Pull mod_devicetable.h header split from Uwe Kleine-König:
"Split <linux/mod_devicetable.h> in per subsystem headers
<linux/mod_devicetable.h> is included transitively in nearly every
driver in an x86_64 allmodconfig build of v7.1:
$ find drivers -name \*.o -not -name \*.mod.o | wc -l
21330
$ find drivers -name \*.o.cmd -not -name \*.mod.o.cmd | xargs grep -l mod_devicetable.h | wc -l
17038
The result of this mixture of different and unrelated subsystem
details is that even when touching an obscure device id struct most of
the kernel needs to be recompiled. Given that each driver typically
only needs one or two of these structures, splitting into per
subsystem headers and only including what is really needed reduces the
amount of needed recompilation.
This split is implemented in the first commit and then after some
preparatory work in the following commits, the last two replace
includes of <linux/mod_devicetable.h> by the actually needed more
specific headers.
There are still a few instances left, but the ones with high impact
(that is in headers that are used a lot) and the easy ones (.c files)
are handled. These remaining includes will be addressed during the
next merge window"
* tag 'device-id-rework' of git://git.kernel.org/pub/scm/linux/kernel/git/ukleinek/linux:
Replace <linux/mod_devicetable.h> by more specific <linux/device-id/*.h> (c files)
Replace <linux/mod_devicetable.h> by more specific <linux/device-id/*.h> (headers)
parisc: #include <linux/compiler.h> for unlikely() in <asm/ptrace.h>
media: em28xx: Add include for struct usb_device_id
LoongArch: KVM: Add include defining struct cpu_feature
ALSA: hda/core: Add include defining struct hda_device_id
usb: dwc2: Add include defining struct pci_device_id
platform/x86: int3472: Add include defining struct dmi_system_id
platform/x86: x86-android-tablets: Add include defining struct dmi_system_id
i2c: Let i2c-core.h include <linux/i2c.h>
of: Explicitly include <linux/types.h> and <linux/err.h>
platform/x86: msi-ec: Ensure dmi_system_id is defined
usb: serial: Include <linux/usb.h> in <linux/usb/serial.h>
driver core: platform: Include header for struct platform_device_id
driver: core: Include headers for acpi_device_id and of_device_id for struct device_driver
media: ti: vpe: #include <linux/platform_device.h> explicitly
mod_devicetable.h: Split into per subsystem headers
hws_bwc_matcher_move() allocates a replacement matcher before setting it
as the resize target. If mlx5hws_matcher_resize_set_target() fails, the
replacement matcher is not attached anywhere and is leaked.
Fix the leak by destroying the replacement matcher before returning from
the resize-target failure path.
The bug was first flagged by an experimental analysis tool we are
developing for kernel memory-management bugs while analyzing
v6.13-rc1. The tool is still under development and is not yet publicly
available. Manual inspection confirms that the bug is still
present in v7.1.1.
An x86_64 allyesconfig build showed no new warnings. As we do not have a
mlx5 HWS-capable device to test with, no runtime testing was able to be
performed.
Fixes: 2111bb970c ("net/mlx5: HWS, added backward-compatible API handling")
Cc: stable@vger.kernel.org
Signed-off-by: Dawei Feng <dawei.feng@seu.edu.cn>
Reviewed-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260629064049.3852759-1-dawei.feng@seu.edu.cn
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Depending on the value of chip_version, the correct decode set is selected.
However, the subsequent matching with the t4 encoding type in the if-else
block results in a reassignment, which leads to the loss of support for
t6_decode as well as reinitializing of values t4_decode and t5_decode.
The component history shows that the if-else block previously used for
this purpose, as well as the execution order, was not affected by the
change.
Furthermore, it is suggested by the execution order that the scenario with
overwriting and loss of support will be implemented.
Delete the if-else block.
Fixes: 6df397539c ("cxgb4: Update correct encoding of SGE Ingress DMA States for T6 adapter")
Signed-off-by: Gleb Markov <markov.gi@npc-ksb.ru>
Reviewed-by: Potnuri Bharat Teja <bharat@chelsio.com>
Link: https://patch.msgid.link/20260629130856.1168-1-markov.gi@npc-ksb.ru
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When busy-poll is active, napi_schedule_prep() returns false in
virtqueue_napi_schedule(), so virtqueue_disable_cb() is skipped.
The device may keep firing irqs until reaches virtqueue_napi_complete().
Under load (received == budget), it will lead to a large number
of spurious interrupts.
Fix it by disabling the callback at the virtnet_poll() entry.
This keeps the callback off while we poll and it is re-enabled by
virtqueue_napi_complete() when going idle.
Fixes: ceef438d61 ("virtio_net: remove custom busy_poll")
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Longjun Tang <tanglongjun@kylinos.cn>
Link: https://patch.msgid.link/20260629024230.37325-1-lange_tang@163.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
irq_of_parse_and_map() creates irqdomain mappings that should be
balanced with irq_dispose_mapping(). The driver never called
irq_dispose_mapping(), leaking mappings on probe failure and
device removal.
Fix by adding irq_dispose_mapping() in free_gfar_dev() and
expanding its loop from priv->num_grps to MAXGROUPS so the
error path also catches partially-initialized groups. All
irqinfo pointers are pre-initialized to NULL in gfar_of_init(),
making the NULL-guarded walk in free_gfar_dev() safe for every
scenario.
gfar_parse_group() itself is left as a simple parse function
with no resource management; cleanup is centralized in the
caller's error path.
Assisted-by: opencode:big-pickle
Fixes: b31a1d8b41 ("gianfar: Convert gianfar to an of_platform_driver")
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Link: https://patch.msgid.link/20260626225228.427392-1-rosenp@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
lan743x_hardware_init() calls pci11x1x_strap_get_status() during the
PCI11x1x probe sequence. That helper acquires the Ethernet subsystem
hardware lock via lan743x_hs_syslock_acquire(), which relies on
adapter->eth_syslock_spinlock to serialize access.
The spinlock is currently initialized only after the strap status is
read. With CONFIG_DEBUG_SPINLOCK enabled, taking the zeroed initialized
spinlock can trip the spinlock debug check.
Fix by initializing adapter->eth_syslock_spinlock before reading the
strap status so the probe path never attempts to lock an uninitialized
spinlock.
Fixes: 46b777ad9a ("net: lan743x: Add support to SGMII 1G and 2.5G")
Cc: stable@vger.kernel.org # v6.0+
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: David Thompson <davthompson@nvidia.com>
Reviewed-by: Thangaraj Samynathan<Thangaraj.s@microchip.com>
Link: https://patch.msgid.link/20260626163218.3591486-1-arighi@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In wx_set_vmdq_queues(), the VMDQ mask was not set for the devices not
supporting WX_FLAG_MULTI_64_FUNC, i.e., NGBE devices. A mask of 0 causes
__ALIGN_MASK(1, ~vmdq->mask) to return 0, which incorrectly sets
q_per_pool to 0 in wx_write_qde().
Fix the VMDQ 1-queue mask to 0x7F then ensures that __ALIGN_MASK(1,
~0x7F) correctly evaluates to 1.
Fixes: c52d4b8989 ("net: libwx: Redesign flow when sriov is enabled")
Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://patch.msgid.link/161F704D2C983E2C+20260626092530.551028-1-jiawenwu@trustnetic.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Set the GDM maximum receive size to AIROHA_MAX_RX_SIZE unconditionally
during hardware initialization instead of updating it according to the
configured MTU. This avoids dropping incoming frames that exceed the
current MTU but could still be processed by the networking stack, which
is able to fragment the reply on the TX side (e.g. ICMP echo requests).
Move the per-port MTU configuration to the PPE egress path where it
belongs, and set the tx frame size running airoha_ppe_set_xmit_frame_size()
to dynamically track the maximum MTU across running interfaces sharing
the same PPE instance.
Fix the PPE MTU register addressing to pack two port entries per
register word and add WAN_MTU0 configuration for non-LAN GDM devices.
Fixes: 54d989d58d ("net: airoha: Move min/max packet len configuration in airoha_dev_open()")
Tested-by: Madhur Agrawal <madhur.agrawal@airoha.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260625-airoha-fix-rx-max-len-v1-1-45b9b827358d@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This reverts commit 8fe125892f ("net: phy: sfp: probe for RollBall
I2C-to-MDIO bridge in mdio-i2c").
That commit added a RollBall bridge probe at MDIO bus creation time, in
i2c_mii_init_rollball(), to avoid a multi-minute PHY probe retry loop on
modules without a bridge (e.g. RTL8261BE). The probe runs in SFP_S_INIT,
before genuine RollBall modules have finished their firmware/bridge
initialization, so the bridge does not yet answer CMD_READ/CMD_DONE. The
probe times out, mdio_protocol is set to MDIO_I2C_NONE, and PHY detection
is then skipped for genuine RollBall modules that worked before the commit.
This was confirmed on hardware by Maxime Chevallier and Aleksander
Bajkowski: their RollBall modules no longer detect a PHY, and work again
on v7.0 (before the bridge probing was introduced). The Sashiko static
review flagged the same path.
Deferring the probe to PHY discovery time does not fix it either: at that
point a slow module may still be initializing, so the probe still returns
-ENODEV. A proper fix needs per-module init timing (a longer module_t_wait
or a per-module quirk, per SFF-8472 the host must also wait at least 300 ms
after insertion), which requires genuine RollBall hardware to develop and
validate. Revert to restore the previous, working behaviour in the meantime.
The RTL8261BE retry-loop latency that the reverted commit addressed is
handled in our downstream tree, so reverting upstream is safe on our side.
Fixes: 8fe125892f ("net: phy: sfp: probe for RollBall I2C-to-MDIO bridge in mdio-i2c")
Reported-by: Aleksander Bajkowski <olek2@wp.pl>
Suggested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://lore.kernel.org/netdev/20260624084814.20972-1-petr.wozniak@gmail.com/
Signed-off-by: Petr Wozniak <petr.wozniak@gmail.com>
Tested-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/23e3931915c3ed2a14cec95f1490e43d30b225e8.1782581445.git.petr.wozniak@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sfp_i2c_mdiobus_create() allocates the I2C MDIO bus with mdio_i2c_alloc(),
a plain (non-devm) allocation, and registers it. sfp_i2c_mdiobus_destroy()
only unregisters the bus and clears sfp->i2c_mii without calling
mdiobus_free(). As the only reference to the bus is then cleared, the
struct mii_bus is leaked.
This is hit whenever a copper/RollBall SFP module that instantiated an MDIO
bus is removed: sfp_sm_main() takes the global teardown path and calls
sfp_i2c_mdiobus_destroy(). sfp_cleanup(), on driver unbind, frees
sfp->i2c_mii directly, which is why the leak only triggered on module
hot-removal and not on unbind.
Free the bus in sfp_i2c_mdiobus_destroy() to match the allocation done in
sfp_i2c_mdiobus_create().
Fixes: e85b1347ac ("net: sfp: create/destroy I2C mdiobus before PHY probe/after PHY release")
Signed-off-by: Petr Wozniak <petr.wozniak@gmail.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://patch.msgid.link/312bde8176fc429aa89524e3be250137f034ba84.1782581445.git.petr.wozniak@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
genelink_rx_fixup() splits an aggregated RX frame into its individual
packets, using a per-packet length taken from device-supplied data. That
length is only bounded by GL_MAX_PACKET_LEN (1514); it is never compared
against how many bytes were actually received.
A malicious GeneLink (GL620A) device can therefore send a short URB whose
header claims packet_count > 1 and a first packet of up to 1514 bytes.
skb_put_data(gl_skb, packet->packet_data, size);
then copies past the end of the receive buffer and hands the adjacent slab
contents up the network stack, an out-of-bounds read that leaks kernel heap.
No privilege is required: the path runs in the usbnet RX softirq as soon as
the interface is up.
BUG: KASAN: slab-out-of-bounds in genelink_rx_fixup (drivers/net/usb/gl620a.c:112)
Read of size 1514 at addr ffff888011309708 by task ksoftirqd/0/14
Call Trace:
...
__asan_memcpy (mm/kasan/shadow.c:105)
genelink_rx_fixup (include/linux/skbuff.h:2814 drivers/net/usb/gl620a.c:112)
usbnet_bh (drivers/net/usb/usbnet.c:572 drivers/net/usb/usbnet.c:1589)
process_one_work (kernel/workqueue.c:3322)
bh_worker (kernel/workqueue.c:3405)
tasklet_action (kernel/softirq.c:965)
handle_softirqs (kernel/softirq.c:622)
run_ksoftirqd (kernel/softirq.c:1076)
...
skb_pull() already verifies that the requested length fits the buffer and
returns NULL otherwise. Move it ahead of the copy and check its result, so
a packet that overruns the received data is rejected before it is read.
Well-formed frames, whose packets are fully present, are unaffected.
Fixes: 47ee3051c8 ("[PATCH] USB: usbnet (5/9) module for genesys gl620a cables")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Link: https://patch.msgid.link/20260627205353.4000788-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When an offloaded MACsec RX SC is deleted, macsec_del_rxsc_ctx() freed
the per-SC metadata_dst with metadata_dst_free(), which kfree()s the
object unconditionally and ignores the dst reference count. The RX
datapath in mlx5e_macsec_offload_handle_rx_skb() looks up the SC under
rcu_read_lock() via xa_load(), takes a reference with dst_hold() and
attaches the dst to the skb with skb_dst_set(). A reader that already
obtained the rx_sc pointer can race with the delete path and operate on
freed memory.
Fix the owner side by dropping the reference with dst_release() instead
of freeing unconditionally, and convert the RX datapath to
dst_hold_safe() so a reader racing the SC delete cannot attach a dst
whose last reference was just dropped; only attach it when a reference
was actually taken.
mlx5e_macsec_add_rxsc() also published sc_xarray_element via xa_alloc()
before rx_sc->md_dst was allocated and initialised, so a datapath reader
that looked the SC up by fs_id could observe rx_sc with md_dst still
NULL or, on weakly-ordered architectures, a non-NULL md_dst pointer
whose contents were not yet visible. NULL-check the xa_load() result and
md_dst on the datapath, and reorder add_rxsc() so the xa_alloc() publish
happens only after md_dst is fully initialised; the xarray RCU publish
then pairs with the rcu_read_lock()/xa_load() in the datapath.
Note: macsec_del_rxsc_ctx() also kfree()s rx_sc->sc_xarray_element
without an RCU grace period while the same datapath reads it under
rcu_read_lock(); that is a separate pre-existing issue left to a
follow-up patch.
Found by 0sec automated security-research tooling (https://0sec.ai).
Fixes: b7c9400cbc ("net/mlx5e: Implement MACsec Rx data path using MACsec skb_metadata_dst")
Cc: stable@vger.kernel.org
Signed-off-by: Doruk Tan Ozturk <doruk@0sec.ai>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260627223059.29917-1-doruk@0sec.ai
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mux_dl_adb_decode() walks a chain of aggregated datagram tables using
offsets and lengths taken from the modem. first_table_index,
next_table_index, table_length, datagram_index and datagram_length are
all device supplied le values. Only first_table_index was checked, and
only for being non zero. The decoder then formed adth = block +
adth_index and read the table header and the datagram entries with no
bound against the received skb. A modem that reports an index or a
length past the downlink buffer makes the decoder read out of bounds.
The buffer is IPC_MEM_MAX_DL_MUX_LITE_BUF_SIZE and skb->len is at most
that, so skb->len is the real limit, but none of these in band offsets
were checked against it.
The table chain is also followed with no forward progress check. The loop
takes the next table from adth->next_table_index and stops only when that
reaches zero. A modem can stage two tables that point at each other, so
the loop never ends. It runs in softirq and clones the skb on every pass.
Validate every device offset and length against skb->len before use.
The block header must fit. Each table header, on entry and after every
next_table_index, must lie inside the skb. The datagram table must fit.
Each datagram index and length must stay inside the skb. The header
padding must not exceed the datagram length so the receive length does
not wrap. Require each next_table_index to move forward so the chain
cannot cycle.
This was reproduced under KASAN as a slab out of bounds read on a normal
downlink receive once the iosm net device is up.
Fixes: 1f52d7b622 ("net: wwan: iosm: Enable M.2 7360 WWAN card support")
Suggested-by: Loic Poulain <loic.poulain@oss.qualcomm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Maoyi Xie <maoyixie.tju@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Loic Poulain <loic.poulain@oss.qualcomm.com>
Link: https://patch.msgid.link/178236824878.3259367.5389624724479864947@maoyixie.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fbnic_tx_lso() calls skb_cow_head() which may reallocate the skb
including the shared info. We can't use the pointer calculated
before the call.
BUG: KASAN: slab-use-after-free in fbnic_tx_lso.isra.0+0x668/0x8e0
Read of size 4 at addr ff110000262edd98 by task swapper/5/0
Call Trace:
fbnic_tx_lso.isra.0+0x668/0x8e0
fbnic_xmit_frame+0x622/0xba0
dev_hard_start_xmit+0xf4/0x620
Allocated by task 8653:
__alloc_skb+0x11e/0x5f0
alloc_skb_with_frags+0xcc/0x6c0
sock_alloc_send_pskb+0x327/0x3f0
__ip_append_data+0x188b/0x47a0
ip_make_skb+0x24a/0x300
udp_sendmsg+0x14d2/0x21e0
Freed by task 0:
kfree+0x123/0x5a0
pskb_expand_head+0x36c/0xfa0
fbnic_tx_lso.isra.0+0x500/0x8e0
fbnic_xmit_frame+0x622/0xba0
dev_hard_start_xmit+0xf4/0x620
sch_direct_xmit+0x25b/0x1100
The buggy address belongs to the object at ff110000262edc40
which belongs to the cache skbuff_small_head of size 640
The buggy address is located 344 bytes inside of
freed 640-byte region [ff110000262edc40, ff110000262ede
Link: https://netdev.bots.linux.dev/logs/vmksft/fbnic-qemu-dbg/results/705762/15-uso-py/stderr
Fixes: b0b0f52042 ("eth: fbnic: support TCP segmentation offload")
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://patch.msgid.link/20260625160508.3327986-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Map xmit skb fragments using skb_frag_dma_map() instead of
dma_map_single(skb_frag_address()). skb_frag_address() relies on
page_address() to obtain a kernel virtual address, which is not
guaranteed to work for all page types (e.g. highmem pages or
user-pinned pages from MSG_ZEROCOPY).
skb_frag_dma_map() maps the fragment directly via its struct page and
offset through dma_map_page(), avoiding the need for a kernel virtual
address entirely.
Introduce an enum airoha_dma_map_type to track how each queue entry was
mapped (single vs page), so that the matching unmap function is called
on completion and in error paths.
Fixes: 23020f0493 ("net: airoha: Introduce ethernet support for EN7581 SoC")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Harshitha Ramamurthy <hramamurthy@google.com>
Link: https://patch.msgid.link/20260625-airoha-eth-skb_frag_dma_map-v1-1-31d9e460aae6@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The size of xdp_redirect_arr array is ENETC_MAX_SKB_FRAGS. However, the
number of fragments contained in xdp_frame may be greater than or equal
to ENETC_MAX_SKB_FRAGS, which will cause the access to xdp_redirect_arr
to be out of bounds.
Fixes: 9d2b68cc10 ("net: enetc: add support for XDP_REDIRECT")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20260626073244.2168214-1-wei.fang@oss.nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
__pse_control_release() drops psec->ps via devm_regulator_put(), which
only succeeds if the devres entry added by the matching
devm_regulator_get_exclusive() is still present on pcdev->dev at the
time the pse_control's kref hits zero.
That assumption does not hold when the controller is unbound while a
pse_control still has consumers: pcdev->dev's devres list is released
LIFO, so every per-attach regulator-GET devres runs (and
regulator_put()s the underlying regulator) before
pse_controller_unregister() itself is invoked. Any later
pse_control_put() from that unbind path then reads psec->ps as a
dangling pointer inside devm_regulator_put() and WARNs at
drivers/regulator/devres.c:232 (devres_release() fails to find the
already-released match).
The pse_control's consumer handle is logically scoped to the
pse_control's refcount, not to pcdev->dev's devres lifetime. Switch to
the plain regulator_get_exclusive() / regulator_put() pair so the
regulator put in __pse_control_release() no longer depends on the
controller's devres still being present. No change to the
regulator-framework-visible refcount or lifetime of the underlying
regulator: a single get paired with a single put. The existing
devm_regulator_register() for the per-PI rails is unchanged (those ARE
correctly scoped to the controller's lifetime).
This addresses only the regulator handle. The same unbind-while-held
scenario also leaves __pse_control_release() reading psec->pcdev->pi[]
and psec->pcdev->owner after pse_controller_unregister() has freed
pcdev->pi, because the controller does not drain its outstanding
pse_control references on unregister. That wider pse_control vs
pcdev lifetime problem pre-dates this change and is addressed by the
PSE controller notifier series, which drains phydev->psec on
PSE_UNREGISTERED before pcdev->pi is freed.
Link: https://lore.kernel.org/netdev/20260620112440.1734404-1-github@szelinsky.de/
Fixes: d83e13761d ("net: pse-pd: Use regulator framework within PSE framework")
Signed-off-by: Corey Leavitt <corey@leavitt.info>
Acked-by: Kory Maincent <kory.maincent@bootlin.com>
Signed-off-by: Carlo Szelinsky <github@szelinsky.de>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260624204017.2752934-1-github@szelinsky.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Jakub Kicinski:
"Including fixes from netfilter and IPsec.
Current release - regressions:
- do not acquire dev->tx_global_lock in netdev_watchdog_up()
- ethtool: keep rtnl_lock for ops using ethtool_op_get_link()
- fix deadlock in nested UP notifier events
Current release - new code bugs:
- eth:
- cn20k: fix subbank free list indexing for search order
- airoha: fix BQL underflow in shared QDMA TX ring
Previous releases - regressions:
- netfilter:
- flowtable: fix offloaded ct timeout never being extended
- nf_conncount: prevent connlimit drops for early confirmed ct
Previous releases - always broken:
- require CAP_NET_ADMIN in the originating netns when modifying
cross-netns devices
- report NAPI thread PID in the caller's pid namespace
- mac802154: fix dirty frag in in-place crypto for IOT radios
- sctp: hold socket lock when dumping endpoints in sctp_diag, avoid
an overflow
- eth: gve: fix header buffer corruption with header-split and HW-GRO
- af_key: initialize alg_key_len for IPComp states, prevent OOB read"
* tag 'net-7.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (213 commits)
selftests: bonding: add a test for VLAN propagation over a bonded real device
vlan: defer real device state propagation to netdev_work
net: add the driver-facing netdev_work scheduling API
net: turn the rx_mode work into a generic netdev_work facility
net: ethtool: keep rtnl_lock for ops using ethtool_op_get_link()
rxrpc: Fix rxrpc_rotate_tx_rotate() to check there's something to rotate
rxrpc: Fix leak of released call in recvmsg(MSG_PEEK)
rxrpc: Fix socket notification race
rxrpc: Fix potential infinite loop in rxrpc_recvmsg()
rxrpc: Fix oob challenge leak in cleanup after notification failure
rxrpc: Fix the reception of a reply packet before data transmission
afs: Fix uncancelled rxrpc OOB message handler
afs: Fix further netns teardown to cancel the preallocation charger
rxrpc: Fix double unlock in rxrpc_recvmsg()
rxrpc: Fix leak of connection from OOB challenge
rxrpc: Fix ACKALL packet handling
net: hns3: differentiate autoneg default values between copper and fiber
net: hns3: fix permanent link down deadlock after reset
net: hns3: refactor MAC autoneg and speed configuration
net: hns3: unify copper port ksettings configuration path
...
Breno reports following splats on mlx5:
RTNL: assertion failed at net/core/dev.c (2241)
WARNING: net/core/dev.c:2241 at netif_state_change+0xed/0x130, CPU#5: ethtool/1335
RIP: 0010:netif_state_change+0xf9/0x130
Call Trace:
<TASK>
__linkwatch_sync_dev+0xea/0x120
ethtool_op_get_link+0xe/0x20
__ethtool_get_link+0x26/0x40
linkstate_prepare_data+0x51/0x200
ethnl_default_doit+0x213/0x470
genl_family_rcv_msg_doit+0xdd/0x110
Looks like I missed ethtool_op_get_link() trying to sync linkwatch,
which needs rtnl_lock. Not all drivers do this - bnxt doesn't,
it just returns the link state, so add an opt-in bit.
Reported-by: Breno Leitao <leitao@debian.org>
Fixes: 45079e0013 ("net: ethtool: optionally skip rtnl_lock on Netlink path for GET ops")
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Breno Leitao <leitao@debian.org>
Acked-by: Harshitha Ramamurthy <hramamurthy@google.com>
Link: https://patch.msgid.link/20260624190439.2521219-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix a link loss issue during driver initialization on optical ports
connected to forced-mode (non-autoneg) remote switches.
Previously, during driver probe or initialization, hclge_configure()
blindly hardcoded hdev->hw.mac.req_autoneg to AUTONEG_ENABLE for all
media types. While this is necessary for copper (BASE-T) ports to
establish a link, many high-speed optical (fiber) ports in data
centers are connected to switches running in forced mode (fixed speed,
autoneg disabled). Forcing autoneg on these optical ports during
initialization causes a permanent link failure since the remote end
refuses to respond to autoneg pulses.
Fix this by implementing media-type differentiated initialization in
hclge_init_ae_dev(). Copper ports continue to default to
AUTONEG_ENABLE, while optical ports strictly inherit the preset
autoneg status pre-configured by the firmware (hdev->hw.mac.autoneg),
preserving native compatibility with forced-mode network environments.
Fixes: 05eb60e964 ("net: hns3: using user configure after hardware reset")
Signed-off-by: Shuaisong Yang <yangshuaisong@h-partners.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Link: https://patch.msgid.link/20260624141319.271439-5-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix a critical race condition deadlock where the network interface
remains permanently Link Down after a hardware reset under specific
ethtool sequences.
This issue exclusively manifests in firmware-controlled PHY topologies
where the driver relies on the IMP firmware to arbitrate link parameters.
Standard devices driven by the kernel's native PHY_LIB are unaffected.
The deadlock occurs via the following path:
1. User disables autoneg and forces an unmatched speed, forcing link
down: `ethtool -s ethx autoneg off speed 10 duplex full`
2. User re-enables autoneg: `ethtool -s ethx autoneg on`. The netdev
stack passes cmd->base.speed as SPEED_UNKNOWN (0xffffffff).
3. Driver saves req_autoneg=1, but before the interface can link up,
a hardware reset is triggered.
4. During reset recovery, MAC init reads the un-synchronized runtime
state mac.autoneg (which is still 0/OFF), misinterprets it as
forced mode, and pushes the cached SPEED_UNKNOWN into the hardware
registers, causing the MAC firmware state machine to freeze.
Meanwhile, PHY init reads req_autoneg=1 and enables PHY autoneg.
Since the MAC is frozen with 0xffffffff and PHY is running autoneg,
they mismatch permanently.
Fix this by:
1. Intercepting SPEED_UNKNOWN/DUPLEX_UNKNOWN in
hclge_set_phy_link_ksettings() and hclge_cfg_mac_speed_dup_h() to
prevent it from corrupting the driver's cached valid configuration.
2. Save req_autoneg in hclge_set_autoneg().
3. Aligning the state judgment in hclge_set_autoneg_speed_dup() to use
req_autoneg instead of the un-synchronized runtime mac.autoneg,
ensuring both MAC and PHY consistently enter the autoneg branch to
eliminate configuration discrepancies during reset recovery.
Fixes: 05eb60e964 ("net: hns3: using user configure after hardware reset")
Signed-off-by: Shuaisong Yang <yangshuaisong@h-partners.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Link: https://patch.msgid.link/20260624141319.271439-4-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Extract the MAC autoneg and speed/duplex/lane configuration logic out
of hclge_mac_init() and encapsulate it into a new dedicated helper
function hclge_set_autoneg_speed_dup().
In the init path (hclge_init_ae_dev), this helper is now called after
hclge_update_port_info() so that firmware-reported autoneg values are
already populated before applying the link configuration.
Introduce a separate req_lane_num field in struct hclge_mac to isolate
the user-requested lane count from mac.lane_num, which firmware may
overwrite via hclge_get_sfp_info() with stale values from a prior link
lifecycle (e.g., lane_num=4 from 100G). During probe, req_lane_num is
initialized to 0, which instructs firmware to auto-select the correct
lane count for the current speed, rather than reusing the firmware-
reported mac.lane_num that may be inconsistent with the target speed.
This prevents probe failures from mismatched (speed, lane_num) pairs.
In the reset path (hclge_reset_ae_dev), it runs immediately after
hclge_mac_init(), using the previously cached req_* values to restore
the link without re-querying firmware.
Signed-off-by: Shuaisong Yang <yangshuaisong@h-partners.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Link: https://patch.msgid.link/20260624141319.271439-3-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Refactor hns3_set_link_ksettings() and hclge_set_phy_link_ksettings()
to unify the configuration path for copper ports.
Previously, netdevs with a native kernel phy attached bypassed the main
MAC parameter caching logic and returned early via
phy_ethtool_ksettings_set(). This prevented the driver from updating
hdev->hw.mac.req_xxx variables for kernel PHY setups, leaving them
out-of-sync during reset recovery.
Clean this up by routing all copper port configurations through
ops->set_phy_link_ksettings(), and perform driver-level or kernel-level
PHY arbitration inside hclge_set_phy_link_ksettings() via
hnae3_dev_phy_imp_supported(). This ensures that the user's intended link
profiles (req_speed, req_duplex, req_autoneg) are uniformly recorded
across all copper and fiber deployment topologies, laying the groundwork
for stable reset recovery.
For copper ports where neither IMP firmware nor a kernel PHY is available
(e.g. PHY_INEXISTENT), hclge_set_phy_link_ksettings() returns -ENODEV.
In hns3_set_link_ksettings(), this is caught so the configuration falls
through to the existing MAC-level path (check_ksettings_param ->
cfg_mac_speed_dup_h), preserving compatibility with PHY-less copper
deployments.
Signed-off-by: Shuaisong Yang <yangshuaisong@h-partners.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Link: https://patch.msgid.link/20260624141319.271439-2-shaojijie@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Before the commit 7553911210 ("net: mana: Allocate MSI-X vectors
dynamically"), all the MANA IRQs were assigned statically and together
during early driver load.
After this commit, the IRQ allocation for MANA was done in two phases.
HWC IRQ allocated earlier and then, queue IRQs dynamically added at a
later point. By this time, the IRQ weights on vCPUs can become imbalanced
and if IRQ count is greater than the vCPU count the topology aware IRQ
distribution logic in MANA can cause multiple MANA IRQs to land on the
same vCPUs, while other sibling vCPUs have none (case 1).
On SMP enabled, low-vCPU systems, this becomes a bigger problem as the
softIRQ handling overhead of two IRQs on the same vCPUs becomes much more
than their overheads if they were spread across sibling vCPUs.
In such cases when many parallel TCP connections are tested, the
throughput drops significantly.
Fix the affinity assignment logic, in cases where the IRQ count is greater
than the vCPU count and when IRQs are added dynamically, by utilizing all
the vCPUs irrespective of their NUMA/core bindings (case 2).
The results of setting the affinity and hint to NULL were also studied,
and we observed that, with this logic if there are pre-existing IRQs
allocated on the VM (apart from MANA), during MANA IRQs allocation, it
leads to clustering of the MANA queue IRQs again (case 3).
=======================================================
Case 1: without this patch
=======================================================
4 vcpu(2 cores), 5 MANA IRQs (1 HWC + 4 Queue)
TYPE effective vCPU aff
=======================================================
IRQ0: HWC 0
IRQ1: mana_q1 0
IRQ2: mana_q2 2
IRQ3: mana_q3 0
IRQ4: mana_q4 3
%soft on each vCPU(mpstat -P ALL 1) on receiver
vCPU 0 1 2 3
=======================================================
pass 1: 38.85 0.03 24.89 24.65
pass 2: 39.15 0.03 24.57 25.28
pass 3: 40.36 0.03 23.20 23.17
=======================================================
Case 2: with this patch
=======================================================
4 vcpu(2 cores), 5 MANA IRQs (1 HWC + 4 Queue)
TYPE effective vCPU aff
=======================================================
IRQ0: HWC 0
IRQ1: mana_q1 0
IRQ2: mana_q2 1
IRQ3: mana_q3 2
IRQ4: mana_q4 3
%soft on each vCPU(mpstat -P ALL 1) on receiver
vCPU 0 1 2 3
=======================================================
pass 1: 15.42 15.85 14.99 14.51
pass 2: 15.53 15.94 15.81 15.93
pass 3: 16.41 16.35 16.40 16.36
=======================================================
Case 3: with affinity set to NULL
=======================================================
4 vCPU(2 cores), 5 MANA IRQs (1 HWC + 4 Queue)
TYPE effective vCPU aff
=======================================================
IRQ0: HWC 0
IRQ1: mana_q1 2
IRQ2: mana_q2 3
IRQ3: mana_q3 2
IRQ4: mana_q4 3
=======================================================
Throughput Impact(in Gbps, same env)
=======================================================
TCP conn with patch w/o patch aff NULL
20480 15.65 7.73 5.25
10240 15.63 8.93 5.77
8192 15.64 9.69 7.16
6144 15.64 13.16 9.33
4096 15.69 15.75 13.50
2048 15.69 15.83 13.61
1024 15.71 15.28 13.60
Fixes: 7553911210 ("net: mana: Allocate MSI-X vectors dynamically")
Cc: stable@vger.kernel.org
Co-developed-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
Signed-off-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
Signed-off-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Yury Norov <ynorov@nvidia.com>
Link: https://patch.msgid.link/20260624072138.1632849-1-shradhagupta@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sparx5_register_notifier_blocks() registers the switchdev blocking
notifier before allocating the ordered workqueue. If the workqueue
allocation fails, the error path unregisters the switchdev and netdevice
notifiers, but leaves the blocking notifier registered.
Add a separate error label for the workqueue allocation failure path and
unregister the switchdev blocking notifier there.
Fixes: d6fce51419 ("net: sparx5: add switching support")
Cc: stable@vger.kernel.org
Signed-off-by: Haoxiang Li <haoxiang_li2024@163.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260623115714.2192074-1-haoxiang_li2024@163.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
For i.MX94 series, all the standalone ENETCs do not support SR-IOV, so
pf->caps.num_vsi is zero. This leads to a divide-by-zero in
enetc4_default_rings_allocation() when distributing rings among PF and
VFs.
Division by zero is undefined behavior in C. On ARM64, the UDIV/SDIV
instructions silently return zero rather than raising an exception, so
the issue does not cause a visible crash. However, relying on this
behavior is incorrect and poses a cross-platform compatibility risk.
Add an explicit check for num_vsi == 0 and return early after the PF's
rings have been configured.
Fixes: 2d673b0e2f ("net: enetc: add standalone ENETC support for i.MX94")
Signed-off-by: Wei Fang <wei.fang@nxp.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Link: https://patch.msgid.link/20260624072726.1238903-1-wei.fang@oss.nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2026-06-22 (ice, i40e, e1000e)
For ice:
Dawid changes call to release control VSI during reset to prevent
leaking it.
Lukasz fixes flow control error check to check value rather than treat
is as bitmap values.
Paul makes link related errors non-fatal to probe to allow for recovery
in certain NVM update situations.
Marcin moves netif_keep_dst() to only be called once when entering
switchdev mode.
ZhaoJinming adds a cleanup path for ice_dpll_init_info() to prevent
memory leaks on error path.
For i40e:
Mohamed Khalfella corrects argument passed in macro to match the
one provided to the macro.
For e1000e:
Dima resolves power state issues by adjusting value of PLL clock gate
and re-enabling K1; a quirk table is added to keep it off for known bad
systems.
* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
e1000e: Reconfigure PLL clock gate timeout and re-enable K1 on Meteor Lake
i40e: Fix i40e_debug() to use struct i40e_hw argument
ice: dpll: fix memory leak in ice_dpll_init_info error paths
ice: dpll: set pointers to NULL after kfree in ice_dpll_deinit_info
ice: call netif_keep_dst() once when entering switchdev mode
ice: fix ice_init_link() error return preventing probe
ice: fix AQ error code comparison in ice_set_pauseparam()
ice: fix FDIR CTRL VSI resource leak in ice_reset_all_vfs()
====================
Link: https://patch.msgid.link/20260622220059.2471844-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mac->phy_node is acquired via of_parse_phandle() in spl2sw_probe() and
stored in the mac private data, transferring ownership of the
device_node reference to mac. On driver removal, spl2sw_phy_remove()
disconnects the PHY but never drops that reference, so each
probe-then-remove cycle leaks one of_node refcount per port permanently.
Drop the reference after phy_disconnect(). While at it, remove the
redundant inner "if (ndev)" check; comm->ndev[i] was just verified
non-NULL on the line above.
Compile-tested only; no SP7021 hardware available.
Fixes: fd3040b939 ("net: ethernet: Add driver for Sunplus SP7021")
Signed-off-by: Shitalkumar Gandhi <shitalkumar.gandhi@cambiumnetworks.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/f3bdd4c91f3e2269b4e256075f9dc70808b1b8e9.1782195965.git.shitalkumar.gandhi@cambiumnetworks.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
gem_init_one() calls gem_remove_one() when register_netdev() fails.
gem_remove_one() unregisters and frees resources owned by the net_device,
including the DMA block, MMIO mapping, PCI regions, and the net_device
itself. gem_init_one() then falls through to its own cleanup labels and
frees the same resources again.
Keep the register_netdev() error path in gem_init_one(): clear drvdata so
PM/remove paths do not see a half-registered device, remove the NAPI
instance added during probe, and let the existing cleanup labels release
the resources once.
The issue was found by a local static-analysis checker for probe error
paths. The reported path was manually inspected before sending this fix.
Compile-tested with CONFIG_SUNGEM=y. Runtime testing was not performed
because no sungem hardware is available.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Ruoyu Wang <ruoyuw560@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260623025759.3468566-1-ruoyuw560@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Configurations with mlx5 built-in but macsec=m fail to link:
x86_64-linux-ld: drivers/infiniband/hw/mlx5/macsec.o: in function `mlx5r_add_gid_macsec_operations':
macsec.c:(.text+0x77d): undefined reference to `macsec_netdev_is_offloaded'
x86_64-linux-ld: drivers/infiniband/hw/mlx5/macsec.o: in function `mlx5r_del_gid_macsec_operations':
macsec.c:(.text+0xe81): undefined reference to `macsec_netdev_is_offloaded'
Fix the dependency so this configuration cannot happen.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://patch.msgid.link/20260622124229.2444502-1-arnd@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
kalmia_rx_fixup() computes usb_packet_length = skb->len - (2 *
KALMIA_HEADER_LENGTH) as a u16, guarded only by a pre-loop check that
skb->len is at least KALMIA_HEADER_LENGTH, which is 6. A device can
deliver a short bulk-IN frame with skb->len in the 6 to 11 range, or
leave a short trailing remainder on a later loop iteration. Either case
underflows usb_packet_length to about 65530.
That bypasses the usb_packet_length < ether_packet_length truncation path.
The device-supplied ether_packet_length, a le16 up to 65535 read from
header_start[2], then drives a memcmp() and the following skb_trim() and
skb_pull() past the end of the rx buffer. The rx buffer is hard_mtu * 10,
which is 14000 bytes. That is an out of bounds read.
Require both the start and end framing headers to be present before
subtracting them, on every loop iteration.
Fixes: d40261236e ("net/usb: Add Samsung Kalmia driver for Samsung GT-B3730")
Cc: stable@vger.kernel.org
Signed-off-by: Maoyi Xie <maoyixie.tju@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/178211531778.2216480.12637613349790980750@maoyixie.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Even with both paths gated on gs->gro_hint, geneve_gro_complete()
re-derives the inner dispatch type and length from the packet and the
current gs->gro_hint, independently of geneve_gro_receive(). The two can
disagree if gs->gro_hint flips under a concurrent geneve_quiesce()/
geneve_unquiesce() (sk_user_data is NULL across a synchronize_net()), or if
the re-read option bytes differ from the ones receive parsed.
geneve_gro_receive() already records the inner network header position in
NAPI_GRO_CB()->inner_network_offset. Have geneve_gro_complete() compute the
offset it is about to dispatch at, adding ETH_HLEN in the ETH_P_TEB case
where eth_gro_complete() steps over the inner MAC header, and bail out if
it lands past inner_network_offset.
Use a lower bound rather than exact equality: between gh_len and the inner
L3 header, geneve_gro_receive() may also have pulled an inner VLAN tag
(vlan_gro_receive() advances the recorded offset past it), which only moves
inner_network_offset further out. A valid frame therefore always satisfies
inner_nh <= inner_network_offset, while a gh_len inflated by a hint
gro_receive() did not honour dispatches past the validated inner header,
i.e. the out-of-bounds completion. Only the latter is rejected.
Fixes: fd0dd79657 ("geneve: use GRO hint option in the RX path")
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Weiming Shi <bestswngs@gmail.com>
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Link: https://patch.msgid.link/20260618032622.484720-2-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
geneve_gro_receive() reads the GRO hint through geneve_sk_gro_hint_off(),
which honours it only when the socket enabled IFLA_GENEVE_GRO_HINT
(gs->gro_hint). geneve_gro_complete() instead calls the low-level
geneve_opt_gro_hint_off() and acts on the hint unconditionally.
On a tunnel without the hint, receive aggregates the frames as plain
ETH_P_TEB while complete still honours an attacker-supplied hint option: it
inflates gh_len by gro_hint->nested_hdr_len (u8) and redirects the dispatch
type, so the inner gro_complete handler runs at nhoff + gh_len, an offset
receive never pulled nor validated, reading out of bounds of the skb head:
BUG: KASAN: slab-out-of-bounds in ipv6_gro_complete (net/ipv6/ip6_offload.c:196)
Read of size 1 at addr ffff88800fe91980 by task exploit/153
ipv6_gro_complete (net/ipv6/ip6_offload.c:196)
geneve_gro_complete (drivers/net/geneve.c:965)
udp_gro_complete (net/ipv4/udp_offload.c:940)
inet_gro_complete (net/ipv4/af_inet.c:1621)
__gro_flush (net/core/gro.c:306)
Gate the complete path on gs->gro_hint too via geneve_sk_gro_hint_off(), so
both paths agree. Tunnels that enable the hint are unaffected.
Fixes: fd0dd79657 ("geneve: use GRO hint option in the RX path")
Reported-by: Weiming Shi <bestswngs@gmail.com>
Reported-by: Kyle Zeng <kylebot@openai.com>
Signed-off-by: Xiang Mei <xmei5@asu.edu>
Link: https://patch.msgid.link/20260618032622.484720-1-xmei5@asu.edu
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
On Marvell MPIC platforms (Armada 370/XP/38x), mvneta uses a percpu
IRQ disable/enable scheme for NAPI: the ISR (mvneta_percpu_isr) calls
disable_percpu_irq() to mask the MPIC per-CPU interrupt and schedules
NAPI poll, which calls enable_percpu_irq() on completion to unmask.
If suspend occurs while NAPI poll is pending (between
disable_percpu_irq in the ISR and enable_percpu_irq in poll
completion), the interrupt is never re-enabled:
1. mvneta_percpu_isr: disable_percpu_irq() + napi_schedule()
=> MPIC masked, percpu_enabled cpumask bit cleared
2. NAPI poll does not complete before suspend proceeds
(on PREEMPT_RT this is highly likely since softirqs run in
ksoftirqd which gets frozen; on non-RT it can happen when
softirq processing is deferred to ksoftirqd)
3. mvneta_stop_dev => napi_disable(): cancels the pending poll
without executing the completion path
4. suspend_device_irqs => IRQCHIP_MASK_ON_SUSPEND: masks MPIC
(already masked, but records IRQS_SUSPENDED)
5. Resume: mpic_resume checks irq_percpu_is_enabled() => false
(bit was cleared in step 1) => skips unmask
6. mvneta_start_dev only restores device-level INTR_NEW_MASK,
does not touch the MPIC per-CPU mask
Result: MPIC per-CPU interrupt stays masked permanently. The NIC
generates interrupts (INTR_NEW_CAUSE != 0) but the CPU never
receives them, causing complete loss of network connectivity.
Fix by calling on_each_cpu(mvneta_percpu_enable) in the resume path
to unconditionally unmask the MPIC per-CPU interrupt regardless of
pre-suspend state.
Fixes: 12bb03b436 ("net: mvneta: Handle per-cpu interrupts")
Signed-off-by: Yun Zhou <yun.zhou@windriver.com>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://patch.msgid.link/20260622074350.1666290-1-yun.zhou@windriver.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
CGX per-lmac debugfs seq readers obtained struct rvu via
pci_get_drvdata(pci_get_device(..., PCI_DEVID_OCTEONTX2_RVU_AF, ...)),
which leaks a PCI device reference on every read. Store rvu and the CGX
handle in debugfs inode private data when creating stats, mac_filter,
and fwdata files (one context per CGX), and use debugfs aux numbers for
fwdata so lmac_id matches the other CGX debugfs entries.
Fixes: f967488d09 ("octeontx2-af: Add per CGX port level NIX Rx/Tx counters")
Fixes: dbc52debf9 ("octeontx2-af: Debugfs support for DMAC filters")
Fixes: 49f02e6877 ("Octeontx2-af: Debugfs support for firmware data")
Cc: Linu Cherian <lcherian@marvell.com>
Reported-by: Yuho Choi <dbgh9129@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Link: https://patch.msgid.link/20260622034229.2254145-1-rkannoth@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
NIX maximum number of LFs can be set via devlink command
but that can be done before assigning any LFs to a PF/VF.
The condition used to check whether any LFs are assigned is
incorrect. This patch fixes that condition.
Fixes: dd78428786 ("octeontx2-af: Add new devlink param to configure maximum usable NIX block LFs")
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Link: https://patch.msgid.link/1782082853-6941-1-git-send-email-sbhatta@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When multiple netdevs share a QDMA TX ring and one device is stopped,
netdev_tx_reset_subqueue() zeroes that device's BQL counters while its
pending skbs remain in the shared HW TX ring. When NAPI later completes
those skbs via netdev_tx_completed_queue(), the already-zeroed
dql->num_queued counter underflows.
Fix the issue:
- Remove netdev_tx_reset_subqueue() from airoha_dev_stop() so pending
skbs are completed naturally by NAPI with proper BQL accounting.
- Rework airoha_qdma_tx_cleanup() to disable TX DMA, flush BQL
counters, DMA-unmap and free all pending skbs while skb->dev
references are still valid. Use a per-queue flushing flag checked
under q->lock in airoha_dev_xmit() to prevent races between teardown
and transmit. Call airoha_qdma_stop_napi() before
airoha_qdma_tx_cleanup() at the call sites.
- Move DMA engine start into probe. Split DMA teardown so TX DMA is
disabled in airoha_qdma_tx_cleanup() and RX DMA in
airoha_qdma_cleanup().
- Remove qdma->users counter since DMA lifetime is now tied to
probe/cleanup rather than per-netdev open/stop.
Fixes: a9c2ca61fe ("net: airoha: Support multiple net_devices for a single FE GDM port")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260620-airoha-bql-fixes-v3-1-76b95374e63e@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
On RTL8127A connected to a link partner that advertises 10000baseT
speed cannot be changed to anything other than 10000baseT as 10GbE
is always advertised regardless of any setting. Fix this by
clearing MDIO_AN_10GBT_CTRL_ADV10G bit in rtl822x_config_aneg()'s
call to phy_modify_mmd_changed().
Fixes: 83d9623161 ("net: phy: realtek: add RTL8127-internal PHY")
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: Jan Klos <honza.klos@gmail.com>
Link: https://patch.msgid.link/20260620011956.37181-1-honza.klos@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
subbank_srch_order[i] is the physical subbank at search-order slot i,
so each subbank's arr_idx must be i (its slot), not
subbank_srch_order[sb->idx]. The old logic mis-keyed xa_sb_free
and broke allocation traversal order.
Populate arr_idx and xa_sb_free in a single pass over the search
order after subbank structs are initialized.
Fixes: 7ac9d4c407 ("octeontx2-af: npc: cn20k: add subbank search order control")
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260619095100.1864440-1-rkannoth@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit d7709812e1 ("net: mana: hardening: Validate adapter_mtu from
MANA_QUERY_DEV_CONFIG") rejected any adapter_mtu value smaller than
ETH_MIN_MTU + ETH_HLEN, including 0, returning -EPROTO and failing
mana_probe().
Some older PF firmware versions still in the field report
adapter_mtu as 0 in the MANA_QUERY_DEV_CONFIG response. With the
hardening check in place, the MANA VF driver now fails to load on
those hosts, breaking networking entirely for guests.
MANA hardware always supports the standard Ethernet MTU. Treat a
reported adapter_mtu of 0 as "the PF did not advertise a value" and
fall back to ETH_FRAME_LEN, the same value used for the pre-V2
message version path. Only jumbo frames remain unavailable until
the PF reports a valid MTU.
Other small-but-nonzero bogus values are still rejected, preserving
the original protection against the unsigned-subtraction wrap that
would otherwise let ndev->max_mtu underflow to a huge value.
Fixes: d7709812e1 ("net: mana: hardening: Validate adapter_mtu from MANA_QUERY_DEV_CONFIG")
Signed-off-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260619055348.467224-1-ernis@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
