Fedor Pchelkin [Mon, 30 Mar 2026 18:45:40 +0000 (21:45 +0300)]
net: macb: fix clk handling on PCI glue driver removal
platform_device_unregister() may still want to use the registered clks
during runtime resume callback.
Note that there is a commit d82d5303c4c5 ("net: macb: fix use after free
on rmmod") that addressed the similar problem of clk vs platform device
unregistration but just moved the bug to another place.
Save the pointers to clks into local variables for reuse after platform
device is unregistered.
BUG: KASAN: use-after-free in clk_prepare+0x5a/0x60
Read of size 8 at addr ffff888104f85e00 by task modprobe/597
Christos Longros [Sun, 29 Mar 2026 07:45:50 +0000 (09:45 +0200)]
wifi: rtw89: fix typo "frome" -> "from" in rx_freq_frome_ie
The chip_info field rx_freq_frome_ie (RX frequency from Information
Element) has a typo. The function that uses it is already spelled
correctly: rtw89_core_update_rx_freq_from_ie. Rename the field to
match.
Julian Braha [Sun, 29 Mar 2026 18:53:48 +0000 (19:53 +0100)]
net: microchip: dead code cleanup in kconfig for FDMA
The Kconfig in the parent directory already has the first 'if NET_VENDOR_MICROCHIP'
gating the inclusion of this Kconfig, meaning that the second
'if NET_VENDOR_MICROCHIP' condition is effectively dead code.
I propose removing the second 'if NET_VENDOR_MICROCHIP' in
drivers/net/ethernet/microchip/fdma/Kconfig
This dead code was found by kconfirm, a static analysis tool for Kconfig.
Srujana Challa [Thu, 26 Mar 2026 14:23:44 +0000 (19:53 +0530)]
virtio_net: clamp rss_max_key_size to NETDEV_RSS_KEY_LEN
rss_max_key_size in the virtio spec is the maximum key size supported by
the device, not a mandatory size the driver must use. Also the value 40
is a spec minimum, not a spec maximum.
The current code rejects RSS and can fail probe when the device reports a
larger rss_max_key_size than the driver buffer limit. Instead, clamp the
effective key length to min(device rss_max_key_size, NETDEV_RSS_KEY_LEN)
and keep RSS enabled.
This keeps probe working on devices that advertise larger maximum key sizes
while respecting the netdev RSS key buffer size limit.
Fixes: 3f7d9c1964fc ("virtio_net: Add hash_key_length check") Cc: stable@vger.kernel.org Signed-off-by: Srujana Challa <schalla@marvell.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Link: https://patch.msgid.link/20260326142344.1171317-1-schalla@marvell.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Yucheng Lu [Tue, 31 Mar 2026 08:00:21 +0000 (16:00 +0800)]
net/sched: sch_netem: fix out-of-bounds access in packet corruption
In netem_enqueue(), the packet corruption logic uses
get_random_u32_below(skb_headlen(skb)) to select an index for
modifying skb->data. When an AF_PACKET TX_RING sends fully non-linear
packets over an IPIP tunnel, skb_headlen(skb) evaluates to 0.
Passing 0 to get_random_u32_below() takes the variable-ceil slow path
which returns an unconstrained 32-bit random integer. Using this
unconstrained value as an offset into skb->data results in an
out-of-bounds memory access.
Fix this by verifying skb_headlen(skb) is non-zero before attempting
to corrupt the linear data area. Fully non-linear packets will silently
bypass the corruption logic.
Fixes: c865e5d99e25 ("[PKT_SCHED] netem: packet corruption option") Reported-by: Yifan Wu <yifanwucs@gmail.com> Reported-by: Juefei Pu <tomapufckgml@gmail.com> Signed-off-by: Yuan Tan <tanyuan98@outlook.com> Signed-off-by: Xin Liu <bird@lzu.edu.cn> Signed-off-by: Yuhang Zheng <z1652074432@gmail.com> Signed-off-by: Yucheng Lu <kanolyc@gmail.com> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org> Link: https://patch.msgid.link/45435c0935df877853a81e6d06205ac738ec65fa.1774941614.git.kanolyc@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Jakub Kicinski [Thu, 2 Apr 2026 02:19:35 +0000 (19:19 -0700)]
Merge tag 'nf-26-04-01' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf
Pablo Neira Ayuso says:
====================
Netfilter fixes for net
The following patchset contains Netfilter fixes for net. Note that most
of the bugs fixed here are >5 years old. The large PR is not due to an
increase in regressions.
1) Flowtable hardware offload support in IPv6 can lead to out-of-bounds
when populating the rule action array when combined with double-tagged
vlan. Bump the maximum number of actions from 16 to 24 and check that
such limit is never reached, otherwise bail out. This bugs stems from
the original flowtable hardware offload support.
2) nfnetlink_log does not include the netlink header size of the trailing
NLMSG_DONE message when calculating the skb size. From Florian Westphal.
3) Reject names in xt_cgroup and xt_rateest extensions which are not
nul-terminated. Also from Florian.
4) Use nla_strcmp in ipset lookup by set name, since IPSET_ATTR_NAME and
IPSET_ATTR_NAMEREF are of NLA_STRING type. From Florian Westphal.
5) When unregistering conntrack helpers, pass the helper that is going
away so the expectation cleanup is done accordingly, otherwise UaF is
possible when accessing expectation that refer to the helper that is
gone. From Qi Tang.
6) Zero expectation NAT fields to address leaking kernel memory through
the expectation netlink dump when unset. Also from Qi Tang.
7) Use the master conntrack helper when creating expectations via
ctnetlink, ignore the suggested helper through CTA_EXPECT_HELP_NAME.
This allows to address a possible read of kernel memory off the
expectation object boundary.
8) Fix incorrect release of the hash bucket logic in ipset when the
bucket is empty, leading to shrinking the hash bucket to size 0
which deals to out-of-bound write in next element additions.
From Yifan Wu.
9) Allow the use of x_tables extensions that explicitly declare
NFPROTO_ARP support only. This is to avoid an incorrect hook number
validation due to non-overlapping arp and inet hook number
definitions.
10) Reject immediate NF_QUEUE verdict in nf_tables. The userspace
nft tool always uses the nft_queue expression for queueing.
This ensures this verdict cannot be used for the arp family,
which does supported this.
* tag 'nf-26-04-01' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
netfilter: nf_tables: reject immediate NF_QUEUE verdict
netfilter: x_tables: restrict xt_check_match/xt_check_target extensions for NFPROTO_ARP
netfilter: ipset: drop logically empty buckets in mtype_del
netfilter: ctnetlink: ignore explicit helper on new expectations
netfilter: ctnetlink: zero expect NAT fields when CTA_EXPECT_NAT absent
netfilter: nf_conntrack_helper: pass helper to expect cleanup
netfilter: ipset: use nla_strcmp for IPSET_ATTR_NAME attr
netfilter: x_tables: ensure names are nul-terminated
netfilter: nfnetlink_log: account for netlink header size
netfilter: flowtable: strictly check for maximum number of actions
====================
We've added 2 non-merge commits during the last 2 day(s) which contain
a total of 3 files changed, 139 insertions(+), 23 deletions(-).
The main changes are:
1) skb_dst_drop(skb) when bpf prog does a encap or decap,
from Jakub Kicinski
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next:
selftests/bpf: Test that dst is cleared on same-protocol encap
net: Clear the dst when performing encap / decap
====================
bpf: sockmap: Fix use-after-free of sk->sk_socket in sk_psock_verdict_data_ready().
syzbot reported use-after-free of AF_UNIX socket's sk->sk_socket
in sk_psock_verdict_data_ready(). [0]
In unix_stream_sendmsg(), the peer socket's ->sk_data_ready() is
called after dropping its unix_state_lock().
Although the sender socket holds the peer's refcount, it does not
prevent the peer's sock_orphan(), and the peer's sk_socket might
be freed after one RCU grace period.
Let's fetch the peer's sk->sk_socket and sk->sk_socket->ops under
RCU in sk_psock_verdict_data_ready().
[0]:
BUG: KASAN: slab-use-after-free in sk_psock_verdict_data_ready+0xec/0x590 net/core/skmsg.c:1278
Read of size 8 at addr ffff8880594da860 by task syz.4.1842/11013
Julian Braha [Mon, 30 Mar 2026 21:32:58 +0000 (22:32 +0100)]
ppp: dead code cleanup in Kconfig
There is already an 'if PPP' condition wrapping several config options
e.g. PPP_MPPE and PPPOE, making the 'depends on PPP' statement for each of
these a duplicate dependency (dead code).
I propose leaving the outer 'if PPP...endif' and removing the individual
'depends on PPP' statement from each option.
This dead code was found by kconfirm, a static analysis tool for Kconfig.
Weiming Shi [Mon, 30 Mar 2026 16:32:38 +0000 (00:32 +0800)]
rds: ib: reject FRMR registration before IB connection is established
rds_ib_get_mr() extracts the rds_ib_connection from conn->c_transport_data
and passes it to rds_ib_reg_frmr() for FRWR memory registration. On a
fresh outgoing connection, ic is allocated in rds_ib_conn_alloc() with
i_cm_id = NULL because the connection worker has not yet called
rds_ib_conn_path_connect() to create the rdma_cm_id. When sendmsg() with
RDS_CMSG_RDMA_MAP is called on such a connection, the sendmsg path parses
the control message before any connection establishment, allowing
rds_ib_post_reg_frmr() to dereference ic->i_cm_id->qp and crash the
kernel.
The existing guard in rds_ib_reg_frmr() only checks for !ic (added in
commit 9e630bcb7701), which does not catch this case since ic is allocated
early and is always non-NULL once the connection object exists.
Add a check in rds_ib_get_mr() that verifies ic, i_cm_id, and qp are all
non-NULL before proceeding with FRMR registration, mirroring the guard
already present in rds_ib_post_inv(). Return -ENODEV when the connection
is not ready, which the existing error handling in rds_cmsg_send() converts
to -EAGAIN for userspace retry and triggers rds_conn_connect_if_down() to
start the connection worker.
Fixes: 1659185fb4d0 ("RDS: IB: Support Fastreg MR (FRMR) memory registration mode") Reported-by: Xiang Mei <xmei5@asu.edu> Signed-off-by: Weiming Shi <bestswngs@gmail.com> Reviewed-by: Allison Henderson <achender@kernel.org> Link: https://patch.msgid.link/20260330163237.2752440-2-bestswngs@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When two CPUs process Router Advertisement packets for the same router
simultaneously, they can both arrive at fib6_metric_set() with the same
fib6_info pointer whose fib6_metrics still points to dst_default_metrics.
Fix this by:
- Set val for p->metrics before published via cmpxchg() so the metrics
value is ready before the pointer becomes visible to other CPUs.
- Replace the plain pointer store with cmpxchg() and free the allocation
safely when competition failed.
- Add READ_ONCE()/WRITE_ONCE() for metrics[] setting in the non-default
metrics path to prevent compiler-based data races.
Fixes: d4ead6b34b67 ("net/ipv6: move metrics from dst to rt6_info") Reported-by: Fei Liu <feliu@redhat.com> Reviewed-by: Jiayuan Chen <jiayuan.chen@linux.dev> Signed-off-by: Hangbin Liu <liuhangbin@gmail.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Link: https://patch.msgid.link/20260331-b4-fib6_metric_set-kmemleak-v3-1-88d27f4d8825@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
i2c: xiic: skip input clock setup on non-OF systems
Currently Linux does not implement ACPI ClockInput() resource to describe
clocks, unlike DT. However the xiic driver is happy if something
magically enables the clock before the driver probes, and does not
turn it off again. The clock should always be considered optional for
ACPI.
Switch from i2c_add_adapter() to i2c_add_numbered_adapter() to enable
platforms to specify fixed I2C bus numbers via the platform device ID.
This allows systems to maintain consistent bus numbering across reboots.
On platforms where the device ID is PLATFORM_DEVID_NONE (the default),
the adapter falls back to dynamic allocation, preserving backward
compatibility.
Anthony Pighin [Tue, 31 Mar 2026 18:26:32 +0000 (14:26 -0400)]
i2c: imx: zero-initialize dma_slave_config for eDMA
commit 66d88e16f204 ("dmaengine: fsl-edma: read/write multiple registers
in cyclic transactions") causes fsl_edma_fill_tcd() to read
dst_port_window_size and src_port_window_size when building transfer
control descriptors.
Initialize the structure so unset fields are explicitly zero.
Aaron Tomlin [Wed, 25 Mar 2026 00:11:59 +0000 (20:11 -0400)]
fs/resctrl: Add "*" shorthand to set io_alloc CBM for all domains
Configuring the io_alloc_cbm interface requires an explicit domain ID for each
cache domain. On systems with high core counts and numerous cache clusters,
this requirement becomes cumbersome for automation and management tasks that
aim to apply a uniform policy.
Introduce a wildcard domain ID selector "*" for the io_alloc_cbm interface.
This enables users to set the same Capacity Bitmask (CBM) across all cache
domains in a single operation.
Aaron Tomlin [Wed, 25 Mar 2026 00:11:58 +0000 (20:11 -0400)]
fs/resctrl: Report invalid domain ID when parsing io_alloc_cbm
The last_cmd_status file is intended to report details about the most recent
resctrl filesystem operation, specifically to aid in diagnosing failures.
However, when parsing io_alloc_cbm, if a user provides a domain ID that does
not exist in the resource, the operation fails with -EINVAL without updating
last_cmd_status. This results in inconsistent behaviour where the system call
returns an error, but last_cmd_status misleadingly reports "ok", leaving the
user unaware that the failure was caused by an invalid domain ID.
Write an error message to last_cmd_status when the target domain ID cannot
be found.
i2c: rtl9300: introduce new function properties to driver data
Due to the very nature of differences between RTL9607C i2c controller
and RTL9300 / RTL9310 that are incompatible with each other in some areas
of this driver, for example in clock configuration, channel configuration
and initialization at the end of the probe, introduce new function
properties to the driver data struct to handle those differences.
With these new properties, create configuration functions for RTL9300 and
RTL9310 and assign them to their respective driver data structs.
i2c: rtl9300: introduce clk struct for upcoming rtl9607 support
In RTL9607C i2c controller, there is 10 bit CLK_DIV field for
setting the clock of i2c interface which depends on the rate
of i2c clk (which seems be fixed to 62.5MHz according to Realtek SDK).
Introduce the clk struct and the respective F_CLK_DIV and clk_div
which are going to be used in the upcoming patch for rtl9607c i2c
controller support addition.
devm_clk_get_optional_enabled() function was used for cleaner code
as it automatically returns NULL if the clk is not present, which is
going to be the case for RTL9300 and RTL9310 i2c controllers.
dt-bindings: i2c: realtek,rtl9301-i2c: extend for clocks and RTL9607C support
Add the "realtek,rtl9607-i2c" compatible for i2c controller on the
RTL9607C SoC series.
Add a clocks property to the properties to describe the i2c reference
clock and make it available for all the compatibles. This i2c reference
clock is assumed to be coming from switchcore region via Lexra bus as
the other SoC peripherals.
According to the info available about the existing devices, they also
have the i2c master controller clocks.
RTL9607C requires the "realtek,scl" and "clocks" to be specified
and so handle it under separate if check for "realtek,rtl9607-i2c".
i2c: rtl9300: introduce a property for 8 bit width reg address
In RTL9607C i2c controller, in order to indicate that the width of
memory address is 8 bits, 0 is written to MEM_ADDR_WIDTH field as
opposed to 1 for RTL9300 and RTL9310.
Introduce a new property to a driver data to indicate what value
need to written to MEM_ADDR_WIDTH field for this case.
i2c: rtl9300: introduce F_BUSY to the reg_fields struct
In RTL9607C i2c controller the busy check operation is done on the
separate bit of the command register as opposed to self clearing
command trigger bit on the rtl9300 and rtl9310 i2c controllers.
Introduce a new F_BUSY field to the reg_fields struct for that
and change the regmap read poll function to use F_BUSY
instead of I2C_TRIG.
i2c: rtl9300: introduce max length property to driver data
In RTL9607C i2c controller, theoretical maximum the data length
can be is 4 bytes as opposed to 16 bytes on rtl9300 and rtl9310.
Introduce a new property to the driver data struct for that.
Adjust if statement in prepare_xfer function to follow that new
property instead of the hardcoded value.
Merge tag 'renesas-arm-soc-for-v7.1-tag1' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel into soc/arm
Renesas ARM SoC updates for v7.1
- Use the of_phandle_args_equal() helper.
* tag 'renesas-arm-soc-for-v7.1-tag1' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel:
ARM: shmobile: rcar-gen2: Use of_phandle_args_equal() helper
Merge tag 'imx-soc-7.1' of git://git.kernel.org/pub/scm/linux/kernel/git/frank.li/linux into soc/drivers
i.MX SoC update for 7.1:
- Updates MAINTAINERS file to include i.MX team coverage for ARM NXP platforms
- Sets default values for OPACR (Off-Platform Peripheral Access Control
Register) in the i.MX AIPSTZ bus driver
* tag 'imx-soc-7.1' of git://git.kernel.org/pub/scm/linux/kernel/git/frank.li/linux:
MAINTAINERS: Add i.MX team to all arm NXP platforms
bus: imx-aipstz: set default value for opacr registers
Jan Kantert [Fri, 27 Feb 2026 11:11:34 +0000 (12:11 +0100)]
i2c: rtl9300: add support for 50 kHz and 2.5 MHz bus speeds
Some SFP modules on certain switches (for example the ONTi ONT-S508CL-8S and
XikeStor SKS8300-8X) exhibit unreliable I2C communication at the currently
supported speeds. Add support for 50 kHz and 2.5 MHz I2C bus modes on the
RTL9300 to improve compatibility with these devices.
Merge tag 'renesas-drivers-for-v7.1-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel into soc/drivers
Renesas driver updates for v7.1 (take two)
- Mark remaining rz_sysc_init_data structures __initconst.
* tag 'renesas-drivers-for-v7.1-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel:
soc: renesas: r9a09g056-sys: Mark rzv2n_sys_init_data as __initconst
soc: renesas: r9a09g047-sys: Mark rzg3e_sys_init_data as __initconst
soc: renesas: r9a09g057-sys: Mark rzv2h_sys_init_data as __initconst
Julian Ruess [Wed, 25 Mar 2026 13:31:24 +0000 (14:31 +0100)]
vfio/ism: Implement vfio_pci driver for ISM devices
Add a vfio_pci variant driver for the s390-specific Internal Shared
Memory (ISM) devices used for inter-VM communication.
This enables the development of vfio-pci-based user space drivers for
ISM devices.
On s390, kernel primitives such as ioread() and iowrite() are switched
over from function-handle-based PCI load/stores instructions to PCI
memory-I/O (MIO) loads/stores when these are available and not
explicitly disabled. Since these instructions cannot be used with ISM
devices, ensure that classic function-handle-based PCI instructions are
used instead.
The driver is still required even when MIO instructions are disabled, as
the ISM device relies on the PCI store block (PCISTB) instruction to
perform write operations.
Stores are not fragmented, therefore one ioctl corresponds to exactly
one PCISTB instruction. User space must ensure to not write more than
4096 bytes at once to an ISM BAR which is the maximum payload of the
PCISTB instruction.
Reviewed-by: Alexandra Winter <wintera@linux.ibm.com> Reviewed-by: Niklas Schnelle <schnelle@linux.ibm.com> Signed-off-by: Julian Ruess <julianr@linux.ibm.com> Reviewed-by: Farhan Ali <alifm@linux.ibm.com> Link: https://lore.kernel.org/r/20260325-vfio_pci_ism-v8-2-ddc504cde914@linux.ibm.com Signed-off-by: Alex Williamson <alex@shazbot.org>
Let's modify the awk script to add the values for each line and print
the total. The variable 'i' has a number of input lines that have valid
output and variable 'c' has the sum of actual counter values. That way
it should work on any platforms.
Reviewed-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 31 Mar 2026 18:05:20 +0000 (11:05 -0700)]
perf tests: Write test files to tmpdir
Writing to the test output files in the current working directory can
fail in various contexts such as continual test. Other tests write to
a mktemp-ed file, make the "perf script task-analyszer tests" follow
this convention too. Currently this isn't possible for the perf.data
file due to a lack of perf script support, add a variable for when
this support is available.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 31 Mar 2026 18:29:48 +0000 (11:29 -0700)]
libperf cpumap: Make index and nr types unsigned
The index into the cpumap array and the number of entries within the
array can never be negative, so let's make them unsigned. This is
prompted by reports that gcc 13 with -O6 is giving a
alloc-size-larger-than errors. The change makes the cpumap changes and
then updates the declaration of index variables throughout perf and
libperf to be unsigned. The two things are hard to separate as
compiler warnings about mixing signed and unsigned types breaks the
build.
Reported-by: Chingbin Li <liqb365@163.com> Closes: https://lore.kernel.org/lkml/20260212025127.841090-1-liqb365@163.com/ Tested-by: Chingbin Li <liqb365@163.com> Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Merge tag 'stm32-bus-firewall-for-7.1-1' of git://git.kernel.org/pub/scm/linux/kernel/git/atorgue/stm32 into soc/drivers
STM32 Firewall bus for v7.1, round 1
Highlights:
----------
Stm32 SoCs embed debug peripherals such as Coresight. These peripherals
can monitor the activity of the cores. Because of that, they can be
used only if some features in the debug configuration are enabled.
Else, errors or firewall exceptions can be observed. Similarly to
the ETZPC(on stm32mp1x platforms) or the RIFSC(on stm32mp2x platforms),
debug-related peripherals access can be assessed at bus level to
prevent these issues from happening.
The debug configuration can only be accessed by the secure world.
That means that a service must be implemented in the secure world for
the kernel to check the firewall configuration. On OpenSTLinux, it is
done through a Debug access PTA in OP-TEE [1].
To represent the debug peripherals present on a dedicated debug bus,
create a debug bus node in the device tree and the associated driver
that will interact with this PTA.
Plus some fixes.
* tag 'stm32-bus-firewall-for-7.1-1' of git://git.kernel.org/pub/scm/linux/kernel/git/atorgue/stm32:
pinctrl: stm32: add firewall checks before probing the HDP driver
drivers: bus: add the stm32 debug bus driver
bus: stm32_firewall: add stm32_firewall_get_grant_all_access() API
bus: stm32_firewall: allow check on different firewall controllers
dt-bindings: bus: document the stm32 debug bus
dt-bindings: pinctrl: document access-controllers property for stm32 HDP
dt-bindings: document access-controllers property for coresight peripherals
bus: rifsc: fix RIF configuration check for peripherals
bus: rifsc: Replace snprintf("%s") with strscpy
bus: stm32_firewall: Simplify with scoped for each OF child loop
bus: firewall: move stm32_firewall header file in include folder
Merge tag 'scmi-updates-7.1' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux into soc/drivers
Arm SCMI updates for v7.1
This batch mainly improves SCMI robustness on systems where the SCP does
not generate completion interrupts, and includes two small follow-up
cleanups in the SCMI core.
The main functional change adds support for the new DT property
'arm,no-completion-irq'. When present for mailbox/shared-memory based
SCMI implementations, the driver forces SCMI operations into polling
mode so affected platforms can continue to operate even with broken
firmware interrupt behavior.
In addition, it
- replaces open-coded size rounding in the base protocol path with
round_up() for clarity, with no functional change
- updates the SCMI quirk snippet macro implementation so quirk handlers
can use break and continue directly when invoked inside loop contexts
* tag 'scmi-updates-7.1' of git://git.kernel.org/pub/scm/linux/kernel/git/sudeep.holla/linux:
firmware: arm_scmi: Support loop control in quirk code snippets
firmware: arm_scmi: Use round_up() for base protocol list size calculation
firmware: arm_scmi: Implement arm,no-completion-irq property
dt-bindings: firmware: arm,scmi: Document arm,no-completion-irq property
Merge tag 'memory-controller-drv-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/krzk/linux-mem-ctrl into soc/drivers
Memory controller drivers for v7.1
1. TegraMC:
- Few fixes for older issues - missing clock on Tegra264,
missing enabling of DLL for Tegra30 and Tegra124.
- Simplify the code in a few places.
- Rework handling interrupts on different variants and add support for
error logging on Tegra 264.
2. Drop Baikal SoC bt1-l2-ctl driver, because SoC support is being
removed tree-wide.
* tag 'memory-controller-drv-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/krzk/linux-mem-ctrl:
memory: tegra: Add MC error logging support for Tegra264
memory: tegra: Prepare for supporting multiple intmask registers
memory: tegra: Group SoC specific fields
memory: tegra: Add support for multiple IRQs
memory: tegra: Group register and fields
memory: tegra: Group error handling related registers
memory: tegra-mc: Use %pe format
memory: tegra-mc: Simplify printing PTR_ERR with dev_err_probe
memory: tegra-mc: Drop tegra_mc_setup_latency_allowance() return value
memory: renesas-rpc-if: Simplify printing PTR_ERR with dev_err_probe
memory: brcmstb_memc: Expand LPDDR4 check to cover for LPDDR5
dt-bindings: cache: bt1-l2-ctl: Remove unused bindings
memory: bt1-l2-ctl: Remove not-going-to-be-supported code for Baikal SoC
memory: tegra30-emc: Fix dll_change check
memory: tegra124-emc: Fix dll_change check
memory: tegra: Add support for DBB clock on Tegra264
Merge tag 'samsung-drivers-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/krzk/linux into soc/drivers
Samsung SoC drivers for v7.1
Few cleanups in ACPM firmware drivers, used on Google GS101 and newer
Samsung Exynos SoCs. Notable change is removing 'const' in
'struct acpm_handle' pointers, because even though the code does not
modify pointed data, it immediately drops the const via cast. Also code
is not logically readable when a reference getters/putters (e.g.
acpm_handle_put()) take a pointer to const, because the meaning of "get"
and "put" implies changing the memory, even if that changeable field is
outside of pointed data.
* tag 'samsung-drivers-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/krzk/linux:
firmware: exynos-acpm: Drop fake 'const' on handle pointer
dt-bindings: firmware: google,gs101-acpm-ipc: add S2MPG11 secondary PMIC
firmware: exynos-acpm: Count acpm_xfer buffers with __counted_by_ptr
firmware: exynos-acpm: Count number of commands in acpm_xfer
firmware: exynos-acpm: Use unsigned int for acpm_pmic_linux_errmap index
Julian Ruess [Wed, 25 Mar 2026 13:31:23 +0000 (14:31 +0100)]
vfio/pci: Rename vfio_config_do_rw() to vfio_pci_config_rw_single() and export it
A follow-up patch adds a new variant driver for s390 ISM devices. Since
this device uses a 256 TiB BAR 0 that is never mapped, the variant
driver needs its own ISM_VFIO_PCI_OFFSET_MASK. To minimally mirror the
functionality of vfio_pci_config_rw() with such a custom mask, export
vfio_config_do_rw(). To better distinguish the now exported function
from vfio_pci_config_rw(), rename it to vfio_pci_config_rw_single()
emphasizing that it does a single config space read or write.
Merge tag 'qcom-arm64-fixes-for-7.0-2' of https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux into arm/fixes
More Qualcomm Arm64 DeviceTree fixes for v7.0
The shuffling of reset and wake GPIO properties across various Hamoa
devices left things in an incomplete state, fix this.
Add the missing "ranges" property to the QCM2290 MDSS DeviceTree binding
example, to fix the validation warning that was introduced by the
previous fix.
Merge tag 'renesas-fixes-for-v7.0-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel into arm/fixes
Renesas fixes for v7.0 (take two)
- Fix TFA BL31 memory corruption on Sparrow Hawk.
* tag 'renesas-fixes-for-v7.0-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel:
arm64: dts: renesas: sparrow-hawk: Reserve first 128 MiB of DRAM
Merge tag 'qcom-arm32-for-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux into soc/dt
Qualcomm Arm32 DeviceTree updates for v7.1
Qualcomm APQ8084 is incomplete and hasn't seen functional contributions
since 2016, so drop the platform (for now?). Also drop a number of
unused IPQ-related dtsi files.
Lastly clean up the RPM bus clocks in MSM8974 interconnect nodes.
* tag 'qcom-arm32-for-7.1' of https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux:
ARM: dts: qcom: msm8974: Drop RPM bus clocks
ARM: dts: qcom: drop apq8084 support
ARM: dts: qcom: Drop unused .dtsi
Merge tag 'thead-dt-for-v7.1-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/fustini/linux into soc/dt
T-HEAD Devicetrees for 7.1, part 2
Additional updates to T-Head device trees for v7.1:
- Enable the display pipeline for the TH1520-based BeagleV Ahead board
by adding the HDMI connector node, connecting it to the HDMI
controller, and activating the DPU and HDMI nodes.
- Add coefficients to the TH1520 PVT node as the values in the TH1520
manual differ from the defaults in the driver.
* tag 'thead-dt-for-v7.1-p2' of git://git.kernel.org/pub/scm/linux/kernel/git/fustini/linux:
riscv: dts: thead: beaglev-ahead: enable HDMI output
riscv: dts: thead: th1520: add coefficients to the PVT node
ftrace: Use kallsyms binary search for single-symbol lookup
When ftrace_lookup_symbols() is called with a single symbol (cnt == 1),
use kallsyms_lookup_name() for O(log N) binary search instead of the
full linear scan via kallsyms_on_each_symbol().
ftrace_lookup_symbols() was designed for batch resolution of many
symbols in a single pass. For large cnt this is efficient: a single
O(N) walk over all symbols with O(log cnt) binary search into the
sorted input array. But for cnt == 1 it still decompresses all ~200K
kernel symbols only to match one.
kallsyms_lookup_name() uses the sorted kallsyms index and needs only
~17 decompressions for a single lookup.
This is the common path for kprobe.session with exact function names,
where libbpf sends one symbol per BPF_LINK_CREATE syscall.
If binary lookup fails (duplicate symbol names where the first match
is not ftrace-instrumented), the function falls through to the existing
linear scan path.
Before (cnt=1, 50 kprobe.session programs):
Attach: 858 ms (kallsyms_expand_symbol 25% of CPU)
hkbinbin [Tue, 31 Mar 2026 05:39:16 +0000 (05:39 +0000)]
Bluetooth: hci_sync: fix stack buffer overflow in hci_le_big_create_sync
hci_le_big_create_sync() uses DEFINE_FLEX to allocate a
struct hci_cp_le_big_create_sync on the stack with room for 0x11 (17)
BIS entries. However, conn->num_bis can hold up to HCI_MAX_ISO_BIS (31)
entries — validated against ISO_MAX_NUM_BIS (0x1f) in the caller
hci_conn_big_create_sync(). When conn->num_bis is between 18 and 31,
the memcpy that copies conn->bis into cp->bis writes up to 14 bytes
past the stack buffer, corrupting adjacent stack memory.
This is trivially reproducible: binding an ISO socket with
bc_num_bis = ISO_MAX_NUM_BIS (31) and calling listen() will
eventually trigger hci_le_big_create_sync() from the HCI command
sync worker, causing a KASAN-detectable stack-out-of-bounds write:
BUG: KASAN: stack-out-of-bounds in hci_le_big_create_sync+0x256/0x3b0
Write of size 31 at addr ffffc90000487b48 by task kworker/u9:0/71
Fix this by changing the DEFINE_FLEX count from the incorrect 0x11 to
HCI_MAX_ISO_BIS, which matches the maximum number of BIS entries that
conn->bis can actually carry.
Fixes: 42ecf1947135 ("Bluetooth: ISO: Do not emit LE BIG Create Sync if previous is pending") Cc: stable@vger.kernel.org Signed-off-by: hkbinbin <hkbinbinbin@gmail.com> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Oleh Konko [Tue, 31 Mar 2026 11:52:13 +0000 (11:52 +0000)]
Bluetooth: SMP: derive legacy responder STK authentication from MITM state
The legacy responder path in smp_random() currently labels the stored
STK as authenticated whenever pending_sec_level is BT_SECURITY_HIGH.
That reflects what the local service requested, not what the pairing
flow actually achieved.
For Just Works/Confirm legacy pairing, SMP_FLAG_MITM_AUTH stays clear
and the resulting STK should remain unauthenticated even if the local
side requested HIGH security. Use the established MITM state when
storing the responder STK so the key metadata matches the pairing result.
This also keeps the legacy path aligned with the Secure Connections code,
which already treats JUST_WORKS/JUST_CFM as unauthenticated.
Fixes: fff3490f4781 ("Bluetooth: Fix setting correct authentication information for SMP STK") Cc: stable@vger.kernel.org Signed-off-by: Oleh Konko <security@1seal.org> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Oleh Konko [Tue, 31 Mar 2026 11:52:12 +0000 (11:52 +0000)]
Bluetooth: SMP: force responder MITM requirements before building the pairing response
smp_cmd_pairing_req() currently builds the pairing response from the
initiator auth_req before enforcing the local BT_SECURITY_HIGH
requirement. If the initiator omits SMP_AUTH_MITM, the response can
also omit it even though the local side still requires MITM.
tk_request() then sees an auth value without SMP_AUTH_MITM and may
select JUST_CFM, making method selection inconsistent with the pairing
policy the responder already enforces.
When the local side requires HIGH security, first verify that MITM can
be achieved from the IO capabilities and then force SMP_AUTH_MITM in the
response in both rsp.auth_req and auth. This keeps the responder auth bits
and later method selection aligned.
Fixes: 2b64d153a0cc ("Bluetooth: Add MITM mechanism to LE-SMP") Cc: stable@vger.kernel.org Suggested-by: Luiz Augusto von Dentz <luiz.dentz@gmail.com> Signed-off-by: Oleh Konko <security@1seal.org> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
mesh_send() currently bounds MGMT_OP_MESH_SEND by total command
length, but it never verifies that the bytes supplied for the
flexible adv_data[] array actually match the embedded adv_data_len
field. MGMT_MESH_SEND_SIZE only covers the fixed header, so a
truncated command can still pass the existing 20..50 byte range
check and later drive the async mesh send path past the end of the
queued command buffer.
Keep rejecting zero-length and oversized advertising payloads, but
validate adv_data_len explicitly and require the command length to
exactly match the flexible array size before queueing the request.
Fixes: b338d91703fa ("Bluetooth: Implement support for Mesh") Reported-by: Keenan Dong <keenanat2000@gmail.com> Signed-off-by: Keenan Dong <keenanat2000@gmail.com> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Pauli Virtanen [Sun, 29 Mar 2026 13:43:02 +0000 (16:43 +0300)]
Bluetooth: hci_event: fix potential UAF in hci_le_remote_conn_param_req_evt
hci_conn lookup and field access must be covered by hdev lock in
hci_le_remote_conn_param_req_evt, otherwise it's possible it is freed
concurrently.
Extend the hci_dev_lock critical section to cover all conn usage.
Fixes: 95118dd4edfec ("Bluetooth: hci_event: Use of a function table to handle LE subevents") Signed-off-by: Pauli Virtanen <pav@iki.fi> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Pauli Virtanen [Sun, 29 Mar 2026 13:43:01 +0000 (16:43 +0300)]
Bluetooth: hci_conn: fix potential UAF in set_cig_params_sync
hci_conn lookup and field access must be covered by hdev lock in
set_cig_params_sync, otherwise it's possible it is freed concurrently.
Take hdev lock to prevent hci_conn from being deleted or modified
concurrently. Just RCU lock is not suitable here, as we also want to
avoid "tearing" in the configuration.
Fixes: a091289218202 ("Bluetooth: hci_conn: Fix hci_le_set_cig_params") Signed-off-by: Pauli Virtanen <pav@iki.fi> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Keenan Dong [Sat, 28 Mar 2026 08:46:47 +0000 (16:46 +0800)]
Bluetooth: MGMT: validate LTK enc_size on load
Load Long Term Keys stores the user-provided enc_size and later uses
it to size fixed-size stack operations when replying to LE LTK
requests. An enc_size larger than the 16-byte key buffer can therefore
overflow the reply stack buffer.
Reject oversized enc_size values while validating the management LTK
record so invalid keys never reach the stored key state.
Fixes: 346af67b8d11 ("Bluetooth: Add MGMT handlers for dealing with SMP LTK's") Reported-by: Keenan Dong <keenanat2000@gmail.com> Signed-off-by: Keenan Dong <keenanat2000@gmail.com> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Commit 5df5dafc171b ("Bluetooth: hci_uart: Fix another race during
initialization") fixed a race for hci commands sent during initialization.
However, there is still a race that happens if an hci event from one of
these commands is received before HCI_UART_REGISTERED has been set at
the end of hci_uart_register_dev(). The event will be ignored which
causes the command to fail with a timeout in the log:
"Bluetooth: hci0: command 0x1003 tx timeout"
This is because the hci event receive path (hci_uart_tty_receive ->
h4_recv) requires HCI_UART_REGISTERED to be set in h4_recv(), while the
hci command transmit path (hci_uart_send_frame -> h4_enqueue) only
requires HCI_UART_PROTO_INIT to be set in hci_uart_send_frame().
The check for HCI_UART_REGISTERED was originally added in commit c2578202919a ("Bluetooth: Fix H4 crash from incoming UART packets")
to fix a crash caused by hu->hdev being null dereferenced. That can no
longer happen: once HCI_UART_PROTO_INIT is set in hci_uart_register_dev()
all pointers (hu, hu->priv and hu->hdev) are valid, and
hci_uart_tty_receive() already calls h4_recv() on HCI_UART_PROTO_INIT
or HCI_UART_PROTO_READY.
Remove the check for HCI_UART_REGISTERED in h4_recv() to fix the race
condition.
Fixes: 5df5dafc171b ("Bluetooth: hci_uart: Fix another race during initialization") Signed-off-by: Jonathan Rissanen <jonathan.rissanen@axis.com> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Bluetooth: hci_sync: Fix UAF in le_read_features_complete
This fixes the following backtrace caused by hci_conn being freed
before le_read_features_complete but after
hci_le_read_remote_features_sync so hci_conn_del -> hci_cmd_sync_dequeue
is not able to prevent it:
==================================================================
BUG: KASAN: slab-use-after-free in instrument_atomic_read_write include/linux/instrumented.h:96 [inline]
BUG: KASAN: slab-use-after-free in atomic_dec_and_test include/linux/atomic/atomic-instrumented.h:1383 [inline]
BUG: KASAN: slab-use-after-free in hci_conn_drop include/net/bluetooth/hci_core.h:1688 [inline]
BUG: KASAN: slab-use-after-free in le_read_features_complete+0x5b/0x340 net/bluetooth/hci_sync.c:7344
Write of size 4 at addr ffff8880796b0010 by task kworker/u9:0/52
The buggy address belongs to the object at ffff8880796b0000
which belongs to the cache kmalloc-8k of size 8192
The buggy address is located 16 bytes inside of
freed 8192-byte region [ffff8880796b0000, ffff8880796b2000)
Memory state around the buggy address: ffff8880796aff00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc ffff8880796aff80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff8880796b0000: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
^ ffff8880796b0080: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8880796b0100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
==================================================================
Fixes: a106e50be74b ("Bluetooth: HCI: Add support for LL Extended Feature Set") Reported-by: syzbot+87badbb9094e008e0685@syzkaller.appspotmail.com Tested-by: syzbot+87badbb9094e008e0685@syzkaller.appspotmail.com Closes: https://syzbot.org/bug?extid=87badbb9094e008e0685 Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com> Signed-off-by: Pauli Virtanen <pav@iki.fi>
Pauli Virtanen [Wed, 25 Mar 2026 19:07:43 +0000 (21:07 +0200)]
Bluetooth: hci_sync: hci_cmd_sync_queue_once() return -EEXIST if exists
hci_cmd_sync_queue_once() needs to indicate whether a queue item was
added, so caller can know if callbacks are called, so it can avoid
leaking resources.
Change the function to return -EEXIST if queue item already exists.
Modify all callsites to handle that.
Signed-off-by: Pauli Virtanen <pav@iki.fi> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Oleh Konko [Thu, 26 Mar 2026 17:31:24 +0000 (17:31 +0000)]
Bluetooth: hci_event: move wake reason storage into validated event handlers
hci_store_wake_reason() is called from hci_event_packet() immediately
after stripping the HCI event header but before hci_event_func()
enforces the per-event minimum payload length from hci_ev_table.
This means a short HCI event frame can reach bacpy() before any bounds
check runs.
Rather than duplicating skb parsing and per-event length checks inside
hci_store_wake_reason(), move wake-address storage into the individual
event handlers after their existing event-length validation has
succeeded. Convert hci_store_wake_reason() into a small helper that only
stores an already-validated bdaddr while the caller holds hci_dev_lock().
Use the same helper after hci_event_func() with a NULL address to
preserve the existing unexpected-wake fallback semantics when no
validated event handler records a wake address.
Annotate the helper with __must_hold(&hdev->lock) and add
lockdep_assert_held(&hdev->lock) so future call paths keep the lock
contract explicit.
Call the helper from hci_conn_request_evt(), hci_conn_complete_evt(),
hci_sync_conn_complete_evt(), le_conn_complete_evt(),
hci_le_adv_report_evt(), hci_le_ext_adv_report_evt(),
hci_le_direct_adv_report_evt(), hci_le_pa_sync_established_evt(), and
hci_le_past_received_evt().
Fixes: 2f20216c1d6f ("Bluetooth: Emit controller suspend and resume events") Cc: stable@vger.kernel.org Signed-off-by: Oleh Konko <security@1seal.org> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Cen Zhang [Thu, 26 Mar 2026 15:16:45 +0000 (23:16 +0800)]
Bluetooth: SCO: fix race conditions in sco_sock_connect()
sco_sock_connect() checks sk_state and sk_type without holding
the socket lock. Two concurrent connect() syscalls on the same
socket can both pass the check and enter sco_connect(), leading
to use-after-free.
The buggy scenario involves three participants and was confirmed
with additional logging instrumentation:
Thread A (connect): HCI disconnect: Thread B (connect):
Thread B revives a BT_CLOSED + SOCK_ZAPPED socket back to
BT_CONNECT. Subsequent cleanup triggers double sock_put() and
use-after-free. Meanwhile conn1 is leaked as it was orphaned
when sco_conn_del() cleared the association.
Fix this by:
- Moving lock_sock() before the sk_state/sk_type checks in
sco_sock_connect() to serialize concurrent connect attempts
- Fixing the sk_type != SOCK_SEQPACKET check to actually
return the error instead of just assigning it
- Adding a state re-check in sco_connect() after lock_sock()
to catch state changes during the window between the locks
- Adding sco_pi(sk)->conn check in sco_chan_add() to prevent
double-attach of a socket to multiple connections
- Adding hci_conn_drop() on sco_chan_add failure to prevent
HCI connection leaks
Fixes: 9a8ec9e8ebb5 ("Bluetooth: SCO: Fix possible circular locking dependency on sco_connect_cfm") Signed-off-by: Cen Zhang <zzzccc427@gmail.com> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Pauli Virtanen [Wed, 25 Mar 2026 19:07:46 +0000 (21:07 +0200)]
Bluetooth: hci_sync: call destroy in hci_cmd_sync_run if immediate
hci_cmd_sync_run() may run the work immediately if called from existing
sync work (otherwise it queues a new sync work). In this case it fails
to call the destroy() function.
On immediate run, make it behave same way as if item was queued
successfully: call destroy, and return 0.
The only callsite is hci_abort_conn() via hci_cmd_sync_run_once(), and
this changes its return value. However, its return value is not used
except as the return value for hci_disconnect(), and nothing uses the
return value of hci_disconnect(). Hence there should be no behavior
change anywhere.
Fixes: c898f6d7b093b ("Bluetooth: hci_sync: Introduce hci_cmd_sync_run/hci_cmd_sync_run_once") Signed-off-by: Pauli Virtanen <pav@iki.fi> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Add a kernel module that benchmarks queue_work() throughput on an
unbound workqueue to measure pool->lock contention under different
affinity scope configurations (cache vs cache_shard).
The module spawns N kthreads (default: num_online_cpus()), each bound
to a different CPU. All threads start simultaneously and queue work
items, measuring the latency of each queue_work() call. Results are
reported as p50/p90/p95 latencies for each affinity scope.
The affinity scope is switched between runs via the workqueue's sysfs
affinity_scope attribute (WQ_SYSFS), avoiding the need for any new
exported symbols.
The module runs as __init-only, returning -EAGAIN to auto-unload,
and can be re-run via insmod.
tools/workqueue: add CACHE_SHARD support to wq_dump.py
The WQ_AFFN_CACHE_SHARD affinity scope was added to the kernel but
wq_dump.py was not updated to enumerate it. Add the missing constant
lookup and include it in the affinity scopes iteration so that drgn
output shows the CACHE_SHARD pod topology alongside the other scopes.
workqueue: set WQ_AFFN_CACHE_SHARD as the default affinity scope
Set WQ_AFFN_CACHE_SHARD as the default affinity scope for unbound
workqueues. On systems where many CPUs share one LLC, the previous
default (WQ_AFFN_CACHE) collapses all CPUs to a single worker pool,
causing heavy spinlock contention on pool->lock.
WQ_AFFN_CACHE_SHARD subdivides each LLC into smaller groups, providing
a better balance between locality and contention. Users can revert to
the previous behavior with workqueue.default_affinity_scope=cache.
On systems with 8 or fewer cores per LLC, CACHE_SHARD produces a single
shard covering the entire LLC, making it functionally identical to the
previous CACHE default. The sharding only activates when an LLC has more
than 8 cores.
On systems where many CPUs share one LLC, unbound workqueues using
WQ_AFFN_CACHE collapse to a single worker pool, causing heavy spinlock
contention on pool->lock. For example, Chuck Lever measured 39% of
cycles lost to native_queued_spin_lock_slowpath on a 12-core shared-L3
NFS-over-RDMA system.
The existing affinity hierarchy (cpu, smt, cache, numa, system) offers
no intermediate option between per-LLC and per-SMT-core granularity.
Add WQ_AFFN_CACHE_SHARD, which subdivides each LLC into groups of at
most wq_cache_shard_size cores (default 8, tunable via boot parameter).
Shards are always split on core (SMT group) boundaries so that
Hyper-Threading siblings are never placed in different pods. Cores are
distributed across shards as evenly as possible -- for example, 36 cores
in a single LLC with max shard size 8 produces 5 shards of 8+7+7+7+7
cores.
The implementation follows the same comparator pattern as other affinity
scopes: precompute_cache_shard_ids() pre-fills the cpu_shard_id[] array
from the already-initialized WQ_AFFN_CACHE and WQ_AFFN_SMT topology,
and cpus_share_cache_shard() is passed to init_pod_type().
Benchmark on NVIDIA Grace (72 CPUs, single LLC, 50k items/thread), show
cache_shard delivers ~5x the throughput and ~6.5x lower p50 latency
compared to cache scope on this 72-core single-LLC system.
Matthew Brost [Wed, 1 Apr 2026 01:07:39 +0000 (18:07 -0700)]
workqueue: Add pool_workqueue to pending_pwqs list when unplugging multiple inactive works
In unplug_oldest_pwq(), the first inactive work item on the
pool_workqueue is activated correctly. However, if multiple inactive
works exist on the same pool_workqueue, subsequent works fail to
activate because wq_node_nr_active.pending_pwqs is empty — the list
insertion is skipped when the pool_workqueue is plugged.
Fix this by checking for additional inactive works in
unplug_oldest_pwq() and updating wq_node_nr_active.pending_pwqs
accordingly.
Fixes: 4c065dbce1e8 ("workqueue: Enable unbound cpumask update on ordered workqueues") Cc: stable@vger.kernel.org Cc: Carlos Santa <carlos.santa@intel.com> Cc: Ryan Neph <ryanneph@google.com> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Waiman Long <longman@redhat.com> Cc: linux-kernel@vger.kernel.org Signed-off-by: Matthew Brost <matthew.brost@intel.com> Signed-off-by: Tejun Heo <tj@kernel.org> Acked-by: Waiman Long <longman@redhat.com>
Eric Biggers [Wed, 1 Apr 2026 00:33:31 +0000 (17:33 -0700)]
lib/crypto: arm64: Assume a little-endian kernel
Since support for big-endian arm64 kernels was removed, the CPU_LE()
macro now unconditionally emits the code it is passed, and the CPU_BE()
macro now unconditionally discards the code it is passed.
Simplify the assembly code in lib/crypto/arm64/ accordingly.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
Since commit aefbab8e77eb ("arm64: fpsimd: Preserve/restore kernel mode
NEON at context switch"), kernel-mode NEON sections have been
preemptible on arm64. And since commit 7dadeaa6e851 ("sched: Further
restrict the preemption modes"), voluntary preemption is no longer
supported on arm64 either. Therefore, there's no longer any need to
limit the length of kernel-mode NEON sections on arm64.
MIPS: mm: Rewrite TLB uniquification for the hidden bit feature
Before the introduction of the EHINV feature, which lets software mark
TLB entries invalid, certain older implementations of the MIPS ISA were
equipped with an analogous bit, as a vendor extension, which however is
hidden from software and only ever set at reset, and then any software
write clears it, making the intended TLB entry valid.
This feature makes it unsafe to read a TLB entry with TLBR, modify the
page mask, and write the entry back with TLBWI, because this operation
will implicitly clear the hidden bit and this may create a duplicate
entry, as with the presence of the hidden bit there is no guarantee all
the entries across the TLB are unique each.
Usually the firmware has already uniquified TLB entries before handing
control over, in which case we only need to guarantee at bootstrap no
clash will happen with the VPN2 values chosen in local_flush_tlb_all().
However with systems such as Mikrotik RB532 we get handed the TLB as at
reset, with the hidden bit set across the entries and possibly duplicate
entries present. This then causes a machine check exception when page
sizes are reset in r4k_tlb_uniquify() and prevents the system from
booting.
Rewrite the algorithm used in r4k_tlb_uniquify() then such as to avoid
the reuse of ASID/VPN values across the TLB. Get rid of global entries
first as they may be blocking the entire address space, e.g. 16 256MiB
pages will exhaust the whole address space of a 32-bit CPU and a single
big page can exhaust the 32-bit compatibility space on a 64-bit CPU.
Details of the algorithm chosen are given across the code itself.
Fixes: 9f048fa48740 ("MIPS: mm: Prevent a TLB shutdown on initial uniquification") Signed-off-by: Maciej W. Rozycki <macro@orcam.me.uk> Cc: stable@vger.kernel.org # v6.18+ Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
projects / thirdparty / linux.git / log
