Some qdiscs like taprio turn out to be actually pretty reliant on a well
configured stab, to not underestimate the skb transmission time (by
properly accounting for L1 overhead).

In a future change, taprio will need the stab, if configured by the
user, to be available at ops->init() time. It will become even more
important in upcoming work, when the overhead will be used for the
queueMaxSDU calculation that is passed to an offloading driver.

However, rcu_assign_pointer(sch->stab, stab) is called right after
ops->init(), making it unavailable, and I don't really see a good reason
for that.

Move it earlier, which nicely seems to simplify the error handling path
as well.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

taprio_dequeue_from_txq() looks at the entry->end_time to determine
whether the skb will overrun its traffic class gate, as if at the end of
the schedule entry there surely is a "gate close" event for it. Hint:
maybe there isn't.

For each schedule entry, introduce an array of kernel times which
actually tracks when in the future will there be an *actual* gate close
event for that traffic class, and use that in the guard band overrun
calculation.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently taprio assumes that the budget for a traffic class expires at
the end of the current interval as if the next interval contains a "gate
close" event for this traffic class.

This is, however, an unfounded assumption. Allow schedule entry
intervals to be fused together for a particular traffic class by
calculating the budget until the gate *actually* closes.

This means we need to keep budgets per traffic class, and we also need
to update the budget consumption procedure.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

There is a confusion in terms in taprio which makes what is called
"close_time" to be actually used for 2 things:

1. determining when an entry "closes" such that transmitted skbs are
   never allowed to overrun that time (?!)
2. an aid for determining when to advance and/or restart the schedule
   using the hrtimer

It makes more sense to call this so-called "close_time" "end_time",
because it's not clear at all to me what "closes". Future patches will
hopefully make better use of the term "to close".

This is an absolutely mechanical change.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Current taprio code operates on a very simplistic (and incorrect)
assumption: that egress scheduling for a traffic class can only take
place for the duration of the current interval, or i.o.w., it assumes
that at the end of each schedule entry, there is a "gate close" event
for all traffic classes.

As an example, traffic sent with the schedule below will be jumpy, even
though all 8 TC gates are open, so there is absolutely no "gate close"
event (effectively a transition from BIT(tc)==1 to BIT(tc)==0 in
consecutive schedule entries):

tc qdisc replace dev veth0 parent root taprio \
	num_tc 2 \
	map 0 1 \
	queues 1@0 1@1 \
	base-time 0 \
	sched-entry S 0xff 4000000000 \
	clockid CLOCK_TAI \
	flags 0x0

This qdisc simply does not have what it takes in terms of logic to
*actually* compute the durations of traffic classes. Also, it does not
recognize the need to use this information on a per-traffic-class basis:
it always looks at entry->interval and entry->close_time.

This change proposes that each schedule entry has an array called
tc_gate_duration[tc]. This holds the information: "for how long will
this traffic class gate remain open, starting from *this* schedule
entry". If the traffic class gate is always open, that value is equal to
the cycle time of the schedule.

We'll also need to keep track, for the purpose of queueMaxSDU[tc]
calculation, what is the maximum time duration for a traffic class
having an open gate. This gives us directly what is the maximum sized
packet that this traffic class will have to accept. For everything else
it has to qdisc_drop() it in qdisc_enqueue().
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Current taprio software implementation is haunted by the shadow of the
igb/igc hardware model. It iterates over child qdiscs in increasing
order of TXQ index, therefore giving higher xmit priority to TXQ 0 and
lower to TXQ N. According to discussions with Vinicius, that is the
default (perhaps even unchangeable) prioritization scheme used for the
NICs that taprio was first written for (igb, igc), and we have a case of
two bugs canceling out, resulting in a functional setup on igb/igc, but
a less sane one on other NICs.

To the best of my understanding, taprio should prioritize based on the
traffic class, so it should really dequeue starting with the highest
traffic class and going down from there. We get to the TXQ using the
tc_to_txq[] netdev property.

TXQs within the same TC have the same (strict) priority, so we should
pick from them as fairly as we can. We can achieve that by implementing
something very similar to q->curband from multiq_dequeue().

Since igb/igc really do have TXQ 0 of higher hardware priority than
TXQ 1 etc, we need to preserve the behavior for them as well. We really
have no choice, because in txtime-assist mode, taprio is essentially a
software scheduler towards offloaded child tc-etf qdiscs, so the TXQ
selection really does matter (not all igb TXQs support ETF/SO_TXTIME,
says Kurt Kanzenbach).

To preserve the behavior, we need a capability bit so that taprio can
determine if it's running on igb/igc, or on something else. Because igb
doesn't offload taprio at all, we can't piggyback on the
qdisc_offload_query_caps() call from taprio_enable_offload(), but
instead we need a separate call which is also made for software
scheduling.

Introduce two static keys to minimize the performance penalty on systems
which only have igb/igc NICs, and on systems which only have other NICs.
For mixed systems, taprio will have to dynamically check whether to
dequeue using one prioritization algorithm or using the other.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Simplify taprio_dequeue_from_txq() by noticing that we can goto one call
earlier than the previous skb_found label. This is possible because
we've unified the treatment of the child->ops->dequeue(child) return
call, we always try other TXQs now, instead of abandoning the root
dequeue completely if we failed in the peek() case.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Future changes will refactor the TXQ selection procedure, and a lot of
stuff will become messy, the indentation of the bulk of the dequeue
procedure would increase, etc.

Break out the bulk of the function into a new one, which knows the TXQ
(child qdisc) we should perform a dequeue from.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

This changes the handling of an unlikely condition to not stop dequeuing
if taprio failed to dequeue the peeked skb in taprio_dequeue().

I've no idea when this can happen, but the only side effect seems to be
that the atomic_sub_return() call right above will have consumed some
budget. This isn't a big deal, since either that made us remain without
any budget (and therefore, we'd exit on the next peeked skb anyway), or
we could send some packets from other TXQs.

I'm making this change because in a future patch I'll be refactoring the
dequeue procedure to simplify it, and this corner case will have to go
away.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

There isn't any code in the network stack which calls taprio_peek().
We only see qdisc->ops->peek() being called on child qdiscs of other
classful qdiscs, never from the generic qdisc code. Whereas taprio is
never a child qdisc, it is always root.

This snippet of a comment from qdisc_peek_dequeued() seems to confirm:

	/* we can reuse ->gso_skb because peek isn't called for root qdiscs */

Since I've been known to be wrong many times though, I'm not completely
removing it, but leaving a stub function in place which emits a warning.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Kurt Kanzenbach <kurt@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Horatiu Vultur says:

====================
net: micrel: Add support for lan8841 PHY

Add support for lan8841 PHY.

The first patch add the support for lan8841 PHY which can run at
10/100/1000Mbit. It also has support for other features, but they are not
added in this series.

The second patch updates the documentation for the dt-bindings which is
similar to the ksz9131.

v3->v4:
- add space between defines and function names
- inside lan8841_config_init use only ret variable

v2->v3:
- reuse ksz9131_config_init
- allow only open-drain configuration
- change from single patch to a patch series

v1->v2:
- Remove hardcoded values
- Fix typo in commit message
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The lan8841 has the same bindings as ksz9131, so just reuse the entire
section of ksz9131.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The LAN8841 is completely integrated triple-speed (10BASE-T/ 100BASE-TX/
1000BASE-T) Ethernet physical layer transceivers for transmission and
reception of data on standard CAT-5, as well as CAT-5e and CAT-6,
unshielded twisted pair (UTP) cables.
The LAN8841 offers the industry-standard GMII/MII as well as the RGMII.
Some of the features of the PHY are:
- Wake on LAN
- Auto-MDIX
- IEEE 1588-2008 (V2)
- LinkMD Capable diagnosis

Currently the patch offers support only for link configuration.
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add flower filter packet statistics. This will just read the TCAM
counter of the rule, which mention how many packages were hit by this
rule.
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tony Nguyen says:

====================
ice: various virtualization cleanups

Jacob Keller says:

This series contains a variety of refactors and cleanups in the VF code for
the ice driver. Its primary focus is cleanup and simplification of the VF
operations and addition of a few new operations that will be required by
Scalable IOV, as well as some other refactors needed for the handling of VF
subfunctions.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  ice: remove unnecessary virtchnl_ether_addr struct use
  ice: introduce .irq_close VF operation
  ice: introduce clear_reset_state operation
  ice: convert vf_ops .vsi_rebuild to .create_vsi
  ice: introduce ice_vf_init_host_cfg function
  ice: add a function to initialize vf entry
  ice: Pull common tasks into ice_vf_post_vsi_rebuild
  ice: move ice_vf_vsi_release into ice_vf_lib.c
  ice: move vsi_type assignment from ice_vsi_alloc to ice_vsi_cfg
  ice: refactor VSI setup to use parameter structure
  ice: drop unnecessary VF parameter from several VSI functions
  ice: fix function comment referring to ice_vsi_alloc
  ice: Add more usage of existing function ice_get_vf_vsi(vf)
====================

Link: https://lore.kernel.org/r/20230206214813.20107-1-anthony.l.nguyen@intel.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

The callback devlink_nl_cmd_health_reporter_diagnose_doit() miss
devlink_fmsg_free(), which leads to memory leak.

Fix it by adding devlink_fmsg_free().

Fixes: e994a75f ("devlink: remove reporter reference counting")
Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/1675698976-45993-1-git-send-email-moshe@nvidia.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Move the nfp_net_get_port_mac_by_hwinfo() check to ahead in the
get/set_eeprom() functions to in order to check for a VF netdev, which
this function does not support.

It is debatable if this is a fix or an enhancement, and we have chosen
to go for the latter. It does address a problem introduced by
commit 74b4f173 ("nfp: flower: change get/set_eeprom logic and enable for flower reps").
However, the ethtool->len == 0 check avoids the problem manifesting as a
run-time bug (NULL pointer dereference of app).
Signed-off-by: James Hershaw <james.hershaw@corigine.com>
Reviewed-by: Louis Peens <louis.peens@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Link: https://lore.kernel.org/r/20230206154836.2803995-1-simon.horman@corigine.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Petr Machata says:

====================
mlxsw: Misc devlink changes

This patchset adjusts mlxsw to recent devlink changes in net-next.

Patch #1 removes a devl_param_driverinit_value_set() call that was
unnecessary, but now additionally triggers a WARN_ON.

Patches #2-#4 are non-functional preparations for the following patches.

Patch #5 fixes a use-after-free that is triggered while changing network
namespaces.

Patch #6 makes mlxsw consistent with netdevsim by having mlxsw register
its devlink instance before its sub-objects. It helps us avoid a warning
described in the commit message.
====================

Link: https://lore.kernel.org/r/cover.1675692666.git.petrm@nvidia.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Recent changes made it possible to register the devlink instance before
its sub-objects and under the instance lock. Among other things, it
allows us to avoid warnings such as this one [1]. The warning is
generated because a buggy firmware is generating a health event during
driver initialization, before the devlink instance is registered.

Move the registration of the devlink instance to the beginning of the
initialization flow to avoid such problems.

A similar change was implemented in netdevsim in commit 82a3aef2
("netdevsim: move devlink registration under the instance lock").

[1]
WARNING: CPU: 3 PID: 49 at net/devlink/leftover.c:7509 devlink_recover_notify.constprop.0+0xaf/0xc0
[...]
Call Trace:
 <TASK>
 devlink_health_report+0x45/0x1d0
 mlxsw_core_health_event_work+0x24/0x30 [mlxsw_core]
 process_one_work+0x1db/0x390
 worker_thread+0x49/0x3b0
 kthread+0xe5/0x110
 ret_from_fork+0x1f/0x30
 </TASK>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Cited commit added 'DEVLINK_CMD_PARAM_DEL' notifications whenever the
network namespace of the devlink instance is changed. Specifically, the
notifications are generated after calling reload_down(), but before
calling reload_up(). At this stage, the data structures accessed while
reading the value of the "acl_region_rehash_interval" devlink parameter
are uninitialized, resulting in a use-after-free [1].

Fix by moving the registration and unregistration of the devlink
parameter to the TCAM code where it is actually used. This means that
the parameter is unregistered during reload_down() and then
re-registered during reload_up(), avoiding the use-after-free between
these two operations.

Reproducer:

 # ip netns add test123
 # devlink dev reload pci/0000:06:00.0 netns test123

[1]
BUG: KASAN: use-after-free in mlxsw_sp_acl_tcam_vregion_rehash_intrvl_get+0xb2/0xd0
Read of size 4 at addr ffff888162fd37d8 by task devlink/1323
[...]
Call Trace:
 <TASK>
 dump_stack_lvl+0x95/0xbd
 print_report+0x181/0x4a1
 kasan_report+0xdb/0x200
 mlxsw_sp_acl_tcam_vregion_rehash_intrvl_get+0xb2/0xd0
 mlxsw_sp_params_acl_region_rehash_intrvl_get+0x32/0x80
 devlink_nl_param_fill.constprop.0+0x29a/0x11e0
 devlink_param_notify.constprop.0+0xb9/0x250
 devlink_notify_unregister+0xbc/0x470
 devlink_reload+0x1aa/0x440
 devlink_nl_cmd_reload+0x559/0x11b0
 genl_family_rcv_msg_doit.isra.0+0x1f8/0x2e0
 genl_rcv_msg+0x558/0x7f0
 netlink_rcv_skb+0x170/0x440
 genl_rcv+0x2d/0x40
 netlink_unicast+0x53f/0x810
 netlink_sendmsg+0x961/0xe80
 __sys_sendto+0x2a4/0x420
 __x64_sys_sendto+0xe5/0x1c0
 do_syscall_64+0x38/0x80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

Fixes: 7d7e9169 ("devlink: move devlink reload notifications back in between _down() and _up() calls")
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Move the initialization and de-initialization code further below in
order to avoid forward declarations in the next patch. No functional
changes.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Move mutex_destroy() to the end to make the function symmetric with
mlxsw_sp_acl_tcam_init(). No functional changes.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Pair mutex_init() with a mutex_destroy() in the error path. Found during
code review. No functional changes.
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The "acl_region_rehash_interval" devlink parameter is a "runtime"
parameter, making the call to devl_param_driverinit_value_set()
pointless. Before cited commit the function simply returned an error
(that was not checked), but now it emits a WARNING [1].

Fix by removing the function call.

[1]
WARNING: CPU: 0 PID: 7 at net/devlink/leftover.c:10974
devl_param_driverinit_value_set+0x8c/0x90
[...]
Call Trace:
 <TASK>
 mlxsw_sp2_params_register+0x83/0xb0 [mlxsw_spectrum]
 __mlxsw_core_bus_device_register+0x5e5/0x990 [mlxsw_core]
 mlxsw_core_bus_device_register+0x42/0x60 [mlxsw_core]
 mlxsw_pci_probe+0x1f0/0x230 [mlxsw_pci]
 local_pci_probe+0x1a/0x40
 work_for_cpu_fn+0xf/0x20
 process_one_work+0x1db/0x390
 worker_thread+0x1d5/0x3b0
 kthread+0xe5/0x110
 ret_from_fork+0x1f/0x30
 </TASK>

Fixes: 85fe0b32 ("devlink: make devlink_param_driverinit_value_set() return void")
Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Petr Machata <petrm@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Add PF driver support for the following:

- Viewing the standardized MAC Merge layer counters.

- Viewing the standardized Ethernet MAC and RMON counters associated
  with the pMAC.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20230206094531.444988-2-vladimir.oltean@nxp.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Add PF driver support for viewing and changing the MAC Merge sublayer
parameters, and seeing the verification state machine's current state.
The verification handshake with the link partner is driven by hardware.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20230206094531.444988-1-vladimir.oltean@nxp.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Yury Norov says:

====================
sched: cpumask: improve on cpumask_local_spread() locality

cpumask_local_spread() currently checks local node for presence of i'th
CPU, and then if it finds nothing makes a flat search among all non-local
CPUs. We can do it better by checking CPUs per NUMA hops.

This has significant performance implications on NUMA machines, for example
when using NUMA-aware allocated memory together with NUMA-aware IRQ
affinity hints.

Performance tests from patch 8 of this series for mellanox network
driver show:

  TCP multi-stream, using 16 iperf3 instances pinned to 16 cores (with aRFS on).
  Active cores: 64,65,72,73,80,81,88,89,96,97,104,105,112,113,120,121

  +-------------------------+-----------+------------------+------------------+
  |                         | BW (Gbps) | TX side CPU util | RX side CPU util |
  +-------------------------+-----------+------------------+------------------+
  | Baseline                | 52.3      | 6.4 %            | 17.9 %           |
  +-------------------------+-----------+------------------+------------------+
  | Applied on TX side only | 52.6      | 5.2 %            | 18.5 %           |
  +-------------------------+-----------+------------------+------------------+
  | Applied on RX side only | 94.9      | 11.9 %           | 27.2 %           |
  +-------------------------+-----------+------------------+------------------+
  | Applied on both sides   | 95.1      | 8.4 %            | 27.3 %           |
  +-------------------------+-----------+------------------+------------------+

  Bottleneck in RX side is released, reached linerate (~1.8x speedup).
  ~30% less cpu util on TX.
====================

Link: https://lore.kernel.org/r/20230121042436.2661843-1-yury.norov@gmail.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Now that we have an iterator-based alternative for a very common case
of using cpumask_local_spread for all cpus in a row, it's worth to
mention that in comment to cpumask_local_spread().
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Reviewed-by: Valentin Schneider <vschneid@redhat.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

In the IRQ affinity hints, replace the binary NUMA preference (local /
remote) with the improved for_each_numa_hop_cpu() API that minds the
actual distances, so that remote NUMAs with short distance are preferred
over farther ones.

This has significant performance implications when using NUMA-aware
allocated memory (follow [1] and derivatives for example).

[1]
drivers/net/ethernet/mellanox/mlx5/core/en_main.c :: mlx5e_open_channel()
   int cpu = cpumask_first(mlx5_comp_irq_get_affinity_mask(priv->mdev, ix));

Performance tests:

TCP multi-stream, using 16 iperf3 instances pinned to 16 cores (with aRFS on).
Active cores: 64,65,72,73,80,81,88,89,96,97,104,105,112,113,120,121

+-------------------------+-----------+------------------+------------------+
|                         | BW (Gbps) | TX side CPU util | RX side CPU util |
+-------------------------+-----------+------------------+------------------+
| Baseline                | 52.3      | 6.4 %            | 17.9 %           |
+-------------------------+-----------+------------------+------------------+
| Applied on TX side only | 52.6      | 5.2 %            | 18.5 %           |
+-------------------------+-----------+------------------+------------------+
| Applied on RX side only | 94.9      | 11.9 %           | 27.2 %           |
+-------------------------+-----------+------------------+------------------+
| Applied on both sides   | 95.1      | 8.4 %            | 27.3 %           |
+-------------------------+-----------+------------------+------------------+

Bottleneck in RX side is released, reached linerate (~1.8x speedup).
~30% less cpu util on TX.

* CPU util on active cores only.

Setups details (similar for both sides):

NIC: ConnectX6-DX dual port, 100 Gbps each.
Single port used in the tests.

$ lscpu
Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
CPU(s):              256
On-line CPU(s) list: 0-255
Thread(s) per core:  2
Core(s) per socket:  64
Socket(s):           2
NUMA node(s):        16
Vendor ID:           AuthenticAMD
CPU family:          25
Model:               1
Model name:          AMD EPYC 7763 64-Core Processor
Stepping:            1
CPU MHz:             2594.804
BogoMIPS:            4890.73
Virtualization:      AMD-V
L1d cache:           32K
L1i cache:           32K
L2 cache:            512K
L3 cache:            32768K
NUMA node0 CPU(s):   0-7,128-135
NUMA node1 CPU(s):   8-15,136-143
NUMA node2 CPU(s):   16-23,144-151
NUMA node3 CPU(s):   24-31,152-159
NUMA node4 CPU(s):   32-39,160-167
NUMA node5 CPU(s):   40-47,168-175
NUMA node6 CPU(s):   48-55,176-183
NUMA node7 CPU(s):   56-63,184-191
NUMA node8 CPU(s):   64-71,192-199
NUMA node9 CPU(s):   72-79,200-207
NUMA node10 CPU(s):  80-87,208-215
NUMA node11 CPU(s):  88-95,216-223
NUMA node12 CPU(s):  96-103,224-231
NUMA node13 CPU(s):  104-111,232-239
NUMA node14 CPU(s):  112-119,240-247
NUMA node15 CPU(s):  120-127,248-255
..

$ numactl -H
..
node distances:
node   0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
  0:  10  11  11  11  12  12  12  12  32  32  32  32  32  32  32  32
  1:  11  10  11  11  12  12  12  12  32  32  32  32  32  32  32  32
  2:  11  11  10  11  12  12  12  12  32  32  32  32  32  32  32  32
  3:  11  11  11  10  12  12  12  12  32  32  32  32  32  32  32  32
  4:  12  12  12  12  10  11  11  11  32  32  32  32  32  32  32  32
  5:  12  12  12  12  11  10  11  11  32  32  32  32  32  32  32  32
  6:  12  12  12  12  11  11  10  11  32  32  32  32  32  32  32  32
  7:  12  12  12  12  11  11  11  10  32  32  32  32  32  32  32  32
  8:  32  32  32  32  32  32  32  32  10  11  11  11  12  12  12  12
  9:  32  32  32  32  32  32  32  32  11  10  11  11  12  12  12  12
 10:  32  32  32  32  32  32  32  32  11  11  10  11  12  12  12  12
 11:  32  32  32  32  32  32  32  32  11  11  11  10  12  12  12  12
 12:  32  32  32  32  32  32  32  32  12  12  12  12  10  11  11  11
 13:  32  32  32  32  32  32  32  32  12  12  12  12  11  10  11  11
 14:  32  32  32  32  32  32  32  32  12  12  12  12  11  11  10  11
 15:  32  32  32  32  32  32  32  32  12  12  12  12  11  11  11  10

$ cat /sys/class/net/ens5f0/device/numa_node
14

Affinity hints (127 IRQs):
Before:
331: 00000000,00000000,00000000,00000000,00010000,00000000,00000000,00000000
332: 00000000,00000000,00000000,00000000,00020000,00000000,00000000,00000000
333: 00000000,00000000,00000000,00000000,00040000,00000000,00000000,00000000
334: 00000000,00000000,00000000,00000000,00080000,00000000,00000000,00000000
335: 00000000,00000000,00000000,00000000,00100000,00000000,00000000,00000000
336: 00000000,00000000,00000000,00000000,00200000,00000000,00000000,00000000
337: 00000000,00000000,00000000,00000000,00400000,00000000,00000000,00000000
338: 00000000,00000000,00000000,00000000,00800000,00000000,00000000,00000000
339: 00010000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
340: 00020000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
341: 00040000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
342: 00080000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
343: 00100000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
344: 00200000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
345: 00400000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
346: 00800000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
347: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001
348: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000002
349: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000004
350: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000008
351: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000010
352: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000020
353: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000040
354: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000080
355: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000100
356: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000200
357: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000400
358: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000800
359: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00001000
360: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00002000
361: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00004000
362: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00008000
363: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00010000
364: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00020000
365: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00040000
366: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00080000
367: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00100000
368: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00200000
369: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00400000
370: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,00800000
371: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,01000000
372: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,02000000
373: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,04000000
374: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,08000000
375: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,10000000
376: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,20000000
377: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,40000000
378: 00000000,00000000,00000000,00000000,00000000,00000000,00000000,80000000
379: 00000000,00000000,00000000,00000000,00000000,00000000,00000001,00000000
380: 00000000,00000000,00000000,00000000,00000000,00000000,00000002,00000000
381: 00000000,00000000,00000000,00000000,00000000,00000000,00000004,00000000
382: 00000000,00000000,00000000,00000000,00000000,00000000,00000008,00000000
383: 00000000,00000000,00000000,00000000,00000000,00000000,00000010,00000000
384: 00000000,00000000,00000000,00000000,00000000,00000000,00000020,00000000
385: 00000000,00000000,00000000,00000000,00000000,00000000,00000040,00000000
386: 00000000,00000000,00000000,00000000,00000000,00000000,00000080,00000000
387: 00000000,00000000,00000000,00000000,00000000,00000000,00000100,00000000
388: 00000000,00000000,00000000,00000000,00000000,00000000,00000200,00000000
389: 00000000,00000000,00000000,00000000,00000000,00000000,00000400,00000000
390: 00000000,00000000,00000000,00000000,00000000,00000000,00000800,00000000
391: 00000000,00000000,00000000,00000000,00000000,00000000,00001000,00000000
392: 00000000,00000000,00000000,00000000,00000000,00000000,00002000,00000000
393: 00000000,00000000,00000000,00000000,00000000,00000000,00004000,00000000
394: 00000000,00000000,00000000,00000000,00000000,00000000,00008000,00000000
395: 00000000,00000000,00000000,00000000,00000000,00000000,00010000,00000000
396: 00000000,00000000,00000000,00000000,00000000,00000000,00020000,00000000
397: 00000000,00000000,00000000,00000000,00000000,00000000,00040000,00000000
398: 00000000,00000000,00000000,00000000,00000000,00000000,00080000,00000000
399: 00000000,00000000,00000000,00000000,00000000,00000000,00100000,00000000
400: 00000000,00000000,00000000,00000000,00000000,00000000,00200000,00000000
401: 00000000,00000000,00000000,00000000,00000000,00000000,00400000,00000000
402: 00000000,00000000,00000000,00000000,00000000,00000000,00800000,00000000
403: 00000000,00000000,00000000,00000000,00000000,00000000,01000000,00000000
404: 00000000,00000000,00000000,00000000,00000000,00000000,02000000,00000000
405: 00000000,00000000,00000000,00000000,00000000,00000000,04000000,00000000
406: 00000000,00000000,00000000,00000000,00000000,00000000,08000000,00000000
407: 00000000,00000000,00000000,00000000,00000000,00000000,10000000,00000000
408: 00000000,00000000,00000000,00000000,00000000,00000000,20000000,00000000
409: 00000000,00000000,00000000,00000000,00000000,00000000,40000000,00000000
410: 00000000,00000000,00000000,00000000,00000000,00000000,80000000,00000000
411: 00000000,00000000,00000000,00000000,00000000,00000001,00000000,00000000
412: 00000000,00000000,00000000,00000000,00000000,00000002,00000000,00000000
413: 00000000,00000000,00000000,00000000,00000000,00000004,00000000,00000000
414: 00000000,00000000,00000000,00000000,00000000,00000008,00000000,00000000
415: 00000000,00000000,00000000,00000000,00000000,00000010,00000000,00000000
416: 00000000,00000000,00000000,00000000,00000000,00000020,00000000,00000000
417: 00000000,00000000,00000000,00000000,00000000,00000040,00000000,00000000
418: 00000000,00000000,00000000,00000000,00000000,00000080,00000000,00000000
419: 00000000,00000000,00000000,00000000,00000000,00000100,00000000,00000000
420: 00000000,00000000,00000000,00000000,00000000,00000200,00000000,00000000
421: 00000000,00000000,00000000,00000000,00000000,00000400,00000000,00000000
422: 00000000,00000000,00000000,00000000,00000000,00000800,00000000,00000000
423: 00000000,00000000,00000000,00000000,00000000,00001000,00000000,00000000
424: 00000000,00000000,00000000,00000000,00000000,00002000,00000000,00000000
425: 00000000,00000000,00000000,00000000,00000000,00004000,00000000,00000000
426: 00000000,00000000,00000000,00000000,00000000,00008000,00000000,00000000
427: 00000000,00000000,00000000,00000000,00000000,00010000,00000000,00000000
428: 00000000,00000000,00000000,00000000,00000000,00020000,00000000,00000000
429: 00000000,00000000,00000000,00000000,00000000,00040000,00000000,00000000
430: 00000000,00000000,00000000,00000000,00000000,00080000,00000000,00000000
431: 00000000,00000000,00000000,00000000,00000000,00100000,00000000,00000000
432: 00000000,00000000,00000000,00000000,00000000,00200000,00000000,00000000
433: 00000000,00000000,00000000,00000000,00000000,00400000,00000000,00000000
434: 00000000,00000000,00000000,00000000,00000000,00800000,00000000,00000000
435: 00000000,00000000,00000000,00000000,00000000,01000000,00000000,00000000
436: 00000000,00000000,00000000,00000000,00000000,02000000,00000000,00000000
437: 00000000,00000000,00000000,00000000,00000000,04000000,00000000,00000000
438: 00000000,00000000,00000000,00000000,00000000,08000000,00000000,00000000
439: 00000000,00000000,00000000,00000000,00000000,10000000,00000000,00000000
440: 00000000,00000000,00000000,00000000,00000000,20000000,00000000,00000000
441: 00000000,00000000,00000000,00000000,00000000,40000000,00000000,00000000
442: 00000000,00000000,00000000,00000000,00000000,80000000,00000000,00000000
443: 00000000,00000000,00000000,00000000,00000001,00000000,00000000,00000000
444: 00000000,00000000,00000000,00000000,00000002,00000000,00000000,00000000
445: 00000000,00000000,00000000,00000000,00000004,00000000,00000000,00000000
446: 00000000,00000000,00000000,00000000,00000008,00000000,00000000,00000000
447: 00000000,00000000,00000000,00000000,00000010,00000000,00000000,00000000
448: 00000000,00000000,00000000,00000000,00000020,00000000,00000000,00000000
449: 00000000,00000000,00000000,00000000,00000040,00000000,00000000,00000000
450: 00000000,00000000,00000000,00000000,00000080,00000000,00000000,00000000
451: 00000000,00000000,00000000,00000000,00000100,00000000,00000000,00000000
452: 00000000,00000000,00000000,00000000,00000200,00000000,00000000,00000000
453: 00000000,00000000,00000000,00000000,00000400,00000000,00000000,00000000
454: 00000000,00000000,00000000,00000000,00000800,00000000,00000000,00000000
455: 00000000,00000000,00000000,00000000,00001000,00000000,00000000,00000000
456: 00000000,00000000,00000000,00000000,00002000,00000000,00000000,00000000
457: 00000000,00000000,00000000,00000000,00004000,00000000,00000000,00000000

After:
331: 00000000,00000000,00000000,00000000,00010000,00000000,00000000,00000000
332: 00000000,00000000,00000000,00000000,00020000,00000000,00000000,00000000
333: 00000000,00000000,00000000,00000000,00040000,00000000,00000000,00000000
334: 00000000,00000000,00000000,00000000,00080000,00000000,00000000,00000000
335: 00000000,00000000,00000000,00000000,00100000,00000000,00000000,00000000
336: 00000000,00000000,00000000,00000000,00200000,00000000,00000000,00000000
337: 00000000,00000000,00000000,00000000,00400000,00000000,00000000,00000000
338: 00000000,00000000,00000000,00000000,00800000,00000000,00000000,00000000
339: 00010000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
340: 00020000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
341: 00040000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
342: 00080000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
343: 00100000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
344: 00200000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
345: 00400000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
346: 00800000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
347: 00000000,00000000,00000000,00000000,00000001,00000000,00000000,00000000
348: 00000000,00000000,00000000,00000000,00000002,00000000,00000000,00000000
349: 00000000,00000000,00000000,00000000,00000004,00000000,00000000,00000000
350: 00000000,00000000,00000000,00000000,00000008,00000000,00000000,00000000
351: 00000000,00000000,00000000,00000000,00000010,00000000,00000000,00000000
352: 00000000,00000000,00000000,00000000,00000020,00000000,00000000,00000000
353: 00000000,00000000,00000000,00000000,00000040,00000000,00000000,00000000
354: 00000000,00000000,00000000,00000000,00000080,00000000,00000000,00000000
355: 00000000,00000000,00000000,00000000,00000100,00000000,00000000,00000000
356: 00000000,00000000,00000000,00000000,00000200,00000000,00000000,00000000
357: 00000000,00000000,00000000,00000000,00000400,00000000,00000000,00000000
358: 00000000,00000000,00000000,00000000,00000800,00000000,00000000,00000000
359: 00000000,00000000,00000000,00000000,00001000,00000000,00000000,00000000
360: 00000000,00000000,00000000,00000000,00002000,00000000,00000000,00000000
361: 00000000,00000000,00000000,00000000,00004000,00000000,00000000,00000000
362: 00000000,00000000,00000000,00000000,00008000,00000000,00000000,00000000
363: 00000000,00000000,00000000,00000000,01000000,00000000,00000000,00000000
364: 00000000,00000000,00000000,00000000,02000000,00000000,00000000,00000000
365: 00000000,00000000,00000000,00000000,04000000,00000000,00000000,00000000
366: 00000000,00000000,00000000,00000000,08000000,00000000,00000000,00000000
367: 00000000,00000000,00000000,00000000,10000000,00000000,00000000,00000000
368: 00000000,00000000,00000000,00000000,20000000,00000000,00000000,00000000
369: 00000000,00000000,00000000,00000000,40000000,00000000,00000000,00000000
370: 00000000,00000000,00000000,00000000,80000000,00000000,00000000,00000000
371: 00000001,00000000,00000000,00000000,00000000,00000000,00000000,00000000
372: 00000002,00000000,00000000,00000000,00000000,00000000,00000000,00000000
373: 00000004,00000000,00000000,00000000,00000000,00000000,00000000,00000000
374: 00000008,00000000,00000000,00000000,00000000,00000000,00000000,00000000
375: 00000010,00000000,00000000,00000000,00000000,00000000,00000000,00000000
376: 00000020,00000000,00000000,00000000,00000000,00000000,00000000,00000000
377: 00000040,00000000,00000000,00000000,00000000,00000000,00000000,00000000
378: 00000080,00000000,00000000,00000000,00000000,00000000,00000000,00000000
379: 00000100,00000000,00000000,00000000,00000000,00000000,00000000,00000000
380: 00000200,00000000,00000000,00000000,00000000,00000000,00000000,00000000
381: 00000400,00000000,00000000,00000000,00000000,00000000,00000000,00000000
382: 00000800,00000000,00000000,00000000,00000000,00000000,00000000,00000000
383: 00001000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
384: 00002000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
385: 00004000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
386: 00008000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
387: 01000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
388: 02000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
389: 04000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
390: 08000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
391: 10000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
392: 20000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
393: 40000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
394: 80000000,00000000,00000000,00000000,00000000,00000000,00000000,00000000
395: 00000000,00000000,00000000,00000000,00000000,00000001,00000000,00000000
396: 00000000,00000000,00000000,00000000,00000000,00000002,00000000,00000000
397: 00000000,00000000,00000000,00000000,00000000,00000004,00000000,00000000
398: 00000000,00000000,00000000,00000000,00000000,00000008,00000000,00000000
399: 00000000,00000000,00000000,00000000,00000000,00000010,00000000,00000000
400: 00000000,00000000,00000000,00000000,00000000,00000020,00000000,00000000
401: 00000000,00000000,00000000,00000000,00000000,00000040,00000000,00000000
402: 00000000,00000000,00000000,00000000,00000000,00000080,00000000,00000000
403: 00000000,00000000,00000000,00000000,00000000,00000100,00000000,00000000
404: 00000000,00000000,00000000,00000000,00000000,00000200,00000000,00000000
405: 00000000,00000000,00000000,00000000,00000000,00000400,00000000,00000000
406: 00000000,00000000,00000000,00000000,00000000,00000800,00000000,00000000
407: 00000000,00000000,00000000,00000000,00000000,00001000,00000000,00000000
408: 00000000,00000000,00000000,00000000,00000000,00002000,00000000,00000000
409: 00000000,00000000,00000000,00000000,00000000,00004000,00000000,00000000
410: 00000000,00000000,00000000,00000000,00000000,00008000,00000000,00000000
411: 00000000,00000000,00000000,00000000,00000000,00010000,00000000,00000000
412: 00000000,00000000,00000000,00000000,00000000,00020000,00000000,00000000
413: 00000000,00000000,00000000,00000000,00000000,00040000,00000000,00000000
414: 00000000,00000000,00000000,00000000,00000000,00080000,00000000,00000000
415: 00000000,00000000,00000000,00000000,00000000,00100000,00000000,00000000
416: 00000000,00000000,00000000,00000000,00000000,00200000,00000000,00000000
417: 00000000,00000000,00000000,00000000,00000000,00400000,00000000,00000000
418: 00000000,00000000,00000000,00000000,00000000,00800000,00000000,00000000
419: 00000000,00000000,00000000,00000000,00000000,01000000,00000000,00000000
420: 00000000,00000000,00000000,00000000,00000000,02000000,00000000,00000000
421: 00000000,00000000,00000000,00000000,00000000,04000000,00000000,00000000
422: 00000000,00000000,00000000,00000000,00000000,08000000,00000000,00000000
423: 00000000,00000000,00000000,00000000,00000000,10000000,00000000,00000000
424: 00000000,00000000,00000000,00000000,00000000,20000000,00000000,00000000
425: 00000000,00000000,00000000,00000000,00000000,40000000,00000000,00000000
426: 00000000,00000000,00000000,00000000,00000000,80000000,00000000,00000000
427: 00000000,00000001,00000000,00000000,00000000,00000000,00000000,00000000
428: 00000000,00000002,00000000,00000000,00000000,00000000,00000000,00000000
429: 00000000,00000004,00000000,00000000,00000000,00000000,00000000,00000000
430: 00000000,00000008,00000000,00000000,00000000,00000000,00000000,00000000
431: 00000000,00000010,00000000,00000000,00000000,00000000,00000000,00000000
432: 00000000,00000020,00000000,00000000,00000000,00000000,00000000,00000000
433: 00000000,00000040,00000000,00000000,00000000,00000000,00000000,00000000
434: 00000000,00000080,00000000,00000000,00000000,00000000,00000000,00000000
435: 00000000,00000100,00000000,00000000,00000000,00000000,00000000,00000000
436: 00000000,00000200,00000000,00000000,00000000,00000000,00000000,00000000
437: 00000000,00000400,00000000,00000000,00000000,00000000,00000000,00000000
438: 00000000,00000800,00000000,00000000,00000000,00000000,00000000,00000000
439: 00000000,00001000,00000000,00000000,00000000,00000000,00000000,00000000
440: 00000000,00002000,00000000,00000000,00000000,00000000,00000000,00000000
441: 00000000,00004000,00000000,00000000,00000000,00000000,00000000,00000000
442: 00000000,00008000,00000000,00000000,00000000,00000000,00000000,00000000
443: 00000000,00010000,00000000,00000000,00000000,00000000,00000000,00000000
444: 00000000,00020000,00000000,00000000,00000000,00000000,00000000,00000000
445: 00000000,00040000,00000000,00000000,00000000,00000000,00000000,00000000
446: 00000000,00080000,00000000,00000000,00000000,00000000,00000000,00000000
447: 00000000,00100000,00000000,00000000,00000000,00000000,00000000,00000000
448: 00000000,00200000,00000000,00000000,00000000,00000000,00000000,00000000
449: 00000000,00400000,00000000,00000000,00000000,00000000,00000000,00000000
450: 00000000,00800000,00000000,00000000,00000000,00000000,00000000,00000000
451: 00000000,01000000,00000000,00000000,00000000,00000000,00000000,00000000
452: 00000000,02000000,00000000,00000000,00000000,00000000,00000000,00000000
453: 00000000,04000000,00000000,00000000,00000000,00000000,00000000,00000000
454: 00000000,08000000,00000000,00000000,00000000,00000000,00000000,00000000
455: 00000000,10000000,00000000,00000000,00000000,00000000,00000000,00000000
456: 00000000,20000000,00000000,00000000,00000000,00000000,00000000,00000000
457: 00000000,40000000,00000000,00000000,00000000,00000000,00000000,00000000
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
[Tweaked API use]
Suggested-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Reviewed-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The recently introduced sched_numa_hop_mask() exposes cpumasks of CPUs
reachable within a given distance budget, wrap the logic for iterating over
all (distance, mask) values inside an iterator macro.
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Reviewed-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Tariq has pointed out that drivers allocating IRQ vectors would benefit
from having smarter NUMA-awareness - cpumask_local_spread() only knows
about the local node and everything outside is in the same bucket.

sched_domains_numa_masks is pretty much what we want to hand out (a cpumask
of CPUs reachable within a given distance budget), introduce
sched_numa_hop_mask() to export those cpumasks.

Link: http://lore.kernel.org/r/20220728191203.4055-1-tariqt@nvidia.comSigned-off-by: Valentin Schneider <vschneid@redhat.com>
Reviewed-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Now after moving all NUMA logic into sched_numa_find_nth_cpu(),
else-branch of cpumask_local_spread() is just a function call, and
we can simplify logic by using ternary operator.

While here, replace BUG() with WARN_ON().
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Peter Lafreniere <peter@n8pjl.ca>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Switch cpumask_local_spread() to use newly added sched_numa_find_nth_cpu(),
which takes into account distances to each node in the system.

For the following NUMA configuration:

root@debian:~# numactl -H
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3
node 0 size: 3869 MB
node 0 free: 3740 MB
node 1 cpus: 4 5
node 1 size: 1969 MB
node 1 free: 1937 MB
node 2 cpus: 6 7
node 2 size: 1967 MB
node 2 free: 1873 MB
node 3 cpus: 8 9 10 11 12 13 14 15
node 3 size: 7842 MB
node 3 free: 7723 MB
node distances:
node   0   1   2   3
  0:  10  50  30  70
  1:  50  10  70  30
  2:  30  70  10  50
  3:  70  30  50  10

The new cpumask_local_spread() traverses cpus for each node like this:

node 0:   0   1   2   3   6   7   4   5   8   9  10  11  12  13  14  15
node 1:   4   5   8   9  10  11  12  13  14  15   0   1   2   3   6   7
node 2:   6   7   0   1   2   3   8   9  10  11  12  13  14  15   4   5
node 3:   8   9  10  11  12  13  14  15   4   5   6   7   0   1   2   3
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Peter Lafreniere <peter@n8pjl.ca>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The function finds Nth set CPU in a given cpumask starting from a given
node.

Leveraging the fact that each hop in sched_domains_numa_masks includes the
same or greater number of CPUs than the previous one, we can use binary
search on hops instead of linear walk, which makes the overall complexity
of O(log n) in terms of number of cpumask_weight() calls.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Peter Lafreniere <peter@n8pjl.ca>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Introduce cpumask_nth_and_andnot() based on find_nth_and_andnot_bit().
It's used in the following patch to traverse cpumasks without storing
intermediate result in temporary cpumask.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Peter Lafreniere <peter@n8pjl.ca>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

In the following patches the function is used to implement in-place bitmaps
traversing without storing intermediate result in temporary bitmaps.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Peter Lafreniere <peter@n8pjl.ca>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Eric Dumazet says:

====================
net: core: use a dedicated kmem_cache for skb head allocs

Our profile data show that using kmalloc(non_const_size)/kfree(ptr)
has a certain cost, because kfree(ptr) has to pull a 'struct page'
in cpu caches.

Using a dedicated kmem_cache for TCP skb->head allocations makes
a difference, both in cpu cycles and memory savings.

This kmem_cache could also be used for GRO skb allocations,
this is left as a future exercise.
====================

Link: https://lore.kernel.org/r/20230206173103.2617121-1-edumazet@google.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

Recent removal of ksize() in alloc_skb() increased
performance because we no longer read
the associated struct page.

We have an equivalent cost at kfree_skb() time.

kfree(skb->head) has to access a struct page,
often cold in cpu caches to get the owning
struct kmem_cache.

Considering that many allocations are small (at least for TCP ones)
we can have our own kmem_cache to avoid the cache line miss.

This also saves memory because these small heads
are no longer padded to 1024 bytes.

CONFIG_SLUB=y
$ grep skbuff_small_head /proc/slabinfo
skbuff_small_head   2907   2907    640   51    8 : tunables    0    0    0 : slabdata     57     57      0

CONFIG_SLAB=y
$ grep skbuff_small_head /proc/slabinfo
skbuff_small_head    607    624    640    6    1 : tunables   54   27    8 : slabdata    104    104      5

Notes:

- After Kees Cook patches and this one, we might
  be able to revert commit
  dbae2b06 ("net: skb: introduce and use a single page frag cache")
  because GRO_MAX_HEAD is also small.

- This patch is a NOP for CONFIG_SLOB=y builds.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

All kmalloc_reserve() callers have to make the same computation,
we can factorize them, to prepare following patch in the series.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

This is a cleanup patch, to prepare following change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>