linux-2.6-microblaze.git
5 weeks agoMerge tag 'amd-drm-next-7.3-2026-08-27' of https://gitlab.freedesktop.org/agd5f/linux...
Dave Airlie [Fri, 28 Aug 2026 03:27:51 +0000 (13:27 +1000)]
Merge tag 'amd-drm-next-7.3-2026-08-27' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-7.3-2026-08-27:

amdgpu:
- MES fixes
- Userq fixes
- UVD fix
- VCE 3 fix
- Enforce isolation fix
- HPD fix for VGA/LVDS
- DML fix
- DCN 6 fixes
- DC gpu reset fix

amdkfd:
- CU occupancy for GFX 11
- CU occupancy for GFX 12/12.1
- Queue bounds checking fix
- SVM fixes
- CRIU bounds checking fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patch.msgid.link/20260827203610.3249084-1-alexander.deucher@amd.com
5 weeks agoMerge tag 'drm-xe-next-fixes-2026-08-27' of https://gitlab.freedesktop.org/drm/xe...
Dave Airlie [Fri, 28 Aug 2026 03:14:31 +0000 (13:14 +1000)]
Merge tag 'drm-xe-next-fixes-2026-08-27' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-next

Cross-subsystem Changes:
- i2c global register definitions as dependency for xe/i2c fixes. (Heikki)

Driver Changes:
- Media workardound (Daniele)
- Add CCS to gt_idle debugfs print (Bala)
- Page fault related fix (Arvind)
- i2c related fixes (Heikki)
- System Controller mailbox bit fix (Anoop)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/apBrVgvZwIRIfuhR@intel.com
5 weeks agodrm/xe/sysctrl: Read mailbox phase bit from hardware
Anoop Vijay [Tue, 25 Aug 2026 17:28:28 +0000 (10:28 -0700)]
drm/xe/sysctrl: Read mailbox phase bit from hardware

The mailbox PHASE bit in SYSCTRL_MB_CTRL is toggled per-message and
was tracked in software as sc->phase_bit, reset to 0 on error paths.
If the cached value ever drifts from what the hardware last saw, all
following messages carry the wrong phase and get silently misread by
firmware.

Drop the cache and read PHASE directly from SYSCTRL_MB_CTRL before
each frame instead, removing xe_sysctrl_mailbox_init() and its call
sites along with it.

Fixes: 1f95f618182b ("drm/xe/xe_sysctrl: Add System Controller mailbox communication support")
Signed-off-by: Anoop Vijay <anoop.c.vijay@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/20260825172827.3801591-2-anoop.c.vijay@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 04984fcdbf6876c940c01026a7404c1e9cc91ba7)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/xe/i2c: Keep the i2c controller always enabled
Heikki Krogerus [Tue, 11 Aug 2026 12:10:08 +0000 (14:10 +0200)]
drm/xe/i2c: Keep the i2c controller always enabled

Some platforms make an assumption that the i2c controller's
enabled state indicates also the power state of the
controller. This can create a problem when the controller is
in disabled state, because the hardware may assume
incorrectly that it is then also in low-power state.

To fix this, the controller is kept enabled by taking over
the IC_ENABLE register. The controller has to be disabled
when the configuration is updated and when the target
address or the slave address are assigned, so disabling it
when IC_CON, IC_TAR or IC_SAR registers are programmed, and
then re-enabling it again.

Fixes: f0e53aadd702 ("drm/xe: Support for I2C attached MCUs")
Cc: stable@vger.kernel.org
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/20260811121008.1493015-4-heikki.krogerus@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 76cc14e2faed1adae20f4ee144ead0e3a7566c49)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/xe/i2c: Fix the interrupt handling
Heikki Krogerus [Tue, 11 Aug 2026 12:10:07 +0000 (14:10 +0200)]
drm/xe/i2c: Fix the interrupt handling

The platforms that support the interrupt from the I2C
adapter can not handle the amount of interrupts the adapter
generates because of the way the IRQ is routed in the
hardware. The I2C controller driver has to be kept in
polling mode because of that.

The AMC MCU can still generate critical alerts that have to
be handled. The interrupt from SMBus Alert is left enabled
and handled separately in the Xe. The alerts from the AMC
will cause the device to be declared wedged for now.

Fixes: f0e53aadd702 ("drm/xe: Support for I2C attached MCUs")
Cc: stable@vger.kernel.org
Reviewed-by: Raag Jadav <raag.jadav@intel.com>
Co-developed-by: Ramesh Babu B <ramesh.babu.b@intel.com>
Signed-off-by: Ramesh Babu B <ramesh.babu.b@intel.com>
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Link: https://patch.msgid.link/20260811121008.1493015-3-heikki.krogerus@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit a55b76b8bc2c49b11d753c1c6d06ec3a2c61c85e)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agoi2c: designware: Global register definitions
Heikki Krogerus [Tue, 11 Aug 2026 12:10:06 +0000 (14:10 +0200)]
i2c: designware: Global register definitions

Moving the register definitions to a global header file
include/linux/designware_i2c.h. That removes the need to
duplicate them in the adaptation layers for this driver
outside of drivers/i2c/busses/. There is at least one of
those in drivers/gpu/drm/xe/xe_i2c.c.

Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Suggested-by: Raag Jadav <raag.jadav@intel.com>
Reviewed-by: Raag Jadav <raag.jadav@intel.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://patch.msgid.link/20260811121008.1493015-2-heikki.krogerus@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 2ab2fb31411a494e4579dfacda986a2672f80e65)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/xe: Reject page faults from non-fault-mode scratch VMs
Arvind Yadav [Thu, 20 Aug 2026 06:54:45 +0000 (12:24 +0530)]
drm/xe: Reject page faults from non-fault-mode scratch VMs

Having scratch enabled does not make a VM capable of handling recoverable
page faults. Allowing scratch VMs through the ASID lookup also admits
dma-fence mode VMs.

If such a VM faults on an already valid VMA, the handler reports success
without fixing the fault, causing the GPU to retry indefinitely.

Only allow fault-mode VMs through the ASID lookup. Fault-mode VMs using
scratch remain supported, while faults from 3D VMs are rejected.

Fixes: ad9843aac91a ("drm/xe/madvise: Implement purgeable buffer object support")
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260820065445.567228-1-arvind.yadav@intel.com
(cherry picked from commit bfb24a06405b652d37831f3fb66b71d33a6605de)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/xe/xe_gt_idle: Add CCS to the powergating info print
Balasubramani Vivekanandan [Wed, 19 Aug 2026 07:34:58 +0000 (13:04 +0530)]
drm/xe/xe_gt_idle: Add CCS to the powergating info print

While reading the main GT powergating info from debugfs, include both
RCS and CCS engine masks.

Fixes: 0914c1e45d3a1 ("drm/xe/xe_gt_idle: add debugfs entry for powergating info")
Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patch.msgid.link/20260819073457.1812722-2-balasubramani.vivekanandan@intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
(cherry picked from commit 8899e413c5ab85443ec9bbc50cffe924c6b596de)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/xe: Do not apply WA 14025883347 to media 3503
Daniele Ceraolo Spurio [Tue, 18 Aug 2026 21:35:20 +0000 (21:35 +0000)]
drm/xe: Do not apply WA 14025883347 to media 3503

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
(cherry picked from commit fae59d5de5de39bc51ac2839f74970312e0c8905)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 weeks agodrm/amd/display: fix dc_lock leak on GPU reset error paths
Linkai Gong [Wed, 19 Aug 2026 05:47:42 +0000 (13:47 +0800)]
drm/amd/display: fix dc_lock leak on GPU reset error paths

On GPU reset, dm_suspend() takes dc_lock and leaves it for dm_resume()
to drop. If amdgpu_dm_commit_zero_streams() or dm_dmub_hw_init() fails,
the function returns with the lock still held. The matching resume path
is then skipped, so every later dc_lock take hangs.

Release the cached DC state and unlock before returning the error.

Fixes: 3cf7a0bc87f0 ("drm/amd/display: Catch failures for amdgpu_dm_commit_zero_streams()")
Fixes: 2b6943df5413 ("drm/amd/display: Pass up errors for reset GPU that fails to init HW")
Cc: stable@vger.kernel.org
Signed-off-by: Linkai Gong <gonglinkai@kylinos.cn>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amd/display: Fix redundant GPUVMEnable checks in dcn6 flip schedule
Srinivasan Shanmugam [Thu, 20 Aug 2026 10:16:01 +0000 (15:46 +0530)]
drm/amd/display: Fix redundant GPUVMEnable checks in dcn6 flip schedule

Inside dcn6_calculate_flip_schedule(), GPUVMEnable is already checked in
the outer if block. But the same GPUVMEnable is checked again in two
inner if blocks inside it. Since GPUVMEnable is always true at that
point, the inner else branches that assign meta_row_height are never
reached.

Remove the redundant inner GPUVMEnable checks and directly assign
dpte_row_height, which is always the correct value here.

Fixes: 7f7d7ea1fa51 ("drm/amd/display: Add new sources for DCN6")
Reported-by: Dan Carpenter <error27@gmail.com>
Cc: Roman Li <roman.li@amd.com>
Cc: Alex Hung <alex.hung@amd.com>
Cc: Tom Chung <chiahsuan.chung@amd.com>
Cc: Aurabindo Pillai <aurabindo.pillai@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: George Zhang <george.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amd/display: Fix wrong bytes-per-pixel value for dml2_422_packed_10
Srinivasan Shanmugam [Thu, 20 Aug 2026 09:59:43 +0000 (15:29 +0530)]
drm/amd/display: Fix wrong bytes-per-pixel value for dml2_422_packed_10

The pixel format dml2_422_packed_10 needs BytePerPixelDETY set to
8.0/3. But it was accidentally placed in the wrong group that sets it to
  4, so the correct value was never used.

This caused wrong DET buffer size and bandwidth calculations whenever
this format was used.

Fix it by moving dml2_422_packed_10 out of the wrong group so it gets
the correct value of 8.0/3.

Fixes: 7f7d7ea1fa51 ("drm/amd/display: Add new sources for DCN6")
Reported-by: Dan Carpenter <error27@gmail.com>
Cc: Roman Li <roman.li@amd.com>
Cc: Alex Hung <alex.hung@amd.com>
Cc: Tom Chung <chiahsuan.chung@amd.com>
Cc: Aurabindo Pillai <aurabindo.pillai@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: George Zhang <george.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdkfd: guard against NULL restore_mqd in CRIU queue restore
Vladimir Marioukhine [Wed, 12 Aug 2026 16:58:12 +0000 (12:58 -0400)]
drm/amdkfd: guard against NULL restore_mqd in CRIU queue restore

Both create_queue_cpsch() and create_queue_nocpsch() unconditionally
call mqd_mgr->restore_mqd() when a CRIU restore is in progress
(qd != NULL), with no NULL guard. On any system where restore_mqd is
not implemented for the given queue type, a user holding
CAP_CHECKPOINT_RESTORE can trigger a kernel NULL pointer dereference
and panic the machine by issuing KFD_IOC_CRIU_OP_RESTORE with a
crafted queue restore object. Note that checkpoint_mqd is likewise
unimplemented on GFX12, so no legitimate CRIU image can reach this
path — only a hand-crafted restore payload.

Add a NULL guard for restore_mqd immediately after mqd_mgr is
resolved, unwinding via the existing error labels and returning
-EOPNOTSUPP if the callback is not implemented. This mirrors the
existing checkpoint_mqd guard in checkpoint_mqd().

Fixes: 48f0bdf4e38e ("drm/amdkfd: Added MQD manager files for GFX12.")
Cc: stable@vger.kernel.org
Signed-off-by: Vladimir Marioukhine <Vladimir.Marioukhine@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdgpu/userq: fix lock missing for userq fence error set
Prike Liang [Fri, 7 Aug 2026 01:46:22 +0000 (09:46 +0800)]
drm/amdgpu/userq: fix lock missing for userq fence error set

amdgpu_userq_fence_driver() and amdgpu_userq_fence_driver_destroy()
don't acquire the dma_fence spinlock, so locking the dma_fence lock
before test the signaled state and set error state to avoid missing
lock assert error.

Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdkfd: Fix the case that vm range is hole at svm_migrate_copy_to_vram
Xiaogang Chen [Sun, 23 Aug 2026 20:47:15 +0000 (15:47 -0500)]
drm/amdkfd: Fix the case that vm range is hole at svm_migrate_copy_to_vram

When migration vm range is hole at cpu side(MIGRATE_PFN_MIGRATE set +
MIGRATE_PFN_VALID unset) driver still allocates device pages. There is no
dma map of src pages and migration. j is 0 and svm_migrate_copy_memory_gart()
will return an uninitialized r. That can trigger out_free_vram_pages to drop
all VRAM just set up.

Initialize r and only call the last svm_migrate_copy_memory_gart if j > 0.

Current code postponed the last page to the final copy. This patch flushes on
the last page when reach to the end of current drm_buddy_block; avoids another
svm_migrate_copy_memory_gart.

Cc: stable@vger.kernel.org
Signed-off-by: Xiaogang Chen <xiaogang.chen@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdkfd: Fix error path at svm_migrate_copy_to_ram
Xiaogang Chen [Sun, 23 Aug 2026 20:22:54 +0000 (15:22 -0500)]
drm/amdkfd: Fix error path at svm_migrate_copy_to_ram

If page migration from device to sys ram fails for some reasons driver needs
release and unlock allocated system pages. To do that driver should use page
physical address, or pfn, then get struct page*. Current driver uses dma
address(for adev) that is not correct with IOMMU enabled, or even in general.

The patch releases and unlocks allocated system pages based on where migration
failed by struct page* of sys ram pages. Also dma_unmap correspodent system
ram pages at error path.

Cc: stable@vger.kernel.org
Signed-off-by: Xiaogang Chen <xiaogang.chen@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amd/display: Log details when failing to register HPD IRQ
Timur Kristóf [Fri, 21 Aug 2026 21:50:59 +0000 (23:50 +0200)]
drm/amd/display: Log details when failing to register HPD IRQ

This should help diagnose HPD IRQ related issues in the future.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Link: https://patch.msgid.link/20260821215059.312868-2-timur.kristof@gmail.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amd/display: Fix HPD consideration for VGA/LVDS connectors on DCE
Timur Kristóf [Fri, 21 Aug 2026 21:50:58 +0000 (23:50 +0200)]
drm/amd/display: Fix HPD consideration for VGA/LVDS connectors on DCE

After a refactor that landed in Linux 7.0,
DC now crashes when it is initialized on GPUs
that have a VGA or LVDS connector. This is because
these connectors have no HPD so the hpd_gpio is NULL
and therefore DC takes the code path meant for
DCN 4.2+ which sets irq_source_hpd = 255 that
causes the subsequent code to try to register
the HPD interrupt, which fails, and causes
a crash.

This commit should be backported to Linux 7.0 and newer.

Cc: stable@vger.kernel.org
Cc: Dmytro Laktyushkin <dmytro.laktyushkin@amd.com>
Cc: Roman Li <roman.li@amd.com>
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/5490
Fixes: def3488eb0fd ("drm/amd/display: refactor HPD to increase flexibility")
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Link: https://patch.msgid.link/20260821215059.312868-1-timur.kristof@gmail.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdgpu: clamp the isolation index for rings outside a partition
Xiang Liu [Fri, 21 Aug 2026 09:41:57 +0000 (17:41 +0800)]
drm/amdgpu: clamp the isolation index for rings outside a partition

adev->isolation[] has one slot per partition, but a ring that is not
assigned to one keeps AMDGPU_XCP_NO_PARTITION, which is ~0, so indexing
the array with it is out of bounds. SDMA submissions hit this on both
the isolation enforcement and the VM flush path and trip UBSAN.

Fall back to the first slot the way the cleaner shader path already
does, and stop taking the address before the ring type check that makes
it relevant.

Cc: stable@vger.kernel.org
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdkfd: Reject zero-sized AQL queue allocations after size halving
Sunday Clement [Thu, 6 Aug 2026 14:59:34 +0000 (10:59 -0400)]
drm/amdkfd: Reject zero-sized AQL queue allocations after size halving

KFD_IOC_ALLOC_MEMORY_OF_GPU with flag
KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM and size=1 triggers the AQL
wraparound workaround (size >>= 1), reducing size to 0. The resulting
zero passes through PAGE_ALIGN(0) = 0 without validation, bypassing the
per-process VRAM quota check in reserve_mem_limit()
(vram_used + 0 > vram_available is always false).

The fix adds post-halving zero-size validation in the primary
allocation path (amdgpu_amdkfd_gpuvm.c). The check happens after size
halving but before reserve_mem_limit(), and uses err_alignment_size
error path to properly clean up the allocated kgd_mem structure and
mutex.

Cc: stable@vger.kernel.org
Signed-off-by: Sunday Clement <Sunday.Clement@amd.com>
Reviewed-by: Alex Deucher <Alexander.Deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdgpu: Fix VCE 3 ring align_mask
David Rosca [Thu, 20 Aug 2026 14:32:51 +0000 (16:32 +0200)]
drm/amdgpu: Fix VCE 3 ring align_mask

The largest frame is 20 dwords, so 0xf mask is too small.
This was always wrong, but we were lucky with the VCE_CMD_END
commands inserted after fence and vm_flush.

Fixes: 8897ea8c761b ("drm/amdgpu: Implement insert_end for VCE 3")
Cc: stable@vger.kernel.org
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: David Rosca <david.rosca@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/kfd: Add CU occupancy support to GFX12.1
David Belanger [Fri, 14 Aug 2026 17:26:17 +0000 (13:26 -0400)]
drm/kfd: Add CU occupancy support to GFX12.1

Port changes from GFX9 to GFX12.1 mostly as-is.
Minor changes to register access code.

Assisted-by: Claude:Sonnet 4.6
Signed-off-by: David Belanger <david.belanger@amd.com>
Reviewed-by: Sreekant Somasekharan <Sreekant.Somasekharan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/kfd: Add CU occupancy support to GFX12
David Belanger [Fri, 14 Aug 2026 17:25:56 +0000 (13:25 -0400)]
drm/kfd: Add CU occupancy support to GFX12

Port changes from GFX9 to GFX12 mostly as-is.
Minor changes to register access code.

Assisted-by: Claude:Sonnet-4-6
Signed-off-by: David Belanger <david.belanger@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Sreekant Somasekharan <Sreekant.Somasekharan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/kfd: Add CU occupancy support to GFX11
David Belanger [Fri, 14 Aug 2026 16:57:18 +0000 (12:57 -0400)]
drm/kfd: Add CU occupancy support to GFX11

Port changes from GFX9 to GFX11 mostly as-is.
Minor changes to register access code.

Assisted-by: Claude:Sonnet-4-6
Signed-off-by: David Belanger <david.belanger@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Sreekant Somasekharan <Sreekant.Somasekharan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdgpu: avoid force-completing uninitialized UVD rings
Bob Zhou [Wed, 19 Aug 2026 06:23:49 +0000 (14:23 +0800)]
drm/amdgpu: avoid force-completing uninitialized UVD rings

uvd_v7_0_sw_init() does not initialize the UVD decode ring for an
SR-IOV VF. However, amdgpu_uvd_resume() unconditionally force-completes
the decode ring when restoring its fence sequence.

Skip fence completion when the fence driver is not initialized.

Fixes: 0a33b11d26c6 ("drm/amdgpu: mark force completed fences with -ECANCELED")
Cc: stable@vger.kernel.org
Signed-off-by: Bob Zhou <bobzhou2@amd.com>
Acked-by: Leo Liu <leo.liu@amd.com>
Acked-by: Frank Min <Frank.Min@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
5 weeks agodrm/amdgpu/userq: lock and validate wptr BOs before reading their GPU offset on restore
Jesse Zhang [Fri, 14 Aug 2026 07:01:44 +0000 (15:01 +0800)]
drm/amdgpu/userq: lock and validate wptr BOs before reading their GPU offset on restore

On resume, amdgpu_userq_vm_validate_and_restore_queue() updates each queue's
wptr GPU address via amdgpu_bo_gpu_offset().

WPTR BOs are VM-mapped, but each BO has its own reservation object and is not
implicitly covered by the VM validation path here. This can leave offset reads
without proper BO locking/placement state and trigger WARN_ONs.
  ------------[ cut here ]------------
  WARNING: amdgpu_object.c:1486 at amdgpu_bo_gpu_offset+0x75/0xa0 [amdgpu], CPU#3: kworker/3:1/116
  Workqueue: events amdgpu_userq_restore_worker [amdgpu]
  RIP: 0010:amdgpu_bo_gpu_offset+0x75/0xa0 [amdgpu]
  Call Trace:
   <TASK>
   amdgpu_userq_vm_validate_and_restore_queue+0x629/0x960 [amdgpu]
   amdgpu_userq_restore_worker+0xa6/0x180 [amdgpu]
   process_scheduled_works+0xa6/0x460
   worker_thread+0x13c/0x290
   kthread+0xfb/0x140
   ret_from_fork+0x1b6/0x2b0
   ret_from_fork_asm+0x1a/0x30
   </TASK>
  ---[ end trace 0000000000000000 ]---
  ------------[ cut here ]------------
  WARNING: amdgpu_object.c:1485 at amdgpu_bo_gpu_offset+0x9a/0xa0 [amdgpu], CPU#2: kworker/2:1/127
  Workqueue: events amdgpu_userq_restore_worker [amdgpu]
  RIP: 0010:amdgpu_bo_gpu_offset+0x9a/0xa0 [amdgpu]

Add each queue's WPTR BO to the drm_exec ww context and validate it to its
allowed placement before the later offset update.

v2:
- Clarify that WPTR BOs are VM-mapped (fix incorrect "not part of VM" wording). (Christian)
- Describe both parts of the fix: lock BO reservations in drm_exec and
  validate BO placement before offset reads.

Acked-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>
5 weeks agodrm/amdgpu/mes: fix the inconsistent indenting for mes_userq_map()
Prike Liang [Wed, 19 Aug 2026 07:23:56 +0000 (15:23 +0800)]
drm/amdgpu/mes: fix the inconsistent indenting for mes_userq_map()

Fix the inconsistent indenting warning for mes_userq_map().

Fixes: d0827dda8fa7 ("drm/amdgpu/mes: refactor the amdgpu_mes_alloc/free_proc|gang()")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202608190252.8XCa0HqR-lkp@intel.com/
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 weeks agoMerge tag 'drm-xe-next-fixes-2026-08-20' of https://gitlab.freedesktop.org/drm/xe...
Dave Airlie [Mon, 24 Aug 2026 01:40:14 +0000 (11:40 +1000)]
Merge tag 'drm-xe-next-fixes-2026-08-20' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-next

Couple commits cleaning up error messages.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/aocSC19By5UIIbTZ@intel.com
6 weeks agoMerge tag 'drm-msm-next-2026-08-01' of https://gitlab.freedesktop.org/drm/msm into...
Dave Airlie [Mon, 24 Aug 2026 01:21:11 +0000 (11:21 +1000)]
Merge tag 'drm-msm-next-2026-08-01' of https://gitlab.freedesktop.org/drm/msm into drm-next

Changes for v7.3

Bindings:
- Added Shikra support
- Document a840, a704, a722

Core:
- Use drm_client buffers for fbdev emulation
- teardown fixes
- ARM32 DMA fixup
- Remove objects from evict list when re-validated
- Bunch of corner case and error path fixes

DPU:
- Dropped dev_pm_opp_set_rate(0) preventing burnout
- Fixed SSPP offsets of Kaanapali

DP:
- Dropped dev_pm_opp_set_rate(0) preventing burnout
- Cleaned up core code in preparation for MST support
- Fixed prepare() to let Pipewire continue in case of the unplugged cable

GPU:
- Add support for a704
- Add support for a722

HDMI:
- Simplifed register access

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <rob.clark@oss.qualcomm.com>
Link: https://patch.msgid.link/CACSVV02Kp=J+w_RjEJbBbQnBYRb+SWdwMvVbCaAL70bq9EBagQ@mail.gmail.com
6 weeks agoMerge tag 'drm-misc-next-fixes-2026-08-20' of https://gitlab.freedesktop.org/drm...
Dave Airlie [Mon, 24 Aug 2026 01:19:43 +0000 (11:19 +1000)]
Merge tag 'drm-misc-next-fixes-2026-08-20' of https://gitlab.freedesktop.org/drm/misc/kernel into drm-next

One patch to remove a WARN in favour of drm_warn.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maxime Ripard <mripard@redhat.com>
Link: https://patch.msgid.link/20260820-fabulous-nimble-woodpecker-ffa695@houat
6 weeks agoMerge tag 'amd-drm-next-7.3-2026-08-19' of https://gitlab.freedesktop.org/agd5f/linux...
Dave Airlie [Sun, 23 Aug 2026 23:07:41 +0000 (09:07 +1000)]
Merge tag 'amd-drm-next-7.3-2026-08-19' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-7.3-2026-08-19:

amdgpu:
- eGPU fixes
- Runtime PM fix
- UserQ fixes
- Backlight fix
- Discovery sysfs fix
- Reset handling fixes
- Buffer func handling fix for xgmi
- VCN boundary check fix
- DC lut handling fixes

amdkfd:
- Fix return value

radeon:
- iMac display fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patch.msgid.link/20260819183622.2406038-1-alexander.deucher@amd.com
6 weeks agodrm/xe: don't WARN on kernel job timeout when device already wedged
Nitin Gote [Fri, 14 Aug 2026 07:41:07 +0000 (13:11 +0530)]
drm/xe: don't WARN on kernel job timeout when device already wedged

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>
6 weeks agoBackMerge tag 'v7.2' into drm-next
Dave Airlie [Thu, 20 Aug 2026 00:58:44 +0000 (10:58 +1000)]
BackMerge tag 'v7.2' into drm-next

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>
6 weeks agodrm/amd/display: validate plane degamma LUT size for private color prop
Harry Wentland [Tue, 4 Aug 2026 21:04:05 +0000 (17:04 -0400)]
drm/amd/display: validate plane degamma LUT size for private color prop

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>
6 weeks agodrm/amd/display: avoid divide-by-zero in __is_lut_linear()
Harry Wentland [Tue, 4 Aug 2026 21:04:04 +0000 (17:04 -0400)]
drm/amd/display: avoid divide-by-zero in __is_lut_linear()

__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>
6 weeks agodrm/amdgpu: handle pipeline sync without a VM fence
Alex Deucher [Fri, 14 Aug 2026 15:39:16 +0000 (11:39 -0400)]
drm/amdgpu: handle pipeline sync without a VM fence

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>
6 weeks agodrm/amdgpu/vcn: fix integer overflow in dec_msg buffer count check
David (Ming Qiang) Wu [Fri, 7 Aug 2026 19:12:14 +0000 (15:12 -0400)]
drm/amdgpu/vcn: fix integer overflow in dec_msg buffer count check

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>
6 weeks agodrm/radeon: fix internal display on iMac11, 1 (RV770/DCE3.1)
Gilles Risch [Mon, 17 Aug 2026 00:43:26 +0000 (02:43 +0200)]
drm/radeon: fix internal display on iMac11, 1 (RV770/DCE3.1)

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>
6 weeks agodrm/amdgpu: fix hang and race in userq destroy
Zhu Lingshan [Fri, 14 Aug 2026 08:37:01 +0000 (16:37 +0800)]
drm/amdgpu: fix hang and race in userq destroy

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>
6 weeks agoamdkfd: let profile_lock_device return an int other than uint32
Zhu Lingshan [Fri, 14 Aug 2026 06:51:32 +0000 (14:51 +0800)]
amdkfd: let profile_lock_device return an int other than uint32

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>
6 weeks agodrm/amdgpu: delay ttm buffer func enablement on xgmi
Pierre-Eric Pelloux-Prayer [Mon, 15 Jun 2026 11:48:23 +0000 (13:48 +0200)]
drm/amdgpu: delay ttm buffer func enablement on xgmi

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>
6 weeks agodrm/amdgpu: check thunderbolt before switcheroo registration
Yang Wang [Wed, 12 Aug 2026 03:08:40 +0000 (11:08 +0800)]
drm/amdgpu: check thunderbolt before switcheroo registration

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>
6 weeks agodrm/amdgpu: force complete the KIQ ring fences on reset
Jesse Zhang [Thu, 13 Aug 2026 04:28:04 +0000 (12:28 +0800)]
drm/amdgpu: force complete the KIQ ring fences on reset

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>
6 weeks agodrm/amdgpu: validate rptr and wptr of a userq
Zhu Lingshan [Thu, 13 Aug 2026 09:36:53 +0000 (17:36 +0800)]
drm/amdgpu: validate rptr and wptr of a userq

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>
6 weeks agodrm/amdgpu: force complete the MES ring fences on reset
Jesse Zhang [Thu, 13 Aug 2026 04:27:53 +0000 (12:27 +0800)]
drm/amdgpu: force complete the MES ring fences on reset

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>
6 weeks agodrm/amdgpu: fix sysfs ip base addr for 64bit in standalone mode
Mukul Joshi [Thu, 13 Aug 2026 15:02:24 +0000 (11:02 -0400)]
drm/amdgpu: fix sysfs ip base addr for 64bit in standalone mode

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>
6 weeks agodrm/amd/display: Scale custom brightness curve from full range
Akhmed Zhitaev [Thu, 13 Aug 2026 17:09:59 +0000 (22:09 +0500)]
drm/amd/display: Scale custom brightness curve from full range

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>
6 weeks agodrm/amdgpu/userq: ignore duplicate BO locks when returning wait fence info
Sunil Khatri [Thu, 13 Aug 2026 08:32:35 +0000 (14:02 +0530)]
drm/amdgpu/userq: ignore duplicate BO locks when returning wait fence info

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>
6 weeks agodrm/amdgpu/userq: ignore duplicate BO locks when counting wait fences
Sunil Khatri [Thu, 13 Aug 2026 08:32:09 +0000 (14:02 +0530)]
drm/amdgpu/userq: ignore duplicate BO locks when counting wait fences

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>
6 weeks agodrm/amdgpu/userq: ignore duplicate BO locks in userq signal ioctl
Sunil Khatri [Thu, 13 Aug 2026 08:29:07 +0000 (13:59 +0530)]
drm/amdgpu/userq: ignore duplicate BO locks in userq signal ioctl

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>
6 weeks agodrm/amdgpu: ensure all userq VAs mapped before restore
Zhu Lingshan [Wed, 22 Jul 2026 10:13:44 +0000 (18:13 +0800)]
drm/amdgpu: ensure all userq VAs mapped before restore

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>
6 weeks agodrm/amdgpu: fix autosuspend cleanup during removal
Guangshuo Li [Sat, 8 Aug 2026 12:09:34 +0000 (20:09 +0800)]
drm/amdgpu: fix autosuspend cleanup during removal

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>
6 weeks agodrm/amdgpu: Disable runtime PM for externally attached dGPUs
Yang Wang [Wed, 5 Aug 2026 03:44:54 +0000 (11:44 +0800)]
drm/amdgpu: Disable runtime PM for externally attached dGPUs

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>
6 weeks agodrm/xe: tests: fix error message in xe_migrate_sanity_test()
Dan Carpenter [Thu, 13 Aug 2026 07:09:47 +0000 (10:09 +0300)]
drm/xe: tests: fix error message in xe_migrate_sanity_test()

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>
7 weeks agoLinux 7.2 master
Linus Torvalds [Sun, 16 Aug 2026 21:32:26 +0000 (14:32 -0700)]
Linux 7.2

7 weeks agoMerge tag 'sched_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 18:15:23 +0000 (11:15 -0700)]
Merge tag 'sched_urgent_for_v7.2' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Borislav Petkov:

 - Make sure a delayed sched entity's runtime stats are updated at the
   right time so that it receives the proper lag compensation

* tag 'sched_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: Update time before requeueing delayed entities

7 weeks agoMerge tag 'timers_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 18:12:13 +0000 (11:12 -0700)]
Merge tag 'timers_urgent_for_v7.2' of git://git./linux/kernel/git/tip/tip

Pull timer fixes from Borislav Petkov:

 - Detect a broken EL2 virtual timer in the bcm2712 SoC boards (RPi5)
   and fallback to the physical one instead

 - Fix a build error with ARM rpc_defconfig and function tracer enabled

* tag 'timers_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  clocksource/drivers/arm_arch_timer: Workaround bcm2712 broken EL2 virtual timer
  tick: Include ktime.h and jiffies.h in linux/tick.h

7 weeks agoMerge tag 'core_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 18:09:37 +0000 (11:09 -0700)]
Merge tag 'core_urgent_for_v7.2' of git://git./linux/kernel/git/tip/tip

Pull rseq fix from Borislav Petkov:

 - Prevent a lockup when rseq grants a timeslice extension

* tag 'core_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  rseq: Prevent hard lockup on granted time slice extension

7 weeks agowifi: mt76: mt7921: refactor regd update to fix recursive mutex deadlock
Charlie-cy Wu [Mon, 29 Jun 2026 08:35:43 +0000 (16:35 +0800)]
wifi: mt76: mt7921: refactor regd update to fix recursive mutex deadlock

Split mt7921_mcu_regd_update() into two functions to prevent recursive
mutex acquisition. Introduce __mt7921_mcu_regd_update() as the internal
implementation that assumes the mutex is already held by the caller,
while mt7921_mcu_regd_update() remains as the external interface that
handles mutex acquisition and release.

This fixes a deadlock issue when mt7921_regd_set_6ghz_power_type() is
called with the device mutex already held. Without this change, calling
mt7921_mcu_regd_update() would attempt to acquire the same mutex again,
causing a recursive lock deadlock.

The __mt7921_mcu_regd_update() function can be safely called when the
caller has already acquired the device mutex, avoiding the deadlock
while maintaining proper synchronization for regulatory domain updates.

Fixes: dc2608cf5224 ("wifi: mt76: mt7921: refactor regulatory notifier flow")
Signed-off-by: Charlie-cy Wu <Charlie-cy.Wu@mediatek.com>
Tested-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Tested-by: Devin Wittmayer <lucid_duck@justthetip.ca>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 weeks agoRevert "i2c: designware: defer probe if child GpioInt controllers are not bound"
Linus Torvalds [Sun, 16 Aug 2026 17:40:14 +0000 (10:40 -0700)]
Revert "i2c: designware: defer probe if child GpioInt controllers are not bound"

This reverts commit 0a4bb2abc3e56d7be6e69b050c88ba52c87e22bf.

This was reported to break the touchpad on at least some Thinkpads, and
while the revert has hit the i2c tree, it hasn't hit mine.  So I'm
reverting it directly just to have this resolved for the imminent 7.2
release.

Reported-by: Thorsten Leemhuis <linux@leemhuis.info>
Link: https://lore.kernel.org/all/b4a4eadb-282f-464c-843a-19d415a34d0c@leemhuis.info/
Cc: Mario Limonciello <mario.limonciello@amd.com>
CC: Hardik Prakash <hardikprakash.official@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 weeks agoMerge tag 'perf_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 17:31:05 +0000 (10:31 -0700)]
Merge tag 'perf_urgent_for_v7.2' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Borislav Petkov:

 - Prevent the use of exited events as group leaders

 - Avoid use-after-free of an event's group leader by promoting detached
   sibling events to standalone entities and correct related accounting
   and state transitions

* tag 'perf_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/core: Fix group leader use-after-free after sibling detach
  perf: Reject exited events as group leaders

7 weeks agoMerge tag 'x86_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 17:28:31 +0000 (10:28 -0700)]
Merge tag 'x86_urgent_for_v7.2' of git://git./linux/kernel/git/tip/tip

Pull x86 fix from Borislav Petkov:

 - Add a proper kernel cmdline option to control the TLB invalidation
   method on x86 prompted mainly by a recent finding on AMD related to
   INVLPGB/TYLBSYNC invalidations.

   Having the command line option is simply another way to alleviate
   the situation short-term

* tag 'x86_urgent_for_v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/CPU: Add a tlbi= cmdline switch

7 weeks agoMerge tag 'block-7.2-20260815' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe...
Linus Torvalds [Sun, 16 Aug 2026 14:00:40 +0000 (07:00 -0700)]
Merge tag 'block-7.2-20260815' of git://git./linux/kernel/git/axboe/linux

Pull block fix from Jens Axboe:
 "A single fix for a regression in this cycle, where drbd would leak
  shared secrets over netlink. This restores the behavior to match
  what we had before"

* tag 'block-7.2-20260815' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux:
  drbd: don't leak the shared secret to unprivileged netlink dumps

7 weeks agoMerge tag 'io_uring-7.2-20260815' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 16 Aug 2026 13:58:27 +0000 (06:58 -0700)]
Merge tag 'io_uring-7.2-20260815' of git://git./linux/kernel/git/axboe/linux

Pull io_uring fix from Jens Axboe:
 "Just a single fix for a potential issue on 32-bit x86 with PAE"

* tag 'io_uring-7.2-20260815' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux:
  io_uring/rsrc: reject overflowing regvec bvec byte counts

7 weeks agodrbd: don't leak the shared secret to unprivileged netlink dumps
Christoph Böhmwalder [Fri, 14 Aug 2026 15:16:17 +0000 (17:16 +0200)]
drbd: don't leak the shared secret to unprivileged netlink dumps

The conversion to explicit netlink serialization dropped the
exclude_sensitive parameter from net_conf_to_skb(), so each caller has
to sanitize by hand. Two dump paths were missed:
drbd_nl_get_connections_dumpit() and the volume-less connection branch
of get_one_status(). Neither op carries GENL_ADMIN_PERM, so any
unprivileged local user could read the CRAM-HMAC secret.

Add a net_conf_to_skb_sanitized() wrapper and route all three callers
through it.

Fixes: 8098eeb693c4 ("drbd: replace genl_magic with explicit netlink serialization")
Reported-by: Vivek Parikh <vivek.parikh@breachx.ai>
Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://patch.msgid.link/20260814151617.73752-1-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
7 weeks agoMerge tag 'soc-fixes-7.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Sat, 15 Aug 2026 15:36:26 +0000 (08:36 -0700)]
Merge tag 'soc-fixes-7.2-3' of git://git./linux/kernel/git/soc/soc

Pull SoC fixes from Arnd Bergmann:
 "These are three last-minute fixes for the 7.2 release, though nothing
  alarming:

   - one error handling fix for optee firmware

   - incorrect i2c data for the apple M3 that was added in 7.2

   - a boot time warning fix for nvidia tegra"

* tag 'soc-fixes-7.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  arm64: tegra: Add EL2 virtual timer interrupt for Tegra194
  arm64: dts: apple: t8122: Fix I2C resources
  optee: ffa: Add NULL check in optee_ffa_lend_protmem

7 weeks agoMerge tag 'for-linus' of https://github.com/openrisc/linux
Linus Torvalds [Sat, 15 Aug 2026 15:05:58 +0000 (08:05 -0700)]
Merge tag 'for-linus' of https://github.com/openrisc/linux

Pull OpenRISC fix from Stafford Horne:
 "A bug fix found by researchers:

   - mask all privileged bits when restoring the supervisor register
     from sigreturn"

* tag 'for-linus' of https://github.com/openrisc/linux:
  openrisc: signal: do not restore privileged SR bits on sigreturn

7 weeks agoopenrisc: signal: do not restore privileged SR bits on sigreturn
Ali Ahmet Memis [Fri, 7 Aug 2026 23:42:30 +0000 (23:42 +0000)]
openrisc: signal: do not restore privileged SR bits on sigreturn

restore_sigcontext() copies the whole supervision register (SR) from the
signal frame and only clears SPR_SR_SM before the value is reloaded into
the hardware SR (through ESR and l.rfe) on the return to user space.  All
other SR bits are left under user control.

An unprivileged task can thus return from a signal handler through a
crafted sigframe that clears SPR_SR_DME.  With the data MMU disabled the
CPU performs no translation or protection on data accesses, so the task
gains read and write access to arbitrary physical memory, a local
privilege escalation.  SPR_SR_IME, SPR_SR_SUMRA, SPR_SR_LEE, SPR_SR_EPH
and the cache-enable bits are exposed the same way.  The ptrace GPR regset
already refuses any change to SR for exactly this reason.

Restore only the arithmetic flag bits (F, CY, OV) from the signal frame
and take every privileged control bit from the SR the kernel saved on
signal entry.

Verified with qemu-system-or1k -M or1k-sim: before this change an
unprivileged PoC clears SPR_SR_DME in rt_sigreturn and writes a marker to
physical address 0x03000000 (beyond the kernel's mem=32M); afterwards the
same PoC receives SIGSEGV and physical memory is unchanged.

Fixes: ac689eb7f9d4 ("OpenRISC: Signal handling")
Cc: stable@vger.kernel.org
Signed-off-by: Ali Ahmet Memis <ali@iusegentoo.com>
Signed-off-by: Stafford Horne <shorne@gmail.com>
7 weeks agoMerge tag 'input-for-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor...
Linus Torvalds [Sat, 15 Aug 2026 04:51:26 +0000 (21:51 -0700)]
Merge tag 'input-for-v7.2-rc7' of git://git./linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:

 - A couple of fixes to the sur40 touchscreen driver to correct
   registration and teardown ordering, and to fix error path
   unwinding when video device registration fails.

* tag 'input-for-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: sur40 - fix V4L error path cleanup
  Input: sur40 - fix input device registration ordering

7 weeks agoMerge tag 'drm-fixes-2026-08-15' of https://gitlab.freedesktop.org/drm/kernel
Linus Torvalds [Fri, 14 Aug 2026 22:48:05 +0000 (15:48 -0700)]
Merge tag 'drm-fixes-2026-08-15' of https://gitlab.freedesktop.org/drm/kernel

Pull drm fixes from Dave Airlie:
 "While this is large for rc8 time but also AI driven fixes is a lot of
  it, we had a more traditional screw up, and a regression was just
  found in the fair scheduling patches that went in back in rc1. This
  reverts the fair scheduler back to an option and sets the default back
  to what it should have been. We might have been a bit overly zealous
  in switching over, but at least it feels more normal than the AI
  driven fixes.

  Apart from the scheduler, it's mostly amdgpu and xe fixes, with some
  misc fixes to the log code and connector code.

  scheduler:
   - revert fair scheduler patches due to regression
   - mark fair as experimental

  connector:
   - fix OOB read in hdmi audio infoframe

  log:
   - fix divide by 0 if module param is set to 0
   - fix OOB read on empty message
   - fix infinite loop for too large scale

  xe:
   - Fix DPT Allocation paths
   - Fixes around UM queue BO
   - Order ring writes before ring tail updates
   - Add termination on resume for PXP
   - Document Sentinel and make CTX_TIMESTAMP read TOCTOU-safe
   - Fix sync entry leak on OA config emit failure
   - Check managed mutex initilization errors
   - Fix min frequency setting
   - Fix xe_device_probe error path

  amdgpu:
   - Bounds checking fix in CS IOCTL
   - Bounds checking fix in GEM IOCTL
   - Display fixes
   - GPUVM fix
   - ASPM fix
   - UVD bounds checking fixes
   - VCE 3 fix
   - BT.2020 fixes
   - NBIF 6.3.1 fix
   - IP discovery fix

  radeon:
   - Runtime pm fix

  amdxdna:
   - skip attempting to populate unmapped pages"

* tag 'drm-fixes-2026-08-15' of https://gitlab.freedesktop.org/drm/kernel: (51 commits)
  drm/log: Fix infinite loop when scale is too large for display
  drm/log: Fix out-of-bounds read on empty message length
  drm/log: Fix division by zero when scale module parameter is 0
  drm/xe: Fix xe_device_probe() failure
  drm/xe: Fix a bug in pc_adjust_freq_bounds()
  drm/xe/oa: Check managed mutex initialization errors
  drm/xe/oa: Fix sync entry leak on OA config emit failure
  drm/xe/lrc: document sentinel and make CTX_TIMESTAMP read TOCTOU-safe
  drm/xe/pxp: add termination on resume
  drm/xe: Order ring writes before ring tail updates
  drm/xe/guc_ads: use uncached mapping for UM queue BO
  drm/xe/guc_ads: allocate UM queues in VRAM on dGFX
  drm/xe/guc_ads: allocate UM queues in a separate BO
  drm/xe: Fix DPT allocation paths.
  accel/amdxdna: Skip unmapped range in aie2_populate_range()
  drm/amdgpu: Prefer default discovery offset
  drm/amdgpu: Reject UVD message with invalid number of h265 refs
  drm/amdgpu: fix nbif 6.3.1 l1 low power not functional
  drm/amd/display: fix BT.2020 YCbCr output CSC matrices for DCE
  drm/amd/display: fix BT.2020 YCbCr limited output CSC matrix
  ...

7 weeks agoMerge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 14 Aug 2026 22:34:14 +0000 (15:34 -0700)]
Merge tag 'clk-fixes-for-linus' of git://git./linux/kernel/git/clk/linux

Pull clk fixes from Stephen Boyd:
 "Fixes for the Qualcomm, Rockchip, and SpacemiT clk drivers:

   - Keep audio working on Rockchip rk3588 by skipping disabling unused
     clks

   - Fix SpacemiT USB2 clk data so they actually work and keep the HDMA
     bus clk enabled to avoid system hangs

   - Avoid clk hangs on Qualcomm Eliza display hardware and revert a
     patch that breaks PCIe on some Qualcomm platforms"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  dt-bindings: clock: Replace bouncing emails
  Revert "clk: qcom: regmap-phy-mux: Rework the implementation"
  clk: spacemit: k3: set hdma clock as critical
  clk: spacemit: k3: fix USB2 bus clock
  clk: qcom: dispcc-eliza: Fix disp_cc_mdss_mdp_clk_src RCG stall on Eliza EVK
  clk: rockchip: rk3588: don't disable unused I2S MCLK output gates

7 weeks agoMerge tag 'spi-fix-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi
Linus Torvalds [Fri, 14 Aug 2026 22:28:29 +0000 (15:28 -0700)]
Merge tag 'spi-fix-v7.2-rc7' of git://git./linux/kernel/git/broonie/spi

Pull spi fixes from Mark Brown:
 "A couple of relatively minor (but as ever important if you're hitting
  them) and straightforward driver specific fixes, plus one new device
  ID documented in the DT bindings for the DesignWare controller"

* tag 'spi-fix-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi:
  spi: virtio: mark device ready before registering the controller
  spi: dw: fix wrong RX_SAMPLE_DLY setting after resume
  spi: dt-bindings: snps,dw-apb-ssi: Document Axiado AX3005

7 weeks agoMerge tag 'regulator-fix-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 14 Aug 2026 22:22:47 +0000 (15:22 -0700)]
Merge tag 'regulator-fix-v7.2-rc7' of git://git./linux/kernel/git/broonie/regulator

Pull regulator fixes from Mark Brown:
 "There's one fix here for a data entry error in the voltage mapping in
  the fp9931 driver, and a device ID addition for a LDO in the Qualcomm
  PM8350b that's just a trivial quirk"

* tag 'regulator-fix-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator:
  regulator: fp9931: Fix VPOS/VNEG voltage selector table
  regulator: qcom-rpmh: Add support for PM8350B
  regulator: dt-bindings: qcom,rpmh: Add support for PM8350B

7 weeks agoMerge tag 'regmap-fix-v7.2-rc7-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 14 Aug 2026 21:59:04 +0000 (14:59 -0700)]
Merge tag 'regmap-fix-v7.2-rc7-2' of git://git./linux/kernel/git/broonie/regmap

Pull regmap fixes from Mark Brown:
 "A couple more fixes for regmap, this time for the SoundWire MBQ
  support:

   - Several drivers omit the readable_reg callback and it's generally
     optional in regmap but the MBQ code had an assumption that one was
     present added in one of the APIs, remove that

   - The timeout and retry intervals were swapped in read_poll_timeout()
     for soundwire-mbq"

* tag 'regmap-fix-v7.2-rc7-2' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: sdw-mbq: don't call an unset readable_reg callback
  regmap: sdw-mbq: Fix swap of timeout and retry times

7 weeks agoMerge tag 'mmc-v7.2-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 14 Aug 2026 21:46:42 +0000 (14:46 -0700)]
Merge tag 'mmc-v7.2-rc2-2' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:

 - atmel-mci: Fix use-after-free in atmci_remove due to race condition

 - loongson2: Fix sg iteration in data reorder functions

 - omap_hsmmc: Fix busy_timeout overflow in ns conversion on 32-bit

 - sdhci:
     - Make tuning_err a signed int
     - Unmap the bounce buffer before device release

* tag 'mmc-v7.2-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: loongson2: Fix sg iteration in data reorder functions
  mmc: omap_hsmmc: fix busy_timeout overflow in ns conversion on 32-bit
  mmc: atmel-mci: Fix use-after-free in atmci_remove due to race condition
  mmc: sdhci: unmap the bounce buffer before device release
  mmc: sdhci: make tuning_err a signed int

7 weeks agoMerge tag 'pmdomain-v7.2-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 14 Aug 2026 21:34:17 +0000 (14:34 -0700)]
Merge tag 'pmdomain-v7.2-rc2-2' of git://git./linux/kernel/git/ulfh/linux-pm

Pull pmdomain fixes from Ulf Hansson:

 - arm: Don't treat performance state 0 as an error

 - mediatek:
   - Fix mt8183 hang on boot
   - Fix potential null pointer dereference
   - Prevent using uninitialized data
   - Avoid setting RTFF's CLK_DIS before NRESTORE

 - qcom: Add missing MXC and MMCX power domains for Eliza

* tag 'pmdomain-v7.2-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/linux-pm:
  pmdomain: mediatek: mfg: initialize prev_o in mtk_mfg_attach_dev()
  pmdomain: qcom: rpmhpd: Add missing MXC and MMCX power domains for Eliza
  pmdomain: arm: Fix -EINVAL from scmi_pd_set_perf_state() on state 0
  pmdomain: mediatek: Fix mt8183 hang on boot
  pmdomain: mediatek: fix remaining %pOF after of_node_put()
  pmdomains: mediatek: Avoid setting RTFF's CLK_DIS before NRESTORE

7 weeks agoMerge tag 'sound-7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
Linus Torvalds [Fri, 14 Aug 2026 17:41:36 +0000 (10:41 -0700)]
Merge tag 'sound-7.2' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai"
 "A handful of small device-specific quirks, regression fixes, and build
  fixes. Nothing really stands out.

   - Fix for a potential UAF in in USB-audio MIDI2 endpoint handling

   - Fix for a mixer regression on SteelSeries Arctis Nova 5 in
     USB-audio

   - ASoC Intel HDMI-In capture, ACPI match table additions and quirks

   - ASoC AMD quirk for HyperX OMEN

   - Fix for ASoC Xilinx about crash in pcm formatter IRQ handlers

   - Fixes for ASoC Realtek rt1320 and rt5645 codecs

   - Fixes for ASoC TI tas2781 and tac5xx2-sdw build errors"

* tag 'sound-7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ASoC: Intel: NVL: Add entry for HDMI-In capture support to non-I2S codec boards.
  ASoC: xilinx: formatter_pcm: pass aud_drv_data to irq handlers
  ASoC: tac5xx2-sdw: select REGMAP_SOUNDWIRE_MBQ
  ASoC: rt1320: run the initialisation preset on the first hardware init
  ASoC: rt5645: Perform the initial jack detect at probe
  ASoC: Intel: sof_rt5682: Add HDMI-In capture with rt5682 support for NVL.
  ASoC: Intel: soc-acpi: Add entry for HDMI_In capture support in NVL match table
  ASoC: amd: yc: Add DMI quirk for HyperX OMEN Gaming Laptop 16-ap1xxx
  ASoC: tas2781: fix clang build error for goto bypassing cleanup variable
  ALSA: usb-audio: Fix mixer regression on SteelSeries Arctis Nova 5
  ALSA: usb: Fix UAF at delayed release of MIDI2 EPs

7 weeks agoMerge tag 'ceph-for-7.2-rc8' of https://github.com/ceph/ceph-client
Linus Torvalds [Fri, 14 Aug 2026 17:18:06 +0000 (10:18 -0700)]
Merge tag 'ceph-for-7.2-rc8' of https://github.com/ceph/ceph-client

Pull ceph fixes from Ilya Dryomov:
 "A handful of tiny fixes, with the main ones being a follow-up for
  CEPH_IOC_SET_LAYOUT{,_POLICY} ioctl permissions check that went into
  rc5 and a userspace compatibility fixup.  The rest mostly harden
  against malformed network input.  All marked for stable"

* tag 'ceph-for-7.2-rc8' of https://github.com/ceph/ceph-client:
  ceph: use the mount idmap for the owner checks in the SET_LAYOUT ioctls
  ceph: fix MDS random selection readiness predicate
  libceph: Avoid using invalid osd indices from primary_temp
  libceph: fix OOB read in decode_watchers() via missing bounds check
  libceph: fix multiple unsafe decodes in decode_locker()
  libceph: tolerate addrvecs with multiple entries of the same type

7 weeks agoMerge tag 'vfs-7.2-rc8.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Linus Torvalds [Fri, 14 Aug 2026 14:58:01 +0000 (07:58 -0700)]
Merge tag 'vfs-7.2-rc8.fixes' of git://git./linux/kernel/git/vfs/vfs

Pull vfs fixes from Christian Brauner:

 - Don't warn when a mount is completed from another user namespace.

   fsopen() records the caller's user namespace in fc->user_ns and
   hands back an ordinary file descriptor. The task that calls
   fsconfig(FSCONFIG_CMD_CREATE) doesn't have to be the one that
   created the context, and mount_capable() lets it through as long
   as the caller has CAP_SYS_ADMIN over fc->user_ns, which anyone in
   an ancestor namespace does. So fc->user_ns != current_user_ns()
   is something an unprivileged user can arrange.

   Both overlayfs and binfmt_misc WARN_ON() that. Overlayfs already
   has the same check as a plain error return in ovl_parse_param().

   Drop the WARN_ON() and just refuse. Add selftests for both cases.

 - Reject pid allocations through dead ancestor pid namespaces.

   Require PIDNS_ADDING in every namespace that will receive the pid
   before publishing any of them. That preserves the invariant that
   free_pid() never decrements pid_allocated in a namespace whose
   child_reaper is no longer live. The existing ENOMEM behavior is
   unchanged.

* tag 'vfs-7.2-rc8.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
  pid: reject allocations through dead ancestor pid namespaces
  selftests/filesystems: test completing a context from another user namespace
  binfmt_misc: don't warn when the mount is completed from another user namespace
  ovl: don't warn when the mount is completed from another user namespace

7 weeks agoMerge tag 'riscv-for-linus-v7.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 14 Aug 2026 14:51:55 +0000 (07:51 -0700)]
Merge tag 'riscv-for-linus-v7.2-rc8' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Paul Walmsley:

 - Fix a fault caused when the RISC-V Zbb-enabled strlen() is executed
   on a string that ends right before a page boundary, when the next
   page is unmapped

 - Fix a race with the misaligned vector performance testing code that
   can prevent the outcome of the test from being stored into the vDSO
   cache

 - Fix a kernel warning generated by the ftrace code when
   ftrace_modify_call_code() runs against a ftrace-traced function where
   a kprobe has already been attached. This shows up in the bpf
   kselftests

* tag 'riscv-for-linus-v7.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: lib: Fix ZBB strnlen reading past count boundary
  riscv: hwprobe: Register unaligned probes before usermode
  riscv: ftrace: Fix ftrace_modify_call failure on kprobed functions

7 weeks agoerofs: fix EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS on some UP platforms
Gao Xiang [Wed, 12 Aug 2026 13:11:43 +0000 (21:11 +0800)]
erofs: fix EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS on some UP platforms

CONFIG_NR_CPUS doesn't define on some UP platforms (e.g.  arm), so this
can cause make oldconfig to loop indefinitely when CONFIG_SMP=n:

  $ make ARCH=arm allmodconfig
  $ sed -i "/CONFIG_SMP=y/d" .config
  $ sed -i "/CONFIG_EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS.*/d" .config

  EROFS LZMA default maximum decompression streams (EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS) [0] (NEW)
  EROFS LZMA default maximum decompression streams (EROFS_FS_ZIP_LZMA_DEFAULT_MAX_STREAMS) [0] (NEW)
  ...

Let's guard NR_CPUS with SMP instead of using a hardcoded arbitrary CPU
uplimit here, similar to commit a3344078101c ("mm: make SPLIT_PTE_PTLOCKS
depend on SMP").

The initial report from SJ Park was for m68k [1] (m68k is the only arch
without NR_CPUS in Kconfig), and that got fixed in commit 1fd495ef09ee
("m68k: Define NR_CPUS to 1")

Reported-by: SJ Park <sj@kernel.org>
Link: https://lore.kernel.org/all/anuyFHLUGDjZWY4K@XiangdeMacBook-Pro.local/T/#u
Closes: https://lore.kernel.org/r/20260728065447.91511-1-sj@kernel.org
Reported-by: Guenter Roeck <groeck7@gmail.com>
Closes: https://lore.kernel.org/r/87853c96-cc8f-49e6-81b1-02bfe409e372@roeck-us.net
Fixes: c9b47e6b2311 ("erofs: cap LZMA stream pool size")
Signed-off-by: Gao Xiang <xiang@kernel.org>
Tested-by: SJ Park <sj@kernel.org>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 weeks agosched: Update time before requeueing delayed entities
Vincent Guittot [Fri, 14 Aug 2026 13:52:41 +0000 (15:52 +0200)]
sched: Update time before requeueing delayed entities

In order to compute the right lag, it is required to update time to 'now'.
Without this, the delayed entity might appear younger than it really is and
receive less compensation for having waited.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
7 weeks agodrm: use drm_warn() in validate_blend_mode_for_alpha_formats()
Leandro Ribeiro [Fri, 31 Jul 2026 15:42:32 +0000 (12:42 -0300)]
drm: use drm_warn() in validate_blend_mode_for_alpha_formats()

Commit 860e748bddcc ("drm: ensure blend mode supported if pixel format
with alpha exposed") introduced a WARN() to let driver developers know
that a previously valid behavior should now be changed.

But WARN() should not be used for that, as it's a kernel warning report
mechanism for conditions that are not expected to happen. It also
produces a stack trace. Instead, a simple warning-level log message
should have been used, as drivers were expected to trigger the
condition.

This is causing problems for fuzzers, as they may stop when encountering
a "BUG:" or "WARNING:" in the logs.

Replace WARN() with drm_warn() in this function, avoiding these issues.

Fixes: 860e748bddcc ("drm: ensure blend mode supported if pixel format with alpha exposed")
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Link: https://patch.msgid.link/20260731154232.37020-2-leandro.ribeiro@collabora.com
Signed-off-by: Daniel Stone <daniels@collabora.com>
7 weeks agoMerge tag 'drm-xe-next-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/xe...
Dave Airlie [Fri, 14 Aug 2026 08:29:25 +0000 (18:29 +1000)]
Merge tag 'drm-xe-next-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-next

Couple drm-ras fixes, a probe failure clean-up fix, a GT freq
boundaries fixes for BMG/CRI and a Media engines/slice fix.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/an4NGzsYN9MOTFII@intel.com
7 weeks agoMerge tag 'drm-misc-next-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm...
Dave Airlie [Fri, 14 Aug 2026 08:24:46 +0000 (18:24 +1000)]
Merge tag 'drm-misc-next-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/misc/kernel into drm-next

A single patch to remove a .orig file introduced by accident.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maxime Ripard <mripard@redhat.com>
Link: https://patch.msgid.link/20260813-shrewd-ocelot-of-research-aa3caa@houat
7 weeks agoMerge tag 'amd-drm-next-7.3-2026-08-12' of https://gitlab.freedesktop.org/agd5f/linux...
Dave Airlie [Fri, 14 Aug 2026 06:31:00 +0000 (16:31 +1000)]
Merge tag 'amd-drm-next-7.3-2026-08-12' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-7.3-2026-08-12:

amdgpu:
- Bounds checking fix in CS IOCTL
- Bounds checking fix in GEM IOCTL
- Display fixes
- GPUVM fix
- ASPM fix
- UVD bounds checking fixes
- VCE 3 fix
- BT.2020 fixes
- NBIF 6.3.1 fix
- IP discovery fix
- SMU metrics reporting fixes
- SMU 15 fixes
- Misc fixes
- MES 11 fix
- Frame size fix for some versions of clang
- Userq fixes
- GTT recovery fix

amdkfd:
- Hibernation fix
- Kernel-doc warning fixes

radeon:
- Runtime pm fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patch.msgid.link/20260812214716.2156632-1-alexander.deucher@amd.com
7 weeks agoMerge tag 'drm-xe-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/xe/kernel...
Dave Airlie [Fri, 14 Aug 2026 04:24:11 +0000 (14:24 +1000)]
Merge tag 'drm-xe-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes

Driver Changes:
- Fix DPT Allocation paths (Maarten)
- Fixes around UM queue BO (Jia)
- Order ring writes before ring tail updates (Matthew Brost)
- Add termination on resume for PXP (Daniele)
- Document Sentinel and make CTX_TIMESTAMP read TOCTOU-safe (Gajendra)
- Fix sync entry leak on OA config emit failure (Linmao Li)
- Check managed mutex initilization errors (Linmao Li)
- Fix min frequency setting (Vinay)
- Fix xe_device_probe error path (Raag)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/an4ZogmPqP2Xtfx3@fedora
7 weeks agoMerge tag 'drm-misc-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/misc...
Dave Airlie [Fri, 14 Aug 2026 03:19:02 +0000 (13:19 +1000)]
Merge tag 'drm-misc-fixes-2026-08-13' of https://gitlab.freedesktop.org/drm/misc/kernel into drm-fixes

drm-misc-fixes for v7.3:
- Revert fair scheduler patches and mark fair policy as experimental due
  to reported regressions.
- Fix OOB read in connector/hdmi infoframe.
- Handle invalid scaling parameters and empty messages in log target.
- Skip attempting to populate unmapped pages in amdxdna.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patch.msgid.link/a9b38792-bdd0-42da-a46a-7a048c26c0c2@linux.intel.com
7 weeks agoMerge tag 'amd-drm-fixes-7.2-2026-08-12' of https://gitlab.freedesktop.org/agd5f...
Dave Airlie [Fri, 14 Aug 2026 02:41:01 +0000 (12:41 +1000)]
Merge tag 'amd-drm-fixes-7.2-2026-08-12' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amd-drm-fixes-7.2-2026-08-12:

amdgpu:
- Bounds checking fix in CS IOCTL
- Bounds checking fix in GEM IOCTL
- Display fixes
- GPUVM fix
- ASPM fix
- UVD bounds checking fixes
- VCE 3 fix
- BT.2020 fixes
- NBIF 6.3.1 fix
- IP discovery fix

radeon:
- Runtime pm fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patch.msgid.link/20260812200720.2155401-1-alexander.deucher@amd.com
7 weeks agoMerge tag 'trace-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
Linus Torvalds [Thu, 13 Aug 2026 23:08:22 +0000 (16:08 -0700)]
Merge tag 'trace-v7.2-rc7' of git://git./linux/kernel/git/trace/linux-trace

Pull tracing fixes from Steven Rostedt:

 - Fix NULL pointer dereference when matching unloaded module wildcard
   event

   The set_event can take events for modules that have not been loaded
   yet. This is done by writing '<event>:mod:<module>'.

   If '<event>' is not added, then it means to add all events in
   <module>. This wildcard is represented by a NULL pointer. If one were
   to try to remove the same module item with a named event it would
   cause a NULL pointer dereference when comparing the NULL with the
   name in strcmp().

      echo ':mod:kvm' > /sys/kernel/tracing/set_event
      echo '!kvm_ack_irq:mod:kvm' >> /sys/kernel/tracing/set_event

   The above will do a strcmp("kvm_ack_irq", NULL) and crash the kernel.

   Test for NULL (wildcard) before doing the strcmp().

 - Fix event data field race in loading two modules at the same time

   When a module loads, its trace events get registered. The fields of
   the events are also dynamically created and added to the events
   fields list. It also will call a function that will look at all the
   events for updates that need to be done. If two modules load at the
   same time, the one that scans all events and their fields may read
   the one being added as the scan doesn't take the event_mutex. This
   may cause a data race.

   Have the scan take the event_mutex to prevent the race.

* tag 'trace-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  tracing: Fix race between update_event_fields and, event_define_fields
  tracing: Fix NULL pointer dereference in module event cache removal

7 weeks agotracing: Fix race between update_event_fields and, event_define_fields
Michael Wu [Mon, 10 Aug 2026 06:32:30 +0000 (14:32 +0800)]
tracing: Fix race between update_event_fields and, event_define_fields

The following sequence may leads race between event_define_fields()
and update_event_fields():

 CPU0 (loads module A)                      CPU1 (loads module B)
 ===============================            ===============================
 load_module(A)                             load_module(B)
   notifier_call_chain                        notifier_call_chain
     trace_module_notify                        trace_module_notify
       mutex_lock(&event_mutex)                   trace_event_update_all()
         trace_module_add_events(A)                 down_write(&trace_event_sem)
            __register_event(call_A)
              __add_event_to_tracers(call_A)
                event_define_fields(call_A)
                  for each f:                         list_for_each_entry(field,
                    list_add(&f->link,                                    &class->fields, link)
                             &class->fields)            field = class->fields->next;

Where access to the class->fields is not protected by the event_mutex in
trace_event_update_all().

This produces the following panic:
   Unable to handle kernel access ... at virtual address 0000000000000018
   pc : update_event_fields+0xf8/0x368
   Call trace:
    update_event_fields+0xf8/0x368
    trace_event_update_all+0x7c/0x2b4
    trace_module_notify+0x4c/0x1dc
    notifier_call_chain+0x84/0x168
    blocking_notifier_call_chain_robust+0x64/0xd4
    load_module+0x10c8/0x123c
    __arm64_sys_finit_module+0x230/0x31c

Fix by taking event_mutex in trace_event_update_all() before
trace_event_sem.

Cc: stable@vger.kernel.org
Fixes: b3bc8547d3be ("tracing: Have TRACE_DEFINE_ENUM affect trace event types as well")
Link: https://patch.msgid.link/2e5730d2-c631-da41-3a3a-ae35bb4895f3@allwinnertech.com
Signed-off-by: Michael Wu <michael@allwinnertech.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
7 weeks agotracing: Fix NULL pointer dereference in module event cache removal
Hui Su [Tue, 11 Aug 2026 17:39:03 +0000 (01:39 +0800)]
tracing: Fix NULL pointer dereference in module event cache removal

A module-only event filter such as ":mod:foo" is cached with a NULL
event_mod->match when foo has not been loaded. If a later write tries to
remove a specific match from the same module, remove_cache_mod() passes
the NULL cached match to strcmp(), causing a NULL pointer dereference.

The issue can be reproduced from userspace:

  echo ':mod:trace_events_kunit_missing' > /sys/kernel/tracing/set_event
  echo '!foo_bar:mod:trace_events_kunit_missing' >> /sys/kernel/tracing/set_event

The second write must be a concatenation (">>") to not include O_TRUNC as
that would cause ftrace_clear_events() to clear the cached modules lines.

The crash was reproduced on x86_64 QEMU while KUnit workers contended on
the event tracing path:

  BUG: kernel NULL pointer dereference, address: 0000000000000000
  #PF: supervisor read access in kernel mode
  RIP: 0010:strcmp+0x10/0x30
  Call Trace:
   __ftrace_set_clr_event_nolock+0x373/0x4a0
   ftrace_set_clr_event+0xf0/0x180
   ftrace_event_write+0xdf/0x110
   vfs_write+0xf6/0x440
   ksys_write+0x68/0xe0
   do_syscall_64+0xf9/0x540
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

Check event_mod->match before comparing it, consistent with the existing
NULL checks for the cached system and event fields. The mismatched removal
continues to return -EINVAL; a broad cached module filter is removed with
"!:mod:<module>".

Cc: stable@vger.kernel.org
Link: https://patch.msgid.link/20260811173902.1927376-2-sh_def@163.com
Fixes: b355247df104 ("tracing: Cache \":mod:\" events for modules not loaded yet")
Reported-by: syzbot+4d3143c8e28f6266c636@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/lkml/6a7a6b7f.9c11d2ce.289b96.00f8.GAE@google.com/
Signed-off-by: Hui Su <sh_def@163.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
7 weeks agospi: virtio: mark device ready before registering the controller
Jasper Wise [Thu, 13 Aug 2026 08:46:18 +0000 (08:46 +0000)]
spi: virtio: mark device ready before registering the controller

virtio_spi_probe() registers the SPI controller with
devm_spi_register_controller(). spi_register_controller() binds a child
inline unless its driver has asked for asynchronous probing, so a
peripheral that performs a transfer during its own probe reaches
virtio_spi_transfer_one(), which kicks the virtqueue before probe has
returned.

The driver never calls virtio_device_ready(), so DRIVER_OK is set on its
behalf by virtio_dev_probe(), only once probe has returned. The virtio
spec is explicit about that ordering in 3.1 Device Initialization:
  |  The driver MUST NOT send any buffer available notifications to the
  |  device before setting DRIVER_OK.

A device that waits for DRIVER_OK before servicing the queue therefore
leaves the transfer unanswered, and virtio_spi_transfer_one() waits for its
completion with no timeout, so probe never returns.

Mark the device ready before registering the controller, as done for the
same reason in commit f5866db64f34 ("virtio_console: enable VQs early") and
commit 1d774589f924 ("i2c: virtio: mark device ready before registering the
adapter").

Fixes: f98cabe3f6cf ("SPI: Add virtio SPI driver")
Signed-off-by: Jasper Wise <jaspwise@amazon.co.uk>
Link: https://patch.msgid.link/20260813084618.613172-1-jaspwise@amazon.co.uk
Signed-off-by: Mark Brown <broonie@kernel.org>
7 weeks agoRevert "wifi: mt76: Disable napi when removing device"
Mikhail Gavrilov [Tue, 4 Aug 2026 12:00:04 +0000 (17:00 +0500)]
Revert "wifi: mt76: Disable napi when removing device"

This reverts commit 13b7e6a96a005c656d38f3da51581deaf9866375.

That commit made mt76_dma_cleanup() disable every RX NAPI instance before
deleting it, to silence WARNs in __netif_napi_del_locked() and
page_pool_disable_direct_recycling() seen when unloading mt7915e with an
MT7916.

On mt7921e and mt7925e the same instances are already disabled earlier,
in mt7921e_unregister_device() and mt7925e_unregister_device(), which
only afterwards call mt792x_dma_cleanup() -> mt76_dma_cleanup().  Each
instance is therefore disabled twice, and napi_disable() is not
idempotent: on return it leaves NAPIF_STATE_SCHED and NAPIF_STATE_NPSVC
set, so the second call spins in usleep_range() forever, waiting for bits
that nobody will clear.

mt7921_pci_shutdown() and mt7925_pci_shutdown() reuse the remove path, so
this is hit on every reboot, poweroff and module unload.  It is silent:
the stuck task keeps sleeping and rescheduling, so neither the hung task
detector nor the lockup detectors fire, and the last line on the console
is "systemd-shutdown[1]: Rebooting."

  task:modprobe        state:D stack:25720 pid:7954  tgid:7954
  Call Trace:
   <TASK>
   __schedule+0x11b8/0x26d0
   schedule+0xe7/0x2f0
   schedule_hrtimeout_range_clock+0x218/0x330
   usleep_range_state+0x133/0x1b0
   napi_disable_locked+0x37d/0x5f0
   napi_disable+0x43/0x80
   mt76_dma_cleanup+0x2b4/0x860 [mt76]
   mt7921_pci_remove+0x17f/0x350 [mt7921e]
   pci_device_remove+0xb6/0x1e0
   device_release_driver_internal+0x38d/0x540
   driver_detach+0xd0/0x1b0
   bus_remove_driver+0x127/0x2d0
   pci_unregister_driver+0x2a/0x280
   __do_sys_delete_module+0x36a/0x5b0
   do_syscall_64+0x11c/0x6d0
   entry_SYSCALL_64_after_hwframe+0x76/0x7e
   </TASK>

Dropping the two driver-side loops instead was tried and rejected: with
them gone, the RX poll can reach mt76_token_release() via
PKT_TYPE_TXRX_NOTIFY and mt7921_mac_tx_free() while
mt76_connac2_tx_token_put() is running idr_destroy(&dev->token) outside
token_lock, which is a use-after-free rather than a hang [1].

Revert for now, so that reboot, poweroff and module unload work again.
The WARNs on mt7915e are a less severe problem than an unbootable
machine, and fixing them belongs in the drivers that delete the NAPI
instances, where each one can pick a point that is safe for its own
teardown order, rather than in the shared mt76_dma_cleanup().

[ This is the "landing soonish" known regression fix mentioned in the
  previous networking merge commit       - Linus ]

Reported-by: Bert Karwatzki <spasswolf@web.de>
Closes: https://lore.kernel.org/all/20260724151419.26014-1-spasswolf@web.de/
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=221818
Link: https://lore.kernel.org/all/20260730050428.GA73812@sol/
Signed-off-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Acked-by: Nicolas Cavallari <nicolas.cavallari@green-communications.fr>
Fixes: 13b7e6a96a00 ("wifi: mt76: Disable napi when removing device")
Tested-by: Devin Wittmayer <lucid_duck@justthetip.ca>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 weeks agoMerge tag 'net-7.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Thu, 13 Aug 2026 15:37:26 +0000 (08:37 -0700)]
Merge tag 'net-7.2-rc8' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Including fixes from netfilter.

  There is a known WiFi/mt76 regression, waiting for a complete fix that
  should land soonish.

  Previous releases - regressions:

   - tcp: fix icsk_ack.ato bitfield overflow

   - af_unix: Unlink scc_entry in unix_del_edge()

   - ipv4: fix use-after-free in fib_nhc_update_mtu()

   - netfilter:
      - ipset: fix refcount race between list:set GC and swap
      - nf_tables_offload: suppress WARN_ON_ONCE for ENOMEM in abort
        path

   - sched: act_ct: fix sk_buff leak when the header checks reject a
     packet

   - sctp: clear new_transport when removing a peer

   - dibs: correct freeing of dmb_clientid_arr

   - ovpn: fix NULL dereference when killing missing key

   - eth:
      - veth: fix queue index used to wake the peer txq in veth_poll
      - ngbe: fix NULL pointer dereference in non-MSI-X interrupt
        enabling
      - gve: fix zero-length skb frag with header-split

  Previous releases - always broken:

   - core: fix skb length accounting after generic XDP frag adjustment

   - af_packet: don't send zero-byte data in tpacket_snd().

   - eth:
      - bnxt: avoid deadlock when canceling IRQ affinity notifier
      - ipvlan: inherit needed_headroom and needed_tailroom from
        phy_dev"

* tag 'net-7.2-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (55 commits)
  l2tp: fix tunnel and session refcount leak on seq_file release
  net/sched: cls_bpf: reject dev-bound programs bound to a different device
  sctp: fix use-after-free of cached ASCONF chunk
  net: ethernet: ti: am65-cpsw-nuss: Fix port_id extraction from SRC TAG
  sctp: clear new_transport when removing a peer
  net/dibs: Correct freeing of dmb_clientid_arr
  net/sched: cls_u32: skip hash tables in u32_bind_class()
  gve: fix NULL dereference due to missing ptp adjfine
  gve: fix zero-length skb frag with header-split
  net/sched: act_api: fix TOCTOU NULL deref on a->goto_chain
  af_packet: Don't send zero-byte data in tpacket_snd().
  tipc: read le->link under the node lock in tipc_node_link_down()
  selftests: tls: cover splice after a failed decrypt
  net/tls: Fail tls_sw_splice_read() after a failed async decrypt
  net: ngbe: fix NULL pointer dereference in non-MSI-X interrupt enabling
  net: tap: fix wrong transport_header when sending VLAN-tagged frame
  net: packet: fix wrong transport_header when sending VLAN-tagged frame
  vxlan: do not arm the ageing timer on a device that is down
  ipv4: fix use-after-free in fib_nhc_update_mtu()
  NTB: ntb_netdev: Preserve RX queue depth on allocation failure
  ...

7 weeks agoMerge tag 'firewire-fixes-7.2-final' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 13 Aug 2026 14:31:21 +0000 (07:31 -0700)]
Merge tag 'firewire-fixes-7.2-final' of git://git./linux/kernel/git/ieee1394/linux1394

Pull firewire fix from Takashi Sakamoto:
 "Fix a NULL pointer dereference in 1394 OHCI PCI driver when probe()
  returns early with an error, as detected by Syzkaller"

* tag 'firewire-fixes-7.2-final' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: ohci: fix NULL pointer dereference in ar_context_release

7 weeks agoMerge tag 'gpio-fixes-for-v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 13 Aug 2026 14:16:58 +0000 (07:16 -0700)]
Merge tag 'gpio-fixes-for-v7.2' of git://git./linux/kernel/git/brgl/linux

Pull gpio fixes from Bartosz Golaszewski:

 - use raw_spinlock_t in gpio-ml-ioh to avoid locking context issues

 - fix a race condition in gpio-ml-ioh by sharing the register locks
   across channels

 - fix a use-after-free bug in unbind path in gpio-sloppy-logic-analyzer

* tag 'gpio-fixes-for-v7.2' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux:
  gpio: sloppy-logic-analyzer: fix use-after-free via debugfs trigger on unbind
  gpio: ml-ioh: share the register lock across channels
  gpio: ml-ioh: use raw_spinlock_t for the register lock
  gpiolib: Check gc->get_direction() before calling gpiod_get_direction()

7 weeks agoMerge tag 'm68k-for-v7.2-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert...
Linus Torvalds [Thu, 13 Aug 2026 14:00:26 +0000 (07:00 -0700)]
Merge tag 'm68k-for-v7.2-tag2' of git://git./linux/kernel/git/geert/linux-m68k

Pull m68k fix from Geert Uytterhoeven:
 "Define NR_CPUS to 1.

  This fixes a long-standing but never critical before oddity on m68k,
  that turned into a serious configuration issue after a recent erofs
  change"

* tag 'm68k-for-v7.2-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k:
  m68k: Define NR_CPUS to 1

7 weeks agodrm/log: Fix infinite loop when scale is too large for display
Shixiong Ou [Wed, 29 Jul 2026 08:48:15 +0000 (16:48 +0800)]
drm/log: Fix infinite loop when scale is too large for display

When scale is large enough that scaled_font exceeds the display
dimensions, rows or columns become 0. A columns value of 0 causes
an infinite loop in drm_log_draw_kmsg_record() because the loop
never decrements len.

Check for zero rows/columns in drm_log_setup_modeset() and return
an error, cleaning up the already allocated buffer to avoid a leak.

Fixes: 8a4b913df427 ("drm/log: Add integer scaling support")
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Reviewed-by: Jocelyn Falempe <jfalempe@redhat.com>
Link: https://patch.msgid.link/20260729084815.692944-1-oushixiong1025@163.com
Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>
7 weeks agodrm/log: Fix out-of-bounds read on empty message length
Shixiong Ou [Wed, 29 Jul 2026 08:45:17 +0000 (16:45 +0800)]
drm/log: Fix out-of-bounds read on empty message length

drm_log_draw_kmsg_record() accesses s[len - 1] to strip the trailing
newline, but len is unsigned int. If len is 0, the subtraction wraps
to UINT_MAX, causing an out-of-bounds read.

Add an early return when len is 0.

Fixes: 25e2c2a3eff5 ("drm/log: Color the timestamp, to improve readability")
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Reviewed-by: Jocelyn Falempe <jfalempe@redhat.com>
Link: https://patch.msgid.link/20260729084520.688087-1-oushixiong1025@163.com
Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>