GitHub/moto-9609/android_kernel_motorola_exynos9610.git
8 years agoMerge branch 'be2net-next'
David S. Miller [Mon, 27 Jun 2016 08:06:31 +0000 (04:06 -0400)]
Merge branch 'be2net-next'

Sathya Perla says:

====================
be2net: patch set

Hi Dave, pls consider commiting the following patches to the net-next tree.
Thanks!

Patch 1 replaces the be_max_eqs() macro with two new macros called
be_max_nic_eqs() and be_max_func_eqs() to clear confusion in that part
of the code.

Patch 2 adds support to configure asymmetric number of rx/tx queues via
ethtool set-channels option.

Patch 3 disables EVB when VFs are not enabled on a BE3 SR-IOV config to
avoid the broadcast echo problem.

Patch 4 updates copyright markings in be2net src files

Patch 5 updates the be2net maintainers' list
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobe2net: update be2net maintainers list
Sathya Perla [Wed, 22 Jun 2016 12:54:57 +0000 (08:54 -0400)]
be2net: update be2net maintainers list

This patch removes Padmanabh's name from the maintainers list as he's no
longer with the company. It also adds the driver name on the headline to
make it easy to lookup the maintainers list by the driver name.

Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobe2net: Change copyright markings in source files
Somnath Kotur [Wed, 22 Jun 2016 12:54:56 +0000 (08:54 -0400)]
be2net: Change copyright markings in source files

This patch updates year and company name in the copyright markings in the
be2net source files.

Signed-off-by: Somnath Kotur <somnath.kotur@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobe2net: Fix broadcast echoes from EVB in BE3
Somnath Kotur [Wed, 22 Jun 2016 12:54:55 +0000 (08:54 -0400)]
be2net: Fix broadcast echoes from EVB in BE3

On SR-IOV profiles, when the user connects a Linux Bridge or OVS to a BE3
vport, they suffer the "broadcast/multicast echo" problem. BE3 EVB echoes
broadcast and multicast packets back to PF's vport confusing the
Linux bridge.  BE3 relies on the src-mac addr being programmed on the
interface to avoid sending back an echo of a broadcast or multicast packet
on a vPort. When a Linux bridge is connected to a BE3, the mac-addr of the
VM behind the bridge doesn't get configured on the vPort and so echo
cancellation doesn't work.
This patch worksaround this problem by disabling the EVB initially
and re-enabling it *only* when SR-IOV is enabled by the user. For the
driver fix to work, the BE3 FW version must be >= 11.1.84.0.

Signed-off-by: Somnath Kotur <somnath.kotur@emulex.com>
Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobe2net: support asymmetric rx/tx queue counts
Sathya Perla [Wed, 22 Jun 2016 12:54:54 +0000 (08:54 -0400)]
be2net: support asymmetric rx/tx queue counts

be2net so far supported creation of RX/TX queues only in pairs.
On configs where rx and tx queue counts are different, creation of only
the lesser number of queues has been supported.

This patch now allows a combination of RX/TX-only channels along with
combined channels. N TX-queues and M RX-queues can be created with the
following cmds:
ethtool -L ethX combined N rx M-N  (when N < M)
ethtool -L ethX combined M tx N-M (when M < N)

Setting both RX-only and TX-only channels is still not supported.
It is mandatory to create atleast one combined channel.

Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobe2net: fix definition of be_max_eqs()
Sathya Perla [Wed, 22 Jun 2016 12:54:53 +0000 (08:54 -0400)]
be2net: fix definition of be_max_eqs()

The EQs available on a function are shared between NIC and RoCE.
The be_max_eqs() macro was so far being used to refer to the max number of
EQs available for NIC. This has caused some confusion in the code. To fix
this confusion this patch introduces a new macro called be_max_nic_eqs()
to refer to the max number of EQs avialable for NIC only and renames
be_max_eqs() to be_max_func_eqs().

Signed-off-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'fec-new-type-device'
David S. Miller [Mon, 27 Jun 2016 08:02:01 +0000 (04:02 -0400)]
Merge branch 'fec-new-type-device'

Andy Duan says:

====================
net: fec: add new type device

Different i.MX SOC FEC support different features like :
    - i.MX6Q/DL FEC does not support AVB and interrupt coalesc
    - i.MX6SX/i.MX7D supports AVB and interrupt coalesc
    - i.MX6UL/ULL does not support AVB, but support interrupt coalesc

Then, add new quirk flag to judge the supported features, and add new
type device for i.MX6UL.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: fec: use a more proper compatible string for i.MX6UL type device
Fugang Duan [Wed, 22 Jun 2016 10:52:36 +0000 (18:52 +0800)]
net: fec: use a more proper compatible string for i.MX6UL type device

i.MX6UL is a member in i.MX series family, the SOC FEC inherits from
i.MX6SX but removes some IP features, lets define a new type for fec
device.

Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: fec: add interrupt coalesc quirk flag
Fugang Duan [Wed, 22 Jun 2016 10:52:35 +0000 (18:52 +0800)]
net: fec: add interrupt coalesc quirk flag

Different i.MX SOC FEC support different features like :
- i.MX6Q/DL FEC does not support AVB and interrupt coalesc
- i.MX6SX/i.MX7D supports AVB and interrupt coalesc
- i.MX6UL/ULL does not support AVB, but support interrupt coalesc

So, add new quirk flag to judge the supported features.

Signed-off-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'rxrpc-rewrite-20160622-2' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Sun, 26 Jun 2016 20:01:54 +0000 (16:01 -0400)]
Merge tag 'rxrpc-rewrite-20160622-2' of git://git./linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Get rid of conn bundle and transport structs

Here's the next part of the AF_RXRPC rewrite.  The primary purpose of this
set is to get rid of the rxrpc_conn_bundle and rxrpc_transport structs.
This simplifies things for future development of the connection handling.

To this end, the following significant changes are made:

 (1) The rxrpc_connection struct is given pointers to the local and peer
     endpoints, inside the rxrpc_conn_parameters struct.  Pointers to the
     transport's copy of these pointers are then redirected to the
     connection struct.

 (2) Exclusive connection handling is fixed.  Exclusive connections should
     do just one call and then be retired.  They are used in security
     negotiations and, I believe, the idea is to avoid reuse of negotiated
     security contexts.

     The current code is doing a single connection per socket and doing all
     the calls over that.  With this change it gets a new connection for
     each call made.

 (3) A new sendmsg() control message marker is added to make individual
     calls operate over exclusive connections.  This should be used in
     future in preference to the sockopt that marks a socket as "exclusive
     connection".

 (4) IDs for client connections initiated by a machine are now allocated
     from a global pool using the IDR facility and are unique across all
     client connections, no matter their destination.  The IDR facility is
     then used to look up a connection on the connection ID alone.  Other
     parameters are then verified afterwards.

     Note that the IDR facility may use a lot of memory if the IDs it holds
     are widely scattered.  Given this, in a future commit, client
     connections will be retired if they are more than a certain distance
     from the last ID allocated.

     The client epoch is advanced by 1 each time the client ID counter
     wraps.  Connections outside the current epoch will also be retired in
     a future commit.

 (5) The connection bundle concept is removed and the client connection
     tree is moved into the local endpoint.  The queue for waiting for a
     call channel is moved to the rxrpc_connection struct as there can only
     be one connection for any particular key going to any particular peer
     now.

 (6) The rxrpc_transport struct is removed and the service connection tree
     is moved into the peer struct.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: stmmac: dwmac-rk: add rk3228-specific data
Xing Zheng [Tue, 21 Jun 2016 12:33:28 +0000 (20:33 +0800)]
net: stmmac: dwmac-rk: add rk3228-specific data

Add constants and callback functions for the dwmac on rk3228/rk3229 socs.
As can be seen, the base structure is the same, only registers and the
bits in them moved slightly.

Signed-off-by: Xing Zheng <zhengxing@rock-chips.com>
Reviewed-by: Heiko Stuebner <heiko@sntech.de>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'net-sched-bulk-dequeue'
David S. Miller [Sat, 25 Jun 2016 16:19:41 +0000 (12:19 -0400)]
Merge branch 'net-sched-bulk-dequeue'

Eric Dumazet says:

====================
net_sched: bulk dequeue and deferred drops

First patch adds an additional parameter to ->enqueue() qdisc method
so that drops can be done outside of critical section
(after locks are released).

Then fq_codel can have a small optimization to reduce number of cache
lines misses during a drop event
(possibly accumulating hundreds of packets to be freed).

A small htb change exports the backlog in class dumps.

Final patch adds bulk dequeue to qdiscs that were lacking this feature.

This series brings a nice qdisc performance increase (more than 80 %
in some cases).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: generalize bulk dequeue
Eric Dumazet [Wed, 22 Jun 2016 06:16:52 +0000 (23:16 -0700)]
net_sched: generalize bulk dequeue

When qdisc bulk dequeue was added in linux-3.18 (commit
5772e9a3463b "qdisc: bulk dequeue support for qdiscs
with TCQ_F_ONETXQUEUE"), it was constrained to some
specific qdiscs.

With some extra care, we can extend this to all qdiscs,
so that typical traffic shaping solutions can benefit from
small batches (8 packets in this patch).

For example, HTB is often used on some multi queue device.
And bonding/team are multi queue devices...

Idea is to bulk-dequeue packets mapping to the same transmit queue.

This brings between 35 and 80 % performance increase in HTB setup
under pressure on a bonding setup :

1) NUMA node contention :   610,000 pps -> 1,110,000 pps
2) No node contention   : 1,380,000 pps -> 1,930,000 pps

Now we should work to add batches on the enqueue() side ;)

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.r.fastabend@intel.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Florian Westphal <fw@strlen.de>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: sch_htb: export class backlog in dumps
Eric Dumazet [Wed, 22 Jun 2016 06:16:51 +0000 (23:16 -0700)]
net_sched: sch_htb: export class backlog in dumps

We already get child qdisc qlen, we also can get its backlog
so that class dumps can report it.

Also replace qstats by a single drop counter, but move it in
a separate cache line so that drops do not dirty useful cache lines.

Tested:

$ tc -s cl sh dev eth0
class htb 1:1 root leaf 3: prio 0 rate 1Gbit ceil 1Gbit burst 500000b cburst 500000b
 Sent 2183346912 bytes 9021815 pkt (dropped 2340774, overlimits 0 requeues 0)
 rate 1001Mbit 517543pps backlog 120758b 499p requeues 0
 lended: 9021770 borrowed: 0 giants: 0
 tokens: 9 ctokens: 9

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: fq_codel: cache skb->truesize into skb->cb
Eric Dumazet [Wed, 22 Jun 2016 06:16:50 +0000 (23:16 -0700)]
net_sched: fq_codel: cache skb->truesize into skb->cb

Now we defer skb drops, it makes sense to keep a copy
of skb->truesize in struct codel_skb_cb to avoid one
cache line miss per dropped skb in fq_codel_drop(),
to reduce latencies a bit further.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet_sched: drop packets after root qdisc lock is released
Eric Dumazet [Wed, 22 Jun 2016 06:16:49 +0000 (23:16 -0700)]
net_sched: drop packets after root qdisc lock is released

Qdisc performance suffers when packets are dropped at enqueue()
time because drops (kfree_skb()) are done while qdisc lock is held,
delaying a dequeue() draining the queue.

Nominal throughput can be reduced by 50 % when this happens,
at a time we would like the dequeue() to proceed as fast as possible.

Even FQ is vulnerable to this problem, while one of FQ goals was
to provide some flow isolation.

This patch adds a 'struct sk_buff **to_free' parameter to all
qdisc->enqueue(), and in qdisc_drop() helper.

I measured a performance increase of up to 12 %, but this patch
is a prereq so that future batches in enqueue() can fly.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'liquidio-next'
David S. Miller [Sat, 25 Jun 2016 16:08:41 +0000 (12:08 -0400)]
Merge branch 'liquidio-next'

Raghu Vatsavayi says:

====================
liquidio: updates and bug fixes

Please consider following patch series for liquidio bug fixes
and updates on top of net-next. Following patches should be
applied in the following order as some of them depend on
earlier patches in the series.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: ddr timeout
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:15 +0000 (22:53 -0700)]
liquidio: ddr timeout

Adds support for ddr_timeout during device init.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: Support priv flag
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:14 +0000 (22:53 -0700)]
liquidio: Support priv flag

This patch adds support for private flags for the driver.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: ptp info
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:13 +0000 (22:53 -0700)]
liquidio: ptp info

This patch has minor changes for proper ptp info retreival.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: New xaui info
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:12 +0000 (22:53 -0700)]
liquidio: New xaui info

This patch adds support for host driver support for new Xaui
interfaces.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: New statistics support
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:11 +0000 (22:53 -0700)]
liquidio: New statistics support

This patch adds extensive support of statistics for data path,
control path and firmware.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: tx rx interrupt moderation
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:10 +0000 (22:53 -0700)]
liquidio: tx rx interrupt moderation

This patch has new tx/rx interrupt moderation defaults of
count/timer for better throughput and utilisation.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: chip reset changes
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:09 +0000 (22:53 -0700)]
liquidio: chip reset changes

This patch resolves the order of chip reset while destroying
the resources by postoponing soft reset in destroy resources
function until all queues are removed properly.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: New unload state
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:08 +0000 (22:53 -0700)]
liquidio: New unload state

This patch adds new state so that the ctrl packets are not sent
to firmware during unload time and only rx packets are allowed.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: Firmware image download
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:07 +0000 (22:53 -0700)]
liquidio: Firmware image download

This patch has firmware image related changes for: firmware
release upon failure, support latest firmware version and
firmware download in 4MB chunks.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: Napi rx/tx traffic
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:06 +0000 (22:53 -0700)]
liquidio: Napi rx/tx traffic

This Patch adds tx buffer handling  to Napi along with RX
traffic. Also separate spinlocks are introduced for handling
iq posting and buffer reclaim so that tx path and tx interrupt
do not compete against each other.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: Vlan filtering
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:05 +0000 (22:53 -0700)]
liquidio: Vlan filtering

This patch adds supports for Vlan filtering for liquidio driver.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: Vlan offloads changes
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:04 +0000 (22:53 -0700)]
liquidio: Vlan offloads changes

This patch adds support for vlan offloads for the driver and
receive header structures are also modified appropriately. Also
requestID will not be used in reveive header any more.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoliquidio: soft command buffer limits
Raghu Vatsavayi [Wed, 22 Jun 2016 05:53:03 +0000 (22:53 -0700)]
liquidio: soft command buffer limits

This patch increases the limits of soft command buffer size and
num command buffers. This patch also has changes for queue macros
and limit related changes for new chips.

Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <rvatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agocxgb4vf: Synchronize access to mailbox
Hariprasad Shenai [Wed, 22 Jun 2016 04:09:29 +0000 (09:39 +0530)]
cxgb4vf: Synchronize access to mailbox

The issue comes when there are multiple threads attempting to use the
mailbox facility at the same time. The issue is the for the Virtual
Function Driver, the only way to get the Virtual Interface statistics
is to issue mailbox commands to ask the firmware for the VI Stats.
And, because the VI Stats command can only retrieve a smallish number of
stats per mailbox command, we have to issue three mailbox commands in quick
succession. When ethtool or netstat command to get interface stats and
interface up/down is run in a loop for every 0.1 sec, we observed
mailbox collisions. And out of the two commands one would fail with
the present code, since we don't queue the second command.

To overcome the above issue, added a queue to access the mailbox.
Whenever a mailbox command is issued add it to the queue. If its at the
head issue the mailbox command, else wait for the existing command to
complete. Usually command takes less than a milli-second to complete.
Also timeout from the loop, if the command under execution takes
long time to run.

In reality, the number of mailbox access collisions is going to be very
rare since no one runs such abusive script.

Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosamples/bpf: set max locked memory to ulimited
William Tu [Wed, 22 Jun 2016 04:05:58 +0000 (21:05 -0700)]
samples/bpf: set max locked memory to ulimited

Signed-off-by: William Tu <u9012063@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: macb: use phy_ethtool_{get|set}_link_ksettings
Philippe Reynes [Tue, 21 Jun 2016 22:32:36 +0000 (00:32 +0200)]
net: ethernet: macb: use phy_ethtool_{get|set}_link_ksettings

There are two generics functions phy_ethtool_{get|set}_link_ksettings,
so we can use them instead of defining the same code in the driver.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: macb: use phydev from struct net_device
Philippe Reynes [Tue, 21 Jun 2016 22:32:35 +0000 (00:32 +0200)]
net: ethernet: macb: use phydev from struct net_device

The private structure contain a pointer to phydev, but the structure
net_device already contain such pointer. So we can remove the pointer
phydev in the private structure, and update the driver to use the
one contained in struct net_device.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: Only set mark and labels with a commit flag.
Jarno Rajahalme [Tue, 21 Jun 2016 21:59:38 +0000 (14:59 -0700)]
openvswitch: Only set mark and labels with a commit flag.

Only set conntrack mark or labels when the commit flag is specified.
This makes sure we can not set them before the connection has been
persisted, as in that case the mark and labels would be lost in an
event of an userspace upcall.

OVS userspace already requires the commit flag to accept setting
ct_mark and/or ct_labels.  Validate for this in the kernel API.

Signed-off-by: Jarno Rajahalme <jarno@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: Set mark and labels before confirming.
Jarno Rajahalme [Tue, 21 Jun 2016 21:59:37 +0000 (14:59 -0700)]
openvswitch: Set mark and labels before confirming.

Set conntrack mark and labels right before committing so that
the initial conntrack NEW event has the mark and labels.

Signed-off-by: Jarno Rajahalme <jarno@ovn.org>
Acked-by: Joe Stringer <joe@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: rename single-chip support
Vivien Didelot [Tue, 21 Jun 2016 16:28:20 +0000 (12:28 -0400)]
net: dsa: mv88e6xxx: rename single-chip support

With the upcoming support for cross-chip operations, it will be hard to
distinguish portions of code supporting a single-chip or a switch fabric
of interconnected chips.

Make the code clearer now, by renaming the mv88e6xxx_priv_state chip
structure to mv88e6xxx_chip. This patch brings no functional changes.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: move driver in its own folder
Vivien Didelot [Tue, 21 Jun 2016 16:28:19 +0000 (12:28 -0400)]
net: dsa: mv88e6xxx: move driver in its own folder

With the upcoming support for cross-chip operations and other mv88e6xxx
enhancements, new files will be added.

Similarly to mlxsw or b53, move mv88e6xxx files into their own folder.

In the meantime, update the MAINTAINERS entry to please checkpatch.pl,
by replacing the invalid 88E6352 entry with 88E6XXX, maintained by
Andrew and myself.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'qed-coalesce'
David S. Miller [Thu, 23 Jun 2016 19:56:54 +0000 (15:56 -0400)]
Merge branch 'qed-coalesce'

Sudarsana Reddy Kalluru says:

====================
qed*: coalesce parameters config support.

The patch series adds the support for config/read of the adapter coalesce
parameters. Patch (1) adds the qed infrastructure/APIs for the support and
patch (2) adds the driver support for following ethtool commands:
ethtool -c|--show-coalesce ethX
ethtool -C|--coalesce ethX [rx-usecs N] [tx-usecs N]
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqede: Add support for coalescing config read/update.
Sudarsana Reddy Kalluru [Tue, 21 Jun 2016 13:36:22 +0000 (09:36 -0400)]
qede: Add support for coalescing config read/update.

Signed-off-by: Sudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoqed: Add support for coalescing config read/update.
Sudarsana Reddy Kalluru [Tue, 21 Jun 2016 13:36:21 +0000 (09:36 -0400)]
qed: Add support for coalescing config read/update.

This patch adds support for configuring the device tx/rx coalescing
timeout values in the order of micro seconds. It also adds APIs for
upper layer drivers for reading/updating the coalescing values.

Signed-off-by: Sudarsana Reddy Kalluru <sudarsana.kalluru@qlogic.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'wireless-drivers-next-for-davem-2016-06-21' of git://git.kernel.org/pub...
David S. Miller [Thu, 23 Jun 2016 19:40:31 +0000 (15:40 -0400)]
Merge tag 'wireless-drivers-next-for-davem-2016-06-21' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers patches for 4.8

Major changes:

ath10k

* enable btcoex support without restarting firmware
* enable ipq4019 support using AHB bus
* add QCA9887 chipset support
* retrieve calibration data from EEPROM, currently only for QCA9887

wil6210

* add pm_notify handling

brcmfmac

* add support for the PCIE devices 43525 and 43465
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx4_en: Add DCB PFC support through CEE netlink commands
Rana Shahout [Tue, 21 Jun 2016 09:43:59 +0000 (12:43 +0300)]
net/mlx4_en: Add DCB PFC support through CEE netlink commands

This patch adds support for reading and updating priority flow
control (PFC) attributes in the driver via netlink.

Signed-off-by: Rana Shahout <ranas@mellanox.com>
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: b53: Fix statistics readings
Florian Fainelli [Tue, 21 Jun 2016 01:26:53 +0000 (18:26 -0700)]
net: dsa: b53: Fix statistics readings

Due to a typo we would always be using the MIB counter width of the
first element of the counter array instead of the current element, and
we would always be accessing the register statistics with a 64-bits
read, while some could be 32-bits. This got unnoticed in testing with
MDIO and SRAB which tolerate doing this, but testing with the SPI bus
revealed bogus values being returned. Fix this by using the proper
iterator here.

Fixes: 967dd82ffc52 ("net: dsa: b53: Add support for Broadcom RoboSwitch")
Reported-by: Jonas Gorski <jogo@openwrt.org>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoof_mdio: Enable fixed PHY support if driver is a module
Ben Hutchings [Tue, 21 Jun 2016 00:17:17 +0000 (01:17 +0100)]
of_mdio: Enable fixed PHY support if driver is a module

The fixed_phy driver doesn't have to be built-in, and it's
important that of_mdio supports it even if it's a module.

Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoti_cpsw: Check for disabled child nodes
Ben Hutchings [Tue, 21 Jun 2016 00:16:31 +0000 (01:16 +0100)]
ti_cpsw: Check for disabled child nodes

Dual MAC devices don't necessarily have both MACs wired up, so ignore
those that are disabled.

Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoopenvswitch: Add packet len info to upcall.
William Tu [Mon, 20 Jun 2016 14:26:17 +0000 (07:26 -0700)]
openvswitch: Add packet len info to upcall.

The commit f2a4d086ed4c ("openvswitch: Add packet truncation support.")
introduces packet truncation before sending to userspace upcall receiver.
This patch passes up the skb->len before truncation so that the upcall
receiver knows the original packet size. Potentially this will be used
by sFlow, where OVS translates sFlow config header=N to a sample action,
truncating packet to N byte in kernel datapath. Thus, only N bytes instead
of full-packet size is copied from kernel to userspace, saving the
kernel-to-userspace bandwidth.

Signed-off-by: William Tu <u9012063@gmail.com>
Cc: Pravin Shelar <pshelar@nicira.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: bgmac: use phy_ethtool_{get|set}_link_ksettings
Philippe Reynes [Sun, 19 Jun 2016 20:37:06 +0000 (22:37 +0200)]
net: ethernet: bgmac: use phy_ethtool_{get|set}_link_ksettings

There are two generics functions phy_ethtool_{get|set}_link_ksettings,
so we can use them instead of defining the same code in the driver.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: bgmac: use phydev from struct net_device
Philippe Reynes [Sun, 19 Jun 2016 20:37:05 +0000 (22:37 +0200)]
net: ethernet: bgmac: use phydev from struct net_device

The private structure contain a pointer to phydev, but the structure
net_device already contain such pointer. So we can remove the pointer
phydev in the private structure, and update the driver to use the
one contained in struct net_device.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: altera_tse: use phy_ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 18 Jun 2016 14:37:21 +0000 (16:37 +0200)]
net: ethernet: altera_tse: use phy_ethtool_{get|set}_link_ksettings

There are two generics functions phy_ethtool_{get|set}_link_ksettings,
so we can use them instead of defining the same code in the driver.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: altera_tse: use phydev from struct net_device
Philippe Reynes [Sat, 18 Jun 2016 14:37:20 +0000 (16:37 +0200)]
net: ethernet: altera_tse: use phydev from struct net_device

The private structure contain a pointer to phydev, but the structure
net_device already contain such pointer. So we can remove the pointer
phydev in the private structure, and update the driver to use the
one contained in struct net_device.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: sun4i-emac: use phy_ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 18 Jun 2016 13:15:40 +0000 (15:15 +0200)]
net: ethernet: sun4i-emac: use phy_ethtool_{get|set}_link_ksettings

There are two generics functions phy_ethtool_{get|set}_link_ksettings,
so we can use them instead of defining the same code in the driver.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: ethernet: sun4i-emac: use phydev from struct net_device
Philippe Reynes [Sat, 18 Jun 2016 13:15:39 +0000 (15:15 +0200)]
net: ethernet: sun4i-emac: use phydev from struct net_device

The private structure contain a pointer to phydev, but the structure
net_device already contain such pointer. So we can remove the pointer
phydev in the private structure, and update the driver to use the
one contained in struct net_device.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agorxrpc: Kill off the rxrpc_transport struct
David Howells [Fri, 17 Jun 2016 09:06:56 +0000 (10:06 +0100)]
rxrpc: Kill off the rxrpc_transport struct

The rxrpc_transport struct is now redundant, given that the rxrpc_peer
struct is now per peer port rather than per peer host, so get rid of it.

Service connection lists are transferred to the rxrpc_peer struct, as is
the conn_lock.  Previous patches moved the client connection handling out
of the rxrpc_transport struct and discarded the connection bundling code.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Kill the client connection bundle concept
David Howells [Fri, 17 Jun 2016 14:42:35 +0000 (15:42 +0100)]
rxrpc: Kill the client connection bundle concept

Kill off the concept of maintaining a bundle of connections to a particular
target service to increase the number of call slots available for any
beyond four for that service (there are four call slots per connection).

This will make cleaning up the connection handling code easier and
facilitate removal of the rxrpc_transport struct.  Bundling can be
reintroduced later if necessary.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Provide more refcount helper functions
David Howells [Mon, 4 Apr 2016 13:00:38 +0000 (14:00 +0100)]
rxrpc: Provide more refcount helper functions

Provide refcount helper functions for connections so that the code doesn't
touch local or connection usage counts directly.

Also make it such that local and peer put functions can take a NULL
pointer.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Make rxrpc_send_packet() take a connection not a transport
David Howells [Fri, 17 Jun 2016 10:53:37 +0000 (11:53 +0100)]
rxrpc: Make rxrpc_send_packet() take a connection not a transport

Make rxrpc_send_packet() take a connection not a transport as part of the
phasing out of the rxrpc_transport struct.

Whilst we're at it, rename the function to rxrpc_send_data_packet() to
differentiate it from the other packet sending functions.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Calls displayed in /proc may in future lack a connection
David Howells [Fri, 17 Jun 2016 10:07:55 +0000 (11:07 +0100)]
rxrpc: Calls displayed in /proc may in future lack a connection

Allocated rxrpc calls displayed in /proc/net/rxrpc_calls may in future be
on the proc list before they're connected or after they've been
disconnected - in which case they may not have a pointer to a connection
struct that can be used to get data from there.

Deal with this by using stuff from the call struct in preference where
possible and printing "no_connection" rather than a peer address if no
connection is assigned.

This change also has the added bonus that the service ID is now taken from
the call rather the connection which will allow per-call service upgrades
to be shown - something required for AuriStor server compatibility.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Validate the net address given to rxrpc_kernel_begin_call()
David Howells [Fri, 17 Jun 2016 10:00:48 +0000 (11:00 +0100)]
rxrpc: Validate the net address given to rxrpc_kernel_begin_call()

Validate the net address given to rxrpc_kernel_begin_call() before using
it.

Whilst this should be mostly unnecessary for in-kernel users, it does clear
the tail of the address struct in case we want to hash or compare the whole
thing.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Use IDR to allocate client conn IDs on a machine-wide basis
David Howells [Mon, 4 Apr 2016 13:00:37 +0000 (14:00 +0100)]
rxrpc: Use IDR to allocate client conn IDs on a machine-wide basis

Use the IDR facility to allocate client connection IDs on a machine-wide
basis so that each client connection has a unique identifier.  When the
connection ID space wraps, we advance the epoch by 1, thereby effectively
having a 62-bit ID space.  The IDR facility is then used to look up client
connections during incoming packet routing instead of using an rbtree
rooted on the transport.

This change allows for the removal of the transport in the future and also
means that client connections can be looked up directly in the data-ready
handler by connection ID.

The ID management code is placed in a new file, conn-client.c, to which all
the client connection-specific code will eventually move.

Note that the IDR tree gets very expensive on memory if the connection IDs
are widely scattered throughout the number space, so we shall need to
retire connections that have, say, an ID more than four times the maximum
number of client conns away from the current allocation point to try and
keep the IDs concentrated.  We will also need to retire connections from an
old epoch.

Also note that, for the moment, a pointer to the transport has to be passed
through into the ID allocation function so that we can take a BH lock to
prevent a locking issue against in-BH lookup of client connections.  This
will go away later when RCU is used for server connections also.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: rxrpc_connection_lock shouldn't be a BH lock, but conn_lock is
David Howells [Tue, 21 Jun 2016 15:10:03 +0000 (16:10 +0100)]
rxrpc: rxrpc_connection_lock shouldn't be a BH lock, but conn_lock is

rxrpc_connection_lock shouldn't be accessed as a BH-excluding lock.  It's
only accessed in a few places and none of those are in BH-context.

rxrpc_transport::conn_lock, however, *is* a BH-excluding lock and should be
accessed so consistently.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Pass sk_buff * rather than rxrpc_host_header * to functions
David Howells [Thu, 16 Jun 2016 12:31:07 +0000 (13:31 +0100)]
rxrpc: Pass sk_buff * rather than rxrpc_host_header * to functions

Pass a pointer to struct sk_buff rather than struct rxrpc_host_header to
functions so that they can in the future get at transport protocol parameters
rather than just RxRPC parameters.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Fix exclusive connection handling
David Howells [Mon, 4 Apr 2016 13:00:37 +0000 (14:00 +0100)]
rxrpc: Fix exclusive connection handling

"Exclusive connections" are meant to be used for a single client call and
then scrapped.  The idea is to limit the use of the negotiated security
context.  The current code, however, isn't doing this: it is instead
restricting the socket to a single virtual connection and doing all the
calls over that.

This is changed such that the socket no longer maintains a special virtual
connection over which it will do all the calls, but rather gets a new one
each time a new exclusive call is made.

Further, using a socket option for this is a poor choice.  It should be
done on sendmsg with a control message marker instead so that calls can be
marked exclusive individually.  To that end, add RXRPC_EXCLUSIVE_CALL
which, if passed to sendmsg() as a control message element, will cause the
call to be done on an single-use connection.

The socket option (RXRPC_EXCLUSIVE_CONNECTION) still exists and, if set,
will override any lack of RXRPC_EXCLUSIVE_CALL being specified so that
programs using the setsockopt() will appear to work the same.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Replace conn->trans->{local,peer} with conn->params.{local,peer}
David Howells [Mon, 4 Apr 2016 13:00:36 +0000 (14:00 +0100)]
rxrpc: Replace conn->trans->{local,peer} with conn->params.{local,peer}

Replace accesses of conn->trans->{local,peer} with
conn->params.{local,peer} thus making it easier for a future commit to
remove the rxrpc_transport struct.

This also reduces the number of memory accesses involved.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: Use structs to hold connection params and protocol info
David Howells [Mon, 4 Apr 2016 13:00:36 +0000 (14:00 +0100)]
rxrpc: Use structs to hold connection params and protocol info

Define and use a structure to hold connection parameters.  This makes it
easier to pass multiple connection parameters around.

Define and use a structure to hold protocol information used to hash a
connection for lookup on incoming packet.  Most of these fields will be
disposed of eventually, including the duplicate local pointer.

Whilst we're at it rename "proto" to "family" when referring to a protocol
family.

Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: fix uninitialized variable use
Arnd Bergmann [Fri, 17 Jun 2016 09:55:22 +0000 (11:55 +0200)]
rxrpc: fix uninitialized variable use

Hashing the peer key was introduced for AF_INET, but gcc
warns about the rxrpc_peer_hash_key function returning uninitialized
data for any other value of srx->transport.family:

net/rxrpc/peer_object.c: In function 'rxrpc_peer_hash_key':
net/rxrpc/peer_object.c:57:15: error: 'p' may be used uninitialized in this function [-Werror=maybe-uninitialized]

Assuming that nothing else can be set here, this changes the
function to just return zero in case of an unknown address
family.

Fixes: be6e6707f6ee ("rxrpc: Rework peer object handling to use hash table and RCU")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David Howells <dhowells@redhat.com>
8 years agorxrpc: checking for IS_ERR() instead of NULL
Dan Carpenter [Sat, 18 Jun 2016 08:44:03 +0000 (11:44 +0300)]
rxrpc: checking for IS_ERR() instead of NULL

rxrpc_lookup_peer_rcu() and rxrpc_lookup_peer() return NULL on error, never
error pointers, so IS_ERR() can't be used.

Fix three callers of those functions.

Fixes: be6e6707f6ee ('rxrpc: Rework peer object handling to use hash table and RCU')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David Howells <dhowells@redhat.com>
8 years agoMerge branch 'mlxsw-next'
David S. Miller [Tue, 21 Jun 2016 09:02:55 +0000 (05:02 -0400)]
Merge branch 'mlxsw-next'

Jiri Pirko says:

====================
mlxsw: Preparation for IPv4 router

Ido says:

This series prepares the driver for IPv4 router support. The router follow-up
patches are available at: https://github.com/jpirko/linux_mlxsw/tree/net-next_queue

Patches 1-9 simplify the netdevice notification block and also add several
checks during PRECHANGEUPPER events against topologies that aren't supported by
the device. This will ensure L3 interfaces are only configured on top of
valid netdevs.

Patches 10-13 contain trivial changes required for the introduction of a generic
FID struct - currently only used for vFIDs - in patch 14. Making the FID
struct generic will allow us to easily associate the underlying FIDs with
their L3-counterparts - Router interfaces (RIFs):

    FID Type        | Used by                         | RIF Type
    --------------------------------------------------------
    FID             | The VLAN-aware bridge           | VLAN
    vFID            | VLAN-unaware bridges            | FID
    rFID            | non-bridged netdevs (follow-up) | Sub-port

Obligatory ASCII art to visualize the above:

                   A.B.C.D
                      +
                      | FID RIF
                      +
                     br0                    E.F.G.H
                      +                        +
                      |                        | VLAN RIF
            +---------+---------+              +
            |                   |            br1.W
            | vFID              |              +
            |                   |              |
 vPort    +-+-+               +-+-+            +
swXpY.Z   |   |               |   |           br1
          +-+-+               +-+-+            +
            |                   |     FID=W    |
            |                   | +------------+------------+
            |                   | |                         |
        +---+---+           +---+-+-+                   +---+---+
        |       |           |       |                   |       |
        |       |           |       |                   |       |
        |       |           |       |                   |       |
        +-------+           +-------+                   +-------+
          swXpY

Patches 15-16 further generalize the struct by exploiting the fact that the
FID is a shared resource among ports. Each FID type is assigned a 'leave'
function that is invoked based on CHANGEUPPER events and takes care of the
necessary clean-up.

Patches 17-22 build upon the previous patches and use the FID struct for the
VLAN-aware bridge and take care of cleaning up FID resources in the 'leave'
functions. For now, these are only FDB records, but later on we'll have to
remove the RIFs associated with these FIDs, which will in turn take care of
routes and neighbours clean-up.

The last patch adds debug prints that proved very useful during the
development of this series.

Tested with the existing L2 recipes:
https://github.com/jpirko/lnst/tree/master/recipes/switchdev
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Add debug prints
Ido Schimmel [Mon, 20 Jun 2016 21:04:21 +0000 (23:04 +0200)]
mlxsw: spectrum: Add debug prints

For debug purposes, it's useful to know the order in which the driver
responds to changes in the topology of its upper devices.

Add debug prints to signal these events.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Free resources upon vPort destruction
Ido Schimmel [Mon, 20 Jun 2016 21:04:20 +0000 (23:04 +0200)]
mlxsw: spectrum: Free resources upon vPort destruction

There are situations in which a vPort is destroyed while still holding
references to device's resources such as FIDs and FDB records. This can
happen, for example, when a VLAN device is deleted while still being
bridged.

Instead of trying to make sure vPort destruction is invoked when it no
longer uses device's resources, just free them upon destruction. This
simplifies the code, as we no longer need to take different situations
into account when events are received - cleanup is taken care of in one
place.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Refactor FDB flushing logic
Ido Schimmel [Mon, 20 Jun 2016 21:04:19 +0000 (23:04 +0200)]
mlxsw: spectrum: Refactor FDB flushing logic

FDB entries are learned using {Port / LAG ID, FID} and therefore should
be flushed whenever a port (vPort) leaves its FID (vFID).

However, when the bridge port is a LAG device (or a VLAN device on top),
then FDB flushing is conditional. Ports removed from such LAG
configurations must not trigger flushing, as other ports might still be
members in the LAG and therefore the bridge port is still active.

The decision whether to flush or not was previously computed in the
netdevice notification block, but in order to flush the entries when a
port leaves its FID this decision should be computed there.

Strip the notification block from this logic and instead move it to one
FDB flushing function that is invoked from both the FID / vFID leave
functions.

When port isn't member in LAG, FDB flushing should always occur.
Otherwise, it should occur only when the last port (vPort) member in the
LAG leaves the FID (vFID).

This will allow us - in the next patch - to simplify the cleanup code
paths that are hit whenever the topology above the port netdevs changes.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Don't count on FID being present
Ido Schimmel [Mon, 20 Jun 2016 21:04:18 +0000 (23:04 +0200)]
mlxsw: spectrum: Don't count on FID being present

Not all vPorts will have FIDs assigned to them, so make sure functions
first test for FID presence.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Add FID get / set functions
Ido Schimmel [Mon, 20 Jun 2016 21:04:17 +0000 (23:04 +0200)]
mlxsw: spectrum: Add FID get / set functions

As previously explained, not all vPorts will be assigned FIDs, so instead
of returning the FID index of a vPort, return a pointer to its FID
struct. This will allow us to know whether it's legal to access the
vPort's FID parameters such as index and device.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Check if port is vPort using its VID
Ido Schimmel [Mon, 20 Jun 2016 21:04:16 +0000 (23:04 +0200)]
mlxsw: spectrum: Check if port is vPort using its VID

When L3 interfaces will be introduced a vPort won't necessarily have a
FID assigned to it. This can happen if it's not member in a bridge (in
which case it's assigned a vFID) or doesn't have an IP address (in which
case it's assigned an rFID).

Therefore, instead check the VID parameter to test whether a port is a
vPort or not.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use per-FID struct for the VLAN-aware bridge
Ido Schimmel [Mon, 20 Jun 2016 21:04:15 +0000 (23:04 +0200)]
mlxsw: spectrum: Use per-FID struct for the VLAN-aware bridge

In a very similar way to the vFIDs, make the first 4K FIDs - used in the
VLAN-aware bridge - use the new FID struct.

Upon first use of the FID by any of the ports do the following:

1) Create the FID
2) Setup a matching flooding entry
3) Create a mapping for the FID

Unlike vFIDs, upon creation of a FID we always create a global
VID-to-FID mapping, so that ports without upper vPorts can use it
instead of creating an explicit {Port, VID} to FID mapping.

When a port leaves a FID the reverse is performed. Whenever the FID's
reference count reaches zero the FID is deleted along with the global
mapping.

The per-FID struct will later allow us to configure L3 interfaces on top
of the VLAN-aware bridge.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Remove unused function argument
Ido Schimmel [Mon, 20 Jun 2016 21:04:14 +0000 (23:04 +0200)]
mlxsw: spectrum: Remove unused function argument

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use join / leave functions for vFID operations
Ido Schimmel [Mon, 20 Jun 2016 21:04:13 +0000 (23:04 +0200)]
mlxsw: spectrum: Use join / leave functions for vFID operations

When a vPort is created or when it joins a bridge we always do the same
set of operations:

1) Create the vFID, if not already created
2) Setup flooding for the vFID
3) Map the {Port, VID} to the vFID

When a vPort is destroyed or when it leaves a bridge the reverse is
performed.

Encapsulate the above in join / leave functions and simplify the code.
FIDs and rFIDs will use a similar set of functions.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Make vFID struct generic
Ido Schimmel [Mon, 20 Jun 2016 21:04:12 +0000 (23:04 +0200)]
mlxsw: spectrum: Make vFID struct generic

Up until now we had a dedicated struct only for vFIDs, but before
introducing support for L3 interfaces we need to make it generic and
use it for all three types of FIDs:

1) FIDs - 0..4K-1, used for the VLAN-aware bridge
2) vFIDs - 4K..15K-1, used for VLAN-unaware bridges
3) rFIDs - 15K..16K-1, used to direct traffic to / from the router in
the device. Will be introduced later in the series.

The three types of L3 interfaces - Router InterFaces, RIFs - that will
be introduced correspond to the three types of FIDs and are configured
using them. Therefore, we'll need to store the links between them as
well as a reference count on the underlying FID, so that the
corresponding RIF will be destroyed when it reaches zero.

Note that the lower 0.5K vFIDs are currently used for for non-bridged
netdevs, so that traffic could be flooded to the CPU port. However, when
rFIDs will be introduced we'll no longer need these and they too will be
used for VLAN-unaware bridges.

Make the vFID struct generic by renaming it and some of its fields. FIDs
will be converted to use it later in the series.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use FID instead of vFID to setup flooding
Ido Schimmel [Mon, 20 Jun 2016 21:04:11 +0000 (23:04 +0200)]
mlxsw: spectrum: Use FID instead of vFID to setup flooding

Use a FID index instead of vFID and ease the transition towards a
generic FID struct.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Create a function to map vPort's FID
Ido Schimmel [Mon, 20 Jun 2016 21:04:10 +0000 (23:04 +0200)]
mlxsw: spectrum: Create a function to map vPort's FID

A FID used by a vPort (vFID, but also rFID later in the series) is
always mapped using {Port, VID} and not only VID as with the 4K FIDs of
the VLAN-aware bridge.

Instead of specifying all the arguments each time, just wrap this
operation using a dedicated function and simplify the code.

As before, the function takes FID as its argument in preparation for a
generic FID struct.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use only one function to create vFIDs
Ido Schimmel [Mon, 20 Jun 2016 21:04:09 +0000 (23:04 +0200)]
mlxsw: spectrum: Use only one function to create vFIDs

Simplify the code and use only one function for vFID creation /
destruction.

Unlike before, the function receives a FID index as its argument and not
a vFID index. Instead of passing 0, now one would need to pass 4K, which
is the first vFID.

This is the first step in creating a generic FID struct that will be
used for all three types of FIDs.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Remove redundant function argument
Ido Schimmel [Mon, 20 Jun 2016 21:04:08 +0000 (23:04 +0200)]
mlxsw: spectrum: Remove redundant function argument

In all call sites 'only_uc' is set to false, so strip it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use DECLARE_BITMAP() macro
Ido Schimmel [Mon, 20 Jun 2016 21:04:07 +0000 (23:04 +0200)]
mlxsw: spectrum: Use DECLARE_BITMAP() macro

There is a macro to do this kind of declarations, so use it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Centralize VLAN-aware bridge ref counting
Ido Schimmel [Mon, 20 Jun 2016 21:04:06 +0000 (23:04 +0200)]
mlxsw: spectrum: Centralize VLAN-aware bridge ref counting

We hold a reference count on the number of ports member in the
VLAN-aware bridge, as we only support one.

Instead of always incrementing / decrementing the reference count after
joining / leaving the bridge, simply do this accounting in the join /
leave functions.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Remove unnecessary function argument
Ido Schimmel [Mon, 20 Jun 2016 21:04:05 +0000 (23:04 +0200)]
mlxsw: spectrum: Remove unnecessary function argument

The argument 'br_dev' is never used, so remove it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Make unlinking functions return void
Ido Schimmel [Mon, 20 Jun 2016 21:04:04 +0000 (23:04 +0200)]
mlxsw: spectrum: Make unlinking functions return void

When responding to unlinking CHANGEUPPER notifications we shouldn't
return any value, as it's not checked by upper layers.

In addition, there's nothing the driver can do in case of failure, so it
should simply continue and try to free as much resources as possible and
not stop on first error.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use WARN_ON() return value
Ido Schimmel [Mon, 20 Jun 2016 21:04:03 +0000 (23:04 +0200)]
mlxsw: spectrum: Use WARN_ON() return value

Instead of checking for a condition and then issue the warning, just do
it in one go and simplify the code.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Remove unnecessary checks from event processing
Ido Schimmel [Mon, 20 Jun 2016 21:04:02 +0000 (23:04 +0200)]
mlxsw: spectrum: Remove unnecessary checks from event processing

When upper device of a VLAN device changes we already made sure it's
a bridge device in PRECHANGEUPPER, so no need to check it's a master
device in CHANGEUPPER.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Forbid LAG slave from having VLAN uppers
Ido Schimmel [Mon, 20 Jun 2016 21:04:01 +0000 (23:04 +0200)]
mlxsw: spectrum: Forbid LAG slave from having VLAN uppers

When a port netdev is put under LAG it cannot have VLAN upper devices,
so forbid that. The LAG device itself can have VLAN upper devices.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Sanitize port netdev upper devices
Ido Schimmel [Mon, 20 Jun 2016 21:04:00 +0000 (23:04 +0200)]
mlxsw: spectrum: Sanitize port netdev upper devices

We currently only support the following upper devices for port netdevs:
1) Bridge
2) LAG (bond / team)
3) VLAN

Any other device is forbidden, so return an error.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agomlxsw: spectrum: Use notifier_from_errno() in notifier block
Ido Schimmel [Mon, 20 Jun 2016 21:03:59 +0000 (23:03 +0200)]
mlxsw: spectrum: Use notifier_from_errno() in notifier block

Instead of checking the error value and returning NOTIFY_BAD, just use
notifier_from_errno() and simplify the code.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'hns-next'
David S. Miller [Tue, 21 Jun 2016 08:52:02 +0000 (04:52 -0400)]
Merge branch 'hns-next'

Yisen Zhuang says:

====================
net: hns: fix some bugs in hns driver

This series includes some bugs fixed. All these patches needs to be
applied after the patchset about ACPI support, so this series is
floated to net-next list.

The patches are:
 > from Daode, fixes about pfc pause frame, getting coaslesce, led
control logic, TSO on|off and tcam table configuration.

 > from Jun He, fix the potential leak to port unavailable

 > from Kejian, fix bug of loopback and failing to test ping6

 > from Qianqian, fix the several typo in hns driver

For more details, please see individual patches.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: bug fix about TSO on|off when there is traffic
Daode Huang [Tue, 21 Jun 2016 03:56:39 +0000 (11:56 +0800)]
net: hns: bug fix about TSO on|off when there is traffic

When enable/disable tso, the driver tries to access the hardware register,
but this operation will cause the port unavalible when there is traffic.
This patch tries to enable TSO when initialize, then control tso through
TSE bit in transmit descriptor.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: change the default coalesce usecs
Daode Huang [Tue, 21 Jun 2016 03:56:38 +0000 (11:56 +0800)]
net: hns: change the default coalesce usecs

The default coalesce timeout is 3us, which is will cause CPU
usage is too high. This patch change it to 50us in order to reduce
CPU usage and the value makes sure network latency also meets requirement.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix sbm default parameters config error
Daode Huang [Tue, 21 Jun 2016 03:56:37 +0000 (11:56 +0800)]
net: hns: fix sbm default parameters config error

The default sbm config parameter leaves little buffer when there is heavy
traffic, which will cause packets drop. This patch changes them to make
enough buffers for handling packets.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix bug that alloc skb fail lead to port unavailable
Jun He [Tue, 21 Jun 2016 03:56:36 +0000 (11:56 +0800)]
net: hns: fix bug that alloc skb fail lead to port unavailable

When hns_nic_poll_rx_skb alloc skb fail, it will break receive cycle and
read new fbd_num to start new receive cycle. It recomputes cycle num is
fbd_num minus clean_count, actually this cycle num is too big because
it drop out receive cycle. It brings about the port unavailable.

So we will goto out when alloc skb fail to fix this bug.

Signed-off-by: Jun He <hjat2005@huawei.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: bug fix about led control logic when link down
Daode Huang [Tue, 21 Jun 2016 03:56:35 +0000 (11:56 +0800)]
net: hns: bug fix about led control logic when link down

The default driver sets anchor led bit to 0 when link down,
actually, the anchor bit should be set to 1, so fixes it when
link status is down.
Secondly, change the return value of cpld_set_led_id to 0, which
means leave the cpld to control led blink frequece other than the
driver itself.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: delete redundancy ring enable operations
Daode Huang [Tue, 21 Jun 2016 03:56:34 +0000 (11:56 +0800)]
net: hns: delete redundancy ring enable operations

When network interface is enabled, the ring enable operation is
conducted twice. This patch deletes the redundancy code of ring enable,
and integrates hnae_ae_ops.toggle_queue_status other functions to
hns_ae_start.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: add get_coalesce_range api for hns
Daode Huang [Tue, 21 Jun 2016 03:56:33 +0000 (11:56 +0800)]
net: hns: add get_coalesce_range api for hns

This patch adds get_coalesce_range api for hns, it shows
range of coalesce usecs and frames that can be set on
this interface.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: hns: fix bug of getting the wrong tcam data
Daode Huang [Tue, 21 Jun 2016 03:56:32 +0000 (11:56 +0800)]
net: hns: fix bug of getting the wrong tcam data

The current driver stores the high bit value of tcam data register
to the tcam data low element, stores the low bit value of tcam data
register to tcam data high element, this patch fixes this bug.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Signed-off-by: Yisen Zhuang <Yisen.Zhuang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>