linux-2.6-block.git
5 years agotcp: add stat of data packet reordering events
Wei Wang [Wed, 1 Aug 2018 00:46:24 +0000 (17:46 -0700)]
tcp: add stat of data packet reordering events

Introduce a new TCP stats to record the number of reordering events seen
and expose it in both tcp_info (TCP_INFO) and opt_stats
(SOF_TIMESTAMPING_OPT_STATS).
Application can use this stats to track the frequency of the reordering
events in addition to the existing reordering stats which tracks the
magnitude of the latest reordering event.

Note: this new stats tracks reordering events triggered by ACKs, which
could often be fewer than the actual number of packets being delivered
out-of-order.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotcp: add dsack blocks received stats
Wei Wang [Wed, 1 Aug 2018 00:46:23 +0000 (17:46 -0700)]
tcp: add dsack blocks received stats

Introduce a new TCP stat to record the number of DSACK blocks received
(RFC4989 tcpEStatsStackDSACKDups) and expose it in both tcp_info
(TCP_INFO) and opt_stats (SOF_TIMESTAMPING_OPT_STATS).

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotcp: add data bytes retransmitted stats
Wei Wang [Wed, 1 Aug 2018 00:46:22 +0000 (17:46 -0700)]
tcp: add data bytes retransmitted stats

Introduce a new TCP stat to record the number of bytes retransmitted
(RFC4898 tcpEStatsPerfOctetsRetrans) and expose it in both tcp_info
(TCP_INFO) and opt_stats (SOF_TIMESTAMPING_OPT_STATS).

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotcp: add data bytes sent stats
Wei Wang [Wed, 1 Aug 2018 00:46:21 +0000 (17:46 -0700)]
tcp: add data bytes sent stats

Introduce a new TCP stat to record the number of bytes sent
(RFC4898 tcpEStatsPerfHCDataOctetsOut) and expose it in both tcp_info
(TCP_INFO) and opt_stats (SOF_TIMESTAMPING_OPT_STATS).

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotcp: add a helper to calculate size of opt_stats
Wei Wang [Wed, 1 Aug 2018 00:46:20 +0000 (17:46 -0700)]
tcp: add a helper to calculate size of opt_stats

This is to refactor the calculation of the size of opt_stats to a helper
function to make the code cleaner and easier for later changes.

Suggested-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ipv4-Control-SKB-reprioritization-after-forwarding'
David S. Miller [Wed, 1 Aug 2018 16:52:31 +0000 (09:52 -0700)]
Merge branch 'ipv4-Control-SKB-reprioritization-after-forwarding'

Petr Machata says:

====================
ipv4: Control SKB reprioritization after forwarding

After IPv4 packets are forwarded, the priority of the corresponding SKB
is updated according to the TOS field of IPv4 header. This overrides any
prioritization done earlier by e.g. an skbedit action or ingress-qos-map
defined at a vlan device.

Such overriding may not always be desirable. Even if the packet ends up
being routed, which implies this is an L3 network node, an administrator
may wish to preserve whatever prioritization was done earlier on in the
pipeline.

Therefore this patch set introduces a sysctl that controls this
behavior, net.ipv4.ip_forward_update_priority. It's value is 1 by
default to preserve the current behavior.

All of the above is implemented in patch #1.

Value changes prompt a new NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE
notification, so that the drivers can hook up whatever logic may depend
on this value. That is implemented in patch #2.

In patches #3 and #4, mlxsw is adapted to recognize the sysctl. On
initialization, the RGCR register that handles router configuration is
set in accordance with the sysctl. The new notification is listened to
and RGCR is reconfigured as necessary.

In patches #5 to #7, a selftest is added to verify that mlxsw reflects
the sysctl value as necessary. The test is expressed in terms of the
recently-introduced ieee_setapp support, and works by observing how DSCP
value gets rewritten depending on packet priority. For this reason, the
test is added to the subdirectory drivers/net/mlxsw. Even though it's
not particularly specific to mlxsw, it's not suitable for running on
soft devices (which don't support the ieee_setapp et.al.).

Changes from v1 to v2:

- In patch #1, init sysctl_ip_fwd_update_priority to 1 instead of true.

Changes from RFC to v1:

- Fix wrong sysctl name in ip-sysctl.txt
- Add notifications
- Add mlxsw support
- Add self test
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: mlxsw: Add test for ip_forward_update_priority
Petr Machata [Tue, 31 Jul 2018 22:39:29 +0000 (00:39 +0200)]
selftests: mlxsw: Add test for ip_forward_update_priority

Verify that with that sysctl turned off, DSCP prioritization and rewrite
works the same way as in qos_dscp_bridge test. However when the sysctl
is charged, there should be a reprioritization after routing stage,
which will be observed by a different DSCP rewrite on egress.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Move DSCP capture to lib.sh
Petr Machata [Tue, 31 Jul 2018 22:39:25 +0000 (00:39 +0200)]
selftests: forwarding: Move DSCP capture to lib.sh

dscp_capture_install() and dscp_capture_uninstall() are going to be
useful for a test added by a following patch, move them therefore to
lib.sh together with related helpers.

While doing so, change the rule preference from mere DSCP value to
DSCP+100 in order to support adding captures of packets with DSCP of 0.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Move lldpad waiting to lib.sh
Petr Machata [Tue, 31 Jul 2018 22:38:59 +0000 (00:38 +0200)]
selftests: forwarding: Move lldpad waiting to lib.sh

The function lldpad_wait() will be useful for a test added by a
following patch. Likewise would the "sleep 5" with its extensive
comment.

Therefore move lldpad_wait() to lib.sh in order to allow reuse. Rename
it to lldpad_app_wait_set() to recognize that what this is intended to
wait on are the pending APP sets.

For the sleeping, add a function lldpad_app_wait_del(). That will serve
to hold the related explanatory comment (which edit for clarity), and as
a token in the caller to identify the sites where this sort of waiting
takes place. That will serve when/if a better way to handle this
business is found.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_router: Handle sysctl_ip_fwd_update_priority
Petr Machata [Tue, 31 Jul 2018 22:38:03 +0000 (00:38 +0200)]
mlxsw: spectrum_router: Handle sysctl_ip_fwd_update_priority

This sysctl setting controls whether packet priority should be updated
after forwarding. Configure RGCR.usp accordingly so that the device is
in sync with the kernel handling.

Note that RGCR doesn't allow changing arbitrary parameters
mid-operation, however "usp" is exempt and can be reconfigured.

Also react to NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE notifications
that signify change in this configuration.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: Extract work-scheduling into a new function
Petr Machata [Tue, 31 Jul 2018 22:37:36 +0000 (00:37 +0200)]
mlxsw: spectrum: Extract work-scheduling into a new function

The boilerplate to schedule NETEVENT_IPV4_MPATH_HASH_UPDATE and
NETEVENT_IPV6_MPATH_HASH_UPDATE handling is almost equivalent to that of
NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE that's coming in the next
patch. The only difference is which actual worker function should be
called. Extract this boilerplate into a named function in order to allow
reuse.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ipv4: Notify about changes to ip_forward_update_priority
Petr Machata [Tue, 31 Jul 2018 22:36:42 +0000 (00:36 +0200)]
net: ipv4: Notify about changes to ip_forward_update_priority

Drivers may make offloading decision based on whether
ip_forward_update_priority is enabled or not. Therefore distribute
netevent notifications to give them a chance to react to a change.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ipv4: Control SKB reprioritization after forwarding
Petr Machata [Tue, 31 Jul 2018 22:36:03 +0000 (00:36 +0200)]
net: ipv4: Control SKB reprioritization after forwarding

After IPv4 packets are forwarded, the priority of the corresponding SKB
is updated according to the TOS field of IPv4 header. This overrides any
prioritization done earlier by e.g. an skbedit action or ingress-qos-map
defined at a vlan device.

Such overriding may not always be desirable. Even if the packet ends up
being routed, which implies this is an L3 network node, an administrator
may wish to preserve whatever prioritization was done earlier on in the
pipeline.

Therefore introduce a sysctl that controls this behavior. Keep the
default value at 1 to maintain backward-compatible behavior.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: add helpers checking if socket can be bound to nonlocal address
Vincent Bernat [Tue, 31 Jul 2018 19:18:11 +0000 (21:18 +0200)]
net: add helpers checking if socket can be bound to nonlocal address

The construction "net->ipv4.sysctl_ip_nonlocal_bind || inet->freebind
|| inet->transparent" is present three times and its IPv6 counterpart
is also present three times. We introduce two small helpers to
characterize these tests uniformly.

Signed-off-by: Vincent Bernat <vincent@bernat.im>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: change Exar/Neterion menu items to be alphabetical
Jon Mason [Tue, 31 Jul 2018 15:56:19 +0000 (11:56 -0400)]
net: change Exar/Neterion menu items to be alphabetical

Neterion was standalone for several years, then acquired by Exar and
shutdown in 11 months without ever making any new Exar branded adapters.
Users would probably think of them as Neterion and not Exar (as there
have been no follow-on adapters and the vast majority ever sold were
under the Neterion name).

6c541b4595a2 ("net: ethernet: Sort Kconfig sourcing alphabetically")
sorted Kconfig sourcing based on directory names, but in a couple cases,
the menu item text is quite different from the directory name and is not
sorted correctly:

  drivers/net/ethernet/neterion/Kconfig    => "Exar devices"

To address that and clear up any confusion about the name, "Exar" was
changed to "Neterion (Exar)" and the relevant entries in the Makefile
and Kconfig were reordered to match the alphabetical organization.

Inspired-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Jon Mason <jdmason@kudzu.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: Use kmemdup to simplify the code
zhong jiang [Tue, 31 Jul 2018 16:50:24 +0000 (00:50 +0800)]
net/tls: Use kmemdup to simplify the code

Kmemdup is better than kmalloc+memcpy. So replace them.

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tipc: remove redundant variables 'tn' and 'oport'
Colin Ian King [Tue, 31 Jul 2018 16:01:37 +0000 (17:01 +0100)]
net/tipc: remove redundant variables 'tn' and 'oport'

Variables 'tn' and 'oport'  are being assigned but are never used hence
they are redundant and can be removed.

Cleans up clang warnings:
warning: variable 'oport' set but not used [-Wunused-but-set-variable]
warning: variable 'tn' set but not used [-Wunused-but-set-variable]

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'be2net-next'
David S. Miller [Wed, 1 Aug 2018 16:39:10 +0000 (09:39 -0700)]
Merge branch 'be2net-next'

Suresh Reddy says:

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

v1->v2 : Modified the subject line and commit log.

Please consider applying these two patches to net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobe2net: Update the driver version to 12.0.0.0
Suresh Reddy [Tue, 31 Jul 2018 15:39:43 +0000 (11:39 -0400)]
be2net: Update the driver version to 12.0.0.0

Signed-off-by: Suresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobe2net: gather debug info and reset adapter (only for Lancer) on a tx-timeout
Suresh Reddy [Tue, 31 Jul 2018 15:39:42 +0000 (11:39 -0400)]
be2net: gather debug info and reset adapter (only for Lancer) on a tx-timeout

This patch handles a TX-timeout as follows:

1) This patch gathers and prints the following info that can
   help in diagnosing the cause of a TX-timeout.
   a) TX queue and completion queue entries.
   b) SKB and TCP/UDP header details.

2) For Lancer NICs (TX-timeout recovery is not supported for
   BE3/Skyhawk-R NICs), it recovers from the TX timeout as follows:

   a) On a TX-timeout, driver sets the PHYSDEV_CONTROL_FW_RESET_MASK
      bit in the PHYSDEV_CONTROL register. Lancer firmware goes into
      an error state and indicates this back to the driver via a bit
      in a doorbell register.
   b) Driver detects this and calls be_err_recover(). DMA is disabled,
      all pending TX skbs are unmapped and freed (be_close()). All rings
      are destroyed (be_clear()).
   c) The driver waits for the FW to re-initialize and re-creates all
      rings along with other data structs (be_resume())

Signed-off-by: Suresh Reddy <suresh.reddy@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx5e: Fix uninitialized variable
Gustavo A. R. Silva [Tue, 31 Jul 2018 14:21:57 +0000 (09:21 -0500)]
net/mlx5e: Fix uninitialized variable

There is a potential execution path in which variable *err* is returned
without being properly initialized previously.

Fix this by initializing variable *err* to 0.

Addresses-Coverity-ID: 1472116 ("Uninitialized scalar variable")
Fixes: 0ec13877ce95 ("net/mlx5e: Gather all XDP pre-requisite checks in a single function")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: Make some functions static
YueHaibing [Tue, 31 Jul 2018 14:12:30 +0000 (22:12 +0800)]
qed: Make some functions static

Fixes the following sparse warning:

drivers/net/ethernet/qlogic/qed/qed_cxt.c:1534:6: warning: symbol 'qed_cm_init_pf' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dev.c:233:4: warning: symbol 'qed_init_qm_get_num_tcs' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dev.c:238:5: warning: symbol 'qed_init_qm_get_num_vfs' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dev.c:246:5: warning: symbol 'qed_init_qm_get_num_pf_rls' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dev.c:264:5: warning: symbol 'qed_init_qm_get_num_vports' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dev.c:276:5: warning: symbol 'qed_init_qm_get_num_pqs' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_mcp.c:573:5: warning: symbol 'qed_mcp_nvm_wr_cmd' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_mcp.c:3012:1: warning: symbol '__qed_mcp_resc_lock' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_dcbx.c:870:6: warning: symbol 'qed_dcbx_aen' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_debug.c:7841:5: warning: symbol 'qed_dbg_nvm_image_length' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_debug.c:7857:5: warning: symbol 'qed_dbg_nvm_image' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_sriov.c:675:6: warning: symbol '_qed_iov_pf_sanity_check' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_sriov.c:690:6: warning: symbol 'qed_iov_pf_sanity_check' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_sriov.c:3982:6: warning: symbol 'qed_iov_pf_get_pending_events' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_vf.c:172:5: warning: symbol '_qed_vf_pf_release' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_rdma.c:137:5: warning: symbol 'qed_rdma_get_sb_id' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_rdma.c:709:5: warning: symbol 'qed_rdma_stop' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_ll2.c:161:6: warning: symbol 'qed_ll2b_complete_rx_packet' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_roce.c:160:6: warning: symbol 'qed_roce_free_cid_pair' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:380:12: warning: symbol 'iwarp_state_names' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:946:1: warning: symbol 'qed_iwarp_parse_private_data' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:971:1: warning: symbol 'qed_iwarp_mpa_reply_arrived' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:2504:1: warning: symbol 'qed_iwarp_ll2_slowpath' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:2806:6: warning: symbol 'qed_iwarp_qp_in_error' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:2827:6: warning: symbol 'qed_iwarp_exception_received' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iwarp.c:2958:1: warning: symbol 'qed_iwarp_connect_complete' was not declared. Should it be static?
drivers/net/ethernet/qlogic/qed/qed_iscsi.c:876:6: warning: symbol 'qed_iscsi_free_connection' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Denis Bolotin <Denis.Bolotin@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agords: Remove IPv6 dependency
Ka-Cheong Poon [Tue, 31 Jul 2018 05:48:42 +0000 (22:48 -0700)]
rds: Remove IPv6 dependency

This patch removes the IPv6 dependency from RDS.

Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agords: rds_ib_recv_alloc_cache() should call alloc_percpu_gfp() instead
Ka-Cheong Poon [Tue, 31 Jul 2018 05:48:41 +0000 (22:48 -0700)]
rds: rds_ib_recv_alloc_cache() should call alloc_percpu_gfp() instead

Currently, rds_ib_conn_alloc() calls rds_ib_recv_alloc_caches()
without passing along the gfp_t flag.  But rds_ib_recv_alloc_caches()
and rds_ib_recv_alloc_cache() should take a gfp_t parameter so that
rds_ib_recv_alloc_cache() can call alloc_percpu_gfp() using the
correct flag instead of calling alloc_percpu().

Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: ti: cpsw: replace unnecessarily macroses on functions
Ivan Khoronzhuk [Mon, 30 Jul 2018 22:05:39 +0000 (01:05 +0300)]
net: ethernet: ti: cpsw: replace unnecessarily macroses on functions

Replace ugly macroses on functions.

Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: remove bogus RCU annotations on socket.wq
Christoph Hellwig [Mon, 30 Jul 2018 07:45:07 +0000 (09:45 +0200)]
net: remove bogus RCU annotations on socket.wq

We never use RCU protection for it, just a lot of cargo-cult
rcu_deference_protects calls.

Note that we do keep the kfree_rcu call for it, as the references through
struct sock are RCU protected and thus might require a grace period before
freeing.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovirtio-net: get rid of unnecessary container of rq stats
Jason Wang [Tue, 31 Jul 2018 09:43:39 +0000 (17:43 +0800)]
virtio-net: get rid of unnecessary container of rq stats

We don't maintain tx counters in rx stats any more. There's no need
for an extra container of rq stats.

Cc: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovirtio-net: correctly update XDP_TX counters
Jason Wang [Tue, 31 Jul 2018 09:43:38 +0000 (17:43 +0800)]
virtio-net: correctly update XDP_TX counters

Commit 5b8f3c8d30a6 ("virtio_net: Add XDP related stats") tries to
count TX XDP stats in virtnet_receive(). This will cause several
issues:

- virtnet_xdp_sq() was called without checking whether or not XDP is
  set. This may cause out of bound access when there's no enough txq
  for XDP.
- Stats were updated even if there's no XDP/XDP_TX.

Fixing this by reusing virtnet_xdp_xmit() for XDP_TX which can counts
TX XDP counter itself and remove the unnecessary tx stats embedded in
rx stats.

Reported-by: syzbot+604f8271211546f5b3c7@syzkaller.appspotmail.com
Fixes: 5b8f3c8d30a6 ("virtio_net: Add XDP related stats")
Cc: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'xsk-improvements-to-RX-queue-check-and-replace'
David S. Miller [Tue, 31 Jul 2018 16:48:22 +0000 (09:48 -0700)]
Merge branch 'xsk-improvements-to-RX-queue-check-and-replace'

Jakub Kicinski says:

====================
xsk: improvements to RX queue check and replace

First 3 patches of my recent RFC.  The first one make the check against
real_num_rx_queues slightly more reliable, while the latter two redefine
XDP_QUERY_XSK_UMEM slightly to disallow replacing UMEM in the driver at
the stack level.

I'm not sure where this lays on the bpf vs net trees scale, but there
should be no conflicts with either tree.
====================

Acked-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoxsk: don't allow umem replace at stack level
Jakub Kicinski [Tue, 31 Jul 2018 03:43:53 +0000 (20:43 -0700)]
xsk: don't allow umem replace at stack level

Currently drivers have to check if they already have a umem
installed for a given queue and return an error if so.  Make
better use of XDP_QUERY_XSK_UMEM and move this functionality
to the core.

We need to keep rtnl across the calls now.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Björn Töpel <bjorn.topel@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoxsk: refactor xdp_umem_assign_dev()
Jakub Kicinski [Tue, 31 Jul 2018 03:43:52 +0000 (20:43 -0700)]
xsk: refactor xdp_umem_assign_dev()

Return early and only take the ref on dev once there is no possibility
of failing.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Björn Töpel <bjorn.topel@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: update real_num_rx_queues even when !CONFIG_SYSFS
Jakub Kicinski [Tue, 31 Jul 2018 03:43:51 +0000 (20:43 -0700)]
net: update real_num_rx_queues even when !CONFIG_SYSFS

We used to depend on real_num_rx_queues as a upper bound for sanity
checks.  For AF_XDP socket validation it's useful if the check behaves
the same regardless of CONFIG_SYSFS setting.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlx5: handle DMA mapping error case for XDP redirect
Jesper Dangaard Brouer [Mon, 30 Jul 2018 17:49:08 +0000 (19:49 +0200)]
mlx5: handle DMA mapping error case for XDP redirect

Commit 58b99ee3e3eb ("net/mlx5e: Add support for XDP_REDIRECT in device-out side")
forgot to return/free the xdp_frame in case the DMA mapping failed, correct this.

Also DMA unmap the frame in case mlx5e_xmit_xdp_frame() fails.

Fixes: 58b99ee3e3eb ("net/mlx5e: Add support for XDP_REDIRECT in device-out side")
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohv_netvsc: Add per-cpu ethtool stats for netvsc
Yidong Ren [Mon, 30 Jul 2018 17:09:45 +0000 (17:09 +0000)]
hv_netvsc: Add per-cpu ethtool stats for netvsc

This patch implements following ethtool stats fields for netvsc:
cpu<n>_tx/rx_packets/bytes
cpu<n>_vf_tx/rx_packets/bytes

Corresponding per-cpu counters already exist in current code. Exposing
these counters will help troubleshooting performance issues.

for_each_present_cpu() was used instead of for_each_possible_cpu().
for_each_possible_cpu() would create very long and useless output.
It is still being used for internal buffer, but not for ethtool
output.

There could be an overflow if cpu was added between ethtool
call netvsc_get_sset_count() and netvsc_get_ethtool_stats() and
netvsc_get_strings(). (still safe if cpu was removed)
ethtool makes these three function calls separately.
As long as we use ethtool, I can't see any clean solution.

Currently and in foreseeable short term, Hyper-V doesn't support
cpu hot-plug. Plus, ethtool is for admin use. Unlikely the admin
would perform such combo operations.

Changes in v2:
  - Remove cpp style comment
  - Resubmit after freeze

Changes in v3:
  - Reimplemented with kvmalloc instead of alloc_percpu

Changes in v4:
  - Fixed inconsistent array size
  - Use kvmalloc_array instead of kvmalloc

Signed-off-by: Yidong Ren <yidren@microsoft.com>
Reviewed-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'selftests-mirror-to-gretap-with-team'
David S. Miller [Mon, 30 Jul 2018 16:47:22 +0000 (09:47 -0700)]
Merge branch 'selftests-mirror-to-gretap-with-team'

Petr Machata says:

====================
A test for mirror-to-gretap with team in UL packet path

This patchset adds a test for "tc action mirred mirror" where the
mirrored-to device is a gretap, and underlay path contains a team
device.

In patch #1 require_command() is added, which should henceforth be used
to declare dependence on a certain tool.

In patch #2, two new functions, team_create() and team_destroy(), are
added to lib.sh.

The newly-added test uses arping, which isn't necessarily available.
Therefore patch #3 introduces $ARPING, and a preexisting test is fixed
to require_command $ARPING.

In patches #4 and #5, two new tests are added. In both cases, a team
device is on egress path of a mirrored packet in a mirror-to-gretap
scenario. In the first one, the team device is in loadbalance mode, in
the second one it's in lacp mode. (The difference in modes necessitates
a different testing strategy, hence two test cases instead of just
parameterizing one.)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Test mirror-to-gretap w/ UL team LACP
Petr Machata [Mon, 30 Jul 2018 14:40:07 +0000 (16:40 +0200)]
selftests: forwarding: Test mirror-to-gretap w/ UL team LACP

This tests mirror-to-gretap when an underlay packet path includes a team
device which is not in loadbalance mode, but in LACP mode. The test
manipulates LAG membership to achieve changes in txability, thus making
sure that a driver that offloads mirror-to-gretap doesn't just consider
upness of a device.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Test mirror-to-gretap w/ UL team
Petr Machata [Mon, 30 Jul 2018 14:40:02 +0000 (16:40 +0200)]
selftests: forwarding: Test mirror-to-gretap w/ UL team

Test for "tc action mirred egress mirror" that mirrors to gretap when
the underlay route points at a VLAN-aware bridge (802.1q), and the
traffic egresses the bridge through a team device. Test upping and
downing individual team device slaves and verify the traffic flows as
expected.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Introduce $ARPING
Petr Machata [Mon, 30 Jul 2018 14:39:57 +0000 (16:39 +0200)]
selftests: forwarding: Introduce $ARPING

Instead of relying on "arping" being installed everywhere under that
name, introduce a variable $ARPING like the other tools do.

Convert an existing test, mirror_gre_vlan_bridge_1q.sh to
require_command $ARPING and then invoke arping through the variable.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: lib: Support team devices
Petr Machata [Mon, 30 Jul 2018 14:39:52 +0000 (16:39 +0200)]
selftests: forwarding: lib: Support team devices

Add team_create() and team_destroy() to manage team netdevices.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: lib: Add require_command()
Petr Machata [Mon, 30 Jul 2018 14:39:46 +0000 (16:39 +0200)]
selftests: forwarding: lib: Add require_command()

The logic for testing whether a certain command is available is used
several times in the current code base. The tests in follow-up patches
add more requirements like that.

Therefore extract the logic into a named function, require_command(),
that can be used directly from lib.sh as well as from any test that
wishes to declare dependence on some command.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agofib_rules: NULL check before kfree is not needed
YueHaibing [Mon, 30 Jul 2018 13:07:24 +0000 (21:07 +0800)]
fib_rules: NULL check before kfree is not needed

kfree(NULL) is safe,so this removes NULL check before freeing the mem

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: mscc: the extended page access register is 16 bits
Quentin Schulz [Mon, 30 Jul 2018 12:53:13 +0000 (14:53 +0200)]
net: phy: mscc: the extended page access register is 16 bits

The Extended Page Access is a 16-bit register, so change the page
parameter of vsc85xx_phy_page_set to a u16.

Signed-off-by: Quentin Schulz <quentin.schulz@bootlin.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: Use socket data_ready callback on record availability
Vakul Garg [Mon, 30 Jul 2018 10:38:33 +0000 (16:08 +0530)]
net/tls: Use socket data_ready callback on record availability

On receipt of a complete tls record, use socket's saved data_ready
callback instead of state_change callback. In function tls_queue(),
the TLS record is queued in encrypted state. But the decryption
happen inline when tls_sw_recvmsg() or tls_sw_splice_read() get invoked.
So it should be ok to notify the waiting context about the availability
of data as soon as we could collect a full TLS record. For new data
availability notification, sk_data_ready callback is more appropriate.
It points to sock_def_readable() which wakes up specifically for EPOLLIN
event. This is in contrast to the socket callback sk_state_change which
points to sock_def_wakeup() which issues a wakeup unconditionally
(without event mask).

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'TC-refactor-act_mirred-packets-re-injection'
David S. Miller [Mon, 30 Jul 2018 16:31:14 +0000 (09:31 -0700)]
Merge branch 'TC-refactor-act_mirred-packets-re-injection'

Paolo Abeni says:

====================
TC: refactor act_mirred packets re-injection

This series is aimed at improving the act_mirred redirect performances.
Such action is used by OVS to represent TC S/W flows, and it's current largest
bottle-neck is the need for a skb_clone() for each packet.

The first 2 patches introduce some cleanup and safeguards to allow extending
tca_result - we will use it to store RCU protected redirect information - and
introduce a clear separation between user-space accessible tcfa_action
values and internal values accessible only by the kernel.
Then a new tcfa_action value is introduced: TC_ACT_REINJECT, similar to
TC_ACT_REDIRECT, but preserving the mirred semantic. Such value is not
accessible from user-space.
The last patch exploits the newly introduced infrastructure in the act_mirred
action, to avoid a skb_clone, when possible.

Overall this the above gives a ~10% performance improvement in forwarding tput,
when using the TC S/W datapath.

v1 -> v2:
 - preserve the rcu lock in act_bpf
 - add and use a new action value to reinject the packets, preserving the mirred
   semantic

v2 -> v3:
 - renamed to new action as TC_ACT_REINJECT
 - TC_ACT_REINJECT is not exposed to user-space

v3 -> v4:
 - dropped the TC_ACT_REDIRECT patch
 - report failure via extack, too
 - rename the new action as TC_ACT_REINSERT
 - skip clone only if the control action don't touch tcf_result

v4 -> v5:
 - fix a couple of build issue reported by kbuild bot
 - dont split messages
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoact_mirred: use TC_ACT_REINSERT when possible
Paolo Abeni [Mon, 30 Jul 2018 12:30:45 +0000 (14:30 +0200)]
act_mirred: use TC_ACT_REINSERT when possible

When mirred is invoked from the ingress path, and it wants to redirect
the processed packet, it can now use the TC_ACT_REINSERT action,
filling the tcf_result accordingly, and avoiding a per packet
skb_clone().

Overall this gives a ~10% improvement in forwarding performance for the
TC S/W data path and TC S/W performances are now comparable to the
kernel openvswitch datapath.

v1 -> v2: use ACT_MIRRED instead of ACT_REDIRECT
v2 -> v3: updated after action rename, fixed typo into the commit
message
v3 -> v4: updated again after action rename, added more comments to
the code (JiriP), skip the optimization if the control action
need to touch the tcf_result (Paolo)
v4 -> v5: fix sparse warning (kbuild bot)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tc: introduce TC_ACT_REINSERT.
Paolo Abeni [Mon, 30 Jul 2018 12:30:44 +0000 (14:30 +0200)]
net/tc: introduce TC_ACT_REINSERT.

This is similar TC_ACT_REDIRECT, but with a slightly different
semantic:
- on ingress the mirred skbs are passed to the target device
network stack without any additional check not scrubbing.
- the rcu-protected stats provided via the tcf_result struct
  are updated on error conditions.

This new tcfa_action value is not exposed to the user-space
and can be used only internally by clsact.

v1 -> v2: do not touch TC_ACT_REDIRECT code path, introduce
 a new action type instead
v2 -> v3:
 - rename the new action value TC_ACT_REINJECT, update the
   helper accordingly
 - take care of uncloned reinjected packets in XDP generic
   hook
v3 -> v4:
 - renamed again the new action value (JiriP)
v4 -> v5:
 - fix build error with !NET_CLS_ACT (kbuild bot)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotc/act: remove unneeded RCU lock in action callback
Paolo Abeni [Mon, 30 Jul 2018 12:30:43 +0000 (14:30 +0200)]
tc/act: remove unneeded RCU lock in action callback

Each lockless action currently does its own RCU locking in ->act().
This allows using plain RCU accessor, even if the context
is really RCU BH.

This change drops the per action RCU lock, replace the accessors
with the _bh variant, cleans up a bit the surrounding code and
documents the RCU status in the relevant header.
No functional nor performance change is intended.

The goal of this patch is clarifying that the RCU critical section
used by the tc actions extends up to the classifier's caller.

v1 -> v2:
 - preserve rcu lock in act_bpf: it's needed by eBPF helpers,
   as pointed out by Daniel

v3 -> v4:
 - fixed some typos in the commit message (JiriP)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/sched: user-space can't set unknown tcfa_action values
Paolo Abeni [Mon, 30 Jul 2018 12:30:42 +0000 (14:30 +0200)]
net/sched: user-space can't set unknown tcfa_action values

Currently, when initializing an action, the user-space can specify
and use arbitrary values for the tcfa_action field. If the value
is unknown by the kernel, is implicitly threaded as TC_ACT_UNSPEC.

This change explicitly checks for unknown values at action creation
time, and explicitly convert them to TC_ACT_UNSPEC. No functional
changes are introduced, but this will allow introducing tcfa_action
values not exposed to user-space in a later patch.

Note: we can't use the above to hide TC_ACT_REDIRECT from user-space,
as the latter is already part of uAPI.

v3 -> v4:
 - use an helper to check for action validity (JiriP)
 - emit an extack for invalid actions (JiriP)
v4 -> v5:
 - keep messages on a single line, drop net_warn (Marcelo)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoliquidio: remove redundant function cn23xx_dump_iq_regs
YueHaibing [Mon, 30 Jul 2018 11:03:41 +0000 (19:03 +0800)]
liquidio: remove redundant function cn23xx_dump_iq_regs

There are no in-tree callers of cn23xx_dump_iq_regs.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'socket-poll-related-cleanups-v2'
David S. Miller [Mon, 30 Jul 2018 16:10:25 +0000 (09:10 -0700)]
Merge branch 'socket-poll-related-cleanups-v2'

Christoph Hellwig says:

====================
socket poll related cleanups v2

A couple of cleanups I stumbled upon when studying the networking
poll code.

Changes since v1:
 - drop a dispute patch from this series (to be sent separately)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: remove sock_poll_busy_flag
Christoph Hellwig [Mon, 30 Jul 2018 07:42:13 +0000 (09:42 +0200)]
net: remove sock_poll_busy_flag

Fold it into the only caller to make the code simpler and easier to read.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: remove sock_poll_busy_loop
Christoph Hellwig [Mon, 30 Jul 2018 07:42:12 +0000 (09:42 +0200)]
net: remove sock_poll_busy_loop

There is no point in hiding this logic in a helper.  Also remove the
useless events != 0 check and only busy loop once we know we actually
have a poll method.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: don not detour through struct sock to find the poll waitqueue
Christoph Hellwig [Mon, 30 Jul 2018 07:42:11 +0000 (09:42 +0200)]
net: don not detour through struct sock to find the poll waitqueue

For any open socket file descriptor sock->sk->sk_wq->wait will always
point to sock->wq->wait.  That means we can do the shorter dereference
and removal a NULL check and don't have to not worry about any RCU
protection.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: simplify sock_poll_wait
Christoph Hellwig [Mon, 30 Jul 2018 07:42:10 +0000 (09:42 +0200)]
net: simplify sock_poll_wait

The wait_address argument is always directly derived from the filp
argument, so remove it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoact_bpf: Use kmemdup instead of duplicating it in tcf_bpf_init_from_ops
YueHaibing [Sat, 28 Jul 2018 10:38:06 +0000 (18:38 +0800)]
act_bpf: Use kmemdup instead of duplicating it in tcf_bpf_init_from_ops

Replace calls to kmalloc followed by a memcpy with a direct call to
kmemdup.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocls_bpf: Use kmemdup instead of duplicating it in cls_bpf_prog_from_ops
YueHaibing [Sat, 28 Jul 2018 10:35:15 +0000 (18:35 +0800)]
cls_bpf: Use kmemdup instead of duplicating it in cls_bpf_prog_from_ops

Replace calls to kmalloc followed by a memcpy with a direct call to
kmemdup.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoact_pedit: remove unnecessary semicolon
YueHaibing [Sat, 28 Jul 2018 10:29:01 +0000 (18:29 +0800)]
act_pedit: remove unnecessary semicolon

net/sched/act_pedit.c:289:2-3: Unneeded semicolon

Remove unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: remove redundant functions qed_get_cm_pq_idx_rl
YueHaibing [Sat, 28 Jul 2018 10:23:36 +0000 (18:23 +0800)]
qed: remove redundant functions qed_get_cm_pq_idx_rl

There are no in-tree callers of qed_get_cm_pq_idx_rl since it be there,
so it can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet-next: mediatek: cleanup unnecessary get chip id and its user
Sean Wang [Sat, 28 Jul 2018 05:35:56 +0000 (13:35 +0800)]
net-next: mediatek: cleanup unnecessary get chip id and its user

Since driver is devicetree-based, all device type and charateristic can be
determined by the compatible string and its data. It's unnecessary to
create another dependent function to check chip ID and then decide whether
the specific funciton is being supported on a certain device. It can be
totally replaced by the existing flag, so a cleanup is made by removing
the function and the only user, HWLRO.

MT2701 also have a missing HWLRO support in old code, so add it the same
patch.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet-next: mediatek: improve more with using dma_zalloc_coherent
Sean Wang [Sat, 28 Jul 2018 05:35:55 +0000 (13:35 +0800)]
net-next: mediatek: improve more with using dma_zalloc_coherent

Improve more in the existing code by reusing dma_zalloc_coherent instead
of dma_alloc_coherent with __GFP_ZERO or superfluous zeroing buffer.

Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosysfs: Fix regression when adding a file to an existing group
Tyler Hicks [Fri, 27 Jul 2018 21:33:27 +0000 (21:33 +0000)]
sysfs: Fix regression when adding a file to an existing group

Commit 5f81880d5204 ("sysfs, kobject: allow creating kobject belonging
to arbitrary users") incorrectly changed the argument passed as the
parent parameter when calling sysfs_add_file_mode_ns(). This caused some
sysfs attribute files to not be added correctly to certain groups.

Fixes: 5f81880d5204 ("sysfs, kobject: allow creating kobject belonging to arbitrary users")
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Heiner Kallweit <hkallweit1@gmail.com>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'mlx5e-updates-2018-07-27' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Sun, 29 Jul 2018 20:08:42 +0000 (13:08 -0700)]
Merge tag 'mlx5e-updates-2018-07-27' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5e-updates-2018-07-27 (Vxlan updates)

This series from Gal and Saeed provides updates to mlx5 vxlan implementation.

Gal, started with three cleanups to reflect the actual hardware vxlan state
- reflect 4789 UDP port default addition to software database
- check maximum number of vxlan  UDP ports
- cleanup an unused member in vxlan work

Then Gal provides performance optimization by replacing the
vxlan radix tree with a hash table.

Measuring mlx5e_vxlan_lookup_port execution time:

                      Radix Tree   Hash Table
     --------------- ------------ ------------
      Single Stream   161 ns       79  ns (51% improvement)
      Multi Stream    259 ns       136 ns (47% improvement)

    Measuring UDP stream packet rate, single fully utilized TX core:
    Radix Tree: 498,300 PPS
    Hash Table: 555,468 PPS (11% improvement)

Next, from Saeed, vxlan refactoring to allow sharing the vxlan table
between different mlx5 netdevice instances like PF and VF representors,
this is done by making mlx5 vxlan interface more generic and decoupling
it from PF netdevice structures and logic, then moving it into mlx5 core
as a low level interface so it can be used by VF representors, which is
illustrated in the last patch of the serious.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: mlxsw: qos_dscp_bridge: Fix
Petr Machata [Fri, 27 Jul 2018 22:48:13 +0000 (00:48 +0200)]
selftests: mlxsw: qos_dscp_bridge: Fix

There are two problems in this test case:

- When indexing in bash associative array, the subscript is interpreted as
  string, not as a variable name to be expanded.

- The keys stored to t0s and t1s are not DSCP values, but priority +
  base (i.e. the logical DSCP value, not the full bitfield value).

In combination these two bugs conspire to make the test just work,
except it doesn't really test anything and always passes.

Fix the above two problems in obvious manner.

Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mtu-related-changes'
David S. Miller [Sun, 29 Jul 2018 19:57:26 +0000 (12:57 -0700)]
Merge branch 'mtu-related-changes'

Stephen Hemminger says:

====================
mtu related changes

While looking at other MTU issues, noticed a couple oppurtunties
for improving user experience.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: report invalid mtu value via netlink extack
Stephen Hemminger [Fri, 27 Jul 2018 20:43:23 +0000 (13:43 -0700)]
net: report invalid mtu value via netlink extack

If an invalid MTU value is set through rtnetlink return extra error
information instead of putting message in kernel log. For other cases
where there is no visible API, keep the error report in the log.

Example:
# ip li set dev enp12s0 mtu 10000
Error: mtu greater than device maximum.

# ifconfig enp12s0 mtu 10000
SIOCSIFMTU: Invalid argument
# dmesg | tail -1
[ 2047.795467] enp12s0: mtu greater than device maximum

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: report min and max mtu network device settings
Stephen Hemminger [Fri, 27 Jul 2018 20:43:22 +0000 (13:43 -0700)]
net: report min and max mtu network device settings

Report the minimum and maximum MTU allowed on a device
via netlink so that it can be displayed by tools like
ip link.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agofailover: change mtu has RTNL
Stephen Hemminger [Fri, 27 Jul 2018 20:43:21 +0000 (13:43 -0700)]
failover: change mtu has RTNL

When changing MTU, RTNL is held so use rtnl_dereference
instead of rcu_dereference.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dcb: add DSCP to comment about priority selector types
Jakub Kicinski [Fri, 27 Jul 2018 20:11:00 +0000 (13:11 -0700)]
net: dcb: add DSCP to comment about priority selector types

Commit ee2059819450 ("net/dcb: Add dscp to priority selector type")
added a define for the new DSCP selector type created by
IEEE 802.1Qcd, but missed the comment enumerating all selector types.
Update the comment.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: ti: cpsw: add missed RX_CTAG feature for second slave
Ivan Khoronzhuk [Fri, 27 Jul 2018 16:54:39 +0000 (19:54 +0300)]
net: ethernet: ti: cpsw: add missed RX_CTAG feature for second slave

Seems it was missed while adding for first net dev in dual-emac mode.

Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'route-add-support-and-selftests-for-directed-broadcast-forwarding'
David S. Miller [Sun, 29 Jul 2018 19:37:06 +0000 (12:37 -0700)]
Merge branch 'route-add-support-and-selftests-for-directed-broadcast-forwarding'

Xin Long says:

====================
route: add support and selftests for directed broadcast forwarding

Patch 1/2 is the feature and 2/2 is the selftest. Check the changelog
on each of them to know the details.

v1->v2:
  - fix a typo in changelog.
  - fix an uapi break that Davide noticed.
  - flush route cache when bc_forwarding is changed.
  - add the selftest for this patch as Ido's suggestion.

v2->v3:
  - fix an incorrect 'if check' in devinet_conf_proc as David Ahern
    noticed.
  - extend the selftest after one David Ahern fix for vrf.

v3->v4:
  - improve the output log in the selftest as David Ahern suggested.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: add a selftest for directed broadcast forwarding
Xin Long [Fri, 27 Jul 2018 08:37:29 +0000 (16:37 +0800)]
selftests: add a selftest for directed broadcast forwarding

As Ido's suggestion, this patch is to add a selftest for directed
broadcast forwarding with vrf. It does the assertion by checking
the src IP of the echo-reply packet in ping_test_from.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoroute: add support for directed broadcast forwarding
Xin Long [Fri, 27 Jul 2018 08:37:28 +0000 (16:37 +0800)]
route: add support for directed broadcast forwarding

This patch implements the feature described in rfc1812#section-5.3.5.2
and rfc2644. It allows the router to forward directed broadcast when
sysctl bc_forwarding is enabled.

Note that this feature could be done by iptables -j TEE, but it would
cause some problems:
  - target TEE's gateway param has to be set with a specific address,
    and it's not flexible especially when the route wants forward all
    directed broadcasts.
  - this duplicates the directed broadcasts so this may cause side
    effects to applications.

Besides, to keep consistent with other os router like BSD, it's also
necessary to implement it in the route rx path.

Note that route cache needs to be flushed when bc_forwarding is
changed.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/ipv6: allow any source address for sendmsg pktinfo with ip_nonlocal_bind
Vincent Bernat [Wed, 25 Jul 2018 11:19:13 +0000 (13:19 +0200)]
net/ipv6: allow any source address for sendmsg pktinfo with ip_nonlocal_bind

When freebind feature is set of an IPv6 socket, any source address can
be used when sending UDP datagrams using IPv6 PKTINFO ancillary
message. Global non-local bind feature was added in commit
35a256fee52c ("ipv6: Nonlocal bind") for IPv6. This commit also allows
IPv6 source address spoofing when non-local bind feature is enabled.

Signed-off-by: Vincent Bernat <vincent@bernat.im>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'linux-can-next-for-4.19-20180727' of ssh://gitolite.kernel.org/pub/scm...
David S. Miller [Sun, 29 Jul 2018 15:35:03 +0000 (08:35 -0700)]
Merge tag 'linux-can-next-for-4.19-20180727' of ssh://gitolite./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2018-01-16

this is a pull request for net-next/master consisting of 38 patches.

Dan Murphy's patch fixes the path to a file in the comment of the CAN
Error Message Frame Mask structure.

A patch by Colin Ian King fixes a typo in the cc770 driver.

The next patch is by me an sorts the Kconfigand Makefile entries of the
CAN-USB driver subdir alphabetically.

The patch by Jakob Unterwurzacher adds support for the UCAN USB-CAN
adapter.

YueHaibing's patch replaces a open coded skb_put()+memset() by
skb_put_zero() in the CAN-dev infrastructure.

Zhu Yi provides a patch to enable multi-queue CAN devices.

Three patches by Luc Van Oostenryck fix the return value of several
driver's xmit function, I contribute a patch for the a fourth driver.

Fabio Estevam's patch switches the flexcan driver to SPDX identifier.

Two patches by Jia-Ju Bai replace the mdelay() by a usleep_range() in
the sja1000 drivers.

The next 6 patches are by Anssi Hannula and refactor the xilinx CAN
driver and add support for the xilinx CAN FD core.

A patch by Gustavo A. R. Silva adds fallthrough annotation to the
peak_usb driver.

5 patches by Stephane Grosjean for the peak CANFD driver do some
cleanups and provide more improvements for further firmware releases.

The remaining 13 patches are by Jimmy Assarsson and the first clean up
the kvaser_usb driver, so that the later patches add support for the
Kvaser USB hydra family.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqed: remove redundant functions qed_set_gft_event_id_cm_hdr
YueHaibing [Fri, 27 Jul 2018 13:24:27 +0000 (21:24 +0800)]
qed: remove redundant functions qed_set_gft_event_id_cm_hdr

There are no in-tree callers of qed_set_gft_event_id_cm_hdr.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoliquidio: remove redundant function cn23xx_dump_vf_iq_regs
YueHaibing [Fri, 27 Jul 2018 11:57:24 +0000 (19:57 +0800)]
liquidio: remove redundant function cn23xx_dump_vf_iq_regs

There are no in-tree callers.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'tls-Fix-improper-revert-in-zerocopy_from_iter'
David S. Miller [Sun, 29 Jul 2018 05:53:31 +0000 (22:53 -0700)]
Merge branch 'tls-Fix-improper-revert-in-zerocopy_from_iter'

Doron Roberts-Kedes says:

====================
tls: Fix improper revert in zerocopy_from_iter

This series fixes the improper iov_iter_revert introcded in
"tls: Fix zerocopy_from_iter iov handling".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotls: Fix improper revert in zerocopy_from_iter
Doron Roberts-Kedes [Thu, 26 Jul 2018 14:59:36 +0000 (07:59 -0700)]
tls: Fix improper revert in zerocopy_from_iter

The current code is problematic because the iov_iter is reverted and
never advanced in the non-error case. This patch skips the revert in the
non-error case. This patch also fixes the amount by which the iov_iter
is reverted. Currently, iov_iter is reverted by size, which can be
greater than the amount by which the iter was actually advanced.
Instead, only revert by the amount that the iter was advanced.

Fixes: 4718799817c5 ("tls: Fix zerocopy_from_iter iov handling")
Signed-off-by: Doron Roberts-Kedes <doronrk@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotls: Remove dead code in tls_sw_sendmsg
Doron Roberts-Kedes [Thu, 26 Jul 2018 14:59:35 +0000 (07:59 -0700)]
tls: Remove dead code in tls_sw_sendmsg

tls_push_record either returns 0 on success or a negative value on failure.
This patch removes code that would only be executed if tls_push_record
were to return a positive value.

Signed-off-by: Doron Roberts-Kedes <doronrk@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mvneta-next'
David S. Miller [Sun, 29 Jul 2018 05:12:55 +0000 (22:12 -0700)]
Merge branch 'mvneta-next'

Gregory CLEMENT says:

====================
A fix and a few improvements on mvneta

This series brings some improvements for the mvneta driver and also
adds a fix.

Compared to the v2, the main change is another patch fixing a bug
in mtu_change.

Changelog:
v1 -> v2

 - In patch 2, use EXPORT_SYMBOL_GPL for mvneta_bm_get and
   mvneta_bm_put to be used in module, reported by kbuild test robot.

 - In patch 4, add the counter to the driver's ethtool state,
   suggested by David Miller.

 - In patch 6, use a single if, suggested by Marcin Wojtas

v2 -> v3

 - Adding a patch fixing the mtu change issue

 - Removing the inline keyword for mvneta_rx_refill() and let the
   comiler decided, suggested by David Miller.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: Improve the buffer allocation method for SWBM
Yelena Krivosheev [Wed, 18 Jul 2018 16:10:57 +0000 (18:10 +0200)]
net: mvneta: Improve the buffer allocation method for SWBM

With system having a small memory (around 256MB), the state "cannot
allocate memory to refill with new buffer" is reach pretty quickly.

By this patch we changed buffer allocation method to a better handling of
this use case by avoiding memory allocation issues.

Signed-off-by: Yelena Krivosheev <yelena@marvell.com>
[gregory: extract from a larger patch]
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: Verify hardware checksum only when offload checksum feature is set
Yelena Krivosheev [Wed, 18 Jul 2018 16:10:56 +0000 (18:10 +0200)]
net: mvneta: Verify hardware checksum only when offload checksum feature is set

If the checksum offload feature is not set, then there is no point to
check the status of the hardware.

[gregory: extract from a larger patch]
Signed-off-by: Yelena Krivosheev <yelena@marvell.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: Allocate page for the descriptor
Gregory CLEMENT [Wed, 18 Jul 2018 16:10:55 +0000 (18:10 +0200)]
net: mvneta: Allocate page for the descriptor

Instead of trying to allocate the exact amount of memory for each
descriptor use a page for each of them, it allows to simplify the
allocation management and increase the performance of the driver.

Based on the work of Yelena Krivosheev <yelena@marvell.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: discriminate error cause for missed packet
Gregory CLEMENT [Wed, 18 Jul 2018 16:10:54 +0000 (18:10 +0200)]
net: mvneta: discriminate error cause for missed packet

In order to improve the diagnostic in case of error, make the distinction
between refill error and skb allocation error. Also make the information
available through the ethtool state.

Based on the work of Yelena Krivosheev <yelena@marvell.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: increase number of buffers in RX and TX queue
Yelena Krivosheev [Wed, 18 Jul 2018 16:10:53 +0000 (18:10 +0200)]
net: mvneta: increase number of buffers in RX and TX queue

The initial values were too small leading to poor performance when using
the software buffer management.

Signed-off-by: Yelena Krivosheev <yelena@marvell.com>
[gregory: extract from a larger patch]
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: remove data pointer usage from device_node structure
Gregory CLEMENT [Wed, 18 Jul 2018 16:10:52 +0000 (18:10 +0200)]
net: mvneta: remove data pointer usage from device_node structure

On year ago Rob Herring wanted to remove the data pointer from the
device_node structure[1]. The mvneta driver seemed to be the only one
which used (abused ?) it. However, the proposal of Rob to remove this
pointer from the driver introduced a regression, and I tested and fixed an
alternative way, but it was never submitted as a proper patch.

Now here it is: Instead of using the device_node structure ->data
pointer, we store the BM private data as the driver data of the BM
platform_device. The core mvneta code can retrieve it by doing a lookup
on which platform_device corresponds to the BM device tree node using
of_find_device_by_node(), and get its driver data

[1]https://www.spinics.net/lists/netdev/msg445197.html

Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: fix mtu change on port without link
Yelena Krivosheev [Wed, 18 Jul 2018 16:10:51 +0000 (18:10 +0200)]
net: mvneta: fix mtu change on port without link

It is incorrect to enable TX/RX queues (call by mvneta_port_up()) for
port without link. Indeed MTU change for interface without link causes TX
queues to stuck.

Fixes: c5aff18204da ("net: mvneta: driver for Marvell Armada 370/XP
network unit")
Signed-off-by: Yelena Krivosheev <yelena@marvell.com>
[gregory.clement: adding Fixes tags and rewording commit log]
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: mvneta: Fix napi structure mixup on armada 3700
Andrew Lunn [Wed, 18 Jul 2018 16:10:50 +0000 (18:10 +0200)]
net: ethernet: mvneta: Fix napi structure mixup on armada 3700

The mvneta Ethernet driver is used on a few different Marvell SoCs.
Some SoCs have per cpu interrupts for Ethernet events. Some SoCs have
a single interrupt, independent of the CPU. The driver handles this by
having a per CPU napi structure when there are per CPU interrupts, and
a global napi structure when there is a single interrupt.

When the napi core calls mvneta_poll(), it passes the napi
instance. This was not being propagated through the call chain, and
instead the per-cpu napi instance was passed to napi_gro_receive()
call. This breaks when there is a single global napi instance.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Fixes: 2636ac3cc2b4 ("net: mvneta: Add network support for Armada 3700 SoC")
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx5e: Issue direct lookup on vxlan ports by vport representors
Saeed Mahameed [Sat, 19 May 2018 12:34:48 +0000 (05:34 -0700)]
net/mlx5e: Issue direct lookup on vxlan ports by vport representors

Remove uplink representor netdevice private structure lookup, and use
mlx5 core handle directly from representor private structure to lookup
vxlan ports.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, move vxlan logic to core driver
Saeed Mahameed [Wed, 9 May 2018 20:28:00 +0000 (13:28 -0700)]
net/mlx5e: Vxlan, move vxlan logic to core driver

Move vxlan logic and objects to mlx5 core dirver.
Since it going to be used from different mlx5 interfaces.
e.g. mlx5e PF NIC netdev and mlx5e E-Switch representors.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, add sync lock for add/del vxlan port
Saeed Mahameed [Tue, 8 May 2018 23:17:06 +0000 (16:17 -0700)]
net/mlx5e: Vxlan, add sync lock for add/del vxlan port

Vxlan API can and will be called from different mlx5 modules, we should
not count on mlx5e private state lock only, hence we introduce a vxlan
private mutex to sync between add/del vxlan port operations.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, return values for add/del port
Saeed Mahameed [Tue, 8 May 2018 21:33:19 +0000 (14:33 -0700)]
net/mlx5e: Vxlan, return values for add/del port

For a better API mlx5_vxlan_{add/del}_port can fail, make them return
error values.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, rename from mlx5e to mlx5
Saeed Mahameed [Tue, 8 May 2018 09:51:23 +0000 (02:51 -0700)]
net/mlx5e: Vxlan, rename from mlx5e to mlx5

Rename vxlan functions from mlx5e_vxlan_* to mlx5_vxlan_*.
Rename mlx5e_vxlan_db to mlx5_vxlan and move it from en.h to vxlan.c
since it is not related to mlx5e anymore.

Allocate mlx5_vxlan structure dynamically in order to make it easier to
move later to core driver and to make it private in vxlan.c.

This is in preparation to move vxlan API to mlx5 core.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, rename struct mlx5e_vxlan to mlx5_vxlan_port
Saeed Mahameed [Tue, 8 May 2018 09:23:49 +0000 (02:23 -0700)]
net/mlx5e: Vxlan, rename struct mlx5e_vxlan to mlx5_vxlan_port

The name mlx5e_vxlan will be used in downstream patch to describe
mlx5 vxlan structure that will replace mlx5e_vxlan_db.

Hence we rename struct mlx5e_vxlan to mlx5_vxlan_port which describes a
mlx5 vxlan port.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, move netdev only logic to en_main.c
Saeed Mahameed [Tue, 8 May 2018 08:49:43 +0000 (01:49 -0700)]
net/mlx5e: Vxlan, move netdev only logic to en_main.c

Create a direct vxlan API to add and delete vxlan ports from HW.
+void mlx5e_vxlan_add_port(struct mlx5e_priv *priv, u16 port);
+void mlx5e_vxlan_del_port(struct mlx5e_priv *priv, u16 port);

And move vxlan_add/del_work to en_main.c since they are netdev only
logic.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, add direct delete function
Saeed Mahameed [Tue, 8 May 2018 01:06:26 +0000 (18:06 -0700)]
net/mlx5e: Vxlan, add direct delete function

Add direct vxlan delete function to be called from vxlan_delete_work.
Needed in downstream patch.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
5 years agonet/mlx5e: Vxlan, cleanup an unused member in vxlan work
Gal Pressman [Tue, 13 Feb 2018 08:31:26 +0000 (10:31 +0200)]
net/mlx5e: Vxlan, cleanup an unused member in vxlan work

Cleanup the sa_family member of the vxlan work, it is unused/needed
anywhere in the code.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Vxlan, replace ports radix-tree with hash table
Gal Pressman [Tue, 26 Dec 2017 16:27:08 +0000 (18:27 +0200)]
net/mlx5e: Vxlan, replace ports radix-tree with hash table

The VXLAN database is accessed in the data path for each VXLAN TX skb in
order to check whether the UDP port is being offloaded or not.
The number of elements in the database is relatively small, we can
simplify the radix-tree to a hash table and speedup the lookup process.

Measuring mlx5e_vxlan_lookup_port execution time:

                  Radix Tree   Hash Table
 --------------- ------------ ------------
  Single Stream   161 ns       79  ns (51% improvement)
  Multi Stream    259 ns       136 ns (47% improvement)

Measuring UDP stream packet rate, single fully utilized TX core:
Radix Tree: 498,300 PPS
Hash Table: 555,468 PPS (11% improvement)

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Vxlan, check maximum number of UDP ports
Gal Pressman [Mon, 25 Dec 2017 16:40:52 +0000 (18:40 +0200)]
net/mlx5e: Vxlan, check maximum number of UDP ports

The NIC has a limited number of offloaded VXLAN UDP ports (usually 4).
Instead of letting the firmware fail when trying to add more ports than
it can handle, let the driver check it on its own.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Vxlan, reflect 4789 UDP port default addition to software database
Gal Pressman [Wed, 17 Jan 2018 09:02:31 +0000 (11:02 +0200)]
net/mlx5e: Vxlan, reflect 4789 UDP port default addition to software database

The hardware offloads 4789 UDP port (default VXLAN port) automatically.
Add it to the software database as well in order to reflect the hardware
state appropriately.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>