openwrt/staging/blogic.git
7 years agosctp: add SCTP_PR_STREAM_STATUS sockopt for prsctp
Xin Long [Sat, 1 Apr 2017 09:07:46 +0000 (17:07 +0800)]
sctp: add SCTP_PR_STREAM_STATUS sockopt for prsctp

Before when implementing sctp prsctp, SCTP_PR_STREAM_STATUS wasn't
added, as it needs to save abandoned_(un)sent for every stream.

After sctp stream reconf is added in sctp, assoc has structure
sctp_stream_out to save per stream info.

This patch is to add SCTP_PR_STREAM_STATUS by putting the prsctp
per stream statistics into sctp_stream_out.

v1->v2:
  fix an indent issue.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'hns-misc-fixes'
David S. Miller [Mon, 3 Apr 2017 21:48:45 +0000 (14:48 -0700)]
Merge branch 'hns-misc-fixes'

Salil Mehta says:

====================
net: hns: Misc. HNS Bug Fixes & Code Improvements

This patch set introduces various HNS bug fixes, optimizations and code
improvements.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Some checkpatch.pl script & warning fixes
Salil [Sat, 1 Apr 2017 11:03:48 +0000 (12:03 +0100)]
net: hns: Some checkpatch.pl script & warning fixes

This patch fixes some checkpatch.pl script caught errors and
warnings during the compilation time.

Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Avoid Hip06 chip TX packet line bug
lipeng [Sat, 1 Apr 2017 11:03:47 +0000 (12:03 +0100)]
net: hns: Avoid Hip06 chip TX packet line bug

There is a bug on Hip06 that tx ring interrupts packets count will be
clear when drivers send data to tx ring, so that the tx packets count
will never upgrade to packets line, and cause the interrupts engendered
was delayed.
Sometimes, it will cause sending performance lower than expected.

To fix this bug, we set tx ring interrupts packets line to 1 forever,
to avoid count clear. And set the gap time to 20us, to solve the problem
that too many interrupts engendered when packets line is 1.

This patch could advance the send performance on ARM  from 6.6G to 9.37G
when an iperf send thread on ARM and an iperf send thread on X86 for XGE.

Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Adjust the SBM module buffer threshold
Kejian Yan [Sat, 1 Apr 2017 11:03:46 +0000 (12:03 +0100)]
net: hns: Adjust the SBM module buffer threshold

HNS needs SMB Buffers to store at least two packets after sending
pause frame because of the link delay. The MTU of HNS is 9728. As
the processor user manual described, the SBM buffer threshold should
be modified.

Reported-by: Ping Zhang <zhangping5@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Simplify the exception sequence in hns_ppe_init()
Kejian Yan [Sat, 1 Apr 2017 11:03:45 +0000 (12:03 +0100)]
net: hns: Simplify the exception sequence in hns_ppe_init()

We need to free all ppe submodule if it fails to initialize ppe by
any fault, so this patch will free all ppe resource before
hns_ppe_init() returns exception situation

Reported-by: JinchuanTian <tianjinchuan1@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Optimise the code in hns_mdio_wait_ready()
Kejian Yan [Sat, 1 Apr 2017 11:03:44 +0000 (12:03 +0100)]
net: hns: Optimise the code in hns_mdio_wait_ready()

This patch fixes the code to clear pclint warning/info.

Reported-by: Ping Zhang <zhangping5@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Clean redundant code from hns_mdio.c file
Kejian Yan [Sat, 1 Apr 2017 11:03:43 +0000 (12:03 +0100)]
net: hns: Clean redundant code from hns_mdio.c file

This patch cleans the redundant code from  hns_mdio.c.

Reported-by: Ping Zhang <zhangping5@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Remove redundant mac table operations
Kejian Yan [Sat, 1 Apr 2017 11:03:42 +0000 (12:03 +0100)]
net: hns: Remove redundant mac table operations

This patch removes redundant functions used only for debugging
purposes.

Reported-by: Weiwei Deng <dengweiwei@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Remove redundant mac_get_id()
Kejian Yan [Sat, 1 Apr 2017 11:03:41 +0000 (12:03 +0100)]
net: hns: Remove redundant mac_get_id()

There is a mac_id in mac control block structure, so the callback
function mac_get_id() is useless. Here we remove this function.

Reported-by: Weiwei Deng <dengweiwei@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Remove the redundant adding and deleting mac function
Kejian Yan [Sat, 1 Apr 2017 11:03:40 +0000 (12:03 +0100)]
net: hns: Remove the redundant adding and deleting mac function

The functions (hns_dsaf_set_mac_mc_entry() and hns_mac_del_mac()) are
not called by any functions. They are dead code in hns. And the same
features are implemented by the patch (the id is 66355f5).

Reported-by: Weiwei Deng <dengweiwei@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Correct HNS RSS key set function
lipeng [Sat, 1 Apr 2017 11:03:39 +0000 (12:03 +0100)]
net: hns: Correct HNS RSS key set function

This patch fixes below ethtool configuration error:

localhost:~ # ethtool -X eth0 hkey XX:XX:XX...
Cannot set Rx flow hash configuration: Operation not supported

Signed-off-by: lipeng <lipeng321@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Replace netif_tx_lock to ring spin lock
lipeng [Sat, 1 Apr 2017 11:03:38 +0000 (12:03 +0100)]
net: hns: Replace netif_tx_lock to ring spin lock

netif_tx_lock is a global spin lock, it will take affect
in all rings in the netdevice. In tx_poll_one process, it can
only lock the current ring, in this case, we define a spin lock
in hnae_ring struct for it.

Signed-off-by: lipeng <lipeng321@huawei.com>
reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Fix to adjust buf_size of ring according to mtu
lipeng [Sat, 1 Apr 2017 11:03:37 +0000 (12:03 +0100)]
net: hns: Fix to adjust buf_size of ring according to mtu

Because buf_size of ring set to 2048, the process of rx_poll_one
can reuse the page, therefore the performance of XGE can improve.
But the chip only supports three bds in one package, so the max mtu
is 6K when it sets to 2048. For better performane in litter mtu, we
need change buf_size according to mtu.

When user change mtu, hns is only change the desc in memory. There
are some desc has been fetched by the chip, these desc can not be
changed by the code. So it needs set the port loopback and send
some packages to let the chip consumes the wrong desc and fetch new
desc.
Because the Pv660 do not support rss indirection, we need add version
check in mtu change process.

Signed-off-by: lipeng <lipeng321@huawei.com>
reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Optimize hns_nic_common_poll for better performance
lipeng [Sat, 1 Apr 2017 11:03:36 +0000 (12:03 +0100)]
net: hns: Optimize hns_nic_common_poll for better performance

After polling less than buget packages, we need check again. If
there are still some packages, we call napi_schedule add softirq
queue, this is not better way. So we return buget value instead
of napi_schedule.

Signed-off-by: lipeng <lipeng321@huawei.com>
reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: bug fix of ethtool show the speed
Daode Huang [Sat, 1 Apr 2017 11:03:35 +0000 (12:03 +0100)]
net: hns: bug fix of ethtool show the speed

When run ethtool ethX on hns driver, the speed will show
as "Unknown". The base.speed is not correct assigned,
this patch fix this bug.

Signed-off-by: Daode Huang <huangdaode@hisilicon.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Remove redundant memset during buffer release
lipeng [Sat, 1 Apr 2017 11:03:34 +0000 (12:03 +0100)]
net: hns: Remove redundant memset during buffer release

Because all members of desc_cb is assigned when xmit one package, so it
can delete in hnae_free_buffer, as follows:
        - "dma, priv, length, type" are assigned in fill_v2_desc.
        - "page_offset, reuse_flag, buf" are not used in tx direction.

Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: Weiwei Deng <dengweiwei@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Optimize the code for GMAC pad and crc Config
lipeng [Sat, 1 Apr 2017 11:03:33 +0000 (12:03 +0100)]
net: hns: Optimize the code for GMAC pad and crc Config

This patch optimises the init configuration code leg
for gmac pad and crc set interface.

Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: JinchuanTian <tianjinchuan1@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Modify GMAC init TX threshold value
lipeng [Sat, 1 Apr 2017 11:03:32 +0000 (12:03 +0100)]
net: hns: Modify GMAC init TX threshold value

This patch reduces GMAC TX threshold value to avoid gmac
hang-up with speed 100M/duplex half.

Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: JinchuanTian <tianjinchuan1@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: hns: Fix the implementation of irq affinity function
lipeng [Sat, 1 Apr 2017 11:03:31 +0000 (12:03 +0100)]
net: hns: Fix the implementation of irq affinity function

This patch fixes the implementation of the IRQ affinity
function. This function is used to create the cpu mask
which eventually is used to initialize the cpu<->queue
association for XPS(Transmit Packet Steering).

Signed-off-by: lipeng <lipeng321@huawei.com>
Signed-off-by: Kejian Yan <yankejian@huawei.com>
Reviewed-by: Yisen Zhuang <yisen.zhuang@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'rds-minor-bug-fixes'
David S. Miller [Mon, 3 Apr 2017 02:41:01 +0000 (19:41 -0700)]
Merge branch 'rds-minor-bug-fixes'

Sowmini Varadhan says:

====================
rds: tcp: couple of minor bug fixes

A couple of minor bugfixes that showed up during testing
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agords: tcp: canonical connection order for all paths with index > 0
Sowmini Varadhan [Fri, 31 Mar 2017 22:56:31 +0000 (15:56 -0700)]
rds: tcp: canonical connection order for all paths with index > 0

The rds_connect_worker() has a bug in the check that enforces the
canonical connection order described in the comments of
rds_tcp_state_change(). The intention is to make sure that all
the multipath connections are always initiated by the smaller IP
address via rds_start_mprds. To achieve this, rds_connection_worker
should check that cp_index > 0.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agords: tcp: allow progress of rds_conn_shutdown if the rds_connection is marked ERROR...
Sowmini Varadhan [Fri, 31 Mar 2017 22:56:30 +0000 (15:56 -0700)]
rds: tcp: allow progress of rds_conn_shutdown if the rds_connection is marked ERROR by an intervening FIN

rds_conn_shutdown() runs in workq context, and marks the rds_connection
as DISCONNECTING before quiescing Tx/Rx paths. However, after all I/O
has quiesced, we may still find the rds_connection state to be
RDS_CONN_ERROR if an intervening FIN was processed in softirq context.

This is not a fatal error: rds_conn_shutdown() should continue the
shutdown, and there is no need to log noisy messages about this event.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosock: correctly test SOCK_TIMESTAMP in sock_recv_ts_and_drops()
Eric Dumazet [Fri, 31 Mar 2017 21:59:25 +0000 (14:59 -0700)]
sock: correctly test SOCK_TIMESTAMP in sock_recv_ts_and_drops()

It seems the code does not match the intent.

This broke packetdrill, and probably other programs.

Fixes: 6c7c98bad488 ("sock: avoid dirtying sk_stamp, if possible")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c: fix build with gcc-4.4.4
Andrew Morton [Fri, 31 Mar 2017 20:09:41 +0000 (13:09 -0700)]
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c: fix build with gcc-4.4.4

drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c: In function 'mlx5e_set_rxfh':
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1067: error: unknown field 'rss' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1067: warning: missing braces around initializer
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1067: warning: (near initialization for 'rrp.<anonymous>')
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1068: error: unknown field 'rss' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1069: warning: excess elements in struct initializer
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c:1069: warning: (near initialization for 'rrp')

gcc-4.4.4 has issues with anonymous union initializers.  Work around this.

Cc: Saeed Mahameed <saeedm@mellanox.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers/net/ethernet/mellanox/mlx5/core/en_main.c: fix build with gcc-4.4.4
Andrew Morton [Fri, 31 Mar 2017 20:09:38 +0000 (13:09 -0700)]
drivers/net/ethernet/mellanox/mlx5/core/en_main.c: fix build with gcc-4.4.4

drivers/net/ethernet/mellanox/mlx5/core/en_main.c: In function 'mlx5e_redirect_rqts':
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2210: error: unknown field 'rqn' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2211: warning: missing braces around initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2211: warning: (near initialization for 'direct_rrp.<anonymous>')
drivers/net/ethernet/mellanox/mlx5/core/en_main.c: In function 'mlx5e_redirect_rqts_to_channels':
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2227: error: unknown field 'rss' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2227: warning: missing braces around initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2227: warning: (near initialization for 'rrp.<anonymous>')
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2227: warning: initialization makes integer from pointer without a cast
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2228: error: unknown field 'rss' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2229: warning: excess elements in struct initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2229: warning: (near initialization for 'rrp')
drivers/net/ethernet/mellanox/mlx5/core/en_main.c: In function 'mlx5e_redirect_rqts_to_drop':
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2238: error: unknown field 'rqn' specified in initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2239: warning: missing braces around initializer
drivers/net/ethernet/mellanox/mlx5/core/en_main.c:2239: warning: (near initialization for 'drop_rrp.<anonymous>')

gcc-4.4.4 has issues with anonymous union initializers.  Work around this.

Cc: Saeed Mahameed <saeedm@mellanox.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: fix cbs configuration
Joao Pinto [Fri, 31 Mar 2017 13:22:02 +0000 (14:22 +0100)]
net: stmmac: fix cbs configuration

Sending again, because forgot to include net-dev.

The QoS IP does not accept AVB capabilities to default/queue 0, this way we
guarantee 75% bandwidth for AVB. This patch assures that only queues >= 1
gets CBS confgured. Additional info was also added to stmmac.txt.

Reported-by: Niklas Cassel <niklas.cassel@axis.com>
Signed-off-by: Joao Pinto <jpinto@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mpls-more-labels'
David S. Miller [Sun, 2 Apr 2017 03:21:45 +0000 (20:21 -0700)]
Merge branch 'mpls-more-labels'

David Ahern says:

====================
net: mpls: Allow users to configure more labels per route

Increase the maximum number of new labels for MPLS routes from 2 to 30.

To keep memory consumption in check, the labels array is moved to the end
of mpls_nh and mpls_iptunnel_encap structs as a 0-sized array. Allocations
use the maximum number of labels across all nexthops in a route for LSR
and the number of labels configured for LWT.

The mpls_route layout is changed to:

   +----------------------+
   | mpls_route           |
   +----------------------+
   | mpls_nh 0            |
   +----------------------+
   | alignment padding    |   4 bytes for odd number of labels; 0 for even
   +----------------------+
   | via[rt_max_alen] 0   |
   +----------------------+
   | alignment padding    |   via's aligned on sizeof(unsigned long)
   +----------------------+
   | ...                  |

Meaning the via follows its mpls_nh providing better locality as the
number of labels increases. UDP_RR tests with namespaces shows no impact
to a modest performance increase with this layout for 1 or 2 labels and
1 or 2 nexthops.

mpls_route allocation size is limited to 4096 bytes allowing on the
order of 30 nexthops with 30 labels (or more nexthops with fewer
labels). LWT encap shares same maximum number of labels as mpls routing.

v3
- initialize n_labels to 0 in case RTA_NEWDST is not defined; detected
  by the kbuild test robot

v2
- updates per Eric's comments
  + added patch to ensure all reads of rt_nhn_alive and nh_flags in
    the packet path use READ_ONCE and all writes via event handlers
    use WRITE_ONCE

  + limit mpls_route size to 4096 (PAGE_SIZE for most arch)

  + mostly killed use of MAX_NEW_LABELS; it exists only for common
    limit between lwt and routing paths
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: Increase max number of labels for lwt encap
David Ahern [Fri, 31 Mar 2017 14:14:04 +0000 (07:14 -0700)]
net: mpls: Increase max number of labels for lwt encap

Alow users to push down more labels per MPLS encap. Similar to LSR case,
move label array to the end of mpls_iptunnel_encap and allocate based on
the number of labels for the route.

For consistency with the LSR case, re-use the same maximum number of
labels.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: bump maximum number of labels
David Ahern [Fri, 31 Mar 2017 14:14:03 +0000 (07:14 -0700)]
net: mpls: bump maximum number of labels

Allow users to push down more labels per MPLS route. With the previous
patches, no memory allocations are based on MAX_NEW_LABELS; the limit
is only used to keep userspace in check.

At this point MAX_NEW_LABELS is only used for mpls_route_config (copying
route data from userspace) and processing nexthops looking for the max
number of labels across the route spec.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: Limit memory allocation for mpls_route
David Ahern [Fri, 31 Mar 2017 14:14:02 +0000 (07:14 -0700)]
net: mpls: Limit memory allocation for mpls_route

Limit memory allocation size for mpls_route to 4096.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: change mpls_route layout
David Ahern [Fri, 31 Mar 2017 14:14:01 +0000 (07:14 -0700)]
net: mpls: change mpls_route layout

Move labels to the end of mpls_nh as a 0-sized array and within mpls_route
move the via for a nexthop after the mpls_nh. The new layout becomes:

   +----------------------+
   | mpls_route           |
   +----------------------+
   | mpls_nh 0            |
   +----------------------+
   | alignment padding    |   4 bytes for odd number of labels; 0 for even
   +----------------------+
   | via[rt_max_alen] 0   |
   +----------------------+
   | alignment padding    |   via's aligned on sizeof(unsigned long)
   +----------------------+
   | ...                  |
   +----------------------+
   | mpls_nh n-1          |
   +----------------------+
   | via[rt_max_alen] n-1 |
   +----------------------+

Memory allocated for nexthop + via is constant across all nexthops and
their via. It is based on the maximum number of labels across all nexthops
and the maximum via length. The size is saved in the mpls_route as
rt_nh_size. Accessing a nexthop becomes rt->rt_nh + index * rt->rt_nh_size.

The offset of the via address from a nexthop is saved as rt_via_offset
so that given an mpls_nh pointer the via for that hop is simply
nh + rt->rt_via_offset.

With prior code, memory allocated per mpls_route with 1 nexthop:
     via is an ethernet address - 64 bytes
     via is an ipv4 address     - 64
     via is an ipv6 address     - 72

With this patch set, memory allocated per mpls_route with 1 nexthop and
1 or 2 labels:
     via is an ethernet address - 56 bytes
     via is an ipv4 address     - 56
     via is an ipv6 address     - 64

The 8-byte reduction is due to the previous patch; the change introduced
by this patch has no impact on the size of allocations for 1 or 2 labels.

Performance impact of this change was examined using network namespaces
with veth pairs connecting namespaces. ns0 inserts the packet to the
label-switched path using an lwt route with encap mpls. ns1 adds 1 or 2
labels depending on test, ns2 (and ns3 for 2-label test) pops the label
and forwards. ns3 (or ns4) for a 2-label is the destination. Similar
series of namespaces used for 2-nexthop test.

Intent is to measure changes to latency (overhead in manipulating the
packet) in the forwarding path. Tests used netperf with UDP_RR.

IPv4:                     current   patches
   1 label, 1 nexthop      29908     30115
   2 label, 1 nexthop      29071     29612
   1 label, 2 nexthop      29582     29776
   2 label, 2 nexthop      29086     29149

IPv6:                     current   patches
   1 label, 1 nexthop      24502     24960
   2 label, 1 nexthop      24041     24407
   1 label, 2 nexthop      23795     23899
   2 label, 2 nexthop      23074     22959

In short, the change has no effect to a modest increase in performance.
This is expected since this patch does not really have an impact on routes
with 1 or 2 labels (the current limit) and 1 or 2 nexthops.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: Convert number of nexthops to u8
David Ahern [Fri, 31 Mar 2017 14:14:00 +0000 (07:14 -0700)]
net: mpls: Convert number of nexthops to u8

Number of nexthops and number of alive nexthops are tracked using an
unsigned int. A route should never have more than 255 nexthops so
convert both to u8. Update all references and intermediate variables
to consistently use u8 as well.

Shrinks the size of mpls_route from 32 bytes to 24 bytes with a 2-byte
hole before the nexthops.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: rt_nhn_alive and nh_flags should be accessed using READ_ONCE
David Ahern [Fri, 31 Mar 2017 14:13:59 +0000 (07:13 -0700)]
net: mpls: rt_nhn_alive and nh_flags should be accessed using READ_ONCE

The number of alive nexthops for a route (rt->rt_nhn_alive) and the
flags for a next hop (nh->nh_flags) are modified by netdev event
handlers. The event handlers run with rtnl_lock held so updates are
always done with the lock held. The packet path accesses the fields
under the rcu lock. Since those fields can change at any moment in
the packet path, both fields should be accessed using READ_ONCE. Updates
to both fields should use WRITE_ONCE.

Update mpls_select_multipath (packet path) and mpls_ifdown and mpls_ifup
(event handlers) accordingly.

Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoudp: use sk_protocol instead of pcflag to detect udplite sockets
Paolo Abeni [Fri, 31 Mar 2017 09:47:39 +0000 (11:47 +0200)]
udp: use sk_protocol instead of pcflag to detect udplite sockets

In the udp_sock struct, the 'forward_deficit' and 'pcflag' fields
share the same cacheline. While the first is dirtied by
udp_recvmsg, the latter is read, possibly several times, by the
bottom half processing to discriminate between udp and udplite
sockets.

With this patch, sk->sk_protocol is used to check is the socket is
really an udplite one, avoiding some cache misses per
packet and improving the performance under udp_flood with
small packet up to 10%.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: fix build error with devlink build as module
Tobias Regnery [Fri, 31 Mar 2017 09:44:52 +0000 (11:44 +0200)]
net: dsa: fix build error with devlink build as module

After commit 96567d5dacf4 ("net: dsa: dsa2: Add basic support of devlink")
I see the following link error with CONFIG_NET_DSA=y and CONFIG_NET_DEVLINK=m:

net/built-in.o: In function 'dsa_register_switch':
(.text+0xe226b): undefined reference to `devlink_alloc'
net/built-in.o: In function 'dsa_register_switch':
(.text+0xe2284): undefined reference to `devlink_register'
net/built-in.o: In function 'dsa_register_switch':
(.text+0xe243e): undefined reference to `devlink_port_register'
net/built-in.o: In function 'dsa_register_switch':
(.text+0xe24e1): undefined reference to `devlink_port_register'
net/built-in.o: In function 'dsa_register_switch':
(.text+0xe24fa): undefined reference to `devlink_port_type_eth_set'
net/built-in.o: In function 'dsa_dst_unapply.part.8':
dsa2.c:(.text.unlikely+0x345): undefined reference to 'devlink_port_unregister'
dsa2.c:(.text.unlikely+0x36c): undefined reference to 'devlink_port_unregister'
dsa2.c:(.text.unlikely+0x38e): undefined reference to 'devlink_port_unregister'
dsa2.c:(.text.unlikely+0x3f2): undefined reference to 'devlink_unregister'
dsa2.c:(.text.unlikely+0x3fb): undefined reference to 'devlink_free'

Fix this by adding a dependency on MAY_USE_DEVLINK so that CONFIG_NET_DSA
get switched to be build as module when CONFIG_NET_DEVLINK=m.

Fixes: 96567d5dacf4 ("net: dsa: dsa2: Add basic support of devlink")
Signed-off-by: Tobias Regnery <tobias.regnery@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'phylib-EEE-updates'
David S. Miller [Sun, 2 Apr 2017 03:04:04 +0000 (20:04 -0700)]
Merge branch 'phylib-EEE-updates'

Russell King says:

====================
phylib EEE updates

This series of patches depends on the previous set of changes, and is
therefore net-next material.

While testing the EEE code, I discovered a number of issues:

1. It is possible to enable advertisment of EEE modes which are not
   supported by the hardware.  We omit to check the supported modes
   and mask off those modes that are not supported before writing the
   EEE advertisment register.

2. We need to restart autonegotiation after a change of the EEE
   advertisment, otherwise the link partner does not see the updated
   EEE modes.

3. SGMII connected PHYs are also capable of supporting EEE.

Through discussion with Florian, it has been decided to remove the check
for the PHY interface mode in patch (3).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: allow EEE with any interface mode
Russell King [Fri, 31 Mar 2017 09:37:18 +0000 (10:37 +0100)]
net: phy: allow EEE with any interface mode

EEE is able to work in any PHY interface mode, there is nothing which
fundamentally restricts it to only a few modes.  For example, EEE works
in SGMII mode with the Marvell 88E1512.

Rather than just adding SGMII mode to the list, Florian suggests
removing the list of interface modes entirely:

  It actually sounds like we should just kill the check entirely,
  it does not appear that any of the interface mode would not
  fundamentally be able to support EEE, because the "lowest" mode
  we support is MII, and even there it's quite possible to support
  EEE.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: restart phy autonegotiation after EEE advertisment change
Russell King [Fri, 31 Mar 2017 09:37:12 +0000 (10:37 +0100)]
net: phy: restart phy autonegotiation after EEE advertisment change

When the EEE advertisment is changed, we should restart autonegotiation
to update the link partner with the new EEE settings.  Add this trigger
but only if the advertisment has changed.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: avoid setting unsupported EEE advertisments
Russell King [Fri, 31 Mar 2017 09:37:07 +0000 (10:37 +0100)]
net: phy: avoid setting unsupported EEE advertisments

We currently allow userspace to set any EEE advertisments it desires,
whether or not the PHY supports them.  For example:

 # ethtool --set-eee eth1 advertise 0xffffffff
 # ethtool --show-eee eth1
 EEE Settings for eth1:
        EEE status: disabled
        Tx LPI: disabled
        Supported EEE link modes:  100baseT/Full
                                   1000baseT/Full
                                   10000baseT/Full
        Advertised EEE link modes:  100baseT/Full
                                    1000baseT/Full
                                    1000baseKX/Full
                                    10000baseT/Full
                                    10000baseKX4/Full
                                    10000baseKR/Full

Clearly, this is not sane, we should only allow link modes that are
supported to be advertised (as we do elsewhere.)  Ensure that we mask
the MDIO_AN_EEE_ADV value with the capabilities retrieved from the
MDIO_PCS_EEE_ABLE register.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-prog-testing-framework'
David S. Miller [Sat, 1 Apr 2017 19:45:58 +0000 (12:45 -0700)]
Merge branch 'bpf-prog-testing-framework'

Alexei Starovoitov says:

====================
bpf: program testing framework

Development and testing of networking bpf programs is quite cumbersome.
Especially tricky are XDP programs that attach to real netdevices and
program development feels like working on the car engine while
the car is in motion.
Another problem is ongoing changes to upstream llvm core
that can introduce an optimization that verifier will not
recognize. llvm bpf backend tests have no ability to run the programs.
To improve this situation introduce BPF_PROG_TEST_RUN command
to test and performance benchmark bpf programs.
It achieves several goals:
- development of xdp and skb based bpf programs can be done
in a canned environment with unit tests
- program performance optimizations can be benchmarked outside of
networking core (without driver and skb costs)
- continuous testing of upstream changes is finally practical

Patches 4,5,6 add C based test cases of various complexity
to cover some sched_cls and xdp features. More tests will
be added in the future. The tests were run on centos7 only.

For now the framework supports only skb and xdp programs. In the future
it can be extended to socket_filter and tracing program types.

More details are in individual patches.

v1->v2:
- rename bpf_program_test_run->bpf_prog_test_run
- add missing #include <linux/bpf.h> since libbpf.h shouldn't depend
on prior includes
- reordered patches 3 and 4 to keep bisect clean
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests/bpf: add l4 load balancer test based on sched_cls
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:43 +0000 (21:45 -0700)]
selftests/bpf: add l4 load balancer test based on sched_cls

this l4lb demo is a comprehensive test case for LLVM codegen and
kernel verifier. It's using fully inlined jhash(), complex packet
parsing and multiple map lookups of different types to stress
llvm and verifier.
The map sizes, map population and test vectors are artificial to
exercise different paths through the bpf program.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests/bpf: add a test for basic XDP functionality
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:42 +0000 (21:45 -0700)]
selftests/bpf: add a test for basic XDP functionality

add C test for xdp_adjust_head(), packet rewrite and map lookups

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoselftests/bpf: add a test for overlapping packet range checks
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:41 +0000 (21:45 -0700)]
selftests/bpf: add a test for overlapping packet range checks

add simple C test case for llvm and verifier range check fix from
commit b1977682a385 ("bpf: improve verifier packet range checks")

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotools/lib/bpf: expose bpf_program__set_type()
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:40 +0000 (21:45 -0700)]
tools/lib/bpf: expose bpf_program__set_type()

expose bpf_program__set_type() to set program type

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotools/lib/bpf: add support for BPF_PROG_TEST_RUN command
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:39 +0000 (21:45 -0700)]
tools/lib/bpf: add support for BPF_PROG_TEST_RUN command

add support for BPF_PROG_TEST_RUN command to libbpf.a

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: introduce BPF_PROG_TEST_RUN command
Alexei Starovoitov [Fri, 31 Mar 2017 04:45:38 +0000 (21:45 -0700)]
bpf: introduce BPF_PROG_TEST_RUN command

development and testing of networking bpf programs is quite cumbersome.
Despite availability of user space bpf interpreters the kernel is
the ultimate authority and execution environment.
Current test frameworks for TC include creation of netns, veth,
qdiscs and use of various packet generators just to test functionality
of a bpf program. XDP testing is even more complicated, since
qemu needs to be started with gro/gso disabled and precise queue
configuration, transferring of xdp program from host into guest,
attaching to virtio/eth0 and generating traffic from the host
while capturing the results from the guest.

Moreover analyzing performance bottlenecks in XDP program is
impossible in virtio environment, since cost of running the program
is tiny comparing to the overhead of virtio packet processing,
so performance testing can only be done on physical nic
with another server generating traffic.

Furthermore ongoing changes to user space control plane of production
applications cannot be run on the test servers leaving bpf programs
stubbed out for testing.

Last but not least, the upstream llvm changes are validated by the bpf
backend testsuite which has no ability to test the code generated.

To improve this situation introduce BPF_PROG_TEST_RUN command
to test and performance benchmark bpf programs.

Joint work with Daniel Borkmann.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: Mock-up driver
Florian Fainelli [Fri, 31 Mar 2017 01:43:21 +0000 (18:43 -0700)]
net: dsa: Mock-up driver

This patch adds support for a DSA mock-up driver which essentially does
the following:

- registers/unregisters 4 fixed PHYs to the slave network devices
- uses eth0 (configurable) as the master netdev
- registers the switch as a fixed MDIO device against the fixed MDIO bus
  at address 31
- includes dynamic debug prints for dsa_switch_ops functions that can be
  enabled to get call traces

This is a good way to test modular builds as well as exercise the DSA
APIs without requiring access to real hardware. This does not test the
data-path, although this could be added later on.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mv88e6xxx-cross-chip-bridging'
David S. Miller [Sat, 1 Apr 2017 19:22:59 +0000 (12:22 -0700)]
Merge branch 'mv88e6xxx-cross-chip-bridging'

Vivien Didelot says:

====================
net: dsa: mv88e6xxx: program cross-chip bridging

The purpose of this patch series is to bring hardware cross-chip
bridging configuration to the DSA layer and the mv88e6xxx DSA driver.

Most recent Marvell switch chips have a Cross-chip Port Based VLAN Table
(PVT) used to restrict to which internal destination port an arbitrary
external source port is allowed to egress frames to.

The current behavior of the mv88e6xxx driver is to program this table
table with all ones, allowing any external ports to egress frames on any
internal ports. This means that carefully crafted Ethernet frames can
potentially bypass the user bridging configuration.

Patches 1 to 7 prepare the setup of this table and factorize the common
bits of both in-chip and cross-chip Marvell bridging code.

Patch 8 adds new optional cross-chip bridging operations to DSA switch.

Patch 9 switches the current behavior to program the table according to
the user bridging configuration when (cross-chip) ports get (un)bridged.

On a ZII Rev B board, bridging together the 3 user ports of both 88E6352
will result in the following PVTs on respectively switch 0 and switch 1:

    External   Internal Ports
    Dev Port   0  1  2  3  4  5  6

     1    0    *  *  *  -  -  *  *
     1    1    *  *  *  -  -  *  *
     1    2    *  *  *  -  -  *  *
     1    3    -  -  -  -  -  *  *
     1    4    -  -  -  -  -  *  *
     1    5    *  *  *  *  *  *  *
     1    6    *  *  *  *  *  *  *

     0    0    *  *  *  -  -  *  *
     0    1    *  *  *  -  -  *  *
     0    2    *  *  *  -  -  *  *
     0    3    -  -  -  -  -  *  *
     0    4    -  -  -  -  -  *  *
     0    5    *  *  *  *  *  *  *
     0    6    *  *  *  *  *  *  *

Changes since v2:
  - Define MV88E6XXX_MAX_PVT_SWITCHES and MV88E6XXX_MAX_PVT_PORTS
  - use mv88e6xxx_g2_misc_4_bit_port instead of the 5-bit variant
  - add Andrew's tags and reword commit 6/9
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: add cross-chip bridging
Vivien Didelot [Thu, 30 Mar 2017 21:37:15 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: add cross-chip bridging

Implement the DSA cross-chip bridging operations by remapping the local
ports an external source port can egress frames to, when this cross-chip
port joins or leaves a bridge.

The PVT is no longer configured with all ones allowing any external
frame to egress any local port. Only DSA and CPU ports, as well as
bridge group members, can egress frames on local ports.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: add cross-chip bridging operations
Vivien Didelot [Thu, 30 Mar 2017 21:37:14 +0000 (17:37 -0400)]
net: dsa: add cross-chip bridging operations

Introduce crosschip_bridge_{join,leave} operations in the dsa_switch_ops
structure, which can be used by switches supporting interconnection.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: remap existing bridge members
Vivien Didelot [Thu, 30 Mar 2017 21:37:13 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: remap existing bridge members

When a local port of a switch chip becomes a member of a bridge group,
we need to reprogram the Cross-chip Port Based VLAN Table (PVT) to allow
existing cross-chip bridge members to egress frames on the new ports.

There is no functional changes yet, since the PVT is still programmed
with all ones, allowing any external port to egress frames locally.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: factorize in-chip bridge map
Vivien Didelot [Thu, 30 Mar 2017 21:37:12 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: factorize in-chip bridge map

Factorize the code in the DSA port_bridge_{join,leave} routines used to
program the port VLAN map of all local ports of a given bridge group.

At the same time shorten the _mv88e6xxx_port_based_vlan_map to get rid
of the old underscore prefix naming convention.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: rework in-chip bridging
Vivien Didelot [Thu, 30 Mar 2017 21:37:11 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: rework in-chip bridging

All ports -- internal and external, for chips featuring a PVT -- have a
mask restricting to which internal ports a frame is allowed to egress.

Now that DSA exposes the number of ports and their bridge devices, it is
possible to extract the code generating the VLAN map and make it generic
so that it can be shared later with the cross-chip bridging code.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: allocate the number of ports
Vivien Didelot [Thu, 30 Mar 2017 21:37:10 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: allocate the number of ports

The current code allocates DSA_MAX_PORTS ports for a Marvell dsa_switch
structure. Provide the exact number of ports so the corresponding
ds->num_ports is accurate.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: program the PVT with all ones
Vivien Didelot [Thu, 30 Mar 2017 21:37:09 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: program the PVT with all ones

The Cross-chip Port Based VLAN Table (PVT) is currently initialized with
all ones, allowing any external ports to egress frames on local ports.

This commit implements the PVT access functions and programs the PVT
with all ones for the local switch ports only, instead of using the Init
operation. The current behavior is unchanged for the moment.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: use 4-bit port for PVT data
Vivien Didelot [Thu, 30 Mar 2017 21:37:08 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: use 4-bit port for PVT data

The Cross-chip Port Based VLAN Table (PVT) supports two indexing modes,
one using 5-bit for device and 4-bit for port, the other using 4-bit for
device and 5-bit for port, configured via the Global 2 Misc register.

Only 4 bits for the source port are needed when interconnecting 88E6xxx
switch devices since they all support less than 16 physical ports. The
full 5 bits are needed when interconnecting a device with 98DXxxx switch
devices since they support more than 16 physical ports.

Add a mv88e6xxx_pvt_setup helper to set the 4-bit port PVT mode, which
will be extended later to also initialize the PVT content.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: move PVT description in info
Vivien Didelot [Thu, 30 Mar 2017 21:37:07 +0000 (17:37 -0400)]
net: dsa: mv88e6xxx: move PVT description in info

Not all Marvell switch chips feature a Cross-chip Port VLAN Table (PVT).

Chips with a PVT use the same implementation, so a new mv88e6xxx_ops
member won't be necessary yet. Add a "pvt" boolean member to the
mv88e6xxx_info structure and kill the obsolete MV88E6XXX_FLAGS_PVT flag.

Add a mv88e6xxx_has_pvt helper to wrap future checks of that condition.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodpaa_eth: use AVOIDBLOCK for Tx confirmation queues
Madalin Bucur [Thu, 30 Mar 2017 13:24:15 +0000 (16:24 +0300)]
dpaa_eth: use AVOIDBLOCK for Tx confirmation queues

The AVOIDBLOCK flag determines the Tx confirmation queues processing
to be redirected to any available CPU when the current one is slow
in processing them. This may result in a higher Tx confirmation
interrupt count but may reduce pressure on a certain CPU that with
the previous setting would process all Tx confirmation frames.

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agofsl/fman: take into account all RGMII modes
Madalin Bucur [Thu, 30 Mar 2017 13:21:40 +0000 (16:21 +0300)]
fsl/fman: take into account all RGMII modes

Accept the internal delay RGMII variants.

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agovxlan: vxlan dev should inherit lowerdev's gso_max_size
Felix Manlunas [Thu, 30 Mar 2017 00:56:43 +0000 (17:56 -0700)]
vxlan: vxlan dev should inherit lowerdev's gso_max_size

vxlan dev currently ignores lowerdev's gso_max_size, which adversely
affects TSO performance of liquidio if it's the lowerdev.  Egress TCP
packets' skb->len often exceed liquidio's advertised gso_max_size.  This
may happen on other NIC drivers.

Fix it by assigning lowerdev's gso_max_size to that of vxlan dev.  Might as
well do likewise for gso_max_segs.

Single flow TSO throughput of liquidio as lowerdev (using iperf3):

    Before the patch:    139 Mbps
    After the patch :   8.68 Gbps
    Percent increase:  6,144 %

Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosock: avoid dirtying sk_stamp, if possible
Paolo Abeni [Thu, 30 Mar 2017 12:03:06 +0000 (14:03 +0200)]
sock: avoid dirtying sk_stamp, if possible

sock_recv_ts_and_drops() unconditionally set sk->sk_stamp for
every packet, even if the SOCK_TIMESTAMP flag is not set in the
related socket.
If selinux is enabled, this cause a cache miss for every packet
since sk->sk_stamp and sk->sk_security share the same cacheline.
With this change sk_stamp is set only if the SOCK_TIMESTAMP
flag is set, and is cleared for the first packet, so that the user
perceived behavior is unchanged.

This gives up to 5% speed-up under udp-flood with small packets.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'ibmvnic-cleanup-resource-handling'
David S. Miller [Thu, 30 Mar 2017 22:58:43 +0000 (15:58 -0700)]
Merge branch 'ibmvnic-cleanup-resource-handling'

Nathan Fontenot says:

====================
ibmvnic: Cleanup resource handling

In order to better manage the resources of the ibmvnic driver, this set of
patches creates a set of initialization and release routines for the
drivers resources. Additionally, some patches do some re-naming of the
affected routines so that there is a common naming scheme in the driver.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Cleanup failure path in ibmvnic_open
Nathan Fontenot [Thu, 30 Mar 2017 06:49:29 +0000 (02:49 -0400)]
ibmvnic: Cleanup failure path in ibmvnic_open

Now that ibmvnic_release_resources will clean up all of our resources
properly, even if they were not allocated, we can just call this
for failues in ibmvnic_open.

This patch also moves the ibmvnic_release_resources() routine up
in the file to avoid creating a forward declaration ad re-names it to
drop the ibmvnic prefix.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Create init/release routines for stats token
Nathan Fontenot [Thu, 30 Mar 2017 06:49:23 +0000 (02:49 -0400)]
ibmvnic: Create init/release routines for stats token

Create an initialization and a release routine for the stats token used by
the ibmvnic driver.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Merge the two release_sub_crq_queue routines
Nathan Fontenot [Thu, 30 Mar 2017 06:49:18 +0000 (02:49 -0400)]
ibmvnic: Merge the two release_sub_crq_queue routines

Keeping two routines for releasing sub crqs, one for when irqs are not
initialized and one for when they are, is a bit of overkill. Merge the
two routines to a common release routine that will check for an irq
and release it if needed.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Create init and release routines for the rx pool
Nathan Fontenot [Thu, 30 Mar 2017 06:49:12 +0000 (02:49 -0400)]
ibmvnic: Create init and release routines for the rx pool

Move the initialization and the release of the rx pool to their own
routines, and update them to do validation.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Create init and release routines for the tx pool
Nathan Fontenot [Thu, 30 Mar 2017 06:49:06 +0000 (02:49 -0400)]
ibmvnic: Create init and release routines for the tx pool

Move the initialization and the release of the tx pool to their own routines,
and update them to do validation. This also adds validation to the release
of the long term buffer.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Create init and release routines for the bounce buffer
Nathan Fontenot [Thu, 30 Mar 2017 06:49:00 +0000 (02:49 -0400)]
ibmvnic: Create init and release routines for the bounce buffer

Move the handling of initialization and releasing the bounce buffer to their
own init and release routines.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Update main crq initialization and release
Nathan Fontenot [Thu, 30 Mar 2017 06:48:54 +0000 (02:48 -0400)]
ibmvnic: Update main crq initialization and release

Update the initialization and release routines for the crq queue so that
we validate the crq queue.

Additionally this updates the naming of the init and release routines
for the crq queue to drop the ibmvnic prefix. This matches the naming
for similar routines in the driver

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: tcp: Refine the __tcp_select_window
Gao Feng [Wed, 29 Mar 2017 22:49:19 +0000 (06:49 +0800)]
net: tcp: Refine the __tcp_select_window

1. Move the "window = tp->rcv_wnd;" into the condition block without
tp->rx_opt.rcv_wscale.
Because it is unnecessary when enable wscale;

2. Use the macro ALIGN instead of two statements.
The two statements are used to make window align to 1<<wscale.
Use the ALIGN is more clearer.

3. Use the rounddown to make codes clearer.

Signed-off-by: Gao Feng <fgao@ikuai8.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: debug ATU Age Time
Vivien Didelot [Wed, 29 Mar 2017 19:38:37 +0000 (15:38 -0400)]
net: dsa: mv88e6xxx: debug ATU Age Time

The ATU ageing time value programmed in the switch is rounded up to the
nearest multiple of its coefficient (variable depending on the model.)

Add a debug message to inform the user about the exact programmed value.

On 6352, "brctl setageing br0 18" gives "AgeTime set to 0x01 (15000 ms)"
while on 6390 we get "AgeTime set to 0x05 (18750 ms)".

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoibmvnic: Remove debugfs support
Nathan Fontenot [Wed, 29 Mar 2017 19:14:55 +0000 (15:14 -0400)]
ibmvnic: Remove debugfs support

The debugfs support in the ibmvnic driver is not, and never has been,
supported. Just remove it.

The work done in the debugfs code for the driver was part of the original
spec for the ibmvnic driver. The corresponding support for this from the
server side was never supported and has been dropped.

Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobonding: refine bond_fold_stats() wrap detection
Eric Dumazet [Wed, 29 Mar 2017 17:45:44 +0000 (10:45 -0700)]
bonding: refine bond_fold_stats() wrap detection

Some device drivers reset their stats at down/up events, possibly
fooling bonding stats, since they operate with relative deltas.

It is nearly not possible to fix drivers, since some of them compute the
tx/rx counters based on per rx/tx queue stats, and the queues can be
reconfigured (ethtool -L) between the down/up sequence.

Lets avoid accumulating 'negative' values that render bonding stats
useless.

It is better to lose small deltas, assuming the bonding stats are
fetched at a reasonable frequency.

Fixes: 5f0c5f73e5ef ("bonding: make global bonding stats more reliable")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoVSOCK: remove unnecessary ternary operator on return value
Colin Ian King [Wed, 29 Mar 2017 15:33:55 +0000 (16:33 +0100)]
VSOCK: remove unnecessary ternary operator on return value

Rather than assign the positive errno values to ret and then
checking if it is positive and flip the sign, just return the
errno value.

Detected by CoverityScan, CID#986649 ("Logically Dead Code")

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: add explicit interrupt.h includes
Florian Westphal [Wed, 29 Mar 2017 15:17:31 +0000 (17:17 +0200)]
drivers: add explicit interrupt.h includes

These files all use functions declared in interrupt.h, but currently rely
on implicit inclusion of this file (via netns/xfrm.h).

That won't work anymore when the flow cache is removed so include that
header where needed.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: stmmac: dwmac-rk: Add handling for RGMII_ID/RXID/TXID
Wadim Egorov [Wed, 29 Mar 2017 12:12:19 +0000 (14:12 +0200)]
net: stmmac: dwmac-rk: Add handling for RGMII_ID/RXID/TXID

ATM dwmac-rk will always set and enable it's internal delay lines.
Using PHY internal delays in combination with the phy-mode
rgmii-id/rxid/txid was not possible. Only rgmii was supported.

Now we can disable rockchip's gmac delay lines and also use
rgmii-id/rxid/txid.

Tested only with a RK3288 based board.

Signed-off-by: Wadim Egorov <w.egorov@phytec.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoRevert "net: stmmac: enable multiple buffers"
LABBE Corentin [Wed, 29 Mar 2017 05:05:40 +0000 (07:05 +0200)]
Revert "net: stmmac: enable multiple buffers"

The commit aff3d9eff843 ("net: stmmac: enable multiple buffers") breaks
numerous boards. while some patch exists for fixing some of it,
dwmac-sunxi is still broken with it.
Since this patch is very huge, it will be better to split it in smaller
part.

Signed-off-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 29 Mar 2017 21:13:09 +0000 (14:13 -0700)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-03-29

This series contains updates to i40e and i40evf only.

Preethi changes the default driver mode of operation to descriptor
write-back for VF.

Alex cleans up and addresses several issues in the way that i40e handles
private flags.  Modifies the driver to use the length of the packet
instead of the DD status bit to determine if a new descriptor is ready
to be processed.  Refactors the driver by pulling the code responsible
for fetching the receive buffer and synchronizing DMA into a single
function.  Also pulled the code responsible for handling buffer
recycling and page counting and distributed it through several functions,
so we can commonize the bits that handle either freeing or recycling the
buffers.  Cleans up the code in preparation for us adding support for
build_skb().  Changed the way we handle the maximum frame size for the
receive path so it is more consistent with other drivers.

Paul enables XL722 to use the direct read/write method since it does not
support the AQ command to read/write the control register.

Christopher fixes a case where we miss an arq element if a new one is
added before we enable interrupts and exit the loop.

Jake cleans up a pointless goto statement.  Also cleaned up a flag that
was not being used.

Carolyn does round 2 for adding a delay to the receive queue to
accommodate the hardware needs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'tipc-socketpair'
David S. Miller [Wed, 29 Mar 2017 21:10:12 +0000 (14:10 -0700)]
Merge branch 'tipc-socketpair'

Parthasarathy Bhuvaragan says:

====================
tipc: add socketpair support

We add socketpair support for connection oriented sockets in
the first patch and for connection less in the second.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotipc: allow rdm/dgram socketpairs
Erik Hugne [Wed, 29 Mar 2017 09:22:17 +0000 (11:22 +0200)]
tipc: allow rdm/dgram socketpairs

for socketpairs using connectionless transport, we cache
the respective node local TIPC portid to use in subsequent
calls to send() in the socket's private data.

Signed-off-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotipc: add support for stream/seqpacket socketpairs
Erik Hugne [Wed, 29 Mar 2017 09:22:16 +0000 (11:22 +0200)]
tipc: add support for stream/seqpacket socketpairs

sockets A and B are connected back-to-back, similar to what
AF_UNIX does.

Signed-off-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: Parthasarathy Bhuvaragan <parthasarathy.bhuvaragan@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvneta: set rx mode during resume if interface is running
Jisheng Zhang [Wed, 29 Mar 2017 08:47:19 +0000 (16:47 +0800)]
net: mvneta: set rx mode during resume if interface is running

I found a bug by:

0. boot and start dhcp client
1. echo mem > /sys/power/state
2. resume back immediately
3. don't touch dhcp client to renew the lease
4. ping the gateway. No acks

Usually, after step2, the DHCP lease isn't expired, so in theory we
should resume all back. But in fact, it doesn't. It turns out
the rx mode isn't resumed correctly. This patch fixes it by adding
mvneta_set_rx_mode(dev) in the resume hook if interface is running.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvneta: add RGMII_RXID and RGMII_TXID support
Jisheng Zhang [Wed, 29 Mar 2017 08:42:26 +0000 (16:42 +0800)]
net: mvneta: add RGMII_RXID and RGMII_TXID support

RGMII_RXID and RGMII_TX_ID share the same GMAC CTRL setting as RGMII
or RGMII_ID.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: veth: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Wed, 29 Mar 2017 06:24:21 +0000 (08:24 +0200)]
net: veth: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge tag 'mlx5e-pedit' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
David S. Miller [Wed, 29 Mar 2017 18:24:21 +0000 (11:24 -0700)]
Merge tag 'mlx5e-pedit' of git://git./linux/kernel/git/saeed/linux

Or Gerlitz says:

====================
mlx5e-pedit 2017-03-28

This series adds support for offloading modifications of packet headers using
ConnectX-5 HW header re-write as an action applied during packet steering.

The offloaded SW mechanism is TC's pedit action. The offloading is
supported for E-Switch steering of VF traffic in the SRIOV
switchdev mode and for NIC (non eswitch) RX.

One use-case for this offload on virtual networks, is when the hypervisor
implements flow based router such as Open-Stack's DVR, where L2 headers
of guest packets re-written with routers' MAC addresses and the IP TTL
is decremented.

Another use case (which can be applied in parallel with routing) is
stateless NAT where guest L3/L4 headers are re-written.

The series is built as follows: the 1st six patches are preperations which
don't yet add new functionality, patches 7-8 add the FW APIs (data-structures
and commands) for header re-write, and patch nine allows offloading driver
to access pedit keys.

The 10th patch is somehow the core of the series, where we translate from
the pedit way to represent set of header modification elements to the FW
API for that same matter.

Once a set of HW modification is established, we register it with the FW
and get a modify header ID. When this ID is used with an action during
packet steering, the HW applies the header modification on the packet.

Patches 11 and 12 implement the above logic as an offload for pedit action
for the NIC and E-Switch use-cases.

I'd like to thanks Elijah Shakkour <elijahs@mellanox.com> for implementing
and helping me testing this functionality on HW simulator, before it could
be done with FW.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mpls: Update lfib_nlmsg_size to skip deleted nexthops
David Ahern [Tue, 28 Mar 2017 22:19:49 +0000 (15:19 -0700)]
net: mpls: Update lfib_nlmsg_size to skip deleted nexthops

A recent commit skips nexthops in a route if the device has been
deleted. Update lfib_nlmsg_size accordingly.

Reported-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Acked-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: phy: Allow building mdio-boardinfo into the kernel
Florian Fainelli [Tue, 28 Mar 2017 19:57:09 +0000 (12:57 -0700)]
net: phy: Allow building mdio-boardinfo into the kernel

mdio-boardinfo contains code that is helpful for platforms to register
specific MDIO bus devices independent of how CONFIG_MDIO_DEVICE or
CONFIG_PHYLIB will be selected (modular or built-in). In order to make
that possible, let's do the following:

- descend into drivers/net/phy/ unconditionally

- make mdiobus_setup_mdiodev_from_board_info() take a callback argument
  which allows us not to expose the internal MDIO board info list and
  mutex, yet maintain the logic within the same file

- relocate the code that creates a MDIO device into
  drivers/net/phy/mdio_bus.c

- build mdio-boardinfo.o into the kernel as soon as MDIO_DEVICE is
  defined (y or m)

Fixes: 90eff9096c01 ("net: phy: Allow splitting MDIO bus/device support from PHYs")
Fixes: 648ea0134069 ("net: phy: Allow pre-declaration of MDIO devices")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoi40e: fix for queue timing delays
Wyborny, Carolyn [Tue, 28 Mar 2017 15:00:48 +0000 (08:00 -0700)]
i40e: fix for queue timing delays

This patch adds a delay to Rx queue disables to accommodate HW needs.

v2: Added missing check for disable only, additional details on the
need for the ugly delay and fixed spacing on comment.

Change-ID: I2864ca667ce5dcc2cc44f8718113b719742a46a1
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Change the way we limit the maximum frame size for Rx
Alexander Duyck [Tue, 14 Mar 2017 17:15:27 +0000 (10:15 -0700)]
i40e/i40evf: Change the way we limit the maximum frame size for Rx

This patch changes the way we handle the maximum frame size for the Rx
path.  Previously we were rounding up to 2K for a 1500 MTU and then brining
the max frame size down to MTU plus a fixed amount.  With this patch
applied what we now do is limit the maximum frame to 1.5K minus the value
for NET_IP_ALIGN for standard MTU, and for any MTU greater than 1500 we
allow up to the maximum frame size.  This makes the behavior more
consistent with the other drivers such as igb which had similar logic.  In
addition it reduces the test matrix for MTU since we only have two max
frame sizes that are handled for Rx now.

Change-ID: I23a9d3c857e7df04b0ef28c64df63e659c013f3f
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Add legacy-rx private flag to allow fallback to old Rx flow
Alexander Duyck [Tue, 14 Mar 2017 17:15:26 +0000 (10:15 -0700)]
i40e/i40evf: Add legacy-rx private flag to allow fallback to old Rx flow

This patch adds a control which will allow us to toggle into and out of the
legacy Rx mode.  The legacy Rx mode is what we currently do when performing
Rx.  As I make further changes what should happen is that the driver will
fall back to the behavior for Rx as of this patch should the "legacy-rx"
flag be set to on.

Change-ID: I0342998849bbb31351cce05f6e182c99174e7751
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Break i40e_fetch_rx_buffer up to allow for reuse of frag code
Alexander Duyck [Tue, 14 Mar 2017 17:15:25 +0000 (10:15 -0700)]
i40e/i40evf: Break i40e_fetch_rx_buffer up to allow for reuse of frag code

This patch is meant to clean up the code in preparation for us adding
support for build_skb.  Specifically we deconstruct i40e_fetch_buffer into
several functions so that those functions can later be reused when we add a
path for build_skb.

Specifically with this change we split out the code for adding a page to an
exiting skb.

Change-ID: Iab1efbab6b8b97cb60ab9fdd0be1d37a056a154d
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Pull out code for cleaning up Rx buffers
Alexander Duyck [Tue, 14 Mar 2017 17:15:24 +0000 (10:15 -0700)]
i40e/i40evf: Pull out code for cleaning up Rx buffers

This patch pulls out the code responsible for handling buffer recycling and
page counting and distributes it through several functions.  This allows us
to commonize the bits that handle either freeing or recycling the buffers.

As far as the page count tracking one change to the logic is that
pagecnt_bias is decremented as soon as we call i40e_get_rx_buffer.  It is
then the responsibility of the function that pulls the data to either
increment the pagecnt_bias if the buffer can be recycled as-is, or to
update page_offset so that we are pointing at the correct location for
placement of the next buffer.

Change-ID: Ibac576360cb7f0b1627f2a993d13c1a8a2bf60af
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Pull code for grabbing and syncing rx_buffer from fetch_buffer
Alexander Duyck [Tue, 14 Mar 2017 17:15:23 +0000 (10:15 -0700)]
i40e/i40evf: Pull code for grabbing and syncing rx_buffer from fetch_buffer

This patch pulls the code responsible for fetching the Rx buffer and
synchronizing DMA into a function, specifically called i40e_get_rx_buffer.

The general idea is to allow for better code reuse by pulling this out of
i40e_fetch_rx_buffer.  We dropped a couple of prefetches since the time
between the prefetch being called and the data being accessed was too small
to be useful.

Change-ID: I4885fce4b2637dbedc8e16431169d23d3d7e79b9
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e/i40evf: Use length to determine if descriptor is done
Alexander Duyck [Tue, 14 Mar 2017 17:15:22 +0000 (10:15 -0700)]
i40e/i40evf: Use length to determine if descriptor is done

This change makes it so that we use the length of the packet instead of the
DD status bit to determine if a new descriptor is ready to be processed.
The obvious advantage is that it cuts down on reads as we don't really even
need the DD bit if going from a 0 to a non-zero value on size is enough to
inform us that the packet has been completed.

Change-ID: Iebdf9cdb36c454ef092df27199b92ad09c374231
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: remove FDIR_REQUIRES_REINIT driver flag
Jacob Keller [Fri, 10 Mar 2017 20:22:05 +0000 (12:22 -0800)]
i40e: remove FDIR_REQUIRES_REINIT driver flag

This flag hasn't been used since commit 1e1be8f622ee ("i40e: ATR policy
change to flush the table to clean stale ATR rules").

Lets simplify things and just remove it.

Change-ID: I76279d84db8a2fd96f445b96aa413059f9256879
Signed-off-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>
7 years agoi40e: remove a useless goto statement
Jacob Keller [Fri, 10 Mar 2017 20:22:04 +0000 (12:22 -0800)]
i40e: remove a useless goto statement

The goto found here for when in MFP mode is pointless. It jumps to the
end of a series of if blocks. However, right after this statement is
a closing '}' for this if block, which will result in the program flow
going to the exact same location as the goto statement indicates. Thus,
regardless of whether we are in MFP mode, the program flow will resume
from the same location.

This arose due to various refactoring which did not notice that this
goto became essentially a no-op.

To properly understand this diff you will need to view a larger context
than is given by default.

Change-ID: I088f73c3831aa5c4e2281380c7a3ce605594300c
Signed-off-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>
7 years agoi40e: Check for new arq elements before leaving the adminq subtask loop
Christopher N Bednarz [Fri, 10 Mar 2017 20:22:03 +0000 (12:22 -0800)]
i40e: Check for new arq elements before leaving the adminq subtask loop

Fix a case where we miss an arq element if a new one is added before we
enable interrupts and exit the arq subtask loop. This occurs frequently
with RDMA running on Windows VF and causes long delays that prevent SMB
from establishing connections.

Change-ID: I3e1c8b2b960c12857d9b8275bea2c1563674392e
Signed-off-by: Christopher N Bednarz <christopher.n.bednarz@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: use register for XL722 control register read/write
Paul M Stillwell Jr [Fri, 10 Mar 2017 20:22:02 +0000 (12:22 -0800)]
i40e: use register for XL722 control register read/write

The XL722 doesn't support the AQ command to read/write the control
register so enable it to bypass the check and use the direct read/write
method.

Change-ID: Iefecc737b57207485c90845af5989d5af518bf16
Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: Clean up handling of private flags
Alexander Duyck [Fri, 10 Mar 2017 20:22:01 +0000 (12:22 -0800)]
i40e: Clean up handling of private flags

This patch cleans up and addresses several issues in the way that i40e
handles private flags. Previously the code was choosing fixed bits and
trying to match them up with strings in a somewhat haphazard way. This
resulted in the possibility for adding a new bit and causing a mismatch as
the private flags are linear bits starting at 0, and the private flags in
the driver were split up over a group specific to the PF and a group that
was global.

What this change does is define an array of structs used to represent the
private flags. Contained within the structs are the bits necessary to know
which flags to set and/or clear depending on the state of the bit. By
doing this we can add new bits in the future with minimal overhead and
avoid creating possible mis-matches should we need to remove a flag based
on compile options.

Change-ID: Ia3214ab04f0ab2f70354ac0997a135f1d01b0acd
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>