tcp: rstreason: fully support in tcp_check_req()

We're going to send an RST due to invalid syn packet which is already
checked whether 1) it is in sequence, 2) it is a retransmitted skb.

As RFC 793 says, if the state of socket is not CLOSED/LISTEN/SYN-SENT,
then we should send an RST when receiving bad syn packet:
"fourth, check the SYN bit,...If the SYN is in the window it is an
error, send a reset"

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://lore.kernel.org/r/20240510122502.27850-6-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: rstreason: handle timewait cases in the receive path

There are two possible cases where TCP layer can send an RST. Since they
happen in the same place, I think using one independent reason is enough
to identify this special situation.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://lore.kernel.org/r/20240510122502.27850-5-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: rstreason: fully support in tcp_rcv_state_process()

Like the previous patch does in this series, finish the conversion map is
enough to let rstreason mechanism work in this function.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://lore.kernel.org/r/20240510122502.27850-4-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: rstreason: fully support in tcp_ack()

Based on the existing skb drop reason, updating the rstreason map can
help us finish the rstreason job in this function.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://lore.kernel.org/r/20240510122502.27850-3-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: rstreason: fully support in tcp_rcv_synsent_state_process()

In this function, only updating the map can finish the job for socket
reset reason because the corresponding drop reasons are ready.

Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://lore.kernel.org/r/20240510122502.27850-2-kerneljasonxing@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-stmmac-add-support-for-rzn1-gmac-devices'

Romain Gantois says:

====================
net: stmmac: Add support for RZN1 GMAC devices

This is version seven of my series that adds support for a Gigabit Ethernet
controller featured in the Renesas r9a06g032 SoC, of the RZ/N1 family. This
GMAC device is based on a Synopsys IP and is compatible with the stmmac driver.

My former colleague Clément Léger originally sent a series for this driver,
but an issue in bringing up the PCS clock had blocked the upstreaming
process. This issue has since been resolved by the following series:

https://lore.kernel.org/all/20240326-rxc_bugfix-v6-0-24a74e5c761f@bootlin.com/

This series consists of a devicetree binding describing the RZN1 GMAC
controller IP, a node for the GMAC1 device in the r9a06g032 SoC device
tree, and the GMAC driver itself which is a glue layer in stmmac.

There are also two patches by Russell that improve pcs initialization handling
in stmmac.
====================

Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-0-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: add support for RZ/N1 GMAC

Add support for the Renesas RZ/N1 GMAC. This support can make use of a
custom RZ/N1 PCS which is fetched by parsing the pcs-handle device tree
property.

Signed-off-by: Clément Léger <clement.leger@bootlin.com>
Co-developed-by: Romain Gantois <romain.gantois@bootlin.com>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-6-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: dwmac-socfpga: use pcs_init/pcs_exit

Use the newly introduced pcs_init() and pcs_exit() operations to
create and destroy the PCS instance at a more appropriate moment during
the driver lifecycle, thereby avoiding publishing a network device to
userspace that has not yet finished its PCS initialisation.

There are other similar issues with this driver which remain
unaddressed, but these are out of scope for this patch.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
[rgantois: removed second parameters of new callbacks]
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-5-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: introduce pcs_init/pcs_exit stmmac operations

Introduce a mechanism whereby platforms can create their PCS instances
prior to the network device being published to userspace, but after
some of the core stmmac initialisation has been completed. This means
that the data structures that platforms need will be available.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Serge Semin <fancer.lancer@gmail.com>
Co-developed-by: Romain Gantois <romain.gantois@bootlin.com>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-4-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: Make stmmac_xpcs_setup() generic to all PCS devices

A pcs_init() callback will be introduced to stmmac in a future patch. This
new function will be called during the hardware initialization phase.
Instead of separately initializing XPCS and PCS components, let's group all
PCS-related hardware initialization logic in the current
stmmac_xpcs_setup() function.

Rename stmmac_xpcs_setup() to stmmac_pcs_setup() and move the conditional
call to stmmac_xpcs_setup() inside the function itself.

Signed-off-by: Serge Semin <fancer.lancer@gmail.com>
Co-developed-by: Romain Gantois <romain.gantois@bootlin.com>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-3-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: Add dedicated XPCS cleanup method

Currently the XPCS handler destruction is performed in the
stmmac_mdio_unregister() method. It doesn't look good because the handler
isn't originally created in the corresponding protagonist
stmmac_mdio_unregister(), but in the stmmac_xpcs_setup() function. In
order to have more coherent MDIO and XPCS setup/cleanup procedures,
let's move the DW XPCS destruction to the dedicated stmmac_pcs_clean()
method.

This method will also be used to cleanup PCS hardware using the
pcs_exit() callback that will be introduced to stmmac in a subsequent
patch.

Signed-off-by: Serge Semin <fancer.lancer@gmail.com>
Co-developed-by: Romain Gantois <romain.gantois@bootlin.com>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Hariprasad Kelam <hkelam@marvell.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-2-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

dt-bindings: net: renesas,rzn1-gmac: Document RZ/N1 GMAC support

The RZ/N1 series of MPUs feature up to two Gigabit Ethernet controllers.
These controllers are based on Synopsys IPs. They can be connected to
RZ/N1 RGMII/RMII converters.

Add a binding that describes these GMAC devices.

Signed-off-by: Clément Léger <clement.leger@bootlin.com>
[rgantois: commit log]
Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Link: https://lore.kernel.org/r/20240513-rzn1-gmac1-v7-1-6acf58b5440d@bootlin.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: flower: validate control flags

This driver currently doesn't support any control flags.

Use flow_rule_match_has_control_flags() to check for control flags,
such as can be set through `tc flower ... ip_flags frag`.

In case any control flags are masked, flow_rule_match_has_control_flags()
sets a NL extended error message, and we return -EOPNOTSUPP.

Only compile-tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240511073705.230507-1-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'virtio_net-rx-enable-premapped-mode-by-default'

Xuan Zhuo says:

====================
virtio_net: rx enable premapped mode by default

Actually, for the virtio drivers, we can enable premapped mode whatever
the value of use_dma_api. Because we provide the virtio dma apis.
So the driver can enable premapped mode unconditionally.

This patch set makes the big mode of virtio-net to support premapped mode.
And enable premapped mode for rx by default.

Based on the following points, we do not use page pool to manage these
    pages:

    1. virtio-net uses the DMA APIs wrapped by virtio core. Therefore,
       we can only prevent the page pool from performing DMA operations, and
       let the driver perform DMA operations on the allocated pages.
    2. But when the page pool releases the page, we have no chance to
       execute dma unmap.
    3. A solution to #2 is to execute dma unmap every time before putting
       the page back to the page pool. (This is actually a waste, we don't
       execute unmap so frequently.)
    4. But there is another problem, we still need to use page.dma_addr to
       save the dma address. Using page.dma_addr while using page pool is
       unsafe behavior.
    5. And we need space the chain the pages submitted once to virtio core.

    More:
        https://lore.kernel.org/all/CACGkMEu=Aok9z2imB_c5qVuujSh=vjj1kx12fy9N7hqyi+M5Ow@mail.gmail.com/

Why we do not use the page space to store the dma?

    http://lore.kernel.org/all/CACGkMEuyeJ9mMgYnnB42=hw6umNuo=agn7VBqBqYPd7GN=+39Q@mail.gmail.com
====================

Link: https://lore.kernel.org/r/20240511031404.30903-1-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

virtio_net: remove the misleading comment

We call the build_skb() actually without copying data.
The comment is misleading. So remove it.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20240511031404.30903-5-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

virtio_net: rx remove premapped failover code

Now, the premapped mode can be enabled unconditionally.

So we can remove the failover code for merge and small mode.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://lore.kernel.org/r/20240511031404.30903-4-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

virtio_net: big mode skip the unmap check

The virtio-net big mode did not enable premapped mode,
so we did not need to check the unmap. And the subsequent
commit will remove the failover code for failing enable
premapped for merge and small mode. So we need to remove
the checking do_dma code in the big mode path.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20240511031404.30903-3-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

virtio_ring: enable premapped mode whatever use_dma_api

Now, we have virtio DMA APIs, the driver can be the premapped
mode whatever the virtio core uses dma api or not.

So remove the limit of checking use_dma_api from
virtqueue_set_dma_premapped().

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20240511031404.30903-2-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2024-05-13

We've added 119 non-merge commits during the last 14 day(s) which contain
a total of 134 files changed, 9462 insertions(+), 4742 deletions(-).

The main changes are:

1) Add BPF JIT support for 32-bit ARCv2 processors, from Shahab Vahedi.

2) Add BPF range computation improvements to the verifier in particular
   around XOR and OR operators, refactoring of checks for range computation
   and relaxing MUL range computation so that src_reg can also be an unknown
   scalar, from Cupertino Miranda.

3) Add support to attach kprobe BPF programs through kprobe_multi link in
   a session mode, meaning, a BPF program is attached to both function entry
   and return, the entry program can decide if the return program gets
   executed and the entry program can share u64 cookie value with return
   program. Session mode is a common use-case for tetragon and bpftrace,
   from Jiri Olsa.

4) Fix a potential overflow in libbpf's ring__consume_n() and improve libbpf
   as well as BPF selftest's struct_ops handling, from Andrii Nakryiko.

5) Improvements to BPF selftests in context of BPF gcc backend,
   from Jose E. Marchesi & David Faust.

6) Migrate remaining BPF selftest tests from test_sock_addr.c to prog_test-
   -style in order to retire the old test, run it in BPF CI and additionally
   expand test coverage, from Jordan Rife.

7) Big batch for BPF selftest refactoring in order to remove duplicate code
   around common network helpers, from Geliang Tang.

8) Another batch of improvements to BPF selftests to retire obsolete
   bpf_tcp_helpers.h as everything is available vmlinux.h,
   from Martin KaFai Lau.

9) Fix BPF map tear-down to not walk the map twice on free when both timer
   and wq is used, from Benjamin Tissoires.

10) Fix BPF verifier assumptions about socket->sk that it can be non-NULL,
    from Alexei Starovoitov.

11) Change BTF build scripts to using --btf_features for pahole v1.26+,
    from Alan Maguire.

12) Small improvements to BPF reusing struct_size() and krealloc_array(),
    from Andy Shevchenko.

13) Fix s390 JIT to emit a barrier for BPF_FETCH instructions,
    from Ilya Leoshkevich.

14) Extend TCP ->cong_control() callback in order to feed in ack and
    flag parameters and allow write-access to tp->snd_cwnd_stamp
    from BPF program, from Miao Xu.

15) Add support for internal-only per-CPU instructions to inline
    bpf_get_smp_processor_id() helper call for arm64 and riscv64 BPF JITs,
    from Puranjay Mohan.

16) Follow-up to remove the redundant ethtool.h from tooling infrastructure,
    from Tushar Vyavahare.

17) Extend libbpf to support "module:<function>" syntax for tracing
    programs, from Viktor Malik.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (119 commits)
  bpf: make list_for_each_entry portable
  bpf: ignore expected GCC warning in test_global_func10.c
  bpf: disable strict aliasing in test_global_func9.c
  selftests/bpf: Free strdup memory in xdp_hw_metadata
  selftests/bpf: Fix a few tests for GCC related warnings.
  bpf: avoid gcc overflow warning in test_xdp_vlan.c
  tools: remove redundant ethtool.h from tooling infra
  selftests/bpf: Expand ATTACH_REJECT tests
  selftests/bpf: Expand getsockname and getpeername tests
  sefltests/bpf: Expand sockaddr hook deny tests
  selftests/bpf: Expand sockaddr program return value tests
  selftests/bpf: Retire test_sock_addr.(c|sh)
  selftests/bpf: Remove redundant sendmsg test cases
  selftests/bpf: Migrate ATTACH_REJECT test cases
  selftests/bpf: Migrate expected_attach_type tests
  selftests/bpf: Migrate wildcard destination rewrite test
  selftests/bpf: Migrate sendmsg6 v4 mapped address tests
  selftests/bpf: Migrate sendmsg deny test cases
  selftests/bpf: Migrate WILDCARD_IP test
  selftests/bpf: Handle SYSCALL_EPERM and SYSCALL_ENOTSUPP test cases
  ...
====================

Link: https://lore.kernel.org/r/20240513134114.17575-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: pcs: lynx: no need to read LPA in lynx_pcs_get_state_2500basex()

Nothing useful is done with the LPA variable in lynx_pcs_get_state_2500basex(),
we can just remove the read.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://lore.kernel.org/r/20240513115345.2452799-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'mlx5-misc-patches'

Tariq Toukan says:

====================
mlx5 misc patches

This series includes patches for the mlx5 driver.

Patch 1 by Shay enables LAG with HCAs of 8 ports.

Patch 2 by Carolina optimizes the safe switch channels operation for the
TX-only changes.

Patch 3 by Parav cleans up some unused code.
====================

Link: https://lore.kernel.org/r/20240512124306.740898-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Remove unused msix related exported APIs

MSIX irq allocation and free APIs are no longer
in use. Hence, remove the dead code.

Signed-off-by: Parav Pandit <parav@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Link: https://lore.kernel.org/r/20240512124306.740898-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5e: Modifying channels number and updating TX queues

It is not appropriate for the mlx5e_num_channels_changed
function to be called solely for updating the TX queues,
even if the channels number has not been changed.

Move the code responsible for updating the TC and TX queues
from mlx5e_num_channels_changed and produce a new function
called mlx5e_update_tc_and_tx_queues. This new function should
only be called when the channels number remains unchanged.

Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512124306.740898-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Enable 8 ports LAG

This patch adds to mlx5 drivers support for 8 ports HCAs.
Starting with ConnectX-8 HCAs with 8 ports are possible.

As most driver parts aren't affected by such configuration most driver
code is unchanged.

Specially the only affected areas are:
- Lag
- Multiport E-Switch
- Single FDB E-Switch

All of the above are already factored in generic way, and LAG and VF LAG
are tested, so all that left is to change a #define and remove checks
which are no longer needed.
However, Multiport E-Switch is not tested yet, so it is left untouched.

This patch will allow to create hardware LAG/VF LAG when all 8 ports are
added to the same bond device.

for example, In order to activate the hardware lag a user can execute
the following:

ip link add bond0 type bond
ip link set bond0 type bond miimon 100 mode 2
ip link set eth2 master bond0
ip link set eth3 master bond0
ip link set eth4 master bond0
ip link set eth5 master bond0
ip link set eth6 master bond0
ip link set eth7 master bond0
ip link set eth8 master bond0
ip link set eth9 master bond0

Where eth2, eth3, eth4, eth5, eth6, eth7, eth8 and eth9 are the PFs of
the same HCA.

Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512124306.740898-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

test: hsr: Extend the hsr_redbox.sh to have more SAN devices connected

After this change the single SAN device (ns3eth1) is now replaced with
two SAN devices - respectively ns4eth1 and ns5eth1.

It is possible to extend this script to have more SAN devices connected
by adding them to ns3br1 bridge.

Signed-off-by: Lukasz Majewski <lukma@denx.de>
Link: https://lore.kernel.org/r/20240510143710.3916631-1-lukma@denx.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-dsa-microchip-dcb-fixes'

Oleksij Rempel says:

====================
net: dsa: microchip: DCB fixes

This patch series address recommendation to rename IPV to IPM to avoid
confusion with IPV name used in 802.1Qci PSFP. And restores default "PCP
only" configuration as source of priorities to avoid possible
regressions.
====================

Link: https://lore.kernel.org/r/20240510053828.2412516-1-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: microchip: dcb: set default apptrust to PCP only

Before DCB support, the KSZ driver had only PCP as source of packet
priority values. To avoid regressions, make PCP only as default value.
User will need enable DSCP support manually.

This patch do not affect other KSZ8 related quirks. User will still be
warned by setting not support configurations for the port 2.

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Acked-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240510053828.2412516-4-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: microchip: dcb: add comments for DSCP related functions

All other functions are commented. Add missing comments to following
functions:
ksz_set_global_dscp_entry()
ksz_port_add_dscp_prio()
ksz_port_del_dscp_prio()

Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Acked-by: Arun Ramadoss <arun.ramadoss@microchip.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240510053828.2412516-3-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: microchip: dcb: rename IPV to IPM

IPV is added and used term in 802.1Qci PSFP and merged into 802.1Q (from
802.1Q-2018) for another functions.

Even it does similar operation holding temporal priority value
internally (as it is named), because KSZ datasheet doesn't use the term
of IPV (Internal Priority Value) and avoiding any confusion later when
PSFP is in the Linux world, it is better to rename IPV to IPM (Internal
Priority Mapping).

In addition, LAN937x documentation already use IPV for 802.1Qci PSFP
related functionality.

Suggested-by: Woojung Huh <Woojung.Huh@microchip.com>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Woojung Huh <woojung.huh@microchip.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240510053828.2412516-2-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

l2tp: Support different protocol versions with same IP/port quadruple

628bc3e5a1be ("l2tp: Support several sockets with same IP/port quadruple")
added support for several L2TPv2 tunnels using the same IP/port quadruple,
but if an L2TPv3 socket exists it could eat all the trafic. We thus have to
first use the version from the packet to get the proper tunnel, and only
then check that the version matches.

Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Reviewed-by: James Chapman <jchapman@katalix.com>
Link: https://lore.kernel.org/r/20240509205812.4063198-1-samuel.thibault@ens-lyon.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ynl: ensure exact-len value is resolved

For type String and Binary we are currently usinig the exact-len
limit value as is without attempting any name resolution.
However, the spec may specify the name of a constant rather than an
actual value, which would result in using the constant name as is
and thus break the policy.

Ensure the limit value is passed to get_limit(), which will always
attempt resolving the name before printing the policy rule.

Signed-off-by: Antonio Quartulli <a@unstable.cc>
Link: https://lore.kernel.org/r/20240510232202.24051-1-a@unstable.cc
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'add-tx-stop-wake-counters'

Daniel Jurgens says:

====================
Add TX stop/wake counters

Several drivers provide TX stop and wake counters via ethtool stats. Add
those to the netdev queue stats, and use them in virtio_net.
====================

Link: https://lore.kernel.org/r/20240510201927.1821109-1-danielj@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

virtio_net: Add TX stopped and wake counters

Add a tx queue stop and wake counters, they are useful for debugging.

$ ./tools/net/ynl/cli.py --spec netlink/specs/netdev.yaml \
--dump qstats-get --json '{"scope": "queue"}'
...
 {'ifindex': 13,
  'queue-id': 0,
  'queue-type': 'tx',
  'tx-bytes': 14756682850,
  'tx-packets': 226465,
  'tx-stop': 113208,
  'tx-wake': 113208},
 {'ifindex': 13,
  'queue-id': 1,
  'queue-type': 'tx',
  'tx-bytes': 18167675008,
  'tx-packets': 278660,
  'tx-stop': 8632,
  'tx-wake': 8632}]

Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Link: https://lore.kernel.org/r/20240510201927.1821109-3-danielj@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

netdev: Add queue stats for TX stop and wake

TX queue stop and wake are counted by some drivers.
Support reporting these via netdev-genl queue stats.

Signed-off-by: Daniel Jurgens <danielj@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Link: https://lore.kernel.org/r/20240510201927.1821109-2-danielj@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: socket option to check for MPTCP fallback to TCP

A way for an application to know if an MPTCP connection fell back to TCP
is to use getsockopt(MPTCP_INFO) and look for errors. The issue with
this technique is that the same errors -- EOPNOTSUPP (IPv4) and
ENOPROTOOPT (IPv6) -- are returned if there was a fallback, *or* if the
kernel doesn't support this socket option. The userspace then has to
look at the kernel version to understand what the errors mean.

It is not clean, and it doesn't take into account older kernels where
the socket option has been backported. A cleaner way would be to expose
this info to the TCP socket level. In case of MPTCP socket where no
fallback happened, the socket options for the TCP level will be handled
in MPTCP code, in mptcp_getsockopt_sol_tcp(). If not, that will be in
TCP code, in do_tcp_getsockopt(). So MPTCP simply has to set the value
1, while TCP has to set 0.

If the socket option is not supported, one of these two errors will be
reported:
- EOPNOTSUPP (95 - Operation not supported) for MPTCP sockets
- ENOPROTOOPT (92 - Protocol not available) for TCP sockets, e.g. on the
  socket received after an 'accept()', when the client didn't request to
  use MPTCP: this socket will be a TCP one, even if the listen socket
  was an MPTCP one.

With this new option, the kernel can return a clear answer to both "Is
this kernel new enough to tell me the fallback status?" and "If it is
new enough, is it currently a TCP or MPTCP socket?" questions, while not
breaking the previous method.

Acked-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://lore.kernel.org/r/20240509-upstream-net-next-20240509-mptcp-tcp_is_mptcp-v1-1-f846df999202@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-gro-remove-network_header-use-move-p-flush-flush_id-calculations-to-l4'

Richard Gobert says:

====================
net: gro: remove network_header use, move p->{flush/flush_id} calculations to L4

The cb fields network_offset and inner_network_offset are used instead of
skb->network_header throughout GRO.

These fields are then leveraged in the next commit to remove flush_id state
from napi_gro_cb, and stateful code in {ipv6,inet}_gro_receive which may be
unnecessarily complicated due to encapsulation support in GRO. These fields
are checked in L4 instead.

3rd patch adds tests for different flush_id flows in GRO.
====================

Link: https://lore.kernel.org/r/20240509190819.2985-1-richardbgobert@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests/net: add flush id selftests

Added flush id selftests to test different cases where DF flag is set or
unset and id value changes in the following packets. All cases where the
packets should coalesce or should not coalesce are tested.

Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240509190819.2985-4-richardbgobert@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: gro: move L3 flush checks to tcp_gro_receive and udp_gro_receive_segment

{inet,ipv6}_gro_receive functions perform flush checks (ttl, flags,
iph->id, ...) against all packets in a loop. These flush checks are used in
all merging UDP and TCP flows.

These checks need to be done only once and only against the found p skb,
since they only affect flush and not same_flow.

This patch leverages correct network header offsets from the cb for both
outer and inner network headers - allowing these checks to be done only
once, in tcp_gro_receive and udp_gro_receive_segment. As a result,
NAPI_GRO_CB(p)->flush is not used at all. In addition, flush_id checks are
more declarative and contained in inet_gro_flush, thus removing the need
for flush_id in napi_gro_cb.

This results in less parsing code for non-loop flush tests for TCP and UDP
flows.

To make sure results are not within noise range - I've made netfilter drop
all TCP packets, and measured CPU performance in GRO (in this case GRO is
responsible for about 50% of the CPU utilization).

perf top while replaying 64 parallel IP/TCP streams merging in GRO:
(gro_receive_network_flush is compiled inline to tcp_gro_receive)
net-next:
        6.94% [kernel] [k] inet_gro_receive
        3.02% [kernel] [k] tcp_gro_receive

patch applied:
        4.27% [kernel] [k] tcp_gro_receive
        4.22% [kernel] [k] inet_gro_receive

perf top while replaying 64 parallel IP/IP/TCP streams merging in GRO (same
results for any encapsulation, in this case inet_gro_receive is top
offender in net-next)
net-next:
        10.09% [kernel] [k] inet_gro_receive
        2.08% [kernel] [k] tcp_gro_receive

patch applied:
        6.97% [kernel] [k] inet_gro_receive
        3.68% [kernel] [k] tcp_gro_receive

Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240509190819.2985-3-richardbgobert@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: gro: use cb instead of skb->network_header

This patch converts references of skb->network_header to napi_gro_cb's
network_offset and inner_network_offset.

Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240509190819.2985-2-richardbgobert@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'ena-driver-changes-may-2024'

David Arinzon says:

====================
ENA driver changes May 2024

This patchset contains several misc and minor
changes to the ENA driver.
====================

Link: https://lore.kernel.org/r/20240512134637.25299-1-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ena: Change initial rx_usec interval

For the purpose of obtaining better CPU utilization,
minimum rx moderation interval is set to 20 usec.

Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512134637.25299-6-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ena: Changes around strscpy calls

strscpy copies as much of the string as possible,
meaning that the destination string will be truncated
in case of no space. As this is a non-critical error in
our case, adding a debug level print for indication.

This patch also removes a -1 which was added to ensure
enough space for NUL, but strscpy destination string is
guaranteed to be NUL-terminted, therefore, the -1 is
not needed.

Signed-off-by: David Arinzon <darinzon@amazon.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512134637.25299-5-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ena: Add validation for completion descriptors consistency

Validate that `first` flag is set only for the first
descriptor in multi-buffer packets.
In case of an invalid descriptor, a reset will occur.
A new reset reason for RX data corruption has been added.

Signed-off-by: Shahar Itzko <itzko@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512134637.25299-4-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ena: Reduce holes in ena_com structures

This patch makes two changes in order to fill holes and
reduce ther overall size of the structures ena_com_dev
and ena_com_rx_ctx.

Signed-off-by: Shahar Itzko <itzko@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512134637.25299-3-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ena: Add a counter for driver's reset failures

This patch adds a counter to the ena_adapter struct in
order to keep track of reset failures.
The counter is incremented every time either ena_restore_device()
or ena_destroy_device() fail.

Signed-off-by: Osama Abboud <osamaabb@amazon.com>
Signed-off-by: David Arinzon <darinzon@amazon.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240512134637.25299-2-darinzon@amazon.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: netfilter: nft_flowtable.sh: bump socat timeout to 1m

Now that this test runs in netdev CI it looks like 10s isn't enough
for debug kernels:
  selftests: net/netfilter: nft_flowtable.sh
  2024/05/10 20:33:08 socat[12204] E write(7, 0x563feb16a000, 8192): Broken pipe
  FAIL: file mismatch for ns1 -> ns2
  -rw------- 1 root root 37345280 May 10 20:32 /tmp/tmp.Am0yEHhNqI
 ...

Looks like socat gets zapped too quickly, so increase timeout to 1m.

Could also reduce tx file size for KSFT_MACHINE_SLOW, but its preferrable
to have same test for both debug and nondebug.

Signed-off-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/20240511064814.561525-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: net: use upstream mtools

Joachim kindly merged the IPv6 support in
https://github.com/troglobit/mtools/pull/2, so we can just use his
version now. A few more fixes subsequently came in for IPv6, so even
better.

Check that the deployed mtools version is 3.0 or above. Note that the
version check breaks compatibility with my fork where I didn't bump the
version, but I assume that won't be a problem.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20240510112856.1262901-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftest: epoll_busy_poll: Fix spelling mistake "couldnt" -> "couldn't"

There is a spelling mistake in a TH_LOG message. Fix it.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240510084811.3299685-1-colin.i.king@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: phy: air_en8811h: reset netdev rules when LED is set manually

Setting LED_OFF via brightness_set should deactivate hw control, so make
sure netdev trigger rules also get cleared in that case.
This fixes unwanted restoration of the default netdev trigger rules and
matches the behaviour when using the 'netdev' trigger without any
hardware offloading.

Fixes: 71e79430117d ("net: phy: air_en8811h: Add the Airoha EN8811H PHY driver")
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Link: https://lore.kernel.org/r/5ed8ea615890a91fa4df59a7ae8311bbdf63cdcf.1715248281.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'nf-next-24-05-12' of git://git./linux/kernel/git/netfilter/nf-next

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next:

Patch #1 skips transaction if object type provides no .update interface.

Patch #2 skips NETDEV_CHANGENAME which is unused.

Patch #3 enables conntrack to handle Multicast Router Advertisements and
 Multicast Router Solicitations from the Multicast Router Discovery
 protocol (RFC4286) as untracked opposed to invalid packets.
 From Linus Luessing.

Patch #4 updates DCCP conntracker to mark invalid as invalid, instead of
 dropping them, from Jason Xing.

Patch #5 uses NF_DROP instead of -NF_DROP since NF_DROP is 0,
 also from Jason.

Patch #6 removes reference in netfilter's sysctl documentation on pickup
 entries which were already removed by Florian Westphal.

Patch #7 removes check for IPS_OFFLOAD flag to disable early drop which
 allows to evict entries from the conntrack table,
 also from Florian.

Patches #8 to #16 updates nf_tables pipapo set backend to allocate
 the datastructure copy on-demand from preparation phase,
 to better deal with OOM situations where .commit step is too late
 to fail. Series from Florian Westphal.

Patch #17 adds a selftest with packetdrill to cover conntrack TCP state
 transitions, also from Florian.

Patch #18 use GFP_KERNEL to clone elements from control plane to avoid
 quick atomic reserves exhaustion with large sets, reporter refers
 to million entries magnitude.

* tag 'nf-next-24-05-12' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf-next:
  netfilter: nf_tables: allow clone callbacks to sleep
  selftests: netfilter: add packetdrill based conntrack tests
  netfilter: nft_set_pipapo: remove dirty flag
  netfilter: nft_set_pipapo: move cloning of match info to insert/removal path
  netfilter: nft_set_pipapo: prepare pipapo_get helper for on-demand clone
  netfilter: nft_set_pipapo: merge deactivate helper into caller
  netfilter: nft_set_pipapo: prepare walk function for on-demand clone
  netfilter: nft_set_pipapo: prepare destroy function for on-demand clone
  netfilter: nft_set_pipapo: make pipapo_clone helper return NULL
  netfilter: nft_set_pipapo: move prove_locking helper around
  netfilter: conntrack: remove flowtable early-drop test
  netfilter: conntrack: documentation: remove reference to non-existent sysctl
  netfilter: use NF_DROP instead of -NF_DROP
  netfilter: conntrack: dccp: try not to drop skb in conntrack
  netfilter: conntrack: fix ct-state for ICMPv6 Multicast Router Discovery
  netfilter: nf_tables: remove NETDEV_CHANGENAME from netdev chain event handler
  netfilter: nf_tables: skip transaction if update object is not implemented
====================

Link: https://lore.kernel.org/r/20240512161436.168973-1-pablo@netfilter.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

bpf: make list_for_each_entry portable

[Changes from V1:
- The __compat_break has been abandoned in favor of
  a more readable can_loop macro that can be used anywhere, including
  loop conditions.]

The macro list_for_each_entry is defined in bpf_arena_list.h as
follows:

  #define list_for_each_entry(pos, head, member) \
for (void * ___tmp = (pos = list_entry_safe((head)->first, \
    typeof(*(pos)), member), \
      (void *)0); \
     pos && ({ ___tmp = (void *)pos->member.next; 1; }); \
     cond_break, \
     pos = list_entry_safe((void __arena *)___tmp, typeof(*(pos)), member))

The macro cond_break, in turn, expands to a statement expression that
contains a `break' statement.  Compound statement expressions, and the
subsequent ability of placing statements in the header of a `for'
loop, are GNU extensions.

Unfortunately, clang implements this GNU extension differently than
GCC:

- In GCC the `break' statement is bound to the containing "breakable"
  context in which the defining `for' appears.  If there is no such
  context, GCC emits a warning: break statement without enclosing `for'
  o `switch' statement.

- In clang the `break' statement is bound to the defining `for'.  If
  the defining `for' is itself inside some breakable construct, then
  clang emits a -Wgcc-compat warning.

This patch adds a new macro can_loop to bpf_experimental, that
implements the same logic than cond_break but evaluates to a boolean
expression.  The patch also changes all the current instances of usage
of cond_break withing the header of loop accordingly.

Tested in bpf-next master.
No regressions.

Signed-off-by: Jose E. Marchesi <jose.marchesi@oracle.com>
Cc: david.faust@oracle.com
Cc: cupertino.miranda@oracle.com
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Link: https://lore.kernel.org/r/20240511212243.23477-1-jose.marchesi@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

bpf: ignore expected GCC warning in test_global_func10.c

The BPF selftest global_func10 in progs/test_global_func10.c contains:

  struct Small {
   long x;
  };

  struct Big {
   long x;
   long y;
  };

  [...]

  __noinline int foo(const struct Big *big)
  {
if (!big)
return 0;

return bpf_get_prandom_u32() < big->y;
  }

  [...]

  SEC("cgroup_skb/ingress")
  __failure __msg("invalid indirect access to stack")
  int global_func10(struct __sk_buff *skb)
  {
const struct Small small = {.x = skb->len };

return foo((struct Big *)&small) ? 1 : 0;
  }

GCC emits a "maybe uninitialized" warning for the code above, because
it knows `foo' accesses `big->y'.

Since the purpose of this selftest is to check that the verifier will
fail on this sort of invalid memory access, this patch just silences
the compiler warning.

Tested in bpf-next master.
No regressions.

Signed-off-by: Jose E. Marchesi <jose.marchesi@oracle.com>
Cc: david.faust@oracle.com
Cc: cupertino.miranda@oracle.com
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240511212349.23549-1-jose.marchesi@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

bpf: disable strict aliasing in test_global_func9.c

The BPF selftest test_global_func9.c performs type punning and breaks
srict-aliasing rules.

In particular, given:

  int global_func9(struct __sk_buff *skb)
  {
int result = 0;

[...]
{
const struct C c = {.x = skb->len, .y = skb->family };

result |= foo((const struct S *)&c);
}
  }

When building with strict-aliasing enabled (the default) the
initialization of `c' gets optimized away in its entirely:

[... no initialization of `c' ...]
r1 = r10
r1 += -40
call foo
w0 |= w6

Since GCC knows that `foo' accesses s->x, we get a "maybe
uninitialized" warning.

On the other hand, when strict-aliasing is disabled GCC only optimizes
away the store to `.y':

r1 = *(u32 *) (r6+0)
*(u32 *) (r10+-40) = r1  ; This is .x = skb->len in `c'
r1 = r10
r1 += -40
call foo
w0 |= w6

In this case the warning is not emitted, because s-> is initialized.

This patch disables strict aliasing in this test when building with
GCC.  clang seems to not optimize this particular code even when
strict aliasing is enabled.

Tested in bpf-next master.

Signed-off-by: Jose E. Marchesi <jose.marchesi@oracle.com>
Cc: david.faust@oracle.com
Cc: cupertino.miranda@oracle.com
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20240511212213.23418-1-jose.marchesi@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Free strdup memory in xdp_hw_metadata

The strdup() function returns a pointer to a new string which is a
duplicate of the string "ifname". Memory for the new string is obtained
with malloc(), and need to be freed with free().

This patch adds this missing "free(saved_hwtstamp_ifname)" in cleanup()
to avoid a potential memory leak in xdp_hw_metadata.c.

Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Link: https://lore.kernel.org/r/af9bcccb96655e82de5ce2b4510b88c9c8ed5ed0.1715417367.git.tanggeliang@kylinos.cn
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Fix a few tests for GCC related warnings.

This patch corrects a few warnings to allow selftests to compile for
GCC.

-- progs/cpumask_failure.c --

progs/bpf_misc.h:136:22: error: ‘cpumask’ is used uninitialized
[-Werror=uninitialized]
  136 | #define __sink(expr) asm volatile("" : "+g"(expr))
      |                      ^~~
progs/cpumask_failure.c:68:9: note: in expansion of macro ‘__sink’
   68 |         __sink(cpumask);

The macro __sink(cpumask) with the '+' contraint modifier forces the
the compiler to expect a read and write from cpumask. GCC detects
that cpumask is never initialized and reports an error.
This patch removes the spurious non required definitions of cpumask.

-- progs/dynptr_fail.c --

progs/dynptr_fail.c:1444:9: error: ‘ptr1’ may be used uninitialized
[-Werror=maybe-uninitialized]
 1444 |         bpf_dynptr_clone(&ptr1, &ptr2);

Many of the tests in the file are related to the detection of
uninitialized pointers by the verifier. GCC is able to detect possible
uninitialized values, and reports this as an error.
The patch initializes all of the previous uninitialized structs.

-- progs/test_tunnel_kern.c --

progs/test_tunnel_kern.c:590:9: error: array subscript 1 is outside
array bounds of ‘struct geneve_opt[1]’ [-Werror=array-bounds=]
  590 |         *(int *) &gopt.opt_data = bpf_htonl(0xdeadbeef);
      |         ^~~~~~~~~~~~~~~~~~~~~~~
progs/test_tunnel_kern.c:575:27: note: at offset 4 into object ‘gopt’ of
size 4
  575 |         struct geneve_opt gopt;

This tests accesses beyond the defined data for the struct geneve_opt
which contains as last field "u8 opt_data[0]" which clearly does not get
reserved space (in stack) in the function header. This pattern is
repeated in ip6geneve_set_tunnel and geneve_set_tunnel functions.
GCC is able to see this and emits a warning.
The patch introduces a local struct that allocates enough space to
safely allow the write to opt_data field.

-- progs/jeq_infer_not_null_fail.c --

progs/jeq_infer_not_null_fail.c:21:40: error: array subscript ‘struct
bpf_map[0]’ is partly outside array bounds of ‘struct <anonymous>[1]’
[-Werror=array-bounds=]
   21 |         struct bpf_map *inner_map = map->inner_map_meta;
      |                                        ^~
progs/jeq_infer_not_null_fail.c:14:3: note: object ‘m_hash’ of size 32
   14 | } m_hash SEC(".maps");

This example defines m_hash in the context of the compilation unit and
casts it to struct bpf_map which is much smaller than the size of struct
bpf_map. It errors out in GCC when it attempts to access an element that
would be defined in struct bpf_map outsize of the defined limits for
m_hash.
This patch disables the warning through a GCC pragma.

This changes were tested in bpf-next master selftests without any
regressions.

Signed-off-by: Cupertino Miranda <cupertino.miranda@oracle.com>
Cc: jose.marchesi@oracle.com
Cc: david.faust@oracle.com
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Link: https://lore.kernel.org/r/20240510183850.286661-2-cupertino.miranda@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

bpf: avoid gcc overflow warning in test_xdp_vlan.c

This patch fixes an integer overflow warning raised by GCC in
xdp_prognum1 of progs/test_xdp_vlan.c:

  GCC-BPF  [test_maps] test_xdp_vlan.bpf.o
progs/test_xdp_vlan.c: In function 'xdp_prognum1':
progs/test_xdp_vlan.c:163:25: error: integer overflow in expression
 '(short int)(((__builtin_constant_p((int)vlan_hdr->h_vlan_TCI)) != 0
   ? (int)(short unsigned int)((short int)((int)vlan_hdr->h_vlan_TCI
   << 8 >> 8) << 8 | (short int)((int)vlan_hdr->h_vlan_TCI << 0 >> 8
   << 0)) & 61440 : (int)__builtin_bswap16(vlan_hdr->h_vlan_TCI)
   & 61440) << 8 >> 8) << 8' of type 'short int' results in '0' [-Werror=overflow]
  163 |                         bpf_htons((bpf_ntohs(vlan_hdr->h_vlan_TCI) & 0xf000)
      |                         ^~~~~~~~~

The problem lies with the expansion of the bpf_htons macro and the
expression passed into it.  The bpf_htons macro (and similarly the
bpf_ntohs macro) expand to a ternary operation using either
__builtin_bswap16 or ___bpf_swab16 to swap the bytes, depending on
whether the expression is constant.

For an expression, with 'value' as a u16, like:

  bpf_htons (value & 0xf000)

The entire (value & 0xf000) is 'x' in the expansion of ___bpf_swab16
and we get as one part of the expanded swab16:

  ((__u16)(value & 0xf000) << 8 >> 8 << 8

This will always evaluate to 0, which is intentional since this
subexpression deals with the byte guaranteed to be 0 by the mask.

However, GCC warns because the precise reason this always evaluates to 0
is an overflow.  Specifically, the plain 0xf000 in the expression is a
signed 32-bit integer, which causes 'value' to also be promoted to a
signed 32-bit integer, and the combination of the 8-bit left shift and
down-cast back to __u16 results in a signed overflow (really a 'warning:
overflow in conversion from int to __u16' which is propegated up through
the rest of the expression leading to the ultimate overflow warning
above), which is a valid warning despite being the intended result of
this code.

Clang does not warn on this case, likely because it performs constant
folding later in the compilation process relative to GCC.  It seems that
by the time clang does constant folding for this expression, the side of
the ternary with this overflow has already been discarded.

Fortunately, this warning is easily silenced by simply making the 0xf000
mask explicitly unsigned.  This has no impact on the result.

Signed-off-by: David Faust <david.faust@oracle.com>
Cc: jose.marchesi@oracle.com
Cc: cupertino.miranda@oracle.com
Cc: Eduard Zingerman <eddyz87@gmail.com>
Cc: Yonghong Song <yonghong.song@linux.dev>
Link: https://lore.kernel.org/r/20240508193512.152759-1-david.faust@oracle.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

tools: remove redundant ethtool.h from tooling infra

Remove the redundant ethtool.h header file from tools/include/uapi/linux.
The file is unnecessary as the system uses the kernel's
include/uapi/linux/ethtool.h directly.

Signed-off-by: Tushar Vyavahare <tushar.vyavahare@intel.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20240508104123.434769-1-tushar.vyavahare@intel.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Merge branch 'retire-progs-test_sock_addr'

Jordan Rife says:

====================
Retire progs/test_sock_addr.c

This patch series migrates remaining tests from bpf/test_sock_addr.c to
prog_tests/sock_addr.c and progs/verifier_sock_addr.c in order to fully
retire the old-style test program and expands test coverage to test
previously untested scenarios related to sockaddr hooks.

This is a continuation of the work started recently during the expansion
of prog_tests/sock_addr.c.

Link: https://lore.kernel.org/bpf/20240429214529.2644801-1-jrife@google.com/T/#u
=======
Patches
=======
* Patch 1 moves tests that check valid return values for recvmsg hooks
  into progs/verifier_sock_addr.c, a new addition to the verifier test
  suite.
* Patches 2-5 lay the groundwork for test migration, enabling
  prog_tests/sock_addr.c to handle more test dimensions.
* Patches 6-11 move existing tests to prog_tests/sock_addr.c.
* Patch 12 removes some redundant test cases.
* Patches 14-17 expand on existing test coverage.
====================

Link: https://lore.kernel.org/r/20240510190246.3247730-1-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Expand ATTACH_REJECT tests

This expands coverage for ATTACH_REJECT tests to include connect_unix,
sendmsg_unix, recvmsg*, getsockname*, and getpeername*.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-18-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Expand getsockname and getpeername tests

This expands coverage for getsockname and getpeername hooks to include
getsockname4, getsockname6, getpeername4, and getpeername6.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-17-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

sefltests/bpf: Expand sockaddr hook deny tests

This patch expands test coverage for EPERM tests to include connect and
bind calls and rounds out the coverage for sendmsg by adding tests for
sendmsg_unix.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-16-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Expand sockaddr program return value tests

This patch expands verifier coverage for program return values to cover
bind, connect, sendmsg, getsockname, and getpeername hooks. It also
rounds out the recvmsg coverage by adding test cases for recvmsg_unix
hooks.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-15-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Retire test_sock_addr.(c|sh)

Fully remove test_sock_addr.c and test_sock_addr.sh, as test coverage
has been fully moved to prog_tests/sock_addr.c.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-14-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Remove redundant sendmsg test cases

Remove these test cases completely, as the same behavior is already
covered by other sendmsg* test cases in prog_tests/sock_addr.c. This
just rewrites the destination address similar to sendmsg_v4_prog and
sendmsg_v6_prog.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-13-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate ATTACH_REJECT test cases

Migrate test case from bpf/test_sock_addr.c ensuring that program
attachment fails when using an inappropriate attach type.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-12-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate expected_attach_type tests

Migrates tests from progs/test_sock_addr.c ensuring that programs fail
to load when the expected attach type does not match.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-11-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate wildcard destination rewrite test

Migrate test case from bpf/test_sock_addr.c ensuring that sendmsg
respects when sendmsg6 hooks rewrite the destination IP with the IPv6
wildcard IP, [::].

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-10-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate sendmsg6 v4 mapped address tests

Migrate test case from bpf/test_sock_addr.c ensuring that sendmsg
returns -ENOTSUPP when sending to an IPv4-mapped IPv6 address to
prog_tests/sock_addr.c.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-9-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate sendmsg deny test cases

This set of tests checks that sendmsg calls are rejected (return -EPERM)
when the sendmsg* hook returns 0. Replace those in bpf/test_sock_addr.c
with corresponding tests in prog_tests/sock_addr.c.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-8-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate WILDCARD_IP test

Move wildcard IP sendmsg test case out of bpf/test_sock_addr.c into
prog_tests/sock_addr.c.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-7-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Handle SYSCALL_EPERM and SYSCALL_ENOTSUPP test cases

In preparation to move test cases from bpf/test_sock_addr.c that expect
system calls to return ENOTSUPP or EPERM, this patch propagates errno
from relevant system calls up to test_sock_addr() where the result can
be checked.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-6-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Handle ATTACH_REJECT test cases

In preparation to move test cases from bpf/test_sock_addr.c that expect
ATTACH_REJECT, this patch adds BPF_SKEL_FUNCS_RAW to generate load and
destroy functions that use bpf_prog_attach() to control the attach_type.

The normal load functions use bpf_program__attach_cgroup which does not
have the same degree of control over the attach type, as
bpf_program_attach_fd() calls bpf_link_create() with the attach type
extracted from prog using bpf_program__expected_attach_type(). It is
currently not possible to modify the attach type before
bpf_program__attach_cgroup() is called, since
bpf_program__set_expected_attach_type() has no effect after the program
is loaded.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-5-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Handle LOAD_REJECT test cases

In preparation to move test cases from bpf/test_sock_addr.c that expect
LOAD_REJECT, this patch adds expected_attach_type and extends load_fn to
accept an expected attach type and a flag indicating whether or not
rejection is expected.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-4-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Use program name for skel load/destroy functions

In preparation to migrate tests from bpf/test_sock_addr.c to
sock_addr.c, update BPF_SKEL_FUNCS so that it generates functions
based on prog_name instead of skel_name. This allows us to differentiate
between programs in the same skeleton.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-3-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

selftests/bpf: Migrate recvmsg* return code tests to verifier_sock_addr.c

This set of tests check that the BPF verifier rejects programs with
invalid return codes (recvmsg4 and recvmsg6 hooks can only return 1).
This patch replaces the tests in test_sock_addr.c with
verifier_sock_addr.c, a new verifier prog_tests for sockaddr hooks, in a
step towards fully retiring test_sock_addr.c.

Signed-off-by: Jordan Rife <jrife@google.com>
Link: https://lore.kernel.org/r/20240510190246.3247730-2-jrife@google.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

riscv, bpf: make some atomic operations fully ordered

The BPF atomic operations with the BPF_FETCH modifier along with
BPF_XCHG and BPF_CMPXCHG are fully ordered but the RISC-V JIT implements
all atomic operations except BPF_CMPXCHG with relaxed ordering.

Section 8.1 of the "The RISC-V Instruction Set Manual Volume I:
Unprivileged ISA" [1], titled, "Specifying Ordering of Atomic
Instructions" says:

| To provide more efficient support for release consistency [5], each
| atomic instruction has two bits, aq and rl, used to specify additional
| memory ordering constraints as viewed by other RISC-V harts.

and

| If only the aq bit is set, the atomic memory operation is treated as
| an acquire access.
| If only the rl bit is set, the atomic memory operation is treated as a
| release access.
|
| If both the aq and rl bits are set, the atomic memory operation is
| sequentially consistent.

Fix this by setting both aq and rl bits as 1 for operations with
BPF_FETCH and BPF_XCHG.

[1] https://riscv.org/wp-content/uploads/2017/05/riscv-spec-v2.2.pdf

Fixes: dd642ccb45ec ("riscv, bpf: Implement more atomic operations for RV64")
Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Reviewed-by: Pu Lehui <pulehui@huawei.com>
Link: https://lore.kernel.org/r/20240505201633.123115-1-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

riscv, bpf: Fix typo in comment

We can use either "instruction" or "insn" in the comment.

Signed-off-by: Xiao Wang <xiao.w.wang@intel.com>
Reviewed-by: Pu Lehui <pulehui@huawei.com>
Link: https://lore.kernel.org/r/20240507111618.437121-1-xiao.w.wang@intel.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

s390/bpf: Emit a barrier for BPF_FETCH instructions

BPF_ATOMIC_OP() macro documentation states that "BPF_ADD | BPF_FETCH"
should be the same as atomic_fetch_add(), which is currently not the
case on s390x: the serialization instruction "bcr 14,0" is missing.
This applies to "and", "or" and "xor" variants too.

s390x is allowed to reorder stores with subsequent fetches from
different addresses, so code relying on BPF_FETCH acting as a barrier,
for example:

  stw [%r0], 1
  afadd [%r1], %r2
  ldxw %r3, [%r4]

may be broken. Fix it by emitting "bcr 14,0".

Note that a separate serialization instruction is not needed for
BPF_XCHG and BPF_CMPXCHG, because COMPARE AND SWAP performs
serialization itself.

Fixes: ba3b86b9cef0 ("s390/bpf: Implement new atomic ops")
Reported-by: Puranjay Mohan <puranjay12@gmail.com>
Closes: https://lore.kernel.org/bpf/mb61p34qvq3wf.fsf@kernel.org/
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Link: https://lore.kernel.org/r/20240507000557.12048-1-iii@linux.ibm.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Merge branch 'bpf-inline-helpers-in-arm64-and-riscv-jits'

Puranjay Mohan says:

====================
bpf: Inline helpers in arm64 and riscv JITs

Changes in v5 -> v6:
arm64 v5: https://lore.kernel.org/all/20240430234739.79185-1-puranjay@kernel.org/
riscv v2: https://lore.kernel.org/all/20240430175834.33152-1-puranjay@kernel.org/
- Combine riscv and arm64 changes in single series
- Some coding style fixes

Changes in v4 -> v5:
v4: https://lore.kernel.org/all/20240429131647.50165-1-puranjay@kernel.org/
- Implement the inlining of the bpf_get_smp_processor_id() in the JIT.

NOTE: This needs to be based on:
https://lore.kernel.org/all/20240430175834.33152-1-puranjay@kernel.org/
to be built.

Manual run of bpf-ci with this series rebased on above:
https://github.com/kernel-patches/bpf/pull/6929

Changes in v3 -> v4:
v3: https://lore.kernel.org/all/20240426121349.97651-1-puranjay@kernel.org/
- Fix coding style issue related to C89 standards.

Changes in v2 -> v3:
v2: https://lore.kernel.org/all/20240424173550.16359-1-puranjay@kernel.org/
- Fixed the xlated dump of percpu mov to "r0 = &(void __percpu *)(r0)"
- Made ARM64 and x86-64 use the same code for inlining. The only difference
  that remains is the per-cpu address of the cpu_number.

Changes in v1 -> v2:
v1: https://lore.kernel.org/all/20240405091707.66675-1-puranjay12@gmail.com/
- Add a patch to inline bpf_get_smp_processor_id()
- Fix an issue in MRS instruction encoding as pointed out by Will
- Remove CONFIG_SMP check because arm64 kernel always compiles with CONFIG_SMP

This series adds the support of internal only per-CPU instructions and inlines
the bpf_get_smp_processor_id() helper call for ARM64 and RISC-V BPF JITs.

Here is an example of calls to bpf_get_smp_processor_id() and
percpu_array_map_lookup_elem() before and after this series on ARM64.

                                         BPF
                                        =====
              BEFORE                                       AFTER
             --------                                     -------

int cpu = bpf_get_smp_processor_id();           int cpu = bpf_get_smp_processor_id();
(85) call bpf_get_smp_processor_id#229032       (85) call bpf_get_smp_processor_id#8

p = bpf_map_lookup_elem(map, &zero);            p = bpf_map_lookup_elem(map, &zero);
(18) r1 = map[id:78]                            (18) r1 = map[id:153]
(18) r2 = map[id:82][0]+65536                   (18) r2 = map[id:157][0]+65536
(85) call percpu_array_map_lookup_elem#313512   (07) r1 += 496
                                                (61) r0 = *(u32 *)(r2 +0)
                                                (35) if r0 >= 0x1 goto pc+5
                                                (67) r0 <<= 3
                                                (0f) r0 += r1
                                                (79) r0 = *(u64 *)(r0 +0)
                                                (bf) r0 = &(void __percpu *)(r0)
                                                (05) goto pc+1
                                                (b7) r0 = 0

                                      ARM64 JIT
                                     ===========

              BEFORE                                       AFTER
             --------                                     -------

int cpu = bpf_get_smp_processor_id();           int cpu = bpf_get_smp_processor_id();
mov     x10, #0xfffffffffffff4d0                mrs     x10, sp_el0
movk    x10, #0x802b, lsl #16                   ldr     w7, [x10, #24]
movk    x10, #0x8000, lsl #32
blr     x10
add     x7, x0, #0x0

p = bpf_map_lookup_elem(map, &zero);            p = bpf_map_lookup_elem(map, &zero);
mov     x0, #0xffff0003ffffffff                 mov     x0, #0xffff0003ffffffff
movk    x0, #0xce5c, lsl #16                    movk    x0, #0xe0f3, lsl #16
movk    x0, #0xca00                             movk    x0, #0x7c00
mov     x1, #0xffff8000ffffffff                 mov     x1, #0xffff8000ffffffff
movk    x1, #0x8bdb, lsl #16                    movk    x1, #0xb0c7, lsl #16
movk    x1, #0x6000                             movk    x1, #0xe000
mov     x10, #0xffffffffffff3ed0                add     x0, x0, #0x1f0
movk    x10, #0x802d, lsl #16                   ldr     w7, [x1]
movk    x10, #0x8000, lsl #32                   cmp     x7, #0x1
blr     x10                                     b.cs    0x0000000000000090
add     x7, x0, #0x0                            lsl     x7, x7, #3
                                                add     x7, x7, x0
                                                ldr     x7, [x7]
                                                mrs     x10, tpidr_el1
                                                add     x7, x7, x10
                                                b       0x0000000000000094
                                                mov     x7, #0x0

              Performance improvement found using benchmark[1]

./benchs/run_bench_trigger.sh glob-arr-inc arr-inc hash-inc

  +---------------+-------------------+-------------------+--------------+
  |      Name     |      Before       |        After      |   % change   |
  |---------------+-------------------+-------------------+--------------|
  | glob-arr-inc  | 23.380 ± 1.675M/s | 25.893 ± 0.026M/s |   + 10.74%   |
  | arr-inc       | 23.928 ± 0.034M/s | 25.213 ± 0.063M/s |   + 5.37%    |
  | hash-inc      | 12.352 ± 0.005M/s | 12.609 ± 0.013M/s |   + 2.08%    |
  +---------------+-------------------+-------------------+--------------+

[1] https://github.com/anakryiko/linux/commit/8dec900975ef

             RISCV64 JIT output for `call bpf_get_smp_processor_id`
            =======================================================

                  Before                           After
                 --------                         -------

           auipc   t1,0x848c                  ld    a5,32(tp)
           jalr    604(t1)
           mv      a5,a0

  Benchmark using [1] on Qemu.

  ./benchs/run_bench_trigger.sh glob-arr-inc arr-inc hash-inc

  +---------------+------------------+------------------+--------------+
  |      Name     |     Before       |       After      |   % change   |
  |---------------+------------------+------------------+--------------|
  | glob-arr-inc  | 1.077 ± 0.006M/s | 1.336 ± 0.010M/s |   + 24.04%   |
  | arr-inc       | 1.078 ± 0.002M/s | 1.332 ± 0.015M/s |   + 23.56%   |
  | hash-inc      | 0.494 ± 0.004M/s | 0.653 ± 0.001M/s |   + 32.18%   |
  +---------------+------------------+------------------+--------------+
====================

Link: https://lore.kernel.org/r/20240502151854.9810-1-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

bpf, arm64: inline bpf_get_smp_processor_id() helper

Inline calls to bpf_get_smp_processor_id() helper in the JIT by emitting
a read from struct thread_info. The SP_EL0 system register holds the
pointer to the task_struct and thread_info is the first member of this
struct. We can read the cpu number from the thread_info.

Here is how the ARM64 JITed assembly changes after this commit:

                                      ARM64 JIT
                                     ===========

              BEFORE                                    AFTER
             --------                                  -------

int cpu = bpf_get_smp_processor_id();        int cpu = bpf_get_smp_processor_id();

mov     x10, #0xfffffffffffff4d0             mrs     x10, sp_el0
movk    x10, #0x802b, lsl #16                ldr     w7, [x10, #24]
movk    x10, #0x8000, lsl #32
blr     x10
add     x7, x0, #0x0

               Performance improvement using benchmark[1]

./benchs/run_bench_trigger.sh glob-arr-inc arr-inc hash-inc

+---------------+-------------------+-------------------+--------------+
|      Name     |      Before       |        After      |   % change   |
|---------------+-------------------+-------------------+--------------|
| glob-arr-inc  | 23.380 ± 1.675M/s | 25.893 ± 0.026M/s |   + 10.74%   |
| arr-inc       | 23.928 ± 0.034M/s | 25.213 ± 0.063M/s |   + 5.37%    |
| hash-inc      | 12.352 ± 0.005M/s | 12.609 ± 0.013M/s |   + 2.08%    |
+---------------+-------------------+-------------------+--------------+

[1] https://github.com/anakryiko/linux/commit/8dec900975ef

Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20240502151854.9810-5-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

arm64, bpf: add internal-only MOV instruction to resolve per-CPU addrs

Support an instruction for resolving absolute addresses of per-CPU
data from their per-CPU offsets. This instruction is internal-only and
users are not allowed to use them directly. They will only be used for
internal inlining optimizations for now between BPF verifier and BPF
JITs.

Since commit 7158627686f0 ("arm64: percpu: implement optimised pcpu
access using tpidr_el1"), the per-cpu offset for the CPU is stored in
the tpidr_el1/2 register of that CPU.

To support this BPF instruction in the ARM64 JIT, the following ARM64
instructions are emitted:

mov dst, src // Move src to dst, if src != dst
mrs tmp, tpidr_el1/2 // Move per-cpu offset of the current cpu in tmp.
add dst, dst, tmp // Add the per cpu offset to the dst.

To measure the performance improvement provided by this change, the
benchmark in [1] was used:

Before:
glob-arr-inc   :   23.597 ± 0.012M/s
arr-inc        :   23.173 ± 0.019M/s
hash-inc       :   12.186 ± 0.028M/s

After:
glob-arr-inc   :   23.819 ± 0.034M/s
arr-inc        :   23.285 ± 0.017M/s
hash-inc       :   12.419 ± 0.011M/s

[1] https://github.com/anakryiko/linux/commit/8dec900975ef

Signed-off-by: Puranjay Mohan <puranjay12@gmail.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20240502151854.9810-4-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

riscv, bpf: inline bpf_get_smp_processor_id()

Inline the calls to bpf_get_smp_processor_id() in the riscv bpf jit.

RISCV saves the pointer to the CPU's task_struct in the TP (thread
pointer) register. This makes it trivial to get the CPU's processor id.
As thread_info is the first member of task_struct, we can read the
processor id from TP + offsetof(struct thread_info, cpu).

          RISCV64 JIT output for `call bpf_get_smp_processor_id`
  ======================================================

                Before                           After
               --------                         -------

         auipc   t1,0x848c                  ld    a5,32(tp)
         jalr    604(t1)
         mv      a5,a0

Benchmark using [1] on Qemu.

./benchs/run_bench_trigger.sh glob-arr-inc arr-inc hash-inc

+---------------+------------------+------------------+--------------+
|      Name     |     Before       |       After      |   % change   |
|---------------+------------------+------------------+--------------|
| glob-arr-inc  | 1.077 ± 0.006M/s | 1.336 ± 0.010M/s |   + 24.04%   |
| arr-inc       | 1.078 ± 0.002M/s | 1.332 ± 0.015M/s |   + 23.56%   |
| hash-inc      | 0.494 ± 0.004M/s | 0.653 ± 0.001M/s |   + 32.18%   |
+---------------+------------------+------------------+--------------+

NOTE: This benchmark includes changes from this patch and the previous
      patch that implemented the per-cpu insn.

[1] https://github.com/anakryiko/linux/commit/8dec900975ef

Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Björn Töpel <bjorn@kernel.org>
Link: https://lore.kernel.org/r/20240502151854.9810-3-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

riscv, bpf: add internal-only MOV instruction to resolve per-CPU addrs

Support an instruction for resolving absolute addresses of per-CPU
data from their per-CPU offsets. This instruction is internal-only and
users are not allowed to use them directly. They will only be used for
internal inlining optimizations for now between BPF verifier and BPF
JITs.

RISC-V uses generic per-cpu implementation where the offsets for CPUs
are kept in an array called __per_cpu_offset[cpu_number]. RISCV stores
the address of the task_struct in TP register. The first element in
task_struct is struct thread_info, and we can get the cpu number by
reading from the TP register + offsetof(struct thread_info, cpu).

Once we have the cpu number in a register we read the offset for that
cpu from address: &__per_cpu_offset + cpu_number << 3. Then we add this
offset to the destination register.

To measure the improvement from this change, the benchmark in [1] was
used on Qemu:

Before:
glob-arr-inc   :    1.127 ± 0.013M/s
arr-inc        :    1.121 ± 0.004M/s
hash-inc       :    0.681 ± 0.052M/s

After:
glob-arr-inc   :    1.138 ± 0.011M/s
arr-inc        :    1.366 ± 0.006M/s
hash-inc       :    0.676 ± 0.001M/s

[1] https://github.com/anakryiko/linux/commit/8dec900975ef

Signed-off-by: Puranjay Mohan <puranjay@kernel.org>
Acked-by: Björn Töpel <bjorn@kernel.org>
Link: https://lore.kernel.org/r/20240502151854.9810-2-puranjay@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

ARC: Add eBPF JIT support

This will add eBPF JIT support to the 32-bit ARCv2 processors. The
implementation is qualified by running the BPF tests on a Synopsys HSDK
board with "ARC HS38 v2.1c at 500 MHz" as the 4-core CPU.

The test_bpf.ko reports 2-10 fold improvements in execution time of its
tests. For instance:

test_bpf: #33 tcpdump port 22 jited:0 704 1766 2104 PASS
test_bpf: #33 tcpdump port 22 jited:1 120  224  260 PASS

test_bpf: #141 ALU_DIV_X: 4294967295 / 4294967295 = 1 jited:0 238 PASS
test_bpf: #141 ALU_DIV_X: 4294967295 / 4294967295 = 1 jited:1  23 PASS

test_bpf: #776 JMP32_JGE_K: all ... magnitudes jited:0 2034681 PASS
test_bpf: #776 JMP32_JGE_K: all ... magnitudes jited:1 1020022 PASS

Deployment and structure
------------------------
The related codes are added to "arch/arc/net":

- bpf_jit.h       -- The interface that a back-end translator must provide
- bpf_jit_core.c  -- Knows how to handle the input eBPF byte stream
- bpf_jit_arcv2.c -- The back-end code that knows the translation logic

The bpf_int_jit_compile() at the end of bpf_jit_core.c is the entrance
to the whole process. Normally, the translation is done in one pass,
namely the "normal pass". In case some relocations are not known during
this pass, some data (arc_jit_data) is allocated for the next pass to
come. This possible next (and last) pass is called the "extra pass".

1. Normal pass       # The necessary pass
     1a. Dry run       # Get the whole JIT length, epilogue offset, etc.
     1b. Emit phase    # Allocate memory and start emitting instructions
2. Extra pass        # Only needed if there are relocations to be fixed
     2a. Patch relocations

Support status
--------------
The JIT compiler supports BPF instructions up to "cpu=v4". However, it
does not yet provide support for:

- Tail calls
- Atomic operations
- 64-bit division/remainder
- BPF_PROBE_MEM* (exception table)

The result of "test_bpf" test suite on an HSDK board is:

hsdk-lnx# insmod test_bpf.ko test_suite=test_bpf

  test_bpf: Summary: 863 PASSED, 186 FAILED, [851/851 JIT'ed]

All the failing test cases are due to the ones that were not JIT'ed.
Categorically, they can be represented as:

  .-----------.------------.-------------.
  | test type |   opcodes  | # of cases  |
  |-----------+------------+-------------|
  | atomic    | 0xC3, 0xDB |         149 |
  | div64     | 0x37, 0x3F |          22 |
  | mod64     | 0x97, 0x9F |          15 |
  `-----------^------------+-------------|
                           | (total) 186 |
                           `-------------'

Setup: build config
-------------------
The following configs must be set to have a working JIT test:

  CONFIG_BPF_JIT=y
  CONFIG_BPF_JIT_ALWAYS_ON=y
  CONFIG_TEST_BPF=m

The following options are not necessary for the tests module,
but are good to have:

  CONFIG_DEBUG_INFO=y             # prerequisite for below
  CONFIG_DEBUG_INFO_BTF=y         # so bpftool can generate vmlinux.h

  CONFIG_FTRACE=y                 #
  CONFIG_BPF_SYSCALL=y            # all these options lead to
  CONFIG_KPROBE_EVENTS=y          # having CONFIG_BPF_EVENTS=y
  CONFIG_PERF_EVENTS=y            #

Some BPF programs provide data through /sys/kernel/debug:
  CONFIG_DEBUG_FS=y
arc# mount -t debugfs debugfs /sys/kernel/debug

Setup: elfutils
---------------
The libdw.{so,a} library that is used by pahole for processing
the final binary must come from elfutils 0.189 or newer. The
support for ARCv2 [1] has been added since that version.

[1]
https://sourceware.org/git/?p=elfutils.git;a=commit;h=de3d46b3e7

Setup: pahole
-------------
The line below in linux/scripts/Makefile.btf must be commented out:

pahole-flags-$(call test-ge, $(pahole-ver), 121) += --btf_gen_floats

Or else, the build will fail:

$ make V=1
  ...
  BTF     .btf.vmlinux.bin.o
pahole -J --btf_gen_floats                    \
       -j --lang_exclude=rust                 \
       --skip_encoding_btf_inconsistent_proto \
       --btf_gen_optimized .tmp_vmlinux.btf
Complex, interval and imaginary float types are not supported
Encountered error while encoding BTF.
  ...
  BTFIDS  vmlinux
./tools/bpf/resolve_btfids/resolve_btfids vmlinux
libbpf: failed to find '.BTF' ELF section in vmlinux
FAILED: load BTF from vmlinux: No data available

This is due to the fact that the ARC toolchains generate
"complex float" DIE entries in libgcc and at the moment, pahole
can't handle such entries.

Running the tests
-----------------
host$ scp /bld/linux/lib/test_bpf.ko arc:
arc # sysctl net.core.bpf_jit_enable=1
arc # insmod test_bpf.ko test_suite=test_bpf
      ...
      test_bpf: #1048 Staggered jumps: JMP32_JSLE_X jited:1 697811 PASS
      test_bpf: Summary: 863 PASSED, 186 FAILED, [851/851 JIT'ed]

Acknowledgments
---------------
- Claudiu Zissulescu for his unwavering support
- Yuriy Kolerov for testing and troubleshooting
- Vladimir Isaev for the pahole workaround
- Sergey Matyukevich for paving the road by adding the interpreter support

Signed-off-by: Shahab Vahedi <shahab@synopsys.com>
Link: https://lore.kernel.org/r/20240430145604.38592-1-list+bpf@vahedi.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Merge branch '40GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2024-05-08 (most Intel drivers)

This series contains updates to i40e, iavf, ice, igb, igc, e1000e, and ixgbe
drivers.

Asbjørn Sloth Tønnesen adds checks against supported flower control flags
for i40e, iavf, ice, and igb drivers.

Michal corrects filters removed during eswitch release for ice.

Corinna Vinschen defers PTP initialization to later in probe so that
netdev log entry is initialized on igc.

Ilpo Järvinen removes a couple of unused, duplicate defines on
e1000e and ixgbe.

* '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  net: e1000e & ixgbe: Remove PCI_HEADER_TYPE_MFD duplicates
  igc: fix a log entry using uninitialized netdev
  ice: remove correct filters during eswitch release
  igb: flower: validate control flags
  ice: flower: validate control flags
  iavf: flower: validate control flags
  i40e: flower: validate control flags
====================

Link: https://lore.kernel.org/r/20240508173342.2760994-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-qede-convert-filter-code-to-use-extack'

Asbjørn Sloth Tønnesen says:

====================
net: qede: convert filter code to use extack

This series converts the filter code in the qede driver
to use NL_SET_ERR_MSG_*(extack, ...) for error handling.

Patch 1-12 converts qede_parse_flow_attr() to use extack,
along with all it's static helper functions.

qede_parse_flow_attr() is used in two places:
- qede_add_tc_flower_fltr()
- qede_flow_spec_to_rule()

In the latter call site extack is faked in the same way as
is done in mlxsw (patch 12).

While the conversion is going on, some error messages are silenced
in between patch 1-12. If wanted could squash patch 1-12 in a v3, but
I felt that it would be easier to review as 12 more trivial patches.

Patch 13 and 14, finishes up by converting qede_parse_actions(),
and ensures that extack is propagated to it, in both call contexts.

v1: https://lore.kernel.org/netdev/20240507104421.1628139-1-ast@fiberby.net/
====================

Link: https://lore.kernel.org/r/20240508143404.95901-1-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_parse_actions()

Convert DP_NOTICE/DP_INFO to NL_SET_ERR_MSG_MOD.

Keep edev around for use with QEDE_RSS_COUNT().

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-15-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: propagate extack through qede_flow_spec_validate()

Pass extack to qede_flow_spec_validate() when called in
qede_flow_spec_to_rule().

Pass extack to qede_parse_actions().

Not converting qede_flow_spec_validate() to use extack for
errors, as it's only called from qede_flow_spec_to_rule(),
where extack is faked into a DP_NOTICE anyway, so opting to
keep DP_VERBOSE/DP_NOTICE usage.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-14-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use faked extack in qede_flow_spec_to_rule()

Since qede_parse_flow_attr() now does error reporting
through extack, then give it a fake extack and extract the
error message afterwards if one was set.

The extracted error message is then passed on through
DP_NOTICE(), including messages that was earlier issued
with DP_INFO().

This fake extack approach is already used by
mlxsw_env_linecard_modules_power_mode_apply() in
drivers/net/ethernet/mellanox/mlxsw/core_env.c

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-13-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_parse_flow_attr()

Convert qede_parse_flow_attr() to take extack,
and drop the edev argument.

Convert DP_NOTICE calls to use NL_SET_ERR_MSG_* instead.

Pass extack in calls to qede_flow_parse_{tcp,udp}_v{4,6}().

In calls to qede_parse_flow_attr(), if extack is
unavailable, then use NULL for now, until a
subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-12-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: add extack in qede_add_tc_flower_fltr()

Define extack locally, to reduce line lengths and aid future users.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-11-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_udp_v4()

Convert qede_flow_parse_udp_v4() to take extack,
and drop the edev argument.

Pass extack in call to qede_flow_parse_v4_common().

In call to qede_flow_parse_udp_v4(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-10-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_udp_v6()

Convert qede_flow_parse_udp_v6() to take extack,
and drop the edev argument.

Pass extack in call to qede_flow_parse_v6_common().

In call to qede_flow_parse_udp_v6(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-9-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_tcp_v4()

Convert qede_flow_parse_tcp_v4() to take extack,
and drop the edev argument.

Pass extack in call to qede_flow_parse_v4_common().

In call to qede_flow_parse_tcp_v4(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-8-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_tcp_v6()

Convert qede_flow_parse_tcp_v6() to take extack,
and drop the edev argument.

Pass extack in call to qede_flow_parse_v6_common().

In call to qede_flow_parse_tcp_v6(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-7-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_v4_common()

Convert qede_flow_parse_v4_common() to take extack,
and drop the edev argument.

Convert DP_NOTICE call to use NL_SET_ERR_MSG_MOD instead.

Pass extack in calls to qede_flow_parse_ports() and
qede_set_v4_tuple_to_profile().

In calls to qede_flow_parse_v4_common(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-6-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_v6_common()

Convert qede_flow_parse_v6_common() to take extack,
and drop the edev argument.

Convert DP_NOTICE call to use NL_SET_ERR_MSG_MOD instead.

Pass extack in calls to qede_flow_parse_ports() and
qede_set_v6_tuple_to_profile().

In calls to qede_flow_parse_v6_common(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-5-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_set_v4_tuple_to_profile()

Convert qede_set_v4_tuple_to_profile() to take extack,
and drop the edev argument.

Convert DP_INFO call to use NL_SET_ERR_MSG_MOD instead.

In calls to qede_set_v4_tuple_to_profile(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-4-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_set_v6_tuple_to_profile()

Convert qede_set_v6_tuple_to_profile() to take extack,
and drop the edev argument.

Convert DP_INFO call to use NL_SET_ERR_MSG_MOD instead.

In calls to qede_set_v6_tuple_to_profile(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-3-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qede: use extack in qede_flow_parse_ports()

Convert qede_flow_parse_ports to use extack,
and drop the edev argument.

Convert DP_NOTICE call to use NL_SET_ERR_MSG_MOD instead.

In calls to qede_flow_parse_ports(), use NULL as extack
for now, until a subsequent patch makes extack available.

Only compile tested.

Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20240508143404.95901-2-ast@fiberby.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>