In i2c_imx_runtime_suspend(), the clock is disabled before switching
the pinctrl state to sleep. If pinctrl_pm_select_sleep_state() fails,
the runtime suspend is aborted but the clock remains disabled, causing
a system crash when the hardware is subsequently accessed.
Fix this by switching the pinctrl state before disabling the clock so
that a pinctrl failure leaves the clock enabled and the hardware
accessible.
In i2c_imx_runtime_resume(), restore the pinctrl state back to sleep
if clk_enable() fails to keep the consistent.
Fixes: 576eba03c9 ("i2c: imx: switch different pinctrl state in different system power status")
Signed-off-by: Carlos Song <carlos.song@nxp.com>
Cc: <stable@vger.kernel.org> # v6.14+
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Signed-off-by: Andi Shyti <andi.shyti@kernel.org>
Link: https://lore.kernel.org/r/20260521065038.2954998-1-carlos.song@oss.nxp.com
Remove the mlx5_mr_pdn() in pagefault_real_mr() by pushing the pdn up, all
the callers use 0 since they don't pass MLX5_PF_FLAGS_ENABLE except the
ioctl reg_mr path which can use the ioctl pd.
Link: https://patch.msgid.link/r/9-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Assisted-by: Codex:gpt-5-5
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Keep pushing the pdn higher to remove more places touching mr->pd:
- XLT combinations that don't use PDN can just pass 0
- Use local pd values instead of mr->pd
- Implicit MR does not have inplace rereg, so the mr->pd is safe
Link: https://patch.msgid.link/r/8-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Assisted-by: Codex:gpt-5-5
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The rereg protocol expects the core code to change mr->pd and synchronize
that change with the atomics and syncs. The driver should not touch it.
mlx5 needed to update it in umr_rereg_pas() because
mlx5r_umr_update_mr_pas() required the updated mr->pd to build the
UMR.
Simply switch mlx5r_umr_update_mr_pas() to use the pdn directly from
the new pd and remove the mr->pd update.
Fixes: 56e11d628c ("IB/mlx5: Added support for re-registration of MRs")
Link: https://patch.msgid.link/r/7-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Assisted-by: Codex:gpt-5-5
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Instead of getting the pdn deep inside the umr code, pass it in from the
top. to_mpd(mr->ibmr.pd)->pdn is not safe due to the rereg races, so all
the call sites need some revision to obtain the pdn in a safe way.
Mark them with mlx5_mr_pdn(); following patches will go through and remove
these.
Cases where the XLT flags are known and do not require the PDN can pass 0,
such as for mlx5_ib_dmabuf_invalidate_cb().
Also extract the DMABUF data_direct special case from inside the UMR code
and into the only place that needs it, pagefault_dmabuf_mr(). The actual
mr was created directly without using the UMR flow. Ultimately this will
be moved into mlx5_ib_init_dmabuf_mr().
Link: https://patch.msgid.link/r/6-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Assisted-by: Codex:gpt-5-5
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The HW only processes mkc fields selected by mkey_mask.
pd, qpn and mkey_7_0 are never selected so they can be left as zero.
This removes a racy read of mr->pd.
Fixes: e73242aa14 ("RDMA/mlx5: Optimize DMABUF mkey page size")
Link: https://patch.msgid.link/r/5-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Sashiko points out that, due to rereg_mr, the PD is actually variable and
all the touches in nldev are racy.
Use mr->device instead of mr->pd->device.
Getting the PD restrack ID is more tricky. To avoid disturbing all the
happy paths, add an rdma_restrack_sync() operation which is sort of like
flush_workqueue() or synchronize_irq(): after it returns, all the old
nldev touches to the mr are gone and everything sees the new PD. This
makes it safe to reach into the PD pointer.
Fixes: da5c850782 ("RDMA/nldev: add driver-specific resource tracking")
Link: https://patch.msgid.link/r/4-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Due to all the child mkeys in the implicit ODP configuration we cannot
change anything in place for the parent mkey. Instead the whole thing
needs to be rebuilt if any change is requested. If the user does not
specify a translation then force the implicit values which will then fall
through the logic into mlx5_ib_reg_user_mr() to allocate a completely new
MR.
Since implicit children were also touching the mr->pd, this removes
another case where the access was racy.
Fixes: ef3642c4f5 ("RDMA/mlx5: Fix error unwinds for rereg_mr")
Link: https://sashiko.dev/#/patchset/20260427-security-bug-fixes-v3-0-4621fa52de0e%40nvidia.com?part=4
Link: https://patch.msgid.link/r/3-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
DMABUF generally relies on the ODP EQ mechanism to safely implement the
move semantics. ODP requires a device-global one time startup of the ODP
machinery when the first MR is created, and this was missed on the DMABUF
path.
Call mlx5r_odp_create_eq() when creating a ODP'able DMABUF.
The core code prevents using IB_ACCESS_ON_DEMAND unless the driver
advertises IB_ODP_SUPPORT, so until now, mlx5r_odp_create_eq() cannot be
called unless the device has ODP support.
However, DMABUF has no such protection and a second bug was allowing
DMABUFs to be created on non-ODP capable HW. Add a guard at the start of
mlx5r_odp_create_eq(). This is necessary here anyhow as the
dev->odp_eq_mutex is not initialized without IB_ODP_SUPPORT.
Link: https://patch.msgid.link/r/2-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Jumping to mlx5_ib_reg_user_mr() without IB_MR_REREG_TRANS set will use
garbage values for start, length, and iova. Recovering the original mr
parameters for ODP and DMABUF to properly recreate it is too hard in this
flow, so just fail it.
Fixes: ef3642c4f5 ("RDMA/mlx5: Fix error unwinds for rereg_mr")
Link: https://patch.msgid.link/r/1-v1-29ebd2c229b5+fd5-ib_mr_pd_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
- use named initializers for arrays of i2c_device_data
-----BEGIN PGP SIGNATURE-----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=fQo7
-----END PGP SIGNATURE-----
gpgsig -----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQScDfrjQa34uOld1VLaeAVmJtMtbgUCaib4kwAKCRDaeAVmJtMt
bjHZAP9jzzE3IVw4QNgRvqgePWvIhzbwaI5tBBorVbugshvA6QD+OuJaPyEYVkJD
Z5gtPs3rrYsTQqNkq6k89XjG6B10Jw4=
=TTSZ
-----END PGP SIGNATURE-----
Merge tag 'at24-updates-for-v7.2-rc1' into i2c/i2c-host
at24 updates for v7.2-rc1
- use named initializers for arrays of i2c_device_data
Commit:
809b997a5c ("x86-64/arm64/powerpc: clean up and rename __copy_from_user_flushcache")
sanitized that any provided copy helper should separately validate
destination and source addresses, but we should also ensure that
anything that is retrieved from a buffer is validated upfront. For ring
provided buffers, always include an access_ok() when grabbing a new
buffer.
Fixes: c7fb19428d ("io_uring: add support for ring mapped supplied buffers")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
s390 selects GENERIC_LOCKBREAK if PREEMPT is enabled. Reason is a historic
18 years old commit [1] which fixed a compile error for PREEMPT enabled
kernels. Back than only PREEMPT_NONE and PREEMPT_VOLUNTARY kernels were
considered to be important for s390. PREEMPT should "just work".
However, since recently PREEMPT is always enabled [2], which also causes
GENERIC_LOCKBREAK to be always enabled. For some workloads this leads to
massive performance degradation; e.g. a simple kernel compile on machines
with many CPUs may take up to four times longer.
To fix this just remove the GENERIC_LOCKBREAK from s390's Kconfig, since
the compile error from 18 years ago does not exist anymore.
[1] commit b6b40c532a ("[S390] Define GENERIC_LOCKBREAK.")
[2] commit 7dadeaa6e8 ("sched: Further restrict the preemption modes")
Cc: stable@vger.kernel.org
Reported-by: Massimiliano Pellizzer <massimiliano.pellizzer@canonical.com>
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Add kernel-doc documentation describing the Intel DP link training
recovery state machine and the sequence of automatic retraining,
fallback selection, and userspace notification.
v2:
- Rebase on dedicated intel-display documentation change.
- Remove unnecessary indent in section bodies. (Jani)
- Add recovery flowcharts. (Jani)
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-23-imre.deak@intel.com
intel_dp_check_link_state() and intel_dp_link_params_valid() perform
only a coarse validation of the link configuration used by the active
mode against the available configurations (as constrained by the link
training fallback code after a previous LT failure). Even if these
coarse checks find a seemingly usable configuration, the modeset check,
which performs full verification, may still fail.
Disallow further autoretrain attempts if an autoretrain modeset fails.
Further attempts would just reuse the same modeset parameters and fail
in the same way. Autoretrain will be reallowed unconditionally when the
sink reports a change in its capabilities. This allows an autoretrain to
proceed once both the link validation and modeset checks confirm a
usable configuration.
Also clarify in intel_dp_check_link_state() and
intel_dp_link_params_valid() that these checks are coarse and that a
full validation is only performed by the subsequent atomic modeset
check.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-21-imre.deak@intel.com
Allocate a local atomic state for the autoretrain modeset. This
prepares for a follow-up change that needs to access the state after
the modeset for sending userspace notifications.
v2: Rebase on upstream drm_atomic_state -> drm_atomic_commit rename.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-20-imre.deak@intel.com
Simplify accessing the DRM atomic state via the intel atomic state
in intel_modeset_commit_pipes_for_atomic_state(), which also allows
dropping the cached DRM state pointer. Also streamline the success/error
return flows.
v2: Rebase on upstream drm_atomic_state -> drm_atomic_commit rename.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-19-imre.deak@intel.com
Factor out a helper modesetting a pipe that accepts an existing atomic
state. This prepares for a follow-up change that needs to allocate its
own atomic state.
v2: Rebase on upstream drm_atomic_state -> drm_atomic_commit rename.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-18-imre.deak@intel.com
Replace the misnamed retrain_disabled flag with a dedicated link
recovery state indicating that no fallback link configurations remain.
This clarifies the meaning of the state: it has always represented the
situation where no further fallback link configurations are available.
While at it, add a TODO comment to the debugfs entry, to expose this
state via a more appropriately named entry.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-17-imre.deak@intel.com
Replace the sequential link training failure counter with an explicit
link recovery state enum.
This makes the recovery states and transitions clearer: idle, automatic
retraining pending, and automatic retraining disabled.
A follow-up change will also move the retrain_disabled flag into this
enum.
v2: (Jani)
- Convert enum intel_dp_link_recovery_state's documentation to be a
non kernel-doc comment.
- Compare against / set enum values explicitly.
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-16-imre.deak@intel.com
Add link_recovery_reset() to make it explicit when link recovery is no
longer needed and the recovery state can be cleared.
This also prepares for replacing the sequential link training failure
counter with an enum in a follow-up change.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-15-imre.deak@intel.com
Add link_recovery_mark_train_failure() to record the failure and make
the link recovery state transition explicit after a link training
failure: recovery can continue with an autoretrain, or must be handed
over to userspace after fallback selection.
This also prepares for replacing the sequential link training failure
counter with an enum in a follow-up change.
v2: (Jani)
- Convert link_recovery_mark_train_failure()'s documentation to be a
non kernel-doc comment.
- Rename can_autoretrain flag to autoretrain_allowed.
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-14-imre.deak@intel.com
Add link_recovery_autoretrain_allowed() to make it clearer what the
condition is about at its callers: queuing work for and starting an
autoretrain is allowed.
This also prepares for replacing the sequential link training failure
counter with an enum in a follow-up change.
v2: Convert link_recovery_autoretrain_allowed()'s documentation to be a
non kernel-doc comment and detail what an allowed autoretrain is.
(Jani)
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-13-imre.deak@intel.com
Add link_recovery_autoretrain_pending() to make it clearer what the
condition is about at its callers: an autoretrain work has been queued.
This also prepares for replacing the sequential link training failure
counter with an enum in a follow-up change.
v2: Remove unnecessary function documentation. (Jani)
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-12-imre.deak@intel.com
Check explicitly for a pending autoretrain by matching
seq_train_failures == 1.
This makes the actual condition clear, since at the point where the
counter is checked it is also below MAX_SEQ_TRAIN_FAILURES.
This also prepares for replacing the counter with an enum in a follow-up
change.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-11-imre.deak@intel.com
Clamp link_training->seq_train_failures to MAX_SEQ_TRAIN_FAILURES to
avoid - an unlikely - overflow. This is ok, because the code only makes
a distinction between the cases where the counter is below or at the
limit.
This also prepares for replacing the counter with an enum in a follow-up
change.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-10-imre.deak@intel.com
After a regular modeset link training failure, the driver attempts to
recover the link via an autoretrain using the same link parameters as
the modeset.
If the autoretrain fails as well, the set of allowed link configurations
is reduced via a fallback mechanism. For further link training, the
modeset parameters will likely need to change. This lowers the required
link bandwidth and allows selecting a link configuration from the
fallback-reduced set. Only userspace can perform such a modeset change.
Therefore, the driver notifies userspace to take over link recovery.
Userspace is expected to continue with the recovery attempt via a
modeset with updated parameters. The driver must not interfere with
these modesets.
link_training->seq_train_failures is set to MAX_SEQ_TRAIN_FAILURES after
the autoretrain fails. If a fallback selection also fails after this,
as no link configurations remain, retrain_disabled is set as well.
retrain_disabled is therefore somewhat misnamed: it indicates that no
fallback is available, not that autoretraining is disabled. This will be
addressed in a follow-up change by renaming the flag. For now, prevent
further autoretrain attempts based on the correct condition:
seq_train_failures == MAX_SEQ_TRAIN_FAILURES.
This also prepares for replacing the counter with an enum in a follow-up
change.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-9-imre.deak@intel.com
Move all state related to link recovery and link training debugging from
struct intel_dp to struct intel_dp_link_training.
This moves towards grouping all link training and recovery state and
logic in a single place and prepares for follow-up changes in the link
recovery state handling.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-8-imre.deak@intel.com
Add helpers to get and set the force retrain state in preparation for
moving the state from the DP struct to the link training state.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-7-imre.deak@intel.com
Retrieve the link_training pointer from the connector and derive the DP
pointer from it in debugfs entries.
This prepares for a follow-up change where values exposed via debugfs
entries will be retrieved from the link training state.
v2: Join unnecessarily wrapped lines. (Jani)
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-6-imre.deak@intel.com
Flush pending connector commits before accessing the link training state
from debugfs. Access to connector state - like the link training state -
that may be updated from an asynchronous commit tail must hold the
connection mutex and wait for the tail to complete. The commit tail
cannot hold the connection mutex, so all other accessors must wait for
it explicitly.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-4-imre.deak@intel.com
Factor out the link training state reset into a helper in
intel_dp_link_training.c to prepare for isolating the link training state
from the generic DP code.
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-3-imre.deak@intel.com
Start isolating the link training state from the generic DP code by
introducing a separate intel_dp_link_training state struct.
Allocate the state so it can remain opaque within its own module.
Follow-up changes will move link training fields from the DP struct to
the link training state.
v2: Remove unnecessary function documentation. (Jani)
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260601093836.3057345-2-imre.deak@intel.com
When a CPU is taken offline and then is brought back online, unaligned
access speed probe always runs even though the unaligned access speed is
already known, wasting CPU cycles.
This is because when a CPU becomes online, the following happen:
1. check_unaligned_access_emulated() is called, which clears
misaligned_access_speed if there is no emulation.
2. check_unaligned_access() is called because misaligned_access_speed is
cleared, wasting CPU cycles determining something already previous
known.
Avoid the redundant access speed probe by stop clearing
misaligned_access_speed in (1). If access speed is already known, just
reuse it.
On my Visionfive 2, this reduces CPU bring-up time from 26ms to 0.8ms.
Signed-off-by: Nam Cao <namcao@linutronix.de>
Link: https://patch.msgid.link/aa5755142537d462a9e3d2074d82ad4eef6774ba.1780002199.git.namcao@linutronix.de
Signed-off-by: Paul Walmsley <pjw@kernel.org>
When booting with unaligned_scalar_speed=fast,
fast_unaligned_access_speed_key is initialized incorrectly.
The key is currently derived from the fast_misaligned_access cpumask, but
that mask is only populated when the unaligned access speed probe runs.
Specifying unaligned_scalar_speed=fast skips the probe entirely, leaving
the mask uninitialized.
The information tracked by fast_misaligned_access is already available in
the misaligned_access_speed per-CPU variable. Use that to initialize
fast_unaligned_access_speed_key instead and remove the redundant cpumask.
Signed-off-by: Nam Cao <namcao@linutronix.de>
Link: https://patch.msgid.link/2468816ceb433394099a00d7822f819745276b49.1780002199.git.namcao@linutronix.de
Signed-off-by: Paul Walmsley <pjw@kernel.org>
srp_process_rsp() copies sense data from rsp->data + resp_data_len,
where resp_data_len is the full 32-bit value supplied by the SRP target
and is never checked against the number of bytes actually received
(wc->byte_len). The copy length is bounded to SCSI_SENSE_BUFFERSIZE, so
at most 96 bytes are copied, but the source offset is not bounded.
A malicious or compromised SRP target on the InfiniBand/RoCE fabric that
the initiator has logged into can return an SRP_RSP with
SRP_RSP_FLAG_SNSVALID set and a large resp_data_len. The receive buffer
is allocated at the target-chosen max_ti_iu_len, so the source of the
sense copy lands past the bytes actually received; with resp_data_len
near 0xFFFFFFFF it is gigabytes past the buffer and the read faults.
Copy the sense data only if it has not been truncated, that is, only if
the response header, the response data, and the sense region fit within
the bytes actually received; otherwise drop the sense and log. The
in-tree iSER and NVMe-RDMA receive paths already bound their parse by
wc->byte_len; this brings ib_srp into line with them.
Fixes: aef9ec39c4 ("IB: Add SCSI RDMA Protocol (SRP) initiator")
Link: https://patch.msgid.link/r/20260602220457.2542840-1-michael.bommarito@gmail.com
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In drivers/infiniband/ulp/isert/ib_isert.c, isert_login_recv_done()
computes the login request payload length as wc->byte_len minus
ISER_HEADERS_LEN with no lower bound, and login_req_len is a signed int.
A remote iSER initiator can post a login Send work request carrying
fewer than ISER_HEADERS_LEN (76) bytes, so the subtraction underflows
and login_req_len becomes negative.
isert_rx_login_req() then reads that negative length back into a signed
int, takes size = min(rx_buflen, MAX_KEY_VALUE_PAIRS), and because the
min() is signed it keeps the negative value; the value is then passed as
the memcpy() length and sign-extended to a multi-gigabyte size_t. The
copy into the 8192-byte login->req_buf runs far out of bounds and
faults, crashing the target node. The login phase precedes iSCSI
authentication, so no credentials are required to reach this path.
Reject any login PDU shorter than ISER_HEADERS_LEN before the
subtraction, mirroring the existing early return on a failed work
completion, so login_req_len can never go negative. The upper bound was
already safe: a posted login buffer cannot deliver more than
ISER_RX_PAYLOAD_SIZE, so the difference stays at or below
MAX_KEY_VALUE_PAIRS and the existing min() clamps it; only the missing
lower bound needs to be added.
Fixes: b8d26b3be8 ("iser-target: Add iSCSI Extensions for RDMA (iSER) target driver")
Link: https://patch.msgid.link/r/20260602194642.2273217-1-michael.bommarito@gmail.com
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Modify cpufreq_policy_init_qos() introduced previously to use
policy->min/max set in the driver .init() callback as the initial
values for the policy min/max frequency QoS requests, respectively,
so long as they are different from 0 (which means that they have
been updated by the driver). Update the documentation in accordance
with that code change.
This only affects the following drivers:
- gx-suspmod (min)
- cppc-cpufreq (min)
- longrun (min/max)
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
[ rjw: Changelog rewrite ]
Link: https://patch.msgid.link/20260528090913.2759118-5-pierre.gondois@arm.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Prior to commit 521223d8b3 ("cpufreq: Fix initialization of min and
max frequency QoS requests"), drivers were setting policy->min/max and
these values were used as initial policy QoS constraints.
After the above commit, these values are only used temporarily, as
cpufreq_set_policy() ultimately overrides them through:
cpufreq_policy_online()
\-cpufreq_init_policy()
\-cpufreq_set_policy()
\-/* Set policy->min/max */
A subsequent change will restore the previous behavior allowing
drivers to request special min/max QoS frequencies instead of
FREQ_QOS_MIN_DEFAULT_VALUE and FREQ_QOS_MAX_DEFAULT_VALUE, respectively,
if desired. For instance, the CPPC driver wants to advertise the lowest
non-linear frequency that should be used as the initial minimum
frequency QoS request.
However, for this purpose, all drivers setting policy->min/max to
policy->cpuinfo.min/max_freq, respectively, need to be updated so
their initial policy->min/max settings don't limit the frequency
scaling unnecessarily going forward (which would defeat the purpose
of commit 521223d8b3), so do that.
This does not actually alter the observed behavior of all of
the drivers in question because setting policy->min/max to
policy->cpuinfo.min/max_freq, respectively, is not necessary or
even useful any more after a previous change ("cpufreq: Set default
policy->min/max values for all drivers").
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Acked-by: Jie Zhan <zhanjie9@hisilicon.com>
[ rjw: Changelog rewrite ]
Link: https://patch.msgid.link/20260528090913.2759118-4-pierre.gondois@arm.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some drivers set policy->min/max in their .init() callback, but
cpufreq_set_policy() will ultimately override them through:
cpufreq_policy_online()
\-cpufreq_init_policy()
\-cpufreq_set_policy()
\-/* Set policy->min/max */
Thus the policy min/max values set by the drivers are only temporary.
There is an exception if CPUFREQ_NEED_INITIAL_FREQ_CHECK is set and
cpufreq_policy_online() calls __cpufreq_driver_target() which invokes
cpufreq_driver->target().
To prepare for a subsequent change that will remove all initialization
of policy->min/max in driver .init() callbacks if the min/max value is
equal to the corresponding cpuinfo.min/max_freq, set default
policy->min/max values in the core for all drivers.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Jie Zhan <zhanjie9@hisilicon.com>
[ rjw: Edits of the new comment and changelog ]
Link: https://patch.msgid.link/20260528090913.2759118-3-pierre.gondois@arm.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Extract the QoS-related logic from cpufreq_policy_online()
to make that function shorter/simpler.
The logic is placed in cpufreq_policy_init_qos() and is
now executed right after the following calls:
- cpufreq_driver->init()
- cpufreq_table_validate_and_sort()
This facilitats subsequent changes that will, in
cpufreq_policy_init_qos():
- Set a default policy->min/max value for all policies.
- Use the policy->min/max values set by drivers as initial request
values for policy frequency QoS requests.
No functional change.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>
Reviewed-by: Jie Zhan <zhanjie9@hisilicon.com>
[ rjw: Changelog edits ]
Link: https://patch.msgid.link/20260528090913.2759118-2-pierre.gondois@arm.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If IB_MR_REREG_ACCESS changes from RO to RW then the umem has to be
re-evaluated to ensure it is properly pinned as RW. Since the umem is
hidden inside each driver's mr struct add a ib_umem_check_rereg() function
that each driver has to call before processing IB_MR_REREG_ACCESS.
mlx4 has to retain its duplicate ib_access_writable check because it
implements IB_MR_REREG_ACCESS | IB_MR_REREG_TRANS by changing both items
in place sequentially while the MR is live, so it will continue to not
support this combination.
Cc: stable@vger.kernel.org
Fixes: b40656aa7d ("RDMA/umem: remove FOLL_FORCE usage")
Link: https://patch.msgid.link/r/0-v1-06fb1a2d6cf5+107-rereg_access_jgg@nvidia.com
Reported-by: Philip Tsukerman <philiptsukerman@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
KVM has reflected KVM_X86_SNP_VM to userspace since 1dfe571c12
("KVM: SEV: Add initial SEV-SNP support"), and KVM_X86_TDX_VM since
161d34609f ("KVM: TDX: Make TDX VM type supported"). Update the
documentation to reflect this fact.
Fixes: 1dfe571c12 ("KVM: SEV: Add initial SEV-SNP support")
Fixes: 161d34609f ("KVM: TDX: Make TDX VM type supported")
Signed-off-by: Carlos López <clopez@suse.de>
Reviewed-by: Binbin Wu <binbin.wu@linux.intel.com>
Link: https://patch.msgid.link/20260603114504.814647-2-clopez@suse.de
[sean: use one tab instead of two]
Signed-off-by: Sean Christopherson <seanjc@google.com>
Add a regression test where KVM would inadvertently ignore PMU event
filters on writes that change _some_ bits in FIXED_CTR_CTRL, but not the
enable bits for PMCs that are denied to the guest.
Link: https://patch.msgid.link/20260603231905.1738487-3-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
When (conditionally) reprogramming fixed counters, use the hardware value
of FIXED_CTR_CTRL to detect changes, not the guest's original value. For
guests with a mediated PMU, overwriting fixed_ctr_ctrl_hw at the start of
reprogramming without actually reacting to changes in fixed_ctr_ctrl_hw can
lead to KVM ignoring PMU event filters.
E.g. if the guest attempts to enable a fixed PMC that is disallowed, and
then toggles a different PMC in a subsequent WRMSR, KVM will update
pmu->fixed_ctr_ctrl_hw and reprogram the PMC that is changing, but not the
others that are now effectively enabled in pmu->fixed_ctr_ctrl_hw.
Note, the perf-based PMU is unaffected, as it doesn't use fixed_ctr_ctrl_hw
(which is also why keying off fixed_ctr_ctrl_hw works for both PMUs.
Note #2, fixed_ctr_ctrl_hw won't mess up pmc_in_use either, because the
latter isn't used by the mediated PMU. Its purpose is solely to release
perf events that are no longer being actively used, and the meadiated PMU
obviously doesn't create perf events.
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/all/20260528005419.0228F1F00A3A@smtp.kernel.org
Link: https://patch.msgid.link/20260603231905.1738487-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
