Lorenzo Bianconi [Thu, 22 Jan 2026 10:40:00 +0000 (11:40 +0100)]
wifi: mt76: mt7996: Store DMA mapped buffer addresses in mt7996_npu_hw_init()
In order to not always reallocate them during NPU reset, store the DMA
mapped buffer addresses allocated by mt7996_npu_hw_init routine in
mt7996 structure.
Lorenzo Bianconi [Thu, 22 Jan 2026 10:39:58 +0000 (11:39 +0100)]
wifi: mt76: Do not reset idx for NPU tx queues during reset
Do not run reset_q callaback with reset_idx set to true for NPU Tx
queues. This is a preliminary patch to properly manage reset procedure
when NPU offloading is enabled.
Introduce __mt7996_npu_hw_init utility routine in order to run it
holding mt76 mutex and move NPU hw re-initialization before restarting
the NAPIs during device reset.
Lorenzo Bianconi [Thu, 22 Jan 2026 10:39:48 +0000 (11:39 +0100)]
wifi: mt76: mt7996: Fix BAND2 tx queues initialization when NPU is enabled
Fix BAND2 tx queues initialization for MT7990 chipset when NPU is
enabled. This is a preliminary patch to enable NPU offload for MT7996
(Eagle) chipset.
Lorenzo Bianconi [Thu, 22 Jan 2026 10:39:47 +0000 (11:39 +0100)]
wifi: mt76: always enable RRO queues for non-MT7992 chipset
MT7990 NPU binary requires to initialize NPU desc_base after configuring
ring_size. This is a preliminary patch to enable NPU offload for MT7996
(Eagle) chipset.
Lorenzo Bianconi [Thu, 22 Jan 2026 10:39:45 +0000 (11:39 +0100)]
wifi: mt76: mt7996: Fix NPU stop procedure
Move mt7996_npu_hw_stop routine before disabling rx NAPIs in order to
fix NPU stop procedure used during device L1 SER recovery.
Add missing usleep_range in mt7996_npu_hw_stop().
All MT76 rx queues have an associated page_pool even if the queue is not
associated to a NAPI (e.g. WED RRO queues with WED enabled). Destroy the
page_pool running mt76_dma_cleanup routine during module unload.
Moreover returns pages to the page pool if WED is not enabled for WED RRO
queues.
Leon Yen [Thu, 9 Oct 2025 02:01:58 +0000 (10:01 +0800)]
wifi: mt76: mt7921: fix a potential clc buffer length underflow
The buf_len is used to limit the iterations for retrieving the country
power setting and may underflow under certain conditions due to changes
in the power table in CLC.
This underflow leads to an almost infinite loop or an invalid power
setting resulting in driver initialization failure.
Cc: stable@vger.kernel.org Fixes: fa6ad88e023d ("wifi: mt76: mt7921: fix country count limitation for CLC") Signed-off-by: Leon Yen <leon.yen@mediatek.com> Signed-off-by: Ming Yen Hsieh <mingyen.hsieh@mediatek.com> Link: https://patch.msgid.link/20251009020158.1923429-1-mingyen.hsieh@mediatek.com Signed-off-by: Felix Fietkau <nbd@nbd.name>
Rory Little [Thu, 4 Sep 2025 00:07:11 +0000 (17:07 -0700)]
wifi: mt76: mt7921: Place upper limit on station AID
Any station configured with an AID over 20 causes a firmware crash.
This situation occurred in our testing using an AP interface on 7922
hardware, with a modified hostapd, sourced from Mediatek's OpenWRT
feeds.
In stock hostapd, station AIDs begin counting at 1, and this
configuration is prevented with an upper limit on associated stations.
However, the modified hostapd began allocation at 65, which caused the
firmware to crash. This fix does not allow these AIDs to work, but will
prevent the firmware crash.
This crash was only seen on IFTYPE_AP interfaces, and the fix does not
appear to have an effect on IFTYPE_STATION behavior.
Alok Tiwari [Mon, 13 Oct 2025 09:08:24 +0000 (02:08 -0700)]
wifi: mt76: mt7996: fix FCS error flag check in RX descriptor
The mt7996 driver currently checks the MT_RXD3_NORMAL_FCS_ERR bit in
rxd1 whereas other Connac3-based drivers(mt7925) correctly check this
bit in rxd3.
Since the MT_RXD3_NORMAL_FCS_ERR bit is defined in the fourth RX
descriptor word (rxd3), update mt7996 to use the proper descriptor
field. This change aligns mt7996 with mt7925 and the rest of the
Connac3 family.
Fixes: 98686cd21624 ("wifi: mt76: mt7996: add driver for MediaTek Wi-Fi 7 (802.11be) devices") Signed-off-by: Alok Tiwari <alok.a.tiwari@oracle.com> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Link: https://patch.msgid.link/20251013090826.753992-1-alok.a.tiwari@oracle.com Signed-off-by: Felix Fietkau <nbd@nbd.name>
tracing: Adjust cmd_check_undefined to show unexpected undefined symbols
When the check_undefined command in kernel/trace/Makefile fails, there
is no output, making it hard to understand why the build failed. Capture
the output of the $(NM) + grep command and print it when failing to make
it clearer what the problem is.
Fixes: a717943d8ecc ("tracing: Check for undefined symbols in simple_ring_buffer") Signed-off-by: Nathan Chancellor <nathan@kernel.org> Reviewed-by: Vincent Donnefort <vdonnefort@google.com> Acked-by: Arnd Bergmann <arnd@arndb.de> Link: https://patch.msgid.link/20260320-cmd_check_undefined-verbose-v1-1-54fc5b061f94@kernel.org Signed-off-by: Marc Zyngier <maz@kernel.org>
Ming Yen Hsieh [Thu, 4 Sep 2025 03:06:47 +0000 (11:06 +0800)]
wifi: mt76: mt7925: prevent NULL vif dereference in mt7925_mac_write_txwi
Check for a NULL `vif` before accessing `ieee80211_vif_is_mld(vif)` to
avoid a potential kernel panic in scenarios where `vif` might not be
initialized.
Zac Bowling [Tue, 20 Jan 2026 20:10:32 +0000 (12:10 -0800)]
wifi: mt76: fix list corruption in mt76_wcid_cleanup
mt76_wcid_cleanup() was not removing wcid entries from sta_poll_list
before mt76_reset_device() reinitializes the master list. This leaves
stale pointers in wcid->poll_list, causing list corruption when
mt76_wcid_add_poll() later checks list_empty() and tries to add the
entry back.
The fix adds proper cleanup of poll_list in mt76_wcid_cleanup(),
matching how tx_list is already handled. This is similar to what
mt7996_mac_sta_deinit_link() already does correctly.
Fixes list corruption warnings like:
list_add corruption. prev->next should be next (ffffffff...)
Ryder Lee [Wed, 21 Jan 2026 17:41:58 +0000 (09:41 -0800)]
wifi: mt76: mt7996: add support for ERP CTS & HT protection
This patch adds support for handling BSS_CHANGED_ERP_CTS_PROT and
BSS_CHANGED_HT.
With this change, when the Wi-Fi driver needs to adjust its behavior for
compatibility or performance, especially concerning older 11g/n devices,
by enabling or disabling CTS protection frames, often for hidden SSIDs or
to manage legacy clients.
It also introduces debugfs options to manually control protection mode,
allowing users to select betweenno protection, RTS/CTS, and CTS-to-self.
Ryder Lee [Wed, 21 Jan 2026 17:41:57 +0000 (09:41 -0800)]
wifi: mt76: mt7915: fix use_cts_prot support
With this fix, when driver needs to adjust its behavior for compatibility,
especially concerning older 11g/n devices, by enabling or disabling CTS
protection frames, often for hidden SSIDs or to manage legacy clients.
Ryder Lee [Wed, 21 Jan 2026 17:41:56 +0000 (09:41 -0800)]
wifi: mt76: mt7615: fix use_cts_prot support
Driver should not directly write WTBL to prevent overwritten issues.
With this fix, when driver needs to adjust its behavior for compatibility,
especially concerning older 11g/n devices, by enabling or disabling CTS
protection frames, often for hidden SSIDs or to manage legacy clients.
Christian Hewitt [Sat, 27 Dec 2025 11:22:19 +0000 (11:22 +0000)]
wifi: mt7601u: check multiple firmware paths
The linux-firmware repo moved mt7601u.bin from its root folder to
the mediatek sub-folder some time ago, but the driver still tries
to load firmware from the old location. Users might have firmware
in either location so update the driver to check both.
Leon Yen [Mon, 15 Dec 2025 12:22:31 +0000 (20:22 +0800)]
wifi: mt76: mt792x: Fix a potential deadlock in high-load situations
A deadlock may occur between two works, ps_work and mac_work, if their work
functions run simultaneously as they attempt to cancel each other by
calling cancel_delayed_work_sync().
In high-load situations, they are queued but may not have chance to be
executed until the CPUs are released. Once the CPUs are available, there
is a high possibility that the ps_work function and mac_work function will
be executed simultaneously, resulting in a possible deadlock.
This patch replaces cancel_delayed_work_sync() with cancel_delayed_work()
in ps_work to eliminate the deadlock and make the code easier to maintain.
Leon Yen [Thu, 11 Dec 2025 12:38:36 +0000 (20:38 +0800)]
wifi: mt76: mt7925: Fix incorrect MLO mode in firmware control
The selection of MLO mode should depend on the capabilities of the STA
rather than those of the peer AP to avoid compatibility issues with
certain APs, such as Xiaomi BE5000 WiFi7 router.
Sean Wang [Tue, 16 Dec 2025 00:59:30 +0000 (18:59 -0600)]
wifi: mt76: mt7921: Reset ampdu_state state in case of failure in mt76_connac2_tx_check_aggr()
Reset ampdu_state if ieee80211_start_tx_ba_session() fails in
mt76_connac2_tx_check_aggr(), otherwise the driver may incorrectly
assume aggregation is active and skip future BA setup attempts.
wifi: mt76: mt7996: Switch to the secondary link if the default one is removed
Switch to the secondary link if available in mt7996_mac_sta_remove_links
routine if the primary one is removed.
Moreover reset secondary link index for single link scenario.
Quan Zhou [Thu, 27 Nov 2025 07:49:11 +0000 (15:49 +0800)]
wifi: mt76: mt7925: fix AMPDU state handling in mt7925_tx_check_aggr
Previously, the AMPDU state bit for a given TID was set before attempting
to start a BA session, which could result in the AMPDU state being marked
active even if ieee80211_start_tx_ba_session() failed. This patch changes
the logic to only set the AMPDU state bit after successfully starting a BA
session, ensuring proper synchronization between AMPDU state and BA session
status.
This fixes potential issues with aggregation state tracking and improves
compatibility with mac80211 BA session management.
Since some radar specifications maintained by the driver are incorrect
and are now also maintained by the firmware, offload the initialization
procedure to the firmware.
This fixes issues for radar detection rate testings.
StanleyYP Wang [Mon, 15 Dec 2025 06:37:24 +0000 (14:37 +0800)]
wifi: mt76: mt7996: set specific BSSINFO and STAREC commands after channel switch
After channel switch, some tags of BSSINFO (rfch) and STAREC (bfer,
rate_ctrl) commands should also be updated. Otherwise, a BSS might not be
able to transmit with its peer using correct bandwidth.
StanleyYP Wang [Mon, 15 Dec 2025 06:37:23 +0000 (14:37 +0800)]
wifi: mt76: mt7996: fix the behavior of radar detection
RDD_DET_MODE is a firmware command intended for testing and does not
pause TX after radar detection, so remove it from the normal flow;
instead, use the MAC_ENABLE_CTRL firmware command to resume TX after
the radar-triggered channel switch completes.
Thorsten Blum [Sat, 21 Mar 2026 21:05:47 +0000 (22:05 +0100)]
fanotify: replace deprecated strcpy in fanotify_info_copy_{name,name2}
strcpy() has been deprecated [1] because it performs no bounds checking
on the destination buffer, which can lead to buffer overflows. Replace
it with the safer strscpy().
Tzuyi Chang [Tue, 17 Mar 2026 11:54:09 +0000 (19:54 +0800)]
pinctrl: realtek: add rtd1625 pinctrl driver
Add support for Realtek RTD1625 SoC using the realtek common pinctrl driver.
This patch introduces the RTK_PIN_CONFIG_V2 and RTK_PIN_CONFIG_I2C macros,
which are required to describe the specific register layout and electrical
features (such as slew rate and high VIL) of the RTD1625 pins.
Tzuyi Chang [Tue, 17 Mar 2026 11:54:08 +0000 (19:54 +0800)]
pinctrl: realtek: add support for slew rate, input voltage and high VIL
Add support for configuring slew rate, input voltage level and high VIL
mode. This involves updating the pin configuration parsing logic to handle
PIN_CONFIG_SLEW_RATE, PIN_CONFIG_INPUT_VOLTAGE_UV and the new custom
property "realtek,high-vil-microvolt".
Reviewed-by: Linus Walleij <linusw@kernel.org> Signed-off-by: Tzuyi Chang <tychang@realtek.com> Co-developed-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Linus Walleij <linusw@kernel.org>
The previous description was misleading because this hardware block is not
a PWM generator. It does not generate a signal with a specific frequency
and duty ratio.
Instead, it provides a fixed nanosecond-level adjustment to the rising/
falling edges of an existing signal.
The property name is kept as 'realtek,duty-cycle' rather than being
renamed to strictly preserve Device Tree ABI backward compatibility.
Add a new generic pin configuration parameter PIN_CONFIG_INPUT_VOLTAGE_UV.
This parameter is used to specify the input voltage level of a pin in
microvolts, which corresponds to the 'input-voltage-microvolt' property
in Device Tree.
Reviewed-by: Linus Walleij <linusw@kernel.org> Signed-off-by: Tzuyi Chang <tychang@realtek.com> Co-developed-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Linus Walleij <linusw@kernel.org>
Add a generic pin configuration property
"input-threshold-voltage-microvolt" to support hardware designs where the
input logic threshold is decoupled from the power supply voltage.
This property allows the pinctrl driver to configure the correct internal
reference voltage for pins that need to accept input signals at a different
voltage level than their power supply. For example, a pin powered by 3.3V
may need to accept 1.8V logic signals.
This defines the reference for VIH (Input High Voltage) and VIL (Input Low
Voltage) thresholds, enabling proper signal detection across different
voltage domains.
Signed-off-by: Tzuyi Chang <tychang@realtek.com> Co-developed-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Yu-Chun Lin <eleanor.lin@realtek.com> Acked-by: Conor Dooley <conor.dooley@microchip.com> Signed-off-by: Linus Walleij <linusw@kernel.org>
Yu-Chun Lin [Tue, 17 Mar 2026 11:54:03 +0000 (19:54 +0800)]
pinctrl: realtek: Fix function signature for config argument
The argument originates from pinconf_to_config_argument(), which returns a
u32. Therefore, the arg parameter should be an unsigned int instead of enum
pin_config_param.
Fixes: e99ce78030db ("pinctrl: realtek: Add common pinctrl driver for Realtek DHC RTD SoCs") Signed-off-by: Yu-Chun Lin <eleanor.lin@realtek.com> Signed-off-by: Linus Walleij <linusw@kernel.org>
gpio: shared: handle pins shared by child nodes of devices
Shared GPIOs may be assigned to child nodes of device nodes which don't
themselves bind to any struct device. We need to pass the firmware node
that is the actual consumer to gpiolib-shared and compare against it
instead of unconditionally using the fwnode of the consumer device.
Fixes: a060b8c511ab ("gpiolib: implement low-level, shared GPIO support") Reported-by: Jon Hunter <jonathanh@nvidia.com> Closes: https://lore.kernel.org/all/921ba8ce-b18e-4a99-966d-c763d22081e2@nvidia.com/ Tested-by: Jon Hunter <jonathanh@nvidia.com> Acked-by: Jon Hunter <jonathanh@nvidia.com> Link: https://patch.msgid.link/20260318-gpio-shared-xlate-v2-2-0ce34c707e81@oss.qualcomm.com Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
OF-based GPIO controller drivers may provide a translation function that
calculates the real chip offset from whatever devicetree sources
provide. We need to take this into account in the shared GPIO management
and call of_xlate() if it's provided and adjust the entry->offset we
initially set when scanning the tree.
To that end: modify the shared GPIO API to take the GPIO chip as
argument on setup (to avoid having to rcu_dereference() it from the GPIO
device) and protect the access to entry->offset with the existing lock.
Michał Winiarski [Tue, 17 Feb 2026 15:41:18 +0000 (16:41 +0100)]
drm/xe/pf: Fix use-after-free in migration restore
When an error is returned from xe_sriov_pf_migration_restore_produce(),
the data pointer is not set to NULL, which can trigger use-after-free
in subsequent .write() calls.
Set the pointer to NULL upon error to fix the problem.
Fixes: 1ed30397c0b92 ("drm/xe/pf: Add support for encap/decap of bitstream to/from packet") Reported-by: Sebastian Österlund <sebastian.osterlund@intel.com> Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/7230 Reviewed-by: Shuicheng Lin <shuicheng.lin@intel.com> Link: https://patch.msgid.link/20260217154118.176902-1-michal.winiarski@intel.com Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Imre Deak [Fri, 20 Mar 2026 09:29:00 +0000 (11:29 +0200)]
drm/i915/dp_tunnel: Fix error handling when clearing stream BW in atomic state
Clearing the DP tunnel stream BW in the atomic state involves getting
the tunnel group state, which can fail. Handle the error accordingly.
This fixes at least one issue where drm_dp_tunnel_atomic_set_stream_bw()
failed to get the tunnel group state returning -EDEADLK, which wasn't
handled. This lead to the ctx->contended warn later in modeset_lock()
while taking a WW mutex for another object in the same atomic state, and
thus within the same already contended WW context.
Moving intel_crtc_state_alloc() later would avoid freeing saved_state on
the error path; this stable patch leaves that simplification for a
follow-up.
Cc: Uma Shankar <uma.shankar@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: <stable@vger.kernel.org> # v6.9+ Fixes: a4efae87ecb2 ("drm/i915/dp: Compute DP tunnel BW during encoder state computation") Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/7617 Reviewed-by: Michał Grzelak <michal.grzelak@intel.com> Reviewed-by: Uma Shankar <uma.shankar@intel.com> Signed-off-by: Imre Deak <imre.deak@intel.com> Link: https://patch.msgid.link/20260320092900.13210-1-imre.deak@intel.com
Andreas Hindborg [Thu, 19 Feb 2026 11:57:45 +0000 (12:57 +0100)]
hrtimer: add usage examples to documentation
Add documentation examples showing various ways to use hrtimers:
- Box-allocated timers with shared state in Arc.
- Arc-allocated timers.
- Stack-based timers for scoped usage.
- Mutable stack-based timers with shared state.
FUJITA Tomonori [Mon, 30 Jun 2025 13:10:11 +0000 (22:10 +0900)]
rust: time: make ClockSource unsafe trait
Mark the ClockSource trait as unsafe and document its safety
requirements. Specifically, implementers must guarantee that their
`ktime_get()` implementation returns a value in the inclusive range
[0, KTIME_MAX].
Update all existing implementations to use `unsafe impl` with
corresponding safety comments.
Note that there could be potential users of a customized clock source [1]
so we don't seal the trait.
Ville Syrjälä [Mon, 16 Mar 2026 16:39:51 +0000 (18:39 +0200)]
drm/i915: Unlink NV12 planes earlier
unlink_nv12_plane() will clobber parts of the plane state
potentially already set up by plane_atomic_check(), so we
must make sure not to call the two in the wrong order.
The problem happens when a plane previously selected as
a Y plane is now configured as a normal plane by user space.
plane_atomic_check() will first compute the proper plane
state based on the userspace request, and unlink_nv12_plane()
later clears some of the state.
This used to work on account of unlink_nv12_plane() skipping
the state clearing based on the plane visibility. But I removed
that check, thinking it was an impossible situation. Now when
that situation happens unlink_nv12_plane() will just WARN
and proceed to clobber the state.
Rather than reverting to the old way of doing things, I think
it's more clear if we unlink the NV12 planes before we even
compute the new plane state.
Cc: stable@vger.kernel.org Reported-by: Khaled Almahallawy <khaled.almahallawy@intel.com> Closes: https://lore.kernel.org/intel-gfx/20260212004852.1920270-1-khaled.almahallawy@intel.com/ Tested-by: Khaled Almahallawy <khaled.almahallawy@intel.com> Fixes: 6a01df2f1b2a ("drm/i915: Remove pointless visible check in unlink_nv12_plane()") Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patch.msgid.link/20260316163953.12905-2-ville.syrjala@linux.intel.com Reviewed-by: Uma Shankar <uma.shankar@intel.com>
(cherry picked from commit 017ecd04985573eeeb0745fa2c23896fb22ee0cc) Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Ville Syrjälä [Fri, 13 Mar 2026 11:07:40 +0000 (13:07 +0200)]
drm/i915: Order OP vs. timeout correctly in __wait_for()
Put the barrier() before the OP so that anything we read out in
OP and check in COND will actually be read out after the timeout
has been evaluated.
Currently the only place where we use OP is __intel_wait_for_register(),
but the use there is precisely susceptible to this reordering, assuming
the ktime_*() stuff itself doesn't act as a sufficient barrier:
drm/i915/gmbus: fix spurious timeout on 512-byte burst reads
When reading exactly 512 bytes with burst read enabled, the
extra_byte_added path breaks out of the inner do-while without
decrementing len. The outer while(len) then re-enters and gmbus_wait()
times out since all data has been delivered. Decrement len before the
break so the outer loop terminates correctly.
Chen-Yu Tsai [Wed, 11 Mar 2026 09:49:28 +0000 (17:49 +0800)]
drm/sun4i: Use backend/mixer as dedicated DMA device
The sun4i DRM driver deals with DMA constraints in a peculiar way.
Instead of using the actual DMA device in various helpers, it justs
reconfigures the DMA constraints of the virtual display device using
the DMA device's device tree node by calling of_dma_configure().
Turns out of_dma_configure() should only be called from bus code.
Lately this also triggers a big warning through of_iommu_configure()
and ultimately __iommu_probe_device():
late IOMMU probe at driver bind, something fishy here!
Now that the GEM DMA helpers have proper support for allocating
and mapping buffers with a dedicated DMA device, switch over to
it as the proper solution.
The mixer change was tested on a Pine H64 model B. The backend change
was only compile tested. Though I don't expect any issues, help testing
on an older device would be appreciated.
Chen-Yu Tsai [Wed, 11 Mar 2026 09:49:27 +0000 (17:49 +0800)]
drm/mediatek: Set dedicated DMA device and drop custom GEM callbacks
In commit 9b54a32c7c6a ("drm/mediatek: mtk_gem: Partial refactor and
use drm_gem_dma_object") the MediaTek DRM driver was refactored to use
drm_gem_dma_object, but custom callbacks were still needed to deal with
using the first device of the pipeline as the DMA device, instead of
the MMSYS device that the DRM driver binds to.
Turns out there is already partial support for dedicated DMA devices in
the DRM subsystem for PRIME imports. The preceding patches add support
for dedicated DMA devices to the GEM DMA helpers.
This allows us to just set the dedicated DMA device for the DRM device,
and drop all the custom GEM callbacks. Also drop the .dma_dev field
from the driver private data as it is no longer needed.
There are slight differences in the mmap helper: the VM_DONTDUMP and
VM_IO flags are no longer set. Both were lifted from drm_gem_mmap_obj().
VM_IO probably doesn't make sense since the buffer is allocated using
dma_alloc_attrs().
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> Acked-by: Chun-Kuang Hu <chunkuang.hu@kernel.org> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Link: https://patch.msgid.link/20260311094929.3393338-4-wenst@chromium.org Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Chen-Yu Tsai [Wed, 11 Mar 2026 09:49:26 +0000 (17:49 +0800)]
drm/gem-dma: Support dedicated DMA device for allocation and mapping
Support for a dedicated DMA device for prime imports was added in commit 143ec8d3f939 ("drm/prime: Support dedicated DMA device for dma-buf imports").
This allowed the DRM driver to provide a dedicated DMA device when its
own underlying device was not capable of DMA, for example when it is a
USB device (the original target) or a virtual device. The latter case is
common on embedded SoCs, on which the display pipeline is composed of
various fixed function blocks, and the DRM device is simply a made-up
device, an address space managing the routing between the blocks, or
whichever block the implementor thought made sense at the time. The
point is that the chosen device is often not the actual device doing
the DMA. Various drivers have used workarounds or reimplemented the
GEM DMA helpers to get the DMA addresses and IOMMUs to work correctly.
Add support for the dedicated DMA device to the GEM DMA helpers.
No existing driver currently uses the GEM DMA helpers and calls
drm_dev_set_dma_dev() to set a dedicated DMA device, so no existing
users should be affected.
Chen-Yu Tsai [Wed, 11 Mar 2026 09:49:25 +0000 (17:49 +0800)]
drm/prime: Limit scatter list size with dedicated DMA device
If a dedicated DMA device is specified for the DRM device, then the
scatter list size limit should pertain to the DMA device.
Use the dedicated DMA device, if given, to limit the scatter list size.
This only applies to drivers that have called drm_dev_set_dma_dev() and
are using drm_prime_pages_to_sg() either directly or through the SHMEM
helpers. At the time of this writing, the former case only includes the
Rockchip DRM driver, while the latter case includes the gud, udl, and
the tiny appletbdrm and gm12u320 drivers.
Ming Lei [Wed, 18 Mar 2026 01:41:12 +0000 (09:41 +0800)]
ublk: move cold paths out of __ublk_batch_dispatch() for icache efficiency
Mark ublk_filter_unused_tags() as noinline since it is only called from
the unlikely(needs_filter) branch. Extract the error-handling block from
__ublk_batch_dispatch() into a new noinline ublk_batch_dispatch_fail()
function to keep the hot path compact and icache-friendly. This also
makes __ublk_batch_dispatch() more readable by separating the error
recovery logic from the normal dispatch flow.
Before: __ublk_batch_dispatch is ~1419 bytes
After: __ublk_batch_dispatch is ~1090 bytes (-329 bytes, -23%)
zhidao su [Sun, 22 Mar 2026 07:35:33 +0000 (15:35 +0800)]
selftests/sched_ext: Add tests for SCX_ENQ_IMMED and scx_bpf_dsq_reenq()
Add three selftests covering features introduced in v7.1:
- dsq_reenq: Verify scx_bpf_dsq_reenq() on user DSQs triggers
ops.enqueue() with SCX_ENQ_REENQ and SCX_TASK_REENQ_KFUNC in
p->scx.flags.
- enq_immed: Verify SCX_OPS_ALWAYS_ENQ_IMMED slow path where tasks
dispatched to a busy CPU's local DSQ are re-enqueued through
ops.enqueue() with SCX_TASK_REENQ_IMMED.
- consume_immed: Verify SCX_ENQ_IMMED via the consume path using
scx_bpf_dsq_move_to_local___v2() with explicit SCX_ENQ_IMMED.
All three tests skip gracefully on kernels that predate the required
features by checking availability via __COMPAT_has_ksym() /
__COMPAT_read_enum() before loading.
Signed-off-by: zhidao su <suzhidao@xiaomi.com> Signed-off-by: Tejun Heo <tj@kernel.org>
Namjae Jeon [Fri, 13 Mar 2026 05:45:58 +0000 (14:45 +0900)]
ksmbd: replace hardcoded hdr2_len with offsetof() in smb2_calc_max_out_buf_len()
After this commit (e2b76ab8b5c9 "ksmbd: add support for read compound"),
response buffer management was changed to use dynamic iov array.
In the new design, smb2_calc_max_out_buf_len() expects the second
argument (hdr2_len) to be the offset of ->Buffer field in the
response structure, not a hardcoded magic number.
Fix the remaining call sites to use the correct offsetof() value.
Cc: stable@vger.kernel.org Fixes: e2b76ab8b5c9 ("ksmbd: add support for read compound") Signed-off-by: Namjae Jeon <linkinjeon@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
Tejun Heo [Sun, 22 Mar 2026 20:33:08 +0000 (10:33 -1000)]
sched_ext: Use irq_work_queue_on() in schedule_deferred()
schedule_deferred() uses irq_work_queue() which always queues on the
calling CPU. The deferred work can run from any CPU correctly, and the
_locked() path already processes remote rqs from the calling CPU. However,
when falling through to the irq_work path, queuing on the target CPU is
preferable as the work can run sooner via IPI delivery rather than waiting
for the calling CPU to re-enable IRQs.
Currently, only reenqueue operations use this path - either BPF-initiated
reenqueue targeting a remote rq, or IMMED reenqueue when the target CPU is
busy running userspace (not in balance or wakeup, so the _locked() fast
paths aren't available). Use irq_work_queue_on() to target the owning CPU.
This improves IMMED reenqueue latency when tasks are dispatched to
remote local DSQs. Testing on a 24-CPU AMD Ryzen 3900X with scx_qmap
-I -F 50 (ALWAYS_ENQ_IMMED, every 50th enqueue forced to prev_cpu's
local DSQ) under heavy mixed load (2x CPU oversubscription, yield and
context-switch pressure, SCHED_FIFO bursts, periodic fork storms, mixed
nice levels, C-states disabled), measuring local DSQ residence time
(insert to remove) over 5 x 120s runs (~1.2M tasks per set):
Nakamura Shuta [Mon, 19 Jan 2026 06:29:25 +0000 (15:29 +0900)]
rust: str: improve safety comment for CString::try_from_fmt
Improve the safety comment for the `inc_len()` call in
`CString::try_from_fmt()` to clarify why `bytes_written()` is
guaranteed not to exceed the buffer capacity.
The current comment states that bytes written is bounded by size,
but does not explain that this invariant is maintained because:
1. The `Formatter` is created with `size` as its capacity limit
2. The `?` operators on `write_fmt` and `write_str` ensure early
return if writing exceeds this limit
Werner Kasselman [Tue, 17 Mar 2026 07:55:37 +0000 (07:55 +0000)]
ksmbd: fix memory leaks and NULL deref in smb2_lock()
smb2_lock() has three error handling issues after list_del() detaches
smb_lock from lock_list at no_check_cl:
1) If vfs_lock_file() returns an unexpected error in the non-UNLOCK
path, goto out leaks smb_lock and its flock because the out:
handler only iterates lock_list and rollback_list, neither of
which contains the detached smb_lock.
2) If vfs_lock_file() returns -ENOENT in the UNLOCK path, goto out
leaks smb_lock and flock for the same reason. The error code
returned to the dispatcher is also stale.
3) In the rollback path, smb_flock_init() can return NULL on
allocation failure. The result is dereferenced unconditionally,
causing a kernel NULL pointer dereference. Add a NULL check to
prevent the crash and clean up the bookkeeping; the VFS lock
itself cannot be rolled back without the allocation and will be
released at file or connection teardown.
Fix cases 1 and 2 by hoisting the locks_free_lock()/kfree() to before
the if(!rc) check in the UNLOCK branch so all exit paths share one
free site, and by freeing smb_lock and flock before goto out in the
non-UNLOCK branch. Propagate the correct error code in both cases.
Fix case 3 by wrapping the VFS unlock in an if(rlock) guard and adding
a NULL check for locks_free_lock(rlock) in the shared cleanup.
Found via call-graph analysis using sqry.
Fixes: e2f34481b24d ("cifsd: add server-side procedures for SMB3") Cc: stable@vger.kernel.org Suggested-by: ChenXiaoSong <chenxiaosong@kylinos.cn> Signed-off-by: Werner Kasselman <werner@verivus.com> Reviewed-by: ChenXiaoSong <chenxiaosong@kylinos.cn> Acked-by: Namjae Jeon <linkinjeon@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
Werner Kasselman [Mon, 16 Mar 2026 11:38:47 +0000 (11:38 +0000)]
ksmbd: fix use-after-free and NULL deref in smb_grant_oplock()
smb_grant_oplock() has two issues in the oplock publication sequence:
1) opinfo is linked into ci->m_op_list (via opinfo_add) before
add_lease_global_list() is called. If add_lease_global_list()
fails (kmalloc returns NULL), the error path frees the opinfo
via __free_opinfo() while it is still linked in ci->m_op_list.
Concurrent m_op_list readers (opinfo_get_list, or direct iteration
in smb_break_all_levII_oplock) dereference the freed node.
2) opinfo->o_fp is assigned after add_lease_global_list() publishes
the opinfo on the global lease list. A concurrent
find_same_lease_key() can walk the lease list and dereference
opinfo->o_fp->f_ci while o_fp is still NULL.
Fix by restructuring the publication sequence to eliminate post-publish
failure:
- Set opinfo->o_fp before any list publication (fixes NULL deref).
- Preallocate lease_table via alloc_lease_table() before opinfo_add()
so add_lease_global_list() becomes infallible after publication.
- Keep the original m_op_list publication order (opinfo_add before
lease list) so concurrent opens via same_client_has_lease() and
opinfo_get_list() still see the in-flight grant.
- Use opinfo_put() instead of __free_opinfo() on err_out so that
the RCU-deferred free path is used.
This also requires splitting add_lease_global_list() to take a
preallocated lease_table and changing its return type from int to void,
since it can no longer fail.
Fixes: 1dfd062caa16 ("ksmbd: fix use-after-free by using call_rcu() for oplock_info") Cc: stable@vger.kernel.org Signed-off-by: Werner Kasselman <werner@verivus.com> Reviewed-by: ChenXiaoSong <chenxiaosong@kylinos.cn> Acked-by: Namjae Jeon <linkinjeon@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
Hyunwoo Kim [Mon, 16 Mar 2026 23:52:01 +0000 (08:52 +0900)]
ksmbd: do not expire session on binding failure
When a multichannel session binding request fails (e.g. wrong password),
the error path unconditionally sets sess->state = SMB2_SESSION_EXPIRED.
However, during binding, sess points to the target session looked up via
ksmbd_session_lookup_slowpath() -- which belongs to another connection's
user. This allows a remote attacker to invalidate any active session by
simply sending a binding request with a wrong password (DoS).
Fix this by skipping session expiration when the failed request was
a binding attempt, since the session does not belong to the current
connection. The reference taken by ksmbd_session_lookup_slowpath() is
still correctly released via ksmbd_user_session_put().
Cc: stable@vger.kernel.org Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com> Acked-by: Namjae Jeon <linkinjeon@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
tools headers: Synchronize linux/build_bug.h with the kernel sources
To pick up the changes in:
6ffd853b0b10e1e2 ("build_bug.h: correct function parameters names in kernel-doc")
That just add some comments, addressing this perf tools build warning:
Warning: Kernel ABI header differences:
diff -u tools/include/linux/build_bug.h include/linux/build_bug.h
Please take a look at tools/include/uapi/README for further info on this
synchronization process.
Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ian Rogers <irogers@google.com> Cc: Randy Dunlap <rdunlap@infradead.org> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Jonathan Corbet [Sun, 22 Mar 2026 21:25:08 +0000 (15:25 -0600)]
Merge branch 'mauro' into docs-mw
This series comes after:
https://lore.kernel.org/linux-doc/cover.1773770483.git.mchehab+huawei@kernel.org/
It basically contains patches I submitted before on a 40+ patch series,
but were less relevant, plus a couple of other minor fixes:
- patch 1 improves one of the CTokenizer unit test, fixing some
potential issues on it;
- patches 2 and 3 contain some improvement/fixes for Sphinx
Python autodoc extension. They basically document c_lex.py;
- The remaining patches:
- create a new class for kernel-doc config;
- fix some internal representations of KdocItem;
- add unit tests for KernelDoc() parser class;
- add support to output KdocItem in YAML, which is a
machine-readable output for all documented kAPI.
None of the patches should affect man or html output.
docs: test_kdoc_parser: add support for dynamic test creation
Use the content of kdoc-test.yaml to generate unittests to
verify that kernel-doc internal methods are parsing C code
and generating output the expected way.
Depending on what is written at the parser file at
kdoc-test.yaml, up to 5 tests can be generated from a single
test entry inside the YAML file:
1. from source to kdoc_item: test KernelDoc class;
2. from kdoc_item to man: test ManOutput class;
3. from kdoc_item to rst: test RestOutput class;
4. from source to man without checking expected KdocItem;
5. from source to rst without checking expected KdocItem.
docs: add a simple kdoc-test.yaml together with a validation tool
Create a simple kdoc-test.yaml to be used to create unit tests for
kernel-doc parser and output classes.
For now, all we want is a simple function mapped on a yaml test
using the defined schema.
To be sure that the schema is followed, add an unittest for
the file, which will also validate that the schema is properly
parsed.
It should be noticed that the .TH definition for the man format
contains a timestamp. We'll need to handle that when dealing with
the actual implementation for the ManOutput class unit tests.
docs: unittests: add a parser to test kernel-doc parser logic
Validating that kernel-doc is parsing data properly is tricky.
Add an unittest skeleton that alllows passing a source code
and check if the corresponding values of export_table and
entries returned by the parser are properly filled.
It works by mocking a file input with the contents of a source
string, an comparing if:
- exports set matches;
- expected KernelItem entries match.
Create a new TestSelfValidate meant to check if the logic
inside KdocParser.run_test() does its job of checking for
differences inside KdocItem.
docs: kdoc_item: add support to generate a KdocItem from a dict
When reading the contents on a KdocItem using YAML, the data
will be imported into a dict.
Add a method to create a new KdocItem from a dict to allow
converting such input into a real KdocItem.
While here, address an issue that, if the class is initialized
with an internal parameter outside the 4 initial arguments,
it would end being added inside other_stuff, which breaks
initializing it from a dict.
projects / thirdparty / linux.git / log
