| Age | Commit message (Collapse) | Author |
|---|
|
nouveau_fence_context_del() drains the uevent work while the event that
feeds it is still armed:
cancel_work_sync(&fctx->uevent_work);
nouveau_fence_context_kill(fctx, 0);
nvif_event_dtor(&fctx->event);
nouveau_fence_wait_uevent_handler() queues the work unconditionally:
schedule_work(&fctx->uevent_work);
return NVIF_EVENT_KEEP;
so a non-stall interrupt arriving after cancel_work_sync() has returned
re-arms the work that was just drained. The window closes in two steps,
neither of which is the drain. The kill blocks the event when the last
fence holding a notify_ref is signalled, which reaches
atomic_xchg(&ntfy->allowed, 0) (nvkm/core/event.c:104), and
nvkm_event_ntfy() skips a ntfy that is not allowed (:183). That stops
further handlers from starting, but not one that is already inside
nvkm_event_ntfy(): the event is created with wait = false
(nouveau_fence.c:201), so nvkm_event_ntfy_block_() leaves it on the list
and never takes event->list_lock. Only nvif_event_dtor() waits that one
out: nvkm_event_ntfy_del() (:141) goes through nvkm_event_ntfy_remove(),
which takes write_lock_irq() on that same list_lock (:84).
Either way the re-arm happens after the drain, and the caller drops its
reference immediately afterwards, for example nv84_fence_context_del():
nouveau_fence_context_del(&fctx->base);
chan->fence = NULL;
nouveau_fence_context_free(&fctx->base);
That is a kref_put() on fctx->fence_ref, so the context outlives the
teardown only while emitted fences still hold a reference of their own.
That is no safety net: whenever none do, the count reaches zero right
there and nouveau_fence_context_put() kfree()s fctx while the work is
still queued. &fctx->uevent_work is embedded in that allocation, so the
workqueue already dereferences freed memory when it picks the item up,
and nouveau_fence_uevent_work() can then take fctx->lock on it. With
CONFIG_DEBUG_OBJECTS_WORK and CONFIG_DEBUG_OBJECTS_FREE, kfree() of a
still-queued work item is reported as the free of an active object.
On live memory the re-armed work has nothing left to do:
nouveau_fence_context_kill() empties fctx->pending and sets fctx->killed
under fctx->lock, nouveau_fence_emit() then returns -ENODEV rather than
queueing anything new, and nouveau_fence_update() only reaches
nvif_event_block() if it signalled something off that list. The defect
is the access to freed memory, not what the work would have found.
Only chips from G84 on can reach this at all: nouveau_fence_context_new()
returns before nvif_event_ctor() when priv->uevent is clear, and
nv84_fence_create() is the only place that sets it.
nv84_fence_context_del() is the context_del for all of those, because
nvc0_fence_create() and gv100_fence_create() build on nv84_fence_create()
and override only context_new.
Use disable_work_sync() instead. It drains the work exactly like
cancel_work_sync() does, and additionally increments the work item's
disable count, after which "any attempt to queue @work will fail and
return %false" (kernel/workqueue.c, disable_work()). The handler's
schedule_work() then has nothing to re-arm, and the teardown order stays
as it is.
Draining a second time after nvif_event_dtor() would close the window as
well, and without the newer API: once nvkm_event_ntfy_remove() has
returned, no handler can start or still be running, so nothing re-arms
the work past that point. disable_work_sync() is preferred here because
it needs one synchronisation point instead of two, it keeps the work
from being queued at all rather than cleaning up after it, and it is
what drm has settled on for this (drm/xe, drm/panthor, drm_pagemap).
Blocking the event rather than the work is not an option: fctx->event is
created with wait = false, so a handler already inside nvkm_event_ntfy()
can still queue the work.
Note for backports: disable_work_sync() arrived in v6.10 with
commit 86898fa6b8cd ("workqueue: Implement disable/enable for (delayed)
work items"), while the fix being corrected here reached 6.6.18 and
6.7.6. linux-6.6.y therefore carries this bug without the API, and this
patch would apply there and then fail to build. A 6.6.y backport wants
the second drain described above instead, as its own patch.
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/all/20260812233022.159301F000E9@smtp.kernel.org/
Fixes: 39126abc5e20 ("nouveau: offload fence uevents work to workqueue")
Cc: <stable@vger.kernel.org> # 6.10.x
Assisted-by: Claude:claude-opus-5
Signed-off-by: Marek Czernohous <marek@czernohous.de>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/178688513268.513871.3468844663561639695@gmail.com
|
|
The database was updated and the WA is no longer listed as applicable
to media 3503, so don't enable it there.
Fixes: c57db41b8d2c ("drm/xe/guc: Add Wa_14025883347 for GuC DMA failure on reset")
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Sk Anirban <sk.anirban@intel.com>
Cc: Badal Nilawar <badal.nilawar@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patch.msgid.link/20260818213520.283063-1-daniele.ceraolospurio@intel.com
|
|
None of panfrost_features.h definitions are used in panfrost_device.c
Reviewed-by: Steven Price <steven.price@arm.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Link: https://patch.msgid.link/20260820-b4-panfrost-fixes-v1-3-1b2c2376f4bc@collabora.com
Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>
|
|
When updating and debug-printing devfreq stats, in the very unlikely
off-chance that total device time is less than 100 ns, division by 0
would happen. Fix the divisor when relevant and rearrange operands.
Reported-by: Claude <noreply@anthropic.com>
Closes: https://gitlab.freedesktop.org/panfrost/linux/-/work_items/88
Fixes: f3ba91228e8e ("drm/panfrost: Add initial panfrost driver")
Reviewed-by: Steven Price <steven.price@arm.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Link: https://patch.msgid.link/20260820-b4-panfrost-fixes-v1-2-1b2c2376f4bc@collabora.com
Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>
|
|
When querying the properties of a BO, whether it was created in
Writeback caching mode for userspace mappings, is recorded in a Panfrost
BO's wb_mmap field.
Reported-by: Claude <noreply@anthropic.com>
Closes: https://gitlab.freedesktop.org/panfrost/linux/-/work_items/88
Fixes: 62eedf1ccba5 ("drm/panfrost: Add flag to map GEM object Write-Back Cacheable")
Reviewed-by: Steven Price <steven.price@arm.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Link: https://patch.msgid.link/20260820-b4-panfrost-fixes-v1-1-1b2c2376f4bc@collabora.com
Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>
|
|
igt@xe_wedged@wedged-at-any-timeout wedges the device in mode 2
(UPON_ANY_HANG_NO_RESET) and then rebinds the driver. During unbind,
a GSC proxy kernel submission can still time out; with the device wedged
and the GuC CT stopped it can never complete, so its kernel job times out.
Tile0: GT1: Kernel-submitted job timed out
WARNING: drivers/gpu/drm/xe/xe_guc_submit.c:...
at guc_exec_queue_timedout_job()
Workqueue: gt-ordered-wq drm_sched_job_timedout
Killed queues skip guc_submit_hint_wedged(), leaving 'wedged' false even
though the device is already wedged. The timeout handler then treats the
kernel queue timeout as unexpected and taints the kernel.
Honour an already-wedged device even for killed queues so the expected
teardown timeout no longer trips the WARN.
Fixes: 5a2f117a80c2 ("drm/xe: Do not wedge device on killed exec queues")
Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patch.msgid.link/20260814074106.92670-2-nitin.r.gote@intel.com
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
(cherry picked from commit a1c1dbd0f047bb05de6aaf6abe9103031179bf19)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
|
|
igt@xe_wedged@wedged-at-any-timeout wedges the device in mode 2
(UPON_ANY_HANG_NO_RESET) and then rebinds the driver. During unbind,
a GSC proxy kernel submission can still time out; with the device wedged
and the GuC CT stopped it can never complete, so its kernel job times out.
Tile0: GT1: Kernel-submitted job timed out
WARNING: drivers/gpu/drm/xe/xe_guc_submit.c:...
at guc_exec_queue_timedout_job()
Workqueue: gt-ordered-wq drm_sched_job_timedout
Killed queues skip guc_submit_hint_wedged(), leaving 'wedged' false even
though the device is already wedged. The timeout handler then treats the
kernel queue timeout as unexpected and taints the kernel.
Honour an already-wedged device even for killed queues so the expected
teardown timeout no longer trips the WARN.
Fixes: 5a2f117a80c2 ("drm/xe: Do not wedge device on killed exec queues")
Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patch.msgid.link/20260814074106.92670-2-nitin.r.gote@intel.com
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
|
|
EXEC_QUEUE_FLAG_PERMANENT used to mark kernel-owned exec queues that were
only torn down at driver unload, so that guc_exec_queue_destroy() could
take a synchronous teardown path for them while user queues went through
the async destroy worker.
Exec-queue teardown now routes every queue through the single
message-based path (with the message layer gating any HW/H2G work behind
drm_dev_enter()/drm_dev_exit() and taking a runtime-PM reference only
while the device is bound), so the PERMANENT flag no longer has any effect.
Drop it and all its users, and renumber the remaining flags to keep the
bit range contiguous.
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260811133753.1742666-6-nitin.r.gote@intel.com
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
|
|
The core_hotunplug "with-load" subtests trigger a "Missing outer runtime
PM protection" warning when an exec queue is torn down from a deferred fd
close after the device has been hot-unplugged:
guc_exec_queue_add_msg
guc_exec_queue_destroy
xe_exec_queue_destroy
xe_file_close
The reference is not actually missing. On unplug the PCI core disables
runtime PM (device_del() -> __pm_runtime_disable()) while the old
drm_device is kept alive by the still-open fd. Once runtime PM is
disabled pm_runtime_get_if_in_use() returns -EINVAL regardless of the
usage count, so xe_pm_runtime_get_noresume() can no longer observe the
held reference and warns.
drm_dev_unplug() always runs before runtime PM is disabled, so gate the
message submission layer on the bound state instead of the PM reference:
- guc_exec_queue_add_msg(): only take the noresume PM reference while
drm_dev_enter() succeeds and record it in the message (MSG_PM_REF), so
the put in guc_exec_queue_process_msg() stays balanced.
- guc_exec_queue_process_msg(): re-check the bound state and skip each
handler's HW access once the device is gone.
- guc_exec_queue_destroy(): always post the CLEANUP message; the cleanup
handler issues the deregister H2G only while bound, registered, fw
running and not wedged, otherwise it tears down on the driver side.
Also document the post-unplug PM-reference hazard in
xe_pm_runtime_get_noresume().
Observed with new IGT core_hotunplug subtests:
igt@core_hotunplug@hotreplug-with-load
igt@core_hotunplug@hotunplug-rescan-with-load
v2:
- Drop the drm_dev_is_unplugged() bypass from guc_exec_queue_destroy()
and instead exclude hot-unplug from the WARN in
xe_pm_runtime_get_noresume().
v3:
- Rework the fix in the message-submission layer instead of touching the
WARN; gate PM/HW work on drm_dev_enter() and route all queues through
the CLEANUP message. (Matthew Brost)
- Prove the root cause (runtime PM disabled on unplug, not a zero
refcount) and record it in the commit message. (Matt B, Raag)
- Also remove EXEC_QUEUE_FLAG_PERMANENT flag in
separate patch (Matthew Brost)
Link: https://patchwork.freedesktop.org/patch/725773/?series=166744&rev=4
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Raag Jadav <raag.jadav@intel.com>
Signed-off-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Tested-by: Varun Gupta <varun.gupta@intel.com>
Link: https://patch.msgid.link/20260811133753.1742666-5-nitin.r.gote@intel.com
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
|
|
Linux 7.2
There was a lot of conflicts this round between fixes and next,
and I'd like to get the merge resolutions that we have in drm-tip.
Signed-off-by: Dave Airlie <airlied@redhat.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Pull crypto update from Herbert Xu:
"API:
- Add af_alg_restrict sysctl and white list
- Fix potential suspend/resume races in hwrng
Algorithms:
- Optimize vli additive operations using compiler builtins in ecc
Drivers:
- Remove unsafe/deprecated algorithms from qce
- Mark qce as BROKEN
- Add runtime PM and interconnect bandwidth scaling support to qce
- Remove crypto_rng from qcom, sun8i and caam
- Fix SG list issues in iaa
- Fix SEV init path bugs in ccp"
* tag 'v7.3-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (122 commits)
crypto: lskcipher - propagate errors from unaligned crypt
crypto: keembay - use crypto_memneq() to compare CCM AEAD tags
crypto: keembay - use crypto_memneq() to compare GCM AEAD tags
crypto: sa2ul - use crypto_memneq() to compare AEAD tag
hwrng: drivers - use named initializers for acpi_device_id
crypto: qce - fix CCM AAD buffer underallocation
crypto: iaa - unmap dst before software fallback on decompress
crypto: iaa - use bounce buffer for multi-sg decompress input
crypto: iaa - avoid counting fallback decompression bytes
crypto: iaa - fall back to software for multi-entry scatterlists
hwrng: core - Stop/start hwrng_fillfn() kthread before/after suspend-resume
crypto: hisilicon/sec2 - fix CCM algorithm long packet failure
crypto: eip93 - use struct_size() and flexible array for ring allocation
crypto: krb5 - use kfree_sensitive() for derived key buffers
crypto: af_alg - Stop after finding name in allowlist
crypto: af_alg - Replace 'bool privileged' with flags
crypto: af_alg - Make cbc(paes) privileged-only
hwrng: imx-rngc - Disable clock on registration failure
crypto: qat - remove dead ADF_HEX code
crypto: qce - simplify qce_handle_request
...
|
|
Xe3p onwards CURSOR_BUF/WM masks have been increased. Update them to
reflect the following.
Bspec: 69847, 69880
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Reviewed-by: Dnyaneshwar Bhadane <dnyaneshwar.bhadane@intel.com>
Link: https://patch.msgid.link/20260724-xe3p_lpd-cursor-dbuf-increase-v1-1-42f2faae2fb5@intel.com
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/driver-core/driver-core
Pull driver core updates from Danilo Krummrich:
"container_of:
- Apply typeof_member(), remove the local __mptr variable to
eliminate variable shadowing warnings on nested container_of()
calls, and remove unnecessary parentheses
core:
- Add driver name to probe debug print for initcall_debug
- Avoid repeatedly printing the same 'Fixed dependency cycle' log
- Unwind device_add() on attribute creation failure in
attribute_container_add_class_device()
- Remove statistics group if encryption group creation fails in
transport_add_class_device()
debugfs:
- Fix lockdown check for mmap_prepare()
- Warn if file creation failed due to uninitialized debugfs
device property:
- Implement fw_devlink support for software nodes by adding
software_node_add_links(), which creates fwnode links from
DEV_PROP_REF properties to enable automatic probe ordering. Add
kunit-managed fwnode helpers and test coverage
- Fix infinite loop in fwnode_for_each_child_node() when the
secondary fwnode has more than one child. Add test cases
- Fix out-of-bounds access in software_node_get_reference_args() when
called with index -1 (UINT_MAX)
- Refactor to use RAII approach with __free()
- Add Bartosz Golaszewski as software node reviewer
firmware loader:
- Fix race where a sysfs fallback request can complete before being
queued as pending, leading to a use-after-free on the next fallback
request
- Reject 0-size built-in firmware and fail the build on empty
firmware files in CONFIG_EXTRA_FIRMWARE
kobject:
- Provide __KOBJ_ATTR() and __KOBJ_ATTR_RO/WO() initialization macros
and allow the constification of kobject attributes, enabling them
to reside in read-only memory
platform:
- Provide platform_device_set_of_node(), platform_device_set_fwnode(),
and platform_device_set_of_node_from_dev() helpers that encapsulate
firmware node reference counting for dynamically allocated platform
devices
Convert all in-tree users that manually assigned dev.of_node or
dev.fwnode, fixing a pre-existing refcount bug in powermac. Switch
to counting references of all firmware node types, not only OF
nodes
- Unify the release path for dynamically allocated platform devices
by removing platform_device_release_full(). Amend the fwnode setter
API contract to warn if a primary software node is overwritten. Add
KUnit tests for correct software node removal on device
unregistration
Rust:
- Auxiliary:
- Add registration_data_with() closure-based API for invariant
ForLt types
- Debugfs:
- Migrate BinaryWriter and BinaryReaderMut trait requirements
from kernel::transmute traits to zerocopy traits
- Device:
- Add BoundInternal device context and InternalBoundContext trait
for bus abstractions that need internal access to a bound
device.
- Make the lifetime on Core and CoreInternal invariant to prevent
coercion to shorter lifetimes
- Devres:
- Fix race between concurrent revokers where the losing revoker
could return before the winning revoker finished dropping the
inner data, causing use-after-free.
- Ensure revocation is complete before the device finishes
unbinding by making the synchronization bidirectional.
- Add DevresLt<F: ForLt>, a wrapper around Devres that shortens
'static back to the caller's borrow scope. Implement ForLt and
CovariantForLt for Bar, IoMem, and ExclusiveIoMem
- Driver:
- Switch from index-based to pointer-based device ID info lookup,
storing static references in driver_data. Centralize device ID
handling in device_id.rs, removing the open-coded ACPI/OF
matching logic and duplicate ID table from driver.rs
- I/O:
- Make I/O regions typed (with a dynamically-sized Region type
for the existing untyped case), create view types representing
subregions of a mapped I/O region, and add io_project!() for
safely creating subviews.
- Split Io into a base trait (IoBase) and an extension trait (Io)
with a blanket implementation, preventing implementers from
overriding provided methods that unsafe code relies on.
- Add a SysMem backend for shared system memory with volatile
access, and make Coherent implement Io via an I/O view type.
Add IoSysMap as sum type of Mmio and SysMem. Add copying
methods (memcpy_{from,to}io()) and read_val()/write_val() for
typed access.
- Replace dma_read!()/dma_write!() with io_read!()/io_write!()
for primitives and copying methods for aggregates; drop the old
macros. Convert nova-core to use I/O projection.
- Fix internal shortcut rule dispatch in the register!() macro,
remove unused rule arguments, and use path fragments for alias
destinations
- IRQ:
- Make irq::Registration compatible with lifetime-bound drivers
by removing the 'static bound on Handler/ThreadedHandler and
replacing Devres<RegistrationInner> with direct
request_irq()/free_irq() calls. Handlers can now directly own
lifetime-bound device resources
- PCI:
- Convert IrqVectorRegistration to a lifetime-annotated owning
type, giving drivers explicit control over the allocation
lifetime. IrqVector embeds a resolved IrqRequest, making the
conversion infallible. Remove the redundant
request_irq()/request_threaded_irq() wrappers from pci::Device.
- Add pci_irq_type() C helper and expose it via irq_type() on
IrqVectorRegistration and IrqVector, returning PCI_IRQ_MSIX,
PCI_IRQ_MSI, or PCI_IRQ_INTX.
- Mark pci::Device refcount methods inline
- Serdev:
- Add Rust abstractions for the serial device bus, including
serdev::Driver trait, serdev::Device wrapping struct
serdev_device, and serdev::Adapter implementing
RegistrationOps. Includes a sample driver. Markus Probst takes
over as serdev maintainer for both C and Rust code
- Misc:
- Split ForLt into a base trait (providing the Of<'a> GAT) and an
unsafe CovariantForLt subtrait guaranteeing covariance,
enabling invariant types (e.g. those containing Mutex<&'bound T>)
to participate in the ForLt abstraction.
- Fix Coherent read past EOF returning -ERANGE instead of zero.
- Fix firmware example UB by avoiding null-pointer ARef
misc:
- Avoid iattr allocation in kernfs listxattr by using
kernfs_iattrs_noalloc().
- Unregister SoC bus on early device registration failure.
- Remove unused DMA_FENCE_TRACE Kconfig symbol.
- Fix /sys/module path in comment.
- Refactor ISA bus init to remove nested blocks.
- Remove redundant nodemask clears in numa_init().
- Add kernel-doc for fwnode_operations and sys_soc.h, mark
internal property data as private for kernel-doc, and add
property.h/fwnode.h to driver-api infrastructure docs.
- Add MAINTAINERS entry for sys_soc.h"
* tag 'driver-core-7.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/driver-core/driver-core: (129 commits)
rust: pci: expose the allocated interrupt type
PCI: Add pci_irq_type() to query the allocated interrupt type
rust: pci: remove request_irq() and request_threaded_irq() from Device
rust: pci: resolve IRQ in index() and embed IrqRequest in IrqVector
rust: pci: convert IrqVectorRegistration to a lifetime-managed owning type
kernfs: avoid iattr allocation in listxattr
rust: serdev: use ThisModule::as_ptr() instead of field access
ACPI/IORT: use platform_device_set_fwnode()
ACPI/APMT: use platform_device_set_fwnode()
firmware_loader: do not queue completed sysfs fallback requests
rust: pci: Mark Device refcount methods inline
rust: irq: make Registration compatible with lifetime-bound drivers
rust: net/phy: remove expansion from doc
rust: dma: return zero for Coherent reads past EOF
rust: io: register: use path fragment for alias destination
rust: io: register: remove unused rule arguments
rust: io: register: dispatch shortcut rules internally
MAINTAINERS: add sys_soc.h to DRIVER CORE
rust: debugfs: remove unsafe blocks from traits impl for Vec
rust: debugfs: migrate debugfs traits requirements to zerocopy
...
|
|
Update PCI error handling path to disconnect the device when
XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET is set. This prevents bus
reset and preserves the device state so that it is available to
the user for debugging.
Signed-off-by: Raag Jadav <raag.jadav@intel.com>
Reviewed-by: Riana Tauro <riana.tauro@intel.com>
Link: https://patch.msgid.link/20260819113519.510853-1-raag.jadav@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
|
|
Unlike the CRTC degamma path, which is guarded by
amdgpu_dm_verify_lut_sizes(), the per-plane degamma LUT size was never
validated before use. __set_dm_plane_degamma() passed the user-supplied
size straight into __is_lut_linear() and, for a non-linear LUT, into
__set_input_tf() -> __drm_lut_to_dc_gamma(), the latter always iterating
MAX_COLOR_LUT_ENTRIES entries regardless of the actual LUT size.
A malformed AMD_PLANE_DEGAMMA_LUT blob (e.g. a single entry) could thus
trigger a divide-by-zero in __is_lut_linear() or an out-of-bounds read in
__drm_lut_to_dc_gamma(). Reject any plane degamma LUT whose size does not
match MAX_COLOR_LUT_ENTRIES, mirroring the invariant the code already
asserts a few lines below (and which the CRTC path enforces).
The AMD_PLANE_DEGAMMA_LUT property is only exposed on builds with
AMD_PRIVATE_COLOR defined.
Fixes: 980f8710075a ("drm/amd/display: add plane degamma TF and LUT support")
Cc: stable@vger.kernel.org
Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Melissa Wen <mwen@igalia.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
__is_lut_linear() computes the expected value of each entry with
expected = i * MAX_DRM_LUT_VALUE / (size - 1);
If it is ever called with a single-entry LUT, size - 1 is zero and the
kernel takes a divide error (#DE). A LUT with fewer than two entries
cannot describe a linear mapping anyway, so return false early instead
of dividing by zero.
Fixes: 086247a4b2fb ("drm/amd/display: Use 4096 lut entries")
Cc: stable@vger.kernel.org
Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Melissa Wen <mwen@igalia.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
If we end up emitting a VM fence keep pipeline sync
associated with that fence. If not, emit them as
part of the IB fence.
v2: fix need_pipe_sync handling
v3: simplify the function
Cc: David Rosca <david.rosca@amd.com>
Fixes: cb1e657ccac8 ("drm/amdgpu: handle GDS and SPM without a VM fence")
Reviewed-by: David Rosca <david.rosca@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
If the supplied msg[2] (num_buffers) is 0x3FFFFFFF, the expression
6 + num_buffers * 4 wraps to 2 and the bounds check passes, letting
the parser loop far past the end of the message BO. Triggering it
additionally requires a ~4GiB mapping so that msg[1] survives the
earlier "header does not fit in BO" check.
Rewrite the test in division form, which is overflow-free by
construction. Also update the message to reflect that msg is invalid.
Fixes: b193019860d6 ("drm/amdgpu/vcn3: Prevent OOB reads when parsing dec msg")
Fixes: 0a78f2bac142 ("drm/amdgpu/vcn4: Prevent OOB reads when parsing dec msg")
Cc: stable@vger.kernel.org
Signed-off-by: David (Ming Qiang) Wu <David.Wu3@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The Apple iMac11,1 (27-inch, Late 2009) uses a Mobility Radeon HD 4850
(RV770/DCE3.1) with a 2560x1440 internal panel on an internal
DisplayPort path. Without this fix the display stays dark under KMS.
This machine suffers from the same issue as iMac10,1 and iMac11,2:
Apple routes the internal display through Link B of the DIG encoder
instead of Link A. Add iMac11,1 to the existing DMI quirk and move
the Apple-specific encoder assignment into its own block, independent
of the DCE version check.
Additionally, the 2560x1440 panel requires RADEON_PLL_USE_FRAC_FB_DIV
and ATOM_ENCODER_CMD_DP_VIDEO_ON, limited to iMac11,1 via dmi_match()
to avoid affecting other boards.
Reviewed-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Gilles Risch <gilles.risch@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
When a queue is hung, the hang_detect_work is the
only way to recover it. However in amdgpu_userq_destroy(),
the hang_detect_work is cancelled too early,
resulting in amdgpu_userq_wait_for_last_fence()
may never return, leaving an uninterruptible dma_fence_wait()
hang there.
To fix this problem, this commit moves the cancelling of
hang_detect_work after amdgpu_userq_wait_for_last_fence(), and it has
to be before the unmap helper, because hang_detect_work resets the
queue, so it races with amdgpu_userq_unmap_helper() for MES operations
and queue state.
This commit splits amdgpu_userq_cleanup() into two parts:
1) amdgpu_userq_detach_doorbell(), which detaches the queue from
userq_doorbell_xa. This has to be called before the cancel, otherwise
the IRQ handlers (for example amdgpu_userq_process_fence_irq)
can re-schedule the hang_detect_work and the cancel is not final.
2) amdgpu_userq_fence_driver_free(), this has to be called after the
unmap helper, because it can release the seq64 slot that the GPU
writes fence values to.
Only one cancel_delayed_work_sync(&queue->hang_detect_work) is needed,
so other redundancies are removed.
Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
profile_lock_device() may return negive error code,
so the type of the return value should be int,
not uint32
Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
When amdgpu_init_minimal_xgmi is used, SDMA engines init
is delayed so amdgpu_ttm_enable_buffer_funcs must be
called later.
Without this, the check for num_buffer_funcs_scheds will
fail and using ttm buffer funcs later will fail.
Given that amdgpu_ttm_enable_buffer_funcs is a no-op if
amdgpu_in_reset() returns true, the call has to occur
after the reset lock is dropped.
Cc: stable@vger.kernel.org
Fixes: e4029f7a9474 ("drm/amdgpu: only use working sdma schedulers for ttm")
Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Introduce a helper to consolidate the vga_switcheroo registration condition
used by the init and fini paths.
Keep the explicit pci_is_thunderbolt_attached() check, as dev_is_removable()
does not provide equivalent coverage for Thunderbolt-attached GPUs.
This ensures such devices remain excluded from switcheroo registration while
preserving the existing PX and Apple gmux handling.
Cc: stable@vger.kernel.org
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Like the MES scheduler ring, the KIQ ring sets no_scheduler = true and uses a
polling fence, so it is skipped by the force-completion loop in
amdgpu_device_pre_asic_reset(). Its hw fence value lives in wb (GTT) memory and
survives a MODE1 reset while fence_drv.sync_seq keeps advancing, so after a
reset the first KIQ submission can poll forever on a seq that is never written
back.
Force complete the KIQ ring fences too so their hw fence is realigned to
sync_seq.
Cc: stable@vger.kernel.org
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
rptr and wptr of a userq are 8 bytes aligned, and may
not placed on a page boundary.
This commit checks whether rptr and wptr are 8 bytes
aligned, and expectes 8 bytes when validates rptr/wptr VA.
With above changes, this commit fixes an regression
in amdgpu_userq_input_va_validate, where
end_addr is caculated by:
check_add_overflow(start_addr, expected_size - 1, &end_addr).
Wptr and rptr are very likely not to be page aligned,
when validating rptr and wptr, if they are located in the last
mapped page(or only one page is mapped)
and expected_size is PAGE_SIZE, end_addr will exceed the last
mapped page, means (end_addr >> AMDGPU_GPU_PAGE_SHIFT) > va_map->last,
and causing an -EINVAL, even it is a valid VA.
Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Fixes: c0122bf2ccb1 ("drm/amdgpu: fix userq VA validation for sub-page buffers")
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The MES scheduler ring has no drm scheduler (no_scheduler = true), so it is
skipped by the force-completion loop in amdgpu_device_pre_asic_reset(). It uses
a polling fence whose hw value lives in wb (GTT) memory and survives a MODE1
reset, while fence_drv.sync_seq keeps advancing for every packet.
When the reset is triggered because MES itself stopped responding, the
timed-out packets advance sync_seq past the last hw fence value MES wrote.
After resume the first MES submission polls forever on a seq that is never
written back, failing the resume and wedging the box on a second reset:
amdgpu: MES ring buffer is full.
amdgpu: *ERROR* ring gfx_0.0.0 test failed (-110)
amdgpu: resume of IP block <gfx_v11_0> failed -110
amdgpu: GPU reset end with ret = -110
Force complete the MES scheduler ring fences together with the scheduler rings
so their hw fence is realigned to sync_seq.
v2: cover all XCCs (one scheduler ring each), not just mes.ring[0].
Cc: stable@vger.kernel.org
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
In standalone mode the ip_discovery sysfs tree is built from a verbatim
copy of the discovery binary taken before reg_base_init() collapses the
64bit base addresses in place. Decoding as 32bit there yields interleaved
zeros. Decode base_address_64[] in that case; keep reading the already
collapsed adev->discovery.bin as-is otherwise.
Fixes: 402e04f11ff7 ("drm/amdgpu: Export ip_discovery sysfs on probe failure")
Cc: stable@vger.kernel.org
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Custom brightness curves use an 8-bit input signal. After exporting the
full PWM range to userspace, the curve normalizer still divides requests
by the physical PWM span. On panels with a nonzero minimum PWM level,
this can produce a curve input greater than 255 and send an invalid
backlight level to DC.
Scale the userspace [0..max] range to the curve's [0..255] range
instead. This retains the full advertised range and keeps the reverse
readback conversion unchanged.
Fixes: 8dbd72cb7900 ("drm/amd/display: Export full brightness range to userspace")
Cc: stable@vger.kernel.org
Signed-off-by: Akhmed Zhitaev <zhitaevakh@gmail.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
(Move to amdgpu_dm_backlight.c)
Link: https://patch.msgid.link/20260813170959.22073-1-zhitaevakh@gmail.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
amdgpu_userq_wait_return_fence_info() calls drm_exec_init() without
DRM_EXEC_IGNORE_DUPLICATES. When the same GEM object appears more than
once across the read/write BO handle lists submitted by userspace,
drm_exec_lock_obj() returns -EALREADY the second time it locks that
object, which aborts the fence resolution pass instead of treating the
repeat as a no-op.
Add DRM_EXEC_IGNORE_DUPLICATES so duplicate objects are silently
skipped on the second lock attempt.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
amdgpu_userq_wait_count_fences() calls drm_exec_init() without
DRM_EXEC_IGNORE_DUPLICATES. When the same GEM object appears more than
once across the read/write BO handle lists submitted by userspace,
drm_exec_lock_obj() returns -EALREADY the second time it locks that
object, which aborts the fence-counting pass instead of treating the
repeat as a no-op.
Add DRM_EXEC_IGNORE_DUPLICATES so duplicate objects are silently
skipped on the second lock attempt.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
amdgpu_userq_signal_ioctl() calls drm_exec_init() without
DRM_EXEC_IGNORE_DUPLICATES. When the same GEM object appears more than
once across the read/write BO handle lists submitted by userspace,
drm_exec_lock_obj() returns -EALREADY the second time it locks that
object, which aborts the ioctl instead of treating the repeat as a
no-op.
Add DRM_EXEC_IGNORE_DUPLICATES so duplicate objects are silently
skipped on the second lock attempt, matching the intended semantics of
locking a set of (possibly overlapping) BOs before publishing a fence
on them.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
amdgpu_userq_buffer_vas_mapped() checks whether all VAs
of a queue are mapped before restoring it.
So that HW won't access any invalid addresses.
Currently, this function assumes all VAs are mapped if
any VA of a queue has been mapped, which is wrong.
This commit fixes this problem by examining all VAs of
a queue and reporting false if any of them is not mapped.
Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
amdgpu_pci_probe() calls pm_runtime_use_autosuspend(), but
amdgpu_pci_remove() does not call the matching
pm_runtime_dont_use_autosuspend().
If the autosuspend delay is set to a negative value while autosuspend
is enabled, the runtime PM core increments usage_count to prevent
runtime suspend. Without calling pm_runtime_dont_use_autosuspend()
during teardown, this reference is not dropped and usage_count remains
unbalanced.
The documentation for pm_runtime_use_autosuspend() also notes that it
is important to undo it with pm_runtime_dont_use_autosuspend() at
driver exit time, unless runtime PM was initially enabled with
devm_pm_runtime_enable().
Add the missing pm_runtime_dont_use_autosuspend() call to the remove
path.
This issue was found by manual code inspection.
Fixes: d38ceaf99ed0 ("drm/amdgpu: add core driver (v4)")
Cc: stable@vger.kernel.org
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Link: https://patch.msgid.link/20260808120934.2813010-1-lgs201920130244@gmail.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
pci_is_thunderbolt_attached() requires an upstream PCI bridge with
is_thunderbolt set from an Intel Thunderbolt VSEC. This does not cover
the affected ASM4242 USB4 PCI hierarchy:
00:02.2
\- 0f:00.0 [1b21:2421]
+- 10:01.0 [1b21:2423] -> 45:00.0 -> 46:00.0
| -> 47:00.0 -> 48:00.0 -> 49:00.0 [1002:7590]
\- 10:03.0 -> 76:00.0 [1b21:2425] USB4 Host Router
The host router is outside the GPU upstream bridge chain, leaving no
ancestor with is_thunderbolt set. PCI core propagates DEVICE_REMOVABLE
below the external-facing PCIe tunnel. Disable Runtime PM when either
pci_is_thunderbolt_attached() or dev_is_removable() is true.
Cc: stable@vger.kernel.org
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Candice Li <candice.li@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
icl_plane_disable_sel_fetch_arm() wrote SEL_FETCH_PLANE_CTL = 0 only when
crtc_state->enable_psr2_sel_fetch was set. If a plane was disabled after
selective fetch had been turned off, the guard fired early and left the
register's enable bit set in hardware.
The bit is harmless until selective fetch is re-enabled. When it is, the
hardware resumes fetching for the now-disabled plane and keeps its old DDB
range reserved.
i9xx_cursor_disable_sel_fetch_arm() has the same guard on SEL_FETCH_CUR_CTL
and is fixed the same way.
v2: Add same check for cursor also. [sashiko]
Cc: stable@vger.kernel.org
Fixes: b1f5279b5981 ("drm/i915/psr: Move plane sel fetch configuration into plane source files")
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/8739
Assisted-by: GitHub-Copilot:claude-opus-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Jouni Högander <jouni.hogander@intel.com>
Signed-off-by: Animesh Manna <animesh.manna@intel.com>
Link: https://patch.msgid.link/20260818095149.2172935-1-nemesa.garg@intel.com
|
|
The verisilicon driver has a custom framebuffer address calculating
helper that the common drm_fb_dma_get_addr() can substitute.
Differences from drm_fb_dma_get_addr():
- Uses drm_format_info_min_pitch() to calculate the horizontal offset;
however the driver does not support any of the blocked formats, so
this just ends up being the same as in drm_fb_dma_get_addr():
"cpp[plane] * y"
- Uses clipped source coordinates instead of non-clipped coordinates
as in drm_fb_dma_get_addr();
For the primary plane this doesn't matter, since the primary plane
must match the output, i.e. it cannot be clipped. Also this driver
doesn't support scaling.
For the cursor plane this seems wrong, as the clipping seems to be
done by the hardware, and thus the buffer address should be unclipped.
As such, it should be fine to use the common helper and drop the custom
code.
Tested-by: Icenowy Zheng <zhengxingda@iscas.ac.cn>
Reviewed-by: Icenowy Zheng <zhengxingda@iscas.ac.cn>
Fixes: 8c4ae2189125 ("drm: verisilicon: add support for cursor planes")
Link: https://patch.msgid.link/20260813065127.341349-1-wenst@chromium.org
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
|
|
The 4th argument of regmap_update_bits() expects a pre-shifted register
value instead of the field value when it's used to set a single field,
however I forgot this when writing the code setting the cursor X/Y
offset fields, thus the write operation doesn't happen at all (because
of the non-shifted values being masked by regmap_update_bits()).
Fix this problem so the cursor is correctly clipped.
The framebuffer address is also clipped so the cursor offset problem was
previously not discovered, and the framebuffer address deserves another
fix for proper cursor behavior when it's clipped.
Fixes: 8c4ae2189125 ("drm: verisilicon: add support for cursor planes")
Signed-off-by: Icenowy Zheng <zhengxingda@iscas.ac.cn>
Reviewed-by: Chen-Yu Tsai <wenst@chromium.org>
Link: https://patch.msgid.link/20260812154829.671777-1-zhengxingda@iscas.ac.cn
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Pull SoC driver updates from Arnd Bergmann:
"The SoC driver changes once more consist of many small fixes and
cleanups, that are to a large part the result of automated testing.
On platform specific drivers, this includes SoC specific code for
xilinx, freescale/nxp, qualcomm, TI, aspeed, omap, tegra, samsung,
rockchip, renesas, ixp4xx. In firmware drivers, we see a similar
picture for SCMI and qcomtee.
Aside from these, we see actual new hardware support in a few areas:
- The Apple platform gets a new driver for low power states
- Updates to Qualcomm platform drivers add several new hardware
specific features and additional SoCs.
- Amlogic SoC support for A1 and T7 is added
- The Mediatek MMSYS driver is refactored as a cleanup"
* tag 'soc-drivers-7.3' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (157 commits)
soc: qcom: make QCOM_PDR_MSG selectable
soc: qcom: ubwc: Fix missing include
soc: qcom: ubwc: Fix link error when QCOM_SMEM=n
media: iris: Guard the QCOM_UBWC_CONFIG select with QCOM_SMEM
drm/msm: Guard the QCOM_UBWC_CONFIG select with QCOM_SMEM
dt-bindings: arm: qcom,ids: Add SoC ID for Snapdragon SDM 850
firmware: xilinx: Clear firmware notifiers across kexec transitions
firmware: xilinx: Release all peripheral devices from firmware
firmware: xilinx: Add support to clear EL3 PM state
firmware: xilinx: Propagate actual error from feature check
firmware: xilinx: Use TF-A feature check for TF-A-specific APIs
bus: fsl-mc: drop unused assignment of acpi_device_id::driver_data
soc: fsl: qe: check platform_driver_register() in qe_ic_of_init()
phy: lynx-10g: use RCW override procedure for dynamic protocol change
soc: fsl: guts: implement the RCW override procedure
dt-bindings: fsl: layerscape-dcfg: define DCFG_DCSR region
soc: fsl: guts: make fsl_soc_data available after fsl_guts_init()
soc: fsl: guts: make it easier to determine on which SoC we are running
soc: fsl: guts: add a central fsl_guts_read() function
soc: fsl: guts: add a global structure to hold state
...
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 cpuid updates from Borislav Petkov:
- Get rid of static_cpu_has() - one less API to care about testing CPU
features
- Unify the handling of CPU core types (performance, efficient, etc) by
mapping the vendor-specific types to Linux ones
- Continuation of the work of Ahmed Darwish to centralize CPUID leaf
representation
* tag 'x86_cpu_for_v7.3_rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/CPU: Rename struct cpuid_read_output to struct cpuid_output
x86/cpu/scattered: Sort it properly
x86/cpu: Use parsed CPUID(0x1)
x86/lib: Add CPUID(0x1) family and model calculation
x86/cpu: Use parsed CPUID(0x0)
x86/cpu/transmeta: Rescan CPUID(0x1) after modifying capabilities
x86/topology: Add TOPO_CPU_TYPE_LOW_POWER
x86/topology: Name the AMD core-type values
x86/topo: Map vendor CPU types to generic Linux such types
x86/bugs: Don't use cpu-type matching in cpu_vuln_blacklist
x86/cpu: Hide and rename static_cpu_has()
|
|
NVAC (MCP79/MCP7A) uses g94_pci_func, whose .msi_rearm is
nv40_pci_msi_rearm(): a re-arm write through the MMIO mirror of PCI
config space. On this IGP that path is unreliable; when a re-arm is
missed the interrupt line stays dead, command submission times out and
the GPU appears hung until reboot. On an Apple Mac mini (early 2009,
MCP79, boot0 0x0ac080b1) this showed as sporadic fifo timeouts and GPU
hangs under load unless MSI was disabled via config=NvMSI=0.
Give NVAC its own pci func that re-arms through real PCI config space
(nv46_pci_msi_rearm) instead. This follows existing precedent: nv46.c
documents the MMIO-mirror re-arm as broken on several related parts,
and commit 5112abc6a433 ("drm/nouveau/pci/g92: Fix rearm") fixed g92
the same way while moving the remaining chipsets, NVAC included, into
the newly added shared g94 table, where NVAC stayed on the MMIO path.
This change completes that fix for NVAC. The sibling IGP NVAA
(MCP77/MCP78) has MSI disabled entirely as "reported broken" in
nvkm_pci_new_(); NVAC works correctly once the re-arm goes through
config space, so disabling MSI is not necessary.
Only NVAC is switched: that is the hardware this has been validated
on. The other users of g94_pci_func (G94/G96/G98/GT2xx and the
MCP77/MCP89 IGPs) keep their current behavior; MCP77 and MCP89
plausibly want the same treatment but were not tested.
Tested on the Mac mini as a daily driver for two months with MSI
enabled and zero fifo timeouts. Independently confirmed stable on an
iMac9,1 (MCP79) running 6.12.90 with the v1 form of this change (the
same one-line functional switch, applied to that kernel's g94
implementation).
Fixes: 5112abc6a433 ("drm/nouveau/pci/g92: Fix rearm")
Cc: <stable@vger.kernel.org> # v6.16+
Tested-by: Fab Stz <fabstz-it@yahoo.fr>
Assisted-by: Claude:claude-opus-4-7
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Marek Czernohous <marek@czernohous.de>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/178688574401.522643.12812428444459119913@gmail.com
|
|
The SVM fault handler nouveau_svm_fault() looks up each fault's
nouveau_svmm under svm->mutex, caches it in the fault array, then drops
svm->mutex and dereferences the svmm across blocking faults
(mmu_interval_notifier_insert(), hmm_range_fault()).
nouveau_svmm is not refcounted. On DRM file close nouveau_svmm_fini() ->
mmu_notifier_put() frees it asynchronously via call_srcu() without
waiting for the handler, and mmget_not_zero() pins the mm, not the svmm.
flush_work() on the fault buffer otherwise runs only at device removal
and suspend, never on close, so a fault handled concurrently with a close
can dereference a freed svmm.
The instance is already unlinked by nouveau_svmm_part() earlier in the
same close, so no new fault can resolve to it. Drain the handler in
nouveau_svmm_fini() before the free, guarding the flush with a NULL check
on drm->svm: on device teardown nouveau_svm_fini() runs first and frees
drm->svm (NULL) after blocking the notify and flushing the buffer, so the
handler is already drained; on per-client close drm->svm is alive and the
flush runs.
Fixes: eeaf06ac1a55 ("drm/nouveau/svm: initial support for shared virtual memory")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260814-nouveau-svm-svmm-uaf-v2-1-8e6590519f07@gmail.com
|
|
nouveau_abi16_ioctl_notifierobj_alloc() passes the userspace-controlled
info->size to nvkm_mm_head() as both size_max and size_min without a lower
bound. A zero size satisfies the allocator's "e - s < size_min" gate
(size_min == 0 makes the unsigned comparison inert) and yields a
zero-length node, after which
args.limit = ntfy->node->offset + ntfy->node->length - 1;
underflows (offset 0, length 0 -> 0xffffffff) into an oversized ~4 GiB DMA
window. The dma object constructor only rejects start > limit, so the
validly ordered [base, base + 0xffffffff] range passes and is programmed
into the GPU DMA context. The ioctl is DRM_RENDER_ALLOW, so any render
node client can trigger this on pre-Fermi hardware.
Reject a zero-size request before allocating anything.
Fixes: ebb945a94bba ("drm/nouveau: port all engines to new engine module format")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260813-nouveau-abi16-notifierobj-zero-size-v1-1-20dee38077cb@gmail.com
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/ojeda/linux
Pull Rust updates from Miguel Ojeda:
"Toolchain and infrastructure:
- Warn when using 'bindgen' < 0.72.1 with 'libclang' >= 22, since
that combination may fail to build. It includes a probe for the bug
in case 'bindgen' happens to be patched, and tests
In parallel, Nathan updated the instructions for the kernel.org
LLVM+Rust toolchains so that the latest version of 'bindgen' is
installed, which should avoid some of these situations
- Support testing 'rust_is_available.sh' with 'bash' as '/bin/sh'
- Fix an objtool warning by adding one more 'noreturn' function for
Rust 1.99.0 (expected 2026-10-01)
- Fix build error in the 'rusttest' target due to ambiguity when the
'rustc-dev' component is installed, which was uncovered by the work
to support Rust's GCC backend ('rustc_codegen_gcc')
- Fix future Clang warnings in the upcoming powerpc support due to
macro redefinitions in the UAPI helper header by including the
arch-aware 'ioctl.h' header
'kernel' crate:
- Rework module ownership support:
- Move the module-related types into a new 'module' module and
make the 'THIS_MODULE' pointer a constant of 'ModuleMetadata'
so that modules can provide the pointer in const contexts, and
add a 'this_module' 'const fn' to retrieve it
This was enabled by upstream Rust's work on the 'const_mut_refs'
and 'const_refs_to_static' features which were stabilized back
in Rust 1.83.0
- Teach '#[vtable]' to associate implementations with their
owning module, defaulting to the local one, including fallbacks
for doctests, uses within the 'kernel' crate (like upcoming
KUnit '#[test]'s for DRM) and 'rusttest'
- Set 'fops.owner' from the module pointer for DRM and
miscdevice
- Migrate Rust Binder and configfs away from the old
'THIS_MODULE' 'static' and finally remove it from the 'module!'
macro
- 'num' module:
- Add the new 'casts' module for lossless integer conversions
Rust's 'core' library's 'From' implementations do not cover
conversions that are not portable or future-proof. However, the
kernel supports a narrower set of architectures, which makes it
helpful to provide more infallible conversions, instead of
having developers use 'as' casts, which carry the risk of
silently losing data
This goes along with previous work we did to avoid casts in
Rust kernel code since they are more powerful than needed
Thus, provide safe 'const' conversion functions (e.g.
'usize_as_u64' and 'u64_into_u8'), as well as the
'FromSafeCast' and 'IntoSafeCast' extension traits that provide
conversions that are known to be lossless in the kernel, and an
'arch' submodule defining conversions that are known to be
lossless on particular architectures (e.g. 64-bit platforms).
For instance:
// Conversion in const context.
const USIZED_CONST: usize = u8_as_usize(255u8);
// Non-const conversions.
let a = u64::from_safe_cast(4096usize);
let b: u64 = 4096usize.into_safe_cast();
- Add 'Bounded::shr_exact' method in the vein of 'try_shrink'
which shifts a bounded right only if it loses no set bits
- Fix unsoundness issue in the 'Bounded::shr' method by
rejecting, at compile-time, shifts of at least the type's bit
width
- 'fmt' module:
- Route '{:p}' raw pointer formatting through the kernel's hashed
'%p' format to prevent address leaks, including support for
width and padding. Include tests for both 'no_hash_pointers'
case and the default (hashed) one
- Fix the '{:p}' forwarding implementation, which could print the
address of a temporary stack variable
- 'time' module:
- Make 'Delta' generic over its time unit, with a default unit of
nanoseconds ('Nsec'), preserving the existing behavior. Then,
add a 'Jiffy' time unit
- Add the 'Delta::as_millis_ceil()' method
- Fix 'as_micros_ceil()' rounding near 'i64::MAX', which could
yield a result one microsecond too small
- 'sync' module:
- Implement 'ForeignOwnable' for 'ARef<T>', allowing C code to
own an 'ARef<T>'
- Add a safe abstraction for 'rcu_barrier()'
- 'error' module: add all of the remaining error codes, except the
deprecated compatibility aliases
- 'bug' module:
- Fix build error on UML in 'warn_on!' for callers from within
the 'kernel' crate
- Fix future 'dead_code' warning on arm and loongarch64 and under
'CONFIG_BUG=n' in 'warn_on!', which would trigger with the
upcoming SRCU abstractions
- Fix future build error in 'rusttest' on cross-compilation
cases, which would trigger when 'warn_on!' has callers inside
the 'kernel' crate
- 'bitfield' module: fix build error for the upcoming support for
Rust's GCC backend ('rustc_codegen_gcc') by always inlining a
couple conversions used in tests
'pin-init' crate:
- User-visible changes:
- Merge the '__pinned_init' and '__init' methods and make 'Init'
a marker trait
- Introduce public APIs 'raw_init' and 'raw_try_init' to prevent
users from needing to invoke the internal '__pinned_init' and
'__init' methods
- Emit errors for duplicate '#[pin]' attributes
- Link 'Zeroable::zeroed' and 'pin_init::zeroed' in documentation
- Other changes:
- Fix unwind safety issues
- Clean up lint 'allow' and 'expect's
- Overhaul '#[cfg]' handling to pave the way for tuple structs
and self-referential structs
- Mark many functions as '#[inline]' for better codegen with '-C
opt-level=s' ('CC_OPTIMIZE_FOR_SIZE')
'MAINTAINERS':
- Update 'MODULE SUPPORT' to cover the new 'module' module
And some other fixes, cleanups and improvements"
* tag 'rust-7.3' of git://git.kernel.org/pub/scm/linux/kernel/git/ojeda/linux: (54 commits)
rust: add functions and traits for lossless integer conversions
rust: kernel: add `LocalModule` fallback for `#[vtable]` `impl`s
rust: fmt: route {:p} through HashedPtr to prevent address leaks
rust: fmt: fix {:p} printing stack addresses
rust: module: update MAINTAINERS to cover module.rs
rust: macros: remove `THIS_MODULE` static from `module!`
rust_binder: use `LocalModule` for `THIS_MODULE`
rust: configfs: use `LocalModule` for `THIS_MODULE`
rust: miscdevice: set fops.owner from driver module pointer
rust: drm: set fops.owner from driver module pointer
rust: macros: auto-insert OwnerModule in #[vtable]
rust: doctest: add LocalModule fallback for #[vtable] ThisModule
rust: module: add `THIS_MODULE` const to `ModuleMetadata` trait
rust: module: move module types into `module.rs`
rust: num: add Bounded::shr_exact
rust: num: reject Bounded::shr overshifts at build time
rust: num: use const_assert! in Bounded
rust: uapi: replace direct asm-generic/ioctl.h include with linux/ioctl.h
rust: time: add Delta::as_millis_ceil()
rust: time: add jiffies time unit for Delta
...
|
|
drm_sched_entity_is_idle() contains a badly documented memory barrier
and an invalid lockless access to entity->stopped. Moreover, it also
accesses entity->list which is at all other places protected by
entity->lock.
This function is in no way performance critical, so it is safer, more
readable and more maintainable to take the spinlock.
Add locking to drm_sched_entity_is_idle().
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Philipp Stanner <phasta@kernel.org>
Link: https://patch.msgid.link/20260813094539.2407911-2-phasta@kernel.org
|
|
QCOM_UBWC_CONFIG is about to depend on QCOM_SMEM. Guard the DRM_MSM
select with 'if ARCH_QCOM && QCOM_SMEM', and add
'depends on QCOM_SMEM || QCOM_SMEM=n' so a built-in driver is never
selected against a missing or modular SMEM.
Fixes: 1b445022d1d0 ("soc: qcom: ubwc: Get HBB from SMEM")
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Acked-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Tested-by: Nathan Chancellor <nathan@kernel.org> # build
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
|
|
This is supposed to print the error code but there is a copy and
paste bug so it prints "bo" instead of "err".
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Dan Carpenter <error27@gmail.com>
Link: https://patch.msgid.link/an1tu0z3T-qX1ogn@stanley.mountain
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 28a4198c52a1468fc1b620a9837557ea1dc1766d)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
|
|
Follow up on the previous devm_drm_bridge_add() conversion by
converting the remaining manual cleanup in chipone-icn6211 to the
managed helper.
Suggested-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260521212529.151493-1-osama.abdelkader@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
As it turns out, Turing isn't the only architecture that needs this. On
this Dell Precision 7780 with an AD103 GPU, along with pretty much every
other laptop I tested, runtime PM is still somewhat unreliable. At first
glance it seems as if it's fixed, but lowering the autosuspend delay to
500ms and then doing a stress test of suspend/resume cycles on the GPU ends
up causing everything to start timing out.
After quite a lot of digging, I eventually landed back on this magic
timeout in r535_gsp_fini(). As it turns out, increasing the timeout ends up
fixing the runtime PM issues as far as I can tell, even during intense
stress testing.
Unfortunately after spending quite a bit of time trying to dig through
OpenRM to figure out what this magic sleep is actually doing, I've also
come up short with any reasonable explanation. In lieu of that, I'm going
to include the observations I did make while trying to figure this out in
hopes someone eventually does figure this out:
* The magic sleep has to occur after fbsr is initialized. Performing it at
any time before that doesn't appear to work.
* In situations where runtime PM starts getting flaky, some rather
interesting visual effects end up happening on occasion before the GPU
fully falls over. In particular, squares that look like the result of an
incomplete blitting operation to a tiled buffer end up showing up on
applications like vkcube. Interestingly enough, they remain in precisely
the same place between runtime PM cycles until the GPU falls over - even
when restarting vkcube multiple times, and even when vkcube is actively
updating the screen. Even more interestingly, they're not limited to a
specific framebuffer - you can see the squares changing as the cube
rotates around.
We cannot however, say that this is likely to be a incomplete fbsr
operation. The magic sleep happens before fbsr is actually saved (which
happens on the GSP unload), so it's something else.
* During a short bit of testing with a desktop that I have, the magic sleep
seemed to make no difference to whether or not suspend/resume works. It
seems to generally work almost always. So we can assume this is likely
exclusive to runtime PM, not S3.
As well, here's a list of the things I tried before settling on the magic
sleep:
* Hooking up NV2080_CTRL_CMD_INTERNAL_GCX_ENTRY_PREREQUISITE and then
blocking runtime PM until OpenRM signals that GC6/GCOFF is ready appears
to make no difference.
* Hooking up some (maybe not all, unsure about that part) bits of comptag
saving including:
* Fetching static memsys information from GSP
* Adding the size of the comptag storage to the fbsr data
* Adding a GA103+ workaround for disabling raw compression mode during
fbsr (it doesn't seem like it applies for any systems I tried it on
anyhow)
* Setting bPreserveVideoMemoryAllocations=1 in GspSystemInfo
So, until we can figure this out properly - just sleep for longer.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org> # v6.16+
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patch.msgid.link/20260814194542.781955-5-lyude@redhat.com
|
|
There's two flags that we've never been setting when asking GSP to suspend
the GPU, which OpenRM does set:
GPU_STATE_FLAGS_PRESERVING
GPU_STATE_FLAGS_PM_TRANSITION
These flags aren't -supposed- to do much in GSP, they're mostly used by
OpenRM itself for state tracking. The only thing they do from GSP's side is
control whether or not a single display related workaround is applied
during suspend.
But as it turns out, that single workaround is actually quite crucial for
getting runtime PM working with nouveau - and without it set we end up
seeing a lot more failures with runtime PM resume. So, let's start setting
it.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org> # v6.16+
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patch.msgid.link/20260814194542.781955-4-lyude@redhat.com
|
|
Previously, it looked as if we were able to fix suspend/resume on some
desktops by setting Gcoff based on whether or not we were entering runtime
PM. This was a mistake though - the only time suspend/resume would end up
actually working was if Gcoff = 0.
It seems like it's likely the main reason for this is the FBSR GcOff
argument actually controls GSP's behavior with regards to which buffers it
decides to save across suspend/resume. When GcOff = 1, RM reserved regions
are saved unless they are marked as LOST_ON_SUSPEND, and RM channel-context
and kernel-client buffers are also saved -including- when they are
LOST_ON_SUSPEND. This means with GcOff = 1, we end up having GSP save and
restore buffers that actually need to be reinitialized on resume - causing
the failures we're setting.
Thanks to John Hubbard from Nvidia for providing some background on what
these options do in the GSP firmware do!
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org> # v6.16+
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patch.msgid.link/20260814194542.781955-3-lyude@redhat.com
|
