This issue applies for the following qdiscs: hhf, fq, fq_codel, and
fq_pie, and occurs in their change handlers when adjusting to the new
limit. The problem is the following in the values passed to the
subsequent qdisc_tree_reduce_backlog call given a tbf parent:
When the tbf parent runs out of tokens, skbs of these qdiscs will
be placed in gso_skb. Their peek handlers are qdisc_peek_dequeued,
which accounts for both qlen and backlog. However, in the case of
qdisc_dequeue_internal, ONLY qlen is accounted for when pulling
from gso_skb. This means that these qdiscs are missing a
qdisc_qstats_backlog_dec when dropping packets to satisfy the
new limit in their change handlers.
One can observe this issue with the following (with tc patched to
support a limit of 0):
export TARGET=fq
tc qdisc del dev lo root
tc qdisc add dev lo root handle 1: tbf rate 8bit burst 100b latency 1ms
tc qdisc replace dev lo handle 3: parent 1:1 $TARGET limit 1000
echo ''; echo 'add child'; tc -s -d qdisc show dev lo
ping -I lo -f -c2 -s32 -W0.001 127.0.0.1 2>&1 >/dev/null
echo ''; echo 'after ping'; tc -s -d qdisc show dev lo
tc qdisc change dev lo handle 3: parent 1:1 $TARGET limit 0
echo ''; echo 'after limit drop'; tc -s -d qdisc show dev lo
tc qdisc replace dev lo handle 2: parent 1:1 sfq
echo ''; echo 'post graft'; tc -s -d qdisc show dev lo
The second to last show command shows 0 packets but a positive
number (74) of backlog bytes. The problem becomes clearer in the
last show command, where qdisc_purge_queue triggers
qdisc_tree_reduce_backlog with the positive backlog and causes an
underflow in the tbf parent's backlog (4096 Mb instead of 0).
To fix this issue, the codepath for all clients of qdisc_dequeue_internal
has been simplified: codel, pie, hhf, fq, fq_pie, and fq_codel.
qdisc_dequeue_internal handles the backlog adjustments for all cases that
do not directly use the dequeue handler.
The old fq_codel_change limit adjustment loop accumulated the arguments to
the subsequent qdisc_tree_reduce_backlog call through the cstats field.
However, this is confusing and error prone as fq_codel_dequeue could also
potentially mutate this field (which qdisc_dequeue_internal calls in the
non gso_skb case), so we have unified the code here with other qdiscs.
Fixes: 2d3cbfd6d5 ("net_sched: Flush gso_skb list too during ->change()")
Fixes: 4b549a2ef4 ("fq_codel: Fair Queue Codel AQM")
Fixes: 10239edf86 ("net-qdisc-hhf: Heavy-Hitter Filter (HHF) qdisc")
Signed-off-by: William Liu <will@willsroot.io>
Reviewed-by: Savino Dicanosa <savy@syst3mfailure.io>
Link: https://patch.msgid.link/20250812235725.45243-1-will@willsroot.io
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When set multicast_query_interval to a large value, the local variable
'time' in br_multicast_send_query() may overflow. If the time is smaller
than jiffies, the timer will expire immediately, and then call mod_timer()
again, which creates a loop and may trigger the following soft lockup
issue.
watchdog: BUG: soft lockup - CPU#1 stuck for 221s! [rb_consumer:66]
CPU: 1 UID: 0 PID: 66 Comm: rb_consumer Not tainted 6.16.0+ #259 PREEMPT(none)
Call Trace:
<IRQ>
__netdev_alloc_skb+0x2e/0x3a0
br_ip6_multicast_alloc_query+0x212/0x1b70
__br_multicast_send_query+0x376/0xac0
br_multicast_send_query+0x299/0x510
br_multicast_query_expired.constprop.0+0x16d/0x1b0
call_timer_fn+0x3b/0x2a0
__run_timers+0x619/0x950
run_timer_softirq+0x11c/0x220
handle_softirqs+0x18e/0x560
__irq_exit_rcu+0x158/0x1a0
sysvec_apic_timer_interrupt+0x76/0x90
</IRQ>
This issue can be reproduced with:
ip link add br0 type bridge
echo 1 > /sys/class/net/br0/bridge/multicast_querier
echo 0xffffffffffffffff >
/sys/class/net/br0/bridge/multicast_query_interval
ip link set dev br0 up
The multicast_startup_query_interval can also cause this issue. Similar to
the commit 99b4061095 ("net: bridge: mcast: add and enforce query
interval minimum"), add check for the query interval maximum to fix this
issue.
Link: https://lore.kernel.org/netdev/20250806094941.1285944-1-wangliang74@huawei.com/
Link: https://lore.kernel.org/netdev/20250812091818.542238-1-wangliang74@huawei.com/
Fixes: d902eee43f ("bridge: Add multicast count/interval sysfs entries")
Suggested-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Wang Liang <wangliang74@huawei.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Acked-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20250813021054.1643649-1-wangliang74@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Submit multiple descriptors in axienet_rx_cb() to fill Rx skb ring. This
ensures the ring "catches up" on previously missed allocations.
Increment Rx skb ring head pointer after BD is successfully allocated.
Previously, head pointer was incremented before verifying if descriptor is
successfully allocated and has valid entries, which could lead to ring
state inconsistency if descriptor setup failed.
These changes improve reliability by maintaining adequate descriptor
availability and ensuring proper ring buffer state management.
Fixes: 6a91b846af ("net: axienet: Introduce dmaengine support")
Signed-off-by: Suraj Gupta <suraj.gupta2@amd.com>
Link: https://patch.msgid.link/20250813135559.1555652-1-suraj.gupta2@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull networking fixes from Paolo Abeni:
"Including fixes from Netfilter and IPsec.
Current release - regressions:
- netfilter: nft_set_pipapo:
- don't return bogus extension pointer
- fix null deref for empty set
Current release - new code bugs:
- core: prevent deadlocks when enabling NAPIs with mixed kthread
config
- eth: netdevsim: Fix wild pointer access in nsim_queue_free().
Previous releases - regressions:
- page_pool: allow enabling recycling late, fix false positive
warning
- sched: ets: use old 'nbands' while purging unused classes
- xfrm:
- restore GSO for SW crypto
- bring back device check in validate_xmit_xfrm
- tls: handle data disappearing from under the TLS ULP
- ptp: prevent possible ABBA deadlock in ptp_clock_freerun()
- eth:
- bnxt: fill data page pool with frags if PAGE_SIZE > BNXT_RX_PAGE_SIZE
- hv_netvsc: fix panic during namespace deletion with VF
Previous releases - always broken:
- netfilter: fix refcount leak on table dump
- vsock: do not allow binding to VMADDR_PORT_ANY
- sctp: linearize cloned gso packets in sctp_rcv
- eth:
- hibmcge: fix the division by zero issue
- microchip: fix KSZ8863 reset problem"
* tag 'net-6.17-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (54 commits)
net: usb: asix_devices: add phy_mask for ax88772 mdio bus
net: kcm: Fix race condition in kcm_unattach()
selftests: net/forwarding: test purge of active DWRR classes
net/sched: ets: use old 'nbands' while purging unused classes
bnxt: fill data page pool with frags if PAGE_SIZE > BNXT_RX_PAGE_SIZE
netdevsim: Fix wild pointer access in nsim_queue_free().
net: mctp: Fix bad kfree_skb in bind lookup test
netfilter: nf_tables: reject duplicate device on updates
ipvs: Fix estimator kthreads preferred affinity
netfilter: nft_set_pipapo: fix null deref for empty set
selftests: tls: test TCP stealing data from under the TLS socket
tls: handle data disappearing from under the TLS ULP
ptp: prevent possible ABBA deadlock in ptp_clock_freerun()
ixgbe: prevent from unwanted interface name changes
devlink: let driver opt out of automatic phys_port_name generation
net: prevent deadlocks when enabling NAPIs with mixed kthread config
net: update NAPI threaded config even for disabled NAPIs
selftests: drv-net: don't assume device has only 2 queues
docs: Fix name for net.ipv4.udp_child_hash_entries
riscv: dts: thead: Add APB clocks for TH1520 GMACs
...
Without setting phy_mask for ax88772 mdio bus, current driver may create
at most 32 mdio phy devices with phy address range from 0x00 ~ 0x1f.
DLink DUB-E100 H/W Ver B1 is such a device. However, only one main phy
device will bind to net phy driver. This is creating issue during system
suspend/resume since phy_polling_mode() in phy_state_machine() will
directly deference member of phydev->drv for non-main phy devices. Then
NULL pointer dereference issue will occur. Due to only external phy or
internal phy is necessary, add phy_mask for ax88772 mdio bus to workarnoud
the issue.
Closes: https://lore.kernel.org/netdev/20250806082931.3289134-1-xu.yang_2@nxp.com
Fixes: e532a096be ("net: usb: asix: ax88772: add phylib support")
Cc: stable@vger.kernel.org
Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Oleksij Rempel <o.rempel@pengutronix.de>
Link: https://patch.msgid.link/20250811092931.860333-1-xu.yang_2@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Davide Caratti says:
====================
ets: use old 'nbands' while purging unused classes
- patch 1/2 fixes a NULL dereference in the control path of sch_ets qdisc
- patch 2/2 extends kselftests to verify effectiveness of the above fix
====================
Link: https://patch.msgid.link/cover.1755016081.git.dcaratti@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tony Nguyen says:
====================
ixgbe: bypass devlink phys_port_name generation
Jedrzej adds option to skip phys_port_name generation and opts
ixgbe into it as some configurations rely on pre-devlink naming
which could end up broken as a result.
* '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
ixgbe: prevent from unwanted interface name changes
devlink: let driver opt out of automatic phys_port_name generation
====================
Link: https://patch.msgid.link/20250812205226.1984369-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The data page pool always fills the HW rx ring with pages. On arm64 with
64K pages, this will waste _at least_ 32K of memory per entry in the rx
ring.
Fix by fragmenting the pages if PAGE_SIZE > BNXT_RX_PAGE_SIZE. This
makes the data page pool the same as the header pool.
Tested with iperf3 with a small (64 entries) rx ring to encourage buffer
circulation.
Fixes: cd1fafe7da ("eth: bnxt: add support rx side device memory TCP")
Reviewed-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David Wei <dw@davidwei.uk>
Link: https://patch.msgid.link/20250812182907.1540755-1-dw@davidwei.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Florian Westphal says:
====================
Netfilter fixes for net
The following patchset contains Netfilter fixes for *net*:
1) I managed to add a null dereference crash in nft_set_pipapo
in the current development cycle, was not caught by CI
because the avx2 implementation is fine, but selftest
splats when run on non-avx2 host.
2) Fix the ipvs estimater kthread affinity, was incorrect
since 6.14. From Frederic Weisbecker.
3) nf_tables should not allow to add a device to a flowtable
or netdev chain more than once -- reject this.
From Pablo Neira Ayuso. This has been broken for long time,
blamed commit dates from v5.8.
* tag 'nf-25-08-13' of https://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
netfilter: nf_tables: reject duplicate device on updates
ipvs: Fix estimator kthreads preferred affinity
netfilter: nft_set_pipapo: fix null deref for empty set
====================
Link: https://patch.msgid.link/20250813113800.20775-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pull erofs fixes from Gao Xiang:
- Align FSDAX enablement among multiple devices
- Fix EROFS_FS_ZIP_ACCEL build dependency again to prevent forcing
CRYPTO{,_DEFLATE}=y even if EROFS=m
- Fix atomic context detection to properly launch kworkers on demand
- Fix block count statistics for 48-bit addressing support
* tag 'erofs-for-6.17-rc2-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs:
erofs: fix block count report when 48-bit layout is on
erofs: fix atomic context detection when !CONFIG_DEBUG_LOCK_ALLOC
erofs: Do not select tristate symbols from bool symbols
erofs: Fallback to normal access if DAX is not supported on extra device
Pull RCU fix from Neeraj Upadhyay:
"Fix a regression introduced by commit b41642c877 ("rcu: Fix
rcu_read_unlock() deadloop due to IRQ work") which results in boot
hang as reported by kernel test bot at [1].
This issue happens because RCU re-initializes the deferred QS IRQ work
everytime it is queued. With commit b41642c877, the IRQ work
re-initialization can happen while it is already queued. This results
in IRQ work being requeued to itself. When IRQ work finally fires, as
it is requeued to itself, it is repeatedly executed and results in
hang.
Fix this with initializing the IRQ work only once before the CPU
boots"
Link: https://lore.kernel.org/rcu/202508071303.c1134cce-lkp@intel.com/ [1]
* tag 'rcu.fixes.6.17' of git://git.kernel.org/pub/scm/linux/kernel/git/rcu/linux:
rcu: Fix racy re-initialization of irq_work causing hangs
Pull misc fixes from Andrew Morton:
"12 hotfixes. 5 are cc:stable and the remainder address post-6.16
issues or aren't considered necessary for -stable kernels.
10 of these fixes are for MM"
* tag 'mm-hotfixes-stable-2025-08-12-20-50' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
proc: proc_maps_open allow proc_mem_open to return NULL
mm/mremap: avoid expensive folio lookup on mremap folio pte batch
userfaultfd: fix a crash in UFFDIO_MOVE when PMD is a migration entry
mm: pass page directly instead of using folio_page
selftests/proc: fix string literal warning in proc-maps-race.c
fs/proc/task_mmu: hold PTL in pagemap_hugetlb_range and gather_hugetlb_stats
mm/smaps: fix race between smaps_hugetlb_range and migration
mm: fix the race between collapse and PT_RECLAIM under per-vma lock
mm/kmemleak: avoid soft lockup in __kmemleak_do_cleanup()
MAINTAINERS: add Masami as a reviewer of hung task detector
mm/kmemleak: avoid deadlock by moving pr_warn() outside kmemleak_lock
kasan/test: fix protection against compiler elision
A chain/flowtable update with duplicated devices in the same batch is
possible. Unfortunately, netdev event path only removes the first
device that is found, leaving unregistered the hook of the duplicated
device.
Check if a duplicated device exists in the transaction batch, bail out
with EEXIST in such case.
WARNING is hit when unregistering the hook:
[49042.221275] WARNING: CPU: 4 PID: 8425 at net/netfilter/core.c:340 nf_hook_entry_head+0xaa/0x150
[49042.221375] CPU: 4 UID: 0 PID: 8425 Comm: nft Tainted: G S 6.16.0+ #170 PREEMPT(full)
[...]
[49042.221382] RIP: 0010:nf_hook_entry_head+0xaa/0x150
Fixes: 78d9f48f7f ("netfilter: nf_tables: add devices to existing flowtable")
Fixes: b9703ed44f ("netfilter: nf_tables: support for adding new devices to an existing netdev chain")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
The estimator kthreads' affinity are defined by sysctl overwritten
preferences and applied through a plain call to the scheduler's affinity
API.
However since the introduction of managed kthreads preferred affinity,
such a practice shortcuts the kthreads core code which eventually
overwrites the target to the default unbound affinity.
Fix this with using the appropriate kthread's API.
Fixes: d1a8919758 ("kthread: Default affine kthread to its preferred NUMA node")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Florian Westphal <fw@strlen.de>
Blamed commit broke the check for a null scratch map:
- if (unlikely(!m || !*raw_cpu_ptr(m->scratch)))
+ if (unlikely(!raw_cpu_ptr(m->scratch)))
This should have been "if (!*raw_ ...)".
Use the pattern of the avx2 version which is more readable.
This can only be reproduced if avx2 support isn't available.
Fixes: d8d871a35c ("netfilter: nft_set_pipapo: merge pipapo_get/lookup")
Signed-off-by: Florian Westphal <fw@strlen.de>
TLS expects that it owns the receive queue of the TCP socket.
This cannot be guaranteed in case the reader of the TCP socket
entered before the TLS ULP was installed, or uses some non-standard
read API (eg. zerocopy ones). Replace the WARN_ON() and a buggy
early exit (which leaves anchor pointing to a freed skb) with real
error handling. Wipe the parsing state and tell the reader to retry.
We already reload the anchor every time we (re)acquire the socket lock,
so the only condition we need to avoid is an out of bounds read
(not having enough bytes in the socket for previously parsed record len).
If some data was read from under TLS but there's enough in the queue
we'll reload and decrypt what is most likely not a valid TLS record.
Leading to some undefined behavior from TLS perspective (corrupting
a stream? missing an alert? missing an attack?) but no kernel crash
should take place.
Reported-by: William Liu <will@willsroot.io>
Reported-by: Savino Dicanosa <savy@syst3mfailure.io>
Link: https://lore.kernel.org/tFjq_kf7sWIG3A7CrCg_egb8CVsT_gsmHAK0_wxDPJXfIzxFAMxqmLwp3MlU5EHiet0AwwJldaaFdgyHpeIUCS-3m3llsmRzp9xIOBR4lAI=@syst3mfailure.io
Fixes: 84c61fe1a7 ("tls: rx: do not use the standard strparser")
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250807232907.600366-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzbot reported the following ABBA deadlock:
CPU0 CPU1
---- ----
n_vclocks_store()
lock(&ptp->n_vclocks_mux) [1]
(physical clock)
pc_clock_adjtime()
lock(&clk->rwsem) [2]
(physical clock)
...
ptp_clock_freerun()
ptp_vclock_in_use()
lock(&ptp->n_vclocks_mux) [3]
(physical clock)
ptp_clock_unregister()
posix_clock_unregister()
lock(&clk->rwsem) [4]
(virtual clock)
Since ptp virtual clock is registered only under ptp physical clock, both
ptp_clock and posix_clock must be physical clocks for ptp_vclock_in_use()
to lock &ptp->n_vclocks_mux and check ptp->n_vclocks.
However, when unregistering vclocks in n_vclocks_store(), the locking
ptp->n_vclocks_mux is a physical clock lock, but clk->rwsem of
ptp_clock_unregister() called through device_for_each_child_reverse()
is a virtual clock lock.
Therefore, clk->rwsem used in CPU0 and clk->rwsem used in CPU1 are
different locks, but in lockdep, a false positive occurs because the
possibility of deadlock is determined through lock-class.
To solve this, lock subclass annotation must be added to the posix_clock
rwsem of the vclock.
Reported-by: syzbot+7cfb66a237c4a5fb22ad@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=7cfb66a237c4a5fb22ad
Fixes: 73f37068d5 ("ptp: support ptp physical/virtual clocks conversion")
Signed-off-by: Jeongjun Park <aha310510@gmail.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20250728062649.469882-1-aha310510@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently when adding devlink port, phys_port_name is automatically
generated within devlink port initialization flow. As a result adding
devlink port support to driver may result in forced changes of interface
names, which breaks already existing network configs.
This is an expected behavior but in some scenarios it would not be
preferable to provide such limitation for legacy driver not being able to
keep 'pre-devlink' interface name.
Add flag no_phys_port_name to devlink_port_attrs struct which indicates
if devlink should not alter name of interface.
Suggested-by: Jiri Pirko <jiri@resnulli.us>
Link: https://lore.kernel.org/all/nbwrfnjhvrcduqzjl4a2jafnvvud6qsbxlvxaxilnryglf4j7r@btuqrimnfuly/
Signed-off-by: Jedrzej Jagielski <jedrzej.jagielski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Pull habanalabs fix from Al Viro:
"Yet another use-after-free fix due to dma_buf_fd() misuse"
* tag 'pull-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
habanalabs: fix UAF in export_dmabuf()
Pull btrfs fixes from David Sterba:
- fix bug in qgroups reporting incorrect usage for higher level qgroups
- in zoned mode, do not select metadata group as finish target
- convert xarray lock to RCU when trying to release extent buffer to
avoid a deadlock
- do not allow relocation on partially dropped subvolumes, which is
normally not possible but has been reported on old filesystems
- in tree-log, report errors on missing block group when unaccounting
log tree extent buffers
- with large folios, fix range length when processing ordered extents
* tag 'for-6.17-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: fix iteration bug in __qgroup_excl_accounting()
btrfs: zoned: do not select metadata BG as finish target
btrfs: do not allow relocation of partially dropped subvolumes
btrfs: error on missing block group when unaccounting log tree extent buffers
btrfs: fix wrong length parameter for btrfs_cleanup_ordered_extents()
btrfs: make btrfs_cleanup_ordered_extents() support large folios
btrfs: fix subpage deadlock in try_release_subpage_extent_buffer()
- Add a mitigation for a cache coherency vulnerability when running an
SNP guest which makes sure all cache lines belonging to a 4K page are
evicted after latter has been converted to a guest-private page
[ SNP: Secure Nested Paging - not to be confused with Single Nucleotide
Polymorphism, which is the more common use of that TLA. I am on a
mission to write out the more obscure TLAs in order to keep track of
them.
Because while math tells us that there are only about 17k different
combinations of three-letter acronyms using English letters (26^3), I
am convinced that somehow Intel, AMD and ARM have together figured out
new mathematics, and have at least a million different TLAs that they
use. - Linus ]
* tag 'snp_cache_coherency' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/sev: Evict cache lines during SNP memory validation
Steffen Klassert says:
====================
pull request (net): ipsec 2025-08-11
1) Fix flushing of all states in xfrm_state_fini.
From Sabrina Dubroca.
2) Fix some IPsec software offload features. These
got lost with some recent HW offload changes.
From Sabrina Dubroca.
Please pull or let me know if there are problems.
* tag 'ipsec-2025-08-11' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec:
udp: also consider secpath when evaluating ipsec use for checksumming
xfrm: bring back device check in validate_xmit_xfrm
xfrm: restore GSO for SW crypto
xfrm: flush all states in xfrm_state_fini
====================
Link: https://patch.msgid.link/20250811092008.731573-1-steffen.klassert@secunet.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Jakub Kicinski says:
====================
net: prevent deadlocks and mis-configuration with per-NAPI threaded config
Running the test added with a recent fix on a driver with persistent
NAPI config leads to a deadlock. The deadlock is fixed by patch 3,
patch 2 is I think a more fundamental problem with the way we
implemented the config.
I hope the fix makes sense, my own thinking is definitely colored
by my preference (IOW how the per-queue config RFC was implemented).
v1: https://lore.kernel.org/20250808014952.724762-1-kuba@kernel.org
====================
Link: https://patch.msgid.link/20250809001205.1147153-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The following order of calls currently deadlocks if:
- device has threaded=1; and
- NAPI has persistent config with threaded=0.
netif_napi_add_weight_config()
dev->threaded == 1
napi_kthread_create()
napi_enable()
napi_restore_config()
napi_set_threaded(0)
napi_stop_kthread()
while (NAPIF_STATE_SCHED)
msleep(20)
We deadlock because disabled NAPI has STATE_SCHED set.
Creating a thread in netif_napi_add() just to destroy it in
napi_disable() is fairly ugly in the first place. Let's read
both the device config and the NAPI config in netif_napi_add().
Fixes: e6d7626881 ("net: Update threaded state in napi config in netif_set_threaded")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Joe Damato <joe@dama.to>
Link: https://patch.msgid.link/20250809001205.1147153-4-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
We have to make sure that all future NAPIs will have the right threaded
state when the state is configured on the device level.
We chose not to have an "unset" state for threaded, and not to wipe
the NAPI config clean when channels are explicitly disabled.
This means the persistent config structs "exist" even when their NAPIs
are not instantiated.
Differently put - the NAPI persistent state lives in the net_device
(ncfg == struct napi_config):
,--- [napi 0] - [napi 1]
[dev] | |
`--- [ncfg 0] - [ncfg 1]
so say we a device with 2 queues but only 1 enabled:
,--- [napi 0]
[dev] |
`--- [ncfg 0] - [ncfg 1]
now we set the device to threaded=1:
,---------- [napi 0 (thr:1)]
[dev(thr:1)] |
`---------- [ncfg 0 (thr:1)] - [ncfg 1 (thr:?)]
Since [ncfg 1] was not attached to a NAPI during configuration we
skipped it. If we create a NAPI for it later it will have the old
setting (presumably disabled). One could argue if this is right
or not "in principle", but it's definitely not how things worked
before per-NAPI config..
Fixes: 2677010e77 ("Add support to set NAPI threaded for individual NAPI")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Joe Damato <joe@dama.to>
Link: https://patch.msgid.link/20250809001205.1147153-3-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The test is implicitly assuming the device only has 2 queues.
A real device will likely have more. The exact problem is that
because NAPIs get added to the list from the head, the netlink
dump reports them in reverse order. So the naive napis[0] will
actually likely give us the _last_ NAPI, not the first one.
Re-enable all the NAPIs instead of hard-coding 2 in the test.
This way the NAPIs we operated on will always reappear,
doesn't matter where they were in the registration order.
Fixes: e6d7626881 ("net: Update threaded state in napi config in netif_set_threaded")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Joe Damato <joe@dama.to>
Link: https://patch.msgid.link/20250809001205.1147153-2-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Yao Zi says:
====================
Fix broken link with TH1520 GMAC when linkspeed changes
It's noted that on TH1520 SoC, the GMAC's link becomes broken after
the link speed is changed (for example, running ethtool -s eth0 speed
100 on the peer when negotiated to 1Gbps), but the GMAC could function
normally if the speed is brought back to the initial.
Just like many other SoCs utilizing STMMAC IP, we need to adjust the TX
clock supplying TH1520's GMAC through some SoC-specific glue registers
when linkspeed changes. But it's found that after the full kernel
startup, reading from them results in garbage and writing to them makes
no effect, which is the cause of broken link.
Further testing shows perisys-apb4-hclk must be ungated for normal
access to Th1520 GMAC APB glue registers, which is neither described in
dt-binding nor acquired by the driver.
This series expands the dt-binding of TH1520's GMAC to allow an extra
"APB glue registers interface clock", instructs the driver to acquire
and enable the clock, and finally supplies CLK_PERISYS_APB4_HCLK for
TH1520's GMACs in SoC devicetree.
v2: https://lore.kernel.org/netdev/20250801091240.46114-1-ziyao@disroot.org/
v1: https://lore.kernel.org/all/20250729093734.40132-1-ziyao@disroot.org/
====================
Link: https://patch.msgid.link/20250808093655.48074-2-ziyao@disroot.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
It's necessary to adjust the MAC TX clock when the linkspeed changes,
but it's noted such adjustment always fails on TH1520 SoC, and reading
back from APB glue registers that control clock generation results in
garbage, causing broken link.
With some testing, it's found a clock must be ungated for access to APB
glue registers. Without any consumer, the clock is automatically
disabled during late kernel startup. Let's get and enable it if it's
described in devicetree.
For backward compatibility with older devicetrees, probing won't fail if
the APB clock isn't found. In this case, we emit a warning since the
link will break if the speed changes.
Fixes: 33a1a01e3a ("net: stmmac: Add glue layer for T-HEAD TH1520 SoC")
Signed-off-by: Yao Zi <ziyao@disroot.org>
Tested-by: Drew Fustini <fustini@kernel.org>
Reviewed-by: Drew Fustini <fustini@kernel.org>
Link: https://patch.msgid.link/20250808093655.48074-4-ziyao@disroot.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
reset_gpio is claimed in mdiobus_register_device(), but it is not
released in mdiobus_unregister_device(). It is instead only
released when the whole MDIO bus is unregistered.
When a device uses the reset_gpio property, it becomes impossible
to unregister it and register it again, because the GPIO remains
claimed.
This patch resolves that issue.
Fixes: bafbdd527d ("phylib: Add device reset GPIO support") # see notes
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Cc: Csókás Bence <csokas.bence@prolan.hu>
[ csokas.bence: Resolve rebase conflict and clarify msg ]
Signed-off-by: Buday Csaba <buday.csaba@prolan.hu>
Link: https://patch.msgid.link/20250807135449.254254-2-csokas.bence@prolan.hu
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
TJA1103/04/20/21 support both C22 and C45 accessing methods.
The TJA11xx driver has implemented the match_phy_device() API.
However, it does not handle the C45 ID. If C45 was used to access
TJA11xx, match_phy_device() would always return false due to
phydev->phy_id only used by C22 being empty, resulting in the
generic phy driver being used for TJA11xx PHYs.
Therefore, check phydev->c45_ids.device_ids[MDIO_MMD_PMAPMD] when
using C45.
Fixes: 1b76b2497a ("net: phy: nxp-c45-tja11xx: simplify .match_phy_device OP")
Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Link: https://patch.msgid.link/20250807040832.2455306-1-xiaoning.wang@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The commit 65c6604725 ("proc: fix the issue of proc_mem_open returning
NULL") caused proc_maps_open() to return -ESRCH when proc_mem_open()
returns NULL. This breaks legitimate /proc/<pid>/maps access for kernel
threads since kernel threads have NULL mm_struct.
The regression causes perf to fail and exit when profiling a kernel
thread:
# perf record -v -g -p $(pgrep kswapd0)
...
couldn't open /proc/65/task/65/maps
This patch partially reverts the commit to fix it.
Link: https://lkml.kernel.org/r/20250807165455.73656-1-wjl.linux@gmail.com
Fixes: 65c6604725 ("proc: fix the issue of proc_mem_open returning NULL")
Signed-off-by: Jialin Wang <wjl.linux@gmail.com>
Cc: Penglei Jiang <superman.xpt@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
It was discovered in the attached report that commit f822a9a81a ("mm:
optimize mremap() by PTE batching") introduced a significant performance
regression on a number of metrics on x86-64, most notably
stress-ng.bigheap.realloc_calls_per_sec - indicating a 37.3% regression in
number of mremap() calls per second.
I was able to reproduce this locally on an intel x86-64 raptor lake
system, noting an average of 143,857 realloc calls/sec (with a stddev of
4,531 or 3.1%) prior to this patch being applied, and 81,503 afterwards
(stddev of 2,131 or 2.6%) - a 43.3% regression.
During testing I was able to determine that there was no meaningful
difference in efforts to optimise the folio_pte_batch() operation, nor
checking folio_test_large().
This is within expectation, as a regression this large is likely to
indicate we are accessing memory that is not yet in a cache line (and
perhaps may even cause a main memory fetch).
The expectation by those discussing this from the start was that
vm_normal_folio() (invoked by mremap_folio_pte_batch()) would likely be
the culprit due to having to retrieve memory from the vmemmap (which
mremap() page table moves does not otherwise do, meaning this is
inevitably cold memory).
I was able to definitively determine that this theory is indeed correct
and the cause of the issue.
The solution is to restore part of an approach previously discarded on
review, that is to invoke pte_batch_hint() which explicitly determines,
through reference to the PTE alone (thus no vmemmap lookup), what the PTE
batch size may be.
On platforms other than arm64 this is currently hardcoded to return 1, so
this naturally resolves the issue for x86-64, and for arm64 introduces
little to no overhead as the pte cache line will be hot.
With this patch applied, we move from 81,503 realloc calls/sec to 138,701
(stddev of 496.1 or 0.4%), which is a -3.6% regression, however accounting
for the variance in the original result, this is broadly restoring
performance to its prior state.
Link: https://lkml.kernel.org/r/20250807185819.199865-1-lorenzo.stoakes@oracle.com
Fixes: f822a9a81a ("mm: optimize mremap() by PTE batching")
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202508071609.4e743d7c-lkp@intel.com
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Pedro Falcato <pfalcato@suse.de>
Reviewed-by: Barry Song <baohua@kernel.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Dev Jain <dev.jain@arm.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
This change resolves non literal string format warning invoked for
proc-maps-race.c while compiling.
proc-maps-race.c:205:17: warning: format not a string literal and no format arguments [-Wformat-security]
205 | printf(text);
| ^~~~~~
proc-maps-race.c:209:17: warning: format not a string literal and no format arguments [-Wformat-security]
209 | printf(text);
| ^~~~~~
proc-maps-race.c: In function `print_last_lines':
proc-maps-race.c:224:9: warning: format not a string literal and no format arguments [-Wformat-security]
224 | printf(start);
| ^~~~~~
Add string format specifier %s for the printf calls in both
print_first_lines() and print_last_lines() thus resolving the warnings.
The test executes fine after this change thus causing no effect to the
functional behavior of the test.
Link: https://lkml.kernel.org/r/20250804225633.841777-1-hsukrut3@gmail.com
Fixes: aadc099c48 ("selftests/proc: add verbose mode for /proc/pid/maps tearing tests")
Signed-off-by: Sukrut Heroorkar <hsukrut3@gmail.com>
Acked-by: Suren Baghdasaryan <surenb@google.com>
Cc: David Hunter <david.hunter.linux@gmail.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
dwc_eth_dwmac_probe() gets bulk clocks, and then prepares and enables
them. Unfortunately, if dwc_eth_dwmac_config_dt() or stmmac_dvr_probe()
fail, we leave the clocks prepared and enabled. Fix this by using
devm_clk_bulk_get_all_enabled() to combine the steps and provide devm
based release of the prepare and enable state.
This also fixes a similar leakin dwc_eth_dwmac_remove() which wasn't
correctly retrieving the struct plat_stmmacenet_data. This becomes
unnecessary.
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Simon Horman <horms@kernel.org>
Fixes: a045e40645 ("net: stmmac: refactor clock management in EQoS driver")
Link: https://patch.msgid.link/E1ukM1X-0086qu-Td@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
