs

While we're at it, now that qedf will be aware of the available CQs
we can add some validation on the inputs it provides.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoqed: Correct order of wwnn and wwpn
Mintz, Yuval [Fri, 2 Jun 2017 05:58:30 +0000 (08:58 +0300)]
qed: Correct order of wwnn and wwpn

Driver reads values via HSI splitting this 8-byte into 2 32-bit
values and builds a single u64 field - but it does so by shifting
the lower field instead of the higher.
Luckily, we still don't use these fields for anything - but we're about
to start.

Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoteam: add macro MODULE_ALIAS_TEAM_MODE for team mode alias
Zhang Shengju [Thu, 1 Jun 2017 07:37:02 +0000 (15:37 +0800)]
team: add macro MODULE_ALIAS_TEAM_MODE for team mode alias

Add a new macro MODULE_ALIAS_TEAM_MODE to unify and simplify the
declaration of team mode alias.

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-tagger-simplification'
David S. Miller [Thu, 1 Jun 2017 21:34:57 +0000 (17:34 -0400)]
Merge branch 'dsa-tagger-simplification'

Vivien Didelot says:

====================
net: dsa: tagger simplification

This series clarifies the hot path, removes the labels in tagging
implementations, and frees the original SKB in the xmit caller.

Changes in v3:
  - drop removal of usused rcv arguments because they will be used later
  - include the new ksz tagging implementation
  - add reviewers' tags

Changes in v2:
  - do not remove tagger function copies
  - document hot path requirements
  - make netdev_uses_dsa simpler
  - add reviewers' tags
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: factor skb freeing on xmit
Vivien Didelot [Thu, 1 Jun 2017 20:07:15 +0000 (16:07 -0400)]
net: dsa: factor skb freeing on xmit

As of a86d8becc3f0 ("net: dsa: Factor bottom tag receive functions"),
the rcv caller frees the original SKB in case or error.

Be symmetric with that and make the xmit caller do the same.

At the same time, fix the checkpatch NULL comparison check:

        CHECK: Comparison to NULL could be written "!nskb"
    #208: FILE: net/dsa/tag_trailer.c:35:
    + if (nskb == NULL)

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: remove out_drop label in taggers rcv
Vivien Didelot [Thu, 1 Jun 2017 20:07:14 +0000 (16:07 -0400)]
net: dsa: remove out_drop label in taggers rcv

Many rcv functions from net/dsa/tag_*.c have a useless out_drop goto
label which simply returns NULL. Kill it in favor of the obvious.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: remove dsa_uses_tagged_protocol
Vivien Didelot [Thu, 1 Jun 2017 20:07:13 +0000 (16:07 -0400)]
net: dsa: remove dsa_uses_tagged_protocol

Since dev->dsa_ptr is a pointer to a dsa_switch_tree, there is no need
to have another inline helper just to check rcv.

Remove dsa_uses_tagged_protocol and check dsa_ptr && dsa_ptr->rcv
together at the same time.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: do not cast dst
Vivien Didelot [Thu, 1 Jun 2017 20:07:12 +0000 (16:07 -0400)]
net: dsa: do not cast dst

dsa_ptr is not a void pointer anymore since Nov 2011, as of cf50dcc24f82
("dsa: Change dsa_uses_{dsa, trailer}_tags() into inline functions"),
but an explicit dsa_switch_tree pointer, thus remove the (void *) cast.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: comment hot path requirements
Vivien Didelot [Thu, 1 Jun 2017 20:07:11 +0000 (16:07 -0400)]
net: dsa: comment hot path requirements

The DSA layer uses inline helpers and copy of the tagging functions for
faster access in hot path. Add comments to detail that.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoi40evf: update i40evf.txt with new content
Jesse Brandeburg [Thu, 11 May 2017 18:23:21 +0000 (11:23 -0700)]
i40evf: update i40evf.txt with new content

The addition of the AVF and virtchnl code to the i40evf driver
means we should update the i40evf.txt file with the most up to date
information.

It seems this file hasn't been updated in a while, so the
changes cover a little more than just AVF, but it's all only
in the i40evf.txt.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40evf: Add support for Adaptive Virtual Function
Preethi Banala [Thu, 11 May 2017 18:23:20 +0000 (11:23 -0700)]
i40evf: Add support for Adaptive Virtual Function

Add device ID define and mac_type assignment needed for
Adaptive Virtual Function (VF Base Mode Support).

Also, update version to v3.0.0 in order to indicate
clearly that this is the first driver supporting the AVF
device ID.

Signed-off-by: Preethi Banala <preethi.banala@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agovirtchnl: Add compile time static asserts to validate structure sizes
Sridhar Samudrala [Thu, 11 May 2017 18:23:19 +0000 (11:23 -0700)]
virtchnl: Add compile time static asserts to validate structure sizes

This uses preprocessor tricks to make sure that a divide by zero
occurs if a struct changes size outside the expected number of
bytes.

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agovirtchnl: Add pad fields to a couple of structures
Sridhar Samudrala [Thu, 11 May 2017 18:23:18 +0000 (11:23 -0700)]
virtchnl: Add pad fields to a couple of structures

This removes holes and makes structure sizes consistent across
32 and 64 bit builds.

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/virtchnl: move function to virtchnl
Jesse Brandeburg [Thu, 11 May 2017 18:23:17 +0000 (11:23 -0700)]
i40e/virtchnl: move function to virtchnl

This moves a function that is needed for the virtchnl interface
from the i40e PF driver over to the virtchnl.h file.

It was manually verified that the function in question is unchanged
except for the function name and function header, which explains
the slight difference in the number of lines removed/added.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agovirtchnl: finish conversion to virtchnl interface
Jesse Brandeburg [Thu, 11 May 2017 18:23:16 +0000 (11:23 -0700)]
virtchnl: finish conversion to virtchnl interface

This patch implements the complete version of the virtchnl.h file
with final renames, and fixes the related code in i40e and i40evf.

It also expands comments, and adds details on the usage of
certain fields.

In addition, due to the changes a couple of casts are needed
to prevent errors found by sparse after renaming some fields.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40evf/virtchnl: whitespace cleanups
Jesse Brandeburg [Thu, 11 May 2017 18:23:15 +0000 (11:23 -0700)]
i40evf/virtchnl: whitespace cleanups

This patch fixes up a bunch of whitespace issues introduced
by the previous automated change of name from i40e to virtchnl.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/virtchnl: refactor code for validate checks
Jesse Brandeburg [Thu, 11 May 2017 18:23:14 +0000 (11:23 -0700)]
i40e/virtchnl: refactor code for validate checks

This change updates the arguments passed to the validate function
and fixes the caller, as well as uses the new return values added to
virtchnl.h

One other minor tweak, remove a duplicate set to zero of valid_len.

This is in preparation for moving the function to virtchnl.h.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>