Conflicts were all overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull networking fixes from David Miller:

 1) Fix reference count leaks in various parts of batman-adv, from Xiyu
    Yang.

 2) Update NAT checksum even when it is zero, from Guillaume Nault.

 3) sk_psock reference count leak in tls code, also from Xiyu Yang.

 4) Sanity check TCA_FQ_CODEL_DROP_BATCH_SIZE netlink attribute in
    fq_codel, from Eric Dumazet.

 5) Fix panic in choke_reset(), also from Eric Dumazet.

 6) Fix VLAN accel handling in bnxt_fix_features(), from Michael Chan.

 7) Disallow out of range quantum values in sch_sfq, from Eric Dumazet.

 8) Fix crash in x25_disconnect(), from Yue Haibing.

 9) Don't pass pointer to local variable back to the caller in
    nf_osf_hdr_ctx_init(), from Arnd Bergmann.

10) Wireguard should use the ECN decap helper functions, from Toke
    Høiland-Jørgensen.

11) Fix command entry leak in mlx5 driver, from Moshe Shemesh.

12) Fix uninitialized variable access in mptcp's
    subflow_syn_recv_sock(), from Paolo Abeni.

13) Fix unnecessary out-of-order ingress frame ordering in macsec, from
    Scott Dial.

14) IPv6 needs to use a global serial number for dst validation just
    like ipv4, from David Ahern.

15) Fix up PTP_1588_CLOCK deps, from Clay McClure.

16) Missing NLM_F_MULTI flag in gtp driver netlink messages, from
    Yoshiyuki Kurauchi.

17) Fix a regression in that dsa user port errors should not be fatal,
    from Florian Fainelli.

18) Fix iomap leak in enetc driver, from Dejin Zheng.

19) Fix use after free in lec_arp_clear_vccs(), from Cong Wang.

20) Initialize protocol value earlier in neigh code paths when
    generating events, from Roman Mashak.

21) netdev_update_features() must be called with RTNL mutex in macsec
    driver, from Antoine Tenart.

22) Validate untrusted GSO packets even more strictly, from Willem de
    Bruijn.

23) Wireguard decrypt worker needs a cond_resched(), from Jason
    Donenfeld.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (111 commits)
  net: flow_offload: skip hw stats check for FLOW_ACTION_HW_STATS_DONT_CARE
  MAINTAINERS: put DYNAMIC INTERRUPT MODERATION in proper order
  wireguard: send/receive: use explicit unlikely branch instead of implicit coalescing
  wireguard: selftests: initalize ipv6 members to NULL to squelch clang warning
  wireguard: send/receive: cond_resched() when processing worker ringbuffers
  wireguard: socket: remove errant restriction on looping to self
  wireguard: selftests: use normal kernel stack size on ppc64
  net: ethernet: ti: am65-cpsw-nuss: fix irqs type
  ionic: Use debugfs_create_bool() to export bool
  net: dsa: Do not leave DSA master with NULL netdev_ops
  net: dsa: remove duplicate assignment in dsa_slave_add_cls_matchall_mirred
  net: stricter validation of untrusted gso packets
  seg6: fix SRH processing to comply with RFC8754
  net: mscc: ocelot: ANA_AUTOAGE_AGE_PERIOD holds a value in seconds, not ms
  net: dsa: ocelot: the MAC table on Felix is twice as large
  net: dsa: sja1105: the PTP_CLK extts input reacts on both edges
  selftests: net: tcp_mmap: fix SO_RCVLOWAT setting
  net: hsr: fix incorrect type usage for protocol variable
  net: macsec: fix rtnl locking issue
  net: mvpp2: cls: Prevent buffer overflow in mvpp2_ethtool_cls_rule_del()
  ...

This patch adds FLOW_ACTION_HW_STATS_DONT_CARE which tells the driver
that the frontend does not need counters, this hw stats type request
never fails. The FLOW_ACTION_HW_STATS_DISABLED type explicitly requests
the driver to disable the stats, however, if the driver cannot disable
counters, it bails out.

TCA_ACT_HW_STATS_* maintains the 1:1 mapping with FLOW_ACTION_HW_STATS_*
except by disabled which is mapped to FLOW_ACTION_HW_STATS_DISABLED
(this is 0 in tc). Add tc_act_hw_stats() to perform the mapping between
TCA_ACT_HW_STATS_* and FLOW_ACTION_HW_STATS_*.

Fixes: 319a1d19 ("flow_offload: check for basic action hw stats type")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Commit 9b038086 ("docs: networking: convert DIM to RST") added a new
file entry to DYNAMIC INTERRUPT MODERATION to the end, and not following
alphabetical order.

So, ./scripts/checkpatch.pl -f MAINTAINERS complains:

  WARNING: Misordered MAINTAINERS entry - list file patterns in alphabetic
  order
  #5966: FILE: MAINTAINERS:5966:
  +F:      lib/dim/
  +F:      Documentation/networking/net_dim.rst

Reorder the file entries to keep MAINTAINERS nicely ordered.
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jason A. Donenfeld says:

====================
wireguard fixes for 5.7-rc5

With Ubuntu and Debian having backported this into their kernels, we're
finally seeing testing from places we hadn't seen prior, which is nice.
With that comes more fixes:

1) The CI for PPC64 was running with extremely small stacks for 64-bit,
   causing spurious crashes in surprising places.

2) There's was an old leftover routing loop restriction, which no longer
   makes sense given the queueing architecture, and was causing problems
   for people who really did want nested routing.

3) Not yielding our kthread on CONFIG_PREEMPT_VOLUNTARY systems caused
   RCU stalls and other issues, reported by Wang Jian, with the fix
   suggested by Sultan Alsawaf.

4) Clang spewed warnings in a selftest for CONFIG_IPV6=n, reported by
   Arnd Bergmann.

5) A complicated if statement was simplified to an assignment while also
   making the likely/unlikely hinting more correct and simple, and
   increasing readability, suggested by Sultan.

Patches (2) and (3) have Fixes: lines and are probably good candidates
for stable.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

It's very unlikely that send will become true. It's nearly always false
between 0 and 120 seconds of a session, and in most cases becomes true
only between 120 and 121 seconds before becoming false again. So,
unlikely(send) is clearly the right option here.

What happened before was that we had this complex boolean expression
with multiple likely and unlikely clauses nested. Since this is
evaluated left-to-right anyway, the whole thing got converted to
unlikely. So, we can clean this up to better represent what's going on.

The generated code is the same.
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Without setting these to NULL, clang complains in certain
configurations that have CONFIG_IPV6=n:

In file included from drivers/net/wireguard/ratelimiter.c:223:
drivers/net/wireguard/selftest/ratelimiter.c:173:34: error: variable 'skb6' is uninitialized when used here [-Werror,-Wuninitialized]
                ret = timings_test(skb4, hdr4, skb6, hdr6, &test_count);
                                               ^~~~
drivers/net/wireguard/selftest/ratelimiter.c:123:29: note: initialize the variable 'skb6' to silence this warning
        struct sk_buff *skb4, *skb6;
                                   ^
                                    = NULL
drivers/net/wireguard/selftest/ratelimiter.c:173:40: error: variable 'hdr6' is uninitialized when used here [-Werror,-Wuninitialized]
                ret = timings_test(skb4, hdr4, skb6, hdr6, &test_count);
                                                     ^~~~
drivers/net/wireguard/selftest/ratelimiter.c:125:22: note: initialize the variable 'hdr6' to silence this warning
        struct ipv6hdr *hdr6;
                            ^

We silence this warning by setting the variables to NULL as the warning
suggests.
Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Users with pathological hardware reported CPU stalls on CONFIG_
PREEMPT_VOLUNTARY=y, because the ringbuffers would stay full, meaning
these workers would never terminate. That turned out not to be okay on
systems without forced preemption, which Sultan observed. This commit
adds a cond_resched() to the bottom of each loop iteration, so that
these workers don't hog the core. Note that we don't need this on the
napi poll worker, since that terminates after its budget is expended.
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reported-by: Wang Jian <larkwang@gmail.com>
Fixes: e7096c13 ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

It's already possible to create two different interfaces and loop
packets between them. This has always been possible with tunnels in the
kernel, and isn't specific to wireguard. Therefore, the networking stack
already needs to deal with that. At the very least, the packet winds up
exceeding the MTU and is discarded at that point. So, since this is
already something that happens, there's no need to forbid the not very
exceptional case of routing a packet back to the same interface; this
loop is no different than others, and we shouldn't special case it, but
rather rely on generic handling of loops in general. This also makes it
easier to do interesting things with wireguard such as onion routing.

At the same time, we add a selftest for this, ensuring that both onion
routing works and infinite routing loops do not crash the kernel. We
also add a test case for wireguard interfaces nesting packets and
sending traffic between each other, as well as the loop in this case
too. We make sure to send some throughput-heavy traffic for this use
case, to stress out any possible recursion issues with the locks around
workqueues.

Fixes: e7096c13 ("net: WireGuard secure network tunnel")
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

While at some point it might have made sense to be running these tests
on ppc64 with 4k stacks, the kernel hasn't actually used 4k stacks on
64-bit powerpc in a long time, and more interesting things that we test
don't really work when we deviate from the default (16k). So, we stop
pushing our luck in this commit, and return to the default instead of
the minimum.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The K3 INTA driver, which is source TX/RX IRQs for CPSW NUSS, defines IRQs
triggering type as EDGE by default, but triggering type for CPSW NUSS TX/RX
IRQs has to be LEVEL as the EDGE triggering type may cause unnecessary IRQs
triggering and NAPI scheduling for empty queues. It was discovered with
RT-kernel.

Fix it by explicitly specifying CPSW NUSS TX/RX IRQ type as
IRQF_TRIGGER_HIGH.

Fixes: 93a76530 ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver")
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The addition of sja1105_port_status_ether structure into the
statistics causes the frame size to go over the warning limit:

drivers/net/dsa/sja1105/sja1105_ethtool.c:421:6: error: stack frame size of 1104 bytes in function 'sja1105_get_ethtool_stats' [-Werror,-Wframe-larger-than=]

Use dynamic allocation to avoid this.

Fixes: 336aa67b ("net: dsa: sja1105: show more ethtool statistics counters for P/Q/R/S")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently bool ionic_cq.done_color is exported using
debugfs_create_u8(), which requires a cast, preventing further compiler
checks.

Fix this by switching to debugfs_create_bool(), and dropping the cast.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>

Oleksij Rempel says:

====================
provide support for PHY master/slave configuration

changes v6:
- use NL_SET_ERR_MSG_ATTR in ethnl_update_linkmodes
- add sanity checks in the ioctl interface
- use bool for ethnl_validate_master_slave_cfg()

changes v5:
- set MASTER_SLAVE_CFG_UNSUPPORTED as default value
- send a netlink error message on validation error
- more code fixes

changes v4:
- rename port_mode to master_slave
- move validation code to net/ethtool/linkmodes.c
- add UNSUPPORTED state and avoid sending unsupported fields
- more formatting and naming fixes
- tja11xx: support only force mode
- tja11xx: mark state as unsupported

changes v3:
- provide separate field for config and state.
- make state rejected on set
- add validation

changes v2:
- change names. Use MASTER_PREFERRED instead of MULTIPORT
- configure master/slave only on request. Default configuration can be
  provided by PHY or eeprom
- status and configuration to the user space.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The TJA11xx PHYs have a vendor specific Master/Slave configuration bit,
which is not compatible with IEEE 803.2-2018 spec for 100Base-T1
devices. So, provide a custom config_ange call back to solve this
problem.
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

This UAPI is needed for BroadR-Reach 100BASE-T1 devices. Due to lack of
auto-negotiation support, we needed to be able to configure the
MASTER-SLAVE role of the port manually or from an application in user
space.

The same UAPI can be used for 1000BASE-T or MultiGBASE-T devices to
force MASTER or SLAVE role. See IEEE 802.3-2018:
22.2.4.3.7 MASTER-SLAVE control register (Register 9)
22.2.4.3.8 MASTER-SLAVE status register (Register 10)
40.5.2 MASTER-SLAVE configuration resolution
45.2.1.185.1 MASTER-SLAVE config value (1.2100.14)
45.2.7.10 MultiGBASE-T AN control 1 register (Register 7.32)

The MASTER-SLAVE role affects the clock configuration:

-------------------------------------------------------------------------------
When the  PHY is configured as MASTER, the PMA Transmit function shall
source TX_TCLK from a local clock source. When configured as SLAVE, the
PMA Transmit function shall source TX_TCLK from the clock recovered from
data stream provided by MASTER.

iMX6Q                     KSZ9031                XXX
------\                /-----------\        /------------\
      |                |           |        |            |
 MAC  |<----RGMII----->| PHY Slave |<------>| PHY Master |
      |<--- 125 MHz ---+-<------/  |        | \          |
------/                \-----------/        \------------/
                                               ^
                                                \-TX_TCLK

-------------------------------------------------------------------------------

Since some clock or link related issues are only reproducible in a
specific MASTER-SLAVE-role, MAC and PHY configuration, it is beneficial
to provide generic (not 100BASE-T1 specific) interface to the user space
for configuration flexibility and trouble shooting.
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Devulapally Shiva Krishna says:

====================
Crypto/chcr: Fix issues regarding algorithm implementation in driver

The following series of patches fixes the issues which came during
self-tests with CONFIG_CRYPTO_MANAGER_EXTRA_TESTS enabled.

Patch 1: Fixes gcm(aes) hang issue and rfc4106-gcm encryption issue.
Patch 2: Fixes ctr, cbc, xts and rfc3686-ctr extra test failures.
Patch 3: Fixes ccm(aes) extra test failures.
Patch 4: Added support for 48 byte-key_len in aes_xts.
Patch 5: fix for hmac(sha) extra test failure.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The hmac(sha) test fails for a zero length source text data.
For hmac(sha) minimum length of the data must be of block-size.
So fix this by including the data_len for the last block.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Added support for 48 byte key length for aes-xts.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The ccm(aes) test fails when req->assoclen > ~240bytes.

The problem is the value assigned to auth_offset is wrong.
As auth_offset is unsigned char, it can take max value as 255.
So fix it by making it unsigned int.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This solves the following issues observed during self test when
CONFIG_CRYPTO_MANAGER_EXTRA_TESTS is enabled.

1. Added fallback for cbc, ctr and rfc3686 if req->nbytes is zero
and for xts added a fallback case if req->nbytes is not multiple of 16.

2. In case of cbc-aes, solved wrong iv update. When
chcr_cipher_fallback() is called, used req->info pointer instead of
reqctx->iv.

3. In cbc-aes decryption there was a wrong result. This occurs when
chcr_cipher_fallback() is called from chcr_handle_cipher_resp().
In the fallback function iv(req->info) used is wrongly updated.
So use the initial iv for this case.

4)In case of ctr-aes encryption observed wrong result. In adjust_ctr_overflow()
there is condition which checks if ((bytes / AES_BLOCK_SIZE) > c),
where c is the number of blocks which can be processed without iv overflow,
but for the above bytes (req->nbytes < 32 , not a multiple of 16) this
condition fails and the 2nd block is corrupted as it requires the rollover iv.
So added a '=' condition in this to take care of this.

5)In rfc3686-ctr there was wrong result observed. This occurs when
chcr_cipher_fallback() is called from chcr_handle_cipher_resp().
Here also copying initial_iv in init_iv pointer for handling the fallback
case correctly.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fixes two issues observed during self tests with
CONFIG_CRYPTO_MANAGER_EXTRA_TESTS enabled.

1. gcm(aes) hang issue , that happens during decryption.
2. rfc4106-gcm-aes-chcr encryption unexpectedly succeeded.

For gcm-aes decryption , authtag is not mapped due to
sg_nents_for_len(upto size: assoclen+ cryptlen - authsize).
So fix it by dma_mapping authtag.
Also replaced sg_nents() to sg_nents_for_len() in case of aead_dma_unmap().

For rfc4106-gcm-aes-chcr, used crypto_ipsec_check_assoclen() for checking
the validity of assoclen.
Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Devulapally Shiva Krishna <shiva@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alex Elder says:

====================
net: ipa: kill endpoint stop workaround

It turns out that a workaround that performs a small DMA operation
between retried attempts to stop a GSI channel is not needed for any
supported hardware.  The hardware quirk that required the extra DMA
operation was fixed after IPA v3.1.  So this series gets rid of that
workaround code, along with some other code that was only present to
support it.

NOTE:  This series depends on (and includes/duplicates) another patch
       that has already been committed in the net tree:
         713b6ebb net: ipa: fix a bug in ipa_endpoint_stop()
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

A recent commit removed the only use of ipa_cmd_dma_task_32b_addr_add().
This function (and the IPA immediate command it implements) is no
longer needed, so get rid of it, along with all of the definitions
associated with it.  Isolate its removal in a commit so it can be
easily added back again if needed.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The previous commit made ipa_endpoint_stop() be a trivial wrapper
around gsi_channel_stop().  Since it no longer does anything
special, just open-code it in the three places it's used.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The only reason ipa_endpoint_stop() had a retry loop was that the
just-removed workaround required an IPA DMA command to occur between
attempts.  The gsi_channel_stop() call that implements the stop does
its own retry loop, to cover a channel's transition from started to
stop-in-progress to stopped state.

Get rid of the unnecessary retry loop in ipa_endpoint_stop().
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

In ipa_endpoint_stop(), a workaround is used for IPA version 3.5.1
where a 1-byte DMA request is issued between GSI channel stop
retries.

It turns out that this workaround is only required for IPA versions
3.1 and 3.2, and we don't support those.  So remove the call to
ipa_endpoint_stop_rx_dma() in that function.  That leaves that
function unused, so get rid of it.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

In ipa_endpoint_stop(), for TX endpoints we set the number of retries
to 0.  When we break out of the loop, retries being 0 means we return
EIO rather than the value of ret (which should be 0).

Fix this by using a non-zero retry count for both RX and TX
channels, and just break out of the loop after calling
gsi_channel_stop() for TX channels.  This way only RX channels
will retry, and the retry count will be non-zero at the end
for TX channels (so the proper value gets returned).
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from commit 713b6ebb)
Signed-off-by: David S. Miller <davem@davemloft.net>

Alex Elder says:

====================
net: ipa: kill endpoint delay mode workaround

A "delay mode" feature was put in place to work around a problem
where packets could passed to the modem before it was ready to
handle them.  That problem no longer exists, and we don't need the
workaround any more so get rid of it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

A "delay mode" feature was put in place to work around a problem
that was observed during development of the upstream IPA driver.  It
used TX endpoint "delay mode" in order to prevent transmitting
packets toward the modem before it was ready.

A race condition that would explain the problem has long since been
fixed, and we have concluded that the "delay mode" feature is no
longer required.  So get rid of it.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Create a new helper function that encapsulates enabling or disabling
suspend on an RX endpoint.  It returns the previous state of the
endpoint (true means suspend mode was enabled).

Create another function that handles enabling or disabling delay mode
on a TX endpoint.  Delay mode does not work correctly on IPA version
4.2, so we don't currently use it (and shouldn't).

We only set delay mode in one case, and although we don't expect an
endpoint to already be in delay mode, it doesn't really matter if it
was.  So the delay function doesn't return a value.

Stop issuing warnings if the previous suspend or delay mode state
differs from what is expected.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Change ipa_endpoint_init_ctrl() so it returns the previous state
(whether suspend or delay mode was enabled) rather than indicating
whether the request caused a change in state.  This makes it easier
to understand what's happening where called.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alex Elder says:

====================
net: ipa: limit special reset handling

Some special handling done during channel reset should only be done
for IPA hardare version 3.5.1.  This series generalizes the meaning
of a flag passed to indicate special behavior, then has the special
handling be used only when appropriate.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

In gsi_channel_reset(), RX channels are subjected to two consecutive
CHANNEL_RESET commands.  This workaround should only be used for IPA
version 3.5.1, and for newer hardware "can lead to unwanted behavior."

Only issue the second CHANNEL_RESET command for legacy hardware.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

In several places, a Boolean flag is used in the GSI code to
indicate whether the "doorbell engine" should be enabled or not
when a channel is configured.  This is basically done to abstract
this property from the IPA version; the GSI code doesn't otherwise
"know" what the IPA hardware version is.  The doorbell engine is
enabled only for IPA v3.5.1, not for IPA v4.0 and later.

The next patch makes another change that affects behavior during
channel reset (which also involves programming the channel).  It
also distinguishes IPA v3.5.1 hardware from newer hardware.

Rather than creating another flag whose value matches the "db_enable"
value, just rename "db_enable" to be "legacy" so it can be used to
signal more than just the special doorbell handling.
Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

When ndo_get_phys_port_name() for the CPU port was added we introduced
an early check for when the DSA master network device in
dsa_master_ndo_setup() already implements ndo_get_phys_port_name(). When
we perform the teardown operation in dsa_master_ndo_teardown() we would
not be checking that cpu_dp->orig_ndo_ops was successfully allocated and
non-NULL initialized.

With network device drivers such as virtio_net, this leads to a NPD as
soon as the DSA switch hanging off of it gets torn down because we are
now assigning the virtio_net device's netdev_ops a NULL pointer.

Fixes: da7b9e9b ("net: dsa: Add ndo_get_phys_port_name() for CPU port")
Reported-by: Allen Pais <allen.pais@oracle.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Allen Pais <allen.pais@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This was caused by a poor merge conflict resolution on my side. The
"act = &cls->rule->action.entries[0];" assignment was already present in
the code prior to the patch mentioned below.

Fixes: e13c2075 ("net: dsa: refactor matchall mirred action to separate function")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
tcp: minor adjustments for low pacing rates

After pacing horizon addition, we have to adjust how we arm rto
timer, otherwise we might freeze very low pacing rate flows.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As hinted in prior change ("tcp: refine tcp_pacing_delay()
for very low pacing rates"), it is probably best arming
the xmit timer only when all the packets have been scheduled,
rather than when the head of rtx queue has been re-sent.

This does matter for flows having extremely low pacing rates,
since their tp->tcp_wstamp_ns could be far in the future.

Note that the regular xmit path has a stronger limit
in tcp_small_queue_check(), meaning it is less likely to
go beyond the pacing horizon.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

With the addition of horizon feature to sch_fq, we noticed some
suboptimal behavior of extremely low pacing rate TCP flows, especially
when TCP is not aware of a drop happening in lower stacks.

Back in commit 3f80e08f ("tcp: add tcp_reset_xmit_timer() helper"),
tcp_pacing_delay() was added to estimate an extra delay to add to standard
rto timers.

This patch removes the skb argument from this helper and
tcp_reset_xmit_timer() because it makes more sense to simply
consider the time at which next packet is allowed to be sent,
instead of the time of whatever packet has been sent.

This avoids arming RTO timer too soon and removes
spurious horizon drops.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>