For the static and progress context params WQEs, do the edge
filling separately.
This improves the WQ utilization, code readability, and reduces
the chance of future bugs.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

After previous modifications, the offloads are no longer called one by
one, the pi is calculated and the wqe is cleared on between of TLS and
IPSEC offloads, which doesn't quite fit mlx5e_accel_handle_tx's purpose.

This patch splits mlx5e_accel_handle_tx into two functions that
correspond to two logical phases of running offloads:

1. Before fetching a WQE. Here runs the code that can post WQEs on its
own, before the main WQE is fetched. It's the main part of TLS offload.

2. After fetching a WQE. Here runs the code that updates the WQE's
fields, but can't post other WQEs any more. It's a minor part of TLS
offload that sets the tisn field in the cseg, and eseg-based offloads
(currently IPSEC, and later patches will move GENEVE and checksum
offloads there, too).

It allows to make mlx5e_xmit take care of all actions needed to transmit
a packet in the right order, improve the structure of the code and
reduce unnecessary operations. The structure will be further improved in
the following patches (all eseg-based offloads will be moved to a single
place, and reserving space for the main WQE will happen between phase 1
and phase 2 of offloads to eliminate unneeded data movements).
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

mlx5e_udp_gso_handle_tx_skb updates the length field in the UDP header
in case of GSO. It doesn't interfere with other offloads, so do it first
to simplify further restructuring of the code. This way we'll make all
independent modifications to the SKB before starting to work with WQEs.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Raed Salem <raeds@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

TLS offload may write a 32-bit field (tisn) to the cseg of the WQE. To
do that, it receives pi and wqe pointers. As TLS offload may also send
additional WQEs, it has to update pi and wqe, and in many cases it even
doesn't use pi calculated before and wqe zeroed before and does it
itself. Also, mlx5e_sq_xmit has to copy the whole cseg if it goes to the
mlx5e_fill_sq_frag_edge flow. This all is not efficient.

It's more efficient to do the following:

1. Just return tisn from TLS offload and make the caller fill it in a
more appropriate place.

2. Calculate pi and clear wqe after calling TLS offload.

3. If TLS offload has to send WQEs, calculate pi and clear wqe just
before that. It's already done in all places anyway, so this commit
allows to remove some redundant memsets and calls.

Copying of cseg will be eliminated in one of the following commits, and
all other stuff is done here.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

IPSEC offload needs to modify the eseg of the WQE that is being filled,
but it receives a pointer to the whole WQE. To make the contract
stricter, pass only the pointer to the eseg of that WQE. This commit is
preparation for the following refactoring of offloads in the TX path and
for the MPWQE support.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

mlx5e_sq_xmit and mlx5i_sq_xmit always return NETDEV_TX_OK. Drop the
return value to simplify the code.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Both INNOVA and ConnectX TLS offloads perform the same checks in the
beginning. Unify them to reduce repeating code. Do WARN_ON_ONCE on
netdev mismatch and finish with an error in both offloads, not only in
the ConnectX one.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

TLS and IPSEC offloads currently return struct sk_buff *, but the value
is either NULL or the same skb that was passed as a parameter. Return
bool instead to provide stronger guarantees to the calling code (it
won't need to support handling a different SKB that could be potentially
returned before this change) and to simplify restructuring this code in
the following commits.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

This merge includes updates to bonding driver needed for the rdma stack,
to avoid conflicts with the RDMA branch.

Maor Gottlieb Says:

====================
Bonding: Add support to get xmit slave

The following series adds support to get the LAG master xmit slave by
introducing new .ndo - ndo_get_xmit_slave. Every LAG module can
implement it and it first implemented in the bond driver.
This is follow-up to the RFC discussion [1].

The main motivation for doing this is for drivers that offload part
of the LAG functionality. For example, Mellanox Connect-X hardware
implements RoCE LAG which selects the TX affinity when the resources
are created and port is remapped when it goes down.

The first part of this patchset introduces the new .ndo and add the
support to the bonding module.

The second part adds support to get the RoCE LAG xmit slave by building
skb of the RoCE packet based on the AH attributes and call to the new
.ndo.

The third part change the mlx5 driver driver to set the QP's affinity
port according to the slave which found by the .ndo.
====================
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

The variable tx_done is being assigned with a value that is never read
as the function returns a few statements later.  The assignment is
redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The variable size is being assigned with a value that is never read,
the assignment is redundant and cab be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The variable ret is being assigned with a value that is never read,
the assignment is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Currently the value for 'off' is computed using a multiplication and
a couple of statements later off is being incremented by len and
this value is never read.  Clean up the code by removing the
multiplication and just increment off by len on each iteration.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Heiner Kallweit says:

====================
r8169: sync few functionalities with vendor driver

Add few helpers (with names copied from vendor drivers) to make clearer
what the respective code is doing. In addition improve reset preparation
for chips from RTL8168g.
====================
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Sync the reset preparation for chips from RTL8168g with the r8168 and
r8125 vendor drivers.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Add a helper for waiting for FIFO's to be empty, again the name is
borrowed from the vendor driver.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Add a helper for setting RXDV_GATED_EN, the 2ms delay is copied from
the vendor driver.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Create a helper for this waiting function, name of the helper is
borrowed from the vendor driver. In addition don't return in the two
hw_init functions if the first wait runs into a timeout, there's no
benefit in doing so.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

When building arm64 allyesconfig:

drivers/net/ipa/ipa_endpoint.c: In function 'ipa_endpoint_stop_rx_dma':
drivers/net/ipa/ipa_endpoint.c:1274:13: error: 'IPA_ENDPOINT_STOP_RX_SIZE' undeclared (first use in this function)
drivers/net/ipa/ipa_endpoint.c:1274:13: note: each undeclared identifier is reported only once for each function it appears in
drivers/net/ipa/ipa_endpoint.c:1289:2: error: implicit declaration of function 'ipa_cmd_dma_task_32b_addr_add' [-Werror=implicit-function-declaration]
drivers/net/ipa/ipa_endpoint.c:1291:45: error: 'ENDPOINT_STOP_DMA_TIMEOUT' undeclared (first use in this function)
drivers/net/ipa/ipa_endpoint.c: In function 'ipa_endpoint_stop':
drivers/net/ipa/ipa_endpoint.c:1309:16: error: 'IPA_ENDPOINT_STOP_RX_RETRIES' undeclared (first use in this function)

These functions were removed in a series, merged in as
commit 33395f4a ("Merge branch 'net-ipa-kill-endpoint-stop-workaround'").

Remove them again so that the build works properly.

Fixes: 3793faad ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net")
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

We currently have to adjust ipv6 route gc_thresh/max_size depending
on number of cpus on a server, this makes very little sense.

If the kernels sets /proc/sys/net/ipv6/route/gc_thresh to 1024
and /proc/sys/net/ipv6/route/max_size to 4096, then we better
not track the percpu dst that our implementation uses.

Only routes not added (directly or indirectly) by the admin
should be tracked and limited.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: David Ahern <dsahern@kernel.org>
Cc: Maciej Żenczykowski <maze@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/net/dsa/sja1105/sja1105_vl.c:468:6: warning: variable ‘prev_time’ set but not used [-Wunused-but-set-variable]
  u32 prev_time = 0;
      ^~~~~~~~~
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Samuel Zou <zou_wei@huawei.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Use the helper function that wraps the calls to platform_get_resource()
and devm_ioremap_resource() together.
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The type of dispatch is u8 which is always '<=' 0xff, so the
dispatch <= 0xff is always true, we can remove this comparison.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

percpu_counter_add() uses a default batch size which is quite big
on platforms with 256 cpus. (2*256 -> 512)

This means dst_entries_get_fast() can be off by +/- 2*(nr_cpus^2)
(131072 on servers with 256 cpus)

Reduce the batch size to something more reasonable, and
add logic to ip6_dst_gc() to call dst_entries_get_slow()
before calling the _very_ expensive fib6_run_gc() function.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Now sch_fq has horizon feature, we want to allow QUIC/UDP applications
to use EDT model so that pacing can be offloaded to the kernel (sch_fq)
or the NIC.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
bonding: report transmit status to callers

First patches cleanup netpoll, and make sure it provides tx status to its users.

Last patch changes bonding to not pretend packets were sent without error.

By providing more accurate status, TCP stack can avoid adding more
packets if the slave qdisc is already full.

This came while testing latest horizon feature in sch_fq, with
very low pacing rate flows, but should benefit hosts under stress.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, bonding always returns NETDEV_TX_OK to its caller.

It is worth trying to be more accurate : TCP for instance
can have different recovery strategies if it can have more
precise status, if packet was dropped by slave qdisc.

This is especially important when host is under stress.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

netpoll_send_skb() callers seem to leak skb if
the np pointer is NULL. While this should not happen, we
can make the code more robust.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Some callers want to know if the packet has been sent or
dropped, to inform upper stacks.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There is no need to inline this helper, as we intend to add more
code in this function.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

netpoll_send_skb_on_dev() can get the device pointer directly from np->dev

Rename it to __netpoll_send_skb()

Following patch will move netpoll_send_skb() out-of-line.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The unsigned variable val is being checked for an error by checking
if it is less than zero. This can never occur because val is unsigned.
Fix this by making val a plain int.

Addresses-Coverity: ("Unsigned compared against zero")
Fixes: bdbdac76 ("ethtool: provide UAPI for PHY master/slave configuration.")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes gcc '-Wunused-but-set-variable' warning:

net/smc/smc_llc.c: In function 'smc_llc_cli_conf_link':
net/smc/smc_llc.c:753:31: warning:
 variable 'del_llc' set but not used [-Wunused-but-set-variable]
  struct smc_llc_msg_del_link *del_llc;
                               ^
net/smc/smc_llc.c: In function 'smc_llc_process_srv_delete_link':
net/smc/smc_llc.c:1311:33: warning:
 variable 'del_llc_resp' set but not used [-Wunused-but-set-variable]
    struct smc_llc_msg_del_link *del_llc_resp;
                                 ^
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

so tcp_is_sack/reno checks are removed from tcp_mark_head_lost.
Signed-off-by: zhang kai <zhangkaiheb@126.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The NL_SET_ERR_MSG_MOD macro is used to report a string describing an
error message to userspace via the netlink extended ACK structure. It
should not have a trailing newline.

Add a cocci script which catches cases where the newline marker is
present. Using this script, fix the handful of cases which accidentally
included a trailing new line.

I couldn't figure out a way to get a patch mode working, so this script
only implements context, report, and org.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Grygorii Strashko says:

====================
net: ethernet: ti: am65x-cpts: follow up dt bindings update

This series is follow update for  TI A65x/J721E Common platform time sync (CPTS)
driver [1] to implement  DT bindings review comments from
Rob Herring <robh@kernel.org> [2].
 - "reg" and "compatible" properties are made required for CPTS DT nodes which
   also required to change K3 CPSW driver to use of_platform_device_create()
   instead of of_platform_populate() for proper CPTS and MDIO initialization
 - minor DT bindings format changes
 - K3 CPTS example added to K3 MCU CPSW bindings

[1] https://lwn.net/Articles/819313/
[2] https://lwn.net/ml/linux-kernel/20200505040419.GA8509@bogus/
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Update CPTS node following DT binding update:
 - add reg and compatible properties
 - fix node name
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch follows K3 CPTS review comments from Rob Herring
<robh@kernel.org>.
 - "reg" and "compatible" properties are required now
 - minor format changes
 - K3 CPTS example added to K3 MCU CPSW bindings

Cc: Rob Herring <robh@kernel.org>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The MCU CPSW expected to populate only MDIO device, but follow up patches
will add "compatible" property to the MCU CPSW CPTS node which will cause
creation of CPTS device and MCU CPSW init failure. Hence, switch to use
of_platform_device_create() instead of of_platform_populate() for MDIO
device population.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Taehee Yoo says:

====================
hsr: hsr code refactoring

There are some unnecessary routine in the hsr module.
This patch removes these routines.

The first patch removes incorrect comment.
The second patch removes unnecessary WARN_ONCE() macro.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>