netpoll_setup() decides whether to auto-populate the local source
address by testing np->local_ip.ip, which only inspects the first 4
bytes of the union inet_addr storage.
For an IPv6 netpoll whose caller-supplied local address has a zero
high-32 bits (::1, ::<suffix>, IPv4-mapped ::ffff:a.b.c.d, etc.), this
misdetects the address as unset (which they are not, but the first
4 bytes are empty), calls netpoll_take_ipv6() and overwrites it with
whatever matching link-local/global address the device happens to expose
first.
Introduce a helper netpoll_local_ip_unset() that picks the correct
family-aware test (ipv6_addr_any() for IPv6, !.ip for IPv4) and use it
from netpoll_setup().
Reproducer is something like:
echo "::2" > local_ip
echo 1 > enabled
cat local_ip
# before this fix: 2001:db8::1 (caller-supplied ::2 was clobbered)
# after this fix: ::2
Fixes: b7394d2429 ("netpoll: prepare for ipv6")
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260424-netpoll_fix-v1-1-3a55348c625f@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_clamp_probe0_to_user_timeout() computes remaining time in jiffies
using subtraction with an unsigned lvalue. If elapsed probing time
exceeds the configured TCP_USER_TIMEOUT, the underflow yields a large
value.
This ends up re-arming the probe timer for a full backoff interval
instead of expiring immediately, delaying connection teardown beyond
the configured timeout.
Fix this by preventing underflow so user-set timeout expiration is
handled correctly without extending the probe timer.
Fixes: 344db93ae3 ("tcp: make TCP_USER_TIMEOUT accurate for zero window probes")
Link: https://lore.kernel.org/r/20260414013634.43997-1-ahacigu.linux@gmail.com
Signed-off-by: Altan Hacigumus <ahacigu.linux@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260424014639.54110-1-ahacigu.linux@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some physical adapters on Power systems do not support segmentation
offload when the MSS is less than 224 bytes. Attempting to send such
packets causes the adapter to freeze, stopping all traffic until
manually reset.
Implement ndo_features_check to disable GSO for packets with small MSS
values. The network stack will perform software segmentation instead.
The 224-byte minimum matches ibmvnic
commit <f10b09ef687f> ("ibmvnic: Enforce stronger sanity checks
on GSO packets")
which uses the same physical adapters in SEA configurations.
The issue occurs specifically when the hardware attempts to perform
segmentation (gso_segs > 1) with a small MSS. Single-segment GSO packets
(gso_segs == 1) do not trigger the problematic LSO code path and are
transmitted normally without segmentation.
Add an ndo_features_check callback to disable GSO when MSS < 224 bytes.
Also call vlan_features_check() to ensure proper handling of VLAN packets,
particularly QinQ (802.1ad) configurations where the hardware parser may
not support certain offload features.
Validated using iptables to force small MSS values. Without the fix,
the adapter freezes. With the fix, packets are segmented in software
and transmission succeeds. Comprehensive regression testing completedd
(MSS tests, performance, stability).
Fixes: 8641dd8579 ("ibmveth: Add support for TSO")
Cc: stable@vger.kernel.org
Reviewed-by: Brian King <bjking1@linux.ibm.com>
Tested-by: Shaik Abdulla <shaik.abdulla1@ibm.com>
Tested-by: Naveed Ahmed <naveedaus@in.ibm.com>
Signed-off-by: Mingming Cao <mmc@linux.ibm.com>
Link: https://patch.msgid.link/20260424162917.65725-1-mmc@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
neigh_xmit always releases the skb, except when no neighbour table is
found. But even the first added user of neigh_xmit (mpls) relied on
neigh_xmit to release the skb (or queue it for tx).
sashiko reported:
If neigh_xmit() is called with an uninitialized neighbor table (for
example, NEIGH_ND_TABLE when IPv6 is disabled), it returns -EAFNOSUPPORT
and bypasses its internal out_kfree_skb error path. Because the return
value of neigh_xmit() is ignored here, does this leak the SKB?
Assume full ownership and remove the last code path that doesn't
xmit or free skb.
Fixes: 4fd3d7d9e8 ("neigh: Add helper function neigh_xmit")
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260424145843.74055-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
With CONFIG_IP_MROUTE_MULTIPLE_TABLES=n, ipmr_fib_lookup()
does not check if net->ipv4.mrt is NULL.
Since default_device_exit_batch() is called after ->exit_rtnl(),
a device could receive IGMP packets and access net->ipv4.mrt
during/after ipmr_rules_exit_rtnl().
If ipmr_rules_exit_rtnl() had already cleared it and freed the
memory, the access would trigger null-ptr-deref or use-after-free.
Let's fix it by using RCU helper and free mrt after RCU grace
period.
In addition, check_net(net) is added to mroute_clean_tables()
and ipmr_cache_unresolved() to synchronise via mfc_unres_lock.
This prevents ipmr_cache_unresolved() from putting skb into
c->_c.mfc_un.unres.unresolved after mroute_clean_tables()
purges it.
For the same reason, timer_shutdown_sync() is moved after
mroute_clean_tables().
Since rhltable_destroy() holds mutex internally, rcu_work is
used, and it is placed as the first member because rcu_head
must be placed within <4K offset. mr_table is alraedy 3864
bytes without rcu_work.
Note that IP6MR is not yet converted to ->exit_rtnl(), so this
change is not needed for now but will be.
Fixes: b22b018674 ("ipmr: Convert ipmr_net_exit_batch() to ->exit_rtnl().")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260423053456.4097409-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzbot reported a kernel BUG triggered from pn_socket_sendmsg() via
pn_socket_autobind():
kernel BUG at net/phonet/socket.c:213!
RIP: 0010:pn_socket_autobind net/phonet/socket.c:213 [inline]
RIP: 0010:pn_socket_sendmsg+0x240/0x250 net/phonet/socket.c:421
Call Trace:
sock_sendmsg_nosec+0x112/0x150 net/socket.c:797
__sock_sendmsg net/socket.c:812 [inline]
__sys_sendto+0x402/0x590 net/socket.c:2280
...
pn_socket_autobind() calls pn_socket_bind() with port 0 and, on
-EINVAL, assumes the socket was already bound and asserts that the
port is non-zero:
err = pn_socket_bind(sock, ..., sizeof(struct sockaddr_pn));
if (err != -EINVAL)
return err;
BUG_ON(!pn_port(pn_sk(sock->sk)->sobject));
return 0; /* socket was already bound */
However pn_socket_bind() also returns -EINVAL when sk->sk_state is not
TCP_CLOSE, even when the socket has never been bound and pn_port() is
still 0. In that case the BUG_ON() fires and panics the kernel from a
user-triggerable path.
Treat the "bind returned -EINVAL but pn_port() is still 0" case as a
regular error and propagate -EINVAL to the caller instead of crashing.
Existing callers already translate a non-zero return from
pn_socket_autobind() into -ENOBUFS/-EAGAIN, so returning -EINVAL here
only changes behaviour from panic to a normal errno.
Fixes: ba113a94b7 ("Phonet: common socket glue")
Reported-by: syzbot+706f5eb79044e686c794@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=706f5eb79044e686c794
Suggested-by: Remi Denis-Courmont <courmisch@gmail.com>
Signed-off-by: Morduan Zang <zhangdandan@uniontech.com>
Signed-off-by: zhanjun <zhanjun@uniontech.com>
Acked-by: Rémi Denis-Courmont <remi@remlab.net>
Link: https://patch.msgid.link/87A8960A2045AF3C+20260423010557.138124-1-zhangdandan@uniontech.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Weiming Shi says:
====================
net/sched: taprio: fix NULL pointer dereference in class dump
Patch 1/2 is the fix: replace NULL entries in q->qdiscs[] with the
global &noop_qdisc singleton so that control-plane dump paths, as well
as the existing NULL guards in the data-plane enqueue/dequeue paths,
cannot deref a NULL child qdisc.
Patch 2/2 is a tdc regression test that drives the graft + delete +
class-dump sequence on a multi-queue netdevsim device. It panics the
vulnerable kernel and passes on the fixed one.
====================
Link: https://patch.msgid.link/20260422161958.2517539-2-bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a regression test for the NULL pointer dereference fixed in the
previous commit. Before the fix, taprio_graft() stored NULL into
q->qdiscs[cl - 1] when an explicitly grafted child qdisc was deleted
via RTM_DELQDISC; the next RTM_GETTCLASS dump then crashed the kernel
in taprio_dump_class() while reading child->handle.
The test installs a taprio root qdisc on a multi-queue netdevsim
device, grafts a pfifo child onto class 8001:1, deletes that child,
and then performs a class dump. On a fixed kernel the dump succeeds
and all eight taprio classes are listed; on an unpatched kernel the
class dump crashes, which surfaces as a test failure.
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260422161958.2517539-4-bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When a TAPRIO child qdisc is deleted via RTM_DELQDISC, taprio_graft()
is called with new == NULL and stores NULL into q->qdiscs[cl - 1].
Subsequent RTM_GETTCLASS dump operations walk all classes via
taprio_walk() and call taprio_dump_class(), which calls taprio_leaf()
returning the NULL pointer, then dereferences it to read child->handle,
causing a kernel NULL pointer dereference.
The bug is reachable with namespace-scoped CAP_NET_ADMIN on any kernel
with CONFIG_NET_SCH_TAPRIO enabled. On systems with unprivileged user
namespaces enabled, an unprivileged local user can trigger a kernel
panic by creating a taprio qdisc inside a new network namespace,
grafting an explicit child qdisc, deleting it, and requesting a class
dump. The RTM_GETTCLASS dump itself requires no capability.
Oops: general protection fault, probably for non-canonical address 0xdffffc0000000007: 0000 [#1] SMP KASAN NOPTI
KASAN: null-ptr-deref in range [0x0000000000000038-0x000000000000003f]
RIP: 0010:taprio_dump_class (net/sched/sch_taprio.c:2478)
Call Trace:
<TASK>
tc_fill_tclass (net/sched/sch_api.c:1966)
qdisc_class_dump (net/sched/sch_api.c:2326)
taprio_walk (net/sched/sch_taprio.c:2514)
tc_dump_tclass_qdisc (net/sched/sch_api.c:2352)
tc_dump_tclass_root (net/sched/sch_api.c:2370)
tc_dump_tclass (net/sched/sch_api.c:2431)
rtnl_dumpit (net/core/rtnetlink.c:6864)
netlink_dump (net/netlink/af_netlink.c:2325)
rtnetlink_rcv_msg (net/core/rtnetlink.c:6959)
netlink_rcv_skb (net/netlink/af_netlink.c:2550)
</TASK>
Fix this by substituting &noop_qdisc when new is NULL in
taprio_graft(), a common pattern used by other qdiscs (e.g.,
multiq_graft()) to ensure the q->qdiscs[] slots are never NULL.
This makes control-plane dump paths safe without requiring individual
NULL checks.
Since the data-plane paths (taprio_enqueue and taprio_dequeue_from_txq)
previously had explicit NULL guards that would drop/skip the packet
cleanly, update those checks to test for &noop_qdisc instead. Without
this, packets would reach taprio_enqueue_one() which increments the root
qdisc's qlen and backlog before calling the child's enqueue; noop_qdisc
drops the packet but those counters are never rolled back, permanently
inflating the root qdisc's statistics.
After this change *old can be a valid qdisc, NULL, or &noop_qdisc.
Only call qdisc_put(*old) in the first case to avoid decreasing
noop_qdisc's refcount, which was never increased.
Fixes: 665338b2a7 ("net/sched: taprio: dump class stats for the actual q->qdiscs[]")
Reported-by: Xiang Mei <xmei5@asu.edu>
Signed-off-by: Weiming Shi <bestswngs@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Tested-by: Weiming Shi <bestswngs@gmail.com>
Link: https://patch.msgid.link/20260422161958.2517539-3-bestswngs@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The main channel Received Signal Strength Indicator (RSSI) measurement
is used to determine whether an RF field is present or not. RSSI != 0
is interpreted as an RF Field is present. This does not take RF noise
and measurement inaccuracy into account, and results in false positives
in the field.
Define a noise level and make sure the RF field is only interpreted as
present when the RSSI is above the noise level.
Fixes: 851ee3cbf8 ("NFC: trf7970a: Don't turn on RF if there is already an RF field")
Signed-off-by: Paul Geurts <paul.geurts@prodrive-technologies.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Reviewed-by: Mark Greer <mgreer@animalcreek.com>
Link: https://patch.msgid.link/20260422100930.581237-1-paul.geurts@prodrive-technologies.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move init_completion(&spisg->completion) to before devm_request_irq()
to avoid a potential race condition where an interrupt could fire
before the completion structure is initialized.
Fixes: cef9991e04 ("spi: Add Amlogic SPISG driver")
Signed-off-by: Felix Gu <ustc.gu@gmail.com>
Link: https://patch.msgid.link/20260428-amlogic-spisg-v1-1-8eecc3b446d6@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
When rtl8150_start_xmit() fails to submit the tx URB, the URB is never
handed to the USB core and write_bulk_callback() will not run. The
driver returns NETDEV_TX_OK, which tells the networking stack that the
skb has been consumed, but nothing actually frees the skb on this
error path:
dev->tx_skb = skb;
...
if ((res = usb_submit_urb(dev->tx_urb, GFP_ATOMIC))) {
...
/* no kfree_skb here */
}
return NETDEV_TX_OK;
This leaks the skb on every submit failure and also leaves dev->tx_skb
pointing at memory that the driver itself may later free, which is
fragile.
Free the skb with dev_kfree_skb_any() in the error path and clear
dev->tx_skb so no stale pointer is left behind.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Morduan Zang <zhangdandan@uniontech.com>
Link: https://patch.msgid.link/E7D3E1C013C5A859+20260424015517.9574-1-zhangdandan@uniontech.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzbot reported a KASAN slab-use-after-free read in rtl8150_start_xmit()
when accessing skb->len for tx statistics after usb_submit_urb() has
been called:
BUG: KASAN: slab-use-after-free in rtl8150_start_xmit+0x71f/0x760
drivers/net/usb/rtl8150.c:712
Read of size 4 at addr ffff88810eb7a930 by task kworker/0:4/5226
The URB completion handler write_bulk_callback() frees the skb via
dev_kfree_skb_irq(dev->tx_skb). The URB may complete on another CPU
in softirq context before usb_submit_urb() returns in the submitter,
so by the time the submitter reads skb->len the skb has already been
queued to the per-CPU completion_queue and freed by net_tx_action():
CPU A (xmit) CPU B (USB completion softirq)
------------ ------------------------------
dev->tx_skb = skb;
usb_submit_urb() --+
|-------> write_bulk_callback()
| dev_kfree_skb_irq(dev->tx_skb)
| net_tx_action()
| napi_skb_cache_put() <-- free
netdev->stats.tx_bytes |
+= skb->len; <-- UAF read
Fix it by caching skb->len before submitting the URB and using the
cached value when updating the tx_bytes counter.
The pre-existing tx_bytes semantics are preserved: the counter tracks
the original frame length (skb->len), not the ETH_ZLEN/USB-alignment
padded "count" value that is handed to the device. Changing that
would be a user-visible accounting change and is out of scope for
this UAF fix.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Reported-by: syzbot+3f46c095ac0ca048cb71@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/69e69ee7.050a0220.24bfd3.002b.GAE@google.com/
Closes: https://syzkaller.appspot.com/bug?extid=3f46c095ac0ca048cb71
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Zhan Jun <zhanjun@uniontech.com>
Link: https://patch.msgid.link/809895186B866C10+20260423004913.136655-1-zhangdandan@uniontech.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ipv6_rpl_srh_rcv() decompresses an RFC 6554 Source Routing Header, swaps
the next segment into ipv6_hdr->daddr, recompresses, then pulls the old
header and pushes the new one plus the IPv6 header back. The
recompressed header can be larger than the received one when the swap
reduces the common-prefix length the segments share with daddr (CmprI=0,
CmprE>0, seg[0][0] != daddr[0] gives the maximum +8 bytes).
pskb_expand_head() was gated on segments_left == 0, so on earlier
segments the push consumed unchecked headroom. Once skb_push() leaves
fewer than skb->mac_len bytes in front of data,
skb_mac_header_rebuild()'s call to:
skb_set_mac_header(skb, -skb->mac_len);
will store (data - head) - mac_len into the u16 mac_header field, which
wraps to ~65530, and the following memmove() writes mac_len bytes ~64KiB
past skb->head.
A single AF_INET6/SOCK_RAW/IPV6_HDRINCL packet over lo with a two
segment type-3 SRH (CmprI=0, CmprE=15) reaches headroom 8 after one
pass; KASAN reports a 14-byte OOB write in ipv6_rthdr_rcv.
Fix this by expanding the head whenever the remaining room is less than
the push size plus mac_len, and request that much extra so the rebuilt
MAC header fits afterwards.
Fixes: 8610c7c6e3 ("net: ipv6: add support for rpl sr exthdr")
Cc: stable <stable@kernel.org>
Reported-by: Anthropic
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Link: https://patch.msgid.link/2026042133-gout-unvented-1bd9@gregkh
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
RCU readers that identified a net device as a VRF port using
netif_is_l3_slave() assume that a subsequent call to
netdev_master_upper_dev_get_rcu() will return a VRF device. They then
continue to dereference its l3mdev operations.
This assumption is not always correct and can result in a NPD [1]. There
is no RCU synchronization when removing a port from a VRF, so it is
possible for an RCU reader to see a new master device (e.g., a bridge)
that does not have l3mdev operations.
Fix by adding RCU synchronization after clearing the IFF_L3MDEV_SLAVE
flag. Skip this synchronization when a net device is removed from a VRF
as part of its deletion and when the VRF device itself is deleted. In
the latter case an RCU grace period will pass by the time RTNL is
released.
[1]
BUG: kernel NULL pointer dereference, address: 0000000000000000
[...]
RIP: 0010:l3mdev_fib_table_rcu (net/l3mdev/l3mdev.c:181)
[...]
Call Trace:
<TASK>
l3mdev_fib_table_by_index (net/l3mdev/l3mdev.c:201 net/l3mdev/l3mdev.c:189)
__inet_bind (net/ipv4/af_inet.c:499 (discriminator 3))
inet_bind_sk (net/ipv4/af_inet.c:469)
__sys_bind (./include/linux/file.h:62 (discriminator 1) ./include/linux/file.h:83 (discriminator 1) net/socket.c:1951 (discriminator 1))
__x64_sys_bind (net/socket.c:1969 (discriminator 1) net/socket.c:1967 (discriminator 1) net/socket.c:1967 (discriminator 1))
do_syscall_64 (arch/x86/entry/syscall_64.c:63 (discriminator 1) arch/x86/entry/syscall_64.c:94 (discriminator 1))
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:130)
Fixes: fdeea7be88 ("net: vrf: Set slave's private flag before linking")
Reported-by: Haoze Xie <royenheart@gmail.com>
Reported-by: Yifan Wu <yifanwucs@gmail.com>
Reported-by: Juefei Pu <tomapufckgml@gmail.com>
Reported-by: Yuan Tan <yuantan098@gmail.com>
Closes: https://lore.kernel.org/netdev/20260419145332.3988923-1-n05ec@lzu.edu.cn/
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://patch.msgid.link/20260423063607.1208202-1-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fq_codel_dump_stats() acquires the qdisc spinlock a bit too late.
Move this acquisition before we fill tc_fq_pie_xstats with live data.
Alternative would be to add READ_ONCE() and WRITE_ONCE() annotations,
but the spinlock is needed anyway to scan q->new_flows and q->old_flows.
Fixes: ec97ecf1eb ("net: sched: add Flow Queue PIE packet scheduler")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260423063527.2568262-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
choke_dump_stats() only runs with RTNL held.
It reads fields that can be changed in qdisc fast path.
Add READ_ONCE()/WRITE_ONCE() annotations.
Fixes: edb09eb17e ("net: sched: do not acquire qdisc spinlock in qdisc/class stats dump")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260423062839.2524324-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The transmit loop in airoha_dev_xmit() reads fragment address and length
during its final iteration, when the loop index equals
skb_shinfo(skb)->nr_frags, at which point the fragment data is
uninitialized. While these values are never consumed, the read itself is
unsafe and may trigger a page fault. Fix this by avoiding the fragment
read on the last iteration.
Additionally, move the skb pointer from the first to the last used packet
descriptor, so that airoha_qdma_tx_napi_poll() defers freeing the skb
until the final descriptor is processed.
Fixes: 23020f0493 ("net: airoha: Introduce ethernet support for EN7581 SoC")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260424-airoha-xmit-fix-read-frag-v1-1-fdc0a83c79e8@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Do not wake every netdev TX queue across all ports sharing the QDMA
running netif_tx_wake_all_queues routine in airoha_qdma_wake_netdev_txqs()
but only the ones that are mapped the specific QDMA stopped hw TX queue.
This patch can potentially avoid waking already stopped netdev TX queues
that are mapped to a different QDMA hw TX queue.
Introduce airoha_qdma_get_txq utility routine.
Fixes: b94769eb2f ("net: airoha: Fix possible TX queue stall in airoha_qdma_tx_napi_poll()")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260421-airoha-wake_netdev_txqs-optmization-v1-1-e0be95115d53@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently, airoha_eth driver updates the CPU index register prior of
verifying whether the number of free descriptors has fallen below the
threshold.
Move net_device TX queue length check before updating the TX CPU index
in order to update TX CPU index even if there are more packets to be
transmitted but the net_device TX queue is going to be stopped
accounting the inflight packets.
Fixes: 1d30417410 ("net: airoha: Implement BQL support")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260421-airoha-xmit-stop-condition-v1-1-e670d6a48467@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix a possible BQL imbalance in airoha_dev_xmit(), where inflight
packets are accounted only for the AIROHA_NUM_TX_RING netdev TX
queues. The queue index is computed as:
qid = skb_get_queue_mapping(skb) % ARRAY_SIZE(qdma->q_tx)
txq = netdev_get_tx_queue(dev, qid);
However, airoha_qdma_tx_napi_poll() accounts completions across all
netdev TX queues (num_tx_queues), leading to inconsistent BQL
accounting.
Also reset all netdev TX queues in the ndo_stop callback.
Fixes: 1d30417410 ("net: airoha: Implement BQL support")
Fixes: c9f947769b ("net: airoha: Reset BQL stopping the netdevice")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://patch.msgid.link/20260421-airoha-fix-bql-v1-1-f135afe4275b@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Stephen Hemminger says:
====================
netem: bug fixes
These bugs were found when doing AI-assisted review of sch_netem.c
during investigation of the packet duplication recursion problem
addressed in Jamal's series.
The fixes cover:
- probability gaps in the 4-state Markov loss model
- queue limit not accounting for reordered packets
- PRNG reseeded on every tc change, breaking reproducibility
- slot configuration not validated (inverted ranges, negative
delays, negative limits)
- slot delay arithmetic overflow for ranges above ~2.1 seconds
- negative latency and jitter wrapping to huge time_to_send
values via u64 arithmetic
====================
Link: https://patch.msgid.link/20260418032027.900913-1-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reject requests with negative latency or jitter.
A negative value added to current timestamp (u64) wraps
to an enormous time_to_send, disabling dequeue.
The original UAPI used u32 for these values; the conversion to 64-bit
time values via TCA_NETEM_LATENCY64 and TCA_NETEM_JITTER64
allowed signed values to reach the kernel without validation.
Jitter is already silently clamped by an abs() in netem_change();
that abs() can be removed in a follow-up once this rejection is in
place.
Fixes: 99803171ef ("netem: add uapi to express delay and jitter in nanoseconds")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-7-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
get_slot_next() computes a random delay between min_delay and
max_delay using:
get_random_u32() * (max_delay - min_delay) >> 32
This overflows signed 64-bit arithmetic when the delay range exceeds
approximately 2.1 seconds (2^31 nanoseconds), producing a negative
result that effectively disables slot-based pacing. This is a
realistic configuration for WAN emulation (e.g., slot 1s 5s).
Use mul_u64_u32_shr() which handles the widening multiply without
overflow.
Fixes: 0a9fe5c375 ("netem: slotting with non-uniform distribution")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-6-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reject slot configurations that have no defensible meaning:
- negative min_delay or max_delay
- min_delay greater than max_delay
- negative dist_delay or dist_jitter
- negative max_packets or max_bytes
Negative or out-of-order delays underflow in get_slot_next(),
producing garbage intervals. Negative limits trip the per-slot
accounting (packets_left/bytes_left <= 0) on the first packet of
every slot, defeating the rate-limiting half of the slot feature.
Note that dist_jitter has been silently coerced to its absolute
value by get_slot() since the feature was introduced; rejecting
negatives here converts that silent coercion into -EINVAL. The
abs() can be removed in a follow-up.
Fixes: 836af83b54 ("netem: support delivering packets in delayed time slots")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-5-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
netem_change() unconditionally reseeds the PRNG on every tc change
command. If TCA_NETEM_PRNG_SEED is not specified, a new random seed
is generated, destroying reproducibility for users who set a
deterministic seed on a previous change.
Move the initial random seed generation to netem_init() and only
reseed in netem_change() when TCA_NETEM_PRNG_SEED is explicitly
provided by the user.
Fixes: 4072d97ddc ("netem: add prng attribute to netem_sched_data")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-4-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The queue limit check in netem_enqueue() uses q->t_len which only
counts packets in the internal tfifo. Packets placed in sch->q by
the reorder path (__qdisc_enqueue_head) are not counted, allowing
the total queue occupancy to exceed sch->limit under reordering.
Include sch->q.qlen in the limit check.
Fixes: f8d4bc4550 ("net/sched: netem: account for backlog updates from child qdisc")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-3-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The 4-state Markov chain in loss_4state() has gaps at the boundaries
between transition probability ranges. The comparisons use:
if (rnd < a4)
else if (a4 < rnd && rnd < a1 + a4)
When rnd equals a boundary value exactly, neither branch matches and
no state transition occurs. The redundant lower-bound check (a4 < rnd)
is already implied by being in the else branch.
Remove the unnecessary lower-bound comparisons so the ranges are
contiguous and every random value produces a transition, matching
the GI (General and Intuitive) loss model specification.
This bug goes back to original implementation of this model.
Fixes: 661b79725f ("netem: revised correlated loss generator")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260418032027.900913-2-stephen@networkplumber.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Syzbot reports a KMSAN uninit-value originating from
nsim_dev_trap_skb_build, with the allocation also
being performed in the same function.
Fix this by calling skb_put_zero instead of skb_put to
guarantee zero initialization of the whole IP header.
Closes: https://syzkaller.appspot.com/bug?extid=23d7fcd204e3837866ff
Fixes: da58f90f11 ("netdevsim: Add devlink-trap support")
Signed-off-by: Nikola Z. Ivanov <zlatistiv@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260426201434.742030-1-zlatistiv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
- Fix UAF race in psi pressure_write() against cgroup file release by
extending cgroup_mutex coverage and ordering of->priv access after
cgroup_kn_lock_live().
- Fix integer overflow in rdmacg_try_charge() when usage equals INT_MAX
by performing the increment in s64.
- Fix asymmetric DL bandwidth accounting on cpuset attach rollback by
recording the CPU used by dl_bw_alloc() so cancel_attach() returns
the reservation to the same root domain.
- Fix nr_dying_subsys_* race that briefly showed 0 in cgroup.stat after
rmdir by incrementing from kill_css() instead of offline_css().
- Typo fix in cgroup-v2 documentation.
-----BEGIN PGP SIGNATURE-----
iIQEABYKACwWIQTfIjM1kS57o3GsC/uxYfJx3gVYGQUCae+xjw4cdGpAa2VybmVs
Lm9yZwAKCRCxYfJx3gVYGaIUAQD2hJ7ELRDXAtXzL1Ck1zH8vESvbX8syFfkSO6L
IgtovQEA4Tk7/RIO3HfBxBjgp6Q5vo7C7Biz4ye7fCu/ry7x3Qk=
=pypQ
-----END PGP SIGNATURE-----
Merge tag 'cgroup-for-7.1-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup
Pull cgroup fixes from Tejun Heo:
- Fix UAF race in psi pressure_write() against cgroup file release by
extending cgroup_mutex coverage and ordering of->priv access after
cgroup_kn_lock_live()
- Fix integer overflow in rdmacg_try_charge() when usage equals INT_MAX
by performing the increment in s64
- Fix asymmetric DL bandwidth accounting on cpuset attach rollback by
recording the CPU used by dl_bw_alloc() so cancel_attach() returns
the reservation to the same root domain
- Fix nr_dying_subsys_* race that briefly showed 0 in cgroup.stat after
rmdir by incrementing from kill_css() instead of offline_css()
- Typo fix in cgroup-v2 documentation
* tag 'cgroup-for-7.1-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
docs: cgroup: fix typo 'protetion' -> 'protection'
cgroup: Increment nr_dying_subsys_* from rmdir context
cgroup/cpuset: record DL BW alloc CPU for attach rollback
cgroup/rdma: fix integer overflow in rdmacg_try_charge()
sched/psi: fix race between file release and pressure write
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEq1nRK9aeMoq1VSgcnJ2qBz9kQNkFAmnvhuAACgkQnJ2qBz9k
QNn7NQgAhNZiUl5HQ2/+6XjrhrbaLTiwYLFYX36MlVcy6jZZ7PABEn4Iq+dKn19L
EQ6WL97yEAOV8A3wdXLnm3J/euf37JlWApuNWqmgA5ZsJ0p14nXKbhd/jcqSf5LF
K4afO0OlN6WrPPJxxY2KfiaElgf9bPAhSkX/JvV5pEVyz0CcAjFfgwmNhxrYxIY1
DfQJ/2w7G1VdgpxeO1kVW5REH5NvbsWj9IQSSRS9r1HzTa7E28e3Zn75XWcdab3I
pt3E03nuUiyfVTmJXi2/HGNb40XZjH5TCeDrsbjo759ZdiPIvWDUpcTx0+5acOaj
b039wWZKKBFTag4KA4yPMkEV37ROiA==
=+sU8
-----END PGP SIGNATURE-----
Merge tag 'fs_for_v7.1-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
Pull isofs and udf fixes from Jan Kara:
"Several isofs and udf fixes"
* tag 'fs_for_v7.1-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
docs: isofs: replace dead ECMA-119 FTP link
udf: reject descriptors with oversized CRC length
isofs: use QSTR_LEN() in isofs_cmp
isofs: validate block number from NFS file handle in isofs_export_iget
isofs: validate Rock Ridge CE continuation extent against volume size
ax_spi_fill_tx_fifo() can be called from ax_spi_irq() which is a hard
irq handler. Replace usleep_range(10, 10) with udelay(10) in atomic
context.
Fixes: e75a6b00ad ("spi: axiado: Add driver for Axiado SPI DB controller")
Signed-off-by: Felix Gu <ustc.gu@gmail.com>
Link: https://patch.msgid.link/20260428-axiado-v1-1-cd767500af72@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
-----BEGIN PGP SIGNATURE-----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=ajKs
-----END PGP SIGNATURE-----
Merge tag 'for-7.1-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
- space reservation fixes:
- correctly undo 'may_use' accounting for remap tree
- avoid double decrement of 'may_use' when submitting async io
- actually enable the shutdown ioctl callback (not just the superblock
ops)
- raid stripe tree fixes when deleting extents
- add missing error handling
- fix various incorrect values set
- fix transaction state when removing a directory, possibly leading to
EIO during log replay
- additional b-tree node key checks during metadata readahead
- error handling and transaction abort updates
* tag 'for-7.1-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: fix double-decrement of bytes_may_use in submit_one_async_extent()
btrfs: check return value of btrfs_partially_delete_raid_extent()
btrfs: handle -EAGAIN from btrfs_duplicate_item and refresh stale leaf pointer
btrfs: replace ASSERT with proper error handling in stripe lookup fallback
btrfs: fix wrong min_objectid in btrfs_previous_item() call
btrfs: fix raid stripe search missing entries at leaf boundaries
btrfs: copy devid in btrfs_partially_delete_raid_extent()
btrfs: handle unexpected free-space-tree key types
btrfs: fix missing last_unlink_trans update when removing a directory
btrfs: don't clobber errors in add_remap_tree_entries()
btrfs: enable shutdown ioctl for non-experimental builds
btrfs: apply first key check for readahead when possible
btrfs: abort transaction in do_remap_reloc_trans() on failure
btrfs: fix bytes_may_use leak in do_remap_reloc_trans()
btrfs: fix bytes_may_use leak in move_existing_remap()
Move lsm_get_xattr_slot() below the SBLABEL_MNT check so we don't leave
a NULL-named slot in the array when returning -EOPNOTSUPP; filesystem
initxattrs() callbacks stop iterating at the first NULL ->name, silently
dropping xattrs installed by later LSMs.
Cc: stable@vger.kernel.org
Signed-off-by: David Windsor <dwindsor@gmail.com>
Signed-off-by: Paul Moore <paul@paul-moore.com>
SELinux socket state lives in the composite LSM socket blob.
sock_has_perm() and nlmsg_sock_has_extended_perms() currently
dereference sk->sk_security directly, which assumes the SELinux socket
blob is at offset zero.
In stacked configurations that assumption does not hold. If another LSM
allocates socket blob storage before SELinux, these helpers may read the
wrong blob and feed invalid SID and class values into AVC checks.
Use selinux_sock() instead of accessing sk->sk_security directly.
Fixes: d1d991efaf ("selinux: Add netlink xperm support")
Cc: stable@vger.kernel.org # v6.13+
Signed-off-by: Zongyao Chen <ZongYao.Chen@linux.alibaba.com>
Signed-off-by: Paul Moore <paul@paul-moore.com>
If a bus has hotplug slots that implement the slot's reset_slot callback,
it is not safe to do the non-slot specific bus reset, so don't fallback to
it. If a slot reset does fail, the subsequent bus reset will attempt a 2nd
link reset on top of previous and fail to handle the hotplug events.
Fixes: 8238cb69c0 ("PCI: Make reset_subordinate hotplug safe")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Link: https://patch.msgid.link/20260421150644.3543733-1-kbusch@meta.com
The cdrom core never calls set_disk_ro() for a registered device, so
BLKROGET on a CD-ROM device always returns 0 (writable), even when the
drive has no write capabilities and writes will inevitably fail. This
causes problems for userspace that relies on BLKROGET to determine
whether a block device is read-only. For example, systemd's loop device
setup uses BLKROGET to decide whether to create a loop device with
LO_FLAGS_READ_ONLY. Without the read-only flag, writes pass through the
loop device to the CD-ROM and fail with I/O errors. systemd-fsck
similarly checks BLKROGET to decide whether to run fsck in no-repair
mode (-n).
The write-capability bits in cdi->mask come from two different sources:
CDC_DVD_RAM and CDC_CD_RW are populated by the driver from the MODE
SENSE capabilities page (page 0x2A) before register_cdrom() is called,
while CDC_MRW_W and CDC_RAM require the MMC GET CONFIGURATION command
and were only probed by cdrom_open_write() at device open time. This
meant that any attempt to compute the writable state from the full
mask at probe time was incorrect, because the GET CONFIGURATION bits
were still unset (and cdi->mask is initialized such that capabilities
are assumed present).
Fix this by factoring the GET CONFIGURATION probing out of
cdrom_open_write() into a new exported helper,
cdrom_probe_write_features(), and having sr call it from sr_probe()
right after get_capabilities() has populated the MODE SENSE bits.
register_cdrom() then calls set_disk_ro() based on the full
write-capability mask (CDC_DVD_RAM | CDC_MRW_W | CDC_RAM | CDC_CD_RW)
so the block layer reflects the drive's actual write support. The
feature queries used (CDF_MRW and CDF_RWRT via GET CONFIGURATION with
RT=00) report drive-level capabilities that are persistent across
media, so a single probe before register_cdrom() is sufficient and the
redundant probe at open time is dropped.
With set_disk_ro() now accurate, the long-vestigial cd->writeable flag
in sr can go: get_capabilities() used to set cd->writeable based on
the same four mask bits, but because CDC_MRW_W and CDC_RAM default to
"capability present" in cdi->mask and aren't touched by MODE SENSE,
the condition that gated cd->writeable was always true, making it
unconditionally 1. Replace the corresponding gate in sr_init_command()
with get_disk_ro(cd->disk), which turns a previously no-op check into
a real one and also catches kernel-internal bio writers that bypass
blkdev_write_iter()'s bdev_read_only() check.
The sd driver (SCSI disks) does not have this problem because it
checks the MODE SENSE Write Protect bit and calls set_disk_ro()
accordingly. The sr driver cannot use the same approach because the
MMC specification does not define the WP bit in the MODE SENSE
device-specific parameter byte for CD-ROM devices.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Daan De Meyer <daan@amutable.com>
Reviewed-by: Phillip Potter <phil@philpotter.co.uk>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Phillip Potter <phil@philpotter.co.uk>
Link: https://patch.msgid.link/20260427210139.1400-2-phil@philpotter.co.uk
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The header references struct notifier_block but neither includes
linux/notifier.h nor contains the relevant forward declaration.
Add the latter for correctness.
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
[ rjw: Subject tweak ]
Link: https://patch.msgid.link/20260427112238.132419-1-bartosz.golaszewski@oss.qualcomm.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The HP OMEN 16 Gaming Laptop (board name 8A44) has a mux-less hybrid
GPU configuration with AMD Rembrandt (Radeon 680M) and NVIDIA GA104
(RTX 3070 Ti). The internal eDP panel is wired to the AMD iGPU.
When Nouveau loads without GSP firmware, the ACPI video backlight
device (acpi_video0) gets registered alongside the native AMD
backlight (amdgpu_bl2). In this state, writes to amdgpu_bl2 update
the software brightness value but fail to change the physical panel
brightness.
Force native backlight to prevent acpi_video0 from registering.
Confirmed that booting with acpi_backlight=native resolves the
issue.
Cc: All applicable <stable@vger.kernel.org>
Signed-off-by: Shivam Kalra <shivamkalra98@zohomail.in>
Link: https://patch.msgid.link/20260426-omen-16-backlight-fix-v1-1-62364f268ea6@zohomail.in
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Since rtc_tm_to_ktime() may overflow for large RTC time values and
full second granularity is sufficient in timer value computations
in acpi_tad_rtc_set_alarm() and acpi_tad_rtc_read_alarm(), use
rtc_tm_to_time64() instead of that function, which also allows the
computations to be simplified.
Moreover, U32_MAX is a special "timer disabled" value, so make
acpi_tad_rtc_set_alarm() reject it when attempting to program the
alarm timers.
Fixes: 7572dcabe3 ("ACPI: TAD: Add alarm support to the RTC class device interface")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Link: https://patch.msgid.link/3414608.aeNJFYEL58@rafael.j.wysocki
The code in acpi_tad_remove() needs to run after the unregistration of
the devres-managed RTC class device so that it doesn't race with the
class callbacks of the latter.
To make that happen, pass it to devm_add_action_or_reset() before
registering the RTC class device.
Fixes: 7572dcabe3 ("ACPI: TAD: Add alarm support to the RTC class device interface")
Fixes: 8a1e7f4b17 ("ACPI: TAD: Add RTC class device interface")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://patch.msgid.link/14001754.uLZWGnKmhe@rafael.j.wysocki
Recent commit 93afe8ba9b ("ACPI: TAD: Use dev_groups in struct
device_driver") switched over the ACPI TAD driver to using device
attribute groups instead of creating and removing the device sysfs
attributes directly, but it might go one step farther and use the
__ATTRIBUTE_GROUPS() macro which would reduce the code size slightly.
Do it now.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
[ rjw: Fixed typo in the changelog ]
Link: https://patch.msgid.link/1961102.tdWV9SEqCh@rafael.j.wysocki
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When concurrently bringing up and down two SMT threads of a physical
core, many warning call traces occur as below:
The issue timeline is as follows:
1. When the system starts,
cpufreq: CPU: 220, policy->related_cpus: 220-221, policy->cpus: 220-221
2. Offline CPU 220 and CPU 221.
3. Online CPU 220
- CPU 221 is now offline, as acpi_get_psd_map() use
for_each_online_cpu(), so the cpu_data->shared_cpu_map,
policy->cpus, and related_cpus has only CPU 220.
cpufreq: CPU: 220, policy->related_cpus: 220, policy->cpus: 220
4. Offline CPU 220
5. Online CPU 221, the below call trace occurs:
- Since CPU 220 and CPU 221 share one policy, and
policy->related_cpus = 220 after step 3, so CPU 221
is not in policy->related_cpus but
per_cpu(cpufreq_cpu_data, cpu221) is not NULL.
After reverting commit 56eb0c0ed3 ("ACPI: CPPC: Fix remaining
for_each_possible_cpu() to use online CPUs"), the issue disappeared.
The _PSD (P-State Dependency) defines the hardware-level dependency of
frequency control across CPU cores. Since this relationship is a physical
attribute of the hardware topology, it remains constant regardless of the
online or offline status of the CPUs.
Using for_each_online_cpu() in acpi_get_psd_map() is problematic. If a
CPU is offline, it will be excluded from the shared_cpu_map.
Consequently, if that CPU is brought online later, the kernel will fail
to recognize it as part of any shared frequency domain.
Switch back to for_each_possible_cpu() to ensure that all cores defined
in the ACPI tables are correctly mapped into their respective performance
domains from the start. This aligns with the logic of policy->related_cpus,
which must encompass all potentially available cores in the domain to
prevent logic gaps during CPU hotplug operations.
To resolve the original issue regarding the "nosmt" or "nosmt=force"
boot parameter, as send_pcc_cmd() function already does if (!desc)
continue, so reverting that loop back to for_each_possible_cpu() is ok,
only need to change the match_cpc_ptr NULL case in acpi_get_psd_map() to
continue as Sean suggested.
How to reproduce, on arm64 machine with SMT support which use acpi cppc
cpufreq driver:
bash test.sh 220 & bash test.sh 221 &
The test.sh is as below:
while true
do
echo 0 > /sys/devices/system/cpu/cpu${1}/online
sleep 0.5
cat /sys/devices/system/cpu/cpu${1}/cpufreq/related_cpus
echo 1 > /sys/devices/system/cpu/cpu${1}/online
cat /sys/devices/system/cpu/cpu${1}/cpufreq/related_cpus
done
CPU: 221 PID: 1119 Comm: cpuhp/221 Kdump: loaded Not tainted 6.6.0debug+ #5
Hardware name: To be filled by O.E.M. S920X20/BC83AMDA01-7270Z, BIOS 20.39 09/04/2024
pstate: a1400009 (NzCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
pc : cpufreq_online+0x8ac/0xa90
lr : cpuhp_cpufreq_online+0x18/0x30
sp : ffff80008739bce0
x29: ffff80008739bce0 x28: 0000000000000000 x27: ffff28400ca32200
x26: 0000000000000000 x25: 0000000000000003 x24: ffffd483503ff000
x23: ffffd483504051a0 x22: ffffd48350024a00 x21: 00000000000000dd
x20: 000000000000001d x19: ffff28400ca32000 x18: 0000000000000000
x17: 0000000000000020 x16: ffffd4834e6a3fc8 x15: 0000000000000020
x14: 0000000000000008 x13: 0000000000000001 x12: 00000000ffffffff
x11: 0000000000000040 x10: ffffd48350430728 x9 : ffffd4834f087c78
x8 : 0000000000000001 x7 : ffff2840092bdf00 x6 : ffffd483504264f0
x5 : ffffd48350405000 x4 : ffff283f7f95cc60 x3 : 0000000000000000
x2 : ffff53bc2f94b000 x1 : 00000000000000dd x0 : 0000000000000000
Call trace:
cpufreq_online+0x8ac/0xa90
cpuhp_cpufreq_online+0x18/0x30
cpuhp_invoke_callback+0x128/0x580
cpuhp_thread_fun+0x110/0x1b0
smpboot_thread_fn+0x140/0x190
kthread+0xec/0x100
ret_from_fork+0x10/0x20
---[ end trace 0000000000000000 ]---
Cc: All applicable <stable@vger.kernel.org>
Fixes: 56eb0c0ed3 ("ACPI: CPPC: Fix remaining for_each_possible_cpu() to use online CPUs")
Co-developed-by: Sean Kelley <skelley@nvidia.com>
Signed-off-by: Sean Kelley <skelley@nvidia.com>
Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
[ rjw: Changelog edits ]
Link: https://patch.msgid.link/20260417040112.3727756-1-ruanjinjie@huawei.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
