GitHub/moto-9609/android_kernel_motorola_exynos9610.git
8 years agofsl/fman: return a phy_dev pointer from init
Madalin Bucur [Tue, 24 May 2016 13:33:54 +0000 (16:33 +0300)]
fsl/fman: return a phy_dev pointer from init

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
8 years agofsl/fman: simplify device tree reads
Madalin Bucur [Mon, 16 May 2016 13:57:14 +0000 (16:57 +0300)]
fsl/fman: simplify device tree reads

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
8 years agofsl/fman: use of_get_phy_mode()
Madalin Bucur [Thu, 7 Apr 2016 09:50:16 +0000 (12:50 +0300)]
fsl/fman: use of_get_phy_mode()

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
8 years agofsl/fman: small fixes
Madalin Bucur [Tue, 22 Mar 2016 08:27:16 +0000 (10:27 +0200)]
fsl/fman: small fixes

Make module params static, proper NULL checks, remove __iomem label
when misused.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
8 years agofsl/fman: fix loadable module compilation
Igal Liberman [Sat, 9 Jan 2016 21:16:33 +0000 (23:16 +0200)]
fsl/fman: fix loadable module compilation

Signed-off-by: Igal Liberman <igal.liberman@freescale.com>
8 years agofsl/fman: split lines over 80 characters
Madalin Bucur [Tue, 19 Jul 2016 07:56:17 +0000 (10:56 +0300)]
fsl/fman: split lines over 80 characters

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
8 years agonet: phy: Add Edge-rate driver for Microsemi PHYs.
Raju Lakkaraju [Mon, 3 Oct 2016 07:23:13 +0000 (12:53 +0530)]
net: phy: Add Edge-rate driver for Microsemi PHYs.

Edge-rate:
As system and networking speeds increase, a signal's output transition,
also know as the edge rate or slew rate (V/ns), takes on greater importance
because high-speed signals come with a price. That price is an assortment of
interference problems like ringing on the line, signal overshoot and
undershoot, extended signal settling times, crosstalk noise, transmission
line reflections, false signal detection by the receiving device and
electromagnetic interference (EMI) -- all of which can negate the potential
gains designers are seeking when they try to increase system speeds through
the use of higher performance logic devices. The fact is, faster signaling
edge rates can cause a higher level of electrical noise or other type of
interference that can actually lead to slower line speeds and lower maximum
system frequencies. This parameter allow the board designers to change the
driving strange, and thereby change the EMI behavioral.

Edge-rate parameters (vddmac, edge-slowdown) get from Device Tree.

Tested on Beaglebone Black with VSC 8531 PHY.

Signed-off-by: Raju Lakkaraju <Raju.Lakkaraju@microsemi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agovmxnet3: Wake queue from reset work
Benjamin Poirier [Mon, 3 Oct 2016 02:47:50 +0000 (10:47 +0800)]
vmxnet3: Wake queue from reset work

vmxnet3_reset_work() expects tx queues to be stopped (via
vmxnet3_quiesce_dev -> netif_tx_disable). However, this races with the
netif_wake_queue() call in netif_tx_timeout() such that the driver's
start_xmit routine may be called unexpectedly, triggering one of the BUG_ON
in vmxnet3_map_pkt with a stack trace like this:

RIP: 0010:[<ffffffffa00cf4bc>] vmxnet3_map_pkt+0x3ac/0x4c0 [vmxnet3]
 [<ffffffffa00cf7e0>] vmxnet3_tq_xmit+0x210/0x4e0 [vmxnet3]
 [<ffffffff813ab144>] dev_hard_start_xmit+0x2e4/0x4c0
 [<ffffffff813c956e>] sch_direct_xmit+0x17e/0x1e0
 [<ffffffff813c96a7>] __qdisc_run+0xd7/0x130
 [<ffffffff813a6a7a>] net_tx_action+0x10a/0x200
 [<ffffffff810691df>] __do_softirq+0x11f/0x260
 [<ffffffff81472fdc>] call_softirq+0x1c/0x30
 [<ffffffff81004695>] do_softirq+0x65/0xa0
 [<ffffffff81069b89>] local_bh_enable_ip+0x99/0xa0
 [<ffffffffa031ff36>] destroy_conntrack+0x96/0x110 [nf_conntrack]
 [<ffffffff813d65e2>] nf_conntrack_destroy+0x12/0x20
 [<ffffffff8139c6d5>] skb_release_head_state+0xb5/0xf0
 [<ffffffff8139d299>] skb_release_all+0x9/0x20
 [<ffffffff8139cfe9>] __kfree_skb+0x9/0x90
 [<ffffffffa00d0069>] vmxnet3_quiesce_dev+0x209/0x340 [vmxnet3]
 [<ffffffffa00d020a>] vmxnet3_reset_work+0x6a/0xa0 [vmxnet3]
 [<ffffffff8107d7cc>] process_one_work+0x16c/0x350
 [<ffffffff810804fa>] worker_thread+0x17a/0x410
 [<ffffffff810848c6>] kthread+0x96/0xa0
 [<ffffffff81472ee4>] kernel_thread_helper+0x4/0x10

Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 4 Oct 2016 03:28:50 +0000 (23:28 -0400)]
Merge branch '100GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2016-10-02

This series contains updates to fm10k only.

Jake fixes an issue where PTP applications requesting software timestamps
may complain that the requested mode is not supported, so add a generic
callback for those drivers that have software transmit timestamp support
enabled.  Then provides a trivial cleanup where a code was not wrapped
properly.  Got make sure that code looks good in a 80 character limit.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoi40e: avoid NULL pointer dereference and recursive errors on early PCI error
Guilherme G Piccoli [Mon, 3 Oct 2016 07:31:12 +0000 (00:31 -0700)]
i40e: avoid NULL pointer dereference and recursive errors on early PCI error

Although rare, it's possible to hit PCI error early on device
probe, meaning possibly some structs are not entirely initialized,
and some might even be completely uninitialized, leading to NULL
pointer dereference.

The i40e driver currently presents a "bad" behavior if device hits
such early PCI error: firstly, the struct i40e_pf might not be
attached to pci_dev yet, leading to a NULL pointer dereference on
access to pf->state.

Even checking if the struct is NULL and avoiding the access in that
case isn't enough, since the driver cannot recover from PCI error
that early; in our experiments we saw multiple failures on kernel
log, like:

  [549.664] i40e 0007:01:00.1: Initial pf_reset failed: -15
  [549.664] i40e: probe of 0007:01:00.1 failed with error -15
  [...]
  [871.644] i40e 0007:01:00.1: The driver for the device stopped because the
  device firmware failed to init. Try updating your NVM image.
  [871.644] i40e: probe of 0007:01:00.1 failed with error -32
  [...]
  [872.516] i40e 0007:01:00.0: ARQ: Unknown event 0x0000 ignored

Between the first probe failure (error -15) and the second (error -32)
another PCI error happened due to the first bad probe. Also, driver
started to flood console with those ARQ event messages.

This patch will prevent these issues by allowing error recovery
mechanism to remove the failed device from the system instead of
trying to recover from early PCI errors during device probe.

CC: <stable@vger.kernel.org>
Signed-off-by: Guilherme G Piccoli <gpiccoli@linux.vnet.ibm.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 4 Oct 2016 03:26:18 +0000 (23:26 -0400)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2016-10-03

This series contains fixes to i40e only.

Stefan Assmann provides the changes in this series to resolve an issue
where when we run out of MSIx vectors, iWARP gets disabled automatically.
First adds a check for "no vectors left" during MSIx vector allocation
for VMDq, which will prevent more vectors being allocated than available.
Then fixed the MSIx vector redistribution when we reach the hardware limit
for vectors so that additional features like VMDq, iWARP, etc do not get
starved for vectors because the PF is hogging all the resources.  Lastly,
fix the issue for flow director by moving the check for the reaching the
vector limit earlier in the code so that a decision can be made on
disabling flow director.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'qed-qedr-infrastructure'
David S. Miller [Tue, 4 Oct 2016 03:22:52 +0000 (23:22 -0400)]
Merge branch 'qed-qedr-infrastructure'

Yuval Mintz says:

====================
qed*: Add qedr infrastructure support

In the last couple of weeks we've been sending RFCs for the qedr
driver - the RoCE driver for QLogic FastLinQ 4xxxx line of adapters.
Latest RFC can be found at [1].

At Doug's advice [2], we've decided to split the series into two:
 - first part contains the qed backbone that's necessary for all the
configurations relating to the qedr driver, as well as the qede
infrastructure that is used for communication between the qedr and qede.
 - Second part consists of the actual qedr driver and introduces almost
no changes to qed/qede.

This is the first of said two parts. The second half would be sent
later this week.

The only 'oddity' in the devision are the Kconfig options -
As this series introduces both LL2 and QEDR-based logic in qed/qede,
I wanted to add the CONFIG_INFINIBAND_QEDR option here [with default n].
Otherwise, a lot of the code introduced would be dead-code [won't even
be compiled] until qedr is accepted.
As a result I've placed the config option in an odd place - under
qlogic's Kconfig. The second series would then remove that option
and add it in its correct place under the infiniband Kconfig.
[I'm fine with pushing it there to begin with, but I didn't want to
'contaminate' non-qlogic configuration files with half-baked options].

Dave - I don't think you were E-mailed with Doug's suggestion.
I think the notion was to have the two halves accepted side-by-side,
but actually the first has no dependency issues, so it's also
possible to simply take this first to net-next, and push the qedr
into rdma once it's merged. But it's basically up to you and Doug;
We'd align with whatever suits you best.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add RoCE ll2 & GSI support
Ram Amrani [Sat, 1 Oct 2016 19:00:01 +0000 (22:00 +0300)]
qed: Add RoCE ll2 & GSI support

Add the RoCE-specific LL2 logic [as well as GSI support] over
the 'generic' LL2 interface.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add support for memory registeration verbs
Ram Amrani [Sat, 1 Oct 2016 19:00:00 +0000 (22:00 +0300)]
qed: Add support for memory registeration verbs

Add slowpath configuration support for user, dma and memory
regions registration.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add support for QP verbs
Ram Amrani [Sat, 1 Oct 2016 18:59:59 +0000 (21:59 +0300)]
qed: Add support for QP verbs

Add support for the slowpath configurations of Queue Pair verbs
which adds, deletes, modifies and queries Queue Pairs.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: PD,PKEY and CQ verb support
Ram Amrani [Sat, 1 Oct 2016 18:59:58 +0000 (21:59 +0300)]
qed: PD,PKEY and CQ verb support

Add support for the configurations of the protection domain and
completion queues.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add support for RoCE hw init
Ram Amrani [Sat, 1 Oct 2016 18:59:57 +0000 (21:59 +0300)]
qed: Add support for RoCE hw init

This adds the backbone required for the various HW initalizations
which are necessary for the qedr driver - FW notification, resource
initializations, etc.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqede: Add qedr framework
Ram Amrani [Sat, 1 Oct 2016 18:59:56 +0000 (21:59 +0300)]
qede: Add qedr framework

Adds a skeletal implementation of the qede RoCE driver -
The qedr has some dependencies of the state of the underlying base
interface. This adds some logic required with mutual registrations
and the ability to pass updates on 'intresting' events.

Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add Light L2 support
Yuval Mintz [Sat, 1 Oct 2016 18:59:55 +0000 (21:59 +0300)]
qed: Add Light L2 support

Other protocols beside the networking driver need the ability
of passing some L2 traffic, usually [although not limited] for the
purpose of some management traffic.

Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoi40e: fix sideband flow director vector allocation
Stefan Assmann [Mon, 19 Sep 2016 11:37:51 +0000 (13:37 +0200)]
i40e: fix sideband flow director vector allocation

Currently if the MSI-X vector limit is reached the sideband flow
director gets disabled. A bit too early to make that decision, as
vectors may get re-distributed. So move the check further back.

Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: fix MSI-X vector redistribution if hw limit is reached
Stefan Assmann [Mon, 19 Sep 2016 11:37:50 +0000 (13:37 +0200)]
i40e: fix MSI-X vector redistribution if hw limit is reached

The driver allocates 1 vector per CPU thread and the current hardware
limit for vectors is 129 per PF. On systems with 128 or more threads
this currently means all vectors are used by the PF leaving no room for
additional features like VMDq, iWARP, etc...
The code that should redistribute the vectors in this case is broken and
never triggers. Fixed the code so that it actually triggers if the
hardware limit is reached and adjust the number of queue pairs
accordingly.
Also the number of initially requested iWARP vectors was not properly
saved when the vector limit was reached, and therefore always zero.

Comparison with debug statement.
Before:
i40e 0000:2d:00.0: VMDq disabled, not enough MSI-X vectors
i40e 0000:2d:00.0: IWARP disabled, not enough MSI-X vectors
i40e 00.0 MSI-X vector distribution: PF 128, VMDq 0, FDSB 0, iWARP 0
After:
i40e 0000:2d:00.0: MSI-X vector limit reached, attempting to redistribute vectors
i40e 00.0 MSI-X vector distribution: PF 78, VMDq 8, FDSB 0, iWARP 42

Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: check if vectors are already depleted when doing VMDq allocation
Stefan Assmann [Mon, 19 Sep 2016 11:37:49 +0000 (13:37 +0200)]
i40e: check if vectors are already depleted when doing VMDq allocation

During MSI-X vector allocation for VMDq, a check for "no vectors left"
was missing, add it. This prevents more vectors to be allocated than
available.

Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoptp: Fix resource leak in case of error
Christophe Jaillet [Sun, 2 Oct 2016 07:04:16 +0000 (09:04 +0200)]
ptp: Fix resource leak in case of error

A call to 'ida_simple_remove()' is missing in the error handling path.

This as been spotted with the following coccinelle script which tries to
detect missing 'ida_simple_remove()' call in error handling paths.

///////////////
@@
expression x;
identifier l;
@@

*   x = ida_simple_get(...);
    ...
    if (...) {
    ...
    }
    ...
    if (...) {
       ...
       goto l;
    }
    ...
*   l: ... when != ida_simple_remove(...);

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: qcom/emac: fix return value check in emac_sgmii_config()
Wei Yongjun [Sat, 1 Oct 2016 09:12:29 +0000 (09:12 +0000)]
net: qcom/emac: fix return value check in emac_sgmii_config()

In case of error, the function ioremap() returns NULL pointer
not ERR_PTR(). The IS_ERR() test in the return value check
should be replaced with NULL test.

Also add check for return value of platform_get_resource().

Fixes: 54e19bc74f33 ("net: qcom/emac: do not use devm on internal
phy pdev")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: skbuff: Limit skb_vlan_pop/push() to expect skb->data at mac header
Shmulik Ladkani [Thu, 29 Sep 2016 09:10:41 +0000 (12:10 +0300)]
net: skbuff: Limit skb_vlan_pop/push() to expect skb->data at mac header

skb_vlan_pop/push were too generic, trying to support the cases where
skb->data is at mac header, and cases where skb->data is arbitrarily
elsewhere.

Supporting an arbitrary skb->data was complex and bogus:
 - It failed to unwind skb->data to its original location post actual
   pop/push.
   (Also, semantic is not well defined for unwinding: If data was into
    the eth header, need to use same offset from start; But if data was
    at network header or beyond, need to adjust the original offset
    according to the push/pull)
 - It mangled the rcsum post actual push/pop, without taking into account
   that the eth bytes might already have been pulled out of the csum.

Most callers (ovs, bpf) already had their skb->data at mac_header upon
invoking skb_vlan_pop/push.
Last caller that failed to do so (act_vlan) has been recently fixed.

Therefore, to simplify things, no longer support arbitrary skb->data
inputs for skb_vlan_pop/push().

skb->data is expected to be exactly at mac_header; WARN otherwise.

Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Pravin Shelar <pshelar@ovn.org>
Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: act_vlan: Push skb->data to mac_header prior calling skb_vlan_*() functions
Shmulik Ladkani [Thu, 29 Sep 2016 09:10:40 +0000 (12:10 +0300)]
net/sched: act_vlan: Push skb->data to mac_header prior calling skb_vlan_*() functions

Generic skb_vlan_push/skb_vlan_pop functions don't properly handle the
case where the input skb data pointer does not point at the mac header:

- They're doing push/pop, but fail to properly unwind data back to its
  original location.
  For example, in the skb_vlan_push case, any subsequent
  'skb_push(skb, skb->mac_len)' calls make the skb->data point 4 bytes
  BEFORE start of frame, leading to bogus frames that may be transmitted.

- They update rcsum per the added/removed 4 bytes tag.
  Alas if data is originally after the vlan/eth headers, then these
  bytes were already pulled out of the csum.

OTOH calling skb_vlan_push/skb_vlan_pop with skb->data at mac_header
present no issues.

act_vlan is the only caller to skb_vlan_*() that has skb->data pointing
at network header (upon ingress).
Other calles (ovs, bpf) already adjust skb->data at mac_header.

This patch fixes act_vlan to point to the mac_header prior calling
skb_vlan_*() functions, as other callers do.

Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Pravin Shelar <pshelar@ovn.org>
Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agofm10k: wrap long line for alloc_workqueue
Jacob Keller [Thu, 25 Aug 2016 21:15:39 +0000 (14:15 -0700)]
fm10k: wrap long line for alloc_workqueue

Trivial change here to cleanup a checkpatch.pl warning that got
introduced when changing to alloc_workqueue.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agonet: mvmdio: do not clk_disable_unprepare() NULL clock
Alexey Khoroshilov [Fri, 30 Sep 2016 21:56:37 +0000 (00:56 +0300)]
net: mvmdio: do not clk_disable_unprepare() NULL clock

There is no need to clk_disable_unprepare(dev->clk)
before it was initialized.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'rxrpc-rewrite-20160930' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Mon, 3 Oct 2016 06:02:17 +0000 (02:02 -0400)]
Merge tag 'rxrpc-rewrite-20160930' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: More fixes and adjustments

This set of patches contains some more fixes and adjustments:

 (1) Actually display the retransmission indication previously added to the
     tx_data trace.

 (2) Switch to Congestion Avoidance mode properly at cwnd==ssthresh rather
     than relying on detection during an overshoot and correction.

 (3) Reduce ssthresh to the peer's declared receive window.

 (4) The offset field in rxrpc_skb_priv can be dispensed with and the error
     field is no longer used.  Get rid of them.

 (5) Keep the call timeouts as ktimes rather than jiffies to make it easier
     to deal with RTT-based timeout values in future.  Rounding to jiffies
     is still necessary when the system timer is set.

 (6) Fix the call timer handling to avoid retriggering of expired timeout
     actions.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'ovs-mpls'
David S. Miller [Mon, 3 Oct 2016 06:00:29 +0000 (02:00 -0400)]
Merge branch 'ovs-mpls'

Jiri Benc says:

====================
openvswitch: mpls fix and clean up

Convert to the new mpls skb layout the last remaining place in openvswitch,
forgotten on the mpls GSO rework. The GSO rework also allows for some
cleanup in the third patch.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: use mpls_hdr
Jiri Benc [Fri, 30 Sep 2016 17:08:07 +0000 (19:08 +0200)]
openvswitch: use mpls_hdr

skb_mpls_header is equivalent to mpls_hdr now. Use the existing helper
instead.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agompls: move mpls_hdr to a common location
Jiri Benc [Fri, 30 Sep 2016 17:08:06 +0000 (19:08 +0200)]
mpls: move mpls_hdr to a common location

This will be also used by openvswitch.

Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: mpls: set network header correctly on key extract
Jiri Benc [Fri, 30 Sep 2016 17:08:05 +0000 (19:08 +0200)]
openvswitch: mpls: set network header correctly on key extract

After the 48d2ab609b6b ("net: mpls: Fixups for GSO"), MPLS handling in
openvswitch was changed to have network header pointing to the start of the
MPLS headers and inner_network_header pointing after the MPLS headers.

However, key_extract was missed by the mentioned commit, causing incorrect
headers to be set when a MPLS packet just enters the bridge or after it is
recirculated.

Fixes: 48d2ab609b6b ("net: mpls: Fixups for GSO")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum_router: avoid potential uninitialized data usage
Arnd Bergmann [Fri, 30 Sep 2016 16:17:10 +0000 (18:17 +0200)]
mlxsw: spectrum_router: avoid potential uninitialized data usage

If fi->fib_nhs is zero, the router interface pointer is uninitialized, as shown by
this warning:

drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c: In function 'mlxsw_sp_router_fib_event':
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:1674:21: error: 'r' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c:1643:23: note: 'r' was declared here

This changes the loop so we handle the case the same way as finding no router
interface pointer attached to one of the nexthops to ensure we always
trap here instead of using uninitialized data.

Fixes: b45f64d16d45 ("mlxsw: spectrum_router: Use FIB notifications instead of switchdev calls")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: shut up maybe-uninitialized warning
Arnd Bergmann [Fri, 30 Sep 2016 16:17:09 +0000 (18:17 +0200)]
net/mlx5e: shut up maybe-uninitialized warning

Build-testing this driver with -Wmaybe-uninitialized gives a new false-positive
warning that I can't really explain:

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c: In function 'mlx5e_configure_flower':
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c:509:3: error: 'old_attr' may be used uninitialized in this function [-Werror=maybe-uninitialized]

It's obvious from the code that 'old_attr' is initialized whenever 'old'
is non-NULL here. The warning appears with all versions I tested from gcc-4.7
through gcc-6.1, and I could not come up with a way to rewrite the function
in a more readable way that avoids the warning, so I'm adding another
initialization to shut it up.

Fixes: 8b32580df1cb ("net/mlx5e: Add TC vlan action for SRIOV offloads")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agofm10k: use generic ethtool_op_get_ts_info callback
Jacob Keller [Thu, 25 Aug 2016 21:06:54 +0000 (14:06 -0700)]
fm10k: use generic ethtool_op_get_ts_info callback

This generic callback is for drivers which have software Tx timestamp
support enabled. Without this, PTP applications requesting software
timestamps may complain that the requested mode is not supported.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agocxgb4: unexport cxgb4_dcb_enabled
Arnd Bergmann [Fri, 30 Sep 2016 16:15:33 +0000 (18:15 +0200)]
cxgb4: unexport cxgb4_dcb_enabled

A recent cleanup marked cxgb4_dcb_enabled as 'static', which is correct, but this ignored
how the symbol is also exported. In addition, the export can be compiled out when modules
are disabled, causing a harmless compiler warning in configurations for which it is not
used at all:

drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c:282:12: error: 'cxgb4_dcb_enabled' defined but not used [-Werror=unused-function]

This removes the export and moves the function into the correct #ifdef so we only build
it when there are users.

Fixes: 50935857f878 ("cxgb4: mark symbols static where possible")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: rtnl: avoid uninitialized data in IFLA_VF_VLAN_LIST handling
Arnd Bergmann [Fri, 30 Sep 2016 16:13:49 +0000 (18:13 +0200)]
net: rtnl: avoid uninitialized data in IFLA_VF_VLAN_LIST handling

With the newly added support for IFLA_VF_VLAN_LIST netlink messages,
we get a warning about potential uninitialized variable use in
the parsing of the user input when enabling the -Wmaybe-uninitialized
warning:

net/core/rtnetlink.c: In function 'do_setvfinfo':
net/core/rtnetlink.c:1756:9: error: 'ivvl$' may be used uninitialized in this function [-Werror=maybe-uninitialized]

I have not been able to prove whether it is possible to arrive in
this code with an empty IFLA_VF_VLAN_LIST block, but if we do,
then ndo_set_vf_vlan gets called with uninitialized arguments.

This adds an explicit check for an empty list, making it obvious
to the reader and the compiler that this cannot happen.

Fixes: 79aab093a0b5 ("net: Update API for VF vlan protocol 802.1ad support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: pktgen: fix pkt_size
Paolo Abeni [Fri, 30 Sep 2016 14:56:45 +0000 (16:56 +0200)]
net: pktgen: fix pkt_size

The commit 879c7220e828 ("net: pktgen: Observe needed_headroom
of the device") increased the 'pkt_overhead' field value by
LL_RESERVED_SPACE.
As a side effect the generated packet size, computed as:

/* Eth + IPh + UDPh + mpls */
datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
  pkt_dev->pkt_overhead;

is decreased by the same value.
The above changed slightly the behavior of existing pktgen users,
and made the procfs interface somewhat inconsistent.
Fix it by restoring the previous pkt_overhead value and using
LL_RESERVED_SPACE as extralen in skb allocation.
Also, change pktgen_alloc_skb() to only partially reserve
the headroom to allow the caller to prefetch from ll header
start.

v1 -> v2:
 - fixed some typos in the comments

Fixes: 879c7220e828 ("net: pktgen: Observe needed_headroom of the device")
Suggested-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: fec: set mac address unconditionally
Gavin Schenk [Fri, 30 Sep 2016 09:46:10 +0000 (11:46 +0200)]
net: fec: set mac address unconditionally

If the mac address origin is not dt, you can only safely assign a mac
address after "link up" of the device. If the link is off the clocks are
disabled and because of issues assigning registers when clocks are off the
new mac address cannot be written in .ndo_set_mac_address() on some soc's.
This fix sets the mac address unconditionally in fec_restart(...) and
ensures consistency between fec registers and the network layer.

Signed-off-by: Gavin Schenk <g.schenk@eckelmann.de>
Acked-by: Fugang Duan <fugang.duan@nxp.com>
Acked-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Fixes: 9638d19e4816 ("net: fec: add netif status check before set mac address")
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: mediatek: mark symbols static where possible
Baoyou Xie [Fri, 30 Sep 2016 07:48:50 +0000 (15:48 +0800)]
net: ethernet: mediatek: mark symbols static where possible

We get 2 warnings when building kernel with W=1:
drivers/net/ethernet/mediatek/mtk_eth_soc.c:2041:5: warning: no previous prototype for 'mtk_get_link_ksettings' [-Wmissing-prototypes]
drivers/net/ethernet/mediatek/mtk_eth_soc.c:2052:5: warning: no previous prototype for 'mtk_set_link_ksettings' [-Wmissing-prototypes]

In fact, these functions are only used in the file in which they are
declared and don't need a declaration, but can be made static.
So this patch marks these functions with 'static'.

Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agocxgb4: mark cxgb_setup_tc() static
Baoyou Xie [Fri, 30 Sep 2016 07:34:25 +0000 (15:34 +0800)]
cxgb4: mark cxgb_setup_tc() static

We get 1 warning when building kernel with W=1:
drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c:2715:5: warning: no previous prototype for 'cxgb_setup_tc' [-Wmissing-prototypes]

In fact, this function is only used in the file in which it is
declared and don't need a declaration, but can be made static.
so this patch marks this function with 'static'.

Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6 addrconf: remove addrconf_sysctl_hop_limit()
Maciej Żenczykowski [Thu, 29 Sep 2016 07:33:43 +0000 (00:33 -0700)]
ipv6 addrconf: remove addrconf_sysctl_hop_limit()

This is an effective no-op in terms of user observable behaviour.

By preventing the overwrite of non-null extra1/extra2 fields
in addrconf_sysctl() we can enable the use of proc_dointvec_minmax().

This allows us to eliminate the constant min/max (1..255) trampoline
function that is addrconf_sysctl_hop_limit().

This is nice because it simplifies the code, and allows future
sysctls with constant min/max limits to also not require trampolines.

We still can't eliminate the trampoline for mtu because it isn't
actually a constant (it depends on other tunables of the device)
and thus requires at-write-time logic to enforce range.

Signed-off-by: Maciej Żenczykowski <maze@google.com>
Acked-by: Erik Kline <ek@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonetfilter: bridge: clarify bridge/netfilter message
Stefan Agner [Wed, 28 Sep 2016 22:05:28 +0000 (15:05 -0700)]
netfilter: bridge: clarify bridge/netfilter message

When using bridge without bridge netfilter enabled the message
displayed is rather confusing and leads to belive that a deprecated
feature is in use. Use IS_MODULE to be explicit that the message only
affects users which use bridge netfilter as module and reword the
message.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Mon, 3 Oct 2016 01:17:07 +0000 (21:17 -0400)]
Merge git://git./linux/kernel/git/davem/net

Three sets of overlapping changes.  Nothing serious.

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoLinux 4.8
Linus Torvalds [Sun, 2 Oct 2016 23:24:33 +0000 (16:24 -0700)]
Linux 4.8

8 years agoMerge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 2 Oct 2016 22:23:00 +0000 (15:23 -0700)]
Merge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 "Three relatively small fixes for ARM:

   - Roger noticed that dma_max_pfn() was calculating the upper limit
     wrongly, by adding the PFN offset of memory twice.

   - A fix from Robin to correct parsing of MPIDR values when the
     address size is larger than one BE32 unit.

   - A fix from Srinivas to ensure that we do not rely on the boot
     loader (or previous Linux kernel) setting the translation table
     base register a certain way in the decompressor, which can lead to
     crashes"

* 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: 8618/1: decompressor: reset ttbcr fields to use TTBR0 on ARMv7
  ARM: 8617/1: dma: fix dma_max_pfn()
  ARM: 8616/1: dt: Respect property size when parsing CPUs

8 years agoARM: 8618/1: decompressor: reset ttbcr fields to use TTBR0 on ARMv7
Srinivas Ramana [Fri, 30 Sep 2016 14:03:31 +0000 (15:03 +0100)]
ARM: 8618/1: decompressor: reset ttbcr fields to use TTBR0 on ARMv7

If the bootloader uses the long descriptor format and jumps to
kernel decompressor code, TTBCR may not be in a right state.
Before enabling the MMU, it is required to clear the TTBCR.PD0
field to use TTBR0 for translation table walks.

The commit dbece45894d3a ("ARM: 7501/1: decompressor:
reset ttbcr for VMSA ARMv7 cores") does the reset of TTBCR.N, but
doesn't consider all the bits for the size of TTBCR.N.

Clear TTBCR.PD0 field and reset all the three bits of TTBCR.N to
indicate the use of TTBR0 and the correct base address width.

Fixes: dbece45894d3 ("ARM: 7501/1: decompressor: reset ttbcr for VMSA ARMv7 cores")
Acked-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Srinivas Ramana <sramana@codeaurora.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
8 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 2 Oct 2016 18:04:29 +0000 (11:04 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "The last regression fixes for 4.8 final:

   - Two patches addressing the fallout of the CR4 optimizations which
     caused CR4-less machines to fail.

   - Fix the VDSO build on big endian machines

   - Take care of FPU initialization if no CPUID is available otherwise
     task struct size ends up being zero

   - Fix up context tracking in case load_gs_index fails"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/entry/64: Fix context tracking state warning when load_gs_index fails
  x86/boot: Initialize FPU and X86_FEATURE_ALWAYS even if we don't have CPUID
  x86/vdso: Fix building on big endian host
  x86/boot: Fix another __read_cr4() case on 486
  x86/init: Fix cr4_init_shadow() on CR4-less machines

8 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Sun, 2 Oct 2016 17:53:38 +0000 (10:53 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:
 "Another round of fixes:

   - CM: Fix mips_cm_max_vp_width for non-MT kernels on MT systems
   - CPS: Avoid BUG() when offlining pre-r6 CPUs
   - DEC: Avoid gas warnings due to suspicious instruction scheduling by
     manually expanding assembler macros.
   - FTLB: Fix configuration by moving confiuguratoin after probing
   - FTLB: clear execution hazard after changing FTLB enable
   - Highmem: Fix detection of unsupported highmem with cache aliases
   - I6400: Don't touch FTLBP chicken bits
   - microMIPS: Fix BUILD_ROLLBACK_PROLOGUE
   - Malta: Fix IOCU disable switch read for MIPS64
   - Octeon: Fix probing of devices attached to GPIO lines
   - uprobes: Misc small fixes"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: CM: Fix mips_cm_max_vp_width for non-MT kernels on MT systems
  MIPS: Fix detection of unsupported highmem with cache aliases
  MIPS: Malta: Fix IOCU disable switch read for MIPS64
  MIPS: Fix BUILD_ROLLBACK_PROLOGUE for microMIPS
  MIPS: clear execution hazard after changing FTLB enable
  MIPS: Configure FTLB after probing TLB sizes from config4
  MIPS: Stop setting I6400 FTLBP
  MIPS: DEC: Avoid la pseudo-instruction in delay slots
  MIPS: Octeon: mark GPIO controller node not populated after IRQ init.
  MIPS: uprobes: fix use of uninitialised variable
  MIPS: uprobes: remove incorrect set_orig_insn
  MIPS: fix uretprobe implementation
  MIPS: smp-cps: Avoid BUG() when offlining pre-r6 CPUs

8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
Linus Torvalds [Sun, 2 Oct 2016 17:42:26 +0000 (10:42 -0700)]
Merge git://git./linux/kernel/git/davem/sparc

Pull sparc fixes from David Miller:

 1) Fix section mismatches in some builds, from Paul Gortmaker.

 2) Need to count huge zero page mappings when doing TSB sizing, from
    Mike Kravetz.

 3) Fix handing of cpu_possible_mask when nr_cpus module option is
    specified, from Atish Patra.

 4) Don't allocate irq stacks until nr_irqs has been processed, also
    from Atish Patra.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
  sparc64: Fix non-SMP build.
  sparc64: Fix irq stack bootmem allocation.
  sparc64: Fix cpu_possible_mask if nr_cpus is set
  sparc64 mm: Fix more TSB sizing issues
  sparc64: fix section mismatch in find_numa_latencies_for_group

8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sun, 2 Oct 2016 17:36:41 +0000 (10:36 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix wrong TCP checksums on MTU probing when checksum offloading is
    disabled, from Douglas Caetano dos Santos.

 2) Fix qdisc backlog updates in qfq and sfb schedulers, from Cong Wang.

 3) Route lookup flow key protocol value is wrong in ip6gre_xmit_other(),
    fix from Lance Richardson.

 4) Scheduling while atomic in multicast routing code of ipv4 and ipv6,
    fix from Nikolay Aleksandrov.

 5) Fix packet alignment in fec driver, from Eric Nelson.

 6) Fix perf regression in sctp due to struct layout and cache misses,
    from Xin Long.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  sctp: fix the issue sctp_diag uses lock_sock in rcu_read_lock
  sctp: change to check peer prsctp_capable when using prsctp polices
  sctp: remove prsctp_param from sctp_chunk
  sctp: move sent_count to the memory hole in sctp_chunk
  tg3: Avoid NULL pointer dereference in tg3_io_error_detected()
  act_ife: Fix false encoding
  act_ife: Fix external mac header on encode
  VSOCK: Don't dec ack backlog twice for rejected connections
  Revert "net: ethernet: bcmgenet: use phydev from struct net_device"
  net: fec: align IP header in hardware
  net: fec: remove QUIRK_HAS_RACC from i.mx27
  net: fec: remove QUIRK_HAS_RACC from i.mx25
  ipmr, ip6mr: fix scheduling while atomic and a deadlock with ipmr_get_route
  ip6_gre: fix flowi6_proto value in ip6gre_xmit_other()
  tcp: fix a compile error in DBGUNDO()
  tcp: fix wrong checksum calculation on MTU probing
  sch_sfb: keep backlog updated with qlen
  sch_qfq: keep backlog updated with qlen
  can: dev: fix deadlock reported after bus-off

8 years agoMIPS: CM: Fix mips_cm_max_vp_width for non-MT kernels on MT systems
Paul Burton [Fri, 30 Sep 2016 16:25:01 +0000 (17:25 +0100)]
MIPS: CM: Fix mips_cm_max_vp_width for non-MT kernels on MT systems

When discovering the number of VPEs per core, smp_num_siblings will be
incorrect for kernels built without support for the MIPS MultiThreading
(MT) ASE running on systems which implement said ASE. This leads to
accesses to VPEs in secondary cores being performed incorrectly since
mips_cm_vp_id calculates the wrong ID to write to the local "other"
registers. Fix this by examining the number of VPEs in the core as
reported by the CM.

This patch presumes that the number of VPEs will be the same in each
core of the system. As this path only applies to systems with CM version
2.5 or lower, and this property is true of all such known systems, this
is likely to be fine but is described in a comment for good measure.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14338/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
8 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 1 Oct 2016 14:37:15 +0000 (07:37 -0700)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fix from James Bottomley:
 "One final fix before 4.8.

  There was a memory leak triggered by turning scsi mq off due to the
  fact that we assume on host release that the already running hosts
  weren't mq based because that's the state of the global flag (even
  though they were).

  Fix it by tracking this on a per host host basis"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: Avoid that toggling use_blk_mq triggers a memory leak

8 years agonet: Use ns_capable_noaudit() when determining net sysctl permissions
Tyler Hicks [Fri, 30 Sep 2016 22:24:31 +0000 (15:24 -0700)]
net: Use ns_capable_noaudit() when determining net sysctl permissions

The capability check should not be audited since it is only being used
to determine the inode permissions. A failed check does not indicate a
violation of security policy but, when an LSM is enabled, a denial audit
message was being generated.

The denial audit message caused confusion for some application authors
because root-running Go applications always triggered the denial. To
prevent this confusion, the capability check in net_ctl_permissions() is
switched to the noaudit variant.

BugLink: https://launchpad.net/bugs/1465724
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Signed-off-by: James Morris <james.l.morris@oracle.com>
[dtor: reapplied after e79c6a4fc923 ("net: make net namespace sysctls
belong to container's owner") accidentally reverted the change.]
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Sat, 1 Oct 2016 04:25:09 +0000 (21:25 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input fix from Dmitry Torokhov:
 "One small change to make joydev (which is used by older games) to bind
  to devices that export Z axis but not X or Y (such as TRC rudder)"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: joydev - recognize devices with Z axis as joysticks

8 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Fri, 30 Sep 2016 22:51:10 +0000 (15:51 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge more fixes from Andrew Morton:
 "Three fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  include/linux/property.h: fix typo/compile error
  ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
  mm: workingset: fix crash in shadow node shrinker caused by replace_page_cache_page()

8 years agoinclude/linux/property.h: fix typo/compile error
John Youn [Fri, 30 Sep 2016 22:11:35 +0000 (15:11 -0700)]
include/linux/property.h: fix typo/compile error

This fixes commit d76eebfa175e ("include/linux/property.h: fix build
issues with gcc-4.4.4").

With that commit we get the following compile error when using the
PROPERTY_ENTRY_INTEGER_ARRAY macro.

 include/linux/property.h:201:39: error: `u32_data' undeclared (first
                 use in this function)
  PROPERTY_ENTRY_INTEGER_ARRAY(_name_, u32, _val_)
                                       ^
 include/linux/property.h:193:17: note: in definition of macro
                 `PROPERTY_ENTRY_INTEGER_ARRAY'
  { .pointer = { _type_##_data = _val_ } },  \
                 ^

This needs a '.' to reference the union member.  It seems this was just
overlooked here since it is done correctly in similar constructs in
other parts of the original commit.

This fix is in preparation of upcoming commits that will use this macro.

Fixes: commit d76eebfa175e ("include/linux/property.h: fix build issues with gcc-4.4.4")
Link: http://lkml.kernel.org/r/2de3b929290d88a723ed829a3e3cbd02044714df.1475114627.git.johnyoun@synopsys.com
Signed-off-by: John Youn <johnyoun@synopsys.com>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Eric Ren [Fri, 30 Sep 2016 22:11:32 +0000 (15:11 -0700)]
ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()

The testcase "mmaptruncate" of ocfs2-test deadlocks occasionally.

In this testcase, we create a 2*CLUSTER_SIZE file and mmap() on it;
there are 2 process repeatedly performing the following operations
respectively: one is doing memset(mmaped_addr + 2*CLUSTER_SIZE - 1, 'a',
1), while the another is playing ftruncate(fd, 2*CLUSTER_SIZE) and then
ftruncate(fd, CLUSTER_SIZE) again and again.

This is the backtrace when the deadlock happens:

   __wait_on_bit_lock+0x50/0xa0
   __lock_page+0xb7/0xc0
   ocfs2_write_begin_nolock+0x163f/0x1790 [ocfs2]
   ocfs2_page_mkwrite+0x1c7/0x2a0 [ocfs2]
   do_page_mkwrite+0x66/0xc0
   handle_mm_fault+0x685/0x1350
   __do_page_fault+0x1d8/0x4d0
   trace_do_page_fault+0x37/0xf0
   do_async_page_fault+0x19/0x70
   async_page_fault+0x28/0x30

In ocfs2_write_begin_nolock(), we first grab the pages and then allocate
disk space for this write; ocfs2_try_to_free_truncate_log() will be
called if -ENOSPC is returned; if we're lucky to get enough clusters,
which is usually the case, we start over again.

But in ocfs2_free_write_ctxt() the target page isn't unlocked, so we
will deadlock when trying to grab the target page again.

Also, -ENOMEM might be returned in ocfs2_grab_pages_for_write().
Another deadlock will happen in __do_page_mkwrite() if
ocfs2_page_mkwrite() returns non-VM_FAULT_LOCKED, and along with a
locked target page.

These two errors fail on the same path, so fix them by unlocking the
target page manually before ocfs2_free_write_ctxt().

Jan Kara helps me clear out the JBD2 part, and suggest the hint for root
cause.

Changes since v1:
1. Also put ENOMEM error case into consideration.

Link: http://lkml.kernel.org/r/1474173902-32075-1-git-send-email-zren@suse.com
Signed-off-by: Eric Ren <zren@suse.com>
Reviewed-by: He Gang <ghe@suse.com>
Acked-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: workingset: fix crash in shadow node shrinker caused by replace_page_cache_page()
Johannes Weiner [Fri, 30 Sep 2016 22:11:29 +0000 (15:11 -0700)]
mm: workingset: fix crash in shadow node shrinker caused by replace_page_cache_page()

Antonio reports the following crash when using fuse under memory pressure:

  kernel BUG at /build/linux-a2WvEb/linux-4.4.0/mm/workingset.c:346!
  invalid opcode: 0000 [#1] SMP
  Modules linked in: all of them
  CPU: 2 PID: 63 Comm: kswapd0 Not tainted 4.4.0-36-generic #55-Ubuntu
  Hardware name: System manufacturer System Product Name/P8H67-M PRO, BIOS 3904 04/27/2013
  task: ffff88040cae6040 ti: ffff880407488000 task.ti: ffff880407488000
  RIP: shadow_lru_isolate+0x181/0x190
  Call Trace:
    __list_lru_walk_one.isra.3+0x8f/0x130
    list_lru_walk_one+0x23/0x30
    scan_shadow_nodes+0x34/0x50
    shrink_slab.part.40+0x1ed/0x3d0
    shrink_zone+0x2ca/0x2e0
    kswapd+0x51e/0x990
    kthread+0xd8/0xf0
    ret_from_fork+0x3f/0x70

which corresponds to the following sanity check in the shadow node
tracking:

  BUG_ON(node->count & RADIX_TREE_COUNT_MASK);

The workingset code tracks radix tree nodes that exclusively contain
shadow entries of evicted pages in them, and this (somewhat obscure)
line checks whether there are real pages left that would interfere with
reclaim of the radix tree node under memory pressure.

While discussing ways how fuse might sneak pages into the radix tree
past the workingset code, Miklos pointed to replace_page_cache_page(),
and indeed there is a problem there: it properly accounts for the old
page being removed - __delete_from_page_cache() does that - but then
does a raw raw radix_tree_insert(), not accounting for the replacement
page.  Eventually the page count bits in node->count underflow while
leaving the node incorrectly linked to the shadow node LRU.

To address this, make sure replace_page_cache_page() uses the tracked
page insertion code, page_cache_tree_insert().  This fixes the page
accounting and makes sure page-containing nodes are properly unlinked
from the shadow node LRU again.

Also, make the sanity checks a bit less obscure by using the helpers for
checking the number of pages and shadows in a radix tree node.

Fixes: 449dd6984d0e ("mm: keep page cache radix tree nodes in check")
Link: http://lkml.kernel.org/r/20160919155822.29498-1-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Antonio SJ Musumeci <trapexit@spawn.link>
Debugged-by: Miklos Szeredi <miklos@szeredi.hu>
Cc: <stable@vger.kernel.org> [3.15+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoMAINTAINERS: Switch to kernel.org email address for Javi Merino
Javi Merino [Fri, 30 Sep 2016 12:14:28 +0000 (13:14 +0100)]
MAINTAINERS: Switch to  email address for Javi Merino

Change my email address to my kernel.org account instead of the ARM one.

Signed-off-by: Javi Merino <javi.merino@arm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agorxrpc: Fix the call timer handling
David Howells [Fri, 30 Sep 2016 08:13:50 +0000 (09:13 +0100)]
rxrpc: Fix the call timer handling

The call timer's concept of a call timeout (of which there are three) that
is inactive is that it is the timeout has the same expiration time as the
call expiration timeout (the expiration timer is never inactive).  However,
I'm not resetting the timeouts when they expire, leading to repeated
processing of expired timeouts when other timeout events occur.

Fix this by:

 (1) Move the timer expiry detection into rxrpc_set_timer() inside the
     locked section.  This means that if a timeout is set that will expire
     immediately, we deal with it immediately.

 (2) If a timeout is at or before now then it has expired.  When an expiry
     is detected, an event is raised, the timeout is automatically
     inactivated and the event processor is queued.

 (3) If a timeout is at or after the expiry timeout then it is inactive.
     Inactive timeouts do not contribute to the timer setting.

 (4) The call timer callback can now just call rxrpc_set_timer() to handle
     things.

 (5) The call processor work function now checks the event flags rather
     than checking the timeouts directly.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Keep the call timeouts as ktimes rather than jiffies
David Howells [Mon, 26 Sep 2016 21:12:49 +0000 (22:12 +0100)]
rxrpc: Keep the call timeouts as ktimes rather than jiffies

Keep that call timeouts as ktimes rather than jiffies so that they can be
expressed as functions of RTT.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Remove error from struct rxrpc_skb_priv as it is unused
David Howells [Fri, 30 Sep 2016 12:42:31 +0000 (13:42 +0100)]
rxrpc: Remove error from struct rxrpc_skb_priv as it is unused

Remove error from struct rxrpc_skb_priv as it is no longer used.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: The offset field in struct rxrpc_skb_priv is unnecessary
David Howells [Fri, 30 Sep 2016 12:26:03 +0000 (13:26 +0100)]
rxrpc: The offset field in struct rxrpc_skb_priv is unnecessary

The offset field in struct rxrpc_skb_priv is unnecessary as the value can
always be calculated.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Reduce ssthresh to peer's receive window
David Howells [Fri, 30 Sep 2016 08:33:27 +0000 (09:33 +0100)]
rxrpc: Reduce ssthresh to peer's receive window

When we receive an ACK from the peer that tells us what the peer's receive
window (rwind) is, we should reduce ssthresh to rwind if rwind is smaller
than ssthresh.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Switch to Congestion Avoidance mode at cwnd==ssthresh
David Howells [Fri, 30 Sep 2016 08:26:12 +0000 (09:26 +0100)]
rxrpc: Switch to Congestion Avoidance mode at cwnd==ssthresh

Switch to Congestion Avoidance mode at cwnd == ssthresh rather than relying
on cwnd getting incremented beyond ssthresh and the window size, the mode
being shifted and then cwnd being corrected.

We need to make sure we switch into CA mode so that we stop marking every
packet for ACK.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agox86/entry/64: Fix context tracking state warning when load_gs_index fails
Wanpeng Li [Fri, 30 Sep 2016 01:01:06 +0000 (09:01 +0800)]
x86/entry/64: Fix context tracking state warning when load_gs_index fails

This warning:

 WARNING: CPU: 0 PID: 3331 at arch/x86/entry/common.c:45 enter_from_user_mode+0x32/0x50
 CPU: 0 PID: 3331 Comm: ldt_gdt_64 Not tainted 4.8.0-rc7+ #13
 Call Trace:
  dump_stack+0x99/0xd0
  __warn+0xd1/0xf0
  warn_slowpath_null+0x1d/0x20
  enter_from_user_mode+0x32/0x50
  error_entry+0x6d/0xc0
  ? general_protection+0x12/0x30
  ? native_load_gs_index+0xd/0x20
  ? do_set_thread_area+0x19c/0x1f0
  SyS_set_thread_area+0x24/0x30
  do_int80_syscall_32+0x7c/0x220
  entry_INT80_compat+0x38/0x50

... can be reproduced by running the GS testcase of the ldt_gdt test unit in
the x86 selftests.

do_int80_syscall_32() will call enter_form_user_mode() to convert context
tracking state from user state to kernel state. The load_gs_index() call
can fail with user gsbase, gsbase will be fixed up and proceed if this
happen.

However, enter_from_user_mode() will be called again in the fixed up path
though it is context tracking kernel state currently.

This patch fixes it by just fixing up gsbase and telling lockdep that IRQs
are off once load_gs_index() failed with user gsbase.

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1475197266-3440-1-git-send-email-wanpeng.li@hotmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agox86/boot: Initialize FPU and X86_FEATURE_ALWAYS even if we don't have CPUID
Andy Lutomirski [Wed, 28 Sep 2016 23:06:33 +0000 (16:06 -0700)]
x86/boot: Initialize FPU and X86_FEATURE_ALWAYS even if we don't have CPUID

Otherwise arch_task_struct_size == 0 and we die.  While we're at it,
set X86_FEATURE_ALWAYS, too.

Reported-by: David Saggiorato <david@saggiorato.net>
Tested-by: David Saggiorato <david@saggiorato.net>
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave@sr71.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Fixes: aaeb5c01c5b ("x86/fpu, sched: Introduce CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT and use it on x86")
Link: http://lkml.kernel.org/r/8de723afbf0811071185039f9088733188b606c9.1475103911.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
8 years agox86/vdso: Fix building on big endian host
Segher Boessenkool [Thu, 29 Sep 2016 11:51:00 +0000 (11:51 +0000)]
x86/vdso: Fix building on big endian host

We need to call GET_LE to read hdr->e_type.

Fixes: 57f90c3dfc75 ("x86/vdso: Error out if the vDSO isn't a valid DSO")
Reported-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Segher Boessenkool <segher@kernel.crashing.org>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: linux-next@vger.kernel.org
Link: http://lkml.kernel.org/r/20160929193442.GA16617@gate.crashing.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
8 years agox86/boot: Fix another __read_cr4() case on 486
Andy Lutomirski [Thu, 29 Sep 2016 19:48:11 +0000 (12:48 -0700)]
x86/boot: Fix another __read_cr4() case on 486

The condition for reading CR4 was wrong: there are some CPUs with
CPUID but not CR4.  Rather than trying to make the condition exact,
use __read_cr4_safe().

Fixes: 18bc7bd523e0 ("x86/boot: Synchronize trampoline_cr4_features and mmu_cr4_features directly")
Reported-by: david@saggiorato.net
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Reviewed-by: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Link: http://lkml.kernel.org/r/8c453a61c4f44ab6ff43c29780ba04835234d2e5.1475178369.git.luto@kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
8 years agorxrpc: Actually display the tx_data trace retransmission note
David Howells [Fri, 30 Sep 2016 07:50:42 +0000 (08:50 +0100)]
rxrpc: Actually display the tx_data trace retransmission note

Actually display in the tx_data trace the retransmission note added in a
previous patch.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agomlx5: Add ndo_poll_controller() implementation
Calvin Owens [Thu, 29 Sep 2016 04:46:39 +0000 (21:46 -0700)]
mlx5: Add ndo_poll_controller() implementation

This implements ndo_poll_controller in net_device_ops callbacks for mlx5,
which is necessary to use netconsole with this driver.

Acked-By: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Calvin Owens <calvinowens@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonfp: bpf: zero extend 4 byte context loads
Jakub Kicinski [Wed, 28 Sep 2016 22:47:37 +0000 (23:47 +0100)]
nfp: bpf: zero extend 4 byte context loads

Set upper 32 bits of destination register to zeros after
load from the context structure.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosctp: fix the issue sctp_diag uses lock_sock in rcu_read_lock
Xin Long [Wed, 28 Sep 2016 18:55:44 +0000 (02:55 +0800)]
sctp: fix the issue sctp_diag uses lock_sock in rcu_read_lock

When sctp dumps all the ep->assocs, it needs to lock_sock first,
but now it locks sock in rcu_read_lock, and lock_sock may sleep,
which would break rcu_read_lock.

This patch is to get and hold one sock when traversing the list.
After that and get out of rcu_read_lock, lock and dump it. Then
it will traverse the list again to get the next one until all
sctp socks are dumped.

For sctp_diag_dump_one, it fixes this issue by holding asoc and
moving cb() out of rcu_read_lock in sctp_transport_lookup_process.

Fixes: 8f840e47f190 ("sctp: add the sctp_diag.c file")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'sctp-fixes'
David S. Miller [Fri, 30 Sep 2016 06:07:10 +0000 (02:07 -0400)]
Merge branch 'sctp-fixes'

Xin Long says:

====================
sctp: a bunch of fixes for prsctp polices

This patchset is to fix 2 issues for prsctp polices:

  1. patch 1 and 2 fix "netperf-Throughput_Mbps -37.2% regression" issue
     when overloading the CPU.

  2. patch 3 fix "prsctp polices should check both sides' prsctp_capable,
     instead of only local side".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosctp: change to check peer prsctp_capable when using prsctp polices
Xin Long [Wed, 28 Sep 2016 18:37:28 +0000 (02:37 +0800)]
sctp: change to check peer prsctp_capable when using prsctp polices

Now before using prsctp polices, sctp uses asoc->prsctp_enable to
check if prsctp is enabled. However asoc->prsctp_enable is set only
means local host support prsctp, sctp should not abandon packet if
peer host doesn't enable prsctp.

So this patch is to use asoc->peer.prsctp_capable to check if prsctp
is enabled on both side, instead of asoc->prsctp_enable, as asoc's
peer.prsctp_capable is set only when local and peer both enable prsctp.

Fixes: a6c2f792873a ("sctp: implement prsctp TTL policy")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosctp: remove prsctp_param from sctp_chunk
Xin Long [Wed, 28 Sep 2016 18:37:27 +0000 (02:37 +0800)]
sctp: remove prsctp_param from sctp_chunk

Now sctp uses chunk->prsctp_param to save the prsctp param for all the
prsctp polices, we didn't need to introduce prsctp_param to sctp_chunk.
We can just use chunk->sinfo.sinfo_timetolive for RTX and BUF polices,
and reuse msg->expires_at for TTL policy, as the prsctp polices and old
expires policy are mutual exclusive.

This patch is to remove prsctp_param from sctp_chunk, and reuse msg's
expires_at for TTL and chunk's sinfo.sinfo_timetolive for RTX and BUF
polices.

Note that sctp can't use chunk's sinfo.sinfo_timetolive for TTL policy,
as it needs a u64 variables to save the expires_at time.

This one also fixes the "netperf-Throughput_Mbps -37.2% regression"
issue.

Fixes: a6c2f792873a ("sctp: implement prsctp TTL policy")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosctp: move sent_count to the memory hole in sctp_chunk
Xin Long [Wed, 28 Sep 2016 18:37:26 +0000 (02:37 +0800)]
sctp: move sent_count to the memory hole in sctp_chunk

Now pahole sctp_chunk, it has 2 memory holes:
   struct sctp_chunk {
struct list_head           list;
atomic_t                   refcnt;
/* XXX 4 bytes hole, try to pack */
...
long unsigned int          prsctp_param;
int                        sent_count;
/* XXX 4 bytes hole, try to pack */

This patch is to move up sent_count to fill the 1st one and eliminate
the 2nd one.

It's not just another struct compaction, it also fixes the "netperf-
Throughput_Mbps -37.2% regression" issue when overloading the CPU.

Fixes: a6c2f792873a ("sctp: implement prsctp TTL policy")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlx4: remove unused fields
David Decotigny [Wed, 28 Sep 2016 18:00:04 +0000 (11:00 -0700)]
mlx4: remove unused fields

This also can address following UBSAN warnings:
[   36.640343] ================================================================================
[   36.648772] UBSAN: Undefined behaviour in drivers/net/ethernet/mellanox/mlx4/fw.c:857:26
[   36.656853] shift exponent 64 is too large for 32-bit type 'int'
[   36.663348] ================================================================================
[   36.671783] ================================================================================
[   36.680213] UBSAN: Undefined behaviour in drivers/net/ethernet/mellanox/mlx4/fw.c:861:27
[   36.688297] shift exponent 35 is too large for 32-bit type 'int'
[   36.694702] ================================================================================

Tested:
  reboot with UBSAN, no warning.

Signed-off-by: David Decotigny <decot@googlers.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6 addrconf: implement RFC7559 router solicitation backoff
Maciej Żenczykowski [Wed, 28 Sep 2016 06:57:58 +0000 (23:57 -0700)]
ipv6 addrconf: implement RFC7559 router solicitation backoff

This implements:
  https://tools.ietf.org/html/rfc7559

Backoff is performed according to RFC3315 section 14:
  https://tools.ietf.org/html/rfc3315#section-14

We allow setting /proc/sys/net/ipv6/conf/*/router_solicitations
to a negative value meaning an unlimited number of retransmits,
and we make this the new default (inline with the RFC).

We also add a new setting:
  /proc/sys/net/ipv6/conf/*/router_solicitation_max_interval
defaulting to 1 hour (per RFC recommendation).

Signed-off-by: Maciej Żenczykowski <maze@google.com>
Acked-by: Erik Kline <ek@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'net_proc_perf'
David S. Miller [Fri, 30 Sep 2016 05:50:57 +0000 (01:50 -0400)]
Merge branch 'net_proc_perf'

Jia He says:

====================
Reduce cache miss for snmp_fold_field

In a PowerPc server with large cpu number(160), besides commit
a3a773726c9f ("net: Optimize snmp stat aggregation by walking all
the percpu data at once"), I watched several other snmp_fold_field
callsites which would cause high cache miss rate.

test source code:
================
My simple test case, which read from the procfs items endlessly:
/***********************************************************/
int main(int argc, char **argv)
{
        int i;
        int fd = -1 ;
        int rdsize = 0;
        char buf[LINELEN+1];

        buf[LINELEN] = 0;
        memset(buf,0,LINELEN);

        if(1 >= argc) {
                printf("file name empty\n");
                return -1;
        }

        fd = open(argv[1], O_RDWR, 0644);
        if(0 > fd){
                printf("open error\n");
                return -2;
        }

        for(i=0;i<0xffffffff;i++) {
                while(0 < (rdsize = read(fd,buf,LINELEN))){
                        //nothing here
                }

                lseek(fd, 0, SEEK_SET);
        }

        close(fd);
        return 0;
}
/**********************************************************/

compile and run:
================
gcc test.c -o test

perf stat -d -e cache-misses ./test /proc/net/snmp
perf stat -d -e cache-misses ./test /proc/net/snmp6
perf stat -d -e cache-misses ./test /proc/net/sctp/snmp
perf stat -d -e cache-misses ./test /proc/net/xfrm_stat

before the patch set:
====================
 Performance counter stats for 'system wide':

         355911097      cache-misses                                                 [40.08%]
        2356829300      L1-dcache-loads                                              [60.04%]
         355642645      L1-dcache-load-misses     #   15.09% of all L1-dcache hits   [60.02%]
         346544541      LLC-loads                                                    [59.97%]
            389763      LLC-load-misses           #    0.11% of all LL-cache hits    [40.02%]

       6.245162638 seconds time elapsed

After the patch set:
===================
 Performance counter stats for 'system wide':

         194992476      cache-misses                                                 [40.03%]
        6718051877      L1-dcache-loads                                              [60.07%]
         194871921      L1-dcache-load-misses     #    2.90% of all L1-dcache hits   [60.11%]
         187632232      LLC-loads                                                    [60.04%]
            464466      LLC-load-misses           #    0.25% of all LL-cache hits    [39.89%]

       6.868422769 seconds time elapsed
The cache-miss rate can be reduced from 15% to 2.9%

changelog
=========
v6:
- correct v5
v5:
- order local variables from longest to shortest line
v4:
- move memset into one block of if statement in snmp6_seq_show_item
- remove the changes in netstat_seq_show considerred the stack usage is too large
v3:
- introduce generic interface (suggested by Marcelo Ricardo Leitner)
- use max_t instead of self defined macro (suggested by David Miller)
v2:
- fix bug in udplite statistics.
- snmp_seq_show is split into 2 parts
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: Suppress the "Comparison to NULL could be written" warnings
Jia He [Fri, 30 Sep 2016 03:29:04 +0000 (11:29 +0800)]
net: Suppress the "Comparison to NULL could be written" warnings

This is to suppress the checkpatch.pl warning "Comparison to NULL
could be written". No functional changes here.

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoipv6: Remove useless parameter in __snmp6_fill_statsdev
Jia He [Fri, 30 Sep 2016 03:29:03 +0000 (11:29 +0800)]
ipv6: Remove useless parameter in __snmp6_fill_statsdev

The parameter items(is always ICMP6_MIB_MAX) is useless for __snmp6_fill_statsdev

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoproc: Reduce cache miss in xfrm_statistics_seq_show
Jia He [Fri, 30 Sep 2016 03:29:02 +0000 (11:29 +0800)]
proc: Reduce cache miss in xfrm_statistics_seq_show

This is to use the generic interfaces snmp_get_cpu_field{,64}_batch to
aggregate the data by going through all the items of each cpu sequentially.

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoproc: Reduce cache miss in sctp_snmp_seq_show
Jia He [Fri, 30 Sep 2016 03:29:01 +0000 (11:29 +0800)]
proc: Reduce cache miss in sctp_snmp_seq_show

This is to use the generic interfaces snmp_get_cpu_field{,64}_batch to
aggregate the data by going through all the items of each cpu sequentially.

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoproc: Reduce cache miss in snmp6_seq_show
Jia He [Fri, 30 Sep 2016 03:29:00 +0000 (11:29 +0800)]
proc: Reduce cache miss in snmp6_seq_show

This is to use the generic interfaces snmp_get_cpu_field{,64}_batch to
aggregate the data by going through all the items of each cpu sequentially.

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoproc: Reduce cache miss in snmp_seq_show
Jia He [Fri, 30 Sep 2016 03:28:59 +0000 (11:28 +0800)]
proc: Reduce cache miss in snmp_seq_show

This is to use the generic interfaces snmp_get_cpu_field{,64}_batch to
aggregate the data by going through all the items of each cpu sequentially.
Then snmp_seq_show is split into 2 parts to avoid build warning "the frame
size" larger than 1024.

Signed-off-by: Jia He <hejianet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet:snmp: Introduce generic interfaces for snmp_get_cpu_field{, 64}
Jia He [Fri, 30 Sep 2016 03:28:58 +0000 (11:28 +0800)]
net:snmp: Introduce generic interfaces for snmp_get_cpu_field{, 64}

This is to introduce the generic interfaces for snmp_get_cpu_field{,64}.
It exchanges the two for-loops for collecting the percpu statistics data.
This can aggregate the data by going through all the items of each cpu
sequentially.

Signed-off-by: Jia He <hejianet@gmail.com>
Suggested-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'rxrpc-rewrite-20160929' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 30 Sep 2016 05:46:36 +0000 (01:46 -0400)]
Merge tag 'rxrpc-rewrite-20160929' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Fixes and adjustments

This set of patches contains some fixes and adjustments:

 (1) Connections for exclusive calls are being reused because the check to
     work out whether to set RXRPC_CONN_DONT_REUSE is checking where the
     parameters will be copied to (but haven't yet).

 (2) Make Tx loss-injection go through the normal return, so the state gets
     set correctly for what the code thinks it has done.

     Note lost Tx packets in the tx_data trace rather than the skb
     tracepoint.

 (3) Activate channels according to the current state from within the
     channel_lock to avoid someone changing it on us.

 (4) Reduce the local endpoint's services list to a single pointer as we
     don't allow service AF_RXRPC sockets to share UDP ports with other
     AF_RXRPC sockets, so there can't be more than one element in the list.

 (5) Request more ACKs in slow-start mode to help monitor the state driving
     the window configuration.

 (6) Display the serial number of the packet being ACK'd rather than the
     ACK packet's own serial number in the congestion trace as this can be
     related to other entries in the trace.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'wireless-drivers-next-for-davem-2016-09-29' of git://git.kernel.org/pub...
David S. Miller [Fri, 30 Sep 2016 05:31:51 +0000 (01:31 -0400)]
Merge tag 'wireless-drivers-next-for-davem-2016-09-29' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

wireless-drivers-next patches for 4.9

Major changes:

iwlwifi

* work for new hardware support continues
* dynamic queue allocation stabilization
* improvements in the MSIx code
* multiqueue support work continues
* new firmware version support (API 26)
* add 8275 series support
* add 9560 series support
* add support for MU-MIMO sniffer
* add support for RRM by scan
* add support for "reverse" rx packet injection faking hw descriptors
* migrate to devm memory allocation handling
* Remove support for older firmwares (API older than -17 and -22)

wl12xx

* support booting the same rootfs with both wl12xx and wl18xx

hostap

* mark the driver as obsolete

ath9k

* disable RNG by default
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotg3: Avoid NULL pointer dereference in tg3_io_error_detected()
Milton Miller [Thu, 29 Sep 2016 16:24:08 +0000 (13:24 -0300)]
tg3: Avoid NULL pointer dereference in tg3_io_error_detected()

While the driver is probing the adapter, an error may occur before the
netdev structure is allocated and attached to pci_dev. In this case,
not only netdev isn't available, but the tg3 private structure is also
not available as it is just math from the NULL pointer, so dereferences
must be skipped.

The following trace is seen when the error is triggered:

  [1.402247] Unable to handle kernel paging request for data at address 0x00001a99
  [1.402410] Faulting instruction address: 0xc0000000007e33f8
  [1.402450] Oops: Kernel access of bad area, sig: 11 [#1]
  [1.402481] SMP NR_CPUS=2048 NUMA PowerNV
  [1.402513] Modules linked in:
  [1.402545] CPU: 0 PID: 651 Comm: eehd Not tainted 4.4.0-36-generic #55-Ubuntu
  [1.402591] task: c000001fe4e42a20 ti: c000001fe4e88000 task.ti: c000001fe4e88000
  [1.402742] NIP: c0000000007e33f8 LR: c0000000007e3164 CTR: c000000000595ea0
  [1.402787] REGS: c000001fe4e8b790 TRAP: 0300   Not tainted  (4.4.0-36-generic)
  [1.402832] MSR: 9000000100009033 <SF,HV,EE,ME,IR,DR,RI,LE>  CR: 28000422  XER: 20000000
  [1.403058] CFAR: c000000000008468 DAR: 0000000000001a99 DSISR: 42000000 SOFTE: 1
  GPR00: c0000000007e3164 c000001fe4e8ba10 c0000000015c5e00 0000000000000000
  GPR04: 0000000000000001 0000000000000000 0000000000000039 0000000000000299
  GPR08: 0000000000000000 0000000000000001 c000001fe4e88000 0000000000000006
  GPR12: 0000000000000000 c00000000fb40000 c0000000000e6558 c000003ca1bffd00
  GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
  GPR20: 0000000000000000 0000000000000000 0000000000000000 c000000000d52768
  GPR24: c000000000d52740 0000000000000100 c000003ca1b52000 0000000000000002
  GPR28: 0000000000000900 0000000000000000 c00000000152a0c0 c000003ca1b52000
  [1.404226] NIP [c0000000007e33f8] tg3_io_error_detected+0x308/0x340
  [1.404265] LR [c0000000007e3164] tg3_io_error_detected+0x74/0x340

This patch avoids the NULL pointer dereference by moving the access after
the netdev NULL pointer check on tg3_io_error_detected(). Also, we add a
check for netdev being NULL on tg3_io_resume() [suggested by Michael Chan].

Fixes: 0486a063b1ff ("tg3: prevent ifup/ifdown during PCI error recovery")
Fixes: dfc8f370316b ("net/tg3: Release IRQs on permanent error")
Tested-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
Signed-off-by: Milton Miller <miltonm@us.ibm.com>
Signed-off-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
Acked-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'dsa-global-cosmetics'
David S. Miller [Fri, 30 Sep 2016 05:26:04 +0000 (01:26 -0400)]
Merge branch 'dsa-global-cosmetics'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: Global (1) cosmetics

The Global (1) internal SMI device of Marvell switches is a set of
registers providing support to different units for MAC addresses (ATU),
VLANs (VTU), PHY polling (PPU), etc.

Chips (like 88E6060) may use a different address for it, or have
subtleties in the units (e.g. different number of databases, changing
how registers must be accessed), making it hard to maintain properly.

This patchset is a first step to polish the Global (1) support, with no
functional changes though. Here's basically what it does:

  - add helpers to access Global1 registers (same for Global2)
  - remove a few family checks (VTU/STU FID registers)
  - s/mv88e6xxx_vtu_stu_entry/mv88e6xxx_vtu_entry/
  - add a per-chip mv88e6xxx_ops structure of function pointers:

  struct mv88e6xxx_ops {
        int (*get_eeprom)(struct mv88e6xxx_chip *chip,
                          struct ethtool_eeprom *eeprom, u8 *data);
        int (*set_eeprom)(struct mv88e6xxx_chip *chip,
                          struct ethtool_eeprom *eeprom, u8 *data);

        int (*set_switch_mac)(struct mv88e6xxx_chip *chip, u8 *addr);

        int (*phy_read)(struct mv88e6xxx_chip *chip, int addr, int reg,
                        u16 *val);
        int (*phy_write)(struct mv88e6xxx_chip *chip, int addr, int reg,
                         u16 val);
  };

Future patchsets will add ATU/VTU ops, software reset, etc.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: add eeprom ops
Vivien Didelot [Thu, 29 Sep 2016 16:22:02 +0000 (12:22 -0400)]
net: dsa: mv88e6xxx: add eeprom ops

Remove EEPROM flags in favor of new {get,set}_eeprom chip-wide
functions in the mv88e6xxx_ops structure.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: add set_switch_mac to ops
Vivien Didelot [Thu, 29 Sep 2016 16:22:01 +0000 (12:22 -0400)]
net: dsa: mv88e6xxx: add set_switch_mac to ops

Add a set_switch_mac chip-wide function to mv88e6xxx_ops and remove
MV88E6XXX_FLAG_G2_SWITCH_MAC flags.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: add chip-wide ops
Vivien Didelot [Thu, 29 Sep 2016 16:22:00 +0000 (12:22 -0400)]
net: dsa: mv88e6xxx: add chip-wide ops

Introduce a mv88e6xxx_ops structure to describe supported chip-wide
functions and assign the correct variant to the chip models.

For the moment, add only PHY access routines. This allows to get rid of
the PHY ops structures and the usage of PHY flags.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: rename mv88e6xxx_ops
Vivien Didelot [Thu, 29 Sep 2016 16:21:59 +0000 (12:21 -0400)]
net: dsa: mv88e6xxx: rename mv88e6xxx_ops

The mv88e6xxx_ops is used to describe how to access the chip registers.
It can be through SMI (via an MDIO bus), or via another interface such
as crafted remote management frames.

The correct BUS operations structure is chosen at runtime, depending on
the chip address and connectivity.

We will need the mv88e6xxx_ops name for future chip-wide operation
structure, thus rename mv88e6xxx_ops to more explicit mv88e6xxx_bus_ops.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: rename mv88e6xxx_vtu_stu_entry
Vivien Didelot [Thu, 29 Sep 2016 16:21:58 +0000 (12:21 -0400)]
net: dsa: mv88e6xxx: rename mv88e6xxx_vtu_stu_entry

The STU (if the switch has one) is abstracted and accessed through the
VTU operations and data registers.

Thus rename the mv88e6xxx_vtu_stu_entry struct to mv88e6xxx_vtu_entry.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: add mv88e6xxx_num_ports helper
Vivien Didelot [Thu, 29 Sep 2016 16:21:57 +0000 (12:21 -0400)]
net: dsa: mv88e6xxx: add mv88e6xxx_num_ports helper

Add an mv88e6xxx_num_ports helper instead of digging in the chip info
structure.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: expose mv88e6xxx_num_databases
Vivien Didelot [Thu, 29 Sep 2016 16:21:56 +0000 (12:21 -0400)]
net: dsa: mv88e6xxx: expose mv88e6xxx_num_databases

The mv88e6xxx_num_databases will be used by shared code, so move it
inline to the header file.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>