In the debug and resolution of an issue involving forced use of bounce
buffers, 7170130e4c72 ("x86/mm/init: Handle the special case of device
private pages in add_pages(), to not increase max_pfn and trigger
dma_addressing_limited() bounce buffers"). It would have been easier
to debug the issue if dma_addressing_limited() had debug information
about the device not being able to address all of memory and thus forcing
all accesses through a bounce buffer. Please see[2]
Implement dev_dbg to debug the potential use of bounce buffers
when we hit the condition. When swiotlb is used,
dma_addressing_limited() is used to determine the size of maximum dma
buffer size in dma_direct_max_mapping_size(). The debug prints could be
triggered in that check as well (when enabled).
Link: https://lore.kernel.org/lkml/20250401000752.249348-1-balbirs@nvidia.com/ Link: https://lore.kernel.org/lkml/20250310112206.4168-1-spasswolf@web.de/ Cc: Marek Szyprowski <m.szyprowski@samsung.com> Cc: Robin Murphy <robin.murphy@arm.com> Cc: "Christian König" <christian.koenig@amd.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Kees Cook <kees@kernel.org> Cc: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: Bert Karwatzki <spasswolf@web.de> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Balbir Singh <balbirs@nvidia.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com> Link: https://lore.kernel.org/r/20250414113752.3298276-1-balbirs@nvidia.com Signed-off-by: Sasha Levin <sashal@kernel.org>
syzbot reports a data-race when accessing the event_triggered, here is the
simplified stack when the issue occurred:
==================================================================
BUG: KCSAN: data-race in virtqueue_disable_cb / virtqueue_enable_cb_delayed
write to 0xffff8881025bc452 of 1 bytes by task 3288 on cpu 0:
virtqueue_enable_cb_delayed+0x42/0x3c0 drivers/virtio/virtio_ring.c:2653
start_xmit+0x230/0x1310 drivers/net/virtio_net.c:3264
__netdev_start_xmit include/linux/netdevice.h:5151 [inline]
netdev_start_xmit include/linux/netdevice.h:5160 [inline]
xmit_one net/core/dev.c:3800 [inline]
read to 0xffff8881025bc452 of 1 bytes by interrupt on cpu 1:
virtqueue_disable_cb_split drivers/virtio/virtio_ring.c:880 [inline]
virtqueue_disable_cb+0x92/0x180 drivers/virtio/virtio_ring.c:2566
skb_xmit_done+0x5f/0x140 drivers/net/virtio_net.c:777
vring_interrupt+0x161/0x190 drivers/virtio/virtio_ring.c:2715
__handle_irq_event_percpu+0x95/0x490 kernel/irq/handle.c:158
handle_irq_event_percpu kernel/irq/handle.c:193 [inline]
value changed: 0x01 -> 0x00
==================================================================
When the data race occurs, the function virtqueue_enable_cb_delayed() sets
event_triggered to false, and virtqueue_disable_cb_split/packed() reads it
as false due to the race condition. Since event_triggered is an unreliable
hint used for optimization, this should only cause the driver temporarily
suggest that the device not send an interrupt notification when the event
index is used.
Fix this KCSAN reported data-race issue by explicitly tagging the access as
data_racy.
Reported-by: syzbot+efe683d57990864b8c8e@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/67c7761a.050a0220.15b4b9.0018.GAE@google.com/ Signed-off-by: Zhongqiu Han <quic_zhonhan@quicinc.com>
Message-Id: <20250312130412.3516307-1-quic_zhonhan@quicinc.com> Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Acked-by: Jason Wang <jasowang@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Samsung UFS devices require additional time in hibern8 mode before
exiting, beyond the negotiated handshaking phase between the host and
device. Introduce a quirk to increase the PA_HIBERN8TIME parameter by
100 µs, a value derived from experiments, to ensure a proper hibernation
process.
Signed-off-by: Manish Pandey <quic_mapa@quicinc.com> Link: https://lore.kernel.org/r/20250411121630.21330-3-quic_mapa@quicinc.com Reviewed-by: Bean Huo <beanhuo@micron.com> Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
All platforms since Snapdragon 8 Gen1 (SM8450) require using 4-byte
reads to access QFPROM data. While older platforms were more than happy
with 1-byte reads, change the qfprom driver to use 4-byte reads for all
the platforms. Specify stride and word size of 4 bytes. To retain
compatibility with the existing DT and to simplify porting data from
vendor kernels, use fixup_dt_cell_info in order to bump alignment
requirements.
If NVMEM cell uses bit offset or specifies bit truncation, update
raw_len manually (following the cell->bytes update), ensuring that the
NVMEM access is still word-aligned.
Check that the NVMEM cell's raw_len is a aligned to word_size. Otherwise
Otherwise drivers might face incomplete read while accessing the last
part of the NVMEM cell.
If the NVMEM specifies a stride to access data, reading particular cell
might require bit offset that is bigger than one byte. Rework NVMEM core
code to support bit offsets of more than 8 bits.
The RK3588 has an offset into the OTP area where the readable area begins
and automatically adds this to the start address.
Other variants are very much similar to rk3588, just with a different
offset, so move that value into variant-data.
To match the size in bytes, store this value also in bytes and not in
number of blocks.
The function nvmet_pci_epf_delete_cq() unconditionally calls
nvmet_pci_epf_remove_irq_vector() even for completion queues that do not
have interrupts enabled. Furthermore, for completion queues that do
have IRQ enabled, deleting and re-creating the completion queue leaves
the flag NVMET_PCI_EPF_Q_IRQ_ENABLED set, even if the completion queue
is being re-created with IRQ disabled.
Fix these issues by calling nvmet_pci_epf_remove_irq_vector() only if
NVMET_PCI_EPF_Q_IRQ_ENABLED is set and make sure to always clear that
flag.
Fixes: 0faa0fe6f90e ("nvmet: New NVMe PCI endpoint function target driver") Cc: stable@vger.kernel.org Signed-off-by: Damien Le Moal <dlemoal@kernel.org> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Sasha Levin <sashal@kernel.org>
Instead of mapping and unmapping the completion queues memory to the
host PCI address space whenever nvmet_pci_epf_cq_work() is called, map
a completion queue to the host PCI address space when the completion
queue is created with nvmet_pci_epf_create_cq() and unmap it when the
completion queue is deleted with nvmet_pci_epf_delete_cq().
This removes the completion queue mapping/unmapping from
nvmet_pci_epf_cq_work() and significantly increases performance. For
a single job 4K random read QD=1 workload, the IOPS is increased from
23 KIOPS to 25 KIOPS. Some significant throughput increasde for high
queue depth and large IOs workloads can also be seen.
Since the functions nvmet_pci_epf_map_queue() and
nvmet_pci_epf_unmap_queue() are called respectively only from
nvmet_pci_epf_create_cq() and nvmet_pci_epf_delete_cq(), these functions
are removed and open-coded in their respective call sites.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
Stable-dep-of: 85adf2094abb ("nvmet: pci-epf: clear completion queue IRQ flag on delete") Signed-off-by: Sasha Levin <sashal@kernel.org>
The phy-rcar-gen3-usb2 driver exposes four individual PHYs that are
requested and configured by PHY users. The struct phy_ops APIs access the
same set of registers to configure all PHYs. Additionally, PHY settings can
be modified through sysfs or an IRQ handler. While some struct phy_ops APIs
are protected by a driver-wide mutex, others rely on individual
PHY-specific mutexes.
This approach can lead to various issues, including:
1/ the IRQ handler may interrupt PHY settings in progress, racing with
hardware configuration protected by a mutex lock
2/ due to msleep(20) in rcar_gen3_init_otg(), while a configuration thread
suspends to wait for the delay, another thread may try to configure
another PHY (with phy_init() + phy_power_on()); re-running the
phy_init() goes to the exact same configuration code, re-running the
same hardware configuration on the same set of registers (and bits)
which might impact the result of the msleep for the 1st configuring
thread
3/ sysfs can configure the hardware (though role_store()) and it can
still race with the phy_init()/phy_power_on() APIs calling into the
drivers struct phy_ops
To address these issues, add a spinlock to protect hardware register access
and driver private data structures (e.g., calls to
rcar_gen3_is_any_rphy_initialized()). Checking driver-specific data remains
necessary as all PHY instances share common settings. With this change,
the existing mutex protection is removed and the cleanup.h helpers are
used.
While at it, to keep the code simpler, do not skip
regulator_enable()/regulator_disable() APIs in
rcar_gen3_phy_usb2_power_on()/rcar_gen3_phy_usb2_power_off() as the
regulators enable/disable operations are reference counted anyway.
Commit 08b0ad375ca6 ("phy: renesas: rcar-gen3-usb2: move IRQ registration
to init") moved the IRQ request operation from probe to
struct phy_ops::phy_init API to avoid triggering interrupts (which lead to
register accesses) while the PHY clocks (enabled through runtime PM APIs)
are not active. If this happens, it results in a synchronous abort.
One way to reproduce this issue is by enabling CONFIG_DEBUG_SHIRQ, which
calls free_irq() on driver removal.
Move the IRQ request and free operations back to probe, and take the
runtime PM state into account in IRQ handler. This commit is preparatory
for the subsequent fixes in this series.
[Why & How]
Instead of dropping DRR updates, defer them. This fixes issues where
monitor continues to see incorrect refresh rate after VRR was turned off
by userspace.
Fixes: 32953485c558 ("drm/amd/display: Do not update DRR while BW optimizations pending") Link: https://gitlab.freedesktop.org/drm/amd/-/issues/3546 Reviewed-by: Sun peng Li <sunpeng.li@amd.com> Signed-off-by: John Olender <john.olender@gmail.com> Signed-off-by: Aurabindo Pillai <aurabindo.pillai@amd.com> Signed-off-by: Ray Wu <ray.wu@amd.com> Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 53761b7ecd83e6fbb9f2206f8c980a6aa308c844) Cc: stable@vger.kernel.org Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why&How]
stream->adjust will be overwritten by update->crtc_timing_adjust.
We should set update->crtc_timing_adjust->timing_adjust_pending
and then overwrite stream->adjust.
Reset update->crtc_timing_adjust->timing_adjust_pending after
the assignment.
Reviewed-by: Charlene Liu <charlene.liu@amd.com> Signed-off-by: Zhongwei Zhang <Zhongwei.Zhang@amd.com> Signed-off-by: Tom Chung <chiahsuan.chung@amd.com> Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Stable-dep-of: 874697e12793 ("drm/amd/display: Defer BW-optimization-blocked DRR adjustments") Signed-off-by: Sasha Levin <sashal@kernel.org>
We used to take a lock in tegra186_utmi_bias_pad_power_on() but now we
have moved the lock into the caller. Unfortunately, when we moved the
lock this unlock was left behind and it results in a double unlock.
Delete it now.
Fixes: b47158fb4295 ("phy: tegra: xusb: Use a bitmask for UTMI pad power state tracking") Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Jon Hunter <jonathanh@nvidia.com> Link: https://lore.kernel.org/r/aAjmR6To4EnvRl4G@stanley.mountain Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Tiezhu Yang [Tue, 20 May 2025 06:30:09 +0000 (14:30 +0800)]
perf tools: Fix build error for LoongArch
There exists the following error when building perf tools on LoongArch:
CC util/syscalltbl.o
In file included from util/syscalltbl.c:16:
tools/perf/arch/loongarch/include/syscall_table.h:2:10: fatal error: asm/syscall_table_64.h: No such file or directory
2 | #include <asm/syscall_table_64.h>
| ^~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
This is because the generated syscall header is syscalls_64.h rather
than syscall_table_64.h. The above problem was introduced from v6.14,
then the header syscall_table.h has been removed from mainline tree
in commit af472d3c4454 ("perf syscalltbl: Remove syscall_table.h"),
just fix it only for the linux-6.14.y branch of stable tree.
By the way, no need to fix the mainline tree and there is no upstream
git id for this patch.
How to reproduce:
git clone https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
cd linux && git checkout origin/linux-6.14.y
make JOBS=1 -C tools/perf
Fixes: fa70857a27e5 ("perf tools loongarch: Use syscall table") Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The page allocator tracks the number of zones that have unaccepted memory
using static_branch_enc/dec() and uses that static branch in hot paths to
determine if it needs to deal with unaccepted memory.
Borislav and Thomas pointed out that the tracking is racy: operations on
static_branch are not serialized against adding/removing unaccepted pages
to/from the zone.
WARNING: CPU: 0 PID: 10 at kernel/jump_label.c:276 __static_key_slow_dec_cpuslocked+0x8e/0xa0
The comment around the WARN() explains the problem:
/*
* Warn about the '-1' case though; since that means a
* decrement is concurrent with a first (0->1) increment. IOW
* people are trying to disable something that wasn't yet fully
* enabled. This suggests an ordering problem on the user side.
*/
The effect of this static_branch optimization is only visible on
microbenchmark.
Instead of adding more complexity around it, remove it altogether.
Link: https://lkml.kernel.org/r/20250506133207.1009676-1-kirill.shutemov@linux.intel.com Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Fixes: dcdfdd40fa82 ("mm: Add support for unaccepted memory") Link: https://lore.kernel.org/all/20250506092445.GBaBnVXXyvnazly6iF@fat_crate.local Reported-by: Borislav Petkov <bp@alien8.de> Tested-by: Borislav Petkov (AMD) <bp@alien8.de> Reported-by: Thomas Gleixner <tglx@linutronix.de> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Brendan Jackman <jackmanb@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: <stable@vger.kernel.org> [6.5+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The workqueue used for the reset worker is marked as WQ_MEM_RECLAIM,
while the GSC one isn't (and can't be as we need to do memory
allocations in the gsc worker). Therefore, we can't flush the latter
from the former.
The reason why we had such a flush was to avoid interrupting either
the GSC FW load or in progress GSC proxy operations. GSC proxy
operations fall into 2 categories:
1) GSC proxy init: this only happens once immediately after GSC FW load
and does not support being interrupted. The only way to recover from
an interruption of the proxy init is to do an FLR and re-load the GSC.
2) GSC proxy request: this can happen in response to a request that
the driver sends to the GSC. If this is interrupted, the GSC FW will
timeout and the driver request will be failed, but overall the GSC
will keep working fine.
Flushing the work allowed us to avoid interruption in both cases (unless
the hang came from the GSC engine itself, in which case we're toast
anyway). However, a failure on a proxy request is tolerable if we're in
a scenario where we're triggering a GT reset (i.e., something is already
gone pretty wrong), so what we really need to avoid is interrupting
the init flow, which we can do by polling on the register that reports
when the proxy init is complete (as that ensure us that all the load and
init operations have been completed).
Note that during suspend we still want to do a flush of the worker to
make sure it completes any operations involving the HW before the power
is cut.
v2: fix spelling in commit msg, rename waiter function (Julia)
Fixes: dd0e89e5edc2 ("drm/xe/gsc: GSC FW load") Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4830 Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: John Harrison <John.C.Harrison@Intel.com> Cc: Alan Previn <alan.previn.teres.alexis@intel.com> Cc: <stable@vger.kernel.org> # v6.8+ Reviewed-by: Julia Filipchuk <julia.filipchuk@intel.com> Link: https://lore.kernel.org/r/20250502155104.2201469-1-daniele.ceraolospurio@intel.com
(cherry picked from commit 12370bfcc4f0bdf70279ec5b570eb298963422b5) Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use flush_work() instead of cancel_work_sync() for driver IRQ
workqueues to guarantee that remaining pending work
will be handled.
This resolves two issues that were encountered where a driver was left
in an incorrect state as the bottom-half was canceled:
1. Cancelling context-abort of a job that is still executing and
is causing translation faults which is going to cause additional TDRs
2. Cancelling bottom-half of a DCT (duty-cycle throttling) request
which will cause a device to not be adjusted to an external frequency
request.
Fixes: bc3e5f48b7ee ("accel/ivpu: Use workqueue for IRQ handling") Signed-off-by: Maciej Falkowski <maciej.falkowski@linux.intel.com> Reviewed-by: Lizhi Hou <lizhi.hou@amd.com> Reviewed-by: Jeff Hugo <jeff.hugo@oss.qualcomm.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://lore.kernel.org/r/20250401155755.4049156-1-maciej.falkowski@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Move the ivpu_mmu_discard_events() function to the common portion of
the abort work function. This ensures it is called only once, even if
there are no faulty contexts in context_xa, to guarantee that MMU events
are discarded and new events are not missed.
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Signed-off-by: Karol Wachowski <karol.wachowski@intel.com> Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20250129125636.1047413-4-jacek.lawrynowicz@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Generate recovery when fault from reserved context is detected.
Add Abort (A) bit to reserved (1) SSID to ensure NPU also receives a fault.
There is no way to create a file_priv with reserved SSID
but it is still possible to receive MMU faults from that SSID
as it is a default NPU HW setting. Such situation will occur if
FW freed context related resources but still performed access to DRAM.
Signed-off-by: Karol Wachowski <karol.wachowski@intel.com> Signed-off-by: Maciej Falkowski <maciej.falkowski@linux.intel.com> Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20250107173238.381120-9-maciej.falkowski@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To prevent looping infinitely in MMU event handler we stop
generating new events by removing 'R' (record) bit from context
descriptor, but to ensure this change has effect KMD has to perform
configuration invalidation followed by sync command.
Because of that move parts of the interrupt handler that can take longer
to a thread not to block in interrupt handler for too long.
This includes:
* disabling event queue for the time KMD updates MMU event queue consumer
to ensure proper synchronization between MMU and KMD
* removal of 'R' (record) bit from context descriptor to ensure no more
faults are recorded until that context is destroyed
Signed-off-by: Karol Wachowski <karol.wachowski@intel.com> Signed-off-by: Maciej Falkowski <maciej.falkowski@linux.intel.com> Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20250107173238.381120-8-maciej.falkowski@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Stop dumping consecutive faults from an already faulty context immediately,
instead of waiting for the context abort thread handler (IRQ handler bottom
half) to abort currently executing jobs.
Remove 'R' (record events) bit from context descriptor of a faulty
context to prevent future faults generation.
This change speeds up the IRQ handler by eliminating the need to print the
fault content repeatedly. Additionally, it prevents flooding dmesg with
errors, which was occurring due to the delay in the bottom half of the
handler stopping fault-generating jobs.
Signed-off-by: Karol Wachowski <karol.wachowski@intel.com> Signed-off-by: Maciej Falkowski <maciej.falkowski@linux.intel.com> Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20250107173238.381120-7-maciej.falkowski@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Convert IRQ bottom half from the thread handler into workqueue.
This increases a stability in rare scenarios where driver on
debugging/hardening kernels processes IRQ too slow and misses
some interrupts due to it.
Workqueue handler also gives a very minor performance increase.
The idxd_cleanup() helper cleans up perfmon, interrupts, internals and
so on. Refactor remove call with the idxd_cleanup() helper to avoid code
duplication. Note, this also fixes the missing put_device() for idxd
groups, enginces and wqs.
Fixes: bfe1d56091c1 ("dmaengine: idxd: Init and probe for Intel data accelerators") Cc: stable@vger.kernel.org Suggested-by: Vinicius Costa Gomes <vinicius.gomes@intel.com> Signed-off-by: Shuai Xue <xueshuai@linux.alibaba.com> Reviewed-by: Fenghua Yu <fenghuay@nvidia.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Link: https://lore.kernel.org/r/20250404120217.48772-10-xueshuai@linux.alibaba.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Memory allocated for idxd is not freed if an error occurs during
idxd_pci_probe(). To fix it, free the allocated memory in the reverse
order of allocation before exiting the function in case of an error.
Fixes: bfe1d56091c1 ("dmaengine: idxd: Init and probe for Intel data accelerators") Cc: stable@vger.kernel.org Signed-off-by: Shuai Xue <xueshuai@linux.alibaba.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Reviewed-by: Fenghua Yu <fenghuay@nvidia.com> Link: https://lore.kernel.org/r/20250404120217.48772-8-xueshuai@linux.alibaba.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Memory allocated for idxd is not freed if an error occurs during
idxd_alloc(). To fix it, free the allocated memory in the reverse order
of allocation before exiting the function in case of an error.
The remove call stack is missing idxd cleanup to free bitmap, ida and
the idxd_device. Call idxd_free() helper routines to make sure we exit
gracefully.
Fixes: bfe1d56091c1 ("dmaengine: idxd: Init and probe for Intel data accelerators") Cc: stable@vger.kernel.org Suggested-by: Vinicius Costa Gomes <vinicius.gomes@intel.com> Signed-off-by: Shuai Xue <xueshuai@linux.alibaba.com> Reviewed-by: Fenghua Yu <fenghuay@nvidia.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Link: https://lore.kernel.org/r/20250404120217.48772-9-xueshuai@linux.alibaba.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The idxd_cleanup_internals() function only decreases the reference count
of groups, engines, and wqs but is missing the step to release memory
resources.
To fix this, use the cleanup helper to properly release the memory
resources.
Fixes: ddf742d4f3f1 ("dmaengine: idxd: Add missing cleanup for early error out in probe call") Cc: stable@vger.kernel.org Signed-off-by: Shuai Xue <xueshuai@linux.alibaba.com> Reviewed-by: Fenghua Yu <fenghuay@nvidia.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Link: https://lore.kernel.org/r/20250404120217.48772-6-xueshuai@linux.alibaba.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Memory allocated for groups is not freed if an error occurs during
idxd_setup_groups(). To fix it, free the allocated memory in the reverse
order of allocation before exiting the function in case of an error.
Memory allocated for engines is not freed if an error occurs during
idxd_setup_engines(). To fix it, free the allocated memory in the
reverse order of allocation before exiting the function in case of an
error.
Memory allocated for wqs is not freed if an error occurs during
idxd_setup_wqs(). To fix it, free the allocated memory in the reverse
order of allocation before exiting the function in case of an error.
Currently, a local dma_cap_mask_t variable is used to store device
cap_mask within udma_of_xlate(). However, the DMA_PRIVATE flag in
the device cap_mask can get cleared when the last channel is released.
This can happen right after storing the cap_mask locally in
udma_of_xlate(), and subsequent dma_request_channel() can fail due to
mismatch in the cap_mask. Fix this by removing the local dma_cap_mask_t
variable and directly using the one from the dma_device structure.
As David pointed out, what truly matters for mremap and userfaultfd move
operations is the soft dirty bit. The current comment and
implementation—which always sets the dirty bit for present PTEs and
fails to set the soft dirty bit for swap PTEs—are incorrect. This could
break features like Checkpoint-Restore in Userspace (CRIU).
This patch updates the behavior to correctly set the soft dirty bit for
both present and swap PTEs in accordance with mremap.
Link: https://lkml.kernel.org/r/20250508220912.7275-1-21cnbao@gmail.com Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI") Signed-off-by: Barry Song <v-songbaohua@oppo.com> Reported-by: David Hildenbrand <david@redhat.com> Closes: https://lore.kernel.org/linux-mm/02f14ee1-923f-47e3-a994-4950afb9afcc@redhat.com/ Acked-by: Peter Xu <peterx@redhat.com> Reviewed-by: Suren Baghdasaryan <surenb@google.com> Cc: Lokesh Gidra <lokeshgidra@google.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
During our testing with hugetlb subpool enabled, we observe that
hstate->resv_huge_pages may underflow into negative values. Root cause
analysis reveals a race condition in subpool reservation fallback handling
as follow:
/* Global reservation may fail after subpool allocation */
if (hugetlb_acct_memory(h, gbl_reserve) < 0)
goto out_put_pages;
out_put_pages:
/* This incorrectly restores reservation to subpool */
hugepage_subpool_put_pages(spool, chg);
When hugetlb_acct_memory() fails after subpool allocation, the current
implementation over-commits subpool reservations by returning the full
'chg' value instead of the actual allocated 'gbl_reserve' amount. This
discrepancy propagates to global reservations during subsequent releases,
eventually causing resv_huge_pages underflow.
This problem can be trigger easily with the following steps:
1. reverse hugepage for hugeltb allocation
2. mount hugetlbfs with min_size to enable hugetlb subpool
3. alloc hugepages with two task(make sure the second will fail due to
insufficient amount of hugepages)
4. with for a few seconds and repeat step 3 which will make
hstate->resv_huge_pages to go below zero.
To fix this problem, return corrent amount of pages to subpool during the
fallback after hugepage_subpool_get_pages is called.
Link: https://lkml.kernel.org/r/20250410062633.3102457-1-mawupeng1@huawei.com Fixes: 1c5ecae3a93f ("hugetlbfs: add minimum size accounting to subpools") Signed-off-by: Wupeng Ma <mawupeng1@huawei.com> Tested-by: Joshua Hahn <joshua.hahnjy@gmail.com> Reviewed-by: Oscar Salvador <osalvador@suse.de> Cc: David Hildenbrand <david@redhat.com> Cc: Ma Wupeng <mawupeng1@huawei.com> Cc: Muchun Song <muchun.song@linux.dev> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Modify the check for whether the timer is initialized during IO transfer
when passthrough is used with hybrid polling, to ensure that it's always
setup correctly.
For older/32-bit systems with highmem, don't assume that the pages in
a mapped region are always going to be mapped. If io_region_init_ptr()
finds that the pages are coalescable, also check if the first page is
a HighMem page or not. If it is, fall through to the usual vmap()
mapping rather than attempt to get the unmapped page address.
After a recent change [1] in clang's randstruct implementation to
randomize structures that only contain function pointers, there is an
error because qede_ll_ops get randomized but does not use a designated
initializer for the first member:
drivers/net/ethernet/qlogic/qede/qede_main.c:206:2: error: a randomized struct can only be initialized with a designated initializer
206 | {
| ^
Explicitly initialize the common member using a designated initializer
to fix the build.
The ring buffer is made up of sub buffers (sometimes called pages as they
are by default PAGE_SIZE). It has the following "pages":
"tail page" - this is the page that the next write will write to
"head page" - this is the page that the reader will swap the reader page with.
"reader page" - This belongs to the reader, where it will swap the head
page from the ring buffer so that the reader does not
race with the writer.
The writer may end up on the "reader page" if the ring buffer hasn't
written more than one page, where the "tail page" and the "head page" are
the same.
The persistent ring buffer has meta data that points to where these pages
exist so on reboot it can re-create the pointers to the cpu_buffer
descriptor. But when the commit page is on the reader page, the logic is
incorrect.
The check to see if the commit page is on the reader page checked if the
head page was the reader page, which would never happen, as the head page
is always in the ring buffer. The correct check would be to test if the
commit page is on the reader page. If that's the case, then it can exit
out early as the commit page is only on the reader page when there's only
one page of data in the buffer. There's no reason to iterate the ring
buffer pages to find the "commit page" as it is already found.
On boot up, the dmesg will have:
Ring buffer meta [0] is from previous boot!
Ring buffer meta [1] is from previous boot!
Ring buffer meta [2] is from previous boot!
Ring buffer meta [3] is from previous boot!
Ring buffer meta [4] commit page not found
Ring buffer meta [5] is from previous boot!
Ring buffer meta [6] is from previous boot!
Ring buffer meta [7] is from previous boot!
Where the buffer on CPU 4 had a "commit page not found" error and that
buffer is cleared and reset causing the output to be empty and the data lost.
A warning on driver removal started occurring after commit 9dd05df8403b
("net: warn if NAPI instance wasn't shut down"). Disable tx napi before
deleting it in mt76_dma_cleanup().
Tested with mt7921e but the same pattern can be actually applied to other
mt76 drivers calling mt76_dma_cleanup() during removal. Tx napi is enabled
in their *_dma_init() functions and only toggled off and on again inside
their suspend/resume/reset paths. So it should be okay to disable tx
napi in such a generic way.
Found by Linux Verification Center (linuxtesting.org).
Fixes: 2ac515a5d74f ("mt76: mt76x02: use napi polling for tx cleanup") Cc: stable@vger.kernel.org Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> Tested-by: Ming Yen Hsieh <mingyen.hsieh@mediatek.com> Link: https://patch.msgid.link/20250506115540.19045-1-pchelkin@ispras.ru Signed-off-by: Felix Fietkau <nbd@nbd.name> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A delay unit of 0 is a valid entry, thus it is not valid to check for
unused delays. Instead, check the value field; if that is zero, the
given delay is unset.
Fixes: 4426e6b4ecf6 ("spi: tegra114: Don't fail set_cs_timing when delays are zero") Cc: stable@vger.kernel.org Signed-off-by: Aaron Kling <webgeek1234@gmail.com> Reviewed-by: Jon Hunter <jonathanh@nvidia.com> Link: https://patch.msgid.link/20250506-spi-tegra114-fixup-v1-1-136dc2f732f3@gmail.com Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The response buffer for the CREATE request handled by smb311_posix_mkdir()
is leaked on the error path (goto err_free_rsp_buf) because the structure
pointer *rsp passed to free_rsp_buf() is not assigned until *after* the
error condition is checked.
As *rsp is initialised to NULL, free_rsp_buf() becomes a no-op and the leak
is instead reported by __kmem_cache_shutdown() upon subsequent rmmod of
cifs.ko if (and only if) the error path has been hit.
Pass rsp_iov.iov_base to free_rsp_buf() instead, similar to the code in
other functions in smb2pdu.c for which *rsp is assigned late.
Cc: stable@vger.kernel.org Signed-off-by: Jethro Donaldson <devel@jro.nz> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The REPORT ZONES buffer size is currently limited by the HBA's maximum
segment count to ensure the buffer can be mapped. However, the block
layer further limits the number of iovec entries to 1024 when allocating
a bio.
To avoid allocation of buffers too large to be mapped, further restrict
the maximum buffer size to BIO_MAX_INLINE_VECS.
Replace the UIO_MAXIOV symbolic name with the more contextually
appropriate BIO_MAX_INLINE_VECS.
Fixes: b091ac616846 ("sd_zbc: Fix report zones buffer allocation") Cc: stable@vger.kernel.org Signed-off-by: Steve Siwinski <ssiwinski@atto.com> Link: https://lore.kernel.org/r/20250508200122.243129-1-ssiwinski@atto.com Reviewed-by: Damien Le Moal <dlemoal@kernel.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
phy-rcar-gen3-usb2 driver exports 4 PHYs. The timing registers are common
to all PHYs. There is no need to set them every time a PHY is initialized.
Set timing register only when the 1st PHY is initialized.
It has been observed on the Renesas RZ/G3S SoC that unbinding and binding
the PHY driver leads to role autodetection failures. This issue occurs when
PHY 3 is the first initialized PHY. PHY 3 does not have an interrupt
associated with the USB2_INT_ENABLE register (as
rcar_gen3_int_enable[3] = 0). As a result, rcar_gen3_init_otg() is called
to initialize OTG without enabling PHY interrupts.
To resolve this, add rcar_gen3_is_any_otg_rphy_initialized() and call it in
role_store(), role_show(), and rcar_gen3_init_otg(). At the same time,
rcar_gen3_init_otg() is only called when initialization for a PHY with
interrupt bits is in progress. As a result, the
struct rcar_gen3_phy::otg_initialized is no longer needed.
The KSZ9477 PHY driver contained workarounds for broken EEE capability
advertisements by manually masking supported EEE modes and forcibly
disabling EEE if MICREL_NO_EEE was set.
With proper MAC-side EEE handling implemented via phylink, these quirks
are no longer necessary. Remove MICREL_NO_EEE handling and the use of
ksz9477_get_features().
This simplifies the PHY driver and avoids duplicated EEE management logic.
Phylink expects MAC drivers to provide LPI callbacks to properly manage
Energy Efficient Ethernet (EEE) configuration. On KSZ switches with
integrated PHYs, LPI is internally handled by hardware, while ports
without integrated PHYs have no documented MAC-level LPI support.
Provide dummy mac_disable_tx_lpi() and mac_enable_tx_lpi() callbacks to
satisfy phylink requirements. Also, set default EEE capabilities during
phylink initialization where applicable.
Since phylink can now gracefully handle optional EEE configuration,
remove the need for the MICREL_NO_EEE PHY flag.
This change addresses issues caused by incomplete EEE refactoring
introduced in commit fe0d4fd9285e ("net: phy: Keep track of EEE
configuration"). It is not easily possible to fix all older kernels, but
this patch ensures proper behavior on latest kernels and can be
considered for backporting to stable kernels starting from v6.14.
If device_add() fails, do not use device_unregister() for error
handling. device_unregister() consists two functions: device_del() and
put_device(). device_unregister() should only be called after
device_add() succeeded because device_del() undoes what device_add()
does if successful. Change device_unregister() to put_device() call
before returning from the function.
As comment of device_add() says, 'if device_add() succeeds, you should
call device_del() when you want to get rid of it. If device_add() has
not succeeded, use only put_device() to drop the reference count'.
Found by code review.
Cc: stable@vger.kernel.org Fixes: 53d2a715c240 ("phy: Add Tegra XUSB pad controller support") Signed-off-by: Ma Ke <make24@iscas.ac.cn> Acked-by: Thierry Reding <treding@nvidia.com> Link: https://lore.kernel.org/r/20250303072739.3874987-1-make24@iscas.ac.cn Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current implementation uses bias_pad_enable as a reference count to
manage the shared bias pad for all UTMI PHYs. However, during system
suspension with connected USB devices, multiple power-down requests for
the UTMI pad result in a mismatch in the reference count, which in turn
produces warnings such as:
The root cause was traced back to the dynamic power-down changes
introduced in commit a30951d31b25 ("xhci: tegra: USB2 pad power controls"),
where the UTMI pad was being powered down without verifying its current
state. This unbalanced behavior led to discrepancies in the reference
count.
To rectify this issue, this patch replaces the single reference counter
with a bitmask, renamed to utmi_pad_enabled. Each bit in the mask
corresponds to one of the four USB2 PHYs, allowing us to track each pad's
enablement status individually.
With this change:
- The bias pad is powered on only when the mask is clear.
- Each UTMI pad is powered on or down based on its corresponding bit
in the mask, preventing redundant operations.
- The overall power state of the shared bias pad is maintained
correctly during suspend/resume cycles.
The mutex used to prevent race conditions during UTMI pad enable/disable
operations has been moved from the tegra186_utmi_bias_pad_power_on/off
functions to the parent functions tegra186_utmi_pad_power_on/down. This
change ensures that there are no race conditions when updating the bitmask.
Cc: stable@vger.kernel.org Fixes: a30951d31b25 ("xhci: tegra: USB2 pad power controls") Signed-off-by: Wayne Chang <waynec@nvidia.com> Reviewed-by: Jon Hunter <jonathanh@nvidia.com> Tested-by: Jon Hunter <jonathanh@nvidia.com> Link: https://lore.kernel.org/r/20250408030905.990474-1-waynec@nvidia.com Signed-off-by: Vinod Koul <vkoul@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When shared pages are being converted to private during kdump, additional
checks are performed. They include handling the case of a GHCB page being
contained within a huge page.
Currently, this check incorrectly skips a page just below the GHCB page from
being transitioned back to private during kdump preparation.
This skipped page causes a 0x404 #VC exception when it is accessed later while
dumping guest memory for vmcore generation.
Correct the range to be checked for GHCB contained in a huge page. Also,
ensure that the skipped huge page containing the GHCB page is transitioned
back to private by applying the correct address mask later when changing GHCBs
to private at end of kdump preparation.
[ bp: Massage commit message. ]
Fixes: 3074152e56c9 ("x86/sev: Convert shared memory back to private on kexec") Signed-off-by: Ashish Kalra <ashish.kalra@amd.com> Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de> Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com> Tested-by: Srikanth Aithal <sraithal@amd.com> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/20250506183529.289549-1-Ashish.Kalra@amd.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When kdump is running makedumpfile to generate vmcore and dump SNP guest
memory it touches the VMSA page of the vCPU executing kdump.
It then results in unrecoverable #NPF/RMP faults as the VMSA page is
marked busy/in-use when the vCPU is running and subsequently a causes
guest softlockup/hang.
Additionally, other APs may be halted in guest mode and their VMSA pages
are marked busy and touching these VMSA pages during guest memory dump
will also cause #NPF.
Issue AP_DESTROY GHCB calls on other APs to ensure they are kicked out
of guest mode and then clear the VMSA bit on their VMSA pages.
If the vCPU running kdump is an AP, mark it's VMSA page as offline to
ensure that makedumpfile excludes that page while dumping guest memory.
Fixes: 3074152e56c9 ("x86/sev: Convert shared memory back to private on kexec") Signed-off-by: Ashish Kalra <ashish.kalra@amd.com> Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de> Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com> Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com> Tested-by: Srikanth Aithal <sraithal@amd.com> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/20250428214151.155464-1-Ashish.Kalra@amd.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When using the stacktrace trigger command to trace syscalls, the
preemption count was consistently reported as 1 when the system call
event itself had 0 (".").
The root cause is that the trace framework disables preemption in __DO_TRACE before
invoking the trigger callback.
Use the tracing_gen_ctx_dec() that will accommodate for the increase of
the preemption count in __DO_TRACE when calling the callback. The result
is the accurate reporting of:
If navi_amd_register_client() fails, the previous i2c_dw_probe() call
should be undone by a corresponding i2c_del_adapter() call, as already done
in the remove function.
A new on by default warning in clang [1] aims to flags instances where
const variables without static or thread local storage or const members
in aggregate types are not initialized because it can lead to an
indeterminate value. This is quite noisy for the kernel due to
instances originating from header files such as:
drivers/gpu/drm/i915/gt/intel_ring.h:62:2: error: default initialization of an object of type 'typeof (ring->size)' (aka 'const unsigned int') leaves the object uninitialized [-Werror,-Wdefault-const-init-var-unsafe]
62 | typecheck(typeof(ring->size), next);
| ^
include/linux/typecheck.h:10:9: note: expanded from macro 'typecheck'
10 | ({ type __dummy; \
| ^
include/net/ip.h:478:14: error: default initialization of an object of type 'typeof (rt->dst.expires)' (aka 'const unsigned long') leaves the object uninitialized [-Werror,-Wdefault-const-init-var-unsafe]
478 | if (mtu && time_before(jiffies, rt->dst.expires))
| ^
include/linux/jiffies.h:138:26: note: expanded from macro 'time_before'
138 | #define time_before(a,b) time_after(b,a)
| ^
include/linux/jiffies.h:128:3: note: expanded from macro 'time_after'
128 | (typecheck(unsigned long, a) && \
| ^
include/linux/typecheck.h:11:12: note: expanded from macro 'typecheck'
11 | typeof(x) __dummy2; \
| ^
include/linux/list.h:409:27: warning: default initialization of an object of type 'union (unnamed union at include/linux/list.h:409:27)' with const member leaves the object uninitialized [-Wdefault-const-init-field-unsafe]
409 | struct list_head *next = smp_load_acquire(&head->next);
| ^
include/asm-generic/barrier.h:176:29: note: expanded from macro 'smp_load_acquire'
176 | #define smp_load_acquire(p) __smp_load_acquire(p)
| ^
arch/arm64/include/asm/barrier.h:164:59: note: expanded from macro '__smp_load_acquire'
164 | union { __unqual_scalar_typeof(*p) __val; char __c[1]; } __u; \
| ^
include/linux/list.h:409:27: note: member '__val' declared 'const' here
crypto/scatterwalk.c:66:22: error: default initialization of an object of type 'struct scatter_walk' with const member leaves the object uninitialized [-Werror,-Wdefault-const-init-field-unsafe]
66 | struct scatter_walk walk;
| ^
include/crypto/algapi.h:112:15: note: member 'addr' declared 'const' here
112 | void *const addr;
| ^
fs/hugetlbfs/inode.c:733:24: error: default initialization of an object of type 'struct vm_area_struct' with const member leaves the object uninitialized [-Werror,-Wdefault-const-init-field-unsafe]
733 | struct vm_area_struct pseudo_vma;
| ^
include/linux/mm_types.h:803:20: note: member 'vm_flags' declared 'const' here
803 | const vm_flags_t vm_flags;
| ^
Silencing the instances from typecheck.h is difficult because '= {}' is
not available in older but supported compilers and '= {0}' would cause
warnings about a literal 0 being treated as NULL. While it might be
possible to come up with a local hack to silence the warning for
clang-21+, it may not be worth it since -Wuninitialized will still
trigger if an uninitialized const variable is actually used.
In all audited cases of the "field" variant of the warning, the members
are either not used in the particular call path, modified through other
means such as memset() / memcpy() because the containing object is not
const, or are within a union with other non-const members.
Since this warning does not appear to have a high signal to noise ratio,
just disable it.
With the netvsc driver changed to use vmbus_sendpacket_mpb_desc()
instead of vmbus_sendpacket_pagebuffer(), the latter has no remaining
callers. Remove it.
Cc: <stable@vger.kernel.org> # 6.1.x Signed-off-by: Michael Kelley <mhklinux@outlook.com> Link: https://patch.msgid.link/20250513000604.1396-6-mhklinux@outlook.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
vmbus_sendpacket_mpb_desc() is currently used only by the storvsc driver
and is hardcoded to create a single GPA range. To allow it to also be
used by the netvsc driver to create multiple GPA ranges, no longer
hardcode as having a single GPA range. Allow the calling driver to
specify the rangecount in the supplied descriptor.
Update the storvsc driver to reflect this new approach.
Cc: <stable@vger.kernel.org> # 6.1.x Signed-off-by: Michael Kelley <mhklinux@outlook.com> Link: https://patch.msgid.link/20250513000604.1396-2-mhklinux@outlook.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
init_page_array() now always creates a single page buffer array entry
for the rndis message, even if the rndis message crosses a page
boundary. As such, the number of page buffer array entries used for
the rndis message must no longer be tracked -- it is always just 1.
Remove the rmsg_pgcnt field and use "1" where the value is needed.
Cc: <stable@vger.kernel.org> # 6.1.x Signed-off-by: Michael Kelley <mhklinux@outlook.com> Link: https://patch.msgid.link/20250513000604.1396-5-mhklinux@outlook.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Starting with commit dca5161f9bd0 ("hv_netvsc: Check status in
SEND_RNDIS_PKT completion message") in the 6.3 kernel, the Linux
driver for Hyper-V synthetic networking (netvsc) occasionally reports
"nvsp_rndis_pkt_complete error status: 2".[1] This error indicates
that Hyper-V has rejected a network packet transmit request from the
guest, and the outgoing network packet is dropped. Higher level
network protocols presumably recover and resend the packet so there is
no functional error, but performance is slightly impacted. Commit dca5161f9bd0 is not the cause of the error -- it only added reporting
of an error that was already happening without any notice. The error
has presumably been present since the netvsc driver was originally
introduced into Linux.
The root cause of the problem is that the netvsc driver in Linux may
send an incorrectly formatted VMBus message to Hyper-V when
transmitting the network packet. The incorrect formatting occurs when
the rndis header of the VMBus message crosses a page boundary due to
how the Linux skb head memory is aligned. In such a case, two PFNs are
required to describe the location of the rndis header, even though
they are contiguous in guest physical address (GPA) space. Hyper-V
requires that two rndis header PFNs be in a single "GPA range" data
struture, but current netvsc code puts each PFN in its own GPA range,
which Hyper-V rejects as an error.
The incorrect formatting occurs only for larger packets that netvsc
must transmit via a VMBus "GPA Direct" message. There's no problem
when netvsc transmits a smaller packet by copying it into a pre-
allocated send buffer slot because the pre-allocated slots don't have
page crossing issues.
After commit 14ad6ed30a10 ("net: allow small head cache usage with
large MAX_SKB_FRAGS values") in the 6.14-rc4 kernel, the error occurs
much more frequently in VMs with 16 or more vCPUs. It may occur every
few seconds, or even more frequently, in an ssh session that outputs a
lot of text. Commit 14ad6ed30a10 subtly changes how skb head memory is
allocated, making it much more likely that the rndis header will cross
a page boundary when the vCPU count is 16 or more. The changes in
commit 14ad6ed30a10 are perfectly valid -- they just had the side
effect of making the netvsc bug more prominent.
Current code in init_page_array() creates a separate page buffer array
entry for each PFN required to identify the data to be transmitted.
Contiguous PFNs get separate entries in the page buffer array, and any
information about contiguity is lost.
Fix the core issue by having init_page_array() construct the page
buffer array to represent contiguous ranges rather than individual
pages. When these ranges are subsequently passed to
netvsc_build_mpb_array(), it can build GPA ranges that contain
multiple PFNs, as required to avoid the error "nvsp_rndis_pkt_complete
error status: 2". If instead the network packet is sent by copying
into a pre-allocated send buffer slot, the copy proceeds using the
contiguous ranges rather than individual pages, but the result of the
copying is the same. Also fix rndis_filter_send_request() to construct
a contiguous range, since it has its own page buffer array.
This change has a side benefit in CoCo VMs in that netvsc_dma_map()
calls dma_map_single() on each contiguous range instead of on each
page. This results in fewer calls to dma_map_single() but on larger
chunks of memory, which should reduce contention on the swiotlb.
Since the page buffer array now contains one entry for each contiguous
range instead of for each individual page, the number of entries in
the array can be reduced, saving 208 bytes of stack space in
netvsc_xmit() when MAX_SKG_FRAGS has the default value of 17.
netvsc currently uses vmbus_sendpacket_pagebuffer() to send VMBus
messages. This function creates a series of GPA ranges, each of which
contains a single PFN. However, if the rndis header in the VMBus
message crosses a page boundary, the netvsc protocol with the host
requires that both PFNs for the rndis header must be in a single "GPA
range" data structure, which isn't possible with
vmbus_sendpacket_pagebuffer(). As the first step in fixing this, add a
new function netvsc_build_mpb_array() to build a VMBus message with
multiple GPA ranges, each of which may contain multiple PFNs. Use
vmbus_sendpacket_mpb_desc() to send this VMBus message to the host.
There's no functional change since higher levels of netvsc don't
maintain or propagate knowledge of contiguous PFNs. Based on its
input, netvsc_build_mpb_array() still produces a separate GPA range
for each PFN and the behavior is the same as with
vmbus_sendpacket_pagebuffer(). But the groundwork is laid for a
subsequent patch to provide the necessary grouping.
Cc: <stable@vger.kernel.org> # 6.1.x Signed-off-by: Michael Kelley <mhklinux@outlook.com> Link: https://patch.msgid.link/20250513000604.1396-3-mhklinux@outlook.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The differences in the vendor-approved CPU and GPU OPPs for the standard
Rockchip RK3588 variant [1] and the industrial Rockchip RK3588J variant [2]
come from the latter, presumably, supporting an extended temperature range
that's usually associated with industrial applications, despite the two SoC
variant datasheets specifying the same upper limit for the allowed ambient
temperature for both variants. However, the lower temperature limit is
specified much lower for the RK3588J variant. [1][2]
To be on the safe side and to ensure maximum longevity of the RK3588J SoCs,
only the CPU and GPU OPPs that are declared by the vendor to be always safe
for this SoC variant may be provided. As explained by the vendor [3] and
according to the RK3588J datasheet, [2] higher-frequency/higher-voltage
CPU and GPU OPPs can be used as well, but at the risk of reducing the SoC
lifetime expectancy. Presumably, using the higher OPPs may be safe only
when not enjoying the assumed extended temperature range that the RK3588J,
as an SoC variant targeted specifically at higher-temperature, industrial
applications, is made (or binned) for.
Anyone able to keep their RK3588J-based board outside the above-presumed
extended temperature range at all times, and willing to take the associated
risk of possibly reducing the SoC lifetime expectancy, is free to apply
a DT overlay that adds the higher CPU and GPU OPPs.
With all this and the downstream RK3588(J) DT definitions [4][5] in mind,
let's delete the RK3588J CPU and GPU OPPs that are not considered belonging
to the normal operation mode for this SoC variant. To quote the RK3588J
datasheet [2], "normal mode means the chipset works under safety voltage
and frequency; for the industrial environment, highly recommend to keep in
normal mode, the lifetime is reasonably guaranteed", while "overdrive mode
brings higher frequency, and the voltage will increase accordingly; under
the overdrive mode for a long time, the chipset may shorten the lifetime,
especially in high-temperature condition".
To sum the RK3588J datasheet [2] and the vendor-provided DTs up, [4][5]
the maximum allowed CPU core, GPU and NPU frequencies are as follows:
Unfortunately, when it comes to the actual voltages for the RK3588J CPU and
GPU OPPs, there's a discrepancy between the RK3588J datasheet [2] and the
downstream kernel code. [4][5] The RK3588J datasheet states that "the max.
working voltage of CPU/GPU/NPU is 0.75 V under the normal mode", while the
downstream kernel code actually allows voltage ranges that go up to 0.95 V,
which is still within the voltage range allowed by the datasheet. However,
the RK3588J datasheet also tells us to "strictly refer to the software
configuration of SDK and the hardware reference design", so let's embrace
the voltage ranges provided by the downstream kernel code, which also
prevents the undesirable theoretical outcome of ending up with no usable
OPPs on a particular board, as a result of the board's voltage regulator(s)
being unable to deliver the exact voltages, for whatever reason.
The above-described voltage ranges for the RK3588J CPU OPPs remain taken
from the downstream kernel code [4][5] by picking the highest, worst-bin
values, which ensure that all RK3588J bins will work reliably. Yes, with
some power inevitably wasted as unnecessarily generated heat, but the
reliability is paramount, together with the longevity. This deficiency
may be revisited separately at some point in the future.
The provided RK3588J CPU OPPs follow the slightly debatable "provide only
the highest-frequency OPP from the same-voltage group" approach that's been
established earlier, [6] as a result of the "same-voltage, lower-frequency"
OPPs being considered inefficient from the IPA governor's standpoint, which
may also be revisited separately at some point in the future.
The RK3588 thermal sensor driver only receives interrupts when a
higher-temperature threshold is crossed; it cannot notify when the
sensor cools back off. As a result, the driver must poll for temperature
changes to detect when the conditions for a thermal trip are no longer
met. However, it only does so if the DT enables polling.
Before this patch, the RK1 DT did not enable polling, causing the fan to
continue running at the speed corresponding to the highest temperature
reached.
Follow suit with similar RK3588 boards by setting a polling-delay of
1000ms, enabling the driver to detect when the sensor cools back off,
allowing the fan speed to decrease as appropriate.
Fixes: 7c8ec5e6b9d6 ("arm64: dts: rockchip: Enable automatic fan control on Turing RK1") Cc: stable@kernel.org # v6.13+ Signed-off-by: Sam Edwards <CFSworks@gmail.com> Reviewed-by: Dragan Simic <dsimic@manjaro.org> Link: https://lore.kernel.org/r/20250329165017.3885-1-CFSworks@gmail.com Signed-off-by: Heiko Stuebner <heiko@sntech.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
smp_store_mb() inserts memory barrier after storing operation.
It is different with what the comment is originally aiming so Null
pointer dereference can be happened if memory update is reordered.
Signed-off-by: Hyejeong Choi <hjeong.choi@samsung.com> Fixes: a590d0fdbaa5 ("dma-buf: Update reservation shared_count after adding the new fence") CC: stable@vger.kernel.org Reviewed-by: Christian König <christian.koenig@amd.com> Link: https://lore.kernel.org/r/20250513020638.GA2329653@au1-maretx-p37.eng.sarc.samsung.com Signed-off-by: Christian König <christian.koenig@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Microdia JP001 does not support reading the sample rate which leads to
many lines of "cannot get freq at ep 0x84".
This patch adds the USB ID to quirks.c and avoids those error messages.
usb 7-4: New USB device found, idVendor=0c45, idProduct=636b, bcdDevice= 1.00
usb 7-4: New USB device strings: Mfr=2, Product=1, SerialNumber=3
usb 7-4: Product: JP001
usb 7-4: Manufacturer: JP001
usb 7-4: SerialNumber: JP001
usb 7-4: 3:1: cannot get freq at ep 0x84
A user reported on the Arch Linux Forums that their device is emitting
the following message in the kernel journal, which is fixed by adding
the quirk as submitted in this patch:
> kernel: usb 1-2: current rate 8436480 is different from the runtime rate 48000
There also is an entry for this product line added long time ago.
Their specific device has the following ID:
$ lsusb | grep Audio
Bus 001 Device 002: ID 1101:0003 EasyPass Industrial Co., Ltd Audioengine D1
The function snd_es1968_capture_open() calls the function
snd_pcm_hw_constraint_pow2(), but does not check its return
value. A proper implementation can be found in snd_cx25821_pcm_open().
Add error handling for snd_pcm_hw_constraint_pow2() and propagate its
error code.
The original PPTT code had a bug where the processor subtable length
was not correctly validated when encountering a truncated
acpi_pptt_processor node.
Commit 7ab4f0e37a0f4 ("ACPI PPTT: Fix coding mistakes in a couple of
sizeof() calls") attempted to fix this by validating the size is as
large as the acpi_pptt_processor node structure. This introduced a
regression where the last processor node in the PPTT table is ignored
if it doesn't contain any private resources. That results errors like:
ACPI PPTT: PPTT table found, but unable to locate core XX (XX)
ACPI: SPE must be homogeneous
Furthermore, it fails in a common case where the node length isn't
equal to the acpi_pptt_processor structure size, leaving the original
bug in a modified form.
Correct the regression by adjusting the loop termination conditions as
suggested by the bug reporters. An additional check performed after
the subtable node type is detected, validates the acpi_pptt_processor
node is fully contained in the PPTT table. Repeating the check in
acpi_pptt_leaf_node() is largely redundant as the node is already
known to be fully contained in the table.
The case where a final truncated node's parent property is accepted,
but the node itself is rejected should not be considered a bug.
Fixes: 7ab4f0e37a0f4 ("ACPI PPTT: Fix coding mistakes in a couple of sizeof() calls") Reported-by: Maximilian Heyne <mheyne@amazon.de> Closes: https://lore.kernel.org/linux-acpi/20250506-draco-taped-15f475cd@mheyne-amazon/ Reported-by: Yicong Yang <yangyicong@hisilicon.com> Closes: https://lore.kernel.org/linux-acpi/20250507035124.28071-1-yangyicong@huawei.com/ Signed-off-by: Jeremy Linton <jeremy.linton@arm.com> Tested-by: Yicong Yang <yangyicong@hisilicon.com> Reviewed-by: Sudeep Holla <sudeep.holla@arm.com> Tested-by: Maximilian Heyne <mheyne@amazon.de> Cc: All applicable <stable@vger.kernel.org> # 7ab4f0e37a0f4: ACPI PPTT: Fix coding mistakes ... Link: https://patch.msgid.link/20250508023025.1301030-1-jeremy.linton@arm.com Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If an input changes state during wake-up and is used as an interrupt
source, the IRQ handler reads the volatile input register to clear the
interrupt mask and deassert the IRQ line. However, the IRQ handler is
triggered before access to the register is granted, causing the read
operation to fail.
As a result, the IRQ handler enters a loop, repeatedly printing the
"failed reading register" message, until `pca953x_resume()` is eventually
called, which restores the driver context and enables access to
registers.
Fix by disabling the IRQ line before entering suspend mode, and
re-enabling it after the driver context is restored in `pca953x_resume()`.
An IRQ can be disabled with disable_irq() and still wake the system as
long as the IRQ has wake enabled, so the wake-up functionality is
preserved.
It's expected that we'll encounter temporary exceptions
during aux transactions. Adjust logging from drm_info to
drm_dbg_dp to prevent flooding with unnecessary log messages.
Fixes: 3637e457eb00 ("drm/amd/display: Fix wrong handling for AUX_DEFER case") Cc: Mario Limonciello <mario.limonciello@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Wayne Lin <Wayne.Lin@amd.com> Acked-by: Alex Deucher <alexander.deucher@amd.com> Link: https://lore.kernel.org/r/20250513032026.838036-1-Wayne.Lin@amd.com Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 9a9c3e1fe5256da14a0a307dff0478f90c55fc8c) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[Why]
Now forcing aux->transfer to return 0 when incomplete AUX write is
inappropriate. It should return bytes have been transferred.
[How]
aux->transfer is asked not to change original msg except reply field of
drm_dp_aux_msg structure. Copy the msg->buffer when it's write request,
and overwrite the first byte when sink reply 1 byte indicating partially
written byte number. Then we can return the correct value without
changing the original msg.
Fixes: 3637e457eb00 ("drm/amd/display: Fix wrong handling for AUX_DEFER case") Cc: Mario Limonciello <mario.limonciello@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Ray Wu <ray.wu@amd.com> Signed-off-by: Wayne Lin <Wayne.Lin@amd.com> Signed-off-by: Ray Wu <ray.wu@amd.com> Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 7ac37f0dcd2e0b729fa7b5513908dc8ab802b540) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After process exit to unmap csa and free GPU vm, if signal is accepted
and then waiting to take vm lock is interrupted and return, it causes
memory leaking and below warning backtrace.
Change to use uninterruptible wait lock fix the issue.
Signed-off-by: Philip Yang <Philip.Yang@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 7dbbfb3c171a6f63b01165958629c9c26abf38ab) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since commit 559358282e5b ("drm/fb-helper: Don't use the preferred depth
for the BPP default"), RGB565 displays such as the CFAF240320X no longer
render correctly: colors are distorted and the content is shown twice
horizontally.
This regression is due to the fbdev emulation layer defaulting to 32 bits
per pixel, whereas the display expects 16 bpp (RGB565). As a result, the
framebuffer data is incorrectly interpreted by the panel.
Fix the issue by calling drm_client_setup_with_fourcc() with a format
explicitly selected based on the display's bits-per-pixel value. For 16
bpp, use DRM_FORMAT_RGB565; for other values, fall back to the previous
behavior. This ensures that the allocated framebuffer format matches the
hardware expectations, avoiding color and layout corruption.
Tested on a CFAF240320X display with an RGB565 configuration, confirming
correct colors and layout after applying this patch.
Cc: stable@vger.kernel.org Fixes: 559358282e5b ("drm/fb-helper: Don't use the preferred depth for the BPP default") Signed-off-by: Fabio Estevam <festevam@denx.de> Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de> Reviewed-by: Javier Martinez Canillas <javierm@redhat.com> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> Link: https://lore.kernel.org/r/20250417103458.2496790-1-festevam@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Applying degamma curve to the cursor by default breaks Linux userspace
expectation.
On Linux, AMD display manager enables cursor degamma ROM just for
implict sRGB on HW versions where degamma is split into two blocks:
degamma ROM for pre-defined TFs and `gamma correction` for user/custom
curves, and degamma ROM settings doesn't apply to cursor plane.
The Btrfs documentation states that if the commit value is greater than
300 a warning should be issued. The warning was accidentally lost in the
new mount API update.
Fixes: 6941823cc878 ("btrfs: remove old mount API code") CC: stable@vger.kernel.org # 6.12+ Reviewed-by: Qu Wenruo <wqu@suse.com> Reviewed-by: Anand Jain <anand.jain@oracle.com> Signed-off-by: Kyoji Ogasawara <sawara04.o@gmail.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If btrfs_reserve_extent() fails while submitting an async_extent for a
compressed write, then we fail to call free_async_extent_pages() on the
async_extent and leak its folios. A likely cause for such a failure
would be btrfs_reserve_extent() failing to find a large enough
contiguous free extent for the compressed extent.
I was able to reproduce this by:
1. mount with compress-force=zstd:3
2. fallocating most of a filesystem to a big file
3. fragmenting the remaining free space
4. trying to copy in a file which zstd would generate large compressed
extents for (vmlinux worked well for this)
Step 4. hits the memory leak and can be repeated ad nauseam to
eventually exhaust the system memory.
Fix this by detecting the case where we fallback to uncompressed
submission for a compressed async_extent and ensuring that we call
free_async_extent_pages().
Fixes: 131a821a243f ("btrfs: fallback if compressed IO fails for ENOSPC") CC: stable@vger.kernel.org # 6.1+ Reviewed-by: Filipe Manana <fdmanana@suse.com> Co-developed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Boris Burkov <boris@bur.io> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the discard worker is running and there's currently only one block
group, that block group is a data block group, it's in the unused block
groups discard list and is being used (it got an extent allocated from it
after becoming unused), the worker can end up in an infinite loop if a
transaction abort happens or the async discard is disabled (during remount
or unmount for example).
This happens like this:
1) Task A, the discard worker, is at peek_discard_list() and
find_next_block_group() returns block group X;
2) Block group X is in the unused block groups discard list (its discard
index is BTRFS_DISCARD_INDEX_UNUSED) since at some point in the past
it become an unused block group and was added to that list, but then
later it got an extent allocated from it, so its ->used counter is not
zero anymore;
3) The current transaction is aborted by task B and we end up at
__btrfs_handle_fs_error() in the transaction abort path, where we call
btrfs_discard_stop(), which clears BTRFS_FS_DISCARD_RUNNING from
fs_info, and then at __btrfs_handle_fs_error() we set the fs to RO mode
(setting SB_RDONLY in the super block's s_flags field);
4) Task A calls __add_to_discard_list() with the goal of moving the block
group from the unused block groups discard list into another discard
list, but at __add_to_discard_list() we end up doing nothing because
btrfs_run_discard_work() returns false, since the super block has
SB_RDONLY set in its flags and BTRFS_FS_DISCARD_RUNNING is not set
anymore in fs_info->flags. So block group X remains in the unused block
groups discard list;
5) Task A then does a goto into the 'again' label, calls
find_next_block_group() again we gets block group X again. Then it
repeats the previous steps over and over since there are not other
block groups in the discard lists and block group X is never moved
out of the unused block groups discard list since
btrfs_run_discard_work() keeps returning false and therefore
__add_to_discard_list() doesn't move block group X out of that discard
list.
When this happens we can get a soft lockup report like this:
So fix this by making sure we move a block group out of the unused block
groups discard list when calling __add_to_discard_list().
Fixes: 2bee7eb8bb81 ("btrfs: discard one region at a time in async discard") Link: https://bugzilla.suse.com/show_bug.cgi?id=1242012 CC: stable@vger.kernel.org # 5.10+ Reviewed-by: Boris Burkov <boris@bur.io> Reviewed-by: Daniel Vacek <neelx@suse.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch_uprobe_skip_sstep() returns true if instruction was emulated, that
is to say, there is no need to single step for the emulated instructions.
regs->csr_era will point to the destination address directly after the
exception, so the resume_era related code is redundant, just remove them.
When executing the "perf probe" and "perf stat" test cases about some
cryptographic algorithm, the output shows that "Trace/breakpoint trap".
This is because it uses the software singlestep breakpoint for uprobes
on LoongArch, and no need to use the hardware singlestep. So just remove
the related function call to user_{en,dis}able_single_step() for uprobes
on LoongArch.
How to reproduce:
Please make sure CONFIG_UPROBE_EVENTS is set and openssl supports sm2
algorithm, then execute the following command.
cd tools/perf && make
./perf probe -x /usr/lib64/libcrypto.so BN_mod_mul_montgomery
./perf stat -e probe_libcrypto:BN_mod_mul_montgomery openssl speed sm2
Fix MAX_REG_OFFSET calculation, make it point to the last register
in 'struct pt_regs' and not to the marker itself, which could allow
regs_get_register() to return an invalid offset.
Save and restore CSR.CNTC for hibernation which is similar to suspend.
For host this is unnecessary because sched clock is ensured continuous,
but for kvm guest sched clock isn't enough because rdtime.d should also
be continuous.
Now arch_cpu_idle() is annotated with __cpuidle which means it is in
the .cpuidle.text section, but __arch_cpu_idle() isn't. Thus, fix the
missing .cpuidle.text section assignment for __arch_cpu_idle() in order
to correct backtracing with nmi_backtrace().
The principle is similar to the commit 97c8580e85cf81c ("MIPS: Annotate
cpu_wait implementations with __cpuidle")
When CONFIG_PREEMPT_COUNT is not configured (i.e. CONFIG_PREEMPT_NONE/
CONFIG_PREEMPT_VOLUNTARY), preempt_disable() / preempt_enable() merely
acts as a barrier(). However, in these cases cond_resched() can still
trigger a context switch and modify the CSR.EUEN, resulting in do_fpu()
exception being activated within the kernel-fpu critical sections, as
demonstrated in the following path:
This patch is similar to commit d02198550423a0b (x86/fpu: Improve crypto
performance by making kernel-mode FPU reliably usable in softirqs). It
uses local_bh_disable() instead of preempt_disable() for non-RT kernels
so it can avoid the cond_resched() issue, and also extend the kernel-fpu
application scenarios to the softirq context.
On systems that only have an SRA sensor connected to SFH the sensor
doesn't get enabled due to a bad optimization condition of breaking
the sensor walk loop.
This optimization is unnecessary in the first place because if there
is only one device then the loop only runs once. Drop the condition
and explicitly mark sensor as enabled.
The current HID bpf implementation assumes no output report/request will
go through it after hid_bpf_destroy_device() has been called. This leads
to a bug that unplugging certain types of HID devices causes a cleaned-
up SRCU to be accessed. The bug was previously a hidden failure until a
recent x86 percpu change [1] made it access not-present pages.
The bug will be triggered if the conditions below are met:
A) a device under the driver has some LEDs on
B) hid_ll_driver->request() is uninplemented (e.g., logitech-djreceiver)
If condition A is met, hidinput_led_worker() is always scheduled *after*
hid_bpf_destroy_device().
This is fine when condition B is not met, where hidinput_led_worker()
calls hid_ll_driver->request(). This is the case for most HID drivers,
which implement it or use the generic one from usbhid. The driver itself
or an underlying driver will then abort processing the request.
Otherwise, hidinput_led_worker() tries hid_hw_output_report() and leads
to the bug.
The bug has existed since the introduction [2] of
dispatch_hid_bpf_output_report(). However, the same bug also exists in
dispatch_hid_bpf_raw_requests(), and I've reproduced (no visible effect
because of the lack of [1], but confirmed bpf.destroyed == 1) the bug
against the commit (i.e., the Fixes:) introducing the function. This is
because hidinput_led_worker() falls back to hid_hw_raw_request() when
hid_ll_driver->output_report() is uninplemented (e.g., logitech-
djreceiver).
Fix the issue by returning early in the two mentioned functions if
hid_bpf has been marked as destroyed. Though
dispatch_hid_bpf_device_event() handles input events, and there is no
evidence that it may be called after the destruction, the same check, as
a safety net, is also added to it to maintain the consistency among all
dispatch functions.
The impact of the bug on other architectures is unclear. Even if it acts
as a hidden failure, this is still dangerous because it corrupts
whatever is on the address calculated by SRCU. Thus, CC'ing the stable
list.
[1]: commit 9d7de2aa8b41 ("x86/percpu/64: Use relative percpu offsets")
[2]: commit 9286675a2aed ("HID: bpf: add HID-BPF hooks for
hid_hw_output_report")
Closes: https://lore.kernel.org/all/20250506145548.GGaBoi9Jzp3aeJizTR@fat_crate.local/ Fixes: 8bd0488b5ea5 ("HID: bpf: add HID-BPF hooks for hid_hw_raw_requests") Cc: stable@vger.kernel.org Signed-off-by: Rong Zhang <i@rong.moe> Tested-by: Petr Tesarik <petr@tesarici.cz> Link: https://patch.msgid.link/20250512152420.87441-1-i@rong.moe Signed-off-by: Benjamin Tissoires <bentiss@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After commit 0a65bc27bd64 ("eventpoll: Set epoll timeout if it's in
the future"), the following program would immediately enter a busy
loop in the kernel:
This happens because the given (non-zero) timeout of 1 nanosecond
usually expires before ep_poll() is entered and then
ep_schedule_timeout() returns false, but `timed_out` is never set
because the code line that sets it is skipped. This quickly turns
into a soft lockup, RCU stalls and deadlocks, inflicting severe
headaches to the whole system.
When the timeout has expired, we don't need to schedule a hrtimer, but
we should set the `timed_out` variable. Therefore, I suggest moving
the ep_schedule_timeout() check into the `timed_out` expression
instead of skipping it.
brauner: Note that there was an earlier fix by Joe Damato in response to
my bug report in [1].
UDF maintains total length of all extents in i_lenExtents. Generally we
keep extent lengths (and thus i_lenExtents) block aligned because it
makes the file appending logic simpler. However the standard mandates
that the inode size must match the length of all extents and thus we
trim the last extent when closing the file. To catch possible bugs we
also verify that i_lenExtents matches i_size when evicting inode from
memory. Commit b405c1e58b73 ("udf: refactor udf_next_aext() to handle
error") however broke the code updating i_lenExtents and thus
udf_evict_inode() ended up spewing lots of errors about incorrectly
sized extents although the extents were actually sized properly. Fix the
updating of i_lenExtents to silence the errors.
Fixes: b405c1e58b73 ("udf: refactor udf_next_aext() to handle error") CC: stable@vger.kernel.org Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
projects / thirdparty / kernel / stable.git / log
