* Re: [EXTERNAL] Re: [PATCH net,v5] virtio_net: clamp rss_max_key_size to NETDEV_RSS_KEY_LEN
From: Jakub Kicinski @ 2026-04-01 1:05 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: Paolo Abeni, Srujana Challa, netdev@vger.kernel.org,
virtualization@lists.linux.dev, jasowang@redhat.com,
xuanzhuo@linux.alibaba.com, eperezma@redhat.com,
davem@davemloft.net, edumazet@google.com, Nithin Kumar Dabilpuram,
Shiva Shankar Kommula, stable@vger.kernel.org
In-Reply-To: <20260331104737-mutt-send-email-mst@kernel.org>
On Tue, 31 Mar 2026 10:48:41 -0400 Michael S. Tsirkin wrote:
> > > Thank you for the feedback. In net-next, NETDEV_RSS_KEY_LEN is 256. This fix is
> > > also intended for stable kernels, where NETDEV_RSS_KEY_LEN is 52, and
> > > I added the message to make clamping visible in that case.
> > > I will remove the check and send the next version.
> >
> > I'm sorry, I haven't looked at the historical context when I wrote my
> > previous reply.
> >
> > IMHO the additional check does not make sense in the current net tree.
> > On the flip side stable trees will need it. I suggest:
> >
> > - dropping the check for the 'net' patch
> > - also dropping CC: stable tag
> > - explicitly sending to stable the fix variant including the size check.
> >
> > @Michael: WDYT?
>
> I was the one who suggested it, the extra check is harmless, I'm
> inclined to always have it. Less work than maintaining two patches.
Give us an RB tag please and lets close this one? :)
^ permalink raw reply
* Re: [PATCH 1/1] net: ipv6: flowlabel: defer exclusive option free until RCU teardown
From: Eric Dumazet @ 2026-04-01 1:05 UTC (permalink / raw)
To: Yuan Tan
Cc: Ren Wei, security, netdev, davem, dsahern, kuba, pabeni, horms,
afaerber, mani, yoshfuji, yifanwucs, tomapufckgml, bird,
enjou1224z, zcliangcn
In-Reply-To: <4631f54a-bcc8-454b-9355-15a21f9e1e21@gmail.com>
On Tue, Mar 31, 2026 at 5:41 PM Yuan Tan <yuantan098@gmail.com> wrote:
>
>
> On 3/31/2026 1:52 AM, Eric Dumazet wrote:
> > On Tue, Mar 31, 2026 at 1:34 AM Ren Wei <n05ec@lzu.edu.cn> wrote:
> >> From: Zhengchuan Liang <zcliangcn@gmail.com>
> >>
> >> `ip6fl_seq_show()` walks the global flowlabel hash under the seq-file
> >> RCU read-side lock and prints `fl->opt->opt_nflen` when an option block
> >> is present.
> > Some points :
> >
> > Please do not CC security@ if you submit a public patch, there is
> > absolutely no reason for it.
> >
> > Please do not resend the same version within 24 hours; this creates noise.
> > Your patch wasn't lost; we have many patch reviews and similar bugs to address.
> >
> > Thank you.
>
> We sincerely apologize for the confusion caused by our previous email. Previously, when we sent reports and patches to the security list, we were advised that patches should be submitted to netdev@ for public review. To follow this, we are now sending the full vulnerability details to security@ while additionally submitting the patches to netdev@. Please let us know if this is the correct way to handle such cases.
When/If an issue and a patch is sent publicly, security@ involvement
ends. netdev maintainers take over.
Thus there is no point CCing this list of security officers, which is
currently flooded by many LLM-based findings.
Please help them by keeping their inbox a bit saner.
>
> Regarding the CC list, could you provide further guidance on who we should include? Currently, we are CCing everyone suggested by get_maintainer.pl because we previously received feedback about missing relevant maintainers. For public mailing lists, we should only including netdev@. Should we be more selective?
Yes, folks that are interesting into following netdev changes are
subscribed to netdev@ mailing list.
Unless a patch touches non-networking parts and requires approval from
other branch maintainers, there is no
point CCing other lists.
>
> Finally, regarding our tags, our team has a clear division of labor—some focus on finding vulnerabilities, others on fixing them, followed by an internal review and testing before a dedicated member handles community outreach.
>
You included 8 tags, this seems a bit too much for such a small patch.
> We welcome any guidance and will improve our upcoming patches accordingly. To be honest, we are currently a bit uncertain about the exact workflow the community prefers, and we are eager to align our process with your expectations.
>
^ permalink raw reply
* Re: [PATCH v2] net: ns83820: fix DMA mapping error handling in hard_start_xmit
From: Jakub Kicinski @ 2026-04-01 1:07 UTC (permalink / raw)
To: Wang Jun
Cc: Andrew Lunn, David S . Miller, Eric Dumazet, Paolo Abeni, netdev,
linux-kernel, gszhai, 25125332, 25125283, 23120469
In-Reply-To: <tencent_EB7280CE1394ABF5035CDA078FC60A481906@qq.com>
On Tue, 31 Mar 2026 20:21:41 +0800 Wang Jun wrote:
> This is v2 of the DMA mapping error handling fix for ns83820. Changes since v1:
>
> - Added queue restart check in error path to avoid potential TX queue stall
> (as pointed out by the AI review)
> - Adjusted variable declarations to follow reverse christmas tree order
> - Switched from dma_unmap_single to dma_unmap_page for fragments
>
> Thanks to reviewers for the feedback.
Please do not post new patches in reply to old ones.
Please do not post new patches within 24h of previous posting.
And ask yourself if you trying and failing to post patches for this
ancient driver is a good use of maintainer's time.
^ permalink raw reply
* Re: [PATCH] igc: fix Tx timestamp timeout caused by unlocked TIMINCA write in adj fine
From: Bob Van Valzah @ 2026-04-01 1:14 UTC (permalink / raw)
To: Vadim Fedorenko
Cc: intel-wired-lan, anthony.l.nguyen, netdev, julianstj, jeff,
Lasse Johnsen
In-Reply-To: <B22CCB1A-15FE-4233-9F47-7FE522C8D6FF@VanValzah.Com>
There is one other symptom of this bug that we neglected to report previously: PEROUT stops. That’s PPS OUT for us, so it’s how we first noticed the problem. Recovery was to unload and reload the igc driver.
I take this as further evidence for my hunch that the hardware just reaches some internal “this should never happen” state and the PHC just stops ticking.
Hoping this additional symptom report might help with diagnosis.
Bob
> On Mar 30, 2026, at 12:42 PM, Bob Van Valzah <Bob@VanValzah.Com> wrote:
>
> Thanks for following up Vadim. To be clear, we do not see our patch as a solution to the problem, since it does not eliminate the underlying contention, it just reduces the likelihood.
>
> We now have a bit more hands-on time running the patch and can see its limitations. The driver just throws the timeout error less often now. Moreover, our code calling adjfine() has to be ready for an EBUSY. The patch is certainly not an elegant solution. It may well have worked better, or perhaps been a complete fix, if we had taken tmreg_lock. Sorry if we sent our patch prematurely.
>
> We think the contention happens when a system call made by a clock disciplining daemon like ts2phc changes a PHC's frequency, while a timestamping daemon like ptp4l has requested a timestamp against that same PHC. A plausible explanation is that the hardware fails to produce a timestamp for a packet when it collides with a PHC frequency adjustment, leading to the timestamp timeout. We see this contention 128x more often than the average user because we cranked our PTP sync rate up to 128 Hz. We care about every ns here. The likelihood probably also scales with the number of PTP clients.
>
> Sorry, in retrospect, we may have overstated our case in declaring the root cause as the lack of locking, since our locking only reduces the likelihood of the timestamp timeout. Intel may provide insight to the underlying cause of the timeout, but my hunch is that the collision of a timestamp request against a PHC and a frequency change of the same PHC causes the timestamp request to fail, leading to the timeout. Our repro code makes this happen so reliably that it should be easy for the hardware guys to explain exactly what’s going on. Hopefully, they can advise on a more elegant avoidance strategy than our ham-handed lock.
>
> We have not tried recovering a timestamp following the error. We may have used the term “corrupted” poorly here. Since the error message says “timeout," that could mean the hardware produced no timestamp at all, rather than a corrupted one.
>
> This isn’t a show-stopper for us, but we note that timekeeping daemons with imperfect error handling could react to this situation poorly and perhaps fail silently.
>
> Bob
>
>> On Mar 30, 2026, at 11:39 AM, Vadim Fedorenko <vadim.fedorenko@linux.dev> wrote:
>>
>> On 29/03/2026 04:25, Bob Van Valzah wrote:
>>> Hi,
>>> We found a race in igc_ptp_adjfine_i225() that causes "Tx timestamp
>>> timeout" errors and eventually wedges EXTTS when a PTP grandmaster
>>> (ptp4l with hardware timestamping) runs concurrently with PHC
>>> frequency discipline (any GPSDO calling clock_adjtime ADJ_FREQUENCY).
>>> Root cause: igc_ptp_adjfine_i225() writes IGC_TIMINCA without holding
>>> any lock. Every other PTP clock operation in igc_ptp.c (adjtime,
>>> gettime, settime) holds tmreg_lock, but adjfine does not. When the
>>> increment rate changes while the hardware is capturing a TX timestamp,
>>> the captured value is corrupt. The driver retries for
>>> IGC_PTP_TX_TIMEOUT (15s), then logs the timeout and frees the skb.
>>> Repeated occurrences eventually prevent EXTTS from delivering events.
>>> The attached reproducer (triggers in ~17 seconds on i226):
>>> One thread calling clock_adjtime(ADJ_FREQUENCY) at ~200k/s on the
>>> PHC, another sending UDP packets with SO_TIMESTAMPING requesting
>>> hardware TX timestamps at ~100k/s. A Python reproducer is at:
>>> https://github.com/bobvan/PePPAR-Fix/blob/main/tools/igc_tx_timeout_repro.py
>>> At realistic rates (1 Hz adjfine from a GPSDO + ptp4l at 128 Hz
>>> sync), the race triggers in ~30 minutes.
>>> The attached patch holds ptp_tx_lock around the TIMINCA write and
>>> skips the write if any TX timestamps are pending (tx_tstamp[i].skb
>>> != NULL), returning -EBUSY. This doesn't fully close the hardware
>>> race (a new TX capture can start between the check and the write),
>>> but at realistic rates the residual probability gives ~25 year MTBF
>>> vs ~30 minutes without the patch.
>>> A complete fix would likely require either disabling TX timestamping
>>> around TIMINCA writes (via TSYNCTXCTL), or making the timeout recovery
>>> path more robust so a single corrupt timestamp doesn't wedge the
>>> subsystem. We'd welcome guidance from the igc maintainers on the
>>> preferred approach.
>>> Tested on:
>>> - Intel i226 (TimeHAT v5 board on Raspberry Pi 5)
>>> - Kernel 6.12.62+rpt-rpi-2712 (Raspberry Pi OS)
>>> - Intel out-of-tree igc driver 5.4.0-7642.46
>>> - Stock upstream igc_ptp.c (same code, same bug)
>>> Bob
>>> ---
>>> drivers/net/ethernet/intel/igc/igc_ptp.c | 18 +++++++++++++++++-
>>> 1 file changed, 17 insertions(+), 1 deletion(-)
>>> diff --git a/drivers/net/ethernet/intel/igc/igc_ptp.c b/drivers/net/ethernet/intel/igc/igc_ptp.c
>>> index XXXXXXX..XXXXXXX 100644
>>> --- a/drivers/net/ethernet/intel/igc/igc_ptp.c
>>> +++ b/drivers/net/ethernet/intel/igc/igc_ptp.c
>>> @@ -47,8 +47,10 @@ static int igc_ptp_adjfine_i225(struct ptp_clock_info *ptp, long scaled_ppm)
>>> {
>>> struct igc_adapter *igc = container_of(ptp, struct igc_adapter,
>>> ptp_caps);
>>> struct igc_hw *hw = &igc->hw;
>>> + unsigned long flags;
>>> int neg_adj = 0;
>>> u64 rate;
>>> u32 inca;
>>> + int i;
>>> if (scaled_ppm < 0) {
>>> neg_adj = 1;
>>> @@ -63,7 +65,21 @@ static int igc_ptp_adjfine_i225(struct ptp_clock_info *ptp, long scaled_ppm)
>>> if (neg_adj)
>>> inca |= ISGN;
>>> - wr32(IGC_TIMINCA, inca);
>>> + /* Changing the clock increment rate while a TX timestamp is being
>>> + * captured by the hardware can corrupt the timestamp, causing the
>>> + * driver to report "Tx timestamp timeout" and eventually wedging
>>> + * the EXTTS subsystem. Serialize with pending TX timestamps:
>>> + * skip the rate change if any are in flight.
>>> + */
>>> + spin_lock_irqsave(&igc->ptp_tx_lock, flags);
>>> + for (i = 0; i < IGC_MAX_TX_TSTAMP_REGS; i++) {
>>> + if (igc->tx_tstamp[i].skb) {
>>> + spin_unlock_irqrestore(&igc->ptp_tx_lock, flags);
>>> + return -EBUSY;
>>> + }
>>> + }
>>> + wr32(IGC_TIMINCA, inca);
>>> + spin_unlock_irqrestore(&igc->ptp_tx_lock, flags);
>>
>> It's a bit weird solution, because in this case we may end up having no
>> successful calls to adjfine with high amount of TX timestamp packets in flight. Another problem here is that access to timing registers is
>> guarded by tmreg_lock, but here you use ptp_tx_lock, which protects
>> queue.
>>
>> Were you able to recover "corrupted" time stamps to figure out why they
>> are discarded?
>>
>>
>>> return 0;
>>> }
>>> --
>>> 2.39.2
>
>
^ permalink raw reply
* [PATCH net-next v4 00/14] net: sleepable ndo_set_rx_mode
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni
This series adds a new ndo_set_rx_mode_async callback that enables
drivers to handle address list updates in a sleepable context. The
current ndo_set_rx_mode is called under the netif_addr_lock spinlock
with BHs disabled, which prevents drivers from sleeping. This is
problematic for ops-locked drivers that need to sleep.
The approach:
1. Add snapshot/reconcile infrastructure for address lists
2. Introduce dev_rx_mode_work that takes snapshots under the lock,
drops the lock, calls the driver, then reconciles changes back
3. Move promiscuity handling into the scheduled work as well
4. Convert existing ops-locked drivers to ndo_set_rx_mode_async
5. Add a warning for ops-locked drivers still using ndo_set_rx_mode
6. Add a selftest exercising the team+bridge+macvlan topology that
triggers the addr_lock -> ops_lock ordering issue
v4:
- rebase on https://lore.kernel.org/netdev/20260319005456.82745-1-saeed@kernel.org/T/#u (Cosmin)
- reword ndo_set_rx_mode_async kdoc (Jakub)
- s/EXPORT_SYMBOL/EXPORT_SYMBOL_IF_KUNIT/ (Jakub)
- remove netif_up_and_present (Jakub)
- netif_addr_lists_snapshot + netif_addr_lists_reconcile to better
explain mix-and-match between
ndo_set_rx_mode/ndo_set_rx_mode_async/ndo_change_rx_flags (Jakub)
- s/cancel_work_sync/flush_work/ (Jakub)
- separate commit to cache snapshot entries (Jakub)
- add dev_addr_test_snapshot_benchmark (Jakub)
- dev_addr_test_snapshot_benchmark: 1024 addrs x 1000 snapshots: 89872802 ns total, 89872 ns/iter
- remove redundant bnxt_uc_list_updated (Michael)
- switch to linkwatch-like work stealing (Jakub)
v3:
- module_export(__rtnl_unlock) (nipa)
- s/netdev_uc_count/netdev_hw_addr_list_count/ in bnxt (Aleksandr)
v2:
- wifi: cfg80211: use __rtnl_unlock in nl80211_pre_doit (syzbot)
- simplify mlx5e_sync_netdev_addr for !uc (Cosmin)
- switch to snapshot in bnxt_cfg_rx_mode (Michael)
- add team to net/config (Jakub)
Stanislav Fomichev (14):
net: add address list snapshot and reconciliation infrastructure
net: introduce ndo_set_rx_mode_async and netdev_rx_mode_work
net: move promiscuity handling into netdev_rx_mode_work
net: cache snapshot entries for ndo_set_rx_mode_async
fbnic: convert to ndo_set_rx_mode_async
mlx5: convert to ndo_set_rx_mode_async
bnxt: convert to ndo_set_rx_mode_async
bnxt: use snapshot in bnxt_cfg_rx_mode
iavf: convert to ndo_set_rx_mode_async
netdevsim: convert to ndo_set_rx_mode_async
dummy: convert to ndo_set_rx_mode_async
net: warn ops-locked drivers still using ndo_set_rx_mode
selftests: net: add team_bridge_macvlan rx_mode test
selftests: net: use ip commands instead of teamd in team rx_mode test
Documentation/networking/netdevices.rst | 13 +
drivers/net/dummy.c | 6 +-
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 62 +--
drivers/net/ethernet/intel/iavf/iavf_main.c | 14 +-
.../net/ethernet/mellanox/mlx5/core/en/fs.h | 5 +-
.../net/ethernet/mellanox/mlx5/core/en_fs.c | 32 +-
.../net/ethernet/mellanox/mlx5/core/en_main.c | 13 +-
.../net/ethernet/meta/fbnic/fbnic_netdev.c | 20 +-
.../net/ethernet/meta/fbnic/fbnic_netdev.h | 4 +-
drivers/net/ethernet/meta/fbnic/fbnic_pci.c | 4 +-
drivers/net/ethernet/meta/fbnic/fbnic_rpc.c | 2 +-
drivers/net/netdevsim/netdev.c | 8 +-
include/linux/netdevice.h | 26 ++
net/core/dev.c | 69 +---
net/core/dev.h | 4 +
net/core/dev_addr_lists.c | 373 ++++++++++++++++-
net/core/dev_addr_lists_test.c | 386 +++++++++++++++++-
net/core/dev_api.c | 3 +
net/core/dev_ioctl.c | 6 +-
net/core/rtnetlink.c | 1 +
.../selftests/drivers/net/bonding/lag_lib.sh | 17 +-
.../drivers/net/team/dev_addr_lists.sh | 2 -
tools/testing/selftests/net/config | 3 +
tools/testing/selftests/net/rtnetlink.sh | 44 ++
24 files changed, 975 insertions(+), 142 deletions(-)
--
2.53.0
^ permalink raw reply
* [PATCH net-next v4 01/14] net: add address list snapshot and reconciliation infrastructure
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni, Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Introduce __hw_addr_list_snapshot() and __hw_addr_list_reconcile()
for use by the upcoming ndo_set_rx_mode_async callback.
The async rx_mode path needs to snapshot the device's unicast and
multicast address lists under the addr_lock, hand those snapshots
to the driver (which may sleep), and then propagate any sync_cnt
changes back to the real lists. Two identical snapshots are taken:
a work copy for the driver to pass to __hw_addr_sync_dev() and a
reference copy to compute deltas against.
__hw_addr_list_reconcile() walks the reference snapshot comparing
each entry against the work snapshot to determine what the driver
synced or unsynced. It then applies those deltas to the real list,
handling concurrent modifications:
- If the real entry was concurrently removed but the driver synced
it to hardware (delta > 0), re-insert a stale entry so the next
work run properly unsyncs it from hardware.
- If the entry still exists, apply the delta normally. An entry
whose refcount drops to zero is removed.
# dev_addr_test_snapshot_benchmark: 1024 addrs x 1000 snapshots: 89872802 ns total, 89872 ns/iter
# dev_addr_test_snapshot_benchmark.speed: slow
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
include/linux/netdevice.h | 7 +
net/core/dev.h | 1 +
net/core/dev_addr_lists.c | 112 +++++++++-
net/core/dev_addr_lists_test.c | 362 ++++++++++++++++++++++++++++++++-
4 files changed, 479 insertions(+), 3 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index e15367373f7c..151d6f4fd9b3 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4995,6 +4995,13 @@ void __hw_addr_unsync_dev(struct netdev_hw_addr_list *list,
int (*unsync)(struct net_device *,
const unsigned char *));
void __hw_addr_init(struct netdev_hw_addr_list *list);
+void __hw_addr_flush(struct netdev_hw_addr_list *list);
+int __hw_addr_list_snapshot(struct netdev_hw_addr_list *snap,
+ const struct netdev_hw_addr_list *list,
+ int addr_len);
+void __hw_addr_list_reconcile(struct netdev_hw_addr_list *real_list,
+ struct netdev_hw_addr_list *work,
+ struct netdev_hw_addr_list *ref, int addr_len);
/* Functions used for device addresses handling */
void dev_addr_mod(struct net_device *dev, unsigned int offset,
diff --git a/net/core/dev.h b/net/core/dev.h
index 781619e76b3e..acc925b7b337 100644
--- a/net/core/dev.h
+++ b/net/core/dev.h
@@ -69,6 +69,7 @@ void linkwatch_run_queue(void);
void dev_addr_flush(struct net_device *dev);
int dev_addr_init(struct net_device *dev);
void dev_addr_check(struct net_device *dev);
+void __hw_addr_flush(struct netdev_hw_addr_list *list);
#if IS_ENABLED(CONFIG_NET_SHAPER)
void net_shaper_flush_netdev(struct net_device *dev);
diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index 76c91f224886..284fcfe9389b 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -11,6 +11,7 @@
#include <linux/rtnetlink.h>
#include <linux/export.h>
#include <linux/list.h>
+#include <kunit/visibility.h>
#include "dev.h"
@@ -481,7 +482,7 @@ void __hw_addr_unsync_dev(struct netdev_hw_addr_list *list,
}
EXPORT_SYMBOL(__hw_addr_unsync_dev);
-static void __hw_addr_flush(struct netdev_hw_addr_list *list)
+void __hw_addr_flush(struct netdev_hw_addr_list *list)
{
struct netdev_hw_addr *ha, *tmp;
@@ -492,6 +493,7 @@ static void __hw_addr_flush(struct netdev_hw_addr_list *list)
}
list->count = 0;
}
+EXPORT_SYMBOL_IF_KUNIT(__hw_addr_flush);
void __hw_addr_init(struct netdev_hw_addr_list *list)
{
@@ -501,6 +503,114 @@ void __hw_addr_init(struct netdev_hw_addr_list *list)
}
EXPORT_SYMBOL(__hw_addr_init);
+/**
+ * __hw_addr_list_snapshot - create a snapshot copy of an address list
+ * @snap: destination snapshot list (needs to be __hw_addr_init-initialized)
+ * @list: source address list to snapshot
+ * @addr_len: length of addresses
+ *
+ * Creates a copy of @list with individually allocated entries suitable
+ * for use with __hw_addr_sync_dev() and other list manipulation helpers.
+ * Each entry is allocated with GFP_ATOMIC; must be called under a spinlock.
+ *
+ * Return: 0 on success, -errno on failure.
+ */
+int __hw_addr_list_snapshot(struct netdev_hw_addr_list *snap,
+ const struct netdev_hw_addr_list *list,
+ int addr_len)
+{
+ struct netdev_hw_addr *ha, *entry;
+
+ list_for_each_entry(ha, &list->list, list) {
+ entry = __hw_addr_create(ha->addr, addr_len, ha->type,
+ false, false);
+ if (!entry) {
+ __hw_addr_flush(snap);
+ return -ENOMEM;
+ }
+ entry->sync_cnt = ha->sync_cnt;
+ entry->refcount = ha->refcount;
+
+ list_add_tail(&entry->list, &snap->list);
+ __hw_addr_insert(snap, entry, addr_len);
+ snap->count++;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL_IF_KUNIT(__hw_addr_list_snapshot);
+
+/**
+ * __hw_addr_list_reconcile - sync snapshot changes back and free snapshots
+ * @real_list: the real address list to update
+ * @work: the working snapshot (modified by driver via __hw_addr_sync_dev)
+ * @ref: the reference snapshot (untouched copy of original state)
+ * @addr_len: length of addresses
+ *
+ * Walks the reference snapshot and compares each entry against the work
+ * snapshot to compute sync_cnt deltas. Applies those deltas to @real_list.
+ * Frees both snapshots when done.
+ * Caller must hold netif_addr_lock_bh.
+ */
+void __hw_addr_list_reconcile(struct netdev_hw_addr_list *real_list,
+ struct netdev_hw_addr_list *work,
+ struct netdev_hw_addr_list *ref, int addr_len)
+{
+ struct netdev_hw_addr *ref_ha, *work_ha, *real_ha;
+ int delta;
+
+ list_for_each_entry(ref_ha, &ref->list, list) {
+ work_ha = __hw_addr_lookup(work, ref_ha->addr, addr_len,
+ ref_ha->type);
+ if (work_ha)
+ delta = work_ha->sync_cnt - ref_ha->sync_cnt;
+ else
+ delta = -1;
+
+ if (delta == 0)
+ continue;
+
+ real_ha = __hw_addr_lookup(real_list, ref_ha->addr, addr_len,
+ ref_ha->type);
+ if (!real_ha) {
+ /* The real entry was concurrently removed. If the
+ * driver synced this addr to hardware (delta > 0),
+ * re-insert it as a stale entry so the next work
+ * run unsyncs it from hardware.
+ */
+ if (delta > 0) {
+ real_ha = __hw_addr_create(ref_ha->addr,
+ addr_len,
+ ref_ha->type, false,
+ false);
+ if (real_ha) {
+ real_ha->sync_cnt = 1;
+ real_ha->refcount = 1;
+ list_add_tail_rcu(&real_ha->list,
+ &real_list->list);
+ __hw_addr_insert(real_list, real_ha,
+ addr_len);
+ real_list->count++;
+ }
+ }
+ continue;
+ }
+
+ real_ha->sync_cnt += delta;
+ real_ha->refcount += delta;
+ if (!real_ha->refcount) {
+ rb_erase(&real_ha->node, &real_list->tree);
+ list_del_rcu(&real_ha->list);
+ kfree_rcu(real_ha, rcu_head);
+ real_list->count--;
+ }
+ }
+
+ __hw_addr_flush(work);
+ __hw_addr_flush(ref);
+}
+EXPORT_SYMBOL_IF_KUNIT(__hw_addr_list_reconcile);
+
/*
* Device addresses handling functions
*/
diff --git a/net/core/dev_addr_lists_test.c b/net/core/dev_addr_lists_test.c
index 8e1dba825e94..4cbc241a9ec0 100644
--- a/net/core/dev_addr_lists_test.c
+++ b/net/core/dev_addr_lists_test.c
@@ -8,16 +8,24 @@
static const struct net_device_ops dummy_netdev_ops = {
};
+#define ADDR_A 1
+#define ADDR_B 2
+#define ADDR_C 3
+
struct dev_addr_test_priv {
u32 addr_seen;
+ u32 addr_synced;
+ u32 addr_unsynced;
};
static int dev_addr_test_sync(struct net_device *netdev, const unsigned char *a)
{
struct dev_addr_test_priv *datp = netdev_priv(netdev);
- if (a[0] < 31 && !memchr_inv(a, a[0], ETH_ALEN))
+ if (a[0] < 31 && !memchr_inv(a, a[0], ETH_ALEN)) {
datp->addr_seen |= 1 << a[0];
+ datp->addr_synced |= 1 << a[0];
+ }
return 0;
}
@@ -26,11 +34,22 @@ static int dev_addr_test_unsync(struct net_device *netdev,
{
struct dev_addr_test_priv *datp = netdev_priv(netdev);
- if (a[0] < 31 && !memchr_inv(a, a[0], ETH_ALEN))
+ if (a[0] < 31 && !memchr_inv(a, a[0], ETH_ALEN)) {
datp->addr_seen &= ~(1 << a[0]);
+ datp->addr_unsynced |= 1 << a[0];
+ }
return 0;
}
+static void dev_addr_test_reset(struct net_device *netdev)
+{
+ struct dev_addr_test_priv *datp = netdev_priv(netdev);
+
+ datp->addr_seen = 0;
+ datp->addr_synced = 0;
+ datp->addr_unsynced = 0;
+}
+
static int dev_addr_test_init(struct kunit *test)
{
struct dev_addr_test_priv *datp;
@@ -225,6 +244,339 @@ static void dev_addr_test_add_excl(struct kunit *test)
rtnl_unlock();
}
+/* Snapshot test: basic sync with no concurrent modifications.
+ * Add one address, snapshot, driver syncs it, reconcile propagates
+ * sync_cnt delta back to real list.
+ */
+static void dev_addr_test_snapshot_sync(struct kunit *test)
+{
+ struct net_device *netdev = test->priv;
+ struct netdev_hw_addr_list snap, ref;
+ struct dev_addr_test_priv *datp;
+ struct netdev_hw_addr *ha;
+ u8 addr[ETH_ALEN];
+
+ datp = netdev_priv(netdev);
+
+ rtnl_lock();
+
+ memset(addr, ADDR_A, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+
+ /* Snapshot: ADDR_A has sync_cnt=0, refcount=1 (new) */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_init(&snap);
+ __hw_addr_init(&ref);
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ netif_addr_unlock_bh(netdev);
+
+ /* Driver syncs ADDR_A to hardware */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&snap, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+
+ /* Reconcile: delta=+1 applied to real entry */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ netif_addr_unlock_bh(netdev);
+
+ /* Real entry should now reflect the sync: sync_cnt=1, refcount=2 */
+ KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+ ha = list_first_entry(&netdev->uc.list, struct netdev_hw_addr, list);
+ KUNIT_EXPECT_MEMEQ(test, ha->addr, addr, ETH_ALEN);
+ KUNIT_EXPECT_EQ(test, 1, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 2, ha->refcount);
+
+ /* Second work run: already synced, nothing to do */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+ KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+
+ rtnl_unlock();
+}
+
+/* Snapshot test: ADDR_A synced to hardware, then concurrently removed
+ * from the real list before reconcile runs. Reconcile re-inserts ADDR_A as
+ * a stale entry so the next work run unsyncs it from hardware.
+ */
+static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
+{
+ struct net_device *netdev = test->priv;
+ struct netdev_hw_addr_list snap, ref;
+ struct dev_addr_test_priv *datp;
+ struct netdev_hw_addr *ha;
+ u8 addr[ETH_ALEN];
+
+ datp = netdev_priv(netdev);
+
+ rtnl_lock();
+
+ memset(addr, ADDR_A, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+
+ /* Snapshot: ADDR_A is new (sync_cnt=0, refcount=1) */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_init(&snap);
+ __hw_addr_init(&ref);
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ netif_addr_unlock_bh(netdev);
+
+ /* Driver syncs ADDR_A to hardware */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&snap, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+
+ /* Concurrent removal: user deletes ADDR_A while driver was working */
+ memset(addr, ADDR_A, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_del(netdev, addr));
+ KUNIT_EXPECT_EQ(test, 0, netdev->uc.count);
+
+ /* Reconcile: ADDR_A gone from real list but driver synced it,
+ * so it gets re-inserted as stale (sync_cnt=1, refcount=1).
+ */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ netif_addr_unlock_bh(netdev);
+
+ KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+ ha = list_first_entry(&netdev->uc.list, struct netdev_hw_addr, list);
+ KUNIT_EXPECT_MEMEQ(test, ha->addr, addr, ETH_ALEN);
+ KUNIT_EXPECT_EQ(test, 1, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 1, ha->refcount);
+
+ /* Second work run: stale entry gets unsynced from HW and removed */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_unsynced);
+ KUNIT_EXPECT_EQ(test, 0, netdev->uc.count);
+
+ rtnl_unlock();
+}
+
+/* Snapshot test: ADDR_A was stale (unsynced from hardware by driver),
+ * but concurrently re-added by the user. The re-add bumps refcount of
+ * the existing stale entry. Reconcile applies delta=-1, leaving ADDR_A
+ * as a fresh entry (sync_cnt=0, refcount=1) for the next work run.
+ */
+static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
+{
+ struct net_device *netdev = test->priv;
+ struct netdev_hw_addr_list snap, ref;
+ struct dev_addr_test_priv *datp;
+ struct netdev_hw_addr *ha;
+ u8 addr[ETH_ALEN];
+
+ datp = netdev_priv(netdev);
+
+ rtnl_lock();
+
+ memset(addr, ADDR_A, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+
+ /* Sync ADDR_A to hardware: sync_cnt=1, refcount=2 */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+
+ /* User removes ADDR_A: refcount=1, sync_cnt=1 -> stale */
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_del(netdev, addr));
+
+ /* Snapshot: ADDR_A is stale (sync_cnt=1, refcount=1) */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_init(&snap);
+ __hw_addr_init(&ref);
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ netif_addr_unlock_bh(netdev);
+
+ /* Driver unsyncs stale ADDR_A from hardware */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&snap, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_unsynced);
+
+ /* Concurrent: user re-adds ADDR_A. dev_uc_add finds the existing
+ * stale entry and bumps refcount from 1 -> 2. sync_cnt stays 1.
+ */
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+ KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+
+ /* Reconcile: ref sync_cnt=1 matches real sync_cnt=1, delta=-1
+ * applied. Result: sync_cnt=0, refcount=1 (fresh).
+ */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ netif_addr_unlock_bh(netdev);
+
+ /* Entry survives as fresh: needs re-sync to HW */
+ KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+ ha = list_first_entry(&netdev->uc.list, struct netdev_hw_addr, list);
+ KUNIT_EXPECT_MEMEQ(test, ha->addr, addr, ETH_ALEN);
+ KUNIT_EXPECT_EQ(test, 0, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 1, ha->refcount);
+
+ /* Second work run: fresh entry gets synced to HW */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+
+ rtnl_unlock();
+}
+
+/* Snapshot test: ADDR_A is new (synced by driver), and independent ADDR_B
+ * is concurrently removed from the real list. A's sync delta propagates
+ * normally; B's absence doesn't interfere.
+ */
+static void dev_addr_test_snapshot_add_and_remove(struct kunit *test)
+{
+ struct net_device *netdev = test->priv;
+ struct netdev_hw_addr_list snap, ref;
+ struct dev_addr_test_priv *datp;
+ struct netdev_hw_addr *ha;
+ u8 addr[ETH_ALEN];
+
+ datp = netdev_priv(netdev);
+
+ rtnl_lock();
+
+ /* Add ADDR_A and ADDR_B (will be synced then removed) */
+ memset(addr, ADDR_A, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+ memset(addr, ADDR_B, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+
+ /* Sync both to hardware: sync_cnt=1, refcount=2 */
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+
+ /* Add ADDR_C (new, will be synced by snapshot) */
+ memset(addr, ADDR_C, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_add(netdev, addr));
+
+ /* Snapshot: A,B synced (sync_cnt=1,refcount=2); C new (0,1) */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_init(&snap);
+ __hw_addr_init(&ref);
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ netif_addr_unlock_bh(netdev);
+
+ /* Driver syncs snapshot: ADDR_C is new -> synced; A,B already synced */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&snap, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_C, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+
+ /* Concurrent: user removes addr B while driver was working */
+ memset(addr, ADDR_B, sizeof(addr));
+ KUNIT_EXPECT_EQ(test, 0, dev_uc_del(netdev, addr));
+
+ /* Reconcile: ADDR_C's delta=+1 applied to real list.
+ * ADDR_B's delta=0 (unchanged in snapshot),
+ * so nothing to apply to ADDR_B.
+ */
+ netif_addr_lock_bh(netdev);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ netif_addr_unlock_bh(netdev);
+
+ /* ADDR_A: unchanged (sync_cnt=1, refcount=2)
+ * ADDR_B: refcount went from 2->1 via dev_uc_del (still present, stale)
+ * ADDR_C: sync propagated (sync_cnt=1, refcount=2)
+ */
+ KUNIT_EXPECT_EQ(test, 3, netdev->uc.count);
+ netdev_hw_addr_list_for_each(ha, &netdev->uc) {
+ u8 id = ha->addr[0];
+
+ if (!memchr_inv(ha->addr, id, ETH_ALEN)) {
+ if (id == ADDR_A) {
+ KUNIT_EXPECT_EQ(test, 1, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 2, ha->refcount);
+ } else if (id == ADDR_B) {
+ /* B: still present but now stale */
+ KUNIT_EXPECT_EQ(test, 1, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 1, ha->refcount);
+ } else if (id == ADDR_C) {
+ KUNIT_EXPECT_EQ(test, 1, ha->sync_cnt);
+ KUNIT_EXPECT_EQ(test, 2, ha->refcount);
+ }
+ }
+ }
+
+ /* Second work run: ADDR_B is stale, gets unsynced and removed */
+ dev_addr_test_reset(netdev);
+ __hw_addr_sync_dev(&netdev->uc, netdev, dev_addr_test_sync,
+ dev_addr_test_unsync);
+ KUNIT_EXPECT_EQ(test, 0, datp->addr_synced);
+ KUNIT_EXPECT_EQ(test, 1 << ADDR_B, datp->addr_unsynced);
+ KUNIT_EXPECT_EQ(test, 2, netdev->uc.count);
+
+ rtnl_unlock();
+}
+
+static void dev_addr_test_snapshot_benchmark(struct kunit *test)
+{
+ struct net_device *netdev = test->priv;
+ struct netdev_hw_addr_list snap;
+ s64 duration = 0;
+ ktime_t start;
+ u8 addr[ETH_ALEN];
+ int i, iter;
+
+ rtnl_lock();
+
+ for (i = 0; i < 1024; i++) {
+ memset(addr, 0, sizeof(addr));
+ addr[0] = (i >> 8) & 0xff;
+ addr[1] = i & 0xff;
+ KUNIT_ASSERT_EQ(test, 0, dev_uc_add(netdev, addr));
+ }
+
+ for (iter = 0; iter < 1000; iter++) {
+ netif_addr_lock_bh(netdev);
+ __hw_addr_init(&snap);
+
+ start = ktime_get();
+ KUNIT_ASSERT_EQ(test, 0,
+ __hw_addr_list_snapshot(&snap, &netdev->uc,
+ ETH_ALEN));
+ duration += ktime_to_ns(ktime_sub(ktime_get(), start));
+
+ netif_addr_unlock_bh(netdev);
+ __hw_addr_flush(&snap);
+ }
+
+ kunit_info(test,
+ "1024 addrs x 1000 snapshots: %lld ns total, %lld ns/iter",
+ duration, duration / 1000);
+
+ rtnl_unlock();
+}
+
static struct kunit_case dev_addr_test_cases[] = {
KUNIT_CASE(dev_addr_test_basic),
KUNIT_CASE(dev_addr_test_sync_one),
@@ -232,6 +584,11 @@ static struct kunit_case dev_addr_test_cases[] = {
KUNIT_CASE(dev_addr_test_del_main),
KUNIT_CASE(dev_addr_test_add_set),
KUNIT_CASE(dev_addr_test_add_excl),
+ KUNIT_CASE(dev_addr_test_snapshot_sync),
+ KUNIT_CASE(dev_addr_test_snapshot_remove_during_sync),
+ KUNIT_CASE(dev_addr_test_snapshot_readd_during_unsync),
+ KUNIT_CASE(dev_addr_test_snapshot_add_and_remove),
+ KUNIT_CASE_SLOW(dev_addr_test_snapshot_benchmark),
{}
};
@@ -243,5 +600,6 @@ static struct kunit_suite dev_addr_test_suite = {
};
kunit_test_suite(dev_addr_test_suite);
+MODULE_IMPORT_NS("EXPORTED_FOR_KUNIT_TESTING");
MODULE_DESCRIPTION("KUnit tests for struct netdev_hw_addr_list");
MODULE_LICENSE("GPL");
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 02/14] net: introduce ndo_set_rx_mode_async and netdev_rx_mode_work
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Add ndo_set_rx_mode_async callback that drivers can implement instead
of the legacy ndo_set_rx_mode. The legacy callback runs under the
netif_addr_lock spinlock with BHs disabled, preventing drivers from
sleeping. The async variant runs from a work queue with rtnl_lock and
netdev_lock_ops held, in fully sleepable context.
When __dev_set_rx_mode() sees ndo_set_rx_mode_async, it schedules
netdev_rx_mode_work instead of calling the driver inline. The work
function takes two snapshots of each address list (uc/mc) under
the addr_lock, then drops the lock and calls the driver with the
work copies. After the driver returns, it reconciles the snapshots
back to the real lists under the lock.
Add netif_rx_mode_sync() to opportunistically execute the pending
workqueue update inline, so that rx mode changes are committed
before returning to userspace:
- dev_change_flags (SIOCSIFFLAGS / RTM_NEWLINK)
- dev_set_promiscuity
- dev_set_allmulti
- dev_ifsioc SIOCADDMULTI / SIOCDELMULTI
- do_setlink (RTM_SETLINK)
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
Documentation/networking/netdevices.rst | 9 ++
include/linux/netdevice.h | 16 ++
net/core/dev.c | 46 +-----
net/core/dev.h | 3 +
net/core/dev_addr_lists.c | 194 ++++++++++++++++++++++++
net/core/dev_api.c | 3 +
net/core/dev_ioctl.c | 6 +-
net/core/rtnetlink.c | 1 +
8 files changed, 235 insertions(+), 43 deletions(-)
diff --git a/Documentation/networking/netdevices.rst b/Documentation/networking/netdevices.rst
index 35704d115312..8a488c21fd7c 100644
--- a/Documentation/networking/netdevices.rst
+++ b/Documentation/networking/netdevices.rst
@@ -289,6 +289,15 @@ struct net_device synchronization rules
ndo_set_rx_mode:
Synchronization: netif_addr_lock spinlock.
Context: BHs disabled
+ Notes: Deprecated in favor of ndo_set_rx_mode_async which runs
+ in process context.
+
+ndo_set_rx_mode_async:
+ Synchronization: rtnl_lock() semaphore. In addition, netdev instance
+ lock if the driver implements queue management or shaper API.
+ Context: process (from a work queue)
+ Notes: Async version of ndo_set_rx_mode which runs in process
+ context. Receives snapshots of the unicast and multicast address lists.
ndo_setup_tc:
``TC_SETUP_BLOCK`` and ``TC_SETUP_FT`` are running under NFT locks
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 151d6f4fd9b3..cf3dd3f1eb0f 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1119,6 +1119,16 @@ struct netdev_net_notifier {
* This function is called device changes address list filtering.
* If driver handles unicast address filtering, it should set
* IFF_UNICAST_FLT in its priv_flags.
+ * Cannot sleep, called with netif_addr_lock_bh held.
+ * Deprecated in favor of ndo_set_rx_mode_async.
+ *
+ * void (*ndo_set_rx_mode_async)(struct net_device *dev,
+ * struct netdev_hw_addr_list *uc,
+ * struct netdev_hw_addr_list *mc);
+ * Async version of ndo_set_rx_mode which runs in process context
+ * with rtnl_lock and netdev_lock_ops(dev) held. The uc/mc parameters
+ * are snapshots of the address lists - iterate with
+ * netdev_hw_addr_list_for_each(ha, uc).
*
* int (*ndo_set_mac_address)(struct net_device *dev, void *addr);
* This function is called when the Media Access Control address
@@ -1439,6 +1449,10 @@ struct net_device_ops {
void (*ndo_change_rx_flags)(struct net_device *dev,
int flags);
void (*ndo_set_rx_mode)(struct net_device *dev);
+ void (*ndo_set_rx_mode_async)(
+ struct net_device *dev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc);
int (*ndo_set_mac_address)(struct net_device *dev,
void *addr);
int (*ndo_validate_addr)(struct net_device *dev);
@@ -1903,6 +1917,7 @@ enum netdev_reg_state {
* has been enabled due to the need to listen to
* additional unicast addresses in a device that
* does not implement ndo_set_rx_mode()
+ * @rx_mode_node: List entry for rx_mode work processing
* @uc: unicast mac addresses
* @mc: multicast mac addresses
* @dev_addrs: list of device hw addresses
@@ -2294,6 +2309,7 @@ struct net_device {
unsigned int promiscuity;
unsigned int allmulti;
bool uc_promisc;
+ struct list_head rx_mode_node;
#ifdef CONFIG_LOCKDEP
unsigned char nested_level;
#endif
diff --git a/net/core/dev.c b/net/core/dev.c
index 90d041390e29..b9f9e6a30963 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -9574,7 +9574,7 @@ static void dev_change_rx_flags(struct net_device *dev, int flags)
ops->ndo_change_rx_flags(dev, flags);
}
-static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
+int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
{
unsigned int old_flags = dev->flags;
unsigned int promiscuity, flags;
@@ -9678,46 +9678,6 @@ int netif_set_allmulti(struct net_device *dev, int inc, bool notify)
return 0;
}
-/*
- * Upload unicast and multicast address lists to device and
- * configure RX filtering. When the device doesn't support unicast
- * filtering it is put in promiscuous mode while unicast addresses
- * are present.
- */
-void __dev_set_rx_mode(struct net_device *dev)
-{
- const struct net_device_ops *ops = dev->netdev_ops;
-
- /* dev_open will call this function so the list will stay sane. */
- if (!(dev->flags&IFF_UP))
- return;
-
- if (!netif_device_present(dev))
- return;
-
- if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
- /* Unicast addresses changes may only happen under the rtnl,
- * therefore calling __dev_set_promiscuity here is safe.
- */
- if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
- __dev_set_promiscuity(dev, 1, false);
- dev->uc_promisc = true;
- } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
- __dev_set_promiscuity(dev, -1, false);
- dev->uc_promisc = false;
- }
- }
-
- if (ops->ndo_set_rx_mode)
- ops->ndo_set_rx_mode(dev);
-}
-
-void dev_set_rx_mode(struct net_device *dev)
-{
- netif_addr_lock_bh(dev);
- __dev_set_rx_mode(dev);
- netif_addr_unlock_bh(dev);
-}
/**
* netif_get_flags() - get flags reported to userspace
@@ -12108,6 +12068,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
#endif
mutex_init(&dev->lock);
+ INIT_LIST_HEAD(&dev->rx_mode_node);
dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
setup(dev);
@@ -12212,6 +12173,9 @@ void free_netdev(struct net_device *dev)
kfree(rcu_dereference_protected(dev->ingress_queue, 1));
+ if (netif_rx_mode_clean(dev))
+ dev_put(dev);
+
/* Flush device addresses */
dev_addr_flush(dev);
diff --git a/net/core/dev.h b/net/core/dev.h
index acc925b7b337..50edb380ca94 100644
--- a/net/core/dev.h
+++ b/net/core/dev.h
@@ -146,6 +146,9 @@ int netif_change_carrier(struct net_device *dev, bool new_carrier);
int dev_change_carrier(struct net_device *dev, bool new_carrier);
void __dev_set_rx_mode(struct net_device *dev);
+int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify);
+bool netif_rx_mode_clean(struct net_device *dev);
+void netif_rx_mode_sync(struct net_device *dev);
void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
unsigned int gchanges, u32 portid,
diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index 284fcfe9389b..a05db3ac43ea 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -11,10 +11,18 @@
#include <linux/rtnetlink.h>
#include <linux/export.h>
#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
#include <kunit/visibility.h>
#include "dev.h"
+static void netdev_rx_mode_work(struct work_struct *work);
+
+static LIST_HEAD(rx_mode_list);
+static DEFINE_SPINLOCK(rx_mode_lock);
+static DECLARE_WORK(rx_mode_work, netdev_rx_mode_work);
+
/*
* General list handling functions
*/
@@ -1159,3 +1167,189 @@ void dev_mc_init(struct net_device *dev)
__hw_addr_init(&dev->mc);
}
EXPORT_SYMBOL(dev_mc_init);
+
+static int netif_addr_lists_snapshot(struct net_device *dev,
+ struct netdev_hw_addr_list *uc_snap,
+ struct netdev_hw_addr_list *mc_snap,
+ struct netdev_hw_addr_list *uc_ref,
+ struct netdev_hw_addr_list *mc_ref)
+{
+ int err;
+
+ err = __hw_addr_list_snapshot(uc_snap, &dev->uc, dev->addr_len);
+ if (!err)
+ err = __hw_addr_list_snapshot(uc_ref, &dev->uc, dev->addr_len);
+ if (!err)
+ err = __hw_addr_list_snapshot(mc_snap, &dev->mc,
+ dev->addr_len);
+ if (!err)
+ err = __hw_addr_list_snapshot(mc_ref, &dev->mc, dev->addr_len);
+
+ if (err) {
+ __hw_addr_flush(uc_snap);
+ __hw_addr_flush(uc_ref);
+ __hw_addr_flush(mc_snap);
+ }
+
+ return err;
+}
+
+static void netif_addr_lists_reconcile(struct net_device *dev,
+ struct netdev_hw_addr_list *uc_snap,
+ struct netdev_hw_addr_list *mc_snap,
+ struct netdev_hw_addr_list *uc_ref,
+ struct netdev_hw_addr_list *mc_ref)
+{
+ __hw_addr_list_reconcile(&dev->uc, uc_snap, uc_ref, dev->addr_len);
+ __hw_addr_list_reconcile(&dev->mc, mc_snap, mc_ref, dev->addr_len);
+}
+
+static void netif_rx_mode_run(struct net_device *dev)
+{
+ struct netdev_hw_addr_list uc_snap, mc_snap, uc_ref, mc_ref;
+ const struct net_device_ops *ops = dev->netdev_ops;
+ int err;
+
+ might_sleep();
+ netdev_ops_assert_locked(dev);
+
+ __hw_addr_init(&uc_snap);
+ __hw_addr_init(&mc_snap);
+ __hw_addr_init(&uc_ref);
+ __hw_addr_init(&mc_ref);
+
+ if (!(dev->flags & IFF_UP) || !netif_device_present(dev))
+ return;
+
+ netif_addr_lock_bh(dev);
+ err = netif_addr_lists_snapshot(dev, &uc_snap, &mc_snap,
+ &uc_ref, &mc_ref);
+ if (err) {
+ netdev_WARN(dev, "failed to sync uc/mc addresses\n");
+ netif_addr_unlock_bh(dev);
+ return;
+ }
+ netif_addr_unlock_bh(dev);
+
+ ops->ndo_set_rx_mode_async(dev, &uc_snap, &mc_snap);
+
+ netif_addr_lock_bh(dev);
+ netif_addr_lists_reconcile(dev, &uc_snap, &mc_snap,
+ &uc_ref, &mc_ref);
+ netif_addr_unlock_bh(dev);
+}
+
+static void netdev_rx_mode_work(struct work_struct *work)
+{
+ struct net_device *dev;
+
+ rtnl_lock();
+
+ while (true) {
+ spin_lock_bh(&rx_mode_lock);
+ if (list_empty(&rx_mode_list)) {
+ spin_unlock_bh(&rx_mode_lock);
+ break;
+ }
+ dev = list_first_entry(&rx_mode_list, struct net_device,
+ rx_mode_node);
+ list_del_init(&dev->rx_mode_node);
+ spin_unlock_bh(&rx_mode_lock);
+
+ netdev_lock_ops(dev);
+ netif_rx_mode_run(dev);
+ netdev_unlock_ops(dev);
+ dev_put(dev);
+ }
+
+ rtnl_unlock();
+}
+
+static void netif_rx_mode_queue(struct net_device *dev)
+{
+ spin_lock_bh(&rx_mode_lock);
+ if (list_empty(&dev->rx_mode_node)) {
+ list_add_tail(&dev->rx_mode_node, &rx_mode_list);
+ dev_hold(dev);
+ }
+ spin_unlock_bh(&rx_mode_lock);
+ schedule_work(&rx_mode_work);
+}
+
+/**
+ * __dev_set_rx_mode() - upload unicast and multicast address lists to device
+ * and configure RX filtering.
+ * @dev: device
+ *
+ * When the device doesn't support unicast filtering it is put in promiscuous
+ * mode while unicast addresses are present.
+ */
+void __dev_set_rx_mode(struct net_device *dev)
+{
+ const struct net_device_ops *ops = dev->netdev_ops;
+
+ /* dev_open will call this function so the list will stay sane. */
+ if (!(dev->flags & IFF_UP))
+ return;
+
+ if (!netif_device_present(dev))
+ return;
+
+ if (ops->ndo_set_rx_mode_async) {
+ netif_rx_mode_queue(dev);
+ return;
+ }
+
+ if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
+ if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
+ __dev_set_promiscuity(dev, 1, false);
+ dev->uc_promisc = true;
+ } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
+ __dev_set_promiscuity(dev, -1, false);
+ dev->uc_promisc = false;
+ }
+ }
+
+ if (ops->ndo_set_rx_mode)
+ ops->ndo_set_rx_mode(dev);
+}
+
+void dev_set_rx_mode(struct net_device *dev)
+{
+ netif_addr_lock_bh(dev);
+ __dev_set_rx_mode(dev);
+ netif_addr_unlock_bh(dev);
+}
+
+bool netif_rx_mode_clean(struct net_device *dev)
+{
+ bool clean = false;
+
+ spin_lock_bh(&rx_mode_lock);
+ if (!list_empty(&dev->rx_mode_node)) {
+ list_del_init(&dev->rx_mode_node);
+ clean = true;
+ }
+ spin_unlock_bh(&rx_mode_lock);
+
+ return clean;
+}
+
+/**
+ * netif_rx_mode_sync() - sync rx mode inline
+ * @dev: network device
+ *
+ * Drivers implementing ndo_set_rx_mode_async() have their rx mode callback
+ * executed from a workqueue. This allows the callback to sleep, but means
+ * the hardware update is deferred and may not be visible to userspace
+ * by the time the initiating syscall returns. netif_rx_mode_sync() steals
+ * workqueue update and executes it inline. This preserves the atomicity of
+ * operations to the userspace.
+ */
+void netif_rx_mode_sync(struct net_device *dev)
+{
+ if (netif_rx_mode_clean(dev)) {
+ netif_rx_mode_run(dev);
+ dev_put(dev);
+ }
+}
diff --git a/net/core/dev_api.c b/net/core/dev_api.c
index f28852078aa6..437947dd08ed 100644
--- a/net/core/dev_api.c
+++ b/net/core/dev_api.c
@@ -66,6 +66,7 @@ int dev_change_flags(struct net_device *dev, unsigned int flags,
netdev_lock_ops(dev);
ret = netif_change_flags(dev, flags, extack);
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return ret;
@@ -285,6 +286,7 @@ int dev_set_promiscuity(struct net_device *dev, int inc)
netdev_lock_ops(dev);
ret = netif_set_promiscuity(dev, inc);
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return ret;
@@ -311,6 +313,7 @@ int dev_set_allmulti(struct net_device *dev, int inc)
netdev_lock_ops(dev);
ret = netif_set_allmulti(dev, inc, true);
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return ret;
diff --git a/net/core/dev_ioctl.c b/net/core/dev_ioctl.c
index 7a8966544c9d..f3979b276090 100644
--- a/net/core/dev_ioctl.c
+++ b/net/core/dev_ioctl.c
@@ -586,24 +586,26 @@ static int dev_ifsioc(struct net *net, struct ifreq *ifr, void __user *data,
return err;
case SIOCADDMULTI:
- if (!ops->ndo_set_rx_mode ||
+ if ((!ops->ndo_set_rx_mode && !ops->ndo_set_rx_mode_async) ||
ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
return -EINVAL;
if (!netif_device_present(dev))
return -ENODEV;
netdev_lock_ops(dev);
err = dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return err;
case SIOCDELMULTI:
- if (!ops->ndo_set_rx_mode ||
+ if ((!ops->ndo_set_rx_mode && !ops->ndo_set_rx_mode_async) ||
ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
return -EINVAL;
if (!netif_device_present(dev))
return -ENODEV;
netdev_lock_ops(dev);
err = dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return err;
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index fae8034efbff..f4e5ac70709d 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -3431,6 +3431,7 @@ static int do_setlink(const struct sk_buff *skb, struct net_device *dev,
dev->name);
}
+ netif_rx_mode_sync(dev);
netdev_unlock_ops(dev);
return err;
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 03/14] net: move promiscuity handling into netdev_rx_mode_work
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni, Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Move unicast promiscuity tracking into netdev_rx_mode_work so it runs
under netdev_ops_lock instead of under the addr_lock spinlock. This
is required because __dev_set_promiscuity calls dev_change_rx_flags
and __dev_notify_flags, both of which may need to sleep.
Change ASSERT_RTNL() to netdev_ops_assert_locked() in
__dev_set_promiscuity, netif_set_allmulti and __dev_change_flags
since these are now called from the work queue under the ops lock.
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
Documentation/networking/netdevices.rst | 4 ++
net/core/dev.c | 16 ++---
net/core/dev_addr_lists.c | 82 ++++++++++++++++++-------
3 files changed, 68 insertions(+), 34 deletions(-)
diff --git a/Documentation/networking/netdevices.rst b/Documentation/networking/netdevices.rst
index 8a488c21fd7c..390537323457 100644
--- a/Documentation/networking/netdevices.rst
+++ b/Documentation/networking/netdevices.rst
@@ -299,6 +299,10 @@ struct net_device synchronization rules
Notes: Async version of ndo_set_rx_mode which runs in process
context. Receives snapshots of the unicast and multicast address lists.
+ndo_change_rx_flags:
+ Synchronization: rtnl_lock() semaphore. In addition, netdev instance
+ lock if the driver implements queue management or shaper API.
+
ndo_setup_tc:
``TC_SETUP_BLOCK`` and ``TC_SETUP_FT`` are running under NFT locks
(i.e. no ``rtnl_lock`` and no device instance lock). The rest of
diff --git a/net/core/dev.c b/net/core/dev.c
index b9f9e6a30963..ae1fd30d1ef9 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -9581,7 +9581,7 @@ int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
kuid_t uid;
kgid_t gid;
- ASSERT_RTNL();
+ netdev_ops_assert_locked(dev);
promiscuity = dev->promiscuity + inc;
if (promiscuity == 0) {
@@ -9617,16 +9617,8 @@ int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
dev_change_rx_flags(dev, IFF_PROMISC);
}
- if (notify) {
- /* The ops lock is only required to ensure consistent locking
- * for `NETDEV_CHANGE` notifiers. This function is sometimes
- * called without the lock, even for devices that are ops
- * locked, such as in `dev_uc_sync_multiple` when using
- * bonding or teaming.
- */
- netdev_ops_assert_locked(dev);
+ if (notify)
__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
- }
return 0;
}
@@ -9648,7 +9640,7 @@ int netif_set_allmulti(struct net_device *dev, int inc, bool notify)
unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
unsigned int allmulti, flags;
- ASSERT_RTNL();
+ netdev_ops_assert_locked(dev);
allmulti = dev->allmulti + inc;
if (allmulti == 0) {
@@ -9716,7 +9708,7 @@ int __dev_change_flags(struct net_device *dev, unsigned int flags,
unsigned int old_flags = dev->flags;
int ret;
- ASSERT_RTNL();
+ netdev_ops_assert_locked(dev);
/*
* Set the flags on our device.
diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index a05db3ac43ea..61589537b2d3 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -1204,10 +1204,34 @@ static void netif_addr_lists_reconcile(struct net_device *dev,
__hw_addr_list_reconcile(&dev->mc, mc_snap, mc_ref, dev->addr_len);
}
+/**
+ * netif_uc_promisc_update() - evaluate whether uc_promisc should be toggled.
+ * @dev: device
+ *
+ * Must be called under netif_addr_lock_bh.
+ * Return: +1 to enter promisc, -1 to leave, 0 for no change.
+ */
+static int netif_uc_promisc_update(struct net_device *dev)
+{
+ if (dev->priv_flags & IFF_UNICAST_FLT)
+ return 0;
+
+ if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
+ dev->uc_promisc = true;
+ return 1;
+ }
+ if (netdev_uc_empty(dev) && dev->uc_promisc) {
+ dev->uc_promisc = false;
+ return -1;
+ }
+ return 0;
+}
+
static void netif_rx_mode_run(struct net_device *dev)
{
struct netdev_hw_addr_list uc_snap, mc_snap, uc_ref, mc_ref;
const struct net_device_ops *ops = dev->netdev_ops;
+ int promisc_inc;
int err;
might_sleep();
@@ -1221,22 +1245,39 @@ static void netif_rx_mode_run(struct net_device *dev)
if (!(dev->flags & IFF_UP) || !netif_device_present(dev))
return;
- netif_addr_lock_bh(dev);
- err = netif_addr_lists_snapshot(dev, &uc_snap, &mc_snap,
- &uc_ref, &mc_ref);
- if (err) {
- netdev_WARN(dev, "failed to sync uc/mc addresses\n");
+ if (ops->ndo_set_rx_mode_async) {
+ netif_addr_lock_bh(dev);
+ err = netif_addr_lists_snapshot(dev, &uc_snap, &mc_snap,
+ &uc_ref, &mc_ref);
+ if (err) {
+ netdev_WARN(dev, "failed to sync uc/mc addresses\n");
+ netif_addr_unlock_bh(dev);
+ return;
+ }
+
+ promisc_inc = netif_uc_promisc_update(dev);
+ netif_addr_unlock_bh(dev);
+ } else {
+ netif_addr_lock_bh(dev);
+ promisc_inc = netif_uc_promisc_update(dev);
netif_addr_unlock_bh(dev);
- return;
}
- netif_addr_unlock_bh(dev);
- ops->ndo_set_rx_mode_async(dev, &uc_snap, &mc_snap);
+ if (promisc_inc)
+ __dev_set_promiscuity(dev, promisc_inc, false);
- netif_addr_lock_bh(dev);
- netif_addr_lists_reconcile(dev, &uc_snap, &mc_snap,
- &uc_ref, &mc_ref);
- netif_addr_unlock_bh(dev);
+ if (ops->ndo_set_rx_mode_async) {
+ ops->ndo_set_rx_mode_async(dev, &uc_snap, &mc_snap);
+
+ netif_addr_lock_bh(dev);
+ netif_addr_lists_reconcile(dev, &uc_snap, &mc_snap,
+ &uc_ref, &mc_ref);
+ netif_addr_unlock_bh(dev);
+ } else if (ops->ndo_set_rx_mode) {
+ netif_addr_lock_bh(dev);
+ ops->ndo_set_rx_mode(dev);
+ netif_addr_unlock_bh(dev);
+ }
}
static void netdev_rx_mode_work(struct work_struct *work)
@@ -1287,6 +1328,7 @@ static void netif_rx_mode_queue(struct net_device *dev)
void __dev_set_rx_mode(struct net_device *dev)
{
const struct net_device_ops *ops = dev->netdev_ops;
+ int promisc_inc;
/* dev_open will call this function so the list will stay sane. */
if (!(dev->flags & IFF_UP))
@@ -1295,20 +1337,16 @@ void __dev_set_rx_mode(struct net_device *dev)
if (!netif_device_present(dev))
return;
- if (ops->ndo_set_rx_mode_async) {
+ if (ops->ndo_set_rx_mode_async || ops->ndo_change_rx_flags) {
netif_rx_mode_queue(dev);
return;
}
- if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
- if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
- __dev_set_promiscuity(dev, 1, false);
- dev->uc_promisc = true;
- } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
- __dev_set_promiscuity(dev, -1, false);
- dev->uc_promisc = false;
- }
- }
+ /* Legacy path for non-ops-locked HW devices. */
+
+ promisc_inc = netif_uc_promisc_update(dev);
+ if (promisc_inc)
+ __dev_set_promiscuity(dev, promisc_inc, false);
if (ops->ndo_set_rx_mode)
ops->ndo_set_rx_mode(dev);
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 04/14] net: cache snapshot entries for ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Add a per-device netdev_hw_addr_list cache (rx_mode_addr_cache) that
allows __hw_addr_list_snapshot() and __hw_addr_list_reconcile() to
reuse previously allocated entries instead of hitting GFP_ATOMIC on
every snapshot cycle.
snapshot pops entries from the cache when available, falling back to
__hw_addr_create(). reconcile splices both snapshot lists back into
the cache via __hw_addr_splice(). The cache is flushed in
free_netdev().
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
include/linux/netdevice.h | 7 ++--
net/core/dev.c | 2 ++
net/core/dev_addr_lists.c | 66 ++++++++++++++++++++++++----------
net/core/dev_addr_lists_test.c | 60 +++++++++++++++++++++----------
4 files changed, 96 insertions(+), 39 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index cf3dd3f1eb0f..8f8fa72c2c64 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1918,6 +1918,7 @@ enum netdev_reg_state {
* additional unicast addresses in a device that
* does not implement ndo_set_rx_mode()
* @rx_mode_node: List entry for rx_mode work processing
+ * @rx_mode_addr_cache: Recycled snapshot entries for rx_mode work
* @uc: unicast mac addresses
* @mc: multicast mac addresses
* @dev_addrs: list of device hw addresses
@@ -2310,6 +2311,7 @@ struct net_device {
unsigned int allmulti;
bool uc_promisc;
struct list_head rx_mode_node;
+ struct netdev_hw_addr_list rx_mode_addr_cache;
#ifdef CONFIG_LOCKDEP
unsigned char nested_level;
#endif
@@ -5014,10 +5016,11 @@ void __hw_addr_init(struct netdev_hw_addr_list *list);
void __hw_addr_flush(struct netdev_hw_addr_list *list);
int __hw_addr_list_snapshot(struct netdev_hw_addr_list *snap,
const struct netdev_hw_addr_list *list,
- int addr_len);
+ int addr_len, struct netdev_hw_addr_list *cache);
void __hw_addr_list_reconcile(struct netdev_hw_addr_list *real_list,
struct netdev_hw_addr_list *work,
- struct netdev_hw_addr_list *ref, int addr_len);
+ struct netdev_hw_addr_list *ref, int addr_len,
+ struct netdev_hw_addr_list *cache);
/* Functions used for device addresses handling */
void dev_addr_mod(struct net_device *dev, unsigned int offset,
diff --git a/net/core/dev.c b/net/core/dev.c
index ae1fd30d1ef9..3ddf347dcdd7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -12061,6 +12061,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
mutex_init(&dev->lock);
INIT_LIST_HEAD(&dev->rx_mode_node);
+ __hw_addr_init(&dev->rx_mode_addr_cache);
dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
setup(dev);
@@ -12167,6 +12168,7 @@ void free_netdev(struct net_device *dev)
if (netif_rx_mode_clean(dev))
dev_put(dev);
+ __hw_addr_flush(&dev->rx_mode_addr_cache);
/* Flush device addresses */
dev_addr_flush(dev);
diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index 61589537b2d3..2cff791ce374 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -511,30 +511,50 @@ void __hw_addr_init(struct netdev_hw_addr_list *list)
}
EXPORT_SYMBOL(__hw_addr_init);
+static void __hw_addr_splice(struct netdev_hw_addr_list *dst,
+ struct netdev_hw_addr_list *src)
+{
+ src->tree = RB_ROOT;
+ list_splice_init(&src->list, &dst->list);
+ dst->count += src->count;
+ src->count = 0;
+}
+
/**
* __hw_addr_list_snapshot - create a snapshot copy of an address list
* @snap: destination snapshot list (needs to be __hw_addr_init-initialized)
* @list: source address list to snapshot
* @addr_len: length of addresses
+ * @cache: entry cache to reuse entries from; falls back to GFP_ATOMIC
*
- * Creates a copy of @list with individually allocated entries suitable
- * for use with __hw_addr_sync_dev() and other list manipulation helpers.
- * Each entry is allocated with GFP_ATOMIC; must be called under a spinlock.
+ * Creates a copy of @list reusing entries from @cache when available.
+ * Must be called under a spinlock.
*
* Return: 0 on success, -errno on failure.
*/
int __hw_addr_list_snapshot(struct netdev_hw_addr_list *snap,
const struct netdev_hw_addr_list *list,
- int addr_len)
+ int addr_len, struct netdev_hw_addr_list *cache)
{
struct netdev_hw_addr *ha, *entry;
list_for_each_entry(ha, &list->list, list) {
- entry = __hw_addr_create(ha->addr, addr_len, ha->type,
- false, false);
- if (!entry) {
- __hw_addr_flush(snap);
- return -ENOMEM;
+ if (cache->count) {
+ entry = list_first_entry(&cache->list,
+ struct netdev_hw_addr, list);
+ list_del(&entry->list);
+ cache->count--;
+ memcpy(entry->addr, ha->addr, addr_len);
+ entry->type = ha->type;
+ entry->global_use = false;
+ entry->synced = 0;
+ } else {
+ entry = __hw_addr_create(ha->addr, addr_len, ha->type,
+ false, false);
+ if (!entry) {
+ __hw_addr_flush(snap);
+ return -ENOMEM;
+ }
}
entry->sync_cnt = ha->sync_cnt;
entry->refcount = ha->refcount;
@@ -554,15 +574,17 @@ EXPORT_SYMBOL_IF_KUNIT(__hw_addr_list_snapshot);
* @work: the working snapshot (modified by driver via __hw_addr_sync_dev)
* @ref: the reference snapshot (untouched copy of original state)
* @addr_len: length of addresses
+ * @cache: entry cache to return snapshot entries to for reuse
*
* Walks the reference snapshot and compares each entry against the work
* snapshot to compute sync_cnt deltas. Applies those deltas to @real_list.
- * Frees both snapshots when done.
+ * Returns snapshot entries to @cache for reuse; frees both snapshots.
* Caller must hold netif_addr_lock_bh.
*/
void __hw_addr_list_reconcile(struct netdev_hw_addr_list *real_list,
struct netdev_hw_addr_list *work,
- struct netdev_hw_addr_list *ref, int addr_len)
+ struct netdev_hw_addr_list *ref, int addr_len,
+ struct netdev_hw_addr_list *cache)
{
struct netdev_hw_addr *ref_ha, *work_ha, *real_ha;
int delta;
@@ -614,8 +636,8 @@ void __hw_addr_list_reconcile(struct netdev_hw_addr_list *real_list,
}
}
- __hw_addr_flush(work);
- __hw_addr_flush(ref);
+ __hw_addr_splice(cache, work);
+ __hw_addr_splice(cache, ref);
}
EXPORT_SYMBOL_IF_KUNIT(__hw_addr_list_reconcile);
@@ -1176,14 +1198,18 @@ static int netif_addr_lists_snapshot(struct net_device *dev,
{
int err;
- err = __hw_addr_list_snapshot(uc_snap, &dev->uc, dev->addr_len);
+ err = __hw_addr_list_snapshot(uc_snap, &dev->uc, dev->addr_len,
+ &dev->rx_mode_addr_cache);
if (!err)
- err = __hw_addr_list_snapshot(uc_ref, &dev->uc, dev->addr_len);
+ err = __hw_addr_list_snapshot(uc_ref, &dev->uc, dev->addr_len,
+ &dev->rx_mode_addr_cache);
if (!err)
err = __hw_addr_list_snapshot(mc_snap, &dev->mc,
- dev->addr_len);
+ dev->addr_len,
+ &dev->rx_mode_addr_cache);
if (!err)
- err = __hw_addr_list_snapshot(mc_ref, &dev->mc, dev->addr_len);
+ err = __hw_addr_list_snapshot(mc_ref, &dev->mc, dev->addr_len,
+ &dev->rx_mode_addr_cache);
if (err) {
__hw_addr_flush(uc_snap);
@@ -1200,8 +1226,10 @@ static void netif_addr_lists_reconcile(struct net_device *dev,
struct netdev_hw_addr_list *uc_ref,
struct netdev_hw_addr_list *mc_ref)
{
- __hw_addr_list_reconcile(&dev->uc, uc_snap, uc_ref, dev->addr_len);
- __hw_addr_list_reconcile(&dev->mc, mc_snap, mc_ref, dev->addr_len);
+ __hw_addr_list_reconcile(&dev->uc, uc_snap, uc_ref, dev->addr_len,
+ &dev->rx_mode_addr_cache);
+ __hw_addr_list_reconcile(&dev->mc, mc_snap, mc_ref, dev->addr_len,
+ &dev->rx_mode_addr_cache);
}
/**
diff --git a/net/core/dev_addr_lists_test.c b/net/core/dev_addr_lists_test.c
index 4cbc241a9ec0..0f60c239cecc 100644
--- a/net/core/dev_addr_lists_test.c
+++ b/net/core/dev_addr_lists_test.c
@@ -250,8 +250,8 @@ static void dev_addr_test_add_excl(struct kunit *test)
*/
static void dev_addr_test_snapshot_sync(struct kunit *test)
{
+ struct netdev_hw_addr_list snap, ref, cache;
struct net_device *netdev = test->priv;
- struct netdev_hw_addr_list snap, ref;
struct dev_addr_test_priv *datp;
struct netdev_hw_addr *ha;
u8 addr[ETH_ALEN];
@@ -267,10 +267,13 @@ static void dev_addr_test_snapshot_sync(struct kunit *test)
netif_addr_lock_bh(netdev);
__hw_addr_init(&snap);
__hw_addr_init(&ref);
+ __hw_addr_init(&cache);
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN,
+ &cache));
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN,
+ &cache));
netif_addr_unlock_bh(netdev);
/* Driver syncs ADDR_A to hardware */
@@ -282,7 +285,8 @@ static void dev_addr_test_snapshot_sync(struct kunit *test)
/* Reconcile: delta=+1 applied to real entry */
netif_addr_lock_bh(netdev);
- __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN,
+ &cache);
netif_addr_unlock_bh(netdev);
/* Real entry should now reflect the sync: sync_cnt=1, refcount=2 */
@@ -300,6 +304,7 @@ static void dev_addr_test_snapshot_sync(struct kunit *test)
KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
+ __hw_addr_flush(&cache);
rtnl_unlock();
}
@@ -309,8 +314,8 @@ static void dev_addr_test_snapshot_sync(struct kunit *test)
*/
static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
{
+ struct netdev_hw_addr_list snap, ref, cache;
struct net_device *netdev = test->priv;
- struct netdev_hw_addr_list snap, ref;
struct dev_addr_test_priv *datp;
struct netdev_hw_addr *ha;
u8 addr[ETH_ALEN];
@@ -326,10 +331,13 @@ static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
netif_addr_lock_bh(netdev);
__hw_addr_init(&snap);
__hw_addr_init(&ref);
+ __hw_addr_init(&cache);
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN,
+ &cache));
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN,
+ &cache));
netif_addr_unlock_bh(netdev);
/* Driver syncs ADDR_A to hardware */
@@ -348,7 +356,8 @@ static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
* so it gets re-inserted as stale (sync_cnt=1, refcount=1).
*/
netif_addr_lock_bh(netdev);
- __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN,
+ &cache);
netif_addr_unlock_bh(netdev);
KUNIT_EXPECT_EQ(test, 1, netdev->uc.count);
@@ -365,6 +374,7 @@ static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_unsynced);
KUNIT_EXPECT_EQ(test, 0, netdev->uc.count);
+ __hw_addr_flush(&cache);
rtnl_unlock();
}
@@ -375,8 +385,8 @@ static void dev_addr_test_snapshot_remove_during_sync(struct kunit *test)
*/
static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
{
+ struct netdev_hw_addr_list snap, ref, cache;
struct net_device *netdev = test->priv;
- struct netdev_hw_addr_list snap, ref;
struct dev_addr_test_priv *datp;
struct netdev_hw_addr *ha;
u8 addr[ETH_ALEN];
@@ -402,10 +412,13 @@ static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
netif_addr_lock_bh(netdev);
__hw_addr_init(&snap);
__hw_addr_init(&ref);
+ __hw_addr_init(&cache);
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN,
+ &cache));
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN,
+ &cache));
netif_addr_unlock_bh(netdev);
/* Driver unsyncs stale ADDR_A from hardware */
@@ -425,7 +438,8 @@ static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
* applied. Result: sync_cnt=0, refcount=1 (fresh).
*/
netif_addr_lock_bh(netdev);
- __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN,
+ &cache);
netif_addr_unlock_bh(netdev);
/* Entry survives as fresh: needs re-sync to HW */
@@ -442,6 +456,7 @@ static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
KUNIT_EXPECT_EQ(test, 1 << ADDR_A, datp->addr_synced);
KUNIT_EXPECT_EQ(test, 0, datp->addr_unsynced);
+ __hw_addr_flush(&cache);
rtnl_unlock();
}
@@ -451,8 +466,8 @@ static void dev_addr_test_snapshot_readd_during_unsync(struct kunit *test)
*/
static void dev_addr_test_snapshot_add_and_remove(struct kunit *test)
{
+ struct netdev_hw_addr_list snap, ref, cache;
struct net_device *netdev = test->priv;
- struct netdev_hw_addr_list snap, ref;
struct dev_addr_test_priv *datp;
struct netdev_hw_addr *ha;
u8 addr[ETH_ALEN];
@@ -479,10 +494,13 @@ static void dev_addr_test_snapshot_add_and_remove(struct kunit *test)
netif_addr_lock_bh(netdev);
__hw_addr_init(&snap);
__hw_addr_init(&ref);
+ __hw_addr_init(&cache);
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&snap, &netdev->uc, ETH_ALEN,
+ &cache));
KUNIT_ASSERT_EQ(test, 0,
- __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN));
+ __hw_addr_list_snapshot(&ref, &netdev->uc, ETH_ALEN,
+ &cache));
netif_addr_unlock_bh(netdev);
/* Driver syncs snapshot: ADDR_C is new -> synced; A,B already synced */
@@ -501,7 +519,8 @@ static void dev_addr_test_snapshot_add_and_remove(struct kunit *test)
* so nothing to apply to ADDR_B.
*/
netif_addr_lock_bh(netdev);
- __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN);
+ __hw_addr_list_reconcile(&netdev->uc, &snap, &ref, ETH_ALEN,
+ &cache);
netif_addr_unlock_bh(netdev);
/* ADDR_A: unchanged (sync_cnt=1, refcount=2)
@@ -535,13 +554,14 @@ static void dev_addr_test_snapshot_add_and_remove(struct kunit *test)
KUNIT_EXPECT_EQ(test, 1 << ADDR_B, datp->addr_unsynced);
KUNIT_EXPECT_EQ(test, 2, netdev->uc.count);
+ __hw_addr_flush(&cache);
rtnl_unlock();
}
static void dev_addr_test_snapshot_benchmark(struct kunit *test)
{
struct net_device *netdev = test->priv;
- struct netdev_hw_addr_list snap;
+ struct netdev_hw_addr_list snap, cache;
s64 duration = 0;
ktime_t start;
u8 addr[ETH_ALEN];
@@ -556,6 +576,8 @@ static void dev_addr_test_snapshot_benchmark(struct kunit *test)
KUNIT_ASSERT_EQ(test, 0, dev_uc_add(netdev, addr));
}
+ __hw_addr_init(&cache);
+
for (iter = 0; iter < 1000; iter++) {
netif_addr_lock_bh(netdev);
__hw_addr_init(&snap);
@@ -563,13 +585,15 @@ static void dev_addr_test_snapshot_benchmark(struct kunit *test)
start = ktime_get();
KUNIT_ASSERT_EQ(test, 0,
__hw_addr_list_snapshot(&snap, &netdev->uc,
- ETH_ALEN));
+ ETH_ALEN, &cache));
duration += ktime_to_ns(ktime_sub(ktime_get(), start));
netif_addr_unlock_bh(netdev);
__hw_addr_flush(&snap);
}
+ __hw_addr_flush(&cache);
+
kunit_info(test,
"1024 addrs x 1000 snapshots: %lld ns total, %lld ns/iter",
duration, duration / 1000);
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 05/14] fbnic: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev
Cc: davem, edumazet, kuba, pabeni, Alexander Duyck, kernel-team,
Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert fbnic from ndo_set_rx_mode to ndo_set_rx_mode_async. The
driver's __fbnic_set_rx_mode() now takes explicit uc/mc list
parameters and uses __hw_addr_sync_dev() on the snapshots instead
of __dev_uc_sync/__dev_mc_sync on the netdev directly.
Update callers in fbnic_up, fbnic_fw_config_after_crash,
fbnic_bmc_rpc_check and fbnic_set_mac to pass the real address
lists calling __fbnic_set_rx_mode outside the async work path.
Cc: Alexander Duyck <alexanderduyck@fb.com>
Cc: kernel-team@meta.com
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
.../net/ethernet/meta/fbnic/fbnic_netdev.c | 20 ++++++++++++-------
.../net/ethernet/meta/fbnic/fbnic_netdev.h | 4 +++-
drivers/net/ethernet/meta/fbnic/fbnic_pci.c | 4 ++--
drivers/net/ethernet/meta/fbnic/fbnic_rpc.c | 2 +-
4 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c
index b4b396ca9bce..c406a3b56b37 100644
--- a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c
@@ -183,7 +183,9 @@ static int fbnic_mc_unsync(struct net_device *netdev, const unsigned char *addr)
return ret;
}
-void __fbnic_set_rx_mode(struct fbnic_dev *fbd)
+void __fbnic_set_rx_mode(struct fbnic_dev *fbd,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
bool uc_promisc = false, mc_promisc = false;
struct net_device *netdev = fbd->netdev;
@@ -213,10 +215,10 @@ void __fbnic_set_rx_mode(struct fbnic_dev *fbd)
}
/* Synchronize unicast and multicast address lists */
- err = __dev_uc_sync(netdev, fbnic_uc_sync, fbnic_uc_unsync);
+ err = __hw_addr_sync_dev(uc, netdev, fbnic_uc_sync, fbnic_uc_unsync);
if (err == -ENOSPC)
uc_promisc = true;
- err = __dev_mc_sync(netdev, fbnic_mc_sync, fbnic_mc_unsync);
+ err = __hw_addr_sync_dev(mc, netdev, fbnic_mc_sync, fbnic_mc_unsync);
if (err == -ENOSPC)
mc_promisc = true;
@@ -238,18 +240,21 @@ void __fbnic_set_rx_mode(struct fbnic_dev *fbd)
fbnic_write_tce_tcam(fbd);
}
-static void fbnic_set_rx_mode(struct net_device *netdev)
+static void fbnic_set_rx_mode(struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct fbnic_net *fbn = netdev_priv(netdev);
struct fbnic_dev *fbd = fbn->fbd;
/* No need to update the hardware if we are not running */
if (netif_running(netdev))
- __fbnic_set_rx_mode(fbd);
+ __fbnic_set_rx_mode(fbd, uc, mc);
}
static int fbnic_set_mac(struct net_device *netdev, void *p)
{
+ struct fbnic_net *fbn = netdev_priv(netdev);
struct sockaddr *addr = p;
if (!is_valid_ether_addr(addr->sa_data))
@@ -257,7 +262,8 @@ static int fbnic_set_mac(struct net_device *netdev, void *p)
eth_hw_addr_set(netdev, addr->sa_data);
- fbnic_set_rx_mode(netdev);
+ if (netif_running(netdev))
+ __fbnic_set_rx_mode(fbn->fbd, &netdev->uc, &netdev->mc);
return 0;
}
@@ -551,7 +557,7 @@ static const struct net_device_ops fbnic_netdev_ops = {
.ndo_features_check = fbnic_features_check,
.ndo_set_mac_address = fbnic_set_mac,
.ndo_change_mtu = fbnic_change_mtu,
- .ndo_set_rx_mode = fbnic_set_rx_mode,
+ .ndo_set_rx_mode_async = fbnic_set_rx_mode,
.ndo_get_stats64 = fbnic_get_stats64,
.ndo_bpf = fbnic_bpf,
.ndo_hwtstamp_get = fbnic_hwtstamp_get,
diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h
index 9129a658f8fa..eded20b0e9e4 100644
--- a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h
@@ -97,7 +97,9 @@ void fbnic_time_init(struct fbnic_net *fbn);
int fbnic_time_start(struct fbnic_net *fbn);
void fbnic_time_stop(struct fbnic_net *fbn);
-void __fbnic_set_rx_mode(struct fbnic_dev *fbd);
+void __fbnic_set_rx_mode(struct fbnic_dev *fbd,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc);
void fbnic_clear_rx_mode(struct fbnic_dev *fbd);
void fbnic_phylink_get_pauseparam(struct net_device *netdev,
diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_pci.c b/drivers/net/ethernet/meta/fbnic/fbnic_pci.c
index e3aebbe3656d..6b139cf54256 100644
--- a/drivers/net/ethernet/meta/fbnic/fbnic_pci.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_pci.c
@@ -135,7 +135,7 @@ void fbnic_up(struct fbnic_net *fbn)
fbnic_rss_reinit_hw(fbn->fbd, fbn);
- __fbnic_set_rx_mode(fbn->fbd);
+ __fbnic_set_rx_mode(fbn->fbd, &fbn->netdev->uc, &fbn->netdev->mc);
/* Enable Tx/Rx processing */
fbnic_napi_enable(fbn);
@@ -180,7 +180,7 @@ static int fbnic_fw_config_after_crash(struct fbnic_dev *fbd)
}
fbnic_rpc_reset_valid_entries(fbd);
- __fbnic_set_rx_mode(fbd);
+ __fbnic_set_rx_mode(fbd, &fbd->netdev->uc, &fbd->netdev->mc);
return 0;
}
diff --git a/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c b/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c
index 42a186db43ea..fe95b6f69646 100644
--- a/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c
@@ -244,7 +244,7 @@ void fbnic_bmc_rpc_check(struct fbnic_dev *fbd)
if (fbd->fw_cap.need_bmc_tcam_reinit) {
fbnic_bmc_rpc_init(fbd);
- __fbnic_set_rx_mode(fbd);
+ __fbnic_set_rx_mode(fbd, &fbd->netdev->uc, &fbd->netdev->mc);
fbd->fw_cap.need_bmc_tcam_reinit = false;
}
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 06/14] mlx5: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev
Cc: davem, edumazet, kuba, pabeni, Saeed Mahameed, Tariq Toukan,
Cosmin Ratiu, Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert mlx5 from ndo_set_rx_mode to ndo_set_rx_mode_async. The
driver's mlx5e_set_rx_mode now receives uc/mc snapshots and calls
mlx5e_fs_set_rx_mode_work directly instead of queueing work.
mlx5e_sync_netdev_addr and mlx5e_handle_netdev_addr now take
explicit uc/mc list parameters and iterate with
netdev_hw_addr_list_for_each instead of netdev_for_each_{uc,mc}_addr.
Fallback to netdev's uc/mc in a few places and grab addr lock.
Cc: Saeed Mahameed <saeedm@nvidia.com>
Cc: Tariq Toukan <tariqt@nvidia.com>
Cc: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
.../net/ethernet/mellanox/mlx5/core/en/fs.h | 5 ++-
.../net/ethernet/mellanox/mlx5/core/en_fs.c | 32 ++++++++++++-------
.../net/ethernet/mellanox/mlx5/core/en_main.c | 13 +++++---
3 files changed, 34 insertions(+), 16 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/fs.h b/drivers/net/ethernet/mellanox/mlx5/core/en/fs.h
index c3408b3f7010..091b80a67189 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/fs.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/fs.h
@@ -201,7 +201,10 @@ int mlx5e_add_vlan_trap(struct mlx5e_flow_steering *fs, int trap_id, int tir_nu
void mlx5e_remove_vlan_trap(struct mlx5e_flow_steering *fs);
int mlx5e_add_mac_trap(struct mlx5e_flow_steering *fs, int trap_id, int tir_num);
void mlx5e_remove_mac_trap(struct mlx5e_flow_steering *fs);
-void mlx5e_fs_set_rx_mode_work(struct mlx5e_flow_steering *fs, struct net_device *netdev);
+void mlx5e_fs_set_rx_mode_work(struct mlx5e_flow_steering *fs,
+ struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc);
int mlx5e_fs_vlan_rx_add_vid(struct mlx5e_flow_steering *fs,
struct net_device *netdev,
__be16 proto, u16 vid);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_fs.c b/drivers/net/ethernet/mellanox/mlx5/core/en_fs.c
index fdfe9d1cfe21..12492c4a5d41 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_fs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_fs.c
@@ -609,20 +609,26 @@ static void mlx5e_execute_l2_action(struct mlx5e_flow_steering *fs,
}
static void mlx5e_sync_netdev_addr(struct mlx5e_flow_steering *fs,
- struct net_device *netdev)
+ struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct netdev_hw_addr *ha;
- netif_addr_lock_bh(netdev);
+ if (!uc || !mc) {
+ netif_addr_lock_bh(netdev);
+ mlx5e_sync_netdev_addr(fs, netdev, &netdev->uc, &netdev->mc);
+ netif_addr_unlock_bh(netdev);
+ return;
+ }
mlx5e_add_l2_to_hash(fs->l2.netdev_uc, netdev->dev_addr);
- netdev_for_each_uc_addr(ha, netdev)
+
+ netdev_hw_addr_list_for_each(ha, uc)
mlx5e_add_l2_to_hash(fs->l2.netdev_uc, ha->addr);
- netdev_for_each_mc_addr(ha, netdev)
+ netdev_hw_addr_list_for_each(ha, mc)
mlx5e_add_l2_to_hash(fs->l2.netdev_mc, ha->addr);
-
- netif_addr_unlock_bh(netdev);
}
static void mlx5e_fill_addr_array(struct mlx5e_flow_steering *fs, int list_type,
@@ -724,7 +730,9 @@ static void mlx5e_apply_netdev_addr(struct mlx5e_flow_steering *fs)
}
static void mlx5e_handle_netdev_addr(struct mlx5e_flow_steering *fs,
- struct net_device *netdev)
+ struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct mlx5e_l2_hash_node *hn;
struct hlist_node *tmp;
@@ -736,7 +744,7 @@ static void mlx5e_handle_netdev_addr(struct mlx5e_flow_steering *fs,
hn->action = MLX5E_ACTION_DEL;
if (fs->state_destroy)
- mlx5e_sync_netdev_addr(fs, netdev);
+ mlx5e_sync_netdev_addr(fs, netdev, uc, mc);
mlx5e_apply_netdev_addr(fs);
}
@@ -820,13 +828,15 @@ static void mlx5e_destroy_promisc_table(struct mlx5e_flow_steering *fs)
}
void mlx5e_fs_set_rx_mode_work(struct mlx5e_flow_steering *fs,
- struct net_device *netdev)
+ struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct mlx5e_priv *priv = netdev_priv(netdev);
struct mlx5e_l2_table *ea = &fs->l2;
if (mlx5e_is_uplink_rep(priv)) {
- mlx5e_handle_netdev_addr(fs, netdev);
+ mlx5e_handle_netdev_addr(fs, netdev, uc, mc);
goto update_vport_context;
}
@@ -856,7 +866,7 @@ void mlx5e_fs_set_rx_mode_work(struct mlx5e_flow_steering *fs,
if (enable_broadcast)
mlx5e_add_l2_flow_rule(fs, &ea->broadcast, MLX5E_FULLMATCH);
- mlx5e_handle_netdev_addr(fs, netdev);
+ mlx5e_handle_netdev_addr(fs, netdev, uc, mc);
if (disable_broadcast)
mlx5e_del_l2_flow_rule(fs, &ea->broadcast);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 1238e5356012..45f5bacef255 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4105,11 +4105,13 @@ static void mlx5e_nic_set_rx_mode(struct mlx5e_priv *priv)
queue_work(priv->wq, &priv->set_rx_mode_work);
}
-static void mlx5e_set_rx_mode(struct net_device *dev)
+static void mlx5e_set_rx_mode(struct net_device *dev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct mlx5e_priv *priv = netdev_priv(dev);
- mlx5e_nic_set_rx_mode(priv);
+ mlx5e_fs_set_rx_mode_work(priv->fs, dev, uc, mc);
}
static int mlx5e_set_mac(struct net_device *netdev, void *addr)
@@ -5284,7 +5286,7 @@ const struct net_device_ops mlx5e_netdev_ops = {
.ndo_setup_tc = mlx5e_setup_tc,
.ndo_select_queue = mlx5e_select_queue,
.ndo_get_stats64 = mlx5e_get_stats,
- .ndo_set_rx_mode = mlx5e_set_rx_mode,
+ .ndo_set_rx_mode_async = mlx5e_set_rx_mode,
.ndo_set_mac_address = mlx5e_set_mac,
.ndo_vlan_rx_add_vid = mlx5e_vlan_rx_add_vid,
.ndo_vlan_rx_kill_vid = mlx5e_vlan_rx_kill_vid,
@@ -6269,8 +6271,11 @@ void mlx5e_set_rx_mode_work(struct work_struct *work)
{
struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv,
set_rx_mode_work);
+ struct net_device *dev = priv->netdev;
- return mlx5e_fs_set_rx_mode_work(priv->fs, priv->netdev);
+ netdev_lock_ops(dev);
+ mlx5e_fs_set_rx_mode_work(priv->fs, dev, NULL, NULL);
+ netdev_unlock_ops(dev);
}
/* mlx5e generic netdev management API (move to en_common.c) */
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 07/14] bnxt: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev
Cc: davem, edumazet, kuba, pabeni, Michael Chan, Pavan Chebbi,
Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert bnxt from ndo_set_rx_mode to ndo_set_rx_mode_async.
bnxt_set_rx_mode, bnxt_mc_list_updated and bnxt_uc_list_updated
now take explicit uc/mc list parameters and iterate with
netdev_hw_addr_list_for_each instead of netdev_for_each_{uc,mc}_addr.
The bnxt_cfg_rx_mode internal caller passes the real lists under
netif_addr_lock_bh.
BNXT_RX_MASK_SP_EVENT is still used here, next patch converts to
the direct call.
Cc: Michael Chan <michael.chan@broadcom.com>
Cc: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 31 +++++++++++++----------
1 file changed, 17 insertions(+), 14 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 8ec611bc01ee..11956c7dde23 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -11033,7 +11033,8 @@ static int bnxt_setup_nitroa0_vnic(struct bnxt *bp)
}
static int bnxt_cfg_rx_mode(struct bnxt *);
-static bool bnxt_mc_list_updated(struct bnxt *, u32 *);
+static bool bnxt_mc_list_updated(struct bnxt *, u32 *,
+ const struct netdev_hw_addr_list *);
static int bnxt_init_chip(struct bnxt *bp, bool irq_re_init)
{
@@ -11123,7 +11124,7 @@ static int bnxt_init_chip(struct bnxt *bp, bool irq_re_init)
} else if (bp->dev->flags & IFF_MULTICAST) {
u32 mask = 0;
- bnxt_mc_list_updated(bp, &mask);
+ bnxt_mc_list_updated(bp, &mask, &bp->dev->mc);
vnic->rx_mask |= mask;
}
@@ -13512,17 +13513,17 @@ void bnxt_get_ring_drv_stats(struct bnxt *bp,
bnxt_get_one_ring_drv_stats(bp, stats, &bp->bnapi[i]->cp_ring);
}
-static bool bnxt_mc_list_updated(struct bnxt *bp, u32 *rx_mask)
+static bool bnxt_mc_list_updated(struct bnxt *bp, u32 *rx_mask,
+ const struct netdev_hw_addr_list *mc)
{
struct bnxt_vnic_info *vnic = &bp->vnic_info[BNXT_VNIC_DEFAULT];
- struct net_device *dev = bp->dev;
struct netdev_hw_addr *ha;
u8 *haddr;
int mc_count = 0;
bool update = false;
int off = 0;
- netdev_for_each_mc_addr(ha, dev) {
+ netdev_hw_addr_list_for_each(ha, mc) {
if (mc_count >= BNXT_MAX_MC_ADDRS) {
*rx_mask |= CFA_L2_SET_RX_MASK_REQ_MASK_ALL_MCAST;
vnic->mc_list_count = 0;
@@ -13546,17 +13547,17 @@ static bool bnxt_mc_list_updated(struct bnxt *bp, u32 *rx_mask)
return update;
}
-static bool bnxt_uc_list_updated(struct bnxt *bp)
+static bool bnxt_uc_list_updated(struct bnxt *bp,
+ const struct netdev_hw_addr_list *uc)
{
- struct net_device *dev = bp->dev;
struct bnxt_vnic_info *vnic = &bp->vnic_info[BNXT_VNIC_DEFAULT];
struct netdev_hw_addr *ha;
int off = 0;
- if (netdev_uc_count(dev) != (vnic->uc_filter_count - 1))
+ if (netdev_hw_addr_list_count(uc) != (vnic->uc_filter_count - 1))
return true;
- netdev_for_each_uc_addr(ha, dev) {
+ netdev_hw_addr_list_for_each(ha, uc) {
if (!ether_addr_equal(ha->addr, vnic->uc_list + off))
return true;
@@ -13565,7 +13566,9 @@ static bool bnxt_uc_list_updated(struct bnxt *bp)
return false;
}
-static void bnxt_set_rx_mode(struct net_device *dev)
+static void bnxt_set_rx_mode(struct net_device *dev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct bnxt *bp = netdev_priv(dev);
struct bnxt_vnic_info *vnic;
@@ -13586,7 +13589,7 @@ static void bnxt_set_rx_mode(struct net_device *dev)
if (dev->flags & IFF_PROMISC)
mask |= CFA_L2_SET_RX_MASK_REQ_MASK_PROMISCUOUS;
- uc_update = bnxt_uc_list_updated(bp);
+ uc_update = bnxt_uc_list_updated(bp, uc);
if (dev->flags & IFF_BROADCAST)
mask |= CFA_L2_SET_RX_MASK_REQ_MASK_BCAST;
@@ -13594,7 +13597,7 @@ static void bnxt_set_rx_mode(struct net_device *dev)
mask |= CFA_L2_SET_RX_MASK_REQ_MASK_ALL_MCAST;
vnic->mc_list_count = 0;
} else if (dev->flags & IFF_MULTICAST) {
- mc_update = bnxt_mc_list_updated(bp, &mask);
+ mc_update = bnxt_mc_list_updated(bp, &mask, mc);
}
if (mask != vnic->rx_mask || uc_update || mc_update) {
@@ -13613,7 +13616,7 @@ static int bnxt_cfg_rx_mode(struct bnxt *bp)
bool uc_update;
netif_addr_lock_bh(dev);
- uc_update = bnxt_uc_list_updated(bp);
+ uc_update = bnxt_uc_list_updated(bp, &dev->uc);
netif_addr_unlock_bh(dev);
if (!uc_update)
@@ -15864,7 +15867,7 @@ static const struct net_device_ops bnxt_netdev_ops = {
.ndo_start_xmit = bnxt_start_xmit,
.ndo_stop = bnxt_close,
.ndo_get_stats64 = bnxt_get_stats64,
- .ndo_set_rx_mode = bnxt_set_rx_mode,
+ .ndo_set_rx_mode_async = bnxt_set_rx_mode,
.ndo_eth_ioctl = bnxt_ioctl,
.ndo_validate_addr = eth_validate_addr,
.ndo_set_mac_address = bnxt_change_mac_addr,
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 08/14] bnxt: use snapshot in bnxt_cfg_rx_mode
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev
Cc: davem, edumazet, kuba, pabeni, Michael Chan, Pavan Chebbi,
Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
With the introduction of ndo_set_rx_mode_async (as discussed in [1])
we can call bnxt_cfg_rx_mode directly. Convert bnxt_cfg_rx_mode to
use uc/mc snapshots and move its call in bnxt_sp_task to the
section that resets BNXT_STATE_IN_SP_TASK. Switch to direct call in
bnxt_set_rx_mode.
Link: https://lore.kernel.org/netdev/CACKFLi=5vj8hPqEUKDd8RTw3au5G+zRgQEqjF+6NZnyoNm90KA@mail.gmail.com/ [1]
Cc: Michael Chan <michael.chan@broadcom.com>
Cc: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 33 +++++++++++------------
1 file changed, 15 insertions(+), 18 deletions(-)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 11956c7dde23..48aa9fc53229 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -11032,7 +11032,8 @@ static int bnxt_setup_nitroa0_vnic(struct bnxt *bp)
return rc;
}
-static int bnxt_cfg_rx_mode(struct bnxt *);
+static int bnxt_cfg_rx_mode(struct bnxt *, struct netdev_hw_addr_list *,
+ struct netdev_hw_addr_list *);
static bool bnxt_mc_list_updated(struct bnxt *, u32 *,
const struct netdev_hw_addr_list *);
@@ -11128,7 +11129,7 @@ static int bnxt_init_chip(struct bnxt *bp, bool irq_re_init)
vnic->rx_mask |= mask;
}
- rc = bnxt_cfg_rx_mode(bp);
+ rc = bnxt_cfg_rx_mode(bp, &bp->dev->uc, &bp->dev->mc);
if (rc)
goto err_out;
@@ -13603,24 +13604,17 @@ static void bnxt_set_rx_mode(struct net_device *dev,
if (mask != vnic->rx_mask || uc_update || mc_update) {
vnic->rx_mask = mask;
- bnxt_queue_sp_work(bp, BNXT_RX_MASK_SP_EVENT);
+ bnxt_cfg_rx_mode(bp, uc, mc);
}
}
-static int bnxt_cfg_rx_mode(struct bnxt *bp)
+static int bnxt_cfg_rx_mode(struct bnxt *bp, struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct net_device *dev = bp->dev;
struct bnxt_vnic_info *vnic = &bp->vnic_info[BNXT_VNIC_DEFAULT];
struct netdev_hw_addr *ha;
int i, off = 0, rc;
- bool uc_update;
-
- netif_addr_lock_bh(dev);
- uc_update = bnxt_uc_list_updated(bp, &dev->uc);
- netif_addr_unlock_bh(dev);
-
- if (!uc_update)
- goto skip_uc;
for (i = 1; i < vnic->uc_filter_count; i++) {
struct bnxt_l2_filter *fltr = vnic->l2_filters[i];
@@ -13632,10 +13626,10 @@ static int bnxt_cfg_rx_mode(struct bnxt *bp)
vnic->uc_filter_count = 1;
netif_addr_lock_bh(dev);
- if (netdev_uc_count(dev) > (BNXT_MAX_UC_ADDRS - 1)) {
+ if (netdev_hw_addr_list_count(uc) > (BNXT_MAX_UC_ADDRS - 1)) {
vnic->rx_mask |= CFA_L2_SET_RX_MASK_REQ_MASK_PROMISCUOUS;
} else {
- netdev_for_each_uc_addr(ha, dev) {
+ netdev_hw_addr_list_for_each(ha, uc) {
memcpy(vnic->uc_list + off, ha->addr, ETH_ALEN);
off += ETH_ALEN;
vnic->uc_filter_count++;
@@ -13662,7 +13656,6 @@ static int bnxt_cfg_rx_mode(struct bnxt *bp)
if (test_and_clear_bit(BNXT_STATE_L2_FILTER_RETRY, &bp->state))
netdev_notice(bp->dev, "Retry of L2 filter configuration successful.\n");
-skip_uc:
if ((vnic->rx_mask & CFA_L2_SET_RX_MASK_REQ_MASK_PROMISCUOUS) &&
!bnxt_promisc_ok(bp))
vnic->rx_mask &= ~CFA_L2_SET_RX_MASK_REQ_MASK_PROMISCUOUS;
@@ -14593,6 +14586,7 @@ static void bnxt_ulp_restart(struct bnxt *bp)
static void bnxt_sp_task(struct work_struct *work)
{
struct bnxt *bp = container_of(work, struct bnxt, sp_task);
+ struct net_device *dev = bp->dev;
set_bit(BNXT_STATE_IN_SP_TASK, &bp->state);
smp_mb__after_atomic();
@@ -14606,9 +14600,6 @@ static void bnxt_sp_task(struct work_struct *work)
bnxt_reenable_sriov(bp);
}
- if (test_and_clear_bit(BNXT_RX_MASK_SP_EVENT, &bp->sp_event))
- bnxt_cfg_rx_mode(bp);
-
if (test_and_clear_bit(BNXT_RX_NTP_FLTR_SP_EVENT, &bp->sp_event))
bnxt_cfg_ntp_filters(bp);
if (test_and_clear_bit(BNXT_HWRM_EXEC_FWD_REQ_SP_EVENT, &bp->sp_event))
@@ -14673,6 +14664,12 @@ static void bnxt_sp_task(struct work_struct *work)
/* These functions below will clear BNXT_STATE_IN_SP_TASK. They
* must be the last functions to be called before exiting.
*/
+ if (test_and_clear_bit(BNXT_RX_MASK_SP_EVENT, &bp->sp_event)) {
+ bnxt_lock_sp(bp);
+ bnxt_cfg_rx_mode(bp, &dev->uc, &dev->mc);
+ bnxt_unlock_sp(bp);
+ }
+
if (test_and_clear_bit(BNXT_RESET_TASK_SP_EVENT, &bp->sp_event))
bnxt_reset(bp, false);
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 09/14] iavf: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev
Cc: davem, edumazet, kuba, pabeni, Tony Nguyen, Przemek Kitszel,
Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert iavf from ndo_set_rx_mode to ndo_set_rx_mode_async.
iavf_set_rx_mode now takes explicit uc/mc list parameters and
uses __hw_addr_sync_dev on the snapshots instead of __dev_uc_sync
and __dev_mc_sync.
The iavf_configure internal caller passes the real lists directly.
Cc: Tony Nguyen <anthony.l.nguyen@intel.com>
Cc: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
drivers/net/ethernet/intel/iavf/iavf_main.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/intel/iavf/iavf_main.c b/drivers/net/ethernet/intel/iavf/iavf_main.c
index dad001abc908..485acf402a47 100644
--- a/drivers/net/ethernet/intel/iavf/iavf_main.c
+++ b/drivers/net/ethernet/intel/iavf/iavf_main.c
@@ -1150,14 +1150,18 @@ bool iavf_promiscuous_mode_changed(struct iavf_adapter *adapter)
/**
* iavf_set_rx_mode - NDO callback to set the netdev filters
* @netdev: network interface device structure
+ * @uc: snapshot of uc address list
+ * @mc: snapshot of mc address list
**/
-static void iavf_set_rx_mode(struct net_device *netdev)
+static void iavf_set_rx_mode(struct net_device *netdev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
struct iavf_adapter *adapter = netdev_priv(netdev);
spin_lock_bh(&adapter->mac_vlan_list_lock);
- __dev_uc_sync(netdev, iavf_addr_sync, iavf_addr_unsync);
- __dev_mc_sync(netdev, iavf_addr_sync, iavf_addr_unsync);
+ __hw_addr_sync_dev(uc, netdev, iavf_addr_sync, iavf_addr_unsync);
+ __hw_addr_sync_dev(mc, netdev, iavf_addr_sync, iavf_addr_unsync);
spin_unlock_bh(&adapter->mac_vlan_list_lock);
spin_lock_bh(&adapter->current_netdev_promisc_flags_lock);
@@ -1210,7 +1214,7 @@ static void iavf_configure(struct iavf_adapter *adapter)
struct net_device *netdev = adapter->netdev;
int i;
- iavf_set_rx_mode(netdev);
+ iavf_set_rx_mode(netdev, &netdev->uc, &netdev->mc);
iavf_configure_tx(adapter);
iavf_configure_rx(adapter);
@@ -5153,7 +5157,7 @@ static const struct net_device_ops iavf_netdev_ops = {
.ndo_open = iavf_open,
.ndo_stop = iavf_close,
.ndo_start_xmit = iavf_xmit_frame,
- .ndo_set_rx_mode = iavf_set_rx_mode,
+ .ndo_set_rx_mode_async = iavf_set_rx_mode,
.ndo_validate_addr = eth_validate_addr,
.ndo_set_mac_address = iavf_set_mac,
.ndo_change_mtu = iavf_change_mtu,
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 10/14] netdevsim: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert netdevsim from ndo_set_rx_mode to ndo_set_rx_mode_async.
The callback is a no-op stub so just update the signature and
ops struct wiring.
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
drivers/net/netdevsim/netdev.c | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index c71b8d116f18..73edc4817d62 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -185,7 +185,9 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
return NETDEV_TX_OK;
}
-static void nsim_set_rx_mode(struct net_device *dev)
+static void nsim_set_rx_mode(struct net_device *dev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
}
@@ -593,7 +595,7 @@ static const struct net_shaper_ops nsim_shaper_ops = {
static const struct net_device_ops nsim_netdev_ops = {
.ndo_start_xmit = nsim_start_xmit,
- .ndo_set_rx_mode = nsim_set_rx_mode,
+ .ndo_set_rx_mode_async = nsim_set_rx_mode,
.ndo_set_mac_address = eth_mac_addr,
.ndo_validate_addr = eth_validate_addr,
.ndo_change_mtu = nsim_change_mtu,
@@ -616,7 +618,7 @@ static const struct net_device_ops nsim_netdev_ops = {
static const struct net_device_ops nsim_vf_netdev_ops = {
.ndo_start_xmit = nsim_start_xmit,
- .ndo_set_rx_mode = nsim_set_rx_mode,
+ .ndo_set_rx_mode_async = nsim_set_rx_mode,
.ndo_set_mac_address = eth_mac_addr,
.ndo_validate_addr = eth_validate_addr,
.ndo_change_mtu = nsim_change_mtu,
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 11/14] dummy: convert to ndo_set_rx_mode_async
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni, Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Convert dummy driver from ndo_set_rx_mode to ndo_set_rx_mode_async.
The dummy driver's set_multicast_list is a no-op, so the conversion
is straightforward: update the signature and the ops assignment.
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
drivers/net/dummy.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/net/dummy.c b/drivers/net/dummy.c
index d6bdad4baadd..f8a4eb365c3d 100644
--- a/drivers/net/dummy.c
+++ b/drivers/net/dummy.c
@@ -47,7 +47,9 @@
static int numdummies = 1;
/* fake multicast ability */
-static void set_multicast_list(struct net_device *dev)
+static void set_multicast_list(struct net_device *dev,
+ struct netdev_hw_addr_list *uc,
+ struct netdev_hw_addr_list *mc)
{
}
@@ -87,7 +89,7 @@ static const struct net_device_ops dummy_netdev_ops = {
.ndo_init = dummy_dev_init,
.ndo_start_xmit = dummy_xmit,
.ndo_validate_addr = eth_validate_addr,
- .ndo_set_rx_mode = set_multicast_list,
+ .ndo_set_rx_mode_async = set_multicast_list,
.ndo_set_mac_address = eth_mac_addr,
.ndo_get_stats64 = dummy_get_stats64,
.ndo_change_carrier = dummy_change_carrier,
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 12/14] net: warn ops-locked drivers still using ndo_set_rx_mode
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni, Aleksandr Loktionov
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Now that all in-tree ops-locked drivers have been converted to
ndo_set_rx_mode_async, add a warning in register_netdevice to catch
any remaining or newly added drivers that use ndo_set_rx_mode with
ops locking. This ensures future driver authors are guided toward
the async path.
Also route ops-locked devices through netdev_rx_mode_work even if they
lack rx_mode NDOs, to ensure netdev_ops_assert_locked() does not fire
on the legacy path where only RTNL is held.
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
net/core/dev.c | 5 +++++
net/core/dev_addr_lists.c | 3 ++-
2 files changed, 7 insertions(+), 1 deletion(-)
diff --git a/net/core/dev.c b/net/core/dev.c
index 3ddf347dcdd7..89e31b1187a9 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -11341,6 +11341,11 @@ int register_netdevice(struct net_device *dev)
goto err_uninit;
}
+ if (netdev_need_ops_lock(dev) &&
+ dev->netdev_ops->ndo_set_rx_mode &&
+ !dev->netdev_ops->ndo_set_rx_mode_async)
+ netdev_WARN(dev, "ops-locked drivers should use ndo_set_rx_mode_async\n");
+
ret = netdev_do_alloc_pcpu_stats(dev);
if (ret)
goto err_uninit;
diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index 2cff791ce374..e2d1d671ea56 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -1365,7 +1365,8 @@ void __dev_set_rx_mode(struct net_device *dev)
if (!netif_device_present(dev))
return;
- if (ops->ndo_set_rx_mode_async || ops->ndo_change_rx_flags) {
+ if (ops->ndo_set_rx_mode_async || ops->ndo_change_rx_flags ||
+ netdev_need_ops_lock(dev)) {
netif_rx_mode_queue(dev);
return;
}
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 13/14] selftests: net: add team_bridge_macvlan rx_mode test
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Add a test that exercises the ndo_change_rx_flags path through a
macvlan -> bridge -> team -> dummy stack. This triggers dev_uc_add
under addr_list_lock which flips promiscuity on the lower device.
With the new work queue approach, this must not deadlock.
Link: https://lore.kernel.org/netdev/20260214033859.43857-1-jiayuan.chen@linux.dev/
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
tools/testing/selftests/net/config | 3 ++
tools/testing/selftests/net/rtnetlink.sh | 44 ++++++++++++++++++++++++
2 files changed, 47 insertions(+)
diff --git a/tools/testing/selftests/net/config b/tools/testing/selftests/net/config
index 2a390cae41bf..38611ea11c6b 100644
--- a/tools/testing/selftests/net/config
+++ b/tools/testing/selftests/net/config
@@ -101,6 +101,9 @@ CONFIG_NET_SCH_HTB=m
CONFIG_NET_SCH_INGRESS=m
CONFIG_NET_SCH_NETEM=y
CONFIG_NET_SCH_PRIO=m
+CONFIG_NET_TEAM=y
+CONFIG_NET_TEAM_MODE_ACTIVEBACKUP=y
+CONFIG_NET_TEAM_MODE_LOADBALANCE=y
CONFIG_NET_VRF=y
CONFIG_NF_CONNTRACK=m
CONFIG_NF_CONNTRACK_OVS=y
diff --git a/tools/testing/selftests/net/rtnetlink.sh b/tools/testing/selftests/net/rtnetlink.sh
index 5a5ff88321d5..c499953d4885 100755
--- a/tools/testing/selftests/net/rtnetlink.sh
+++ b/tools/testing/selftests/net/rtnetlink.sh
@@ -23,6 +23,7 @@ ALL_TESTS="
kci_test_encap
kci_test_macsec
kci_test_macsec_vlan
+ kci_test_team_bridge_macvlan
kci_test_ipsec
kci_test_ipsec_offload
kci_test_fdb_get
@@ -636,6 +637,49 @@ kci_test_macsec_vlan()
end_test "PASS: macsec_vlan"
}
+# Test ndo_change_rx_flags call from dev_uc_add under addr_list_lock spinlock.
+# When we are flipping the promisc, make sure it runs on the work queue.
+#
+# https://lore.kernel.org/netdev/20260214033859.43857-1-jiayuan.chen@linux.dev/
+# With (more conventional) macvlan instead of macsec.
+# macvlan -> bridge -> team -> dummy
+kci_test_team_bridge_macvlan()
+{
+ local vlan="test_macv1"
+ local bridge="test_br1"
+ local team="test_team1"
+ local dummy="test_dummy1"
+ local ret=0
+
+ run_cmd ip link add $team type team
+ if [ $ret -ne 0 ]; then
+ end_test "SKIP: team_bridge_macvlan: can't add team interface"
+ return $ksft_skip
+ fi
+
+ run_cmd ip link add $dummy type dummy
+ run_cmd ip link set $dummy master $team
+ run_cmd ip link set $team up
+ run_cmd ip link add $bridge type bridge vlan_filtering 1
+ run_cmd ip link set $bridge up
+ run_cmd ip link set $team master $bridge
+ run_cmd ip link add link $bridge name $vlan \
+ address 00:aa:bb:cc:dd:ee type macvlan mode bridge
+ run_cmd ip link set $vlan up
+
+ run_cmd ip link del $vlan
+ run_cmd ip link del $bridge
+ run_cmd ip link del $team
+ run_cmd ip link del $dummy
+
+ if [ $ret -ne 0 ]; then
+ end_test "FAIL: team_bridge_macvlan"
+ return 1
+ fi
+
+ end_test "PASS: team_bridge_macvlan"
+}
+
#-------------------------------------------------------------------
# Example commands
# ip x s add proto esp src 14.0.0.52 dst 14.0.0.70 \
--
2.53.0
^ permalink raw reply related
* [PATCH net-next v4 14/14] selftests: net: use ip commands instead of teamd in team rx_mode test
From: Stanislav Fomichev @ 2026-04-01 1:19 UTC (permalink / raw)
To: netdev; +Cc: davem, edumazet, kuba, pabeni, Jiri Pirko, Jay Vosburgh
In-Reply-To: <20260401011914.1716692-1-sdf@fomichev.me>
Replace teamd daemon usage with ip link commands for team device
setup. teamd -d daemonizes and returns to the shell before port
addition completes, creating a race: the test may create the macvlan
(and check for its address on a slave) before teamd has finished
adding ports. This makes the test inherently dependent on scheduling
timing.
Using ip commands makes port addition synchronous, removing the race
and making the test deterministic.
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: Jay Vosburgh <jv@jvosburgh.net>
Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
.../selftests/drivers/net/bonding/lag_lib.sh | 17 +++--------------
.../drivers/net/team/dev_addr_lists.sh | 2 --
2 files changed, 3 insertions(+), 16 deletions(-)
diff --git a/tools/testing/selftests/drivers/net/bonding/lag_lib.sh b/tools/testing/selftests/drivers/net/bonding/lag_lib.sh
index bf9bcd1b5ec0..f2e43b6c4c81 100644
--- a/tools/testing/selftests/drivers/net/bonding/lag_lib.sh
+++ b/tools/testing/selftests/drivers/net/bonding/lag_lib.sh
@@ -23,20 +23,9 @@ test_LAG_cleanup()
ip link set dev dummy2 master "$name"
elif [ "$driver" = "team" ]; then
name="team0"
- teamd -d -c '
- {
- "device": "'"$name"'",
- "runner": {
- "name": "'"$mode"'"
- },
- "ports": {
- "dummy1":
- {},
- "dummy2":
- {}
- }
- }
- '
+ ip link add "$name" type team
+ ip link set dev dummy1 master "$name"
+ ip link set dev dummy2 master "$name"
ip link set dev "$name" up
else
check_err 1
diff --git a/tools/testing/selftests/drivers/net/team/dev_addr_lists.sh b/tools/testing/selftests/drivers/net/team/dev_addr_lists.sh
index b1ec7755b783..26469f3be022 100755
--- a/tools/testing/selftests/drivers/net/team/dev_addr_lists.sh
+++ b/tools/testing/selftests/drivers/net/team/dev_addr_lists.sh
@@ -42,8 +42,6 @@ team_cleanup()
}
-require_command teamd
-
trap cleanup EXIT
tests_run
--
2.53.0
^ permalink raw reply related
* Re: [PATCH] net-shapers: free rollback entries using kfree_rcu
From: Jakub Kicinski @ 2026-04-01 1:33 UTC (permalink / raw)
To: Kangzheng Gu
Cc: gregkh, davem, edumazet, pabeni, horms, kees, p, netdev, stable,
linux-kernel
In-Reply-To: <CAKvcANN1OEqXv9fo=cxTEEnq+=qs8NnZBrDTf=FTzdo9rHYJbQ@mail.gmail.com>
On Tue, 31 Mar 2026 15:41:28 +0800 Kangzheng Gu wrote:
> Jakub Kicinski <kuba@kernel.org> 于2026年3月31日周二 09:15写道：
> > If dump can see NOT_VALID entries we have a bigger problem than a UAF
> > don't you think? :/
> I am not sure.
Please experiment and return once you are sure.
netdevsim (netdev simulator) driver supports net_shapers, so you can
easily exercise this code in a VM.
^ permalink raw reply
* Re: [PATCH] net: alteon: Add missing DMA mapping error checks in ace_start_xmit
From: Jakub Kicinski @ 2026-04-01 1:41 UTC (permalink / raw)
To: Joe Damato
Cc: Wang Jun, Jes Sorensen, Andrew Lunn, David S . Miller,
Eric Dumazet, Paolo Abeni, linux-acenic, netdev, linux-kernel,
gszhai, 25125332, 25125283, 23120469, stable
In-Reply-To: <acxcGhzbNsHdK49W@devvm20253.cco0.facebook.com>
On Tue, 31 Mar 2026 16:43:22 -0700 Joe Damato wrote:
> On Tue, Mar 31, 2026 at 09:48:41AM +0800, Wang Jun wrote:
> > The ace_start_xmit function does not check the return value of
> > dma_map_page (via ace_map_tx_skb) and skb_frag_dma_map when building
> > transmit descriptors. If mapping fails, an invalid DMA address is
> > written to the descriptor, which may cause hardware to access
> > illegal memory, leading to system instability or crashes.
> >
> > Add proper dma_mapping_error() checks for all mapping calls. When
> > mapping fails, free the skb, increment the dropped packet counter,
> > and return NETDEV_TX_OK.
> >
> > Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
>
> Is this fixing a bug you've seen in the wild? If not, I'd probably drop the
> fixes tag and send this to net-next instead.
Either it's worth fixing in net or its not worth fixing at all.
My preference would be to try to delete this driver completely.
^ permalink raw reply
* Re: [PATCH net-next 0/7] Decouple receive and transmit enablement in team driver
From: Marc Harvey @ 2026-04-01 1:42 UTC (permalink / raw)
To: Jakub Kicinski
Cc: jiri, andrew+netdev, willemb, maheshb, netdev, linux-kselftest
In-Reply-To: <20260331103308.421aac59@kernel.org>
On Tue, Mar 31, 2026 at 10:33 AM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Tue, 31 Mar 2026 05:33:46 +0000 Marc Harvey wrote:
> > Allow independent control over receive and transmit enablement states
> > for aggregated ports in the team driver.
> >
> > The motivation is that IEE 802.3ad LACP "independent control" can't
> > be implemented for the team driver currently. This was added to the
> > bonding driver in commit 240fd405528b ("bonding: Add independent
> > control state machine").
> >
> > This series also has a few patches that add tests to show that the old
> > coupled enablement still works and that the new decoupled enalbment
> > works as intended (4, 5, and 7).
> >
> > There are three patches with small fixes as well, with the goal of
> > making the final decouplement patch clearer (1, 2, and 3).
>
> The tests don't pass for us, any dependencies we should be aware of?
Ah, I added a check for this using the `require_command` function, but
it seems I used it incorrectly. I will fix this in v2, and probably
remove the pv dependency altogether.
>
> # selftests: drivers/net/team: teamd_activebackup.sh
> # Setting up two-link aggregation for runner activebackup
> # Conf files are /tmp/tmp.zYDeygAAcf and /tmp/tmp.y5dfmFapAj
> # This program is not intended to be run as root.
> # This program is not intended to be run as root.
> # Created team devices
> # PING fd00::2 (fd00::2) 56 data bytes
> # 64 bytes from fd00::2: icmp_seq=1 ttl=64 time=233 ms
> #
> # --- fd00::2 ping statistics ---
> # 1 packets transmitted, 1 received, 0% packet loss, time 0ms
> # rtt min/avg/max/mdev = 233.163/233.163/233.163/0.000 ms/bin/bash: line 1: pv: command not found
> # Packet count for test_team2 was 0
> # Packet count for eth0 was 0
> # Packet count for eth1 was 0
> # Packet count for eth0 was 0
> # Packet count for eth1 was 0
> # TEST: teamd active backup runner test [FAIL]
> # Traffic did not reach team interface in NS2.
> # ./team_lib.sh: line 87: kill: (597) - No such process
> # Tearing down two-link aggregation
> # Failed to kill daemon: Timer expired
> --
> pw-bot: defer
^ permalink raw reply
* RE: [PATCH net-next v1 2/4] r8169: add sfp mode for RTL8116af
From: Javen @ 2026-04-01 1:43 UTC (permalink / raw)
To: Heiner Kallweit, nic_swsd@realtek.com, andrew+netdev@lunn.ch,
davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
pabeni@redhat.com, horms@kernel.org
Cc: netdev@vger.kernel.org, linux-kernel@vger.kernel.org
In-Reply-To: <cd96bc5c-1087-4d13-b556-e1fd94e6322e@gmail.com>
>On 17.03.2026 04:14, Javen wrote:
>>> On 13.03.2026 08:25, Javen wrote:
>>>>> On 02.03.2026 07:32, javen wrote:
>>>>>> From: Javen Xu <javen_xu@realsil.com.cn>
>>>>>>
>>>>>> RTL8116af is a variation of RTL8168fp. It uses SerDes instead of PHY.
>>>>>> But SerDes status will not relect to PHY. So it needs to add sfp
>>>>>> mode for quirk to help reflect SerDes status during PHY read.
>>>>>>
>>>>>
>>>>> Is there any mass market device using this chip version? As far as
>>>>> possible I'd like to avoid adding support for chip versions that
>>>>> never make it to the mass market. This just makes driver
>>>>> maintenance
>>> harder.
>>>>>
>>>>> The patch includes support for 100Mbps fiber mode. Is there any use
>>>>> case for such legacy modes?
>>>>>
>>>>>
>>>>>> Signed-off-by: Javen Xu <javen_xu@realsil.com.cn>
>>>>>> ---
>>>>>> drivers/net/ethernet/realtek/r8169_main.c | 71
>>>>>> ++++++++++++++++++++---
>>>>>> 1 file changed, 62 insertions(+), 9 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/net/ethernet/realtek/r8169_main.c
>>>>>> b/drivers/net/ethernet/realtek/r8169_main.c
>>>>>> index fb2247a20c36..a5c0d3995328 100644
>>>>>> --- a/drivers/net/ethernet/realtek/r8169_main.c
>>>>>> +++ b/drivers/net/ethernet/realtek/r8169_main.c
>>>>>> @@ -726,6 +726,12 @@ enum rtl_dash_type {
>>>>>> RTL_DASH_25_BP,
>>>>>> };
>>>>>>
>>>>>> +enum rtl_sfp_mode {
>>>>>> + RTL_SFP_NONE,
>>>>>> + RTL_SFP_8168_AF,
>>>>>> + RTL_SFP_8127_ATF,
>>>>>> +};
>>>>>> +
>>>>>> struct rtl8169_private {
>>>>>> void __iomem *mmio_addr; /* memory map physical address
>*/
>>>>>> struct pci_dev *pci_dev;
>>>>>> @@ -734,6 +740,7 @@ struct rtl8169_private {
>>>>>> struct napi_struct napi;
>>>>>> enum mac_version mac_version;
>>>>>> enum rtl_dash_type dash_type;
>>>>>> + enum rtl_sfp_mode sfp_mode;
>>>>>> u32 cur_rx; /* Index into the Rx descriptor buffer of next Rx pkt. */
>>>>>> u32 cur_tx; /* Index into the Tx descriptor buffer of next Rx pkt. */
>>>>>> u32 dirty_tx;
>>>>>> @@ -760,7 +767,6 @@ struct rtl8169_private {
>>>>>> unsigned supports_gmii:1;
>>>>>> unsigned aspm_manageable:1;
>>>>>> unsigned dash_enabled:1;
>>>>>> - bool sfp_mode:1;
>>>>>> dma_addr_t counters_phys_addr;
>>>>>> struct rtl8169_counters *counters;
>>>>>> struct rtl8169_tc_offsets tc_offset; @@ -1126,7 +1132,7 @@
>>>>>> static int r8168_phy_ocp_read(struct rtl8169_private *tp, u32 reg)
>>>>>> return 0;
>>>>>>
>>>>>> /* Return dummy MII_PHYSID2 in SFP mode to match SFP PHY
>>>>>> driver
>>> */
>>>>>> - if (tp->sfp_mode && reg == (OCP_STD_PHY_BASE + 2 *
>MII_PHYSID2))
>>>>>> + if (tp->sfp_mode == RTL_SFP_8127_ATF && reg ==
>>>>> (OCP_STD_PHY_BASE
>>>>>> + + 2 * MII_PHYSID2))
>>>>>> return PHY_ID_RTL_DUMMY_SFP & 0xffff;
>>>>>>
>>>>>> RTL_W32(tp, GPHY_OCP, reg << 15); @@ -1270,6 +1276,34 @@
>>>>>> static int r8168g_mdio_read(struct rtl8169_private *tp, int reg)
>>>>>> return r8168_phy_ocp_read(tp, tp->ocp_base + reg * 2); }
>>>>>>
>>>>>> +/* The quirk reflects RTL8116af SerDes status. */ static int
>>>>>> +r8116af_mdio_read_quirk(struct rtl8169_private *tp, int reg) {
>>>>>> + u8 phyStatus = RTL_R8(tp, PHYstatus);
>>>>>> +
>>>>>> + if (!(phyStatus & LinkStatus))
>>>>>> + return 0;
>>>>>> +
>>>>>> + /* BMSR */
>>>>>> + if (tp->ocp_base == OCP_STD_PHY_BASE && reg == MII_BMSR)
>>>>>> + return BMSR_ANEGCOMPLETE | BMSR_LSTATUS;
>>>>>> +
>>>>>> + /* PHYSR */
>>>>>> + if (tp->ocp_base == 0xa430 && reg == 0x12) {
>>>>>> + if (phyStatus & _1000bpsF)
>>>>>> + return 0x0028;
>>>>>> + else if (phyStatus & _100bps)
>>>>>> + return 0x0018;
>>>>>
>>>>> This is a complete hack. Any means to access the SerDes directly?
>>>>
>>>> Hi, Heiner
>>>> Now we find a indirect access to serdes reg. The serdes reg is
>>>> designed
>>> according to IEEE 802.3.
>>>> BIT 2 means Link Status and BIT 5 means ANEGCOMPLETE. We can access
>>>> it through read mac ocp 0xeb14, I wonder whether this indirect
>>>> access can be
>>> accepted?
>>>>
>>> Based on the bit numbers this seems to be the BMSR register. Can all
>>> clause
>>> 22 registers be accessed in a similar way so that a MII bus can be
>>> implemented for accessing the SerDes?
>>
>> We checked our chip documentation and confirmed that the SerDes
>> register file maps all clause 22 registers(e.g. WR 0xeb10<reg_addr>,
>> RD 0xeb14. reg_addr 0x40 is for BMCR,
>> 0x41 for BMSR, 0x42 for PHYSID1, 0x43 for PHYSID2, etc.) As you
>> suggested, I wonder whether the proper approach is to implement a
>> virtual mii_bus and map the standard serdes reg in the new map.
>>
>We'd need to know how the internal PHY and the SerDes play together.
>Whether the usual internal PHY plays any role in communication between
>MAC and SerDes.
>How interface modes are controlled on MAC and SerDes side, etc.
>If Realtek doesn't implement proper layering in hw, and doesn't have enough
>knowledge about mainline network driver structures, and on the other hand
>doesn't want to provide chip documentation, then the situation somewhat
>deadlocks.
>
Hi, Heiner
For RTL8116af, the datapath is structured as MAC->PCS->SerDes.
However, PCS block does not have a standard MDIO interface. Instead, we have to
access the standard reg via MAC OCP.These standard registers are exactly the standard
IEEE 802.3 Clause 22 PHY registers.
Given hardware information above, I would like to ask for your advice on the recommended
approach to support RTL8116af. Or any other information do I need to provide?
Thanks,
Javen Xu
>>> In general it's hard to provide a statement w/o knowing the internal
>>> layout of the IP blocks.
>>> Can you provide any chip documentation / datasheet?
>>> For proper support of such a component chain between MAC and SFP port
>>> the driver would have to be converted to use phylink, what requires
>>> access to the involved components, incl. the SFP I2C bus.
>>>
>>>
>>>> Thanks,
>>>> Javen Xu
>>>>
>>>>>
>>>>>> + }
>>>>>> +
>>>>>> + return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static int r8116af_mdio_read(struct rtl8169_private *tp, int reg) {
>>>>>> + return r8168g_mdio_read(tp, reg) |
>>>>>> +r8116af_mdio_read_quirk(tp, reg); }
>>>>>> +
>>>>>> static void mac_mcu_write(struct rtl8169_private *tp, int reg,
>>>>>> int
>>>>>> value) {
>>>>>> if (reg == 0x1f) {
>>>>>> @@ -1280,6 +1314,13 @@ static void mac_mcu_write(struct
>>>>>> rtl8169_private
>>>>> *tp, int reg, int value)
>>>>>> r8168_mac_ocp_write(tp, tp->ocp_base + reg, value); }
>>>>>>
>>>>>> +static bool rtl_is_8116af(struct rtl8169_private *tp) {
>>>>>> + return tp->mac_version == RTL_GIGA_MAC_VER_52 &&
>>>>>> + (r8168_mac_ocp_read(tp, 0xdc00) & 0x0078) == 0x0030 &&
>>>>>> + (r8168_mac_ocp_read(tp, 0xd006) & 0x00ff) == 0x0000;
>>>>>> +}
>>>>>> +
>>>>>> static int mac_mcu_read(struct rtl8169_private *tp, int reg) {
>>>>>> return r8168_mac_ocp_read(tp, tp->ocp_base + reg); @@
>>>>>> -1386,7
>>>>>> +1427,10 @@ static int rtl_readphy(struct rtl8169_private *tp, int
>>>>>> +location)
>>>>>> case RTL_GIGA_MAC_VER_31:
>>>>>> return r8168dp_2_mdio_read(tp, location);
>>>>>> case RTL_GIGA_MAC_VER_40 ... RTL_GIGA_MAC_VER_LAST:
>>>>>> - return r8168g_mdio_read(tp, location);
>>>>>> + if (tp->sfp_mode == RTL_SFP_8168_AF)
>>>>>> + return r8116af_mdio_read(tp, location);
>>>>>> + else
>>>>>> + return r8168g_mdio_read(tp, location);
>>>>>> default:
>>>>>> return r8169_mdio_read(tp, location);
>>>>>> }
>>>>>> @@ -1575,6 +1619,20 @@ static bool rtl_dash_is_enabled(struct
>>>>> rtl8169_private *tp)
>>>>>> }
>>>>>> }
>>>>>>
>>>>>> +static enum rtl_sfp_mode rtl_get_sfp_mode(struct rtl8169_private
>>>>>> +*tp) {
>>>>>> + if (rtl_is_8125(tp)) {
>>>>>> + u16 data = r8168_mac_ocp_read(tp, 0xd006);
>>>>>> +
>>>>>> + if ((data & 0xff) == 0x07)
>>>>>> + return RTL_SFP_8127_ATF;
>>>>>> + } else if (rtl_is_8116af(tp)) {
>>>>>> + return RTL_SFP_8168_AF;
>>>>>> + }
>>>>>> +
>>>>>> + return RTL_SFP_NONE;
>>>>>> +}
>>>>>> +
>>>>>> static enum rtl_dash_type rtl_get_dash_type(struct
>>>>>> rtl8169_private
>>>>>> *tp) {
>>>>>> switch (tp->mac_version) {
>>>>>> @@ -5693,12 +5751,7 @@ static int rtl_init_one(struct pci_dev
>>>>>> *pdev, const
>>>>> struct pci_device_id *ent)
>>>>>> }
>>>>>> tp->aspm_manageable = !rc;
>>>>>>
>>>>>> - if (rtl_is_8125(tp)) {
>>>>>> - u16 data = r8168_mac_ocp_read(tp, 0xd006);
>>>>>> -
>>>>>> - if ((data & 0xff) == 0x07)
>>>>>> - tp->sfp_mode = true;
>>>>>> - }
>>>>>> + tp->sfp_mode = rtl_get_sfp_mode(tp);
>>>>>>
>>>>>> tp->dash_type = rtl_get_dash_type(tp);
>>>>>> tp->dash_enabled = rtl_dash_is_enabled(tp);
>>>>
>>
^ permalink raw reply
* Re: [PATCH net-next 3/4] ethtool: strset: check nla_len overflow before nla_nest_end
From: Jakub Kicinski @ 2026-04-01 1:46 UTC (permalink / raw)
To: Hangbin Liu
Cc: Donald Hunter, David S. Miller, Eric Dumazet, Paolo Abeni,
Simon Horman, Andrew Lunn, netdev, linux-kernel
In-Reply-To: <20260331-b4-ynl_ethtool-v1-3-dda2a9b55df8@gmail.com>
On Tue, 31 Mar 2026 11:56:13 +0800 Hangbin Liu wrote:
> + if (skb_tail_pointer(skb) - (unsigned char *)strings_attr > U16_MAX)
> + goto nla_put_failure;
bit ugly, let's add a variant of nla_nest_end() which can return
an error on overflow (without the warning from patch 4) ?
> +
> nla_nest_end(skb, strings_attr);
^ permalink raw reply
* Re: [PATCH net-next 1/4] tools: ynl: ethtool: use doit instead of dumpit for per-device GET
From: Jakub Kicinski @ 2026-04-01 1:50 UTC (permalink / raw)
To: Hangbin Liu
Cc: Donald Hunter, David S. Miller, Eric Dumazet, Paolo Abeni,
Simon Horman, Andrew Lunn, netdev, linux-kernel
In-Reply-To: <20260331-b4-ynl_ethtool-v1-1-dda2a9b55df8@gmail.com>
On Tue, 31 Mar 2026 11:56:11 +0800 Hangbin Liu wrote:
> +++ b/tools/net/ynl/pyynl/ethtool.py
We have converted all the samples to selftests so this script is
the last piece of random "PoC" code we still have laying around.
Should we also move it to tests/ ?
If there's a reason to keep it as is -- could you please add that reason
at the top of the file for my future self? I'm pretty sure I already
questioned this script in the past ;)
^ 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