linux-block.git
2 months agonet: phy: phy_caps: Implement link_capabilities lookup by linkmode
Maxime Chevallier [Fri, 7 Mar 2025 17:36:03 +0000 (18:36 +0100)]
net: phy: phy_caps: Implement link_capabilities lookup by linkmode

In several occasions, phylib needs to lookup a set of matching speed and
duplex against a given linkmode set. Instead of relying on the
phy_settings array and thus iterate over the whole linkmodes list, use
the link_capabilities array to lookup these matches, as we aren't
interested in the actual link setting that matches but rather the speed
and duplex for that setting.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-7-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: phy_caps: Introduce phy_caps_valid
Maxime Chevallier [Fri, 7 Mar 2025 17:36:02 +0000 (18:36 +0100)]
net: phy: phy_caps: Introduce phy_caps_valid

With the link_capabilities array, it's trivial to validate a given mask
againts a <speed, duplex> tuple. Create a helper for that purpose, and
use it to replace a phy_settings lookup in phy_check_valid();

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-6-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: phy_caps: Move __set_linkmode_max_speed to phy_caps
Maxime Chevallier [Fri, 7 Mar 2025 17:36:01 +0000 (18:36 +0100)]
net: phy: phy_caps: Move __set_linkmode_max_speed to phy_caps

Convert the __set_linkmode_max_speed to use the link_capabilities array.
This makes it easy to clamp the linkmodes to a given max speed.
Introduce a new helper phy_caps_linkmode_max_speed to replace the
previous one that used phy_settings.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-5-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: phy_caps: Move phy_speeds to phy_caps
Maxime Chevallier [Fri, 7 Mar 2025 17:36:00 +0000 (18:36 +0100)]
net: phy: phy_caps: Move phy_speeds to phy_caps

Use the newly introduced link_capabilities array to derive the list of
possible speeds when given a combination of linkmodes. As
link_capabilities is indexed by speed, we don't have to iterate the
whole phy_settings array.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-4-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: Use an internal, searchable storage for the linkmodes
Maxime Chevallier [Fri, 7 Mar 2025 17:35:59 +0000 (18:35 +0100)]
net: phy: Use an internal, searchable storage for the linkmodes

The canonical definition for all the link modes is in linux/ethtool.h,
which is complemented by the link_mode_params array stored in
net/ethtool/common.h . That array contains all the metadata about each
of these modes, including the Speed and Duplex information.

Phylib and phylink needs that information as well for internal
management of the link, which was done by duplicating that information
in locally-stored arrays and lookup functions. This makes it easy for
developpers adding new modes to forget modifying phylib and phylink
accordingly.

However, the link_mode_params array in net/ethtool/common.c is fairly
inefficient to search through, as it isn't sorted in any manner. Phylib
and phylink perform a lot of lookup operations, mostly to filter modes
by speed and/or duplex.

We therefore introduce the link_caps private array in phy_caps.c, that
indexes linkmodes in a more efficient manner. Each element associated a
tuple <speed, duplex> to a bitfield of all the linkmodes runs at these
speed/duplex.

We end-up with an array that's fairly short, easily addressable and that
it optimised for the typical use-cases of phylib/phylink.

That array is initialized at the same time as phylib. As the
link_mode_params array is part of the net stack, which phylink depends
on, it should always be accessible from phylib.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-3-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: ethtool: Export the link_mode_params definitions
Maxime Chevallier [Fri, 7 Mar 2025 17:35:58 +0000 (18:35 +0100)]
net: ethtool: Export the link_mode_params definitions

link_mode_params contains a lookup table of all 802.3 link modes that
are currently supported with structured data about each mode's speed,
duplex, number of lanes and mediums.

As a preparation for a port representation, export that table for the
rest of the net stack to use.

Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20250307173611.129125-2-maxime.chevallier@bootlin.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agoMerge branch 'net-stmmac-avoid-unnecessary-work-in-stmmac_release-stmmac_dvr_remove'
Paolo Abeni [Mon, 17 Mar 2025 20:36:22 +0000 (21:36 +0100)]
Merge branch 'net-stmmac-avoid-unnecessary-work-in-stmmac_release-stmmac_dvr_remove'

Russell King says:

====================
net: stmmac: avoid unnecessary work in stmmac_release()/stmmac_dvr_remove()

This small series is a subset of a RFC I sent earlier. These two
patches remove code that is unnecessary and/or wrong in these paths.
Details in each commit.
====================

Link: https://patch.msgid.link/Z87bpDd7QYYVU0ML@shell.armlinux.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: stmmac: remove unnecessary stmmac_mac_set() in stmmac_release()
Russell King (Oracle) [Mon, 10 Mar 2025 12:31:30 +0000 (12:31 +0000)]
net: stmmac: remove unnecessary stmmac_mac_set() in stmmac_release()

stmmac_release() calls phylink_stop() and then goes on to call
stmmac_mac_set(, false). However, phylink_stop() will call
stmmac_mac_link_down() before returning, which will do this work.
Remove this unnecessary call.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Tested-by: Furong Xu <0x1207@gmail.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/E1trcI6-005rn8-GV@rmk-PC.armlinux.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: stmmac: remove redundant racy tear-down in stmmac_dvr_remove()
Russell King (Oracle) [Mon, 10 Mar 2025 12:31:25 +0000 (12:31 +0000)]
net: stmmac: remove redundant racy tear-down in stmmac_dvr_remove()

While the network device is registered, it is published to userspace,
and thus userspace can change its state. This means calling
functions such as stmmac_stop_all_dma() and stmmac_mac_set() are
racy.

Moreover, unregister_netdev() will unpublish the network device, and
then if appropriate call the .ndo_stop() method, which is
stmmac_release(). This will first call phylink_stop() which will
synchronously take the link down, resulting in stmmac_mac_link_down()
and stmmac_mac_set(, false) being called.

stmmac_release() will also call stmmac_stop_all_dma().

Consequently, neither of these two functions need to called prior
to unregister_netdev() as that will safely call paths that will
result in this work being done if necessary.

Remove these redundant racy calls.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Furong Xu <0x1207@gmail.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/E1trcI1-005rn2-CZ@rmk-PC.armlinux.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phylink: expand on .pcs_config() method documentation
Russell King (Oracle) [Mon, 10 Mar 2025 11:10:52 +0000 (11:10 +0000)]
net: phylink: expand on .pcs_config() method documentation

Expand on the requirements of the .pcs_config() method documentation,
specifically mentioning that it should cause minimal disruption to
an established link, and that it should return a positive non-zero
value when requiring the .pcs_an_restart() method to be called.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/E1trb24-005oVq-Is@rmk-PC.armlinux.org.uk
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agocdc_ether|r8152: ThinkPad Hybrid USB-C/A Dock quirk
Philipp Hahn [Mon, 10 Mar 2025 10:17:35 +0000 (11:17 +0100)]
cdc_ether|r8152: ThinkPad Hybrid USB-C/A Dock quirk

Lenovo ThinkPad Hybrid USB-C with USB-A Dock (17ef:a359) is affected by
the same problem as the Lenovo Powered USB-C Travel Hub (17ef:721e):
Both are based on the Realtek RTL8153B chip used to use the cdc_ether
driver. However, using this driver, with the system suspended the device
constantly sends pause-frames as soon as the receive buffer fills up.
This causes issues with other devices, where some Ethernet switches stop
forwarding packets altogether.

Using the Realtek driver (r8152) fixes this issue. Pause frames are no
longer sent while the host system is suspended.

Cc: Leon Schuermann <leon@is.currently.online>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Oliver Neukum <oliver@neukum.org> (maintainer:USB CDC ETHERNET DRIVER)
Cc: netdev@vger.kernel.org (open list:NETWORKING DRIVERS)
Link: https://git.kernel.org/netdev/net/c/cb82a54904a9
Link: https://git.kernel.org/netdev/net/c/2284bbd0cf39
Link: https://www.lenovo.com/de/de/p/accessories-and-software/docking/docking-usb-docks/40af0135eu
Signed-off-by: Philipp Hahn <phahn-oss@avm.de>
Reviewed-by: Kory Maincent <kory.maincent@bootlin.com>
Link: https://patch.msgid.link/484336aad52d14ccf061b535bc19ef6396ef5120.1741601523.git.p.hahn@avm.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agostmmac: intel: Fix warning message for return value in intel_tsn_lane_is_available()
Choong Yong Liang [Mon, 10 Mar 2025 05:08:35 +0000 (13:08 +0800)]
stmmac: intel: Fix warning message for return value in intel_tsn_lane_is_available()

Fix the warning "warn: missing error code? 'ret'" in the
intel_tsn_lane_is_available() function.

The function now returns 0 to indicate that a TSN lane was found and
returns -EINVAL when it is not found.

Fixes: a42f6b3f1cc1 ("net: stmmac: configure SerDes according to the interface mode")
Signed-off-by: Choong Yong Liang <yong.liang.choong@linux.intel.com>
Reviewed-by: Kory Maincent <kory.maincent@bootlin.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250310050835.808870-1-yong.liang.choong@linux.intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agoMerge branch 'net-phy-clean-up-phy-package-mmd-access-functions'
Paolo Abeni [Mon, 17 Mar 2025 18:07:56 +0000 (19:07 +0100)]
Merge branch 'net-phy-clean-up-phy-package-mmd-access-functions'

Heiner Kallweit says:

====================
net: phy: clean up PHY package MMD access functions

Move declarations of the functions with users to phylib.h, and remove
unused functions.
====================

Link: https://patch.msgid.link/b624fcb7-b493-461a-a0b5-9ca7e9d767bc@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: remove unused functions phy_package_[read|write]_mmd
Heiner Kallweit [Sun, 9 Mar 2025 20:05:08 +0000 (21:05 +0100)]
net: phy: remove unused functions phy_package_[read|write]_mmd

These functions have never had a user, so remove them.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/5792e2cd-6f0a-4f7d-a5ef-b932f94d82f3@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet: phy: move PHY package MMD access function declarations from phy.h to phylib.h
Heiner Kallweit [Sun, 9 Mar 2025 20:04:14 +0000 (21:04 +0100)]
net: phy: move PHY package MMD access function declarations from phy.h to phylib.h

These functions are used by PHY drivers only, therefore move their
declaration to phylib.h.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/406c8a20-b62e-4ee3-b174-b566724a0876@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agoMerge branch 'mlx5-support-hws-flow-meter-sampler-actions-in-fs-core'
Paolo Abeni [Mon, 17 Mar 2025 17:57:19 +0000 (18:57 +0100)]
Merge branch 'mlx5-support-hws-flow-meter-sampler-actions-in-fs-core'

Tariq Toukan says:

====================
mlx5: Support HWS flow meter/sampler actions in FS core

This series by Moshe adds support for flow meter and flow sampler HW
Steering actions in FS core level. As these actions can be shared by
multiple rules, these patches use refcounts to manage the HWS actions
sharing in FS core level.
====================

Link: https://patch.msgid.link/1741543663-22123-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet/mlx5: fs, add support for dest flow sampler HWS action
Moshe Shemesh [Sun, 9 Mar 2025 18:07:43 +0000 (20:07 +0200)]
net/mlx5: fs, add support for dest flow sampler HWS action

Add support for HW Steering action of flow sampler destination. For each
flow sampler created cache the hws action by sampler id as a key. Hold
refcount for each rule using the cached action.

Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/1741543663-22123-4-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet/mlx5: fs, add support for flow meters HWS action
Moshe Shemesh [Sun, 9 Mar 2025 18:07:42 +0000 (20:07 +0200)]
net/mlx5: fs, add support for flow meters HWS action

Add support for HW Steering action of flow meter range. Flow meters
range can use one HWS action for the whole range. Thus, share a cached
HWS action among rules that use same flow meter object range. Hold
refcount for each rule using the cached action.

Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/1741543663-22123-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agonet/mlx5: fs, add API for sharing HWS action by refcount
Moshe Shemesh [Sun, 9 Mar 2025 18:07:41 +0000 (20:07 +0200)]
net/mlx5: fs, add API for sharing HWS action by refcount

Counters HWS actions are shared using refcount, to create action on
demand by flow steering rule and destroy only when no rules are using
the action. The method is extensible to other HWS action types, such as
flow meter and sampler actions, in the downstream patches.

Add an API to facilitate the reuse of get/put logic for HWS actions
shared by refcount.

Signed-off-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/1741543663-22123-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2 months agoMerge branch 'tcp-accecn'
David S. Miller [Mon, 17 Mar 2025 13:57:14 +0000 (13:57 +0000)]
Merge branch 'tcp-accecn'

Chia-Yu Chang says:

====================
AccECN protocol preparation patch series

Please find the v7

v7 (03-Mar-2025)
- Move 2 new patches added in v6 to the next AccECN patch series

v6 (27-Dec-2024)
- Avoid removing removing the potential CA_ACK_WIN_UPDATE in ack_ev_flags of patch #1 (Eric Dumazet <edumazet@google.com>)
- Add reviewed-by tag in patches #2, #3, #4, #5, #6, #7, #8, #12, #14
- Foloiwng 2 new pathces are added after patch #9 (Patch that adds SKB_GSO_TCP_ACCECN)
  * New patch #10 to replace exisiting SKB_GSO_TCP_ECN with SKB_GSO_TCP_ACCECN in the driver to avoid CWR flag corruption
  * New patch #11 adds AccECN for virtio by adding new negotiation flag (VIRTIO_NET_F_HOST/GUEST_ACCECN) in feature handshake and translating Accurate ECN GSO flag between virtio_net_hdr (VIRTIO_NET_HDR_GSO_ACCECN) and skb header (SKB_GSO_TCP_ACCECN)
- Add detailed changelog and comments in #13 (Eric Dumazet <edumazet@google.com>)
- Move patch #14 to the next AccECN patch series (Eric Dumazet <edumazet@google.com>)

v5 (5-Nov-2024)
- Add helper function "tcp_flags_ntohs" to preserve last 2 bytes of TCP flags of patch #4 (Paolo Abeni <pabeni@redhat.com>)
- Fix reverse X-max tree order of patches #4, #11 (Paolo Abeni <pabeni@redhat.com>)
- Rename variable "delta" as "timestamp_delta" of patch #2 fo clariety
- Remove patch #14 in this series (Paolo Abeni <pabeni@redhat.com>, Joel Granados <joel.granados@kernel.org>)

v4 (21-Oct-2024)
- Fix line length warning of patches #2, #4, #8, #10, #11, #14
- Fix spaces preferred around '|' (ctx:VxV) warning of patch #7
- Add missing CC'ed of patches #4, #12, #14

v3 (19-Oct-2024)
- Fix build error in v2

v2 (18-Oct-2024)
- Fix warning caused by NETIF_F_GSO_ACCECN_BIT in patch #9 (Jakub Kicinski <kuba@kernel.org>)

The full patch series can be found in
https://github.com/L4STeam/linux-net-next/commits/upstream_l4steam/

The Accurate ECN draft can be found in
https://datatracker.ietf.org/doc/html/draft-ietf-tcpm-accurate-ecn-28
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: Pass flags to __tcp_send_ack
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:52 +0000 (23:38 +0100)]
tcp: Pass flags to __tcp_send_ack

Accurate ECN needs to send custom flags to handle IP-ECN
field reflection during handshake.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: add new TCP_TW_ACK_OOW state and allow ECN bits in TOS
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:51 +0000 (23:38 +0100)]
tcp: add new TCP_TW_ACK_OOW state and allow ECN bits in TOS

ECN bits in TOS are always cleared when sending in ACKs in TW. Clearing
them is problematic for TCP flows that used Accurate ECN because ECN bits
decide which service queue the packet is placed into (L4S vs Classic).
Effectively, TW ACKs are always downgraded from L4S to Classic queue
which might impact, e.g., delay the ACK will experience on the path
compared with the other packets of the flow.

Change the TW ACK sending code to differentiate:
- In tcp_v4_send_reset(), commit ba9e04a7ddf4f ("ip: fix tos reflection
  in ack and reset packets") cleans ECN bits for TW reset and this is
  not affected.
- In tcp_v4_timewait_ack(), ECN bits for all TW ACKs are cleaned. But now
  only ECN bits of ACKs for oow data or paws_reject are cleaned, and ECN
  bits of other ACKs will not be cleaned.
- In tcp_v4_reqsk_send_ack(), commit 66b13d99d96a1 ("ipv4: tcp: fix TOS
  value in ACK messages sent from TIME_WAIT") did not clean ECN bits of
  ACKs for oow data or paws_reject. But now the ECN bits rae cleaned for
  these ACKs.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: AccECN support to tcp_add_backlog
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:50 +0000 (23:38 +0100)]
tcp: AccECN support to tcp_add_backlog

AE flag needs to be preserved for AccECN.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agogro: prevent ACE field corruption & better AccECN handling
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:49 +0000 (23:38 +0100)]
gro: prevent ACE field corruption & better AccECN handling

There are important differences in how the CWR field behaves
in RFC3168 and AccECN. With AccECN, CWR flag is part of the
ACE counter and its changes are important so adjust the flags
changed mask accordingly.

Also, if CWR is there, set the Accurate ECN GSO flag to avoid
corrupting CWR flag somewhere.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agogso: AccECN support
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:48 +0000 (23:38 +0100)]
gso: AccECN support

Handling the CWR flag differs between RFC 3168 ECN and AccECN.
With RFC 3168 ECN aware TSO (NETIF_F_TSO_ECN) CWR flag is cleared
starting from 2nd segment which is incompatible how AccECN handles
the CWR flag. Such super-segments are indicated by SKB_GSO_TCP_ECN.
With AccECN, CWR flag (or more accurately, the ACE field that also
includes ECE & AE flags) changes only when new packet(s) with CE
mark arrives so the flag should not be changed within a super-skb.
The new skb/feature flags are necessary to prevent such TSO engines
corrupting AccECN ACE counters by clearing the CWR flag (if the
CWR handling feature cannot be turned off).

If NIC is completely unaware of RFC3168 ECN (doesn't support
NETIF_F_TSO_ECN) or its TSO engine can be set to not touch CWR flag
despite supporting also NETIF_F_TSO_ECN, TSO could be safely used
with AccECN on such NIC. This should be evaluated per NIC basis
(not done in this patch series for any NICs).

For the cases, where TSO cannot keep its hands off the CWR flag,
a GSO fallback is provided by this patch.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: helpers for ECN mode handling
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:47 +0000 (23:38 +0100)]
tcp: helpers for ECN mode handling

Create helpers for TCP ECN modes. No functional changes.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: rework {__,}tcp_ecn_check_ce() -> tcp_data_ecn_check()
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:46 +0000 (23:38 +0100)]
tcp: rework {__,}tcp_ecn_check_ce() -> tcp_data_ecn_check()

Rename tcp_ecn_check_ce to tcp_data_ecn_check as it is
called only for data segments, not for ACKs (with AccECN,
also ACKs may get ECN bits).

The extra "layer" in tcp_ecn_check_ce() function just
checks for ECN being enabled, that can be moved into
tcp_ecn_field_check rather than having the __ variant.

No functional changes.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: extend TCP flags to allow AE bit/ACE field
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:44 +0000 (23:38 +0100)]
tcp: extend TCP flags to allow AE bit/ACE field

With AccECN, there's one additional TCP flag to be used (AE)
and ACE field that overloads the definition of AE, CWR, and
ECE flags. As tcp_flags was previously only 1 byte, the
byte-order stuff needs to be added to it's handling.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: use BIT() macro in include/net/tcp.h
Chia-Yu Chang [Wed, 5 Mar 2025 22:38:43 +0000 (23:38 +0100)]
tcp: use BIT() macro in include/net/tcp.h

Use BIT() macro for TCP flags field and TCP congestion control
flags that will be used by the congestion control algorithm.

No functional changes.

Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Ilpo Järvinen <ij@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: create FLAG_TS_PROGRESS
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:42 +0000 (23:38 +0100)]
tcp: create FLAG_TS_PROGRESS

Whenever timestamp advances, it declares progress which
can be used by the other parts of the stack to decide that
the ACK is the most recent one seen so far.

AccECN will use this flag when deciding whether to use the
ACK to update AccECN state or not.

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2 months agotcp: reorganize tcp_in_ack_event() and tcp_count_delivered()
Ilpo Järvinen [Wed, 5 Mar 2025 22:38:41 +0000 (23:38 +0100)]
tcp: reorganize tcp_in_ack_event() and tcp_count_delivered()

- Move tcp_count_delivered() earlier and split tcp_count_delivered_ce()
  out of it
- Move tcp_in_ack_event() later
- While at it, remove the inline from tcp_in_ack_event() and let
  the compiler to decide

Accurate ECN's heuristics does not know if there is going
to be ACE field based CE counter increase or not until after
rtx queue has been processed. Only then the number of ACKed
bytes/pkts is available. As CE or not affects presence of
FLAG_ECE, that information for tcp_in_ack_event is not yet
available in the old location of the call to tcp_in_ack_event().

Signed-off-by: Ilpo Järvinen <ij@kernel.org>
Signed-off-by: Chia-Yu Chang <chia-yu.chang@nokia-bell-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 months agonet/smc: use the correct ndev to find pnetid by pnetid table
Guangguan Wang [Tue, 4 Mar 2025 12:43:04 +0000 (20:43 +0800)]
net/smc: use the correct ndev to find pnetid by pnetid table

When using smc_pnet in SMC, it will only search the pnetid in the
base_ndev of the netdev hierarchy(both HW PNETID and User-defined
sw pnetid). This may not work for some scenarios when using SMC in
container on cloud environment.
In container, there have choices of different container network,
such as directly using host network, virtual network IPVLAN, veth,
etc. Different choices of container network have different netdev
hierarchy. Examples of netdev hierarchy show below. (eth0 and eth1
in host below is the netdev directly related to the physical device).
            _______________________________
           |   _________________           |
           |  |POD              |          |
           |  |                 |          |
           |  | eth0_________   |          |
           |  |____|         |__|          |
           |       |         |             |
           |       |         |             |
           |   eth1|base_ndev| eth0_______ |
           |       |         |    | RDMA  ||
           | host  |_________|    |_______||
           ---------------------------------
     netdev hierarchy if directly using host network
           ________________________________
           |   _________________           |
           |  |POD  __________  |          |
           |  |    |upper_ndev| |          |
           |  |eth0|__________| |          |
           |  |_______|_________|          |
           |          |lower netdev        |
           |        __|______              |
           |   eth1|         | eth0_______ |
           |       |base_ndev|    | RDMA  ||
           | host  |_________|    |_______||
           ---------------------------------
            netdev hierarchy if using IPVLAN
            _______________________________
           |   _____________________       |
           |  |POD        _________ |      |
           |  |          |base_ndev||      |
           |  |eth0(veth)|_________||      |
           |  |____________|________|      |
           |               |pairs          |
           |        _______|_              |
           |       |         | eth0_______ |
           |   veth|base_ndev|    | RDMA  ||
           |       |_________|    |_______||
           |        _________              |
           |   eth1|base_ndev|             |
           | host  |_________|             |
           ---------------------------------
             netdev hierarchy if using veth
Due to some reasons, the eth1 in host is not RDMA attached netdevice,
pnetid is needed to map the eth1(in host) with RDMA device so that POD
can do SMC-R. Because the eth1(in host) is managed by CNI plugin(such
as Terway, network management plugin in container environment), and in
cloud environment the eth(in host) can dynamically be inserted by CNI
when POD create and dynamically be removed by CNI when POD destroy and
no POD related to the eth(in host) anymore. It is hard to config the
pnetid to the eth1(in host). But it is easy to config the pnetid to the
netdevice which can be seen in POD. When do SMC-R, both the container
directly using host network and the container using veth network can
successfully match the RDMA device, because the configured pnetid netdev
is a base_ndev. But the container using IPVLAN can not successfully
match the RDMA device and 0x03030000 fallback happens, because the
configured pnetid netdev is not a base_ndev. Additionally, if config
pnetid to the eth1(in host) also can not work for matching RDMA device
when using veth network and doing SMC-R in POD.

To resolve the problems list above, this patch extends to search user
-defined sw pnetid in the clc handshake ndev when no pnetid can be found
in the base_ndev, and the base_ndev take precedence over ndev for backward
compatibility. This patch also can unify the pnetid setup of different
network choices list above in container(Config user-defined sw pnetid in
the netdevice can be seen in POD).

Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 months agoselftests: drv-net: fix merge conflicts resolution
Matthieu Baerts (NGI0) [Fri, 14 Mar 2025 08:35:51 +0000 (09:35 +0100)]
selftests: drv-net: fix merge conflicts resolution

After the recent merge between net-next and net, I got some conflicts on
my side because the merge resolution was different from Stephen's one
[1] I applied on my side in the MPTCP tree.

It looks like the code that is now in net-next is using the old way to
retrieve the local and remote addresses. This patch is now using the new
way, like what was in Stephen's email [1].

Also, in get_interface_info(), there were no conflicts in this area,
because that was new code from 'net', but a small adaptation was needed
there as well to get the remote address.

Fixes: 941defcea7e1 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net")
Link: https://lore.kernel.org/20250311115758.17a1d414@canb.auug.org.au
Suggested-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20250314-net-next-drv-net-ping-fix-merge-v1-1-0d5c19daf707@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Paolo Abeni [Thu, 13 Mar 2025 22:01:29 +0000 (23:01 +0100)]
Merge git://git./linux/kernel/git/netdev/net

Cross-merge networking fixes after downstream PR (net-6.14-rc6).

Conflicts:

tools/testing/selftests/drivers/net/ping.py
  75cc19c8ff89 ("selftests: drv-net: add xdp cases for ping.py")
  de94e8697405 ("selftests: drv-net: store addresses in dict indexed by ipver")
https://lore.kernel.org/netdev/20250311115758.17a1d414@canb.auug.org.au/

net/core/devmem.c
  a70f891e0fa0 ("net: devmem: do not WARN conditionally after netdev_rx_queue_restart()")
  1d22d3060b9b ("net: drop rtnl_lock for queue_mgmt operations")
https://lore.kernel.org/netdev/20250313114929.43744df1@canb.auug.org.au/

Adjacent changes:

tools/testing/selftests/net/Makefile
  6f50175ccad4 ("selftests: Add IPv6 link-local address generation tests for GRE devices.")
  2e5584e0f913 ("selftests/net: expand cmsg_ipv6.sh with ipv4")

drivers/net/ethernet/broadcom/bnxt/bnxt.c
  661958552eda ("eth: bnxt: do not use BNXT_VNIC_NTUPLE unconditionally in queue restart logic")
  fe96d717d38e ("bnxt_en: Extend queue stop/start for TX rings")

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge tag 'net-6.14-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 13 Mar 2025 17:58:48 +0000 (07:58 -1000)]
Merge tag 'net-6.14-rc7' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from netfilter, bluetooth and wireless.

  No known regressions outstanding.

  Current release - regressions:

   - wifi: nl80211: fix assoc link handling

   - eth: lan78xx: sanitize return values of register read/write
     functions

  Current release - new code bugs:

   - ethtool: tsinfo: fix dump command

   - bluetooth: btusb: configure altsetting for HCI_USER_CHANNEL

   - eth: mlx5: DR, use the right action structs for STEv3

  Previous releases - regressions:

   - netfilter: nf_tables: make destruction work queue pernet

   - gre: fix IPv6 link-local address generation.

   - wifi: iwlwifi: fix TSO preparation

   - bluetooth: revert "bluetooth: hci_core: fix sleeping function
     called from invalid context"

   - ovs: revert "openvswitch: switch to per-action label counting in
     conntrack"

   - eth:
       - ice: fix switchdev slow-path in LAG
       - bonding: fix incorrect MAC address setting to receive NS
         messages

  Previous releases - always broken:

   - core: prevent TX of unreadable skbs

   - sched: prevent creation of classes with TC_H_ROOT

   - netfilter: nft_exthdr: fix offset with ipv4_find_option()

   - wifi: cfg80211: cancel wiphy_work before freeing wiphy

   - mctp: copy headers if cloned

   - phy: nxp-c45-tja11xx: add errata for TJA112XA/B

   - eth:
       - bnxt: fix kernel panic in the bnxt_get_queue_stats{rx | tx}
       - mlx5: bridge, fix the crash caused by LAG state check"

* tag 'net-6.14-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (65 commits)
  net: mana: cleanup mana struct after debugfs_remove()
  net/mlx5e: Prevent bridge link show failure for non-eswitch-allowed devices
  net/mlx5: Bridge, fix the crash caused by LAG state check
  net/mlx5: Lag, Check shared fdb before creating MultiPort E-Switch
  net/mlx5: Fix incorrect IRQ pool usage when releasing IRQs
  net/mlx5: HWS, Rightsize bwc matcher priority
  net/mlx5: DR, use the right action structs for STEv3
  Revert "openvswitch: switch to per-action label counting in conntrack"
  net: openvswitch: remove misbehaving actions length check
  selftests: Add IPv6 link-local address generation tests for GRE devices.
  gre: Fix IPv6 link-local address generation.
  netfilter: nft_exthdr: fix offset with ipv4_find_option()
  selftests/tc-testing: Add a test case for DRR class with TC_H_ROOT
  net_sched: Prevent creation of classes with TC_H_ROOT
  ipvs: prevent integer overflow in do_ip_vs_get_ctl()
  selftests: netfilter: skip br_netfilter queue tests if kernel is tainted
  netfilter: nf_conncount: Fully initialize struct nf_conncount_tuple in insert_tree()
  wifi: mac80211: fix MPDU length parsing for EHT 5/6 GHz
  qlcnic: fix memory leak issues in qlcnic_sriov_common.c
  rtase: Fix improper release of ring list entries in rtase_sw_reset
  ...

3 months agoMerge tag 'vfs-6.14-rc7.fixes' of gitolite.kernel.org:pub/scm/linux/kernel/git/vfs/vfs
Linus Torvalds [Thu, 13 Mar 2025 17:53:25 +0000 (07:53 -1000)]
Merge tag 'vfs-6.14-rc7.fixes' of gitolite.pub/scm/linux/kernel/git/vfs/vfs

Pull vfs fixes from Christian Brauner:

 - Bring in an RCU pathwalk fix for afs. This is brought in as a merge
   from the vfs-6.15.shared.afs branch that needs this commit and other
   trees already depend on it.

 - Fix vboxfs unterminated string handling.

* tag 'vfs-6.14-rc7.fixes' of gitolite.kernel.org:pub/scm/linux/kernel/git/vfs/vfs:
  vboxsf: Add __nonstring annotations for unterminated strings
  afs: Fix afs_atcell_get_link() to handle RCU pathwalk

3 months agoMerge tag 'nf-25-03-13' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf
Paolo Abeni [Thu, 13 Mar 2025 14:04:26 +0000 (15:04 +0100)]
Merge tag 'nf-25-03-13' of git://git./linux/kernel/git/netfilter/nf

Pablo Neira Ayuso says:

====================
Netfilter/IPVS fixes for net

The following patchset contains Netfilter/IPVS fixes for net:

1) Missing initialization of cpu and jiffies32 fields in conncount,
   from Kohei Enju.

2) Skip several tests in case kernel is tainted, otherwise tests bogusly
   report failure too as they also check for tainted kernel,
   from Florian Westphal.

3) Fix a hyphothetical integer overflow in do_ip_vs_get_ctl() leading
   to bogus error logs, from Dan Carpenter.

4) Fix incorrect offset in ipv4 option match in nft_exthdr, from
   Alexey Kashavkin.

netfilter pull request 25-03-13

* tag 'nf-25-03-13' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
  netfilter: nft_exthdr: fix offset with ipv4_find_option()
  ipvs: prevent integer overflow in do_ip_vs_get_ctl()
  selftests: netfilter: skip br_netfilter queue tests if kernel is tainted
  netfilter: nf_conncount: Fully initialize struct nf_conncount_tuple in insert_tree()
====================

Link: https://patch.msgid.link/20250313095636.2186-1-pablo@netfilter.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: mana: cleanup mana struct after debugfs_remove()
Shradha Gupta [Tue, 11 Mar 2025 10:17:40 +0000 (03:17 -0700)]
net: mana: cleanup mana struct after debugfs_remove()

When on a MANA VM hibernation is triggered, as part of hibernate_snapshot(),
mana_gd_suspend() and mana_gd_resume() are called. If during this
mana_gd_resume(), a failure occurs with HWC creation, mana_port_debugfs
pointer does not get reinitialized and ends up pointing to older,
cleaned-up dentry.
Further in the hibernation path, as part of power_down(), mana_gd_shutdown()
is triggered. This call, unaware of the failures in resume, tries to cleanup
the already cleaned up  mana_port_debugfs value and hits the following bug:

[  191.359296] mana 7870:00:00.0: Shutdown was called
[  191.359918] BUG: kernel NULL pointer dereference, address: 0000000000000098
[  191.360584] #PF: supervisor write access in kernel mode
[  191.361125] #PF: error_code(0x0002) - not-present page
[  191.361727] PGD 1080ea067 P4D 0
[  191.362172] Oops: Oops: 0002 [#1] SMP NOPTI
[  191.362606] CPU: 11 UID: 0 PID: 1674 Comm: bash Not tainted 6.14.0-rc5+ #2
[  191.363292] Hardware name: Microsoft Corporation Virtual Machine/Virtual Machine, BIOS Hyper-V UEFI Release v4.1 11/21/2024
[  191.364124] RIP: 0010:down_write+0x19/0x50
[  191.364537] Code: 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 0f 1f 44 00 00 55 48 89 e5 53 48 89 fb e8 de cd ff ff 31 c0 ba 01 00 00 00 <f0> 48 0f b1 13 75 16 65 48 8b 05 88 24 4c 6a 48 89 43 08 48 8b 5d
[  191.365867] RSP: 0000:ff45fbe0c1c037b8 EFLAGS: 00010246
[  191.366350] RAX: 0000000000000000 RBX: 0000000000000098 RCX: ffffff8100000000
[  191.366951] RDX: 0000000000000001 RSI: 0000000000000064 RDI: 0000000000000098
[  191.367600] RBP: ff45fbe0c1c037c0 R08: 0000000000000000 R09: 0000000000000001
[  191.368225] R10: ff45fbe0d2b01000 R11: 0000000000000008 R12: 0000000000000000
[  191.368874] R13: 000000000000000b R14: ff43dc27509d67c0 R15: 0000000000000020
[  191.369549] FS:  00007dbc5001e740(0000) GS:ff43dc663f380000(0000) knlGS:0000000000000000
[  191.370213] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  191.370830] CR2: 0000000000000098 CR3: 0000000168e8e002 CR4: 0000000000b73ef0
[  191.371557] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  191.372192] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
[  191.372906] Call Trace:
[  191.373262]  <TASK>
[  191.373621]  ? show_regs+0x64/0x70
[  191.374040]  ? __die+0x24/0x70
[  191.374468]  ? page_fault_oops+0x290/0x5b0
[  191.374875]  ? do_user_addr_fault+0x448/0x800
[  191.375357]  ? exc_page_fault+0x7a/0x160
[  191.375971]  ? asm_exc_page_fault+0x27/0x30
[  191.376416]  ? down_write+0x19/0x50
[  191.376832]  ? down_write+0x12/0x50
[  191.377232]  simple_recursive_removal+0x4a/0x2a0
[  191.377679]  ? __pfx_remove_one+0x10/0x10
[  191.378088]  debugfs_remove+0x44/0x70
[  191.378530]  mana_detach+0x17c/0x4f0
[  191.378950]  ? __flush_work+0x1e2/0x3b0
[  191.379362]  ? __cond_resched+0x1a/0x50
[  191.379787]  mana_remove+0xf2/0x1a0
[  191.380193]  mana_gd_shutdown+0x3b/0x70
[  191.380642]  pci_device_shutdown+0x3a/0x80
[  191.381063]  device_shutdown+0x13e/0x230
[  191.381480]  kernel_power_off+0x35/0x80
[  191.381890]  hibernate+0x3c6/0x470
[  191.382312]  state_store+0xcb/0xd0
[  191.382734]  kobj_attr_store+0x12/0x30
[  191.383211]  sysfs_kf_write+0x3e/0x50
[  191.383640]  kernfs_fop_write_iter+0x140/0x1d0
[  191.384106]  vfs_write+0x271/0x440
[  191.384521]  ksys_write+0x72/0xf0
[  191.384924]  __x64_sys_write+0x19/0x20
[  191.385313]  x64_sys_call+0x2b0/0x20b0
[  191.385736]  do_syscall_64+0x79/0x150
[  191.386146]  ? __mod_memcg_lruvec_state+0xe7/0x240
[  191.386676]  ? __lruvec_stat_mod_folio+0x79/0xb0
[  191.387124]  ? __pfx_lru_add+0x10/0x10
[  191.387515]  ? queued_spin_unlock+0x9/0x10
[  191.387937]  ? do_anonymous_page+0x33c/0xa00
[  191.388374]  ? __handle_mm_fault+0xcf3/0x1210
[  191.388805]  ? __count_memcg_events+0xbe/0x180
[  191.389235]  ? handle_mm_fault+0xae/0x300
[  191.389588]  ? do_user_addr_fault+0x559/0x800
[  191.390027]  ? irqentry_exit_to_user_mode+0x43/0x230
[  191.390525]  ? irqentry_exit+0x1d/0x30
[  191.390879]  ? exc_page_fault+0x86/0x160
[  191.391235]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[  191.391745] RIP: 0033:0x7dbc4ff1c574
[  191.392111] Code: c7 00 16 00 00 00 b8 ff ff ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 f3 0f 1e fa 80 3d d5 ea 0e 00 00 74 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 55 48 89 e5 48 83 ec 20 48 89
[  191.393412] RSP: 002b:00007ffd95a23ab8 EFLAGS: 00000202 ORIG_RAX: 0000000000000001
[  191.393990] RAX: ffffffffffffffda RBX: 0000000000000005 RCX: 00007dbc4ff1c574
[  191.394594] RDX: 0000000000000005 RSI: 00005a6eeadb0ce0 RDI: 0000000000000001
[  191.395215] RBP: 00007ffd95a23ae0 R08: 00007dbc50003b20 R09: 0000000000000000
[  191.395805] R10: 0000000000000001 R11: 0000000000000202 R12: 0000000000000005
[  191.396404] R13: 00005a6eeadb0ce0 R14: 00007dbc500045c0 R15: 00007dbc50001ee0
[  191.396987]  </TASK>

To fix this, we explicitly set such mana debugfs variables to NULL after
debugfs_remove() is called.

Fixes: 6607c17c6c5e ("net: mana: Enable debugfs files for MANA device")
Cc: stable@vger.kernel.org
Signed-off-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Michal Kubiak <michal.kubiak@intel.com>
Link: https://patch.msgid.link/1741688260-28922-1-git-send-email-shradhagupta@linux.microsoft.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'mlx5-misc-fixes-2025-03-10'
Paolo Abeni [Thu, 13 Mar 2025 12:11:18 +0000 (13:11 +0100)]
Merge branch 'mlx5-misc-fixes-2025-03-10'

Tariq Toukan says:

====================
mlx5 misc fixes 2025-03-10

This patchset provides misc bug fixes from the team to the mlx5 core and
Eth drivers.
====================

Link: https://patch.msgid.link/1741644104-97767-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5e: Prevent bridge link show failure for non-eswitch-allowed devices
Carolina Jubran [Mon, 10 Mar 2025 22:01:44 +0000 (00:01 +0200)]
net/mlx5e: Prevent bridge link show failure for non-eswitch-allowed devices

mlx5_eswitch_get_vepa returns -EPERM if the device lacks
eswitch_manager capability, blocking mlx5e_bridge_getlink from
retrieving VEPA mode. Since mlx5e_bridge_getlink implements
ndo_bridge_getlink, returning -EPERM causes bridge link show to fail
instead of skipping devices without this capability.

To avoid this, return -EOPNOTSUPP from mlx5e_bridge_getlink when
mlx5_eswitch_get_vepa fails, ensuring the command continues processing
other devices while ignoring those without the necessary capability.

Fixes: 4b89251de024 ("net/mlx5: Support ndo bridge_setlink and getlink")
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Reviewed-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://patch.msgid.link/1741644104-97767-7-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5: Bridge, fix the crash caused by LAG state check
Jianbo Liu [Mon, 10 Mar 2025 22:01:43 +0000 (00:01 +0200)]
net/mlx5: Bridge, fix the crash caused by LAG state check

When removing LAG device from bridge, NETDEV_CHANGEUPPER event is
triggered. Driver finds the lower devices (PFs) to flush all the
offloaded entries. And mlx5_lag_is_shared_fdb is checked, it returns
false if one of PF is unloaded. In such case,
mlx5_esw_bridge_lag_rep_get() and its caller return NULL, instead of
the alive PF, and the flush is skipped.

Besides, the bridge fdb entry's lastuse is updated in mlx5 bridge
event handler. But this SWITCHDEV_FDB_ADD_TO_BRIDGE event can be
ignored in this case because the upper interface for bond is deleted,
and the entry will never be aged because lastuse is never updated.

To make things worse, as the entry is alive, mlx5 bridge workqueue
keeps sending that event, which is then handled by kernel bridge
notifier. It causes the following crash when accessing the passed bond
netdev which is already destroyed.

To fix this issue, remove such checks. LAG state is already checked in
commit 15f8f168952f ("net/mlx5: Bridge, verify LAG state when adding
bond to bridge"), driver still need to skip offload if LAG becomes
invalid state after initialization.

 Oops: stack segment: 0000 [#1] SMP
 CPU: 3 UID: 0 PID: 23695 Comm: kworker/u40:3 Tainted: G           OE      6.11.0_mlnx #1
 Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
 Workqueue: mlx5_bridge_wq mlx5_esw_bridge_update_work [mlx5_core]
 RIP: 0010:br_switchdev_event+0x2c/0x110 [bridge]
 Code: 44 00 00 48 8b 02 48 f7 00 00 02 00 00 74 69 41 54 55 53 48 83 ec 08 48 8b a8 08 01 00 00 48 85 ed 74 4a 48 83 fe 02 48 89 d3 <4c> 8b 65 00 74 23 76 49 48 83 fe 05 74 7e 48 83 fe 06 75 2f 0f b7
 RSP: 0018:ffffc900092cfda0 EFLAGS: 00010297
 RAX: ffff888123bfe000 RBX: ffffc900092cfe08 RCX: 00000000ffffffff
 RDX: ffffc900092cfe08 RSI: 0000000000000001 RDI: ffffffffa0c585f0
 RBP: 6669746f6e690a30 R08: 0000000000000000 R09: ffff888123ae92c8
 R10: 0000000000000000 R11: fefefefefefefeff R12: ffff888123ae9c60
 R13: 0000000000000001 R14: ffffc900092cfe08 R15: 0000000000000000
 FS:  0000000000000000(0000) GS:ffff88852c980000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f15914c8734 CR3: 0000000002830005 CR4: 0000000000770ef0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 PKRU: 55555554
 Call Trace:
  <TASK>
  ? __die_body+0x1a/0x60
  ? die+0x38/0x60
  ? do_trap+0x10b/0x120
  ? do_error_trap+0x64/0xa0
  ? exc_stack_segment+0x33/0x50
  ? asm_exc_stack_segment+0x22/0x30
  ? br_switchdev_event+0x2c/0x110 [bridge]
  ? sched_balance_newidle.isra.149+0x248/0x390
  notifier_call_chain+0x4b/0xa0
  atomic_notifier_call_chain+0x16/0x20
  mlx5_esw_bridge_update+0xec/0x170 [mlx5_core]
  mlx5_esw_bridge_update_work+0x19/0x40 [mlx5_core]
  process_scheduled_works+0x81/0x390
  worker_thread+0x106/0x250
  ? bh_worker+0x110/0x110
  kthread+0xb7/0xe0
  ? kthread_park+0x80/0x80
  ret_from_fork+0x2d/0x50
  ? kthread_park+0x80/0x80
  ret_from_fork_asm+0x11/0x20
  </TASK>

Fixes: ff9b7521468b ("net/mlx5: Bridge, support LAG")
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://patch.msgid.link/1741644104-97767-6-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5: Lag, Check shared fdb before creating MultiPort E-Switch
Shay Drory [Mon, 10 Mar 2025 22:01:42 +0000 (00:01 +0200)]
net/mlx5: Lag, Check shared fdb before creating MultiPort E-Switch

Currently, MultiPort E-Switch is requesting to create a LAG with shared
FDB without checking the LAG is supporting shared FDB.
Add the check.

Fixes: a32327a3a02c ("net/mlx5: Lag, Control MultiPort E-Switch single FDB mode")
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: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://patch.msgid.link/1741644104-97767-5-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5: Fix incorrect IRQ pool usage when releasing IRQs
Shay Drory [Mon, 10 Mar 2025 22:01:41 +0000 (00:01 +0200)]
net/mlx5: Fix incorrect IRQ pool usage when releasing IRQs

mlx5_irq_pool_get() is a getter for completion IRQ pool only.
However, after the cited commit, mlx5_irq_pool_get() is called during
ctrl IRQ release flow to retrieve the pool, resulting in the use of an
incorrect IRQ pool.

Hence, use the newly introduced mlx5_irq_get_pool() getter to retrieve
the correct IRQ pool based on the IRQ itself. While at it, rename
mlx5_irq_pool_get() to mlx5_irq_table_get_comp_irq_pool() which
accurately reflects its purpose and improves code readability.

Fixes: 0477d5168bbb ("net/mlx5: Expose SFs IRQs")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Reviewed-by: Maher Sanalla <msanalla@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Link: https://patch.msgid.link/1741644104-97767-4-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5: HWS, Rightsize bwc matcher priority
Vlad Dogaru [Mon, 10 Mar 2025 22:01:40 +0000 (00:01 +0200)]
net/mlx5: HWS, Rightsize bwc matcher priority

The bwc layer was clamping the matcher priority from 32 bits to 16 bits.
This didn't show up until a matcher was resized, since the initial
native matcher was created using the correct 32 bit value.

The fix also reorders fields to avoid some padding.

Fixes: 2111bb970c78 ("net/mlx5: HWS, added backward-compatible API handling")
Signed-off-by: Vlad Dogaru <vdogaru@nvidia.com>
Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1741644104-97767-3-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet/mlx5: DR, use the right action structs for STEv3
Yevgeny Kliteynik [Mon, 10 Mar 2025 22:01:39 +0000 (00:01 +0200)]
net/mlx5: DR, use the right action structs for STEv3

Some actions in ConnectX-8 (STEv3) have different structure,
and they are handled separately in ste_ctx_v3.
This separate handling was missing two actions: INSERT_HDR
and REMOVE_HDR, which broke SWS for Linux Bridge.
This patch resolves the issue by introducing dedicated
callbacks for the insert and remove header functions,
with version-specific implementations for each STE variant.

Fixes: 4d617b57574f ("net/mlx5: DR, add support for ConnectX-8 steering")
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Itamar Gozlan <igozlan@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1741644104-97767-2-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Paolo Abeni [Thu, 13 Mar 2025 11:40:45 +0000 (12:40 +0100)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

Tariq Toukan says:

====================
mlx5-next updates 2025-03-10

The following pull-request contains common mlx5 updates for your *net-next* tree.
Please pull and let me know of any problem.

* 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
  net/mlx5: Add IFC bits for PPCNT recovery counters group
  net/mlx5: fs, add RDMA TRANSPORT steering domain support
  net/mlx5: Query ADV_RDMA capabilities
  net/mlx5: Limit non-privileged commands
  net/mlx5: Allow the throttle mechanism to be more dynamic
  net/mlx5: Add RDMA_CTRL HW capabilities
====================

Link: https://patch.msgid.link/1741608293-41436-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agodt-bindings: net: Define interrupt constraints for DWMAC vendor bindings
Lad Prabhakar [Sun, 9 Mar 2025 00:33:01 +0000 (00:33 +0000)]
dt-bindings: net: Define interrupt constraints for DWMAC vendor bindings

The `snps,dwmac.yaml` binding currently sets `maxItems: 3` for the
`interrupts` and `interrupt-names` properties, but vendor bindings
selecting `snps,dwmac.yaml` do not impose these limits.

Define constraints for `interrupts` and `interrupt-names` properties in
various DWMAC vendor bindings to ensure proper validation and consistency.

Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Reviewed-by: Rob Herring (Arm) <robh@kernel.org>
Acked-by: Nobuhiro Iwamatsu <nobuhiro1.iwamatsu@toshiba.co.jp>
Link: https://patch.msgid.link/20250309003301.1152228-1-prabhakar.mahadev-lad.rj@bp.renesas.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'net-stmmac-dwmac-rk-validate-grf-and-peripheral-grf-during-probe'
Paolo Abeni [Thu, 13 Mar 2025 09:45:21 +0000 (10:45 +0100)]
Merge branch 'net-stmmac-dwmac-rk-validate-grf-and-peripheral-grf-during-probe'

Jonas Karlman says:

====================
net: stmmac: dwmac-rk: Validate GRF and peripheral GRF during probe

All Rockchip GMAC variants typically write to GRF regs to control e.g.
interface mode, speed and MAC rx/tx delay. Newer SoCs such as RK3576 and
RK3588 use a mix of GRF and peripheral GRF regs. These syscon regmaps is
located with help of a rockchip,grf and rockchip,php-grf phandle.

However, validating the rockchip,grf and rockchip,php-grf syscon regmap
is deferred until e.g. interface mode or speed is configured.

This series change to validate the GRF and peripheral GRF syscon regmap
at probe time to help simplify the SoC specific operations.

This should not introduce any backward compatibility issues as all
GMAC nodes have been added together with a rockchip,grf phandle (and
rockchip,php-grf where required) in their initial commit.
====================

Link: https://patch.msgid.link/20250308213720.2517944-1-jonas@kwiboo.se
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: stmmac: dwmac-rk: Remove unneeded GRF and peripheral GRF checks
Jonas Karlman [Sat, 8 Mar 2025 21:37:15 +0000 (21:37 +0000)]
net: stmmac: dwmac-rk: Remove unneeded GRF and peripheral GRF checks

Now that GRF, and peripheral GRF where needed, is validated at probe
time there is no longer any need to check and log an error in each SoC
specific operation.

Remove unneeded IS_ERR() checks and early bail out from each SoC
specific operation.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250308213720.2517944-4-jonas@kwiboo.se
Reviewed-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: stmmac: dwmac-rk: Validate GRF and peripheral GRF during probe
Jonas Karlman [Sat, 8 Mar 2025 21:37:14 +0000 (21:37 +0000)]
net: stmmac: dwmac-rk: Validate GRF and peripheral GRF during probe

All Rockchip GMAC variants typically write to GRF regs to control e.g.
interface mode, speed and MAC rx/tx delay. Newer SoCs such as RK3576 and
RK3588 use a mix of GRF and peripheral GRF regs. These syscon regmaps is
located with help of a rockchip,grf and rockchip,php-grf phandle.

However, validating the rockchip,grf and rockchip,php-grf syscon regmap
is deferred until e.g. interface mode or speed is configured, inside the
individual SoC specific operations.

Change to validate the rockchip,grf and rockchip,php-grf syscon regmap
at probe time to simplify all SoC specific operations.

This should not introduce any backward compatibility issues as all
GMAC nodes have been added together with a rockchip,grf phandle (and
rockchip,php-grf where required) in their initial commit.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250308213720.2517944-3-jonas@kwiboo.se
Reviewed-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agodt-bindings: net: rockchip-dwmac: Require rockchip,grf and rockchip,php-grf
Jonas Karlman [Sat, 8 Mar 2025 21:37:13 +0000 (21:37 +0000)]
dt-bindings: net: rockchip-dwmac: Require rockchip,grf and rockchip,php-grf

All Rockchip GMAC variants typically write to GRF regs to control e.g.
interface mode, speed and MAC rx/tx delay. Newer SoCs such as RK3562,
RK3576 and RK3588 use a mix of GRF and peripheral GRF regs.

Prior to the commit b331b8ef86f0 ("dt-bindings: net: convert
rockchip-dwmac to json-schema") the property rockchip,grf was listed
under "Required properties". During the conversion this was lost and
rockchip,grf has since then incorrectly been treated as optional and
not as required.

Similarly, when rockchip,php-grf was added to the schema in the
commit a2b77831427c ("dt-bindings: net: rockchip-dwmac: add rk3588 gmac
compatible") it also incorrectly has been treated as optional for all
GMAC variants, when it should have been required for RK3588, and later
also for RK3576.

Update this binding to require rockchip,grf and rockchip,php-grf to
properly reflect that GRF (and peripheral GRF for RK3576/RK3588) is
required to control part of GMAC.

This should not introduce any breakage as all Rockchip GMAC nodes have
been added together with a rockchip,grf phandle (and rockchip,php-grf
where required) in their initial commit.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://patch.msgid.link/20250308213720.2517944-2-jonas@kwiboo.se
Reviewed-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoRevert "openvswitch: switch to per-action label counting in conntrack"
Xin Long [Sat, 8 Mar 2025 18:05:43 +0000 (13:05 -0500)]
Revert "openvswitch: switch to per-action label counting in conntrack"

Currently, ovs_ct_set_labels() is only called for confirmed conntrack
entries (ct) within ovs_ct_commit(). However, if the conntrack entry
does not have the labels_ext extension, attempting to allocate it in
ovs_ct_get_conn_labels() for a confirmed entry triggers a warning in
nf_ct_ext_add():

  WARN_ON(nf_ct_is_confirmed(ct));

This happens when the conntrack entry is created externally before OVS
increments net->ct.labels_used. The issue has become more likely since
commit fcb1aa5163b1 ("openvswitch: switch to per-action label counting
in conntrack"), which changed to use per-action label counting and
increment net->ct.labels_used when a flow with ct action is added.

Since there’s no straightforward way to fully resolve this issue at the
moment, this reverts the commit to avoid breaking existing use cases.

Fixes: fcb1aa5163b1 ("openvswitch: switch to per-action label counting in conntrack")
Reported-by: Jianbo Liu <jianbol@nvidia.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Aaron Conole <aconole@redhat.com>
Link: https://patch.msgid.link/1bdeb2f3a812bca016a225d3de714427b2cd4772.1741457143.git.lucien.xin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: openvswitch: remove misbehaving actions length check
Ilya Maximets [Sat, 8 Mar 2025 00:45:59 +0000 (01:45 +0100)]
net: openvswitch: remove misbehaving actions length check

The actions length check is unreliable and produces different results
depending on the initial length of the provided netlink attribute and
the composition of the actual actions inside of it.  For example, a
user can add 4088 empty clone() actions without triggering -EMSGSIZE,
on attempt to add 4089 such actions the operation will fail with the
-EMSGSIZE verdict.  However, if another 16 KB of other actions will
be *appended* to the previous 4089 clone() actions, the check passes
and the flow is successfully installed into the openvswitch datapath.

The reason for a such a weird behavior is the way memory is allocated.
When ovs_flow_cmd_new() is invoked, it calls ovs_nla_copy_actions(),
that in turn calls nla_alloc_flow_actions() with either the actual
length of the user-provided actions or the MAX_ACTIONS_BUFSIZE.  The
function adds the size of the sw_flow_actions structure and then the
actually allocated memory is rounded up to the closest power of two.

So, if the user-provided actions are larger than MAX_ACTIONS_BUFSIZE,
then MAX_ACTIONS_BUFSIZE + sizeof(*sfa) rounded up is 32K + 24 -> 64K.
Later, while copying individual actions, we look at ksize(), which is
64K, so this way the MAX_ACTIONS_BUFSIZE check is not actually
triggered and the user can easily allocate almost 64 KB of actions.

However, when the initial size is less than MAX_ACTIONS_BUFSIZE, but
the actions contain ones that require size increase while copying
(such as clone() or sample()), then the limit check will be performed
during the reserve_sfa_size() and the user will not be allowed to
create actions that yield more than 32 KB internally.

This is one part of the problem.  The other part is that it's not
actually possible for the userspace application to know beforehand
if the particular set of actions will be rejected or not.

Certain actions require more space in the internal representation,
e.g. an empty clone() takes 4 bytes in the action list passed in by
the user, but it takes 12 bytes in the internal representation due
to an extra nested attribute, and some actions require less space in
the internal representations, e.g. set(tunnel(..)) normally takes
64+ bytes in the action list provided by the user, but only needs to
store a single pointer in the internal implementation, since all the
data is stored in the tunnel_info structure instead.

And the action size limit is applied to the internal representation,
not to the action list passed by the user.  So, it's not possible for
the userpsace application to predict if the certain combination of
actions will be rejected or not, because it is not possible for it to
calculate how much space these actions will take in the internal
representation without knowing kernel internals.

All that is causing random failures in ovs-vswitchd in userspace and
inability to handle certain traffic patterns as a result.  For example,
it is reported that adding a bit more than a 1100 VMs in an OpenStack
setup breaks the network due to OVS not being able to handle ARP
traffic anymore in some cases (it tries to install a proper datapath
flow, but the kernel rejects it with -EMSGSIZE, even though the action
list isn't actually that large.)

Kernel behavior must be consistent and predictable in order for the
userspace application to use it in a reasonable way.  ovs-vswitchd has
a mechanism to re-direct parts of the traffic and partially handle it
in userspace if the required action list is oversized, but that doesn't
work properly if we can't actually tell if the action list is oversized
or not.

Solution for this is to check the size of the user-provided actions
instead of the internal representation.  This commit just removes the
check from the internal part because there is already an implicit size
check imposed by the netlink protocol.  The attribute can't be larger
than 64 KB.  Realistically, we could reduce the limit to 32 KB, but
we'll be risking to break some existing setups that rely on the fact
that it's possible to create nearly 64 KB action lists today.

Vast majority of flows in real setups are below 100-ish bytes.  So
removal of the limit will not change real memory consumption on the
system.  The absolutely worst case scenario is if someone adds a flow
with 64 KB of empty clone() actions.  That will yield a 192 KB in the
internal representation consuming 256 KB block of memory.  However,
that list of actions is not meaningful and also a no-op.  Real world
very large action lists (that can occur for a rare cases of BUM
traffic handling) are unlikely to contain a large number of clones and
will likely have a lot of tunnel attributes making the internal
representation comparable in size to the original action list.
So, it should be fine to just remove the limit.

Commit in the 'Fixes' tag is the first one that introduced the
difference between internal representation and the user-provided action
lists, but there were many more afterwards that lead to the situation
we have today.

Fixes: 7d5437c709de ("openvswitch: Add tunneling interface.")
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
Reviewed-by: Aaron Conole <aconole@redhat.com>
Link: https://patch.msgid.link/20250308004609.2881861-1-i.maximets@ovn.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'gre-fix-regressions-in-ipv6-link-local-address-generation'
Paolo Abeni [Thu, 13 Mar 2025 09:17:46 +0000 (10:17 +0100)]
Merge branch 'gre-fix-regressions-in-ipv6-link-local-address-generation'

Guillaume Nault says:

====================
gre: Fix regressions in IPv6 link-local address generation.

IPv6 link-local address generation has some special cases for GRE
devices. This has led to several regressions in the past, and some of
them are still not fixed. This series fixes the remaining problems,
like the ipv6.conf.<dev>.addr_gen_mode sysctl being ignored and the
router discovery process not being started (see details in patch 1).

To avoid any further regressions, patch 2 adds selftests covering
IPv4 and IPv6 gre/gretap devices with all combinations of currently
supported addr_gen_mode values.
====================

Link: https://patch.msgid.link/cover.1741375285.git.gnault@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoselftests: Add IPv6 link-local address generation tests for GRE devices.
Guillaume Nault [Fri, 7 Mar 2025 19:28:58 +0000 (20:28 +0100)]
selftests: Add IPv6 link-local address generation tests for GRE devices.

GRE devices have their special code for IPv6 link-local address
generation that has been the source of several regressions in the past.

Add selftest to check that all gre, ip6gre, gretap and ip6gretap get an
IPv6 link-link local address in accordance with the
net.ipv6.conf.<dev>.addr_gen_mode sysctl.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Tested-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Petr Machata <petrm@nvidia.com>
Link: https://patch.msgid.link/2d6772af8e1da9016b2180ec3f8d9ee99f470c77.1741375285.git.gnault@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agogre: Fix IPv6 link-local address generation.
Guillaume Nault [Fri, 7 Mar 2025 19:28:53 +0000 (20:28 +0100)]
gre: Fix IPv6 link-local address generation.

Use addrconf_addr_gen() to generate IPv6 link-local addresses on GRE
devices in most cases and fall back to using add_v4_addrs() only in
case the GRE configuration is incompatible with addrconf_addr_gen().

GRE used to use addrconf_addr_gen() until commit e5dd729460ca
("ip/ip6_gre: use the same logic as SIT interfaces when computing v6LL
address") restricted this use to gretap and ip6gretap devices, and
created add_v4_addrs() (borrowed from SIT) for non-Ethernet GRE ones.

The original problem came when commit 9af28511be10 ("addrconf: refuse
isatap eui64 for INADDR_ANY") made __ipv6_isatap_ifid() fail when its
addr parameter was 0. The commit says that this would create an invalid
address, however, I couldn't find any RFC saying that the generated
interface identifier would be wrong. Anyway, since gre over IPv4
devices pass their local tunnel address to __ipv6_isatap_ifid(), that
commit broke their IPv6 link-local address generation when the local
address was unspecified.

Then commit e5dd729460ca ("ip/ip6_gre: use the same logic as SIT
interfaces when computing v6LL address") tried to fix that case by
defining add_v4_addrs() and calling it to generate the IPv6 link-local
address instead of using addrconf_addr_gen() (apart for gretap and
ip6gretap devices, which would still use the regular
addrconf_addr_gen(), since they have a MAC address).

That broke several use cases because add_v4_addrs() isn't properly
integrated into the rest of IPv6 Neighbor Discovery code. Several of
these shortcomings have been fixed over time, but add_v4_addrs()
remains broken on several aspects. In particular, it doesn't send any
Router Sollicitations, so the SLAAC process doesn't start until the
interface receives a Router Advertisement. Also, add_v4_addrs() mostly
ignores the address generation mode of the interface
(/proc/sys/net/ipv6/conf/*/addr_gen_mode), thus breaking the
IN6_ADDR_GEN_MODE_RANDOM and IN6_ADDR_GEN_MODE_STABLE_PRIVACY cases.

Fix the situation by using add_v4_addrs() only in the specific scenario
where the normal method would fail. That is, for interfaces that have
all of the following characteristics:

  * run over IPv4,
  * transport IP packets directly, not Ethernet (that is, not gretap
    interfaces),
  * tunnel endpoint is INADDR_ANY (that is, 0),
  * device address generation mode is EUI64.

In all other cases, revert back to the regular addrconf_addr_gen().

Also, remove the special case for ip6gre interfaces in add_v4_addrs(),
since ip6gre devices now always use addrconf_addr_gen() instead.

Fixes: e5dd729460ca ("ip/ip6_gre: use the same logic as SIT interfaces when computing v6LL address")
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/559c32ce5c9976b269e6337ac9abb6a96abe5096.1741375285.git.gnault@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: hsr: Add KUnit test for PRP
Jaakko Karrenpalo [Fri, 7 Mar 2025 16:17:00 +0000 (18:17 +0200)]
net: hsr: Add KUnit test for PRP

Add unit tests for the PRP duplicate detection

Signed-off-by: Jaakko Karrenpalo <jkarrenpalo@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250307161700.1045-2-jkarrenpalo@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: hsr: Fix PRP duplicate detection
Jaakko Karrenpalo [Fri, 7 Mar 2025 16:16:59 +0000 (18:16 +0200)]
net: hsr: Fix PRP duplicate detection

Add PRP specific function for handling duplicate
packets. This is needed because of potential
L2 802.1p prioritization done by network switches.

The L2 prioritization can re-order the PRP packets
from a node causing the existing implementation to
discard the frame(s) that have been received 'late'
because the sequence number is before the previous
received packet. This can happen if the node is
sending multiple frames back-to-back with different
priority.

Signed-off-by: Jaakko Karrenpalo <jkarrenpalo@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250307161700.1045-1-jkarrenpalo@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonetfilter: nft_exthdr: fix offset with ipv4_find_option()
Alexey Kashavkin [Sat, 1 Mar 2025 21:14:36 +0000 (00:14 +0300)]
netfilter: nft_exthdr: fix offset with ipv4_find_option()

There is an incorrect calculation in the offset variable which causes
the nft_skb_copy_to_reg() function to always return -EFAULT. Adding the
start variable is redundant. In the __ip_options_compile() function the
correct offset is specified when finding the function. There is no need
to add the size of the iphdr structure to the offset.

Fixes: dbb5281a1f84 ("netfilter: nf_tables: add support for matching IPv4 options")
Signed-off-by: Alexey Kashavkin <akashavkin@gmail.com>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 months agonet: cn23xx: fix typos
Janik Haag [Fri, 7 Mar 2025 14:56:49 +0000 (15:56 +0100)]
net: cn23xx: fix typos

This patch fixes a few typos, spelling mistakes, and a bit of grammar,
increasing the comments readability.

Signed-off-by: Janik Haag <janik@aq0.de>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250307145648.1679912-2-janik@aq0.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: hns3: use string choices helper
Jian Shen [Fri, 7 Mar 2025 11:37:33 +0000 (19:37 +0800)]
net: hns3: use string choices helper

Use string choices helper for better readability.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250307113733.819448-1-shaojijie@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge tag 'sched_ext-for-6.14-rc6-fixes' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 12 Mar 2025 21:52:04 +0000 (11:52 -1000)]
Merge tag 'sched_ext-for-6.14-rc6-fixes' of git://git./linux/kernel/git/tj/sched_ext

Pull sched_ext fix from Tejun Heo:
 "BPF schedulers could trigger a crash by passing in an invalid CPU to
  the scx_bpf_select_cpu_dfl() helper.

  Fix it by verifying input validity"

* tag 'sched_ext-for-6.14-rc6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext:
  sched_ext: Validate prev_cpu in scx_bpf_select_cpu_dfl()

3 months agoMerge tag 'spi-fix-v6.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/brooni...
Linus Torvalds [Wed, 12 Mar 2025 21:47:24 +0000 (11:47 -1000)]
Merge tag 'spi-fix-v6.14-rc6' of git://git./linux/kernel/git/broonie/spi

Pull spi fixes from Mark Brown:
 "A couple of driver specific fixes, an error handling fix for the Atmel
  QuadSPI driver and a fix for a nasty synchronisation issue in the data
  path for the Microchip driver which affects larger transfers.

  There's also a MAINTAINERS update for the Samsung driver"

* tag 'spi-fix-v6.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi:
  spi: microchip-core: prevent RX overflows when transmit size > FIFO size
  MAINTAINERS: add tambarus as R for Samsung SPI
  spi: atmel-quadspi: remove references to runtime PM on error path

3 months agonetdevsim: 'support' multi-buf XDP
Jakub Kicinski [Tue, 11 Mar 2025 09:28:20 +0000 (10:28 +0100)]
netdevsim: 'support' multi-buf XDP

Don't error out on large MTU if XDP is multi-buf.
The ping test now tests ping with XDP and high MTU.
netdevsim doesn't actually run the prog (yet?) so
it doesn't matter if the prog was multi-buf..

Reviewed-by: Michal Kubiak <michal.kubiak@intel.com>
Link: https://patch.msgid.link/20250311092820.542148-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoMerge branch 'net-remove-rtnl_lock-from-the-callers-of-queue-apis'
Jakub Kicinski [Wed, 12 Mar 2025 20:32:40 +0000 (13:32 -0700)]
Merge branch 'net-remove-rtnl_lock-from-the-callers-of-queue-apis'

Stanislav Fomichev says:

====================
net: remove rtnl_lock from the callers of queue APIs

All drivers that use queue management APIs already depend on the netdev
lock. Ultimately, we want to have most of the paths that work with
specific netdev to be rtnl_lock-free (ethtool mostly in particular).
Queue API currently has a much smaller API surface, so start with
rtnl_lock from it:

- add mutex to each dmabuf binding (to replace rtnl_lock)
- move netdev lock management to the callers of netdev_rx_queue_restart
  and drop rtnl_lock
====================

Link: https://patch.msgid.link/20250311144026.4154277-1-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet: drop rtnl_lock for queue_mgmt operations
Stanislav Fomichev [Tue, 11 Mar 2025 14:40:26 +0000 (07:40 -0700)]
net: drop rtnl_lock for queue_mgmt operations

All drivers that use queue API are already converted to use
netdev instance lock. Move netdev instance lock management to
the netlink layer and drop rtnl_lock.

Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Mina Almasry. <almasrymina@google.com>
Link: https://patch.msgid.link/20250311144026.4154277-4-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet: add granular lock for the netdev netlink socket
Stanislav Fomichev [Tue, 11 Mar 2025 14:40:25 +0000 (07:40 -0700)]
net: add granular lock for the netdev netlink socket

As we move away from rtnl_lock for queue ops, introduce
per-netdev_nl_sock lock.

Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/20250311144026.4154277-3-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet: create netdev_nl_sock to wrap bindings list
Stanislav Fomichev [Tue, 11 Mar 2025 14:40:24 +0000 (07:40 -0700)]
net: create netdev_nl_sock to wrap bindings list

No functional changes. Next patches will add more granular locking
to netdev_nl_sock.

Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/20250311144026.4154277-2-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet/mlx5: Avoid unnecessary use of comma operator
Simon Horman [Fri, 7 Mar 2025 12:39:33 +0000 (12:39 +0000)]
net/mlx5: Avoid unnecessary use of comma operator

Although it does not seem to have any untoward side-effects,
the use of ';' to separate to assignments seems more appropriate than ','.

Flagged by clang-19 -Wcomma

No functional change intended.
Compile tested only.

Signed-off-by: Simon Horman <horms@kernel.org>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20250307-mlx5-comma-v1-1-934deb6927bb@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoselftests: net: bump GRO timeout for gro/setup_veth
Jakub Kicinski [Mon, 10 Mar 2025 11:08:21 +0000 (12:08 +0100)]
selftests: net: bump GRO timeout for gro/setup_veth

Commit 51bef03e1a71 ("selftests/net: deflake GRO tests") recently
switched to NAPI suspension, and lowered the timeout from 1ms to 100us.
This started causing flakes in netdev-run CI. Let's bump it to 200us.
In a quick test of a debug kernel I see failures with 100us, with 200us
in 5 runs I see 2 completely clean runs and 3 with a single retry
(GRO test will retry up to 5 times).

Reviewed-by: Kevin Krakauer <krakauer@google.com>
Link: https://patch.msgid.link/20250310110821.385621-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoeth: bnxt: add missing netdev lock management to bnxt_dl_reload_up
Stanislav Fomichev [Sun, 9 Mar 2025 21:58:51 +0000 (14:58 -0700)]
eth: bnxt: add missing netdev lock management to bnxt_dl_reload_up

bnxt_dl_reload_up is completely missing instance lock management
which can result in `devlink dev reload` leaving with instance
lock held. Add the missing calls.

Also add netdev_assert_locked to make it clear that the up() method
is running with the instance lock grabbed.

v2:
- add net/netdev_lock.h include to bnxt_devlink.c for netdev_assert_locked

Fixes: 004b5008016a ("eth: bnxt: remove most dependencies on RTNL")
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250309215851.2003708-3-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoeth: bnxt: request unconditional ops lock
Stanislav Fomichev [Sun, 9 Mar 2025 21:58:50 +0000 (14:58 -0700)]
eth: bnxt: request unconditional ops lock

netdev_lock_ops conditionally grabs instance lock when queue_mgmt_ops
is defined. However queue_mgmt_ops support is signaled via FW
so we can sometimes boot without queue_mgmt_ops being set.
This will result in bnxt running without instance lock which
the driver now heavily depends on. Set request_ops_lock to true
unconditionally to always request netdev instance lock.

Fixes: 004b5008016a ("eth: bnxt: remove most dependencies on RTNL")
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250309215851.2003708-2-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoeth: bnxt: switch to netif_close
Stanislav Fomichev [Sun, 9 Mar 2025 21:58:49 +0000 (14:58 -0700)]
eth: bnxt: switch to netif_close

All (error) paths that call dev_close are already holding instance lock,
so switch to netif_close to avoid the deadlock.

v2:
- add missing EXPORT_MODULE for netif_close

Fixes: 004b5008016a ("eth: bnxt: remove most dependencies on RTNL")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250309215851.2003708-1-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet: revert to lockless TC_SETUP_BLOCK and TC_SETUP_FT
Stanislav Fomichev [Sat, 8 Mar 2025 04:47:26 +0000 (20:47 -0800)]
net: revert to lockless TC_SETUP_BLOCK and TC_SETUP_FT

There is a couple of places from which we can arrive to ndo_setup_tc
with TC_SETUP_BLOCK/TC_SETUP_FT:
- netlink
- netlink notifier
- netdev notifier

Locking netdev too deep in this call chain seems to be problematic
(especially assuming some/all of the call_netdevice_notifiers
NETDEV_UNREGISTER) might soon be running with the instance lock).
Revert to lockless ndo_setup_tc for TC_SETUP_BLOCK/TC_SETUP_FT. NFT
framework already takes care of most of the locking. Document
the assumptions.

ndo_setup_tc TC_SETUP_BLOCK
  nft_block_offload_cmd
    nft_chain_offload_cmd
      nft_flow_block_chain
        nft_flow_offload_chain
  nft_flow_rule_offload_abort
    nft_flow_rule_offload_commit
  nft_flow_rule_offload_commit
    nf_tables_commit
      nfnetlink_rcv_batch
        nfnetlink_rcv_skb_batch
  nfnetlink_rcv
nft_offload_netdev_event
  NETDEV_UNREGISTER notifier

ndo_setup_tc TC_SETUP_FT
  nf_flow_table_offload_cmd
    nf_flow_table_offload_setup
      nft_unregister_flowtable_hook
        nft_register_flowtable_net_hooks
  nft_flowtable_update
  nf_tables_newflowtable
    nfnetlink_rcv_batch (.call NFNL_CB_BATCH)
nft_flowtable_update
  nf_tables_newflowtable
nft_flowtable_event
  nf_tables_flowtable_event
    NETDEV_UNREGISTER notifier
      __nft_unregister_flowtable_net_hooks
        nft_unregister_flowtable_net_hooks
  nf_tables_commit
    nfnetlink_rcv_batch (.call NFNL_CB_BATCH)
  __nf_tables_abort
    nf_tables_abort
      nfnetlink_rcv_batch
__nft_release_hook
  __nft_release_hooks
    nf_tables_pre_exit_net -> module unload
  nft_rcv_nl_event
    netlink_register_notifier (oh boy)
      nft_register_flowtable_net_hooks
       nft_flowtable_update
  nf_tables_newflowtable
        nf_tables_newflowtable

Fixes: c4f0f30b424e ("net: hold netdev instance lock during nft ndo_setup_tc")
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
Reported-by: syzbot+0afb4bcf91e5a1afdcad@syzkaller.appspotmail.com
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250308044726.1193222-1-sdf@fomichev.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoMerge branch 'net_sched-prevent-creation-of-classes-with-tc_h_root'
Jakub Kicinski [Wed, 12 Mar 2025 19:51:30 +0000 (12:51 -0700)]
Merge branch 'net_sched-prevent-creation-of-classes-with-tc_h_root'

Cong Wang says:

====================
net_sched: Prevent creation of classes with TC_H_ROOT

This patchset contains a bug fix and its TDC test case.
====================

Link: https://patch.msgid.link/20250306232355.93864-1-xiyou.wangcong@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoselftests/tc-testing: Add a test case for DRR class with TC_H_ROOT
Cong Wang [Thu, 6 Mar 2025 23:23:55 +0000 (15:23 -0800)]
selftests/tc-testing: Add a test case for DRR class with TC_H_ROOT

Integrate the reproduer from Mingi to TDC.

All test results:

1..4
ok 1 0385 - Create DRR with default setting
ok 2 2375 - Delete DRR with handle
ok 3 3092 - Show DRR class
ok 4 4009 - Reject creation of DRR class with classid TC_H_ROOT

Cc: Mingi Cho <mincho@theori.io>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250306232355.93864-3-xiyou.wangcong@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agonet_sched: Prevent creation of classes with TC_H_ROOT
Cong Wang [Thu, 6 Mar 2025 23:23:54 +0000 (15:23 -0800)]
net_sched: Prevent creation of classes with TC_H_ROOT

The function qdisc_tree_reduce_backlog() uses TC_H_ROOT as a termination
condition when traversing up the qdisc tree to update parent backlog
counters. However, if a class is created with classid TC_H_ROOT, the
traversal terminates prematurely at this class instead of reaching the
actual root qdisc, causing parent statistics to be incorrectly maintained.
In case of DRR, this could lead to a crash as reported by Mingi Cho.

Prevent the creation of any Qdisc class with classid TC_H_ROOT
(0xFFFFFFFF) across all qdisc types, as suggested by Jamal.

Reported-by: Mingi Cho <mincho@theori.io>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Fixes: 066a3b5b2346 ("[NET_SCHED] sch_api: fix qdisc_tree_decrease_qlen() loop")
Link: https://patch.msgid.link/20250306232355.93864-2-xiyou.wangcong@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 months agoipvs: prevent integer overflow in do_ip_vs_get_ctl()
Dan Carpenter [Mon, 10 Mar 2025 07:45:53 +0000 (10:45 +0300)]
ipvs: prevent integer overflow in do_ip_vs_get_ctl()

The get->num_services variable is an unsigned int which is controlled by
the user.  The struct_size() function ensures that the size calculation
does not overflow an unsigned long, however, we are saving the result to
an int so the calculation can overflow.

Both "len" and "get->num_services" come from the user.  This check is
just a sanity check to help the user and ensure they are using the API
correctly.  An integer overflow here is not a big deal.  This has no
security impact.

Save the result from struct_size() type size_t to fix this integer
overflow bug.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 months agoselftests: netfilter: skip br_netfilter queue tests if kernel is tainted
Florian Westphal [Tue, 11 Mar 2025 11:52:45 +0000 (12:52 +0100)]
selftests: netfilter: skip br_netfilter queue tests if kernel is tainted

These scripts fail if the kernel is tainted which leads to wrong test
failure reports in CI environments when an unrelated test triggers some
splat.

Check taint state at start of script and SKIP if its already dodgy.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 months agonetfilter: nf_conncount: Fully initialize struct nf_conncount_tuple in insert_tree()
Kohei Enju [Sun, 9 Mar 2025 08:07:38 +0000 (17:07 +0900)]
netfilter: nf_conncount: Fully initialize struct nf_conncount_tuple in insert_tree()

Since commit b36e4523d4d5 ("netfilter: nf_conncount: fix garbage
collection confirm race"), `cpu` and `jiffies32` were introduced to
the struct nf_conncount_tuple.

The commit made nf_conncount_add() initialize `conn->cpu` and
`conn->jiffies32` when allocating the struct.
In contrast, count_tree() was not changed to initialize them.

By commit 34848d5c896e ("netfilter: nf_conncount: Split insert and
traversal"), count_tree() was split and the relevant allocation
code now resides in insert_tree().
Initialize `conn->cpu` and `conn->jiffies32` in insert_tree().

BUG: KMSAN: uninit-value in find_or_evict net/netfilter/nf_conncount.c:117 [inline]
BUG: KMSAN: uninit-value in __nf_conncount_add+0xd9c/0x2850 net/netfilter/nf_conncount.c:143
 find_or_evict net/netfilter/nf_conncount.c:117 [inline]
 __nf_conncount_add+0xd9c/0x2850 net/netfilter/nf_conncount.c:143
 count_tree net/netfilter/nf_conncount.c:438 [inline]
 nf_conncount_count+0x82f/0x1e80 net/netfilter/nf_conncount.c:521
 connlimit_mt+0x7f6/0xbd0 net/netfilter/xt_connlimit.c:72
 __nft_match_eval net/netfilter/nft_compat.c:403 [inline]
 nft_match_eval+0x1a5/0x300 net/netfilter/nft_compat.c:433
 expr_call_ops_eval net/netfilter/nf_tables_core.c:240 [inline]
 nft_do_chain+0x426/0x2290 net/netfilter/nf_tables_core.c:288
 nft_do_chain_ipv4+0x1a5/0x230 net/netfilter/nft_chain_filter.c:23
 nf_hook_entry_hookfn include/linux/netfilter.h:154 [inline]
 nf_hook_slow+0xf4/0x400 net/netfilter/core.c:626
 nf_hook_slow_list+0x24d/0x860 net/netfilter/core.c:663
 NF_HOOK_LIST include/linux/netfilter.h:350 [inline]
 ip_sublist_rcv+0x17b7/0x17f0 net/ipv4/ip_input.c:633
 ip_list_rcv+0x9ef/0xa40 net/ipv4/ip_input.c:669
 __netif_receive_skb_list_ptype net/core/dev.c:5936 [inline]
 __netif_receive_skb_list_core+0x15c5/0x1670 net/core/dev.c:5983
 __netif_receive_skb_list net/core/dev.c:6035 [inline]
 netif_receive_skb_list_internal+0x1085/0x1700 net/core/dev.c:6126
 netif_receive_skb_list+0x5a/0x460 net/core/dev.c:6178
 xdp_recv_frames net/bpf/test_run.c:280 [inline]
 xdp_test_run_batch net/bpf/test_run.c:361 [inline]
 bpf_test_run_xdp_live+0x2e86/0x3480 net/bpf/test_run.c:390
 bpf_prog_test_run_xdp+0xf1d/0x1ae0 net/bpf/test_run.c:1316
 bpf_prog_test_run+0x5e5/0xa30 kernel/bpf/syscall.c:4407
 __sys_bpf+0x6aa/0xd90 kernel/bpf/syscall.c:5813
 __do_sys_bpf kernel/bpf/syscall.c:5902 [inline]
 __se_sys_bpf kernel/bpf/syscall.c:5900 [inline]
 __ia32_sys_bpf+0xa0/0xe0 kernel/bpf/syscall.c:5900
 ia32_sys_call+0x394d/0x4180 arch/x86/include/generated/asm/syscalls_32.h:358
 do_syscall_32_irqs_on arch/x86/entry/common.c:165 [inline]
 __do_fast_syscall_32+0xb0/0x110 arch/x86/entry/common.c:387
 do_fast_syscall_32+0x38/0x80 arch/x86/entry/common.c:412
 do_SYSENTER_32+0x1f/0x30 arch/x86/entry/common.c:450
 entry_SYSENTER_compat_after_hwframe+0x84/0x8e

Uninit was created at:
 slab_post_alloc_hook mm/slub.c:4121 [inline]
 slab_alloc_node mm/slub.c:4164 [inline]
 kmem_cache_alloc_noprof+0x915/0xe10 mm/slub.c:4171
 insert_tree net/netfilter/nf_conncount.c:372 [inline]
 count_tree net/netfilter/nf_conncount.c:450 [inline]
 nf_conncount_count+0x1415/0x1e80 net/netfilter/nf_conncount.c:521
 connlimit_mt+0x7f6/0xbd0 net/netfilter/xt_connlimit.c:72
 __nft_match_eval net/netfilter/nft_compat.c:403 [inline]
 nft_match_eval+0x1a5/0x300 net/netfilter/nft_compat.c:433
 expr_call_ops_eval net/netfilter/nf_tables_core.c:240 [inline]
 nft_do_chain+0x426/0x2290 net/netfilter/nf_tables_core.c:288
 nft_do_chain_ipv4+0x1a5/0x230 net/netfilter/nft_chain_filter.c:23
 nf_hook_entry_hookfn include/linux/netfilter.h:154 [inline]
 nf_hook_slow+0xf4/0x400 net/netfilter/core.c:626
 nf_hook_slow_list+0x24d/0x860 net/netfilter/core.c:663
 NF_HOOK_LIST include/linux/netfilter.h:350 [inline]
 ip_sublist_rcv+0x17b7/0x17f0 net/ipv4/ip_input.c:633
 ip_list_rcv+0x9ef/0xa40 net/ipv4/ip_input.c:669
 __netif_receive_skb_list_ptype net/core/dev.c:5936 [inline]
 __netif_receive_skb_list_core+0x15c5/0x1670 net/core/dev.c:5983
 __netif_receive_skb_list net/core/dev.c:6035 [inline]
 netif_receive_skb_list_internal+0x1085/0x1700 net/core/dev.c:6126
 netif_receive_skb_list+0x5a/0x460 net/core/dev.c:6178
 xdp_recv_frames net/bpf/test_run.c:280 [inline]
 xdp_test_run_batch net/bpf/test_run.c:361 [inline]
 bpf_test_run_xdp_live+0x2e86/0x3480 net/bpf/test_run.c:390
 bpf_prog_test_run_xdp+0xf1d/0x1ae0 net/bpf/test_run.c:1316
 bpf_prog_test_run+0x5e5/0xa30 kernel/bpf/syscall.c:4407
 __sys_bpf+0x6aa/0xd90 kernel/bpf/syscall.c:5813
 __do_sys_bpf kernel/bpf/syscall.c:5902 [inline]
 __se_sys_bpf kernel/bpf/syscall.c:5900 [inline]
 __ia32_sys_bpf+0xa0/0xe0 kernel/bpf/syscall.c:5900
 ia32_sys_call+0x394d/0x4180 arch/x86/include/generated/asm/syscalls_32.h:358
 do_syscall_32_irqs_on arch/x86/entry/common.c:165 [inline]
 __do_fast_syscall_32+0xb0/0x110 arch/x86/entry/common.c:387
 do_fast_syscall_32+0x38/0x80 arch/x86/entry/common.c:412
 do_SYSENTER_32+0x1f/0x30 arch/x86/entry/common.c:450
 entry_SYSENTER_compat_after_hwframe+0x84/0x8e

Reported-by: syzbot+83fed965338b573115f7@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=83fed965338b573115f7
Fixes: b36e4523d4d5 ("netfilter: nf_conncount: fix garbage collection confirm race")
Signed-off-by: Kohei Enju <enjuk@amazon.com>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 months agoMerge tag 'wireless-2025-03-12' of https://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Wed, 12 Mar 2025 13:56:58 +0000 (13:56 +0000)]
Merge tag 'wireless-2025-03-12' of https://git./linux/kernel/git/wireless/wireless

Johannes berg says:

====================
Few more fixes:
 - cfg80211/mac80211
   - stop possible runaway wiphy worker
   - EHT should not use reserved MPDU size bits
   - don't run worker for stopped interfaces
   - fix SA Query processing with MLO
   - fix lookup of assoc link BSS entries
   - correct station flush on unauthorize
 - iwlwifi:
   - TSO fixes
   - fix non-MSI-X platforms
   - stop possible runaway restart worker
 - rejigger maintainers so I'm not CC'ed on
   everything
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 months agowifi: mac80211: fix MPDU length parsing for EHT 5/6 GHz
Benjamin Berg [Tue, 11 Mar 2025 11:17:04 +0000 (12:17 +0100)]
wifi: mac80211: fix MPDU length parsing for EHT 5/6 GHz

The MPDU length is only configured using the EHT capabilities element on
2.4 GHz. On 5/6 GHz it is configured using the VHT or HE capabilities
respectively.

Fixes: cf0079279727 ("wifi: mac80211: parse A-MSDU len from EHT capabilities")
Reviewed-by: Miriam Rachel Korenblit <miriam.rachel.korenblit@intel.com>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Link: https://patch.msgid.link/20250311121704.0634d31f0883.I28063e4d3ef7d296b7e8a1c303460346a30bf09c@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
3 months agoMerge tag 'hyperv-fixes-signed-20250311' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 11 Mar 2025 22:49:51 +0000 (12:49 -1000)]
Merge tag 'hyperv-fixes-signed-20250311' of git://git./linux/kernel/git/hyperv/linux

Pull hyperv fixes from Wei Liu:

 - Patches to fix Hyper-v framebuffer code (Michael Kelley and Saurabh
   Sengar)

 - Fix for Hyper-V output argument to hypercall that changes page
   visibility (Michael Kelley)

 - Fix for Hyper-V VTL mode (Naman Jain)

* tag 'hyperv-fixes-signed-20250311' of git://git.kernel.org/pub/scm/linux/kernel/git/hyperv/linux:
  Drivers: hv: vmbus: Don't release fb_mmio resource in vmbus_free_mmio()
  x86/hyperv: Fix output argument to hypercall that changes page visibility
  fbdev: hyperv_fb: Allow graceful removal of framebuffer
  fbdev: hyperv_fb: Simplify hvfb_putmem
  fbdev: hyperv_fb: Fix hang in kdump kernel when on Hyper-V Gen 2 VMs
  drm/hyperv: Fix address space leak when Hyper-V DRM device is removed
  fbdev: hyperv_fb: iounmap() the correct memory when removing a device
  x86/hyperv/vtl: Stop kernel from probing VTL0 low memory

3 months agoMerge tag 'pinctrl-v6.14-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Tue, 11 Mar 2025 17:29:02 +0000 (07:29 -1000)]
Merge tag 'pinctrl-v6.14-3' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fixes from Linus Walleij:

 - Fix the regmap settings for bcm281xx, this was missing the stride

 - NULL check for the Nuvoton npcm8xx devm_kasprintf()

 - Enable the Spacemit pin controller by default in the SoC config. The
   SoC will not boot without it so this one is pretty much required

* tag 'pinctrl-v6.14-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: spacemit: enable config option
  pinctrl: nuvoton: npcm8xx: Add NULL check in npcm8xx_gpio_fw
  pinctrl: bcm281xx: Fix incorrect regmap max_registers value

3 months agoqlcnic: fix memory leak issues in qlcnic_sriov_common.c
Haoxiang Li [Fri, 7 Mar 2025 09:49:52 +0000 (17:49 +0800)]
qlcnic: fix memory leak issues in qlcnic_sriov_common.c

Add qlcnic_sriov_free_vlans() in qlcnic_sriov_alloc_vlans() if
any sriov_vlans fails to be allocated.
Add qlcnic_sriov_free_vlans() to free the memory allocated by
qlcnic_sriov_alloc_vlans() if "sriov->allowed_vlans" fails to
be allocated.

Fixes: 91b7282b613d ("qlcnic: Support VLAN id config.")
Cc: stable@vger.kernel.org
Signed-off-by: Haoxiang Li <haoxiang_li2024@163.com>
Link: https://patch.msgid.link/20250307094952.14874-1-haoxiang_li2024@163.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agodocs: netdev: add a note on selftest posting
Jakub Kicinski [Thu, 6 Mar 2025 18:05:33 +0000 (10:05 -0800)]
docs: netdev: add a note on selftest posting

We haven't had much discussion on the list about this, but
a handful of people have been confused about rules on
posting selftests for fixes, lately. I tend to post fixes
with their respective selftests in the same series.
There are tradeoffs around size of the net tree and conflicts
but so far it hasn't been a major issue.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Link: https://patch.msgid.link/20250306180533.1864075-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agortase: Fix improper release of ring list entries in rtase_sw_reset
Justin Lai [Thu, 6 Mar 2025 07:05:10 +0000 (15:05 +0800)]
rtase: Fix improper release of ring list entries in rtase_sw_reset

Since rtase_init_ring, which is called within rtase_sw_reset, adds ring
entries already present in the ring list back into the list, it causes
the ring list to form a cycle. This results in list_for_each_entry_safe
failing to find an endpoint during traversal, leading to an error.
Therefore, it is necessary to remove the previously added ring_list nodes
before calling rtase_init_ring.

Fixes: 079600489960 ("rtase: Implement net_device_ops")
Signed-off-by: Justin Lai <justinlai0215@realtek.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250306070510.18129-1-justinlai0215@realtek.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'bonding-fix-incorrect-mac-address-setting'
Paolo Abeni [Tue, 11 Mar 2025 12:19:29 +0000 (13:19 +0100)]
Merge branch 'bonding-fix-incorrect-mac-address-setting'

Hangbin Liu says:

====================
bonding: fix incorrect mac address setting

The mac address on backup slave should be convert from Solicited-Node
Multicast address, not from bonding unicast target address.
====================

Link: https://patch.msgid.link/20250306023923.38777-1-liuhangbin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoselftests: bonding: fix incorrect mac address
Hangbin Liu [Thu, 6 Mar 2025 02:39:23 +0000 (02:39 +0000)]
selftests: bonding: fix incorrect mac address

The correct mac address for NS target 2001:db8::254 is 33:33:ff:00:02:54,
not 33:33:00:00:02:54. The same with client maddress.

Fixes: 86fb6173d11e ("selftests: bonding: add ns multicast group testing")
Acked-by: Jay Vosburgh <jv@jvosburgh.net>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250306023923.38777-3-liuhangbin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agobonding: fix incorrect MAC address setting to receive NS messages
Hangbin Liu [Thu, 6 Mar 2025 02:39:22 +0000 (02:39 +0000)]
bonding: fix incorrect MAC address setting to receive NS messages

When validation on the backup slave is enabled, we need to validate the
Neighbor Solicitation (NS) messages received on the backup slave. To
receive these messages, the correct destination MAC address must be added
to the slave. However, the target in bonding is a unicast address, which
we cannot use directly. Instead, we should first convert it to a
Solicited-Node Multicast Address and then derive the corresponding MAC
address.

Fix the incorrect MAC address setting on both slave_set_ns_maddr() and
slave_set_ns_maddrs(). Since the two function names are similar. Add
some description for the functions. Also only use one mac_addr variable
in slave_set_ns_maddr() to save some code and logic.

Fixes: 8eb36164d1a6 ("bonding: add ns target multicast address to slave device")
Acked-by: Jay Vosburgh <jv@jvosburgh.net>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250306023923.38777-2-liuhangbin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: mctp: unshare packets when reassembling
Matt Johnston [Thu, 6 Mar 2025 02:32:45 +0000 (10:32 +0800)]
net: mctp: unshare packets when reassembling

Ensure that the frag_list used for reassembly isn't shared with other
packets. This avoids incorrect reassembly when packets are cloned, and
prevents a memory leak due to circular references between fragments and
their skb_shared_info.

The upcoming MCTP-over-USB driver uses skb_clone which can trigger the
problem - other MCTP drivers don't share SKBs.

A kunit test is added to reproduce the issue.

Signed-off-by: Matt Johnston <matt@codeconstruct.com.au>
Fixes: 4a992bbd3650 ("mctp: Implement message fragmentation & reassembly")
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20250306-matt-mctp-usb-v1-1-085502b3dd28@codeconstruct.com.au
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agovboxsf: Add __nonstring annotations for unterminated strings
Kees Cook [Mon, 10 Mar 2025 22:25:31 +0000 (15:25 -0700)]
vboxsf: Add __nonstring annotations for unterminated strings

When a character array without a terminating NUL character has a static
initializer, GCC 15's -Wunterminated-string-initialization will only
warn if the array lacks the "nonstring" attribute[1]. Mark the arrays
with __nonstring to and correctly identify the char array as "not a C
string" and thereby eliminate the warning.

This effectively reverts the change in 4e7487245abc ("vboxsf: fix building
with GCC 15"), to add the annotation that has other uses (i.e. warning
if the string is ever used with C string APIs).

Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117178
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Brahmajit Das <brahmajit.xyz@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Kees Cook <kees@kernel.org>
Link: https://lore.kernel.org/r/20250310222530.work.374-kees@kernel.org
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
3 months agonet: switchdev: Convert blocking notification chain to a raw one
Amit Cohen [Wed, 5 Mar 2025 12:15:09 +0000 (14:15 +0200)]
net: switchdev: Convert blocking notification chain to a raw one

A blocking notification chain uses a read-write semaphore to protect the
integrity of the chain. The semaphore is acquired for writing when
adding / removing notifiers to / from the chain and acquired for reading
when traversing the chain and informing notifiers about an event.

In case of the blocking switchdev notification chain, recursive
notifications are possible which leads to the semaphore being acquired
twice for reading and to lockdep warnings being generated [1].

Specifically, this can happen when the bridge driver processes a
SWITCHDEV_BRPORT_UNOFFLOADED event which causes it to emit notifications
about deferred events when calling switchdev_deferred_process().

Fix this by converting the notification chain to a raw notification
chain in a similar fashion to the netdev notification chain. Protect
the chain using the RTNL mutex by acquiring it when modifying the chain.
Events are always informed under the RTNL mutex, but add an assertion in
call_switchdev_blocking_notifiers() to make sure this is not violated in
the future.

Maintain the "blocking" prefix as events are always emitted from process
context and listeners are allowed to block.

[1]:
WARNING: possible recursive locking detected
6.14.0-rc4-custom-g079270089484 #1 Not tainted
--------------------------------------------
ip/52731 is trying to acquire lock:
ffffffff850918d8 ((switchdev_blocking_notif_chain).rwsem){++++}-{4:4}, at: blocking_notifier_call_chain+0x58/0xa0

but task is already holding lock:
ffffffff850918d8 ((switchdev_blocking_notif_chain).rwsem){++++}-{4:4}, at: blocking_notifier_call_chain+0x58/0xa0

other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock((switchdev_blocking_notif_chain).rwsem);
lock((switchdev_blocking_notif_chain).rwsem);

*** DEADLOCK ***
May be due to missing lock nesting notation
3 locks held by ip/52731:
 #0: ffffffff84f795b0 (rtnl_mutex){+.+.}-{4:4}, at: rtnl_newlink+0x727/0x1dc0
 #1: ffffffff8731f628 (&net->rtnl_mutex){+.+.}-{4:4}, at: rtnl_newlink+0x790/0x1dc0
 #2: ffffffff850918d8 ((switchdev_blocking_notif_chain).rwsem){++++}-{4:4}, at: blocking_notifier_call_chain+0x58/0xa0

stack backtrace:
...
? __pfx_down_read+0x10/0x10
? __pfx_mark_lock+0x10/0x10
? __pfx_switchdev_port_attr_set_deferred+0x10/0x10
blocking_notifier_call_chain+0x58/0xa0
switchdev_port_attr_notify.constprop.0+0xb3/0x1b0
? __pfx_switchdev_port_attr_notify.constprop.0+0x10/0x10
? mark_held_locks+0x94/0xe0
? switchdev_deferred_process+0x11a/0x340
switchdev_port_attr_set_deferred+0x27/0xd0
switchdev_deferred_process+0x164/0x340
br_switchdev_port_unoffload+0xc8/0x100 [bridge]
br_switchdev_blocking_event+0x29f/0x580 [bridge]
notifier_call_chain+0xa2/0x440
blocking_notifier_call_chain+0x6e/0xa0
switchdev_bridge_port_unoffload+0xde/0x1a0
...

Fixes: f7a70d650b0b6 ("net: bridge: switchdev: Ensure deferred event delivery on unoffload")
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Tested-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://patch.msgid.link/20250305121509.631207-1-amcohen@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'net-ti-icssg-prueth-add-native-mode-xdp-support'
Paolo Abeni [Tue, 11 Mar 2025 10:10:16 +0000 (11:10 +0100)]
Merge branch 'net-ti-icssg-prueth-add-native-mode-xdp-support'

Meghana Malladi says:

====================
net: ti: icssg-prueth: Add native mode XDP support

This series adds native XDP support using page_pool.
XDP zero copy support is not included in this patch series.

Patch 1/3: Replaces skb with page pool for Rx buffer allocation
Patch 2/3: Adds prueth_swdata struct for SWDATA for all swdata cases
Patch 3/3: Introduces native mode XDP support

v3: https://lore.kernel.org/all/20250224110102.1528552-1-m-malladi@ti.com/
====================

Link: https://patch.msgid.link/20250305101422.1908370-1-m-malladi@ti.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: ti: icssg-prueth: Add XDP support
Roger Quadros [Wed, 5 Mar 2025 10:14:22 +0000 (15:44 +0530)]
net: ti: icssg-prueth: Add XDP support

Add native XDP support. We do not support zero copy yet.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Link: https://patch.msgid.link/20250305101422.1908370-4-m-malladi@ti.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: ti: icssg-prueth: introduce and use prueth_swdata struct for SWDATA
Roger Quadros [Wed, 5 Mar 2025 10:14:21 +0000 (15:44 +0530)]
net: ti: icssg-prueth: introduce and use prueth_swdata struct for SWDATA

We have different cases for SWDATA (skb, page, cmd, etc)
so it is better to have a dedicated data structure for that.
We can embed the type field inside the struct and use it
to interpret the data in completion handlers.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Link: https://patch.msgid.link/20250305101422.1908370-3-m-malladi@ti.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agonet: ti: icssg-prueth: Use page_pool API for RX buffer allocation
Roger Quadros [Wed, 5 Mar 2025 10:14:20 +0000 (15:44 +0530)]
net: ti: icssg-prueth: Use page_pool API for RX buffer allocation

This is to prepare for native XDP support.

The page pool API is more faster in allocating pages than
__alloc_skb(). Drawback is that it works at PAGE_SIZE granularity
so we are not efficient in memory usage.
i.e. we are using PAGE_SIZE (4KB) memory for 1.5KB max packet size.

Signed-off-by: Roger Quadros <rogerq@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Link: https://patch.msgid.link/20250305101422.1908370-2-m-malladi@ti.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoMerge branch 'enic-enable-32-64-byte-cqes-and-get-max-rx-tx-ring-size-from-hw'
Paolo Abeni [Tue, 11 Mar 2025 09:21:41 +0000 (10:21 +0100)]
Merge branch 'enic-enable-32-64-byte-cqes-and-get-max-rx-tx-ring-size-from-hw'

Satish Kharat via says:

====================
enic: enable 32, 64 byte cqes and get max rx/tx ring size from hw

This series enables using the max rx and tx ring sizes read from hw.
For newer hw that can be up to 16k entries. This requires bigger
completion entries for rx queues. This series enables the use of the
32 and 64 byte completion queues entries for enic rx queues on
supported hw versions. This is in addition to the exiting (default)
16 byte rx cqes.

Signed-off-by: Satish Kharat <satishkh@cisco.com>
====================

Link: https://patch.msgid.link/20250304-enic_cleanup_and_ext_cq-v2-0-85804263dad8@cisco.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoenic: get max rq & wq entries supported by hw, 16K queues
Satish Kharat [Wed, 5 Mar 2025 00:56:44 +0000 (19:56 -0500)]
enic: get max rq & wq entries supported by hw, 16K queues

Enables reading the max rq and wq entries supported from the hw.
Enables 16k rq and wq entries on hw that supports.

Co-developed-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Co-developed-by: John Daley <johndale@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Link: https://patch.msgid.link/20250304-enic_cleanup_and_ext_cq-v2-8-85804263dad8@cisco.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
3 months agoenic: cleanup of enic wq request completion path
Satish Kharat [Wed, 5 Mar 2025 00:56:43 +0000 (19:56 -0500)]
enic: cleanup of enic wq request completion path

Cleans up the enic wq request completion path needed for 16k wq size
support.

Co-developed-by: Nelson Escobar <neescoba@cisco.com>
Signed-off-by: Nelson Escobar <neescoba@cisco.com>
Co-developed-by: John Daley <johndale@cisco.com>
Signed-off-by: John Daley <johndale@cisco.com>
Signed-off-by: Satish Kharat <satishkh@cisco.com>
Link: https://patch.msgid.link/20250304-enic_cleanup_and_ext_cq-v2-7-85804263dad8@cisco.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>