The runtime suspend callback drains the running job workqueue before
suspending the device. If a job is still executing and calls
pm_runtime_resume_and_get(), it can deadlock with the runtime suspend
path.
Fix this by moving pm_runtime_resume_and_get() from the job execution
routine to the job submission routine, ensuring the device is resumed
before the job is queued and avoiding the deadlock during runtime
suspend.
Fixes: 063db45183 ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260310180058.336348-1-lizhi.hou@amd.com
The management channel used for firmware control command submission is
currently created after the firmware is started. If channel creation
fails (for example, due to memory allocation failure or workqueue
creation interruption), the firmware remains in a pending state and is
unable to receive any control commands.
To avoid leaving the firmware in this inconsistent state, split
xdna_mailbox_create_channel() into two separate functions so that
resource allocation can be completed before interacting with the
hardware.
xdna_mailbox_alloc_channel()
Allocates memory and initializes the workqueue. This can be called
earlier, before interacting with the hardware.
xdna_mailbox_start_channel()
Performs the hardware interaction required to start the channel.
Rename xdna_mailbox_destroy_channel() to xdna_mailbox_free_channel().
Ensure that xdna_mailbox_stop_channel() and xdna_mailbox_free_channel()
properly unwind the corresponding start and allocation steps, respectively.
Fixes: b87f920b93 ("accel/amdxdna: Support hardware mailbox")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260305062041.3954024-1-lizhi.hou@amd.com
Add the missing major number in npu1_fw_feature_table.
Without the major version specified, the firmware feature check fails,
preventing new firmware commands from being enabled on the NPU1
platform.
With the correct major version populated, the driver properly detects
firmware support and enables the new command.
Fixes: f1eac46fe5 ("accel/amdxdna: Update firmware version check for latest firmware")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260304195012.3616908-1-lizhi.hou@amd.com
mgmt_chann may be set to NULL if the firmware returns an unexpected
error in aie2_send_mgmt_msg_wait(). This can later lead to a NULL
pointer dereference in aie2_hw_stop().
Fix this by introducing a dedicated helper to destroy mgmt_chann
and by adding proper NULL checks before accessing it.
Fixes: b87f920b93 ("accel/amdxdna: Support hardware mailbox")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260226213857.3068474-1-lizhi.hou@amd.com
Newer userspace applications may read the payload of a failed command
to obtain detailed error information. However, the driver and old firmware
versions may not support returning advanced error information.
In this case, initialize the command payload with an invalid value so
userspace can detect that no detailed error information is available.
Fixes: aac243092b ("accel/amdxdna: Add command execution")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260227004841.3080241-1-lizhi.hou@amd.com
The remaining space in a command slot may be smaller than the size of
the command header. Clearing the command header with memset() before
verifying the available slot space can result in an out-of-bounds write
and memory corruption.
Fix this by moving the memset() call after the size validation.
Fixes: 3d32eb7a5e ("accel/amdxdna: Fix cu_idx being cleared by memset() during command setup")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260217185415.1781908-1-lizhi.hou@amd.com
When a hardware context is suspended, the job scheduler is stopped. If a
command is submitted while the context is suspended, the job is queued in
the scheduler but aie2_sched_job_run() is never invoked to restart the
hardware context. As a result, the command hangs.
Fix this by modifying the hardware context suspend routine to keep the job
scheduler running so that queued jobs can trigger context restart properly.
Fixes: aac243092b ("accel/amdxdna: Add command execution")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260211205341.722982-1-lizhi.hou@amd.com
Enabling turbo mode disables hardware clock gating. Suspend requires
hardware clock gating to be re-enabled, otherwise suspend will fail.
Fix this by calling aie2_runtime_cfg() from aie2_hw_stop() to
re-enable clock gating during suspend. Also ensure that firmware is
initialized in aie2_hw_start() before modifying clock-gating
settings during resume.
Fixes: f4d7b8a6bc ("accel/amdxdna: Enhance power management settings")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260211204716.722788-1-lizhi.hou@amd.com
When an application issues a query IOCTL while auto suspend is running,
a deadlock can occur. The query path holds dev_lock and then calls
pm_runtime_resume_and_get(), which waits for the ongoing suspend to
complete. Meanwhile, the suspend callback attempts to acquire dev_lock
and blocks, resulting in a deadlock.
Fix this by releasing dev_lock before calling pm_runtime_resume_and_get()
and reacquiring it after the call completes. Also acquire dev_lock in the
resume callback to keep the locking consistent.
Fixes: 063db45183 ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260211204644.722758-1-lizhi.hou@amd.com
During process termination, several error messages are logged that are
not actual errors but expected conditions when a process is killed or
interrupted. This creates unnecessary noise in the kernel log.
The specific scenarios are:
1. HMM invalidation returns -ERESTARTSYS when the wait is interrupted by
a signal during process cleanup. This is expected when a process is
being terminated and should not be logged as an error.
2. Context destruction returns -ENODEV when the firmware or device has
already stopped, which commonly occurs during cleanup if the device
was already torn down. This is also an expected condition during
orderly shutdown.
Downgrade these expected error conditions from error level to debug level
to reduce log noise while still keeping genuine errors visible.
Fixes: 97f2757383 ("accel/amdxdna: Fix potential NULL pointer dereference in context cleanup")
Reviewed-by: Lizhi Hou <lizhi.hou@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260210164521.1094274-3-mario.limonciello@amd.com
If userspace issues an ioctl to destroy a hardware context that has
already been automatically suspended, the driver may crash because the
mailbox channel pointer is NULL for the suspended context.
Fix this by checking the mailbox channel pointer in aie2_destroy_context()
before accessing it.
Fixes: 97f2757383 ("accel/amdxdna: Fix potential NULL pointer dereference in context cleanup")
Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260206060306.4050531-1-lizhi.hou@amd.com
Preempt commands are only supported when submitted as chained commands.
To ensure preempt support works consistently, always submit commands in
chained command format.
Set force_cmdlist to true so that single commands are filled using the
chained command layout, enabling correct handling of preempt commands.
Fixes: 3a0ff7b98a ("accel/amdxdna: Support preemption requests")
Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260206060251.4050512-1-lizhi.hou@amd.com
Large command buffers may be used, and they do not always need to be
mapped or accessed by the driver. Performing a size check at command BO
creation time unnecessarily rejects valid use cases.
Remove the buffer size check from command BO creation, and defer vmap
and size validation to the paths where the driver actually needs to map
and access the command buffer.
Fixes: ac49797c18 ("accel/amdxdna: Add GEM buffer object management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260206060237.4050492-1-lizhi.hou@amd.com
This converts some of the visually simpler cases that have been split
over multiple lines. I only did the ones that are easy to verify the
resulting diff by having just that final GFP_KERNEL argument on the next
line.
Somebody should probably do a proper coccinelle script for this, but for
me the trivial script actually resulted in an assertion failure in the
middle of the script. I probably had made it a bit _too_ trivial.
So after fighting that far a while I decided to just do some of the
syntactically simpler cases with variations of the previous 'sed'
scripts.
The more syntactically complex multi-line cases would mostly really want
whitespace cleanup anyway.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the exact same thing as the 'alloc_obj()' version, only much
smaller because there are a lot fewer users of the *alloc_flex()
interface.
As with alloc_obj() version, this was done entirely with mindless brute
force, using the same script, except using 'flex' in the pattern rather
than 'objs*'.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
Currently, amdxdna_pm_resume_get() is called during job creation, and
amdxdna_pm_suspend_put() is called when the hardware notifies job
completion. If a job is canceled before it is run, no hardware
completion notification is generated, resulting in an unbalanced
runtime PM resume/suspend pair.
Fix this by moving amdxdna_pm_resume_get() to the job run path, ensuring
runtime PM is only resumed for jobs that are actually executed.
Fixes: 063db45183 ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260204171118.3165607-1-lizhi.hou@amd.com
The suspend routine sets the DPM level to 0, which unintentionally
overwrites the previously saved DPM level. As a result, the device always
resumes with DPM level 0 instead of restoring the original value.
Fix this by ensuring the suspend path does not overwrite the saved DPM
level, allowing the correct DPM level to be restored during resume.
Fixes: f4d7b8a6bc ("accel/amdxdna: Enhance power management settings")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260204171048.3165580-1-lizhi.hou@amd.com
One newly supported command does not require hardware context configuration
to be performed upfront. As a result, checking hardware context status
causes this command to fail incorrectly.
Remove hardware context status handling entirely. For other commands,
if userspace submits a request without configuring the hardware context
first, the firmware will report an error or time out as appropriate.
Fixes: aac243092b ("accel/amdxdna: Add command execution")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260202212450.2681273-1-lizhi.hou@amd.com
Running jobs on a hardware context while it is in the process of
releasing resources can lead to use-after-free and crashes.
Fix this by stopping job scheduling before calling
aie2_release_resource() and restarting it after the release completes.
Additionally, aie2_sched_job_run() now checks whether the hardware
context is still active.
Fixes: 4fd6ca90fc ("accel/amdxdna: Refactor hardware context destroy routine")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260130003255.2083255-1-lizhi.hou@amd.com
Some tests trigger a crash in iommu_sva_unbind_device() due to
accessing iommu_mm after the associated mm structure has been
freed.
Fix this by taking an explicit reference to the mm structure
after successfully binding the device, and releasing it only
after the device is unbound. This ensures the mm remains valid
for the entire SVA bind/unbind lifetime.
Fixes: be462c97b7 ("accel/amdxdna: Add hardware context")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20260128002356.1858122-1-lizhi.hou@amd.com
The latest firmware requires the message DMA buffer to
- have a minimum size of 8K
- use a power-of-two size
- be aligned to the buffer size
- not cross 64M boundary
Update the buffer allocation logic to meet these requirements and support
the latest firmware.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251219014356.2234241-1-lizhi.hou@amd.com
Newer firmware versions prefer temporal sharing only mode. In this mode,
the driver no longer needs to manage AIE array column allocation. Instead,
a new field, num_unused_col, is added to the hardware context creation
request to specify how many columns will not be used by this hardware
context.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251217191150.2145937-1-lizhi.hou@amd.com
amdxdna_flush() was introduced to ensure that the device does not access
a process address space after it has been freed. However, this is no
longer necessary because the driver now increments the mm reference count
when a command is submitted and decrements it only after the command has
completed. This guarantees that the process address space remains valid
for the entire duration of command execution. Remove amdxdna_flush to
simplify the teardown path.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251216031311.2033399-1-lizhi.hou@amd.com
aie_destroy_context() is invoked during error handling in
aie2_create_context(). However, aie_destroy_context() assumes that the
context's mailbox channel pointer is non-NULL. If mailbox channel
creation fails, the pointer remains NULL and calling aie_destroy_context()
can lead to a NULL pointer dereference.
In aie2_create_context(), replace aie_destroy_context() with a function
which request firmware to remove the context created previously.
Fixes: be462c97b7 ("accel/amdxdna: Add hardware context")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251212183244.1826318-1-lizhi.hou@amd.com
The firmware sends a response and interrupts the driver before advancing
the mailbox send ring head pointer. As a result, the driver may observe
the response and attempt to send a new request before the firmware has
updated the head pointer. In this window, the send ring still appears
full, causing the driver to incorrectly fail the send operation.
This race can be triggered more easily in a multithreaded environment,
leading to unexpected and spurious "send ring full" failures.
To address this, poll the send ring head pointer for up to 100us before
returning a full-ring condition. This allows the firmware time to update
the head pointer.
Fixes: b87f920b93 ("accel/amdxdna: Support hardware mailbox")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251211045125.1724604-1-lizhi.hou@amd.com
For one command type, cu_idx is assigned before calling memset() on the
command structure. This results in cu_idx being overwritten, causing the
firmware to receive an incomplete or invalid command and leading to
unexpected command failures.
Fix this by moving the memset() call before initializing cu_idx so that
all fields are populated in the correct order.
Fixes: 71829d7f2f ("accel/amdxdna: Use MSG_OP_CHAIN_EXEC_NPU when supported")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251209211639.1636888-1-lizhi.hou@amd.com
When autosuspend is triggered, driver rpm_on flag is set to indicate that
a suspend/resume is already in progress. However, when a userspace
application submits a command during this narrow window,
amdxdna_pm_resume_get() may incorrectly skip the resume operation because
the rpm_on flag is still set. This results in commands being submitted
while the device has not actually resumed, causing unexpected behavior.
The set_dpm() is called by suspend/resume, it relied on rpm_on flag to
avoid calling into rpm suspend/resume recursivly. So to fix this, remove
the use of the rpm_on flag entirely. Instead, introduce aie2_pm_set_dpm()
which explicitly resumes the device before invoking set_dpm(). With this
change, set_dpm() is called directly inside the suspend or resume execution
path. Otherwise, aie2_pm_set_dpm() is called.
Fixes: 063db45183 ("accel/amdxdna: Enhance runtime power management")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251208165356.1549237-1-lizhi.hou@amd.com
In mailbox_get_msg(), mailbox_reg_read_non_zero() is called to poll for a
non-zero tail pointer. This assumed that a zero value indicates an error.
However, certain corner cases legitimately produce a zero tail pointer.
To handle these cases, remove mailbox_reg_read_non_zero(). The zero tail
pointer will be treated as a valid rewind event.
Reviewed-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251204181603.793824-1-lizhi.hou@amd.com
After issuing a firmware suspend request, the driver must ensure that the
suspend operation has completed before proceeding. Add polling of the
MPNPU_PWAITMODE register to confirm that the firmware has fully entered
the suspended state. This prevents race conditions where subsequent
operations assume the firmware is idle before it has actually completed
its suspend sequence.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251202165427.507414-1-lizhi.hou@amd.com
Hardware context destroy function holds dev_lock while waiting for all jobs
to complete. The timeout job also needs to acquire dev_lock, this leads to
a deadlock.
Fix the issue by temporarily releasing dev_lock before waiting for all
jobs to finish, and reacquiring it afterward.
Fixes: 4fd6ca90fc ("accel/amdxdna: Refactor hardware context destroy routine")
Reviewed-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251107181050.1293125-1-lizhi.hou@amd.com
Failing to set power off indicates an unrecoverable hardware or firmware
error. Update the driver to treat such a failure as a fatal condition
and stop further operations that depend on successful power state
transition.
This prevents undefined behavior when the hardware remains in an
unexpected state after a failed power-off attempt.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251106180521.1095218-1-lizhi.hou@amd.com
Currently, dma_fence_put(job->fence) is called in job notification
callback. However, if a job is canceled, the notification callback is never
invoked, leading to a memory leak. Move dma_fence_put(job->fence)
to the job cleanup function to ensure the fence is always released.
Fixes: aac243092b ("accel/amdxdna: Add command execution")
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251105194140.1004314-1-lizhi.hou@amd.com
The driver checks the firmware version during initialization.If preemption
is supported, the driver configures preemption accordingly and handles
userspace preemption requests. Otherwise, the driver returns an error for
userspace preemption requests.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
Link: https://patch.msgid.link/20251104185340.897560-1-lizhi.hou@amd.com
