mvpp2_bm_switch_buffers() unconditionally calls
mvpp2_bm_pool_update_priv_fc() when switching between per-cpu and
shared buffer pool modes. This function programs CM3 flow control
registers via mvpp2_cm3_read()/mvpp2_cm3_write(), which dereference
priv->cm3_base without any NULL check.
When the CM3 SRAM resource is not present in the device tree (the
third reg entry added by commit 60523583b0 ("dts: marvell: add CM3
SRAM memory to cp11x ethernet device tree")), priv->cm3_base remains
NULL and priv->global_tx_fc is false. Any operation that triggers
mvpp2_bm_switch_buffers(), for example an MTU change that crosses
the jumbo frame threshold, will crash:
Unable to handle kernel NULL pointer dereference at
virtual address 0000000000000000
Mem abort info:
ESR = 0x0000000096000006
EC = 0x25: DABT (current EL), IL = 32 bits
pc : readl+0x0/0x18
lr : mvpp2_cm3_read.isra.0+0x14/0x20
Call trace:
readl+0x0/0x18
mvpp2_bm_pool_update_fc+0x40/0x12c
mvpp2_bm_pool_update_priv_fc+0x94/0xd8
mvpp2_bm_switch_buffers.isra.0+0x80/0x1c0
mvpp2_change_mtu+0x140/0x380
__dev_set_mtu+0x1c/0x38
dev_set_mtu_ext+0x78/0x118
dev_set_mtu+0x48/0xa8
dev_ifsioc+0x21c/0x43c
dev_ioctl+0x2d8/0x42c
sock_ioctl+0x314/0x378
Every other flow control call site in the driver already guards
hardware access with either priv->global_tx_fc or port->tx_fc.
mvpp2_bm_switch_buffers() is the only place that omits this check.
Add the missing priv->global_tx_fc guard to both the disable and
re-enable calls in mvpp2_bm_switch_buffers(), consistent with the
rest of the driver.
Fixes: 3a616b92a9 ("net: mvpp2: Add TX flow control support for jumbo frames")
Signed-off-by: Muhammad Hammad Ijaz <mhijaz@amazon.com>
Reviewed-by: Gunnar Kudrjavets <gunnarku@amazon.com>
Link: https://patch.msgid.link/20260316193157.65748-1-mhijaz@amazon.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In IPSec full offload mode, the device reports an ESN (Extended
Sequence Number) wrap event to the driver. The driver validates this
event by querying the IPSec ASO and checking that the esn_event_arm
field is 0x0, which indicates an event has occurred. After handling
the event, the driver must re-arm the context by setting esn_event_arm
back to 0x1.
A race condition exists in this handling path. After validating the
event, the driver calls mlx5_accel_esp_modify_xfrm() to update the
kernel's xfrm state. This function temporarily releases and
re-acquires the xfrm state lock.
So, need to acknowledge the event first by setting esn_event_arm to
0x1. This prevents the driver from reprocessing the same ESN update if
the hardware sends events for other reason. Since the next ESN update
only occurs after nearly 2^31 packets are received, there's no risk of
missing an update, as it will happen long after this handling has
finished.
Processing the event twice causes the ESN high-order bits (esn_msb) to
be incremented incorrectly. The driver then programs the hardware with
this invalid ESN state, which leads to anti-replay failures and a
complete halt of IPSec traffic.
Fix this by re-arming the ESN event immediately after it is validated,
before calling mlx5_accel_esp_modify_xfrm(). This ensures that any
spurious, duplicate events are correctly ignored, closing the race
window.
Fixes: fef0667893 ("net/mlx5e: Fix ESN update kernel panic")
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260316094603.6999-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The query or updating IPSec offload object is through Access ASO WQE.
The driver uses a single mlx5e_ipsec_aso struct for each PF, which
contains a shared DMA-mapped context for all ASO operations.
A race condition exists because the ASO spinlock is released before
the hardware has finished processing WQE. If a second operation is
initiated immediately after, it overwrites the shared context in the
DMA area.
When the first operation's completion is processed later, it reads
this corrupted context, leading to unexpected behavior and incorrect
results.
This commit fixes the race by introducing a private context within
each IPSec offload object. The shared ASO context is now copied to
this private context while the ASO spinlock is held. Subsequent
processing uses this saved, per-object context, ensuring its integrity
is maintained.
Fixes: 1ed78fc033 ("net/mlx5e: Update IPsec soft and hard limits")
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260316094603.6999-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A lock dependency cycle exists where:
1. mlx5_ib_roce_init -> mlx5_core_uplink_netdev_event_replay ->
mlx5_blocking_notifier_call_chain (takes notifier_rwsem) ->
mlx5e_mdev_notifier_event -> mlx5_netdev_notifier_register ->
register_netdevice_notifier_dev_net (takes rtnl)
=> notifier_rwsem -> rtnl
2. mlx5e_probe -> _mlx5e_probe ->
mlx5_core_uplink_netdev_set (takes uplink_netdev_lock) ->
mlx5_blocking_notifier_call_chain (takes notifier_rwsem)
=> uplink_netdev_lock -> notifier_rwsem
3: devlink_nl_rate_set_doit -> devlink_nl_rate_set ->
mlx5_esw_devlink_rate_leaf_tx_max_set -> esw_qos_devlink_rate_to_mbps ->
mlx5_esw_qos_max_link_speed_get (takes rtnl) ->
mlx5_esw_qos_lag_link_speed_get_locked ->
mlx5_uplink_netdev_get (takes uplink_netdev_lock)
=> rtnl -> uplink_netdev_lock
=> BOOM! (lock cycle)
Fix that by restricting the rtnl-protected section to just the necessary
part, the call to netdev_master_upper_dev_get and speed querying, so
that the last lock dependency is avoided and the cycle doesn't close.
This is safe because mlx5_uplink_netdev_get uses netdev_hold to keep the
uplink netdev alive while its master device is queried.
Use this opportunity to rename the ambiguously-named "hold_rtnl_lock"
argument to "take_rtnl" and remove the "_locked" suffix from
mlx5_esw_qos_lag_link_speed_get_locked.
Fixes: 6b4be64fd9 ("net/mlx5e: Harden uplink netdev access against device unbind")
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260316094603.6999-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2026-03-17 (igc, iavf, libie)
Kohei Enju adds use of helper function to add missing update of
skb->tail when padding is needed for igc.
Zdenek Bouska clears stale XSK timestamps when taking down Tx rings on
igc.
Petr Oros changes handling of iavf VLAN filter handling when an added
VLAN is also on the delete list to which can race and cause the VLAN
filter to not be added.
Michal frees cmd_buf for libie firmware logging to stop memory leaks.
* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
libie: prevent memleak in fwlog code
iavf: fix VLAN filter lost on add/delete race
igc: fix page fault in XDP TX timestamps handling
igc: fix missing update of skb->tail in igc_xmit_frame()
====================
Link: https://patch.msgid.link/20260317211906.115505-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If hardware doesn't support RX Flow Filters, rx_fs_lock spinlock is not
initialized leading to the following assertion splat triggerable via
set_rxnfc callback.
INFO: trying to register non-static key.
The code is fine but needs lockdep annotation, or maybe
you didn't initialize this object before use?
turning off the locking correctness validator.
CPU: 1 PID: 949 Comm: syz.0.6 Not tainted 6.1.164+ #113
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.1-0-g3208b098f51a-prebuilt.qemu.org 04/01/2014
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0x8d/0xba lib/dump_stack.c:106
assign_lock_key kernel/locking/lockdep.c:974 [inline]
register_lock_class+0x141b/0x17f0 kernel/locking/lockdep.c:1287
__lock_acquire+0x74f/0x6c40 kernel/locking/lockdep.c:4928
lock_acquire kernel/locking/lockdep.c:5662 [inline]
lock_acquire+0x190/0x4b0 kernel/locking/lockdep.c:5627
__raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
_raw_spin_lock_irqsave+0x33/0x50 kernel/locking/spinlock.c:162
gem_del_flow_filter drivers/net/ethernet/cadence/macb_main.c:3562 [inline]
gem_set_rxnfc+0x533/0xac0 drivers/net/ethernet/cadence/macb_main.c:3667
ethtool_set_rxnfc+0x18c/0x280 net/ethtool/ioctl.c:961
__dev_ethtool net/ethtool/ioctl.c:2956 [inline]
dev_ethtool+0x229c/0x6290 net/ethtool/ioctl.c:3095
dev_ioctl+0x637/0x1070 net/core/dev_ioctl.c:510
sock_do_ioctl+0x20d/0x2c0 net/socket.c:1215
sock_ioctl+0x577/0x6d0 net/socket.c:1320
vfs_ioctl fs/ioctl.c:51 [inline]
__do_sys_ioctl fs/ioctl.c:870 [inline]
__se_sys_ioctl fs/ioctl.c:856 [inline]
__x64_sys_ioctl+0x18c/0x210 fs/ioctl.c:856
do_syscall_x64 arch/x86/entry/common.c:46 [inline]
do_syscall_64+0x35/0x80 arch/x86/entry/common.c:76
entry_SYSCALL_64_after_hwframe+0x6e/0xd8
A more straightforward solution would be to always initialize rx_fs_lock,
just like rx_fs_list. However, in this case the driver set_rxnfc callback
would return with a rather confusing error code, e.g. -EINVAL. So deny
set_rxnfc attempts directly if the RX filtering feature is not supported
by hardware.
Fixes: ae8223de3d ("net: macb: Added support for RX filtering")
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Link: https://patch.msgid.link/20260316103826.74506-2-pchelkin@ispras.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The ASYNC_EVENT_CMPL_EVENT_ID_DBG_BUF_PRODUCER handler in
bnxt_async_event_process() uses a firmware-supplied 'type' field
directly as an index into bp->bs_trace[] without bounds validation.
The 'type' field is a 16-bit value extracted from DMA-mapped completion
ring memory that the NIC writes directly to host RAM. A malicious or
compromised NIC can supply any value from 0 to 65535, causing an
out-of-bounds access into kernel heap memory.
The bnxt_bs_trace_check_wrap() call then dereferences bs_trace->magic_byte
and writes to bs_trace->last_offset and bs_trace->wrapped, leading to
kernel memory corruption or a crash.
Fix by adding a bounds check and defining BNXT_TRACE_MAX as
DBG_LOG_BUFFER_FLUSH_REQ_TYPE_ERR_QPC_TRACE + 1 to cover all currently
defined firmware trace types (0x0 through 0xc).
Fixes: 84fcd9449f ("bnxt_en: Manage the FW trace context memory")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Junrui Luo <moonafterrain@outlook.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/SYBPR01MB7881A253A1C9775D277F30E9AF42A@SYBPR01MB7881.ausprd01.prod.outlook.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All cmd_buf buffers are allocated and need to be freed after usage.
Add an error unwinding path that properly frees these buffers.
The memory leak happens whenever fwlog configuration is changed. For
example:
$echo 256K > /sys/kernel/debug/ixgbe/0000\:32\:00.0/fwlog/log_size
Fixes: 96a9a9341c ("ice: configure FW logging")
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
When iavf_add_vlan() finds an existing filter in IAVF_VLAN_REMOVE
state, it transitions the filter to IAVF_VLAN_ACTIVE assuming the
pending delete can simply be cancelled. However, there is no guarantee
that iavf_del_vlans() has not already processed the delete AQ request
and removed the filter from the PF. In that case the filter remains in
the driver's list as IAVF_VLAN_ACTIVE but is no longer programmed on
the NIC. Since iavf_add_vlans() only picks up filters in
IAVF_VLAN_ADD state, the filter is never re-added, and spoof checking
drops all traffic for that VLAN.
CPU0 CPU1 Workqueue
---- ---- ---------
iavf_del_vlan(vlan 100)
f->state = REMOVE
schedule AQ_DEL_VLAN
iavf_add_vlan(vlan 100)
f->state = ACTIVE
iavf_del_vlans()
f is ACTIVE, skip
iavf_add_vlans()
f is ACTIVE, skip
Filter is ACTIVE in driver but absent from NIC.
Transition to IAVF_VLAN_ADD instead and schedule
IAVF_FLAG_AQ_ADD_VLAN_FILTER so iavf_add_vlans() re-programs the
filter. A duplicate add is idempotent on the PF.
Fixes: 0c0da0e951 ("iavf: refactor VLAN filter states")
Signed-off-by: Petr Oros <poros@redhat.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
If an XDP application that requested TX timestamping is shutting down
while the link of the interface in use is still up the following kernel
splat is reported:
[ 883.803618] [ T1554] BUG: unable to handle page fault for address: ffffcfb6200fd008
...
[ 883.803650] [ T1554] Call Trace:
[ 883.803652] [ T1554] <TASK>
[ 883.803654] [ T1554] igc_ptp_tx_tstamp_event+0xdf/0x160 [igc]
[ 883.803660] [ T1554] igc_tsync_interrupt+0x2d5/0x300 [igc]
...
During shutdown of the TX ring the xsk_meta pointers are left behind, so
that the IRQ handler is trying to touch them.
This issue is now being fixed by cleaning up the stale xsk meta data on
TX shutdown. TX timestamps on other queues remain unaffected.
Fixes: 15fd021bc4 ("igc: Add Tx hardware timestamp request for AF_XDP zero-copy packet")
Signed-off-by: Zdenek Bouska <zdenek.bouska@siemens.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Tested-by: Avigail Dahan <avigailx.dahan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
igc_xmit_frame() misses updating skb->tail when the packet size is
shorter than the minimum one.
Use skb_put_padto() in alignment with other Intel Ethernet drivers.
Fixes: 0507ef8a03 ("igc: Add transmit and receive fastpath and interrupt handlers")
Signed-off-by: Kohei Enju <kohei@enjuk.jp>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Tested-by: Avigail Dahan <avigailx.dahan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
On some systems (e.g. iMac 20,1 with BCM57766), the tg3 driver reads
a default placeholder mac address (00:10:18:00:00:00) from the
mailbox. The correct value on those systems are stored in the
'local-mac-address' property.
This patch, detect the default value and tries to retrieve
the correct address from the device_get_mac_address
function instead.
The patch has been tested on two different systems:
- iMac 20,1 (BCM57766) model which use the local-mac-address property
- iMac 13,2 (BCM57766) model which can use the mailbox,
NVRAM or MAC control registers
Tested-by: Rishon Jonathan R <mithicalaviator85@gmail.com>
Co-developed-by: Vincent MORVAN <vinc@42.fr>
Signed-off-by: Vincent MORVAN <vinc@42.fr>
Signed-off-by: Paul SAGE <paul.sage@42.fr>
Signed-off-by: Atharva Tiwari <atharvatiwarilinuxdev@gmail.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Link: https://patch.msgid.link/20260314215432.3589-1-atharvatiwarilinuxdev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Do not run airoha_dev_stop routine explicitly in airoha_remove()
since ndo_stop() callback is already executed by unregister_netdev() in
__dev_close_many routine if necessary and, doing so, we will end up causing
an underflow in the qdma users atomic counters. Rely on networking subsystem
to stop the device removing the airoha_eth module.
Fixes: 23020f0493 ("net: airoha: Introduce ethernet support for EN7581 SoC")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260313-airoha-remove-ndo_stop-remove-net-v2-1-67542c3ceeca@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
On certain platforms, such as AMD Versal boards, the tx/rx queue pointer
registers are cleared after suspend, and the rx queue pointer register
is also disabled during suspend if WOL is enabled. Previously, we assumed
that these registers would be restored by macb_mac_link_up(). However,
in commit bf9cf80cab, macb_init_buffers() was moved from
macb_mac_link_up() to macb_open(). Therefore, we should call
macb_init_buffers() to reinitialize the tx/rx queue pointer registers
during resume.
Due to the reset of these two registers, we also need to adjust the
tx/rx rings accordingly. The tx ring will be handled by
gem_shuffle_tx_rings() in macb_mac_link_up(), so we only need to
initialize the rx ring here.
Fixes: bf9cf80cab ("net: macb: Fix tx/rx malfunction after phy link down and up")
Reported-by: Quanyang Wang <quanyang.wang@windriver.com>
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Tested-by: Quanyang Wang <quanyang.wang@windriver.com>
Cc: stable@vger.kernel.org
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260312-macb-versal-v1-2-467647173fa4@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Page recycling was removed from the XDP_DROP path in emac_run_xdp() to
avoid conflicts with AF_XDP zero-copy mode, which uses xsk_buff_free()
instead.
However, this causes a memory leak when running XDP programs that drop
packets in non-zero-copy mode (standard page pool mode). The pages are
never returned to the page pool, leading to OOM conditions.
Fix this by handling cleanup in the caller, emac_rx_packet().
When emac_run_xdp() returns ICSSG_XDP_CONSUMED for XDP_DROP, the
caller now recycles the page back to the page pool. The zero-copy
path, emac_rx_packet_zc() already handles cleanup correctly with
xsk_buff_free().
Fixes: 7a64bb388d ("net: ti: icssg-prueth: Add AF_XDP zero copy for RX")
Signed-off-by: Meghana Malladi <m-malladi@ti.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260311095441.1691636-1-m-malladi@ti.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A potential race condition exists in mana_hwc_destroy_channel() where
hwc->caller_ctx is freed before the HWC's Completion Queue (CQ) and
Event Queue (EQ) are destroyed. This allows an in-flight CQ interrupt
handler to dereference freed memory, leading to a use-after-free or
NULL pointer dereference in mana_hwc_handle_resp().
mana_smc_teardown_hwc() signals the hardware to stop but does not
synchronize against IRQ handlers already executing on other CPUs. The
IRQ synchronization only happens in mana_hwc_destroy_cq() via
mana_gd_destroy_eq() -> mana_gd_deregister_irq(). Since this runs
after kfree(hwc->caller_ctx), a concurrent mana_hwc_rx_event_handler()
can dereference freed caller_ctx (and rxq->msg_buf) in
mana_hwc_handle_resp().
Fix this by reordering teardown to reverse-of-creation order: destroy
the TX/RX work queues and CQ/EQ before freeing hwc->caller_ctx. This
ensures all in-flight interrupt handlers complete before the memory they
access is freed.
Fixes: ca9c54d2d6 ("net: mana: Add a driver for Microsoft Azure Network Adapter (MANA)")
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: Dipayaan Roy <dipayanroy@linux.microsoft.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/abHA3AjNtqa1nx9k@linuxonhyperv3.guj3yctzbm1etfxqx2vob5hsef.xx.internal.cloudapp.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The NIX RAS health reporter recovery routine checks nix_af_rvu_int to
decide whether to re-enable NIX_AF_RAS interrupts. This is the RVU
interrupt status field and is unrelated to RAS events, so the recovery
flow may incorrectly skip re-enabling NIX_AF_RAS interrupts.
Check nix_af_rvu_ras instead before writing NIX_AF_RAS_ENA_W1S.
Fixes: 5ed66306ea ("octeontx2-af: Add devlink health reporters for NIX")
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Link: https://patch.msgid.link/20260310184824.1183651-1-alok.a.tiwari@oracle.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The "rx_filter" member of "hwtstamp_config" structure is an enum field and
does not support bitwise OR combination of multiple filter values. It
causes error while linuxptp application tries to match rx filter version.
Fix this by storing the requested filter type in a new port field.
Fixes: 97248adb5a ("net: ti: am65-cpsw: Update hw timestamping filter for PTPv1 RX packets")
Signed-off-by: Chintan Vankar <c-vankar@ti.com>
Link: https://patch.msgid.link/20260310160940.109822-1-c-vankar@ti.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2026-03-10 (ice, iavf, i40e, e1000e, e1000)
Nikolay Aleksandrov changes return code of RDMA related ice devlink get
parameters when irdma is not enabled to -EOPNOTSUPP as current return
of -ENODEV causes issues with devlink output.
Petr Oros resolves a couple of issues in iavf; freeing PTP resources
before reset and disable. Fixing contention issues with the netdev lock
between reset and some ethtool operations.
Alok Tiwari corrects an incorrect comparison of cloud filter values and
adjust some passed arguments to sizeof() for consistency on i40e.
Matt Vollrath removes an incorrect decrement for DMA error on e1000 and
e1000e drivers.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
e1000/e1000e: Fix leak in DMA error cleanup
i40e: fix src IP mask checks and memcpy argument names in cloud filter
iavf: fix incorrect reset handling in callbacks
iavf: fix PTP use-after-free during reset
drivers: net: ice: fix devlink parameters get without irdma
====================
Link: https://patch.msgid.link/20260310205654.4109072-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The bcmgenet EEE implementation is broken in several ways.
phy_support_eee() is never called, so the PHY never advertises EEE
and phylib never sets phydev->enable_tx_lpi. bcmgenet_mac_config()
checks priv->eee.eee_enabled to decide whether to enable the MAC
LPI logic, but that field is never initialised to true, so the MAC
never enters Low Power Idle even when EEE is negotiated - wasting
the power savings EEE is designed to provide. The only way to get
EEE working at all is a manual 'ethtool --set-eee eth0 eee on' after
every link-up, and even then bcmgenet_get_eee() immediately clobbers
the reported state because phy_ethtool_get_eee() overwrites
eee_enabled and tx_lpi_enabled with the uninitialised PHY eee_cfg
values. Finally, bcmgenet_mac_config() is only called on link-up,
so EEE is never disabled in hardware on link-down.
Fix all of this by removing the MAC-side EEE state tracking
(priv->eee) and aligning with the pattern used by other non-phylink
MAC drivers such as FEC.
Call phy_support_eee() in bcmgenet_mii_probe() so the PHY advertises
EEE link modes and phylib tracks negotiation state. Move the EEE
hardware control to bcmgenet_mii_setup(), which is called on every
link event, and drive it directly from phydev->enable_tx_lpi - the
flag phylib sets when EEE is negotiated and the user has not disabled
it. This enables EEE automatically once the link partner agrees and
disables it cleanly on link-down.
Make bcmgenet_get_eee() and bcmgenet_set_eee() pure passthroughs to
phy_ethtool_get_eee() and phy_ethtool_set_eee(), with the MAC
hardware register read/written for tx_lpi_timer. Drop struct
ethtool_keee eee from struct bcmgenet_priv.
Fixes: fe0d4fd928 ("net: phy: Keep track of EEE configuration")
Link: https://lore.kernel.org/netdev/d352039f-4cbb-41e6-9aeb-0b4f3941b54c@lunn.ch/
Suggested-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Nicolai Buchwitz <nb@tipi-net.de>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Tested-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://patch.msgid.link/20260310054935.1238594-1-nb@tipi-net.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Normal RX/TX interrupts are enabled later, in arc_emac_open(), so probe
should not see interrupt delivery in the usual case. However, hardware may
still present stale or latched interrupt status left by firmware or the
bootloader.
If probe later unwinds after devm_request_irq() has installed the handler,
such a stale interrupt can still reach arc_emac_intr() during teardown and
race with release of the associated net_device.
Avoid that window by putting the device into a known quiescent state before
requesting the IRQ: disable all EMAC interrupt sources and clear any
pending EMAC interrupt status bits. This keeps the change hardware-focused
and minimal, while preventing spurious IRQ delivery from leftover state.
Fixes: e4f2379db6 ("ethernet/arc/arc_emac - Add new driver")
Cc: stable@vger.kernel.org
Signed-off-by: Fan Wu <fanwu01@zju.edu.cn>
Link: https://patch.msgid.link/20260309132409.584966-1-fanwu01@zju.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Quanyang observed that when using an NFS rootfs on an AMD ZynqMp board,
the rootfs may take an extended time to recover after a suspend.
Upon investigation, it was determined that the issue originates from a
problem in the macb driver.
According to the Zynq UltraScale TRM [1], when transmit is disabled,
the transmit buffer queue pointer resets to point to the address
specified by the transmit buffer queue base address register.
In the current implementation, the code merely resets `queue->tx_head`
and `queue->tx_tail` to '0'. This approach presents several issues:
- Packets already queued in the tx ring are silently lost,
leading to memory leaks since the associated skbs cannot be released.
- Concurrent write access to `queue->tx_head` and `queue->tx_tail` may
occur from `macb_tx_poll()` or `macb_start_xmit()` when these values
are reset to '0'.
- The transmission may become stuck on a packet that has already been sent
out, with its 'TX_USED' bit set, but has not yet been processed. However,
due to the manipulation of 'queue->tx_head' and 'queue->tx_tail',
`macb_tx_poll()` incorrectly assumes there are no packets to handle
because `queue->tx_head == queue->tx_tail`. This issue is only resolved
when a new packet is placed at this position. This is the root cause of
the prolonged recovery time observed for the NFS root filesystem.
To resolve this issue, shuffle the tx ring and tx skb array so that
the first unsent packet is positioned at the start of the tx ring.
Additionally, ensure that updates to `queue->tx_head` and
`queue->tx_tail` are properly protected with the appropriate lock.
[1] https://docs.amd.com/v/u/en-US/ug1085-zynq-ultrascale-trm
Fixes: bf9cf80cab ("net: macb: Fix tx/rx malfunction after phy link down and up")
Reported-by: Quanyang Wang <quanyang.wang@windriver.com>
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Cc: stable@vger.kernel.org
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260307-zynqmp-v2-1-6ef98a70e1d0@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If an error is encountered while mapping TX buffers, the driver should
unmap any buffers already mapped for that skb.
Because count is incremented after a successful mapping, it will always
match the correct number of unmappings needed when dma_error is reached.
Decrementing count before the while loop in dma_error causes an
off-by-one error. If any mapping was successful before an unsuccessful
mapping, exactly one DMA mapping would leak.
In these commits, a faulty while condition caused an infinite loop in
dma_error:
Commit 03b1320dfc ("e1000e: remove use of skb_dma_map from e1000e
driver")
Commit 602c0554d7 ("e1000: remove use of skb_dma_map from e1000 driver")
Commit c1fa347f20 ("e1000/e1000e/igb/igbvf/ixgb/ixgbe: Fix tests of
unsigned in *_tx_map()") fixed the infinite loop, but introduced the
off-by-one error.
This issue may still exist in the igbvf driver, but I did not address it
in this patch.
Fixes: c1fa347f20 ("e1000/e1000e/igb/igbvf/ixgb/ixgbe: Fix tests of unsigned in *_tx_map()")
Assisted-by: Claude:claude-4.6-opus
Signed-off-by: Matt Vollrath <tactii@gmail.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Fix following issues in the IPv4 and IPv6 cloud filter handling logic in
both the add and delete paths:
- The source-IP mask check incorrectly compares mask.src_ip[0] against
tcf.dst_ip[0]. Update it to compare against tcf.src_ip[0]. This likely
goes unnoticed because the check is in an "else if" path that only
executes when dst_ip is not set, most cloud filter use cases focus on
destination-IP matching, and the buggy condition can accidentally
evaluate true in some cases.
- memcpy() for the IPv4 source address incorrectly uses
ARRAY_SIZE(tcf.dst_ip) instead of ARRAY_SIZE(tcf.src_ip), although
both arrays are the same size.
- The IPv4 memcpy operations used ARRAY_SIZE(tcf.dst_ip) and ARRAY_SIZE
(tcf.src_ip), Update these to use sizeof(cfilter->ip.v4.dst_ip) and
sizeof(cfilter->ip.v4.src_ip) to ensure correct and explicit copy size.
- In the IPv6 delete path, memcmp() uses sizeof(src_ip6) when comparing
dst_ip6 fields. Replace this with sizeof(dst_ip6) to make the intent
explicit, even though both fields are struct in6_addr.
Fixes: e284fc2804 ("i40e: Add and delete cloud filter")
Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Three driver callbacks schedule a reset and wait for its completion:
ndo_change_mtu(), ethtool set_ringparam(), and ethtool set_channels().
Waiting for reset in ndo_change_mtu() and set_ringparam() was added by
commit c2ed2403f1 ("iavf: Wait for reset in callbacks which trigger
it") to fix a race condition where adding an interface to bonding
immediately after MTU or ring parameter change failed because the
interface was still in __RESETTING state. The same commit also added
waiting in iavf_set_priv_flags(), which was later removed by commit
53844673d5 ("iavf: kill "legacy-rx" for good").
Waiting in set_channels() was introduced earlier by commit 4e5e6b5d9d
("iavf: Fix return of set the new channel count") to ensure the PF has
enough time to complete the VF reset when changing channel count, and to
return correct error codes to userspace.
Commit ef490bbb22 ("iavf: Add net_shaper_ops support") added
net_shaper_ops to iavf, which required reset_task to use _locked NAPI
variants (napi_enable_locked, napi_disable_locked) that need the netdev
instance lock.
Later, commit 7e4d784f58 ("net: hold netdev instance lock during
rtnetlink operations") and commit 2bcf4772e4 ("net: ethtool: try to
protect all callback with netdev instance lock") started holding the
netdev instance lock during ndo and ethtool callbacks for drivers with
net_shaper_ops.
Finally, commit 120f28a6f3 ("iavf: get rid of the crit lock")
replaced the driver's crit_lock with netdev_lock in reset_task, causing
incorrect behavior: the callback holds netdev_lock and waits for
reset_task, but reset_task needs the same lock:
Thread 1 (callback) Thread 2 (reset_task)
------------------- ---------------------
netdev_lock() [blocked on workqueue]
ndo_change_mtu() or ethtool op
iavf_schedule_reset()
iavf_wait_for_reset() iavf_reset_task()
waiting... netdev_lock() <- blocked
This does not strictly deadlock because iavf_wait_for_reset() uses
wait_event_interruptible_timeout() with a 5-second timeout. The wait
eventually times out, the callback returns an error to userspace, and
after the lock is released reset_task completes the reset. This leads to
incorrect behavior: userspace sees an error even though the configuration
change silently takes effect after the timeout.
Fix this by extracting the reset logic from iavf_reset_task() into a new
iavf_reset_step() function that expects netdev_lock to be already held.
The three callbacks now call iavf_reset_step() directly instead of
scheduling the work and waiting, performing the reset synchronously in
the caller's context which already holds netdev_lock. This eliminates
both the incorrect error reporting and the need for
iavf_wait_for_reset(), which is removed along with the now-unused
reset_waitqueue.
The workqueue-based iavf_reset_task() becomes a thin wrapper that
acquires netdev_lock and calls iavf_reset_step(), preserving its use
for PF-initiated resets.
The callbacks may block for several seconds while iavf_reset_step()
polls hardware registers, but this is acceptable since netdev_lock is a
per-device mutex and only serializes operations on the same interface.
v3:
- Remove netif_running() guard from iavf_set_channels(). Unlike
set_ringparam where descriptor counts are picked up by iavf_open()
directly, num_req_queues is only consumed during
iavf_reinit_interrupt_scheme() in the reset path. Skipping the reset
on a down device would silently discard the channel count change.
- Remove dead reset_waitqueue code (struct field, init, and all
wake_up calls) since iavf_wait_for_reset() was the only consumer.
Fixes: 120f28a6f3 ("iavf: get rid of the crit lock")
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Petr Oros <poros@redhat.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Commit 7c01dbfc8a ("iavf: periodically cache PHC time") introduced a
worker to cache PHC time, but failed to stop it during reset or disable.
This creates a race condition where `iavf_reset_task()` or
`iavf_disable_vf()` free adapter resources (AQ) while the worker is still
running. If the worker triggers `iavf_queue_ptp_cmd()` during teardown, it
accesses freed memory/locks, leading to a crash.
Fix this by calling `iavf_ptp_release()` before tearing down the adapter.
This ensures `ptp_clock_unregister()` synchronously cancels the worker and
cleans up the chardev before the backing resources are destroyed.
Fixes: 7c01dbfc8a ("iavf: periodically cache PHC time")
Signed-off-by: Petr Oros <poros@redhat.com>
Reviewed-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
If CONFIG_IRDMA isn't enabled but there are ice NICs in the system, the
driver will prevent full devlink dev param show dump because its rdma get
callbacks return ENODEV and stop the dump. For example:
$ devlink dev param show
pci/0000:82:00.0:
name msix_vec_per_pf_max type generic
values:
cmode driverinit value 2
name msix_vec_per_pf_min type generic
values:
cmode driverinit value 2
kernel answers: No such device
Returning EOPNOTSUPP allows the dump to continue so we can see all devices'
devlink parameters.
Fixes: c24a65b6a2 ("iidc/ice/irdma: Update IDC to support multiple consumers")
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Tested-by: Rinitha S <sx.rinitha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
When operating in 10GBASE-KR mode with auto-negotiation disabled and RX
adaptation enabled, CRC errors can occur during the RX adaptation
process. This happens because the driver continues transmitting and
receiving packets while adaptation is in progress.
Fix this by stopping TX/RX immediately when the link goes down and RX
adaptation needs to be re-triggered, and only re-enabling TX/RX after
adaptation completes and the link is confirmed up. Introduce a flag to
track whether TX/RX was disabled for adaptation so it can be restored
correctly.
This prevents packets from being transmitted or received during the RX
adaptation window and avoids CRC errors from corrupted frames.
The flag tracking the data path state is synchronized with hardware
state in xgbe_start() to prevent stale state after device restarts.
This ensures that after a restart cycle (where xgbe_stop disables
TX/RX and xgbe_start re-enables them), the flag correctly reflects
that the data path is active.
Fixes: 4f3b20bfbb ("amd-xgbe: add support for rx-adaptation")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260306111629.1515676-3-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Given that some platforms may use PHY address 0 (I suppose the PHY may
not treat address 0 as a broadcast address or default response address).
It is possible for some boards to connect multiple PHYs to the same
ENETC MAC, for example:
- a PHY with a non-zero address connects to ENETC MAC through SGMII
interface (selected via DTS_A)
- a PHY with address 0 connects to ENETC MAC through RGMII interface
(selected via DTS_B)
For the case where the ENETC port MDIO is used to manage the PHY, when
switching from DTS_A to DTS_B via soft reboot, LaBCR[MDIO_PHYAD_PRTAD]
must be updated to 0 because the NETCMIX block is not reset during soft
reboot. However, the current driver explicitly skips configuring address
0, causing LaBCR[MDIO_PHYAD_PRTAD] to retain its old value.
Therefore, remove the special-case skip of PHY address 0 so that valid
configurations using address 0 are properly supported.
Fixes: 6633df05f3 ("net: enetc: set the external PHY address in IERB for port MDIO usage")
Fixes: 50bfd9c06f ("net: enetc: set external PHY address in IERB for i.MX94 ENETC")
Reviewed-by: Clark Wang <xiaoning.wang@nxp.com>
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20260305031211.904812-3-wei.fang@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The current netc_get_phy_addr() implementation falls back to PHY address
0 when the "mdio" node or the PHY child node is missing. On i.MX95, this
causes failures when a real PHY is actually assigned address 0 and is
managed through the EMDIO interface. Because the bit 0 of phy_mask will
be set, leading imx95_enetc_mdio_phyaddr_config() to return an error, and
the netc_blk_ctrl driver probe subsequently fails. Fix this by returning
-ENODEV when neither an "mdio" node nor any PHY node is present, it means
that ENETC port MDIO is not used to manage the PHY, so there is no need
to configure LaBCR[MDIO_PHYAD_PRTAD].
Reported-by: Alexander Stein <alexander.stein@ew.tq-group.com>
Closes: https://lore.kernel.org/all/7825188.GXAFRqVoOG@steina-w
Fixes: 6633df05f3 ("net: enetc: set the external PHY address in IERB for port MDIO usage")
Reviewed-by: Clark Wang <xiaoning.wang@nxp.com>
Tested-by: Alexander Stein <alexander.stein@ew.tq-group.com>
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20260305031211.904812-2-wei.fang@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Even if we get a dma_mapping_error() while mapping an RX buffer, we
should still update rx_ring->head to ensure that the buffers we were
able to allocate and map are used. Fix this by breaking out to the
existing code after the loop, analogous to the existing handling for skb
allocation failure.
Fixes: bfec6d7f20 ("net: spacemit: Add K1 Ethernet MAC")
Signed-off-by: Vivian Wang <wangruikang@iscas.ac.cn>
Link: https://patch.msgid.link/20260305-k1-ethernet-more-fixes-v2-1-e4e434d65055@iscas.ac.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
XDP multi-buf programs can modify the layout of the XDP buffer when the
program calls bpf_xdp_pull_data() or bpf_xdp_adjust_tail(). The
referenced commit in the fixes tag corrected the assumption in the mlx5
driver that the XDP buffer layout doesn't change during a program
execution. However, this fix introduced another issue: the dropped
fragments still need to be counted on the driver side to avoid page
fragment reference counting issues.
Such issue can be observed with the
test_xdp_native_adjst_tail_shrnk_data selftest when using a payload of
3600 and shrinking by 256 bytes (an upcoming selftest patch): the last
fragment gets released by the XDP code but doesn't get tracked by the
driver. This results in a negative pp_ref_count during page release and
the following splat:
WARNING: include/net/page_pool/helpers.h:297 at mlx5e_page_release_fragmented.isra.0+0x4a/0x50 [mlx5_core], CPU#12: ip/3137
Modules linked in: [...]
CPU: 12 UID: 0 PID: 3137 Comm: ip Not tainted 6.19.0-rc3+ #12 NONE
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
RIP: 0010:mlx5e_page_release_fragmented.isra.0+0x4a/0x50 [mlx5_core]
[...]
Call Trace:
<TASK>
mlx5e_dealloc_rx_wqe+0xcb/0x1a0 [mlx5_core]
mlx5e_free_rx_descs+0x7f/0x110 [mlx5_core]
mlx5e_close_rq+0x50/0x60 [mlx5_core]
mlx5e_close_queues+0x36/0x2c0 [mlx5_core]
mlx5e_close_channel+0x1c/0x50 [mlx5_core]
mlx5e_close_channels+0x45/0x80 [mlx5_core]
mlx5e_safe_switch_params+0x1a5/0x230 [mlx5_core]
mlx5e_change_mtu+0xf3/0x2f0 [mlx5_core]
netif_set_mtu_ext+0xf1/0x230
do_setlink.isra.0+0x219/0x1180
rtnl_newlink+0x79f/0xb60
rtnetlink_rcv_msg+0x213/0x3a0
netlink_rcv_skb+0x48/0xf0
netlink_unicast+0x24a/0x350
netlink_sendmsg+0x1ee/0x410
__sock_sendmsg+0x38/0x60
____sys_sendmsg+0x232/0x280
___sys_sendmsg+0x78/0xb0
__sys_sendmsg+0x5f/0xb0
[...]
do_syscall_64+0x57/0xc50
This patch fixes the issue by doing page frag counting on all the
original XDP buffer fragments for all relevant XDP actions (XDP_TX ,
XDP_REDIRECT and XDP_PASS). This is basically reverting to the original
counting before the commit in the fixes tag.
As frag_page is still pointing to the original tail, the nr_frags
parameter to xdp_update_skb_frags_info() needs to be calculated
in a different way to reflect the new nr_frags.
Fixes: afd5ba577c ("net/mlx5e: RX, Fix generating skb from non-linear xdp_buff for legacy RQ")
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Amery Hung <ameryhung@gmail.com>
Link: https://patch.msgid.link/20260305142634.1813208-6-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
XDP multi-buf programs can modify the layout of the XDP buffer when the
program calls bpf_xdp_pull_data() or bpf_xdp_adjust_tail(). The
referenced commit in the fixes tag corrected the assumption in the mlx5
driver that the XDP buffer layout doesn't change during a program
execution. However, this fix introduced another issue: the dropped
fragments still need to be counted on the driver side to avoid page
fragment reference counting issues.
The issue was discovered by the drivers/net/xdp.py selftest,
more specifically the test_xdp_native_tx_mb:
- The mlx5 driver allocates a page_pool page and initializes it with
a frag counter of 64 (pp_ref_count=64) and the internal frag counter
to 0.
- The test sends one packet with no payload.
- On RX (mlx5e_skb_from_cqe_mpwrq_nonlinear()), mlx5 configures the XDP
buffer with the packet data starting in the first fragment which is the
page mentioned above.
- The XDP program runs and calls bpf_xdp_pull_data() which moves the
header into the linear part of the XDP buffer. As the packet doesn't
contain more data, the program drops the tail fragment since it no
longer contains any payload (pp_ref_count=63).
- mlx5 device skips counting this fragment. Internal frag counter
remains 0.
- mlx5 releases all 64 fragments of the page but page pp_ref_count is
63 => negative reference counting error.
Resulting splat during the test:
WARNING: CPU: 0 PID: 188225 at ./include/net/page_pool/helpers.h:297 mlx5e_page_release_fragmented.isra.0+0xbd/0xe0 [mlx5_core]
Modules linked in: [...]
CPU: 0 UID: 0 PID: 188225 Comm: ip Not tainted 6.18.0-rc7_for_upstream_min_debug_2025_12_08_11_44 #1 NONE
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:mlx5e_page_release_fragmented.isra.0+0xbd/0xe0 [mlx5_core]
[...]
Call Trace:
<TASK>
mlx5e_free_rx_mpwqe+0x20a/0x250 [mlx5_core]
mlx5e_dealloc_rx_mpwqe+0x37/0xb0 [mlx5_core]
mlx5e_free_rx_descs+0x11a/0x170 [mlx5_core]
mlx5e_close_rq+0x78/0xa0 [mlx5_core]
mlx5e_close_queues+0x46/0x2a0 [mlx5_core]
mlx5e_close_channel+0x24/0x90 [mlx5_core]
mlx5e_close_channels+0x5d/0xf0 [mlx5_core]
mlx5e_safe_switch_params+0x2ec/0x380 [mlx5_core]
mlx5e_change_mtu+0x11d/0x490 [mlx5_core]
mlx5e_change_nic_mtu+0x19/0x30 [mlx5_core]
netif_set_mtu_ext+0xfc/0x240
do_setlink.isra.0+0x226/0x1100
rtnl_newlink+0x7a9/0xba0
rtnetlink_rcv_msg+0x220/0x3c0
netlink_rcv_skb+0x4b/0xf0
netlink_unicast+0x255/0x380
netlink_sendmsg+0x1f3/0x420
__sock_sendmsg+0x38/0x60
____sys_sendmsg+0x1e8/0x240
___sys_sendmsg+0x7c/0xb0
[...]
__sys_sendmsg+0x5f/0xb0
do_syscall_64+0x55/0xc70
The problem applies for XDP_PASS as well which is handled in a different
code path in the driver.
This patch fixes the issue by doing page frag counting on all the
original XDP buffer fragments for all relevant XDP actions (XDP_TX ,
XDP_REDIRECT and XDP_PASS). This is basically reverting to the original
counting before the commit in the fixes tag.
As frag_page is still pointing to the original tail, the nr_frags
parameter to xdp_update_skb_frags_info() needs to be calculated
in a different way to reflect the new nr_frags.
Fixes: 87bcef158a ("net/mlx5e: RX, Fix generating skb from non-linear xdp_buff for striding RQ")
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Cc: Amery Hung <ameryhung@gmail.com>
Reviewed-by: Nimrod Oren <noren@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260305142634.1813208-5-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In case of a TX error CQE, a recovery flow is triggered,
mlx5e_reset_txqsq_cc_pc() resets dma_fifo_cc to 0 but not dma_fifo_pc,
desyncing the DMA FIFO producer and consumer.
After recovery, the producer pushes new DMA entries at the old
dma_fifo_pc, while the consumer reads from position 0.
This causes us to unmap stale DMA addresses from before the recovery.
The DMA FIFO is a purely software construct with no HW counterpart.
At the point of reset, all WQEs have been flushed so dma_fifo_cc is
already equal to dma_fifo_pc. There is no need to reset either counter,
similar to how skb_fifo pc/cc are untouched.
Remove the 'dma_fifo_cc = 0' reset.
This fixes the following WARNING:
WARNING: CPU: 0 PID: 0 at drivers/iommu/dma-iommu.c:1240 iommu_dma_unmap_page+0x79/0x90
Modules linked in: mlx5_vdpa vringh vdpa bonding mlx5_ib mlx5_vfio_pci ipip mlx5_fwctl tunnel4 mlx5_core ib_ipoib geneve ip6_gre ip_gre gre nf_tables ip6_tunnel rdma_ucm ib_uverbs ib_umad vfio_pci vfio_pci_core act_mirred act_skbedit act_vlan vhost_net vhost tap ip6table_mangle ip6table_nat ip6table_filter ip6_tables iptable_mangle cls_matchall nfnetlink_cttimeout act_gact cls_flower sch_ingress vhost_iotlb iptable_raw tunnel6 vfio_iommu_type1 vfio openvswitch nsh rpcsec_gss_krb5 auth_rpcgss oid_registry xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink iptable_nat nf_nat xt_addrtype br_netfilter overlay zram zsmalloc rpcrdma ib_iser libiscsi scsi_transport_iscsi rdma_cm iw_cm ib_cm ib_core fuse [last unloaded: nf_tables]
CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted 6.13.0-rc5_for_upstream_min_debug_2024_12_30_21_33 #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:iommu_dma_unmap_page+0x79/0x90
Code: 2b 4d 3b 21 72 26 4d 3b 61 08 73 20 49 89 d8 44 89 f9 5b 4c 89 f2 4c 89 e6 48 89 ef 5d 41 5c 41 5d 41 5e 41 5f e9 c7 ae 9e ff <0f> 0b 5b 5d 41 5c 41 5d 41 5e 41 5f c3 66 2e 0f 1f 84 00 00 00 00
Call Trace:
<IRQ>
? __warn+0x7d/0x110
? iommu_dma_unmap_page+0x79/0x90
? report_bug+0x16d/0x180
? handle_bug+0x4f/0x90
? exc_invalid_op+0x14/0x70
? asm_exc_invalid_op+0x16/0x20
? iommu_dma_unmap_page+0x79/0x90
? iommu_dma_unmap_page+0x2e/0x90
dma_unmap_page_attrs+0x10d/0x1b0
mlx5e_tx_wi_dma_unmap+0xbe/0x120 [mlx5_core]
mlx5e_poll_tx_cq+0x16d/0x690 [mlx5_core]
mlx5e_napi_poll+0x8b/0xac0 [mlx5_core]
__napi_poll+0x24/0x190
net_rx_action+0x32a/0x3b0
? mlx5_eq_comp_int+0x7e/0x270 [mlx5_core]
? notifier_call_chain+0x35/0xa0
handle_softirqs+0xc9/0x270
irq_exit_rcu+0x71/0xd0
common_interrupt+0x7f/0xa0
</IRQ>
<TASK>
asm_common_interrupt+0x22/0x40
Fixes: db75373c91 ("net/mlx5e: Recover Send Queue (SQ) from error state")
Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260305142634.1813208-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The check on mlx5_esw_host_functions_enabled(esw->dev) for adding VF
peer miss rules is incorrect. These rules match traffic from peer's VFs,
so the local device's host function status is irrelevant. Remove this
check to ensure peer VF traffic is properly handled regardless of local
host configuration.
Also fix the PF peer miss rule deletion to be symmetric with the add
path, so only attempt to delete the rule if it was actually created.
Fixes: 520369ef43 ("net/mlx5: Support disabling host PFs")
Signed-off-by: Carolina Jubran <cjubran@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260305142634.1813208-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
esw->work_queue executes esw_functions_changed_event_handler ->
esw_vfs_changed_event_handler and acquires the devlink lock.
.eswitch_mode_set (acquires devlink lock in devlink_nl_pre_doit) ->
mlx5_devlink_eswitch_mode_set -> mlx5_eswitch_disable_locked ->
mlx5_eswitch_event_handler_unregister -> flush_workqueue deadlocks
when esw_vfs_changed_event_handler executes.
Fix that by no longer flushing the work to avoid the deadlock, and using
a generation counter to keep track of work relevance. This avoids an old
handler manipulating an esw that has undergone one or more mode changes:
- the counter is incremented in mlx5_eswitch_event_handler_unregister.
- the counter is read and passed to the ephemeral mlx5_host_work struct.
- the work handler takes the devlink lock and bails out if the current
generation is different than the one it was scheduled to operate on.
- mlx5_eswitch_cleanup does the final draining before destroying the wq.
No longer flushing the workqueue has the side effect of maybe no longer
cancelling pending vport_change_handler work items, but that's ok since
those are disabled elsewhere:
- mlx5_eswitch_disable_locked disables the vport eq notifier.
- mlx5_esw_vport_disable disarms the HW EQ notification and marks
vport->enabled under state_lock to false to prevent pending vport
handler from doing anything.
- mlx5_eswitch_cleanup destroys the workqueue and makes sure all events
are disabled/finished.
Fixes: f1bc646c9a ("net/mlx5: Use devl_ API in mlx5_esw_offloads_devlink_port_register")
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260305081019.1811100-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The only user of frag_size field in XDP RxQ info is
bpf_xdp_frags_increase_tail(). It clearly expects whole buffer size instead
of DMA write size. Different assumptions in idpf driver configuration lead
to negative tailroom.
To make it worse, buffer sizes are not actually uniform in idpf when
splitq is enabled, as there are several buffer queues, so rxq->rx_buf_size
is meaningless in this case.
Use truesize of the first bufq in AF_XDP ZC, as there is only one. Disable
growing tail for regular splitq.
Fixes: ac8a861f63 ("idpf: prepare structures to support XDP")
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://patch.msgid.link/20260305111253.2317394-8-larysa.zaremba@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The only user of frag_size field in XDP RxQ info is
bpf_xdp_frags_increase_tail(). It clearly expects whole buffer size instead
of DMA write size. Different assumptions in i40e driver configuration lead
to negative tailroom.
Set frag_size to the same value as frame_sz in shared pages mode, use new
helper to set frag_size when AF_XDP ZC is active.
Fixes: a045d2f2d0 ("i40e: set xdp_rxq_info::frag_size")
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://patch.msgid.link/20260305111253.2317394-7-larysa.zaremba@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Current way of handling XDP RxQ info in i40e has a problem, where frag_size
is not updated when xsk_buff_pool is detached or when MTU is changed, this
leads to growing tail always failing for multi-buffer packets.
Couple XDP RxQ info registering with buffer allocations and unregistering
with cleaning the ring.
Fixes: a045d2f2d0 ("i40e: set xdp_rxq_info::frag_size")
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://patch.msgid.link/20260305111253.2317394-6-larysa.zaremba@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
