A deep dive into the AP bus code and zcrypt device driver about the
usage of mutex locking showed that there is one questionable call in
ap_bus.c in function ap_bus_force_rescan(). This function may be
called in kernel and process context. In both contexts only one info
is important: was there a AP bus scan running and did it result in
some updates on the AP devices. So only true/false is returned but no
info like -EINTR. But still the mutex lock call should be
interruptible to be able to kill a user space program blocked forever
on this. So this patch replaces mutex_lock_interruptible() with
mutex_lock_killable() to be able to handle SIGKILL especially in user
space process context.
Suggested-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Harald Freudenberger <freude@linux.ibm.com>
Reviewed-by: Finn Callies <fcallies@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
While at it explicitly specify GNU_EH_FRAME PHDR to be read-only.
Inspired by x86 commit 8717b02b8c ("x86/entry/vdso: Include
GNU_PROPERTY and GNU_STACK PHDRs").
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Jens Remus <jremus@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Commit 2b2a25845d ("s390/vdso: Use $(LD) instead of $(CC) to link
vDSO") accidentally broke the GNU_EH_FRAME program table entry in
the vDSO, causing it to be empty:
$ readelf --program-headers arch/s390/kernel/vdso/vdso.so
...
Program Headers:
Type Offset VirtAddr PhysAddr
FileSiz MemSiz Flags Align
...
GNU_EH_FRAME 0x0000000000000000 0x0000000000000000 0x0000000000000000
0x0000000000000000 0x0000000000000000 0x8
...
Originally, the compiler would implicitly add --eh-frame-hdr when
invoking the linker, but when this Makefile was converted from invoking
the linker via the compiler, to invoking it directly, the option was
missed.
This is the s390 variant of x86 commit cd01544a26 ("x86/vdso: Pass
--eh-frame-hdr to the linker").
Fixes: 2b2a25845d ("s390/vdso: Use $(LD) instead of $(CC) to link vDSO")
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Jens Remus <jremus@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Make the code a bit easier to read by defining SYSCALL_PER_TRAP
instead of passing '1' to __do_syscall().
Suggested-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
PIF_GUEST_FAULT is only used to pass information whether a fault was
caused when executing SIE or when executing host code. Instead of
using ptregs for this, just pass the flag directly as argument to
__do_pgm_check(). This also saves the time required to read the flag
from ptregs, although this likely isn't much as it is already in the
data cache.
Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
unistd_32.h is not build anymore since commit 4ac286c4a8
("s390/syscalls: Switch to generic system call table generation").
Thus drop the superfluous line from the Kbuild file now, too.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
SWIOTLB_DYNAMIC allows the SWIOTLB to grow on demand when the initial
bounce-buffer area is exhausted, instead of relying only on the pool
reserved at boot. This improves behavior under sustained I/O pressure.
In testing, exhaustion of the initial SWIOTLB area was observed under
real workloads such as large file transfers and package installation
workloads, where continued availability of bounce buffering was needed
to avoid guest-visible I/O stalls or failures.
When the current SWIOTLB capacity is exhausted, the dynamic expansion
path schedules asynchronous pool growth and then attempts immediate
non-blocking allocation of transient bounce-buffer memory.
In that path, the current implementation uses the generic DMA coherent
atomic pool for transient bounce-buffer allocations. Selecting
DMA_COHERENT_POOL together with SWIOTLB_DYNAMIC on s390 makes that
atomic fallback path available.
The coherent atomic pool is a generic DMA facility. By default, its size
is 128 KiB per 1 GiB of memory, with a minimum default size of 128 KiB,
and it can be adjusted with the coherent_pool= kernel parameter. The pool
itself can be exhausted, so this improves behavior under pressure, but
does not guarantee that DMA mappings will always succeed.
Select both options from s390 Kconfig so this support is available by
default on s390 systems.
Signed-off-by: Jaehoon Kim <jhkim@linux.ibm.com>
Acked-by: Niklas Schnelle <schnelle@linux.ibm.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Tested-by: Joshua Daley <jdaley@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
In s390 protected virtualization (PV) environments, guests use SWIOTLB
for virtio DMA. Because SWIOTLB_ANY is currently not passed to
swiotlb_init(), the SWIOTLB buffer is allocated from low memory, i.e.
from the first 2G of the physical address space on s390.
This restriction is unnecessary for s390 PV guests and limits the
available address range for the SWIOTLB buffer. In turn, that can limit
the maximum practical SWIOTLB size and make larger allocations more
likely to fail at boot, especially under memory fragmentation.
Pass SWIOTLB_ANY in pv_init() so the SWIOTLB buffer can be allocated
from any suitable memory instead of being restricted to low memory.
Signed-off-by: Jaehoon Kim <jhkim@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Reviewed-by: Halil Pasic <pasic@linux.ibm.com>
Tested-by: Joshua Daley <jdaley@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
[Why]
We need to exit PSR/IPS before programming. Before calling DC for
programming in amdgpu_dm_commit_planes(), there's a
vblank_control_workqueue flush. This waits for IPS and PSR exit. (See
drm_vblank_on/off() > amdgpu_dm_crtc_set_vblank() --queue_work()->
amdgpu_dm_crtc_vblank_control_worker())
Prior to the tagged "Fixes:" change, drm_vblank_get() was called before
the workqueue flush. This ordering ensures that PSR exit occurred before
programming. After the "Fixes:" change, drm_vblank_get() is called after
the workqueue flush, leading to programming while idle optimizations are
still active. This can lead to incorrect flip_pending detection used by
vblank event delivery.
[How]
Split the vblank_get() component of `dm_arm_vblank_event()` into
`dm_arm_vblank_event_pre_programming()`, which is called before
programming. Call it before the vblank_control_workqueue flush.
Includes a drive-by cleanup of prepare_flip_isr(): the only caller is
dm_arm_vblank_event() and it's simple enough to roll-in.
v2: Fix checkpatch formatting warning on
drm_arm_vblank_event_pre_programming() arg alignment.
Fixes: 48ab86360a ("drm/amd/display: check GRPH_FLIP status before sending event")
Cc: stable@vger.kernel.org
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/4141#note_3583205
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5527
Assisted-by: Codex:gpt-5.6-sol
Assisted-by: Claude:opus-5
Suggested-by: David Weber <weber.aulendorf@gmail.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 05984e2952)
APUs use firmware-owned DPM tables and do not support replacement through
pp_table. Generic callbacks can nevertheless expose the sysfs file and
accept an upload before resetting the power management stack.
Treat pp_table as unsupported on APUs. Use the same platform check in the
get and set paths to hide the file and reject uploads.
Fixes: 289921b03f ("drm/amd/powerplay: implement sysfs of pp_table for smu11 (v2)")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 74f28db2db)
Cc: stable@vger.kernel.org
There are a few issues here:
1) After we assign:
chan = dma_request_chan(&pdev->dev, "tx");
Then the error paths need to clean up before returning. The first
error path does a direct return.
2) The error paths check "dcmi->mdma_chan" but that is not assigned
until later so it results in memory leaks. Test "mdma_chan"
instead.
3) The error handling calls dma_release_channel(dcmi->dma_chan) before
"dcmi->dma_chan" has been assigned which leads to a NULL pointer
dereference. Use the "chan" variable instead.
I also moved the call to dma_release_channel() after the call to
dma_release_channel() so it mirrors the allocation code better.
Fixes: bc901885fa ("media: stm32: dcmi: perform dmaengine_slave_config at probe")
Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Alain Volmat <alain.volmat@foss.st.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
The exposure and vertical blanking writes pass the address of the local
ret variable to cci_write() as its error pointer, but there is no earlier
error to propagate: each case is a single standalone write, like the other
controls in the same switch that already pass NULL. In the exposure case
ret is still uninitialized, so a non-zero stack value makes cci_write()
return early without programming the register, and the control write
reports a bogus status. The vertical blanking case is benign today because
ret is zero there, but the construct is equally wrong.
Pass NULL as the error pointer in both cases.
Fixes: be1589e567 ("media: i2c: imx471: Add Sony IMX471 image sensor driver")
Suggested-by: Kate Hsuan <hpa@redhat.com>
Signed-off-by: David Carlier <devnexen@gmail.com>
Reviewed-by: Kate Hsuan <hpa@redhat.com>
Reviewed-by: Tarang Raval <tarang.raval@siliconsignals.io>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
The ISPIF_VFE_m_RDI_INTF_n_PACK_CFG_0_CID_c_PLAIN() macro
declares a parameter named 'c' but uses 'cid' in the macro
body instead.
Rename the parameter to match the identifier used in the
macro body and silence the checkpatch warning:
WARNING: Argument 'c' is not used in function-like macro
No functional change intended.
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Shikra contains the same Camera Subsystem IP as QCM2290. Document the
platform-specific compatible string, using qcom,qcm2290-camss as
fallback.
Unlike QCM2290, Shikra omits the CDM and OPE blocks, requiring only a
single IOMMU context bank instead of four.
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Reviewed-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Signed-off-by: Nihal Kumar Gupta <nihal.gupta@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
The iris encoder driver was not sending
HFI_PROP_TIME_DELTA_BASED_RATE_CONTROL to the firmware during encoder
initialization. Without this property, the firmware defaults to
time-delta-based rate control (enabled), which calculates the output
bitrate from actual frame timing rather than following the configured
bitrate target.
This caused variable bitrate (VBR) encoding to produce ~5x configured
bitrate. For example, with video_bitrate=896000 (896 Kbps), the output
is ~4.4 Mbps instead of the expected ~896 Kbps.
Time-delta-based rate control is designed for variable frame rate (VFR)
scenarios where the encoder adapts to actual frame timing. However, when
an application explicitly configures a bitrate target, the firmware must
follow that target regardless of frame timing.
Fix this by adding the TIME_DELTA_BASED_RC capability with a default value
of 0 (disabled) and sending HFI_PROP_TIME_DELTA_BASED_RATE_CONTROL = 0 to
the firmware during stream-on, allowing the firmware to use the configured
bitrate as the target.
Signed-off-by: Gourav Kumar <gouravk@qti.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Reviewed-by: Vikash Garodia <vikash.garodia@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Add support for the Milos Iris codec. This only supports the variant
found on the SM7635-AB that has half of it's pipes disabled via efuse.
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Add binding for Qualcomm Milos Iris video codec.
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Some Iris platforms support both Gen1 and Gen2 HFI firmware images.
Update the firmware loading logic to handle this generically by
preferring Gen2 when available, while safely falling back to Gen1
when required.
The firmware loading logic is updated with the following priority:
1. Device Tree (`firmware-name`): If specified, load unconditionally.
2. Gen2 default : If no DT override exists, select the Gen2 firmware
descriptor when present and attempt to load the corresponding
firmware image.
3. Gen1 Fallback: If loading the Gen2 firmware fails and a Gen1
descriptor is available, retry with the Gen1 firmware image.
When a platform provides both Gen1 and Gen2 firmware descriptors and the
firmware is loaded via a DT override, the driver detects the
firmware generation at runtime before authentication by inspecting
the firmware data. The firmware is classified as Gen2 if the
QC_IMAGE_VERSION_STRING starts with "vfw" or matches the
"video-firmware.N.M" format with N >= 2.
If a Gen1 firmware image is detected in this case, the driver switches
to the Gen1 firmware descriptor and associated platform data so that
the correct HFI implementation is used.
This change makes firmware generation detection platform‑agnostic,
preserves DT overrides, prefers newer Gen2 firmware when available,
and maintains compatibility with platforms that only support Gen1.
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Reviewed-by: Vikash Garodia <vikash.garodia@oss.qualcomm.com>
Co-developed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Add __counted_by() annotations to flexible array members:
- hfi_capabilities::data, counted by num_capabilities
- hfi_profile_level_supported::profile_level, counted by profile_count
- hfi_resource_ocmem_requirement_info::requirements, counted by num_entries
This improves run-time bounds checking via CONFIG_UBSAN_BOUNDS and
compile-time object size resolution via __builtin_dynamic_object_size().
Assisted-by: Claude:claude-opus-4
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Mohammed EL Kadiri <med08elkadiri@gmail.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
The IRQ is registered as a threaded IRQ.
Using disable_irq_nosync() in iris_vpu_power_off() does not wait
for an already queued threaded IRQ handler to complete before
returning.
As a result, a threaded IRQ handler may still run after the VPU has
been powered down and access hardware registers after power-off.
Replace disable_irq_nosync() with disable_irq() so the power-off path
waits for any in-flight threaded IRQ handler to complete before
returning.
Fixes: bb8a95aa03 ("media: iris: implement power management")
Cc: stable@vger.kernel.org
Suggested-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Initialize the managed OPP pointer at declaration rather than
assigning it in a separate statement.
This avoids a checkpatch warning about a potentially uninitialized
managed pointer.
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Check the return value of pm_runtime_resume_and_get() in
iris_core_deinit().
If runtime PM resume fails, skip hardware power-off operations but
still perform software teardown and state transition. Also skip the
corresponding pm_runtime_put_sync() call to avoid unbalanced runtime
PM references.
Fixes: bb8a95aa03 ("media: iris: implement power management")
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
commit 66c744e28b ("media: venus: assign unique bus_info strings for
encoder and decoder") introduced the naming convention
plat:node-addr:video-codec{enc|dec}. Right now Iris does not replicate this
naming convention.
When we do v4l2-ctrl --list -devices we see:
Iris Decoder (platform:aa00000.video-codec):
/dev/video0
/dev/video1
Enumerate the bus_info field of the capabilities structure for namespace
parity and appropriate differentiation:
Iris Decoder (plat:aa00000.video-codec:dec):
/dev/video0
Iris Encoder (plat:aa00000.video-codec:enc):
/dev/video1
Fixes: 5ad964ad56 ("media: iris: Initialize and deinitialize encoder instance structure")
Cc: stable@vger.kernel.org
Signed-off-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
iris_destroy_internal_buffer() is guaranteed to succeed and never
reports an error. Returning an int is misleading and forces callers to
handle a meaningless status value. Convert it to return void to match
its behavior and simplify callers.
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Without the break the AV1 case falls through, risking unintended behaviour
if new cases are added after it.
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
HFI_PROP_OPB_ENABLE/iris_hfi_gen2_set_opb_enable appeared twice in the
dispatch table, causing the property to be sent to firmware twice on every
config-params call.
Fixes: 2af481a459 ("media: iris: Define AV1-specific platform capabilities and properties")
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Fold the ternary initialiser directly into the variable declaration,
removing the dead store that was immediately overwritten.
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
iris_allow_cmd() incorrectly checks a sub‑state flag using a logical
equality comparison. Since sub_state is a bitmask, this allows STOP to
pass when IRIS_INST_SUB_DRAIN is set alongside other bits, violating the
intended drain semantics. Fix this by using a proper bitmask test.
Fixes: d09100763b ("media: iris: add support for drain sequence")
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Internal buffer table dispatch is duplicated across multiple Iris code
paths, which is error‑prone and makes future changes harder to reason
about.
Consolidate the buffer dispatch logic into a single helper so that table
selection is defined in exactly one place and keep call sites minimal.
No functional change intended.
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
The consumed size is computed after the loop using the num_planes value
from the last iteration for all entries. When entries have different
plane counts, this produces an incorrect total.
Accumulate the actual size during the loop instead.
Fixes: 9edaaa8e3e ("media: venus: hfi_parser: refactor hfi packet parsing logic")
Cc: stable@vger.kernel.org
Signed-off-by: Mohammed EL Kadiri <med08elkadiri@gmail.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
parse_caps() and parse_alloc_mode() return only the size of their fixed
header fields, excluding the flexible array payload. hfi_parser() uses
this return value to advance through the firmware response buffer, so
underreporting causes parser desynchronization.
Return the full consumed size (header + entries), matching the correct
pattern used by parse_profile_level().
Fixes: 9edaaa8e3e ("media: venus: hfi_parser: refactor hfi packet parsing logic")
Cc: stable@vger.kernel.org
Signed-off-by: Mohammed EL Kadiri <med08elkadiri@gmail.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Add hfi_id to gen1 encoder GOP_SIZE cap and replace the set function,
remove the redundant INTRA_PERIOD cap.
Fixes: d22037f3fd ("media: iris: Set platform capabilities to firmware for encoder video device")
Reviewed-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Signed-off-by: Wangao Wang <wangao.wang@oss.qualcomm.com>
Reviewed-by: Vikash Garodia <vikash.garodia@oss.qualcomm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
Use pm_runtime_resume_and_get() in iris_enable_power_domains()
to avoid leaking a runtime PM usage count on failure.
Also ensure pm_runtime_put_sync() is always called in
iris_disable_power_domains(), even when iris_opp_set_rate()
fails, so runtime PM references remain balanced.
Fixes: bb8a95aa03 ("media: iris: implement power management")
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Hungyu Lin <dennylin0707@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
The state‑change debug log in iris_inst_change_state() always prints the
same value for the old and new state, rendering it useless for
debugging. This happens because the state is updated before the log is
emitted.
Log the transition before updating the state so the previous value is
preserved, consistent with the existing sub‑state handling.
Fixes: 11712ce70f ("media: iris: implement vb2 streaming ops")
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
Signed-off-by: Vishnu Reddy <busanna.reddy@oss.qualcomm.com>
Cc: stable@vger.kernel.org
Signed-off-by: Bryan O'Donoghue <bod@kernel.org>
The functions gfx_v11_0_handle_priv_fault and
gfx_v12_0_handle_priv_fault share the same logic for searching and
triggering a scheduler fault on a ring. This patch moves the shared
ring-searching logic to a common function, amdgpu_gfx_handle_priv_fault,
in amdgpu_gfx.c. The hardware-specific decoding of ring IDs remains in
the version-specific files to maintain proper architectural separation.
Signed-off-by: Ulisses Paixao <ulissespaixao@usp.br>
Co-developed-by: Felipe Sousa <felipesousa@usp.br>
Signed-off-by: Felipe Sousa <felipesousa@usp.br>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Enable IP block soft reset as a GPU recovery method for GFX7
graphics and compute rings.
This improves current user experience on all GFX7 chips:
* On Kaveri and Kabini there is currently no working
GPU recovery method so those chips currently require
the user to manually reset the computer when there
was a hang.
* On Hawaii and Bonaire, the current GPU recovery method
always clears the contents of VRAM, which means that
a buggy (hanging) app can crash the whole graphical
session, which is less than ideal.
Using GFX IP block soft reset means that we can now
have a working recovery on GFX7 APUs and we can also
move on from GFX hangs on dGPUs without crashing the
whole system.
Tested with the "hard_reset_cp_wait" test case from the
Hang Test Suite created by Natalie Vock and Konstantin Seurer.
This Vulkan testcase waits for an event that never occurs,
effectively a WAIT_REG_MEM packet that intentionally hangs.
IP block soft reset can resolve that hang and allow
the rest of the system to move on and keep functioning
without needing a full ASIC reset.
Tested on the following chips:
Bonaire (Radeon HD 7790)
Hawaii (Radeon R9 390X)
Kaveri (A10-7850K)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use basically the same implementation as GFX8,
except for the GFX7 specific MQD functions.
Reset every block using the GRBM, then proceed
to reset the GRBM and SEM blocks using the SRBM.
Remove the redundant gfx_v7_0_update_cg() function.
The soft reset now calls the clock and powergating
functions of the IP block instead.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
COND_EXEC tells the CP to discard the dwords following it
when its condition is zero (false).
This is useful for GPU recovery because it can help reduce
collateral damage during GFX IP block soft reset, meaning
that it reduces the likelyhood that we fail some jobs which
are not guilty of the hang as the IP block soft reset
mechanism clears the condition before doing the reset.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clear the WPTR and RPTR at ring initialization.
Additionally clear the ring contents during reset.
After a reset, the ring contents could be "dirty"
and contain packets emitted before the reset.
and thus need to be cleared to prevent the command
processor from executing packets left over in the
ring from before the reset.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This packet is interpreted by the CE (constant engine).
The reason why this packet is emitted is basically to
make sure the CE can't start executing packets from the
next job submission until the current one is finished.
(Note that CE is not utilized by any maintained userspace
driver and is discontinued in new GPUs. It is now also
deprecated in the kernel.)
Implement the emit_switch_buffer() function instead of emitting
them duing emit_ib, emit_pipeline_sync and emit_vm_flush.
It isn't necessary to emit these in both emit_pipeline_sync()
and emit_vm_flush() because amdgpu_vm_flush() already calls
these when calling either of those functions.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Return an error code instead of silently failing.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The gfx_v7_0_cp_compute_resume() function should only return
success when all compute rings are actually functional.
This will be especially important for soft reset which needs
this to know whether the reset was successful.
Note that the gfx_v8_0_cp_test_all_rings() function already
does this on GFX8, here we just follow the same idea.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Call amdgpu_gfx_mqd_sw_init()/_fini() on GFX7 to initialize and
finalize MQD BOs, just like GFX8 and newer; instead of doing
an ad-hoc BO allocation. Introduce the possibility of backing
up the MQD instead of trying to reinitialize every time.
This solves an issue with GFX IP block soft reset where
all compute rings would hang after the reset.
Rename gfx_v7_0_mqd_deactivate() to gfx_v7_0_deactivate_hqd()
to more closely reflect what it does and for consistency
with the GFX8 code.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GFX7 supports KIQ, but amdgpu doesn't use it.
Change amdgpu_gfx_mqd_sw_init() to only allocate the MQD BO
for the KIQ on GFX8 and newer (that is, TOPAZ and newer).
This makes amdgpu_gfx_mqd_sw_init() usable on GFX7 without
any further changes to its functionality.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
We need to exit PSR/IPS before programming. Before calling DC for
programming in amdgpu_dm_commit_planes(), there's a
vblank_control_workqueue flush. This waits for IPS and PSR exit. (See
drm_vblank_on/off() > amdgpu_dm_crtc_set_vblank() --queue_work()->
amdgpu_dm_crtc_vblank_control_worker())
Prior to the tagged "Fixes:" change, drm_vblank_get() was called before
the workqueue flush. This ordering ensures that PSR exit occurred before
programming. After the "Fixes:" change, drm_vblank_get() is called after
the workqueue flush, leading to programming while idle optimizations are
still active. This can lead to incorrect flip_pending detection used by
vblank event delivery.
[How]
Split the vblank_get() component of `dm_arm_vblank_event()` into
`dm_arm_vblank_event_pre_programming()`, which is called before
programming. Call it before the vblank_control_workqueue flush.
Includes a drive-by cleanup of prepare_flip_isr(): the only caller is
dm_arm_vblank_event() and it's simple enough to roll-in.
v2: Fix checkpatch formatting warning on
drm_arm_vblank_event_pre_programming() arg alignment.
Fixes: f64a9be565 ("drm/amd/display: check GRPH_FLIP status before sending event")
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/4141#note_3583205
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5527
Assisted-by: Codex:gpt-5.6-sol
Assisted-by: Claude:opus-5
Suggested-by: David Weber <weber.aulendorf@gmail.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SVM BOs are now migrated back to system memory synchronously from the TTM
eviction path (svm_range_evict_svm_bo), so the per-svm_bo eviction fence
is no longer used.
Remove the eviction fence from svm_range_bo, drop the
amdgpu_amdkfd_fence->svm_bo back pointer and the
amdgpu_amdkfd_evict_svm_bo() helper, and stop special-casing svm_bo
fences in the KFD fence enable_signaling and check_mm paths. Embed struct
amdgpu_bo directly in svm_range_bo with a dedicated svm_range_bo_destroy()
callback, and keep the owning mm via mmgrab()/mmdrop() instead of through
the fence.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
APUs use firmware-owned DPM tables and do not support replacement through
pp_table. Generic callbacks can nevertheless expose the sysfs file and
accept an upload before resetting the power management stack.
Treat pp_table as unsupported on APUs. Use the same platform check in the
get and set paths to hide the file and reject uploads.
Fixes: 289921b03f ("drm/amd/powerplay: implement sysfs of pp_table for smu11 (v2)")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
