* Re: [PATCH net-next v3 0/9] net: dsa: add DSA support for the LAN9645x switch chip family
From: Jakub Kicinski @ 2026-04-13 22:47 UTC (permalink / raw)
To: Jens Emil Schulz Østergaard
Cc: UNGLinuxDriver, Andrew Lunn, Vladimir Oltean, David S. Miller,
Eric Dumazet, Paolo Abeni, Simon Horman, Rob Herring,
Krzysztof Kozlowski, Conor Dooley, Woojung Huh, Russell King,
Steen Hegelund, Daniel Machon, linux-kernel, netdev, devicetree
In-Reply-To: <20260410-dsa_lan9645x_switch_driver_base-v3-0-aadc8595306d@microchip.com>
On Fri, 10 Apr 2026 13:48:36 +0200 Jens Emil Schulz Østergaard wrote:
> This series provides the Microchip LAN9645X Switch driver.
>
> The LAN9645x is a family of chips with ethernet switch functionality and
> multiple peripheral functions. The switch delivers up to 9 ethernet
> ports and 12 Gbps switching bandwidth.
>
> The switch chip has 5 integrated copper PHYs, support for 2x RGMII
> interfaces, 2x SGMII and one QSGMII interface.
>
> The switch chip is from the same design architecture family as ocelot
> and lan966x, and the driver reflects this similarity. However, LAN9645x
> does not have an internal CPU in any package, and must be driven
> externally. For register IO it supports interfaces such as SPI, I2C and
> MDIO.
We're wrapping up the 7.1 PR and doesn't look like Vladimir (or any
other DSA expert) had a chance to review this yet, so let's defer to
the next cycle.
--
pw-bot: defer
^ permalink raw reply
* Re: [PATCH net v2 0/2] octeon_ep_vf: fix napi_build_skb() NULL dereference
From: patchwork-bot+netdevbpf @ 2026-04-13 22:50 UTC (permalink / raw)
To: David CARLIER
Cc: netdev, vburru, sedara, srasheed, sburla, andrew+netdev, davem,
edumazet, kuba, pabeni, horms, linux-kernel, stable
In-Reply-To: <20260409184009.930359-1-devnexen@gmail.com>
Hello:
This series was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Thu, 9 Apr 2026 19:40:07 +0100 you wrote:
> napi_build_skb() can return NULL on allocation failure. In
> __octep_vf_oq_process_rx(), the result is used directly without a
> NULL check in both the single-buffer and multi-fragment paths,
> leading to a NULL pointer dereference.
>
> Patch 1 introduces a helper to deduplicate the ring index advance
> pattern, patch 2 adds the actual NULL checks.
>
> [...]
Here is the summary with links:
- [net,v2,1/2] octeon_ep_vf: introduce octep_vf_oq_next_idx() helper
https://git.kernel.org/netdev/net/c/4e5bc3ff060e
- [net,v2,2/2] octeon_ep_vf: add NULL check for napi_build_skb()
https://git.kernel.org/netdev/net/c/dd66b4285470
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net-next v4 2/2] net: pse-pd: add LED trigger support via notification path
From: Jakub Kicinski @ 2026-04-13 22:51 UTC (permalink / raw)
To: Carlo Szelinsky
Cc: Oleksij Rempel, Kory Maincent, Andrew Lunn, David S . Miller,
Eric Dumazet, Paolo Abeni, Krzysztof Kozlowski,
Krzysztof Kozlowski, Conor Dooley, Rob Herring, netdev,
linux-kernel, linux-leds, kernel test robot
In-Reply-To: <20260410124428.809943-3-github@szelinsky.de>
On Fri, 10 Apr 2026 14:44:28 +0200 Carlo Szelinsky wrote:
> @@ -266,6 +267,23 @@ struct pse_pi {
> int pw_allocated_mW;
> };
>
> +#if IS_ENABLED(CONFIG_LEDS_TRIGGERS)
> +/**
> + * struct pse_pi_led_triggers - LED trigger state for a PSE PI
> + *
> + * @delivering: LED trigger for power delivering state
> + * @enabled: LED trigger for admin enabled state
> + * @last_delivering: cached delivering state for change detection
> + * @last_enabled: cached enabled state for change detection
> + */
> +struct pse_pi_led_triggers {
> + struct led_trigger delivering;
> + struct led_trigger enabled;
> + bool last_delivering;
> + bool last_enabled;
> +};
> +#endif
> +
> /**
avoid wrapping type definitions in ifdefs.
This only forces other code to also be wrapped in ifdefs.
^ permalink raw reply
* Re: [PATCH net-next v4 1/2] net: pse-pd: add devm_pse_poll_helper()
From: Jakub Kicinski @ 2026-04-13 22:50 UTC (permalink / raw)
To: github
Cc: Jakub Kicinski, o.rempel, kory.maincent, andrew+netdev, davem,
edumazet, pabeni, krzk, krzk+dt, conor+dt, robh, netdev,
linux-kernel, linux-leds, lkp
In-Reply-To: <20260410124428.809943-2-github@szelinsky.de>
This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
---
net: pse-pd: add devm_pse_poll_helper()
Extract the common event handling loop from pse_isr() into a shared
pse_handle_events() function, and add a generic poll-based alternative
to the IRQ path for PSE controllers that lack interrupt support.
> Work cancellation on teardown is handled via devm_add_action_or_reset()
> to ensure the delayed work is cancelled before poll_notifs is freed
> by devres, avoiding a use-after-free when devm_pse_poll_helper() is
> called after devm_pse_controller_register() (devres LIFO ordering).
Is it safe to rely on this devm registration order to prevent a
use-after-free?
Standard Linux driver initialization typically sets up interrupts and
pollers before device registration to prevent race conditions where an
unready device is exposed to userspace.
If a driver follows this standard ordering by calling the helper before
registration, devres will unregister the controller first. This will
tear down resources in pse_controller_unregister() while the poll work
remains active.
If the worker executes during this race window, could it access the
already-freed pcdev->pi array or push to the destroyed ntf_fifo? Would
it be safer to explicitly cancel asynchronous event sources within
pse_controller_unregister() instead?
> --- a/drivers/net/pse-pd/pse_core.c
> +++ b/drivers/net/pse-pd/pse_core.c
[ ... ]
> +static void pse_handle_events(struct pse_controller_dev *pcdev,
> + unsigned long *notifs,
> + unsigned long notifs_mask)
> {
> - struct pse_controller_dev *pcdev;
> - unsigned long notifs_mask = 0;
> - struct pse_irq_desc *desc;
> - struct pse_irq *h = data;
> - int ret, i;
> -
> - desc = &h->desc;
> - pcdev = h->pcdev;
> -
> - /* Clear notifs mask */
> - memset(h->notifs, 0, pcdev->nr_lines * sizeof(*h->notifs));
> - mutex_lock(&pcdev->lock);
> - ret = desc->map_event(irq, pcdev, h->notifs, ¬ifs_mask);
> - if (ret || !notifs_mask) {
> - mutex_unlock(&pcdev->lock);
> - return IRQ_NONE;
> - }
> + int i;
>
> for_each_set_bit(i, ¬ifs_mask, pcdev->nr_lines) {
Can this lead to an out-of-bounds read on the stack if a controller
supports more than BITS_PER_LONG ports?
The notifs_mask variable is passed by value to pse_handle_events(), but
for_each_set_bit() takes its address. If pcdev->nr_lines is greater than
BITS_PER_LONG (e.g., a 48-port switch on a 32-bit architecture), the
macro will read past the function argument on the stack into uninitialized
memory.
[ ... ]
> +static irqreturn_t pse_isr(int irq, void *data)
> +{
> + struct pse_controller_dev *pcdev;
> + unsigned long notifs_mask = 0;
> + struct pse_irq *h = data;
> + int ret;
> +
> + pcdev = h->pcdev;
> +
> + /* Clear notifs mask */
> + memset(h->notifs, 0, pcdev->nr_lines * sizeof(*h->notifs));
> + mutex_lock(&pcdev->lock);
> + ret = h->desc.map_event(irq, pcdev, h->notifs, ¬ifs_mask);
Could this also result in a stack out-of-bounds write?
If map_event() uses set_bit() to report an event on a port index greater
than or equal to BITS_PER_LONG, it will write past the 4-byte boundary of
the scalar notifs_mask variable, corrupting adjacent stack memory.
[ ... ]
> +static void pse_poll_worker(struct work_struct *work)
> +{
> + struct pse_controller_dev *pcdev =
> + container_of(work, struct pse_controller_dev,
> + poll_work.work);
> + unsigned long notifs_mask = 0;
> + int ret;
> +
> + memset(pcdev->poll_notifs, 0,
> + pcdev->nr_lines * sizeof(*pcdev->poll_notifs));
> + mutex_lock(&pcdev->lock);
> + ret = pcdev->poll_desc.map_event(0, pcdev, pcdev->poll_notifs,
> + ¬ifs_mask);
Does this share the same out-of-bounds write risk for notifs_mask as the
IRQ handler above?
--
pw-bot: cr
^ permalink raw reply
* Re: [PATCH net-next v4 2/2] net: pse-pd: add LED trigger support via notification path
From: Jakub Kicinski @ 2026-04-13 22:53 UTC (permalink / raw)
To: github
Cc: Jakub Kicinski, o.rempel, kory.maincent, andrew+netdev, davem,
edumazet, pabeni, krzk, krzk+dt, conor+dt, robh, netdev,
linux-kernel, linux-leds, lkp
In-Reply-To: <20260410124428.809943-3-github@szelinsky.de>
This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
---
Date: Fri Apr 10 12:44:28 2026 +0000
net: pse-pd: add LED trigger support via notification path
Add per-PI "delivering" and "enabled" LED triggers to the PSE core
subsystem. LED state is updated from the shared pse_handle_events()
function whenever the IRQ or poll path detects a state change, as well
as from the regulator enable/disable paths so that host-initiated
admin state changes via ethtool are immediately reflected.
Link: https://lore.kernel.org/oe-kbuild-all/202603251254.o5PqMBRU-lkp@intel.com/
Link: https://lore.kernel.org/oe-kbuild-all/202603251250.cuMCk5Yv-lkp@intel.com/
> diff --git a/drivers/net/pse-pd/pse_core.c b/drivers/net/pse-pd/pse_core.c
> index f641a6fa087fb..dfc84340afb9d 100644
> --- a/drivers/net/pse-pd/pse_core.c
> +++ b/drivers/net/pse-pd/pse_core.c
[ ... ]
> +static int pse_led_triggers_register(struct pse_controller_dev *pcdev)
> +{
> + struct device *dev = pcdev->dev;
> + const char *dev_id;
> + int i, ret;
> +
> + dev_id = dev_name(dev);
> +
> + pcdev->pi_led_trigs = devm_kcalloc(dev, pcdev->nr_lines,
> + sizeof(*pcdev->pi_led_trigs),
> + GFP_KERNEL);
> + if (!pcdev->pi_led_trigs)
> + return -ENOMEM;
> +
> + for (i = 0; i < pcdev->nr_lines; i++) {
> + struct pse_pi_led_triggers *trigs = &pcdev->pi_led_trigs[i];
> +
> + /* Skip PIs not described in device tree */
> + if (!pcdev->no_of_pse_pi && !pcdev->pi[i].np)
> + continue;
> +
> + trigs->delivering.name = devm_kasprintf(dev, GFP_KERNEL,
> + "pse-%s:port%d:delivering",
> + dev_id, i);
> + if (!trigs->delivering.name)
> + return -ENOMEM;
> +
> + ret = devm_led_trigger_register(dev, &trigs->delivering);
The commit message states that the initial LED state is queried at
registration time so already-active ports are reflected immediately.
Since no LED is bound to these triggers at registration time, the initial
led_trigger_event() call will be dropped. If an LED is bound later, will
it remain in the default (off) state because there is no .activate callback
configured on the trigger to sync the current state?
Furthermore, since the state is cached in last_delivering and last_enabled,
subsequent updates might not fire new events until the hardware state
actually toggles.
> + if (ret)
> + return ret;
> +
> + trigs->enabled.name = devm_kasprintf(dev, GFP_KERNEL,
> + "pse-%s:port%d:enabled",
> + dev_id, i);
> + if (!trigs->enabled.name)
> + return -ENOMEM;
> +
> + ret = devm_led_trigger_register(dev, &trigs->enabled);
> + if (ret)
> + return ret;
> + }
> +
> + return 0;
> +}
[ ... ]
> @@ -1108,6 +1217,20 @@ int pse_controller_register(struct pse_controller_dev *pcdev)
> if (ret)
> return ret;
>
> + ret = pse_led_triggers_register(pcdev);
> + if (ret) {
> + dev_warn(pcdev->dev, "Failed to register LED triggers: %d\n",
> + ret);
> + }
If devm_led_trigger_register() fails inside pse_led_triggers_register(),
the function returns an error but leaves trigs->delivering.name allocated
and non-NULL.
Because we only log a warning here and continue execution, the cleanup
for pcdev->pi_led_trigs does not occur.
Later, pse_led_update() verifies validity using this check:
if (!trigs->delivering.name)
return;
Since the name is non-NULL, it proceeds to call led_trigger_event() on the
unregistered trigger. Could this result in a NULL pointer dereference,
as the led_cdevs list head within the zero-initialized led_trigger
structure is uninitialized?
> +
> + /* Query initial LED state for all PIs so already-active ports
> + * are reflected immediately without waiting for a hardware event.
> + */
> + for (i = 0; i < pcdev->nr_lines; i++) {
> + if (pcdev->no_of_pse_pi || pcdev->pi[i].np)
> + pse_led_update(pcdev, i);
> + }
The docstring for pse_led_update() specifies that it "Must be called with
pcdev->lock held."
Since this loop calls pse_led_update(pcdev, i) without acquiring
pcdev->lock, and the PI regulators have already been registered and exposed
to consumers earlier in this function, can this race with a concurrent
regulator_enable()?
A concurrent regulator_enable() can trigger pse_pi_enable(), which acquires
the lock and also calls pse_led_update(). This could result in unprotected
calls to the pi_get_pw_status and pi_get_admin_state hardware callbacks,
and concurrent writes to the last_delivering and last_enabled variables.
> +
> mutex_lock(&pse_list_mutex);
> list_add(&pcdev->list, &pse_controller_list);
> mutex_unlock(&pse_list_mutex);
^ permalink raw reply
* Re: [PATCH net-next v3 00/12] BIG TCP for UDP tunnels
From: Jakub Kicinski @ 2026-04-13 22:55 UTC (permalink / raw)
To: Alice Mikityanska
Cc: Daniel Borkmann, David S. Miller, Eric Dumazet, Paolo Abeni,
Xin Long, Willem de Bruijn, David Ahern, Nikolay Aleksandrov,
Shuah Khan, Stanislav Fomichev, Andrew Lunn, Simon Horman,
Florian Westphal, netdev, Alice Mikityanska
In-Reply-To: <20260410150943.993350-1-alice.kernel@fastmail.im>
On Fri, 10 Apr 2026 18:09:31 +0300 Alice Mikityanska wrote:
> This series is a follow-up to "BIG TCP without HBH in IPv6", and it adds
> support for BIG TCP IPv4/IPv6 workloads in vxlan and geneve. Now that
> IPv6 BIG TCP doesn't require stripping the HBH in all various
> combinations in tunneled traffic, adding BIG TCP becomes feasible.
No longer applies, sorry :(
We'll have to revisit after the merge window.
--
pw-bot: cr
^ permalink raw reply
* Re: [PATCH net] net: airoha: Add missing bits in airoha_qdma_cleanup_tx_queue()
From: Jakub Kicinski @ 2026-04-13 22:57 UTC (permalink / raw)
To: lorenzo
Cc: Jakub Kicinski, andrew+netdev, davem, edumazet, pabeni,
linux-arm-kernel, linux-mediatek, netdev
In-Reply-To: <20260410-airoha_qdma_cleanup_tx_queue-fix-net-v1-1-b7171c8f1e78@kernel.org>
This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
---
net: airoha: Add missing bits in airoha_qdma_cleanup_tx_queue()
This commit resets DMA TX descriptors in airoha_qdma_cleanup_tx_queue
and synchronizes TX_DMA_IDX to TX_CPU_IDX to notify the hardware that
the ring is empty.
> diff --git a/drivers/net/ethernet/airoha/airoha_eth.c b/drivers/net/ethernet/airoha/airoha_eth.c
> index 9285a68f435fe..963ab7b8d166b 100644
> --- a/drivers/net/ethernet/airoha/airoha_eth.c
> +++ b/drivers/net/ethernet/airoha/airoha_eth.c
> @@ -1044,13 +1044,17 @@ static int airoha_qdma_init_tx(struct airoha_qdma *qdma)
[ ... ]
> spin_lock_bh(&q->lock);
> for (i = 0; i < q->ndesc; i++) {
> - struct airoha_queue_entry *e = &q->entry[i];
> + struct airoha_qdma_desc *desc = &q->desc[i];
>
> + e = &q->entry[i];
Can this result in a NULL pointer dereference?
If airoha_qdma_init_tx_queue() sets q->ndesc before allocating q->entry and
q->desc, and devm_kzalloc() fails, the error path jumps to
airoha_hw_cleanup().
Because q->ndesc > 0, this cleanup loop would execute and dereference
q->entry[i] and q->desc[i], which may not have been allocated.
[ ... ]
> + /* Reset DMA descriptor */
> + WRITE_ONCE(desc->ctrl, 0);
> + WRITE_ONCE(desc->addr, 0);
> + WRITE_ONCE(desc->data, 0);
> + WRITE_ONCE(desc->msg0, 0);
> + WRITE_ONCE(desc->msg1, 0);
> + WRITE_ONCE(desc->msg2, 0);
Is it safe to modify these DMA descriptors and reset the ring indices below
without waiting for the hardware DMA engine to become idle?
In airoha_dev_stop(), the driver clears GLOBAL_CFG_TX_DMA_EN_MASK to request
the hardware DMA engine to stop, but it does not appear to poll
GLOBAL_CFG_TX_DMA_BUSY_MASK to ensure it has actually stopped.
Modifying the DMA descriptors and internal ring indices while the hardware
is still actively processing in-flight descriptors could potentially lead
to corrupted DMA fetches or IOMMU faults.
> q->queued--;
> }
> +
> + e = list_first_entry(&q->tx_list, struct airoha_queue_entry,
> + list);
> + index = e - q->entry;
> + /* Set TX_DMA_IDX to TX_CPU_IDX to notify the hw the QDMA TX ring is
> + * empty.
> + */
> + airoha_qdma_rmw(qdma, REG_TX_CPU_IDX(qid), TX_RING_CPU_IDX_MASK,
> + FIELD_PREP(TX_RING_CPU_IDX_MASK, index));
> + airoha_qdma_rmw(qdma, REG_TX_DMA_IDX(qid), TX_RING_DMA_IDX_MASK,
> + FIELD_PREP(TX_RING_DMA_IDX_MASK, index));
Could list_first_entry() be called on an empty list here?
If q->tx_list is empty (for example, if allocation failed in
airoha_qdma_init_tx_queue or if no entries had a valid dma_addr), calling
list_first_entry() returns an invalid pointer.
The subsequent pointer arithmetic would produce a garbage index value,
which is then written to the REG_TX_CPU_IDX and REG_TX_DMA_IDX registers.
--
pw-bot: cr
^ permalink raw reply
* Re: [PATCH v2] udp: Force compute_score to always inline
From: patchwork-bot+netdevbpf @ 2026-04-13 23:00 UTC (permalink / raw)
To: Gabriel Krisman Bertazi
Cc: willemdebruijn.kernel, davem, dsahern, edumazet, kuba, pabeni,
kuniyu, horms, netdev, willemb
In-Reply-To: <20260410155936.654915-1-krisman@suse.de>
Hello:
This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Fri, 10 Apr 2026 11:59:36 -0400 you wrote:
> Back in 2024 I reported a 7-12% regression on an iperf3 UDP loopback
> thoughput test that we traced to the extra overhead of calling
> compute_score on two places, introduced by commit f0ea27e7bfe1 ("udp:
> re-score reuseport groups when connected sockets are present"). At the
> time, I pointed out the overhead was caused by the multiple calls,
> associated with cpu-specific mitigations, and merged commit
> 50aee97d1511 ("udp: Avoid call to compute_score on multiple sites") to
> jump back explicitly, to force the rescore call in a single place.
>
> [...]
Here is the summary with links:
- [v2] udp: Force compute_score to always inline
https://git.kernel.org/netdev/net-next/c/b80a95ccf160
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net] net/sched: taprio: fix use-after-free in advance_sched() on schedule switch
From: Jakub Kicinski @ 2026-04-13 23:01 UTC (permalink / raw)
To: vinicius.gomes
Cc: Jakub Kicinski, vladimir.oltean, jhs, jiri, davem, edumazet,
pabeni, horms, netdev, linux-kernel, qjx1298677004
In-Reply-To: <20260410-taprio-user-after-free-fix-net-v1-1-0931b1579988@intel.com>
This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
---
net/sched: taprio: fix use-after-free in advance_sched() on schedule switch
This commit updates advance_sched() to select the next entry from the new
oper schedule immediately after switch_schedules() returns. It relies on the
pre-calculated end_time of the new schedule's first entry instead of
overwriting it with the admin schedule's base time.
> diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
> index f721c03514f60..0316f2dee06ac 100644
> --- a/net/sched/sch_taprio.c
> +++ b/net/sched/sch_taprio.c
[ ... ]
> @@ -972,11 +972,12 @@ static enum hrtimer_restart advance_sched(struct hrtimer *timer)
> }
>
> if (should_change_schedules(admin, oper, end_time)) {
> - /* Set things so the next time this runs, the new
> - * schedule runs.
> - */
> - end_time = sched_base_time(admin);
Does removing this assignment introduce a regression where the new schedule
activates prematurely?
Since the new schedule's first entry has its end_time pre-calculated
as base_time + interval, if we activate it early (before its actual
base_time), does it stretch the entry backwards in time?
Could this cause its gate mask to apply during the time gap, violating the
strictly configured gate durations?
> switch_schedules(q, &admin, &oper);
Can this sequence cause a use-after-free regression due to the RCU grace
period starting too early?
Inside switch_schedules(), call_rcu() is invoked on the old oper
schedule. However, q->current_entry is not updated until later in
advance_sched() at the first_run label:
first_run:
rcu_assign_pointer(q->current_entry, next);
If a concurrent reader calls rcu_dereference(q->current_entry) between
switch_schedules() and this assignment, can it acquire a pointer to the
old schedule after call_rcu() has already started? This would mean the
grace period will not wait for this new reader.
As a separate pre-existing issue, can switch_schedules() be called
concurrently by lockless readers?
If advance_sched() calls it under q->current_entry_lock, but a concurrent
reader like get_packet_txtime() evaluates the switch condition and calls
switch_schedules() without holding locks, can they race?
Could this lead to call_rcu() being called twice on the same old oper
object, potentially corrupting the RCU callback list?
> + /* After changing schedules, the next entry is the first one
> + * in the new schedule, with a pre-calculated end_time.
> + */
> + next = list_first_entry(&oper->entries, struct sched_entry, list);
> + end_time = next->end_time;
> }
>
> next->end_time = end_time;
^ permalink raw reply
* Re: [PATCH net] bnge: return after auxiliary_device_uninit() in error path
From: patchwork-bot+netdevbpf @ 2026-04-13 23:10 UTC (permalink / raw)
To: Greg KH
Cc: netdev, linux-kernel, vikas.gupta, andrew+netdev, davem, edumazet,
kuba, pabeni, stable
In-Reply-To: <2026041124-banshee-molecular-0f70@gregkh>
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sat, 11 Apr 2026 12:45:25 +0200 you wrote:
> When auxiliary_device_add() fails, the error block calls
> auxiliary_device_uninit() but does not return. The uninit drops the
> last reference and synchronously runs bnge_aux_dev_release(), which sets
> bd->auxr_dev = NULL and frees the underlying object. The subsequent
> bd->auxr_dev->net = bd->netdev then dereferences NULL, which is not a
> good thing to have happen when trying to clean up from an error.
>
> [...]
Here is the summary with links:
- [net] bnge: return after auxiliary_device_uninit() in error path
https://git.kernel.org/netdev/net/c/8b0c25528cb6
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH v1 net] tcp: Don't set treq->req_usec_ts in cookie_tcp_reqsk_init().
From: patchwork-bot+netdevbpf @ 2026-04-13 23:10 UTC (permalink / raw)
To: Kuniyuki Iwashima
Cc: edumazet, ncardwell, davem, kuba, pabeni, horms, kuni1840, netdev
In-Reply-To: <20260410235328.1773449-1-kuniyu@google.com>
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Fri, 10 Apr 2026 23:53:27 +0000 you wrote:
> Commit de5626b95e13 ("tcp: Factorise cookie-independent fields
> initialisation in cookie_v[46]_check().") miscategorised
> tcp_rsk(req)->req_usec_ts init to cookie_tcp_reqsk_init(),
> which is used by both BPF/non-BPF SYN cookie reqsk.
>
> Rather, it should have been moved to cookie_tcp_reqsk_alloc() by
> commit 8e7bab6b9652 ("tcp: Factorise cookie-dependent fields
> initialisation in cookie_v[46]_check()") so that only non-BPF SYN
> cookie sets tcp_rsk(req)->req_usec_ts to false.
>
> [...]
Here is the summary with links:
- [v1,net] tcp: Don't set treq->req_usec_ts in cookie_tcp_reqsk_init().
https://git.kernel.org/netdev/net/c/c058bbf05b11
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH iwl-net] ice: fix infinite recursion in ice_cfg_tx_topo via ice_init_dev_hw
From: Jacob Keller @ 2026-04-13 23:43 UTC (permalink / raw)
To: Petr Oros, netdev
Cc: Tony Nguyen, Przemek Kitszel, Andrew Lunn, David S. Miller,
Eric Dumazet, Jakub Kicinski, Paolo Abeni, Aleksandr Loktionov,
Nikolay Aleksandrov, Daniel Zahka, Paul Greenwalt, Dave Ertman,
Michal Swiatkowski, intel-wired-lan, linux-kernel
In-Reply-To: <20260413191420.3524013-1-poros@redhat.com>
On 4/13/2026 12:14 PM, Petr Oros wrote:
> On certain E810 configurations where firmware supports Tx scheduler
> topology switching (tx_sched_topo_comp_mode_en), ice_cfg_tx_topo()
> may need to apply a new 5-layer or 9-layer topology from the DDP
> package. If the AQ command to set the topology fails (e.g. due to
> invalid DDP data or firmware limitations), the global configuration
> lock must still be cleared via a CORER reset.
>
> Commit 86aae43f21cf ("ice: don't leave device non-functional if Tx
> scheduler config fails") correctly fixed this by refactoring
> ice_cfg_tx_topo() to always trigger CORER after acquiring the global
> lock and re-initialize hardware via ice_init_hw() afterwards.
>
> However, commit 8a37f9e2ff40 ("ice: move ice_deinit_dev() to the end
> of deinit paths") later moved ice_init_dev_hw() into ice_init_hw(),
> breaking the reinit path introduced by 86aae43f21cf. This creates an
> infinite recursive call chain:
>
> ice_init_hw()
> ice_init_dev_hw()
> ice_cfg_tx_topo() # topology change needed
> ice_deinit_hw()
> ice_init_hw() # reinit after CORER
> ice_init_dev_hw() # recurse
> ice_cfg_tx_topo()
> ... # stack overflow
>
Oof, ya thats not good. I guess this only happens if the topology needs
to change, so it wouldn't affect many systems where we had already
changed the topology before hand on the old driver.
> Fix by moving ice_init_dev_hw() back out of ice_init_hw() and calling
> it explicitly from ice_probe() and ice_devlink_reinit_up(). The third
> caller, ice_cfg_tx_topo(), intentionally does not need ice_init_dev_hw()
> during its reinit, it only needs the core HW reinitialization. This
> breaks the recursion cleanly without adding flags or guards.
>
> The deinit ordering changes from commit 8a37f9e2ff40 ("ice: move
> ice_deinit_dev() to the end of deinit paths") which fixed slow rmmod
> are preserved, only the init-side placement of ice_init_dev_hw() is
> reverted.
>
> Fixes: 8a37f9e2ff40 ("ice: move ice_deinit_dev() to the end of deinit paths")
> Signed-off-by: Petr Oros <poros@redhat.com>
The fix looks correct to me, and definitely the most elegant.
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
^ permalink raw reply
* Re: [PATCH] net: phy: realtek: use LEDCR page number define on RTL8211F
From: patchwork-bot+netdevbpf @ 2026-04-13 23:50 UTC (permalink / raw)
To: Aleksander Jan Bajkowski
Cc: andrew, hkallweit1, linux, davem, edumazet, kuba, pabeni, daniel,
vladimir.oltean, michael, daniel.braunwarth, ih, rmk+kernel,
netdev, linux-kernel
In-Reply-To: <20260411105150.184577-1-olek2@wp.pl>
Hello:
This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sat, 11 Apr 2026 12:51:45 +0200 you wrote:
> Replace the magic number with an existing define for the LEDCR
> register page number on the RTL8211F.
>
> Signed-off-by: Aleksander Jan Bajkowski <olek2@wp.pl>
> ---
> drivers/net/phy/realtek/realtek_main.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
Here is the summary with links:
- net: phy: realtek: use LEDCR page number define on RTL8211F
https://git.kernel.org/netdev/net-next/c/43a2deae3661
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net-next v2 0/2] net: dsa: mxl862xx: add statistics support
From: patchwork-bot+netdevbpf @ 2026-04-13 23:50 UTC (permalink / raw)
To: Daniel Golle
Cc: andrew, olteanv, davem, edumazet, kuba, pabeni, linux, netdev,
linux-kernel, frankwu, chad, cezary.wilmanski, lxu, yweng, jverdu,
ajayaraman, john
In-Reply-To: <cover.1775951347.git.daniel@makrotopia.org>
Hello:
This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sun, 12 Apr 2026 01:01:48 +0100 you wrote:
> Add per-port RMON statistics support for the MxL862xx DSA driver,
> covering hardware-specific ethtool -S counters, standard IEEE 802.3
> MAC/ctrl/pause statistics, and rtnl_link_stats64 via polled 64-bit
> accumulation.
>
> Changes since v1:
> * trim mxl862xx_mib[] to counters not covered elsewhere only
> * remove histogram counters (moved to .get_rmon_stats)
> * remove RMON error counters (moved to .get_rmon_stats)
> * remove counters already in .get_eth_mac_stats
> * remove counters already in .get_stats64
> * add mxl862xx_rmon_ranges[] and mxl862xx_get_rmon_stats()
>
> [...]
Here is the summary with links:
- [net-next,v2,1/2] net: dsa: mxl862xx: add ethtool statistics support
https://git.kernel.org/netdev/net-next/c/e6295d124644
- [net-next,v2,2/2] net: dsa: mxl862xx: implement .get_stats64
https://git.kernel.org/netdev/net-next/c/a21d33a5265f
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net-next v18 00/15] Begin upstreaming Homa transport protocol
From: John Ousterhout @ 2026-04-13 23:54 UTC (permalink / raw)
To: Jakub Kicinski; +Cc: netdev, pabeni, edumazet, horms
In-Reply-To: <20260412134531.21341692@kernel.org>
On Sun, Apr 12, 2026 at 1:45 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Fri, 10 Apr 2026 13:02:54 -0700 John Ousterhout wrote:
> > This patch series begins the process of upstreaming the Homa transport
> > protocol. Homa is an alternative to TCP for use in datacenter
> > environments. It provides 10-100x reductions in tail latency for short
> > messages relative to TCP. Its benefits are greatest for mixed workloads
> > containing both short and long messages running under high network loads.
> > Homa is not API-compatible with TCP: it is connectionless and message-
> > oriented (but still reliable and flow-controlled). Homa's new API not
> > only contributes to its performance gains, but it also eliminates the
> > massive amount of connection state required by TCP for highly connected
> > datacenter workloads (Homa uses ~ 1 socket per application, whereas
> > TCP requires a separate socket for each peer).
>
> make coccicheck says:
>
> net/homa/homa_peer.c:213:21-22: WARNING opportunity for swap()
Fixed.
^ permalink raw reply
* Re: [PATCH net-next 5/5] selftests: ovpn: align command flow with TAP
From: Jakub Kicinski @ 2026-04-13 23:56 UTC (permalink / raw)
To: antonio
Cc: Jakub Kicinski, netdev, ralf, sd, pabeni, andrew+netdev, davem,
edumazet
In-Reply-To: <20260412221121.410365-6-antonio@openvpn.net>
This is an AI-generated review of your patch. The human sending this
email says: "The second one looks legit, would you prefer to follow
up or respin?"
---
selftests: ovpn: align command flow with TAP
This commit restructures the ovpn selftests to use the TAP infrastructure,
splitting each test into stages with KTAP output, adding command wrappers,
and centralizing behavior control in common.sh.
> diff --git a/tools/testing/selftests/net/ovpn/common.sh b/tools/testing/selftests/net/ovpn/common.sh
> index 2b3f2e5c8cc9d..ec6fea37ceb39 100644
> --- a/tools/testing/selftests/net/ovpn/common.sh
> +++ b/tools/testing/selftests/net/ovpn/common.sh
[ ... ]
> @@ -152,8 +263,8 @@ ovpn_compare_ntfs() {
> received="${OVPN_TMP_JSONS[$1]}"
> diff_file=$(mktemp)
>
> - kill -TERM ${OVPN_LISTENER_PIDS[$1]} || true
> - wait ${OVPN_LISTENER_PIDS[$1]} || true
> + kill -TERM ${OVPN_LISTENER_PIDS[$1]} 2>/dev/null || true
> + wait ${OVPN_LISTENER_PIDS[$1]} 2>/dev/null || true
> printf "Checking notifications for peer ${1}... "
Does this leave the reaped PID in OVPN_LISTENER_PIDS?
Later, ovpn_cleanup() calls ovpn_stop_listener() which iterates over the
array and issues another kill -TERM on the same PID. Since the process was
already reaped by the wait here, could the OS recycle the PID, potentially
causing the second kill to terminate an unrelated system process?
[ ... ]
> +ovpn_cleanup_peer_ns() {
> + local peer="$1"
> +
> + if ! ip netns list | grep -qx "${peer}"; then
> + return 0
> + fi
Can this fail to match and permanently leak network namespaces?
When virtual interfaces are moved into namespaces during setup, ip netns list
often includes an internal ID in its output, such as:
peer0 (id: 0)
Because grep is using the -x flag to enforce an exact whole-line match,
it appears this will fail to match peer0 and return early, skipping the
subsequent namespace deletion.
[ ... ]
> diff --git a/tools/testing/selftests/net/ovpn/test-close-socket.sh b/tools/testing/selftests/net/ovpn/test-close-socket.sh
> index 4f0367c60fdaf..e97affe727383 100755
> --- a/tools/testing/selftests/net/ovpn/test-close-socket.sh
> +++ b/tools/testing/selftests/net/ovpn/test-close-socket.sh
> @@ -9,37 +9,72 @@ set -e
>
> source ./common.sh
Is test-close-socket.sh missing the set -eE flag and the ovpn_stage_err
ERR trap?
The other test scripts were updated to use set -eE and register the trap,
which handles emitting the KTAP failure directive. If a command fails here,
will set -e cause the script to exit immediately and bypass the KTAP
failure reporting entirely?
^ permalink raw reply
* Re: [PATCH net-next 2/5] selftests: ovpn: fail notification check on mismatch
From: Jakub Kicinski @ 2026-04-14 0:00 UTC (permalink / raw)
To: Antonio Quartulli
Cc: netdev, ralf, Sabrina Dubroca, Paolo Abeni, Andrew Lunn,
David S. Miller, Eric Dumazet
In-Reply-To: <20260412221121.410365-3-antonio@openvpn.net>
On Mon, 13 Apr 2026 00:11:18 +0200 Antonio Quartulli wrote:
> compare_ntfs doesn't fail when expected and received notification
> streams diverge.
>
> Fix this bug by trackink the diff exit status explicitly and return it
> to the caller so notification mismatches propagate as test failures.
Hm, this series nicely cleans up test_mark.sh failures
but test_tcp.sh now always fails on debug (slow) kernel
builds with:
# TAP version 13
# 1..12
# ok 1 setup network topology
# ok 2 run baseline data traffic
# ok 3 run LAN traffic behind peer1
# ok 4 run iperf throughput
# ok 5 run key rollout
# ok 6 query peers
# ok 7 query missing peer fails
# ok 8 peer lifecycle and key queries
# ok 9 delete peer while traffic
# ok 10 delete stale keys
# ok 11 check timeout behavior
# Checking notifications for peer 3... failed
# 1,9d0
# < {
# < "name": "peer-del-ntf",
# < "msg": {
# < "peer": {
# < "del-reason": "expired",
# < "id": 12
# < }
# < }
# < }
# validate listener output for peer 3: command failed with rc=1: ovpn_compare_ntfs 3
# not ok 12 validate notification output
# # Totals: pass:11 fail:1 xfail:0 xpass:0 skip:0 error:0
Similar failure in test_symmetric_id_tcp.sh
Only the debug kernels tho, non-debug kernels seem to pass.
So probably some race / slowness.
More runs if you want to compare
https://netdev.bots.linux.dev/contest.html?executor=vmksft-net-extra-dbg&test=test-tcp-sh
This series landed in net-next-2026-04-13--06-00
^ permalink raw reply
* Re: [PATCH net-next 0/3] net: airoha: Preliminary series to support multiple net_devices connected to the same GDM port
From: patchwork-bot+netdevbpf @ 2026-04-14 0:00 UTC (permalink / raw)
To: Lorenzo Bianconi
Cc: andrew+netdev, davem, edumazet, kuba, pabeni, linux-arm-kernel,
linux-mediatek, netdev, xuegang.lu
In-Reply-To: <20260412-airoha-multi-serdes-preliminary-patch-v1-0-08d5b670ca8f@kernel.org>
Hello:
This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sun, 12 Apr 2026 19:13:11 +0200 you wrote:
> EN7581 or AN7583 SoCs support connecting multiple external SerDes (e.g.
> Ethernet or USB SerDes) to GDM3 or GDM4 ports via a hw arbiter that
> manages the traffic in a TDM manner.
> This series introduces some preliminary changes necessary to introduce
> support for multiple net_devices connected to the same Frame Engine (FE)
> GDM port (GDM3 or GDM4).
>
> [...]
Here is the summary with links:
- [net-next,1/3] net: airoha: Rely on net_device pointer in airoha_dev_setup_tc_block signature
https://git.kernel.org/netdev/net-next/c/360d745a5319
- [net-next,2/3] net: airoha: Rely on net_device pointer in HTB callbacks
https://git.kernel.org/netdev/net-next/c/8baf4bf72ef9
- [net-next,3/3] net: airoha: Rely on net_device pointer in ETS callbacks
https://git.kernel.org/netdev/net-next/c/ae32f80018f0
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net] sctp: disable BH before calling udp_tunnel_xmit_skb()
From: patchwork-bot+netdevbpf @ 2026-04-14 0:10 UTC (permalink / raw)
To: Xin Long
Cc: netdev, linux-sctp, davem, kuba, edumazet, pabeni, horms,
marcelo.leitner, bestswngs
In-Reply-To: <c874a8548221dcd56ff03c65ba75a74e6cf99119.1776017727.git.lucien.xin@gmail.com>
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sun, 12 Apr 2026 14:15:27 -0400 you wrote:
> udp_tunnel_xmit_skb() / udp_tunnel6_xmit_skb() are expected to run with
> BH disabled. After commit 6f1a9140ecda ("add xmit recursion limit to
> tunnel xmit functions"), on the path:
>
> udp(6)_tunnel_xmit_skb() -> ip(6)tunnel_xmit()
>
> dev_xmit_recursion_inc()/dec() must stay balanced on the same CPU.
>
> [...]
Here is the summary with links:
- [net] sctp: disable BH before calling udp_tunnel_xmit_skb()
https://git.kernel.org/netdev/net/c/2cd7e6971fc2
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net] sctp: fix missing encap_port propagation for GSO fragments
From: patchwork-bot+netdevbpf @ 2026-04-14 0:10 UTC (permalink / raw)
To: Xin Long
Cc: netdev, linux-sctp, davem, kuba, edumazet, pabeni, horms,
marcelo.leitner
In-Reply-To: <ea65ed61b3598d8b4940f0170b9aa1762307e6c3.1776017631.git.lucien.xin@gmail.com>
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sun, 12 Apr 2026 14:13:51 -0400 you wrote:
> encap_port in SCTP_INPUT_CB(skb) is used by sctp_vtag_verify() for
> SCTP-over-UDP processing. In the GSO case, it is only set on the head
> skb, while fragment skbs leave it 0.
>
> This results in fragment skbs seeing encap_port == 0, breaking
> SCTP-over-UDP connections.
>
> [...]
Here is the summary with links:
- [net] sctp: fix missing encap_port propagation for GSO fragments
https://git.kernel.org/netdev/net/c/bf6f95ae3b8b
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net v3] netfilter: nft_set_rbtree: fix use count leak on transaction abort
From: Pablo Neira Ayuso @ 2026-04-14 0:11 UTC (permalink / raw)
To: Marko Jevtic
Cc: fw, netfilter-devel, phil, coreteam, davem, edumazet, kuba,
pabeni, horms, netdev, linux-kernel
In-Reply-To: <20260412222801.34965-1-marko.jevtic@codereflect.io>
Hi,
On Mon, Apr 13, 2026 at 12:28:01AM +0200, Marko Jevtic wrote:
> nft_rbtree_abort() does not handle elements moved to the expired list
> by inline GC during __nft_rbtree_insert(). When inline GC encounters
> expired elements during overlap detection, it calls
> nft_rbtree_gc_elem_move() which deactivates element data (decrementing
> chain/object use counts), removes the element from the rbtree, and
> queues it for deferred freeing. On commit, these elements are freed
> via nft_rbtree_gc_queue(). On abort, however, the expired list is
> ignored entirely.
>
> This leaves use counts permanently decremented after abort.
Yes, but that is expected?
Expired elements reside in priv->expired, these elements are already
deactivated, ie. removed from the rbtree and chain reference is
decremented.
From abort path, the deactivated element simply remains there until
there is a commit run that gets rid of it.
I can't make any sense of this bug report so far.
Why do you think there is a need to restore an expired element?
^ permalink raw reply
* Re: [PATCH net v2] RDS: Fix memory leak in rds_rdma_extra_size()
From: Allison Henderson @ 2026-04-14 0:15 UTC (permalink / raw)
To: Xiaobo Liu, David S. Miller
Cc: Eric Dumazet, Jakub Kicinski, Paolo Abeni, Simon Horman, netdev,
linux-rdma, rds-devel, linux-kernel
In-Reply-To: <20260413070005.15272-1-cppcoffee@gmail.com>
On Mon, 2026-04-13 at 15:00 +0800, Xiaobo Liu wrote:
> Free iov->iov when copy_from_user() or page count validation fails
> in rds_rdma_extra_size().
>
> This preserves the existing success path and avoids leaking the
> allocated iovec array on error.
>
> Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
I think this looks good now. Thanks Xiaobo.
Reviewed-by: Allison Henderson <achender@kernel.org>
> ---
> net/rds/rdma.c | 28 +++++++++++++++++++++-------
> 1 file changed, 21 insertions(+), 7 deletions(-)
>
> diff --git a/net/rds/rdma.c b/net/rds/rdma.c
> index aa6465dc7..91a20c1e2 100644
> --- a/net/rds/rdma.c
> +++ b/net/rds/rdma.c
> @@ -560,6 +560,7 @@ int rds_rdma_extra_size(struct rds_rdma_args *args,
> struct rds_iovec *vec;
> struct rds_iovec __user *local_vec;
> int tot_pages = 0;
> + int ret = 0;
> unsigned int nr_pages;
> unsigned int i;
>
> @@ -578,16 +579,20 @@ int rds_rdma_extra_size(struct rds_rdma_args *args,
> vec = &iov->iov[0];
>
> if (copy_from_user(vec, local_vec, args->nr_local *
> - sizeof(struct rds_iovec)))
> - return -EFAULT;
> + sizeof(struct rds_iovec))) {
> + ret = -EFAULT;
> + goto out;
> + }
> iov->len = args->nr_local;
>
> /* figure out the number of pages in the vector */
> for (i = 0; i < args->nr_local; i++, vec++) {
>
> nr_pages = rds_pages_in_vec(vec);
> - if (nr_pages == 0)
> - return -EINVAL;
> + if (nr_pages == 0) {
> + ret = -EINVAL;
> + goto out;
> + }
>
> tot_pages += nr_pages;
>
> @@ -595,11 +600,20 @@ int rds_rdma_extra_size(struct rds_rdma_args *args,
> * nr_pages for one entry is limited to (UINT_MAX>>PAGE_SHIFT)+1,
> * so tot_pages cannot overflow without first going negative.
> */
> - if (tot_pages < 0)
> - return -EINVAL;
> + if (tot_pages < 0) {
> + ret = -EINVAL;
> + goto out;
> + }
> }
>
> - return tot_pages * sizeof(struct scatterlist);
> + ret = tot_pages * sizeof(struct scatterlist);
> +
> +out:
> + if (ret < 0) {
> + kfree(iov->iov);
> + iov->iov = NULL;
> + }
> + return ret;
> }
>
> /*
^ permalink raw reply
* Re: [PATCH net-next v05 1/6] hinic3: Add ethtool queue ops
From: Jakub Kicinski @ 2026-04-14 0:18 UTC (permalink / raw)
To: Fan Gong
Cc: Zhu Yikai, netdev, David S. Miller, Eric Dumazet, Paolo Abeni,
Simon Horman, Andrew Lunn, Ioana Ciornei, Mohsin Bashir,
linux-kernel, linux-doc, luosifu, Xin Guo, Zhou Shuai, Wu Like,
Shi Jing, Zheng Jiezhen, Maxime Chevallier
In-Reply-To: <157d5cc6e757ffa77eee01dfdc3f2159dc97905f.1775711066.git.zhuyikai1@h-partners.com>
On Sat, 11 Apr 2026 11:36:59 +0800 Fan Gong wrote:
> Implement following ethtool callback function:
> .get_ringparam
> .set_ringparam
>
> These callbacks allow users to utilize ethtool for detailed
> queue depth configuration and monitoring.
> +static int hinic3_check_ringparam_valid(struct net_device *netdev,
> + const struct ethtool_ringparam *ring)
> +{
> + if (ring->rx_jumbo_pending || ring->rx_mini_pending) {
Can driver actually be called with non-zero values if max is not set?
> + netdev_err(netdev, "Unsupported rx_jumbo_pending/rx_mini_pending\n");
> + return -EINVAL;
> + }
> + if (ring->tx_pending > HINIC3_MAX_TX_QUEUE_DEPTH ||
> + ring->tx_pending < HINIC3_MIN_QUEUE_DEPTH ||
> + ring->rx_pending > HINIC3_MAX_RX_QUEUE_DEPTH ||
> + ring->rx_pending < HINIC3_MIN_QUEUE_DEPTH) {
similar question - do you need to check the upper bound?
kernel should check the input against max returned by .get
> + netdev_err(netdev,
please use extack for errors
> + "Queue depth out of range tx[%d-%d] rx[%d-%d]\n",
> + HINIC3_MIN_QUEUE_DEPTH, HINIC3_MAX_TX_QUEUE_DEPTH,
> + HINIC3_MIN_QUEUE_DEPTH, HINIC3_MAX_RX_QUEUE_DEPTH);
> + return -EINVAL;
> + }
> +
> + return 0;
> +}
> +
> +static int hinic3_set_ringparam(struct net_device *netdev,
> + struct ethtool_ringparam *ring,
> + struct kernel_ethtool_ringparam *kernel_ring,
> + struct netlink_ext_ack *extack)
> +{
> + struct hinic3_nic_dev *nic_dev = netdev_priv(netdev);
> + struct hinic3_dyna_txrxq_params q_params = {};
> + u32 new_sq_depth, new_rq_depth;
> + int err;
> +
> + err = hinic3_check_ringparam_valid(netdev, ring);
> + if (err)
> + return err;
> +
> + new_sq_depth = 1U << ilog2(ring->tx_pending);
> + new_rq_depth = 1U << ilog2(ring->rx_pending);
> + if (new_sq_depth == nic_dev->q_params.sq_depth &&
> + new_rq_depth == nic_dev->q_params.rq_depth)
> + return 0;
> +
> + if (new_sq_depth != ring->tx_pending)
> + netdev_info(netdev, "Requested Tx depth trimmed to %d\n",
> + new_sq_depth);
please use extack for warnings
> + if (new_rq_depth != ring->rx_pending)
> + netdev_info(netdev, "Requested Rx depth trimmed to %d\n",
> + new_rq_depth);
> +
> + netdev_info(netdev, "Change Tx/Rx ring depth from %u/%u to %u/%u\n",
> + nic_dev->q_params.sq_depth, nic_dev->q_params.rq_depth,
> + new_sq_depth, new_rq_depth);
> +
> + if (!netif_running(netdev)) {
> + hinic3_update_qp_depth(netdev, new_sq_depth, new_rq_depth);
> + } else {
> + q_params = nic_dev->q_params;
> + q_params.sq_depth = new_sq_depth;
> + q_params.rq_depth = new_rq_depth;
> +
> + err = hinic3_change_channel_settings(netdev, &q_params);
> + if (err) {
> + netdev_err(netdev, "Failed to change channel settings\n");
> + return err;
> + }
> + }
> +
> + return 0;
> +}
> +
> static const struct ethtool_ops hinic3_ethtool_ops = {
> .supported_coalesce_params = ETHTOOL_COALESCE_USECS |
> ETHTOOL_COALESCE_PKT_RATE_RX_USECS,
> @@ -417,6 +516,8 @@ static const struct ethtool_ops hinic3_ethtool_ops = {
> .get_msglevel = hinic3_get_msglevel,
> .set_msglevel = hinic3_set_msglevel,
> .get_link = ethtool_op_get_link,
> + .get_ringparam = hinic3_get_ringparam,
> + .set_ringparam = hinic3_set_ringparam,
> };
> diff --git a/drivers/net/ethernet/huawei/hinic3/hinic3_main.c b/drivers/net/ethernet/huawei/hinic3/hinic3_main.c
> index 0a888fe4c975..3b470978714a 100644
> --- a/drivers/net/ethernet/huawei/hinic3/hinic3_main.c
> +++ b/drivers/net/ethernet/huawei/hinic3/hinic3_main.c
> @@ -179,6 +179,8 @@ static int hinic3_sw_init(struct net_device *netdev)
> int err;
>
> mutex_init(&nic_dev->port_state_mutex);
> + mutex_init(&nic_dev->channel_cfg_lock);
Why do you need this mutex?
Aren't all the places you take it under rtnl_lock anyway?
> + spin_lock_init(&nic_dev->channel_res_lock);
>
> nic_dev->q_params.sq_depth = HINIC3_SQ_DEPTH;
> nic_dev->q_params.rq_depth = HINIC3_RQ_DEPTH;
> @@ -314,6 +316,15 @@ static void hinic3_link_status_change(struct net_device *netdev,
> bool link_status_up)
> {
> struct hinic3_nic_dev *nic_dev = netdev_priv(netdev);
> + unsigned long flags;
> + bool valid;
> +
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + valid = HINIC3_CHANNEL_RES_VALID(nic_dev);
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + if (!valid)
Why are you checking valid here? What if the state changes immediately
after unlocking?
> + return;
>
> if (link_status_up) {
> if (netif_carrier_ok(netdev))
> diff --git a/drivers/net/ethernet/huawei/hinic3/hinic3_netdev_ops.c b/drivers/net/ethernet/huawei/hinic3/hinic3_netdev_ops.c
> index da73811641a9..d652a5ffdc2c 100644
> --- a/drivers/net/ethernet/huawei/hinic3/hinic3_netdev_ops.c
> +++ b/drivers/net/ethernet/huawei/hinic3/hinic3_netdev_ops.c
> @@ -428,6 +428,85 @@ static void hinic3_vport_down(struct net_device *netdev)
> }
> }
>
> +int
> +hinic3_change_channel_settings(struct net_device *netdev,
> + struct hinic3_dyna_txrxq_params *trxq_params)
> +{
> + struct hinic3_nic_dev *nic_dev = netdev_priv(netdev);
> + struct hinic3_dyna_txrxq_params old_qp_params = {};
> + struct hinic3_dyna_qp_params new_qp_params = {};
> + struct hinic3_dyna_qp_params cur_qp_params = {};
> + bool need_teardown = false;
> + unsigned long flags;
> + int err;
> +
> + mutex_lock(&nic_dev->channel_cfg_lock);
> +
> + hinic3_config_num_qps(netdev, trxq_params);
> +
> + err = hinic3_alloc_channel_resources(netdev, &new_qp_params,
> + trxq_params);
> + if (err) {
> + netdev_err(netdev, "Failed to alloc channel resources\n");
> + mutex_unlock(&nic_dev->channel_cfg_lock);
> + return err;
> + }
> +
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + if (!test_and_set_bit(HINIC3_CHANGE_RES_INVALID, &nic_dev->flags))
> + need_teardown = true;
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + if (need_teardown) {
> + hinic3_vport_down(netdev);
> + hinic3_close_channel(netdev);
> + hinic3_uninit_qps(nic_dev, &cur_qp_params);
> + hinic3_free_channel_resources(netdev, &cur_qp_params,
> + &nic_dev->q_params);
> + }
> +
> + if (nic_dev->num_qp_irq > trxq_params->num_qps)
> + hinic3_qp_irq_change(netdev, trxq_params->num_qps);
> +
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + old_qp_params = nic_dev->q_params;
> + nic_dev->q_params = *trxq_params;
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + hinic3_init_qps(nic_dev, &new_qp_params);
> +
> + err = hinic3_open_channel(netdev);
This "open" function allocates Rx buffers, and fails if it couldn't get
even one. That's no good.
> + if (err)
> + goto err_uninit_qps;
> +
> + err = hinic3_vport_up(netdev);
> + if (err)
> + goto err_close_channel;
> +
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + clear_bit(HINIC3_CHANGE_RES_INVALID, &nic_dev->flags);
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + mutex_unlock(&nic_dev->channel_cfg_lock);
> +
> + return 0;
> +
> +err_close_channel:
> + hinic3_close_channel(netdev);
> +err_uninit_qps:
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + nic_dev->q_params = old_qp_params;
> + clear_bit(HINIC3_CHANGE_RES_INVALID, &nic_dev->flags);
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + hinic3_uninit_qps(nic_dev, &new_qp_params);
> + hinic3_free_channel_resources(netdev, &new_qp_params, trxq_params);
> +
> + mutex_unlock(&nic_dev->channel_cfg_lock);
AI says:
Can this error path lead to memory corruption?
If need_teardown was true, the old channel resources were freed earlier in
the function. If hinic3_open_channel() or hinic3_vport_up() fails, the code
jumps to err_uninit_qps and restores nic_dev->q_params = old_qp_params.
However, it doesn't appear to re-allocate those old resources or mark the
interface as down. Could a subsequent administrative teardown or network
traffic dereference these freed pointers?
> +
> static int hinic3_open(struct net_device *netdev)
> {
> struct hinic3_nic_dev *nic_dev = netdev_priv(netdev);
> @@ -487,16 +566,33 @@ static int hinic3_close(struct net_device *netdev)
> {
> struct hinic3_nic_dev *nic_dev = netdev_priv(netdev);
> struct hinic3_dyna_qp_params qp_params;
> + bool need_teardown = false;
> + unsigned long flags;
>
> if (!test_and_clear_bit(HINIC3_INTF_UP, &nic_dev->flags)) {
> netdev_dbg(netdev, "Netdev already close, do nothing\n");
> return 0;
> }
>
> - hinic3_vport_down(netdev);
> - hinic3_close_channel(netdev);
> - hinic3_uninit_qps(nic_dev, &qp_params);
> - hinic3_free_channel_resources(netdev, &qp_params, &nic_dev->q_params);
> + mutex_lock(&nic_dev->channel_cfg_lock);
> +
> + spin_lock_irqsave(&nic_dev->channel_res_lock, flags);
> + if (!test_and_set_bit(HINIC3_CHANGE_RES_INVALID, &nic_dev->flags))
> + need_teardown = true;
> + spin_unlock_irqrestore(&nic_dev->channel_res_lock, flags);
> +
> + if (need_teardown) {
> + hinic3_vport_down(netdev);
> + hinic3_close_channel(netdev);
> + hinic3_uninit_qps(nic_dev, &qp_params);
> + hinic3_free_channel_resources(netdev, &qp_params,
> + &nic_dev->q_params);
> + }
> +
> + hinic3_free_nicio_res(nic_dev);
> + hinic3_destroy_num_qps(netdev);
> +
> + mutex_unlock(&nic_dev->channel_cfg_lock);
>
> return 0;
> }
> diff --git a/drivers/net/ethernet/huawei/hinic3/hinic3_nic_dev.h b/drivers/net/ethernet/huawei/hinic3/hinic3_nic_dev.h
> index 9502293ff710..55b280888ad8 100644
> --- a/drivers/net/ethernet/huawei/hinic3/hinic3_nic_dev.h
> +++ b/drivers/net/ethernet/huawei/hinic3/hinic3_nic_dev.h
> @@ -10,6 +10,9 @@
> #include "hinic3_hw_cfg.h"
> #include "hinic3_hwdev.h"
> #include "hinic3_mgmt_interface.h"
> +#include "hinic3_nic_io.h"
> +#include "hinic3_tx.h"
> +#include "hinic3_rx.h"
>
> #define HINIC3_VLAN_BITMAP_BYTE_SIZE(nic_dev) (sizeof(*(nic_dev)->vlan_bitmap))
> #define HINIC3_VLAN_BITMAP_SIZE(nic_dev) \
> @@ -20,8 +23,13 @@ enum hinic3_flags {
> HINIC3_MAC_FILTER_CHANGED,
> HINIC3_RSS_ENABLE,
> HINIC3_UPDATE_MAC_FILTER,
> + HINIC3_CHANGE_RES_INVALID,
> };
>
> +#define HINIC3_CHANNEL_RES_VALID(nic_dev) \
> + (test_bit(HINIC3_INTF_UP, &(nic_dev)->flags) && \
> + !test_bit(HINIC3_CHANGE_RES_INVALID, &(nic_dev)->flags))
I don't get why you need to check both of these bits.
Can't there be one bit for "resources valid" ?
And it will only be set while device is up (of course) so no need to
also check UP (this way checking can be atomic without the spin lock).
> enum hinic3_event_work_flags {
> HINIC3_EVENT_WORK_TX_TIMEOUT,
> };
> @@ -129,6 +137,10 @@ struct hinic3_nic_dev {
> struct work_struct rx_mode_work;
> /* lock for enable/disable port */
> struct mutex port_state_mutex;
> + /* lock for channel configuration */
> + struct mutex channel_cfg_lock;
> + /* lock for channel resources */
> + spinlock_t channel_res_lock;
>
> struct list_head uc_filter_list;
> struct list_head mc_filter_list;
> @@ -143,6 +155,10 @@ struct hinic3_nic_dev {
>
> void hinic3_set_netdev_ops(struct net_device *netdev);
> int hinic3_set_hw_features(struct net_device *netdev);
> +int
> +hinic3_change_channel_settings(struct net_device *netdev,
> + struct hinic3_dyna_txrxq_params *trxq_params);
> +
> int hinic3_qps_irq_init(struct net_device *netdev);
> void hinic3_qps_irq_uninit(struct net_device *netdev);
>
> diff --git a/drivers/net/ethernet/huawei/hinic3/hinic3_nic_io.h b/drivers/net/ethernet/huawei/hinic3/hinic3_nic_io.h
> index 12eefabcf1db..3791b9bc865b 100644
> --- a/drivers/net/ethernet/huawei/hinic3/hinic3_nic_io.h
> +++ b/drivers/net/ethernet/huawei/hinic3/hinic3_nic_io.h
> @@ -14,6 +14,10 @@ struct hinic3_nic_dev;
> #define HINIC3_RQ_WQEBB_SHIFT 3
> #define HINIC3_SQ_WQEBB_SIZE BIT(HINIC3_SQ_WQEBB_SHIFT)
>
> +#define HINIC3_MAX_TX_QUEUE_DEPTH 65536
> +#define HINIC3_MAX_RX_QUEUE_DEPTH 16384
> +#define HINIC3_MIN_QUEUE_DEPTH 128
> +
> /* ******************** RQ_CTRL ******************** */
> enum hinic3_rq_wqe_type {
> HINIC3_NORMAL_RQ_WQE = 1,
^ permalink raw reply
* Re: [PATCH net-next] net: airoha: Remove PCE_MC_EN_MASK bit in REG_FE_PCE_CFG configuration
From: patchwork-bot+netdevbpf @ 2026-04-14 0:20 UTC (permalink / raw)
To: Lorenzo Bianconi
Cc: andrew+netdev, davem, edumazet, kuba, pabeni, linux-arm-kernel,
linux-mediatek, netdev
In-Reply-To: <20260412-airoha_fe_init_remove_mc_en_bit-v1-1-7b6a5a25a74d@kernel.org>
Hello:
This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:
On Sun, 12 Apr 2026 11:56:25 +0200 you wrote:
> PCE_MC_EN_MASK bit in REG_FE_PCE_CFG configuration performed in
> airoha_fe_init() is used to duplicate multicast packets and send a copy
> to the CPU when the traffic is offloaded. This is necessary just if
> it is requested by the user. Disable multicast packets duplication by
> default.
>
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
>
> [...]
Here is the summary with links:
- [net-next] net: airoha: Remove PCE_MC_EN_MASK bit in REG_FE_PCE_CFG configuration
https://git.kernel.org/netdev/net-next/c/34e1a98ff2a8
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply
* Re: [PATCH net-next v4 02/10] enic: add admin channel open and close for SR-IOV
From: Jakub Kicinski @ 2026-04-14 0:21 UTC (permalink / raw)
To: devnull+satishkh.cisco.com
Cc: Jakub Kicinski, andrew+netdev, davem, edumazet, pabeni, netdev,
linux-kernel, 20260401-enic-sriov-v2-prep-v4-0-d5834b2ef1b9,
satishkh
In-Reply-To: <20260411-enic-sriov-v2-admin-channel-v2-v4-2-f052326c2a57@cisco.com>
This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
---
enic: add admin channel open and close for SR-IOV
Introduce enic_admin_channel_open() and enic_admin_channel_close() to allocate
and initialize the admin WQ, RQ, two CQs, and one SR-IOV interrupt for the V2
SR-IOV PF-VF mailbox communication.
> diff --git a/drivers/net/ethernet/cisco/enic/enic_admin.c b/drivers/net/ethernet/cisco/enic/enic_admin.c
> new file mode 100644
> index 0000000000000..d1abe6a50095d
> --- /dev/null
> +++ b/drivers/net/ethernet/cisco/enic/enic_admin.c
[ ... ]
> + err = vnic_cq_alloc_with_type(enic->vdev, &enic->admin_cq[0], 0,
> + ENIC_ADMIN_DESC_COUNT,
> + sizeof(struct cq_desc),
> + RES_TYPE_ADMIN_CQ);
> + if (err)
> + goto free_rq;
> +
> + err = vnic_cq_alloc_with_type(enic->vdev, &enic->admin_cq[1], 1,
> + ENIC_ADMIN_DESC_COUNT,
> + 16 << enic->ext_cq,
^^^^^^^^^^^^^^^^^^
Does this code use the wrong descriptor size for the admin RQ's completion
queue?
The admin WQ's completion queue (admin_cq[0]) uses sizeof(struct cq_desc),
which is 16 bytes. If the admin channel handles PF-VF mailbox messages rather
than data packets with extended metadata, will firmware still write standard
16-byte completions here?
If so, won't allocating the ring with an extended size cause the host polling
loop to look for the color bit at the wrong offset (e.g., byte 31 instead of
byte 15), causing it to silently miss incoming admin messages?
> + RES_TYPE_ADMIN_CQ);
> + if (err)
> + goto free_cq0;
--
pw-bot: cr
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox