drm/virtio: Fix return value for VIRTGPU_CONTEXT_PARAM_DEBUG_NAME

The strncpy_from_user() returns number of copied bytes and not zero on
success. The non-zero return value of ioctl is treated as error. Return
zero on success instead of the number of copied bytes.

Fixes: 7add80126bce ("drm/uapi: add explicit virtgpu context debug name")
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20231111224236.890431-1-dmitry.osipenko@collabora.com

iosys-map: Rename locals used inside macros

Widely used variable names can be used by macro users, potentially
leading to name collisions.
Suffix locals used inside the macros with an underscore, to reduce the
collision potential.

Suggested-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231024110710.3039807-1-michal.winiarski@intel.com

drm: Fix flip-task docs

Say that drm_flip_work_commit() is safe to call in atomic context. Turn
the name into a hyperlink.

Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20231101103618.23806-3-tzimmermann@suse.de

drm: Remove struct drm_flip_task from DRM interfaces

Contain struct drm_flip_task and its helper functions
drm_flip_work_allocate_task() and drm_flip_work_queue_task() within
drm_flip_work.c There are no callers outside of the flip-work code.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20231101103618.23806-2-tzimmermann@suse.de

drm/ssd130x: Preallocate format-conversion buffer in atomic_check

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

v6:
* update patch for ssd132x support

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-8-tzimmermann@suse.de

drm/simpledrm: Preallocate format-conversion buffer in atomic_check

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

Also inline drm_plane_helper_atomic_check() into the driver and thus
return early for invisible planes. Avoids memory allocation entirely
in this case.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-6-tzimmermann@suse.de

drm/ofdrm: Preallocate format-conversion buffer in atomic_check

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-5-tzimmermann@suse.de

drm/format-helper: Pass format-conversion state to helpers

Pass an instance of struct drm_format_conv_state to DRM's format
conversion helpers. Update all callers.

Most drivers can use the format-conversion state from their shadow-
plane state. The shadow plane's destroy function releases the
allocated buffer. Drivers will later be able to allocate a buffer
of appropriate size in their plane's atomic_check code.

The gud driver uses a separate thread for committing updates. For
now, the update worker contains its own format-conversion state.

Images in the format-helper tests are small. The tests preallocate
a static page for the temporary buffer. Unloading the module releases
the memory.

v6:
* update patch for ssd132x support
v5:
* avoid using unusupported shadow-plane state in repaper (Noralf)
* fix documentation (Noralf, kernel test robot)
v3:
* store buffer in shadow-plane state (Javier, Maxime)
* replace ARRAY_SIZE() with sizeof() (Jani)

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Javier Martinez Canillas <javierm@redhat.com> # ssd130x
Cc: Noralf Trønnes <noralf@tronnes.org>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: David Lechner <david@lechnology.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-4-tzimmermann@suse.de

drm/atomic-helper: Add format-conversion state to shadow-plane state

Store an instance of struct drm_format_conv_state in the shadow-plane
state struct drm_shadow_plane_state. Many drivers with shadow planes
use DRM's format helpers to copy or convert the framebuffer data to
backing storage in the scanout buffer. The shadow plane provides the
necessary state and manages the conversion's intermediate buffer memory.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-3-tzimmermann@suse.de

drm/format-helper: Cache buffers with struct drm_format_conv_state

Hold temporary memory for format conversion in an instance of struct
drm_format_conv_state. Update internal helpers of DRM's format-conversion
code accordingly. Drivers will later be able to maintain this cache by
themselves.

Besides caching, struct drm_format_conv_state will be useful to hold
additional information for format conversion, such as palette data or
foreground/background colors. This will enable conversion from indexed
color formats to component-based formats.

v5:
* improve documentation (Javier, Noralf)
v3:
* rename struct drm_xfrm_buf to struct drm_format_conv_state
  (Javier)
* remove managed cleanup
* add drm_format_conv_state_copy() for shadow-plane support

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231009141018.11291-2-tzimmermann@suse.de

Revert "drm/sched: Define pr_fmt() for DRM using pr_*()"

From Jani:
The drm_print.[ch] facilities use very few pr_*() calls directly. The
users of pr_*() calls do not necessarily include <drm/drm_print.h> at
all, and really don't have to.

Even the ones that do include it, usually have <linux/...> includes
first, and <drm/...> includes next. Notably, <linux/kernel.h> includes
<linux/printk.h>.

And, of course, <linux/printk.h> defines pr_fmt() itself if not already
defined.

No, it's encouraged not to use pr_*() at all, and prefer drm device
based logging, or device based logging.

This reverts commit 36245bd02e88e68ac5955c2958c968879d7b75a9.

Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/878r75wzm9.fsf@intel.com
Acked-by: Javier Martinez Canillas <javierm@redhat.com>
Reviewed-by: Jani Nikula <jani.nikula@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231111024130.11464-2-ltuikov89@gmail.com

drm/gpuvm: track/lock/validate external/evicted objects

Currently the DRM GPUVM offers common infrastructure to track GPU VA
allocations and mappings, generically connect GPU VA mappings to their
backing buffers and perform more complex mapping operations on the GPU VA
space.

However, there are more design patterns commonly used by drivers, which
can potentially be generalized in order to make the DRM GPUVM represent
a basis for GPU-VM implementations. In this context, this patch aims
at generalizing the following elements.

1) Provide a common dma-resv for GEM objects not being used outside of
   this GPU-VM.

2) Provide tracking of external GEM objects (GEM objects which are
   shared with other GPU-VMs).

3) Provide functions to efficiently lock all GEM objects dma-resv the
   GPU-VM contains mappings of.

4) Provide tracking of evicted GEM objects the GPU-VM contains mappings
   of, such that validation of evicted GEM objects is accelerated.

5) Provide some convinience functions for common patterns.

Big thanks to Boris Brezillon for his help to figure out locking for
drivers updating the GPU VA space within the fence signalling path.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-12-dakr@redhat.com

drm/gpuvm: add an abstraction for a VM / BO combination

Add an abstraction layer between the drm_gpuva mappings of a particular
drm_gem_object and this GEM object itself. The abstraction represents a
combination of a drm_gem_object and drm_gpuvm. The drm_gem_object holds
a list of drm_gpuvm_bo structures (the structure representing this
abstraction), while each drm_gpuvm_bo contains list of mappings of this
GEM object.

This has multiple advantages:

1) We can use the drm_gpuvm_bo structure to attach it to various lists
   of the drm_gpuvm. This is useful for tracking external and evicted
   objects per VM, which is introduced in subsequent patches.

2) Finding mappings of a certain drm_gem_object mapped in a certain
   drm_gpuvm becomes much cheaper.

3) Drivers can derive and extend the structure to easily represent
   driver specific states of a BO for a certain GPUVM.

The idea of this abstraction was taken from amdgpu, hence the credit for
this idea goes to the developers of amdgpu.

Cc: Christian König <christian.koenig@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-11-dakr@redhat.com

drm/gpuvm: reference count drm_gpuvm structures

Implement reference counting for struct drm_gpuvm.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-10-dakr@redhat.com

drm/nouveau: separately allocate struct nouveau_uvmm

Allocate struct nouveau_uvmm separately in preparation for subsequent
commits introducing reference counting for struct drm_gpuvm.

While at it, get rid of nouveau_uvmm_init() as indirection of
nouveau_uvmm_ioctl_vm_init() and perform some minor cleanups.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-9-dakr@redhat.com

drm/gpuvm: add drm_gpuvm_flags to drm_gpuvm

Introduce flags for struct drm_gpuvm, this required by subsequent
commits.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-8-dakr@redhat.com

drm/nouveau: make use of the GPUVM's shared dma-resv

DRM GEM objects private to a single GPUVM can use a shared dma-resv.
Make use of the shared dma-resv of GPUVM rather than a driver specific
one.

The shared dma-resv originates from a "root" GEM object serving as
container for the dma-resv to make it compatible with drm_exec.

In order to make sure the object proving the shared dma-resv can't be
freed up before the objects making use of it, let every such GEM object
take a reference on it.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-7-dakr@redhat.com

drm/gpuvm: add common dma-resv per struct drm_gpuvm

Provide a common dma-resv for GEM objects not being used outside of this
GPU-VM. This is used in a subsequent patch to generalize dma-resv,
external and evicted object handling and GEM validation.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-6-dakr@redhat.com

drm/nouveau: make use of drm_gpuvm_range_valid()

Use drm_gpuvm_range_valid() in order to validate userspace requests.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-5-dakr@redhat.com

drm/gpuvm: export drm_gpuvm_range_valid()

Drivers may use this function to validate userspace requests in advance,
hence export it.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-4-dakr@redhat.com

drm/gpuvm: don't always WARN in drm_gpuvm_check_overflow()

Don't always WARN in drm_gpuvm_check_overflow() and separate it into a
drm_gpuvm_check_overflow() and a dedicated
drm_gpuvm_warn_check_overflow() variant.

This avoids printing warnings due to invalid userspace requests.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-3-dakr@redhat.com

drm/gpuvm: convert WARN() to drm_WARN() variants

Use drm_WARN() and drm_WARN_ON() variants to indicate drivers the
context the failing VM resides in.

Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231108001259.15123-2-dakr@redhat.com

drm/uapi: add explicit virtgpu context debug name

There are two problems with the current method of determining the
virtio-gpu debug name.

1) TASK_COMM_LEN is defined to be 16 bytes only, and this is a
   Linux kernel idiom (see PR_SET_NAME + PR_GET_NAME). Though,
   Android/FreeBSD get around this via setprogname(..)/getprogname(..)
   in libc.

   On Android, names longer than 16 bytes are common.  For example,
   one often encounters a program like "com.android.systemui".

   The virtio-gpu spec allows the debug name to be up to 64 bytes, so
   ideally userspace should be able to set debug names up to 64 bytes.

2) The current implementation determines the debug name using whatever
   task initiated virtgpu.  This is could be a "RenderThread" of a
   larger program, when we actually want to propagate the debug name
   of the program.

To fix these issues, add a new CONTEXT_INIT param that allows userspace
to set the debug name when creating a context.

It takes a null-terminated C-string as the param value. The length of the
string (excluding the terminator) **should** be <= 64 bytes.  Otherwise,
the debug_name will be truncated to 64 bytes.

Link to open-source userspace:
https://android-review.googlesource.com/c/platform/hardware/google/gfxstream/+/2787176

Signed-off-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Josh Simonot <josh.simonot@gmail.com>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231018181727.772-2-gurchetansingh@chromium.org

drm/virtio: use uint64_t more in virtio_gpu_context_init_ioctl

drm_virtgpu_context_set_param defines both param and
value to be u64s.

Signed-off-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Josh Simonot <josh.simonot@gmail.com>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231018181727.772-1-gurchetansingh@chromium.org

drm/panfrost: Set regulators on/off during system sleep on MediaTek SoCs

All of the MediaTek SoCs supported by Panfrost can completely cut power
to the GPU during full system sleep without any user-noticeable delay
in the resume operation, as shown by measurements taken on multiple
MediaTek SoCs (MT8183/86/92/95).

As an example, for MT8195 - a "before" with only runtime PM operations
(so, without turning on/off regulators), and an "after" executing both
the system sleep .resume() handler and .runtime_resume() (so the time
refers to T_Resume + T_Runtime_Resume):

Average Panfrost-only system sleep resume time, before: ~33500ns
Average Panfrost-only system sleep resume time, after:  ~336200ns

Keep in mind that this additional ~308200 nanoseconds delay happens only
in resume from a full system suspend, and not in runtime PM operations,
hence it is acceptable.

Measurements were also taken on MT8186, showing a delay of ~312000 ns.

Testing of this happened on all of the aforementioned MediaTek SoCs, but:
MT8183 got tested only by KernelCI with <=10 suspend/resume cycles
MT8186, MT8192, MT8195 were tested manually with over 100 suspend/resume
cycles with GNOME DE (Mutter + Wayland).

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-7-angelogioacchino.delregno@collabora.com

drm/panfrost: Implement ability to turn on/off regulators in suspend

Some platforms/SoCs can power off the GPU entirely by completely cutting
off power, greatly enhancing battery time during system suspend: add a
new pm_feature GPU_PM_VREG_OFF to allow turning off the GPU regulators
during full suspend only on selected platforms.

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-6-angelogioacchino.delregno@collabora.com

drm/panfrost: Set clocks on/off during system sleep on MediaTek SoCs

All of the MediaTek SoCs supported by Panfrost can switch the clocks
off and on during system sleep to save some power without any user
experience penalty.

Measurements taken on multiple MediaTek SoCs (MT8183/8186/8192/8195)
show that adding this will not prolong the time that is required to
resume the system in any meaningful way.

As an example, for MT8195 - a "before" with only runtime PM operations
(so, without turning on/off GPU clocks), and an "after" executing both
the system sleep .resume() handler and .runtime_resume() (so the time
refers to T_Resume + T_Runtime_Resume):

Average Panfrost-only system sleep resume time, before: ~28000ns
Average Panfrost-only system sleep resume time, after:  ~33500ns

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-5-angelogioacchino.delregno@collabora.com

drm/panfrost: Implement ability to turn on/off GPU clocks in suspend

Currently, the GPU is being internally powered off for runtime suspend
and turned back on for runtime resume through commands sent to it, but
note that the GPU doesn't need to be clocked during the poweroff state,
hence it is possible to save some power on selected platforms.

Add suspend and resume handlers for full system sleep and then add
a new panfrost_gpu_pm enumeration and a pm_features variable in the
panfrost_compatible structure: BIT(GPU_PM_CLK_DIS) will be used to
enable this power saving technique only on SoCs that are able to
safely use it.

Note that this was implemented only for the system sleep case and not
for runtime PM because testing on one of my MediaTek platforms showed
issues when turning on and off clocks aggressively (in PM runtime)
resulting in a full system lockup.

Doing this only for full system sleep never showed issues during my
testing by suspending and resuming the system continuously for more
than 100 cycles.

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-4-angelogioacchino.delregno@collabora.com

drm/panfrost: Tighten polling for soft reset and power on

In many cases, soft reset takes more than 1 microsecond, but definitely
less than 10; moreover in the poweron flow, tilers, shaders and l2 will
become ready (each) in less than 10 microseconds as well.

Even in the cases (at least on my platforms, rarely) in which those take
more than 10 microseconds, it's very unlikely to see both soft reset and
poweron to take more than 70 microseconds.

Shorten the polling delay to 10 microseconds to consistently reduce the
runtime resume time of the GPU.

As an indicative example, measurements taken on a MediaTek MT8195 SoC

Average runtime resume time in nanoseconds before this commit:
GDM, user selection up/down:            88435ns
GDM, Text Entry (typing user/password): 91489ns
GNOME Desktop, idling, GKRELLM running: 73200ns

After this commit:

GDM: user selection up/down:            26690ns
GDM: Text Entry (typing user/password): 27917ns
GNOME Desktop, idling, GKRELLM running: 25304ns

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-3-angelogioacchino.delregno@collabora.com

drm/panfrost: Perform hard reset to recover GPU if soft reset fails

Even though soft reset should ideally never fail, during development of
some power management features I managed to get some bits wrong: this
resulted in GPU soft reset failures, where the GPU was never able to
recover, not even after suspend/resume cycles, meaning that the only
way to get functionality back was to reboot the machine.

Perform a hard reset after a soft reset failure to be able to recover
the GPU during runtime (so, without any machine reboot).

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231109102543.42971-2-angelogioacchino.delregno@collabora.com

drm/panfrost: Really power off GPU cores in panfrost_gpu_power_off()

The layout of the registers {TILER,SHADER,L2}_PWROFF_LO, used to request
powering off cores, is the same as the {TILER,SHADER,L2}_PWRON_LO ones:
this means that in order to request poweroff of cores, we are supposed
to write a bitmask of cores that should be powered off!
This means that the panfrost_gpu_power_off() function has always been
doing nothing.

Fix powering off the GPU by writing a bitmask of the cores to poweroff
to the relevant PWROFF_LO registers and then check that the transition
(from ON to OFF) has finished by polling the relevant PWRTRANS_LO
registers.

While at it, in order to avoid code duplication, move the core mask
logic from panfrost_gpu_power_on() to a new panfrost_get_core_mask()
function, used in both poweron and poweroff.

Fixes: f3ba91228e8e ("drm/panfrost: Add initial panfrost driver")
Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102141507.73481-1-angelogioacchino.delregno@collabora.com

drm/sched: implement dynamic job-flow control

Currently, job flow control is implemented simply by limiting the number
of jobs in flight. Therefore, a scheduler is initialized with a credit
limit that corresponds to the number of jobs which can be sent to the
hardware.

This implies that for each job, drivers need to account for the maximum
job size possible in order to not overflow the ring buffer.

However, there are drivers, such as Nouveau, where the job size has a
rather large range. For such drivers it can easily happen that job
submissions not even filling the ring by 1% can block subsequent
submissions, which, in the worst case, can lead to the ring run dry.

In order to overcome this issue, allow for tracking the actual job size
instead of the number of jobs. Therefore, add a field to track a job's
credit count, which represents the number of credits a job contributes
to the scheduler's credit limit.

Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231110001638.71750-1-dakr@redhat.com

drm/sched: Define pr_fmt() for DRM using pr_*()

Define pr_fmt() as "[drm] " for DRM code using pr_*() facilities, especially
when no devices are available. This makes it easier to browse kernel logs.

Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>
Acked-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231110002659.113208-2-ltuikov89@gmail.com

drm/panel-edp: drm/panel-edp: Add several generic edp panels

Add a few generic edp panels used by mt8186 chromebooks.

Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Acked-by: Maxime Ripard <mripard@kernel.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20231107204611.3082200-4-hsinyi@chromium.org

drm/panel-edp: drm/panel-edp: Fix AUO B116XTN02 name

Rename AUO 0x235c B116XTN02 to B116XTN02.3 according to decoding edid.

Fixes: 3db2420422a5 ("drm/panel-edp: Add AUO B116XTN02, BOE NT116WHM-N21,836X2, NV116WHM-N49 V8.0")
Cc: stable@vger.kernel.org
Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Acked-by: Maxime Ripard <mripard@kernel.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20231107204611.3082200-3-hsinyi@chromium.org

drm/panel-edp: drm/panel-edp: Fix AUO B116XAK01 name and timing

Rename AUO 0x405c B116XAK01 to B116XAK01.0 and adjust the timing of
auo_b116xak01: T3=200, T12=500, T7_max = 50 according to decoding edid
and datasheet.

Fixes: da458286a5e2 ("drm/panel: Add support for AUO B116XAK01 panel")
Cc: stable@vger.kernel.org
Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Acked-by: Maxime Ripard <mripard@kernel.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20231107204611.3082200-2-hsinyi@chromium.org

drm/sched: Qualify drm_sched_wakeup() by drm_sched_entity_is_ready()

Don't "wake up" the GPU scheduler unless the entity is ready, as well as we
can queue to the scheduler, i.e. there is no point in waking up the scheduler
for the entity unless the entity is ready.

Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>
Fixes: bc8d6a9df99038 ("drm/sched: Don't disturb the entity when in RR-mode scheduling")
Reviewed-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231110000123.72565-2-ltuikov89@gmail.com

drm/eld: add helpers to modify the SADs of an ELD

Occasionally it's necessary for drivers to modify the SADs of an ELD,
but it's not so cool to have drivers poke at the ELD buffer directly.

Using the helpers to translate between 3-byte SAD and struct cea_sad,
add ELD helpers to get/set the SADs from/to an ELD.

v2: s/i/sad_index/ (Mitul)

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/8e9a05f2b1e0dd184132d636e1e778e8917ec25d.1698747331.git.jani.nikula@intel.com

drm/edid: add helpers to get/set struct cea_sad from/to 3-byte sad

Add helpers to pack/unpack SADs. Both ways and non-static, as follow-up
work needs them.

v2: Add include to get the declarations

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/21d657ca854ce26423b461c0bb71e7a0727ba437.1698747331.git.jani.nikula@intel.com

drm/edid: use a temp variable for sads to drop one level of dereferences

Use a temporary variable struct cea_sad *, instead of using struct
cea_sad ** directly with the double dereferences. It's arguably easier
on the eyes, and drops a set of parenthesis too.

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/b6e2f295ae5491c2bb0f528508f0f5fca921dc77.1698747331.git.jani.nikula@intel.com

drm/edid: include drm_eld.h only where required

Reduce the dependencies on drm_eld.h. Some files might be able to drop
the dependency on drm_edid.h too with the direct inclusion of drm_eld.h.

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/9f5963ce900d747f3279312c0cd1da599fd83f94.1698747331.git.jani.nikula@intel.com

drm/eld: replace uint8_t with u8

Unify on kernel types.

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/6e048fc4c8a3ebec638ce27b0b8b969a3d2fa8bc.1698747331.git.jani.nikula@intel.com

drm/edid: split out drm_eld.h from drm_edid.h

The drm_edid.[ch] files are starting to be a bit crowded, and with plans
to add more ELD related functionality, it's perhaps cleanest to split
the ELD code out to a header of its own.

Include drm_eld.h from drm_edid.h for starters, and leave it to
follow-up work to only include drm_eld.h where needed.

Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/0c6d631fa1058036d72dd25d1cabc90a7c52490e.1698747331.git.jani.nikula@intel.com

MAINTAINERS: Add Maira to V3D maintainers

I've been contributing to V3D with improvements, reviews, testing and
debugging. Therefore, add myself as a co-maintainer of the V3D driver.

Signed-off-by: Maíra Canal <mcanal@igalia.com>
Acked-by: Melissa Wen <mwen@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231106134201.725805-1-mcanal@igalia.com

drm/todo: Add entry to clean up former seltests suites

Most of those suites are undocumented and aren't really clear about what
they are testing. Let's add a TODO entry as a future task to get started
into KUnit and DRM.

Acked-by: Daniel Vetter <daniel@ffwll.ch>
Link: https://lore.kernel.org/r/20231025132428.723672-2-mripard@kernel.org
Signed-off-by: Maxime Ripard <mripard@kernel.org>

drm/tests: Remove slow tests

Both the drm_buddy and drm_mm tests have been converted from selftest to
kunit recently.

However, a significant portion of them are trying to exert some part of
their API over a huge number of iterations and with random variations of
their parameters.  They are thus more a way to discover new bugs than
actual unit tests.

This is fine in itself but leads to very slow runtime (up to 25s for
some drm_test_mm_reserve and drm_test_mm_insert on a Ryzen 7950x while
running the tests in qemu) which make them a poor fit for kunit.

Let's remove those tests from the drm_mm and drm_buddy test suites for
now, and if it's ever needed we can always create proper unit tests for
them later on.

This made the entire DRM tests execution time (as of v6.6-rc1) come from
65s to less than .5s on a Ryzen 7950x system when running under qemu,
and from 9 minutes to about 4s on a RaspberryPi4.

Acked-by: Daniel Vetter <daniel@ffwll.ch>
Suggested-by: Daniel Vetter <daniel@ffwll.ch>
Link: https://lore.kernel.org/r/20231025132428.723672-1-mripard@kernel.org
Signed-off-by: Maxime Ripard <mripard@kernel.org>

accel/ivpu: Use GEM shmem helper for all buffers

Use struct drm_gem_shmem_object as a base for struct ivpu_bo.
This cuts by 50% the buffer management code.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073156.1301669-5-stanislaw.gruszka@linux.intel.com

accel/ivpu: Remove support for uncached buffers

Usages of DRM_IVPU_BO_UNCACHED should be replaced by DRM_IVPU_BO_WC.
There is no functional benefit from DRM_IVPU_BO_UNCACHED if these
buffers are never mapped to host VM.

This allows to cut the buffer handling code in the kernel driver
by half.

Usage of DRM_IVPU_BO_UNCACHED buffers was removed from user-space
driver and will not be part of first UMD release.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073156.1301669-4-stanislaw.gruszka@linux.intel.com

accel/ivpu: Fix locking in ivpu_bo_remove_all_bos_from_context()

ivpu_bo_remove_all_bos_from_context() could race with ivpu_bo_free()
when prime buffer was closed after vpu device was closed.

Move the bo_list from context to vdev and use a dedicated lock to
sync it. This list is not modified when BO is added/removed from
a context.

Also rename ivpu_bo_free_vpu_addr() to ivpu_bo_unbind() because this
function does more then just free vpu_addr.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073156.1301669-3-stanislaw.gruszka@linux.intel.com

accel/ivpu: Allocate vpu_addr in gem->open() callback

Use gem->open() callback to simplify the code and prepare for gem_shmem
conversion. It is called during handle creation for a gem object,
during prime import and in BO_CREATE ioctl. Hence can be used for vpu_addr
allocation. On the way remove unused bo->user_ptr field.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073156.1301669-2-stanislaw.gruszka@linux.intel.com

MAINTAINERS: Drop Emma Anholt from all M lines.

I am not active in the Linux kernel and don't want to see patches.

Signed-off-by: Emma Anholt <emma@anholt.net>
Acked-by: Maíra Canal <mcanal@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031181648.48675-1-emma@anholt.net

drm/sched: Don't disturb the entity when in RR-mode scheduling

Don't call drm_sched_select_entity() in drm_sched_run_job_queue().  In fact,
rename __drm_sched_run_job_queue() to just drm_sched_run_job_queue(), and let
it do just that, schedule the work item for execution.

The problem is that drm_sched_run_job_queue() calls drm_sched_select_entity()
to determine if the scheduler has an entity ready in one of its run-queues,
and in the case of the Round-Robin (RR) scheduling, the function
drm_sched_rq_select_entity_rr() does just that, selects the _next_ entity
which is ready, sets up the run-queue and completion and returns that
entity. The FIFO scheduling algorithm is unaffected.

Now, since drm_sched_run_job_work() also calls drm_sched_select_entity(), then
in the case of RR scheduling, that would result in drm_sched_select_entity()
having been called twice, which may result in skipping a ready entity if more
than one entity is ready. This commit fixes this by eliminating the call to
drm_sched_select_entity() from drm_sched_run_job_queue(), and leaves it only
in drm_sched_run_job_work().

v2: Rebased on top of Tvrtko's renames series of patches. (Luben)
    Add fixes-tag. (Tvrtko)

Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>
Fixes: f7fe64ad0f22ff ("drm/sched: Split free_job into own work item")
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231107041020.10035-2-ltuikov89@gmail.com

accel/ivpu: Fix compilation with CONFIG_PM=n

Use pm_runtime_status_suspended() instead of dev->power.runtime_status
field that is not available without PM.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://lore.kernel.org/all/20231106130827.1600948-1-jacek.lawrynowicz@linux.intel.com

drm/panel: nt35510: fix typo

Replace 'HFP' with 'HBP'.

Fixes: 899f24ed8d3a ("drm/panel: Add driver for Novatek NT35510-based panels")
Signed-off-by: Dario Binacchi <dario.binacchi@amarulasolutions.com>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20231023090613.1694133-1-dario.binacchi@amarulasolutions.com

drm/v3d: Expose the total GPU usage stats on sysfs

The previous patch exposed the accumulated amount of active time per
client for each V3D queue. But this doesn't provide a global notion of
the GPU usage.

Therefore, provide the accumulated amount of active time for each V3D
queue (BIN, RENDER, CSD, TFU and CACHE_CLEAN), considering all the jobs
submitted to the queue, independent of the client.

This data is exposed through the sysfs interface, so that if the
interface is queried at two different points of time the usage percentage
of each of the queues can be calculated.

Co-developed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Acked-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Melissa Wen <mwen@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230905213416.1290219-3-mcanal@igalia.com

drm/v3d: Implement show_fdinfo() callback for GPU usage stats

This patch exposes the accumulated amount of active time per client
through the fdinfo infrastructure. The amount of active time is exposed
for each V3D queue: BIN, RENDER, CSD, TFU and CACHE_CLEAN.

In order to calculate the amount of active time per client, a CPU clock
is used through the function local_clock(). The point where the jobs has
started is marked and is finally compared with the time that the job had
finished.

Moreover, the number of jobs submitted to each queue is also exposed on
fdinfo through the identifier "v3d-jobs-<queue>".

Co-developed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Acked-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Melissa Wen <mwen@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230905213416.1290219-3-mcanal@igalia.com

drm: Do not round to megabytes for greater than 1MiB sizes in fdinfo stats

It is better not to lose precision and not revert to 1 MiB size
granularity for every size greater than 1 MiB.

Sizes in KiB should not be so troublesome to read (and in fact machine
parsing is I expect the norm here), they align with other api like
/proc/meminfo, and they allow writing tests for the interface without
having to embed drm.ko implementation knowledge into them. (Like knowing
that minimum buffer size one can use for successful verification has to be
1MiB aligned, and on top account for any pre-existing memory utilisation
outside of driver's control.)

But probably even more importantly I think that it is just better to show
the accurate sizes and not arbitrary lose precision for a little bit of a
stretched use case of eyeballing fdinfo text directly.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Rob Clark <robdclark@gmail.com>
Cc: Adrián Larumbe <adrian.larumbe@collabora.com>
Cc: steven.price@arm.com
Reviewed-by: Steven Price <steven.price@arm.com>
Link: https://lore.kernel.org/r/20230927133843.247957-2-tvrtko.ursulin@linux.intel.com
Signed-off-by: Maxime Ripard <mripard@kernel.org>

drm/sched: Drop suffix from drm_sched_wakeup_if_can_queue

Because a) helper is exported to other parts of the scheduler and
b) there isn't a plain drm_sched_wakeup to begin with, I think we can
drop the suffix and by doing so separate the intimiate knowledge
between the scheduler components a bit better.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Luben Tuikov <ltuikov89@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102105538.391648-6-tvrtko.ursulin@linux.intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Rename drm_sched_run_job_queue_if_ready and clarify kerneldoc

"If ready" is not immediately clear what it means - is the scheduler
ready or something else? Drop the suffix, clarify kerneldoc, and employ
the same naming scheme as in drm_sched_run_free_queue:

 - drm_sched_run_job_queue   - enqueues if there is something to enqueue
                               *and* scheduler is ready (can queue)
 - __drm_sched_run_job_queue - low-level helper to simply queue the job

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Luben Tuikov <ltuikov89@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102105538.391648-5-tvrtko.ursulin@linux.intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Rename drm_sched_free_job_queue to be more descriptive

The current name makes it sound like helper will free a queue, while what
it does is it enqueues the free job worker.

Rename it to drm_sched_run_free_queue to align with existing
drm_sched_run_job_queue.

Despite that creating an illusion there are two queues, while in reality
there is only one, at least it creates a consistent naming for the two
enqueuing helpers.

At the same time simplify the "if done" helper by dropping the suffix and
adding a double underscore prefix to the one which just enqueues.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Luben Tuikov <ltuikov89@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102105538.391648-4-tvrtko.ursulin@linux.intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Move free worker re-queuing out of the if block

Whether or not there are more jobs to clean up does not depend on the
existance of the current job, given both drm_sched_get_finished_job and
drm_sched_free_job_queue_if_done take and drop the job list lock.
Therefore it is confusing to make it read like there is a dependency.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Luben Tuikov <ltuikov89@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102105538.391648-3-tvrtko.ursulin@linux.intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Rename drm_sched_get_cleanup_job to be more descriptive

"Get cleanup job" makes it sound like helper is returning a job which will
execute some cleanup, or something, while the kerneldoc itself accurately
says "fetch the next _finished_ job". So lets rename the helper to be self
documenting.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Luben Tuikov <ltuikov89@gmail.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231102105538.391648-2-tvrtko.ursulin@linux.intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

accel/qaic: Support for 0 resize slice execution in BO

Add support to partially execute a slice which is resized to zero.
Executing a zero size slice in a BO should mean that there is no DMA
transfers involved but you should still configure doorbell and semaphores.

For example consider a BO of size 18K and it is sliced into 3 6K slices
and user calls partial execute ioctl with resize as 10K.
slice 0 - size is 6k and offset is 0, so resize of 10K will not cut short
          this slice hence we send the entire slice for execution.
slice 1 - size is 6k and offset is 6k, so resize of 10K will cut short this
          slice and only the first 4k should be DMA along with configuring
          doorbell and semaphores.
slice 2 - size is 6k and offset is 12k, so resize of 10k will cut short
          this slice and no DMA transfer would be involved but we should
          would configure doorbell and semaphores.

This change begs to change the behavior of 0 resize. Currently, 0 resize
partial execute ioctl behaves exactly like execute ioctl i.e. no resize.
After this patch all the slice in BO should behave exactly like slice 2 in
above example.

Refactor copy_partial_exec_reqs() to make it more readable and less
complex.

Signed-off-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231027164330.11978-1-quic_jhugo@quicinc.com

accel/qaic: Quiet array bounds check on DMA abort message

Current wrapper is right-sized to the message being transferred;
however, this is smaller than the structure defining message wrappers
since the trailing element is a union of message/transfer headers of
various sizes (8 and 32 bytes on 32-bit system where issue was
reported). Using the smaller header with a small message
(wire_trans_dma_xfer is 24 bytes including header) ends up being smaller
than a wrapper with the larger header. There are no accesses outside of
the defined size, however they are possible if the larger union member
is referenced.

Abort messages are outside of hot-path and changing the wrapper struct
would require a larger rewrite, so having the memory allocated to the
message be 8 bytes too big is acceptable.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202310182253.bcb9JcyJ-lkp@intel.com/
Signed-off-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231027180810.4873-1-quic_jhugo@quicinc.com

drm/v3d: add brcm,2712-v3d as a compatible V3D device

This is required to get the V3D module to load with Raspberry Pi 5.

Signed-off-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Stefan Wahren <wahrenst@gmx.net>
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073859.25298-5-itoral@igalia.com

dt-bindings: gpu: v3d: Add BCM2712's compatible

BCM2712, Raspberry Pi 5's SoC, contains a V3D core. So add its specific
compatible to the bindings.

Signed-off-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073859.25298-4-itoral@igalia.com

drm/v3d: fix up register addresses for V3D 7.x

This patch updates a number of register addresses that have
been changed in Raspberry Pi 5 (V3D 7.1) and updates the
code to use the corresponding registers and addresses based
on the actual V3D version.

Signed-off-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073859.25298-3-itoral@igalia.com

drm/v3d: update UAPI to match user-space for V3D 7.x

V3D 7.x takes a new parameter to configure TFU jobs that needs
to be provided by user space.

Signed-off-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231031073859.25298-2-itoral@igalia.com

fbdev/simplefb: Add support for generic power-domains

The simple-framebuffer device tree bindings document the power-domains
property, so make sure that simplefb supports it. This ensures that the
power domains remain enabled as long as simplefb is active.

v2: - remove unnecessary call to simplefb_detach_genpds() since that's
      already done automatically by devres
    - fix crash if power-domains property is missing in DT

Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231101172017.3872242-3-thierry.reding@gmail.com

fbdev/simplefb: Support memory-region property

The simple-framebuffer bindings specify that the "memory-region"
property can be used as an alternative to the "reg" property to define
the framebuffer memory used by the display hardware. Implement support
for this in the simplefb driver.

Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231101172017.3872242-2-thierry.reding@gmail.com

drm/sched: Add a helper to queue TDR immediately

Add a helper whereby a driver can invoke TDR immediately.

v2:
 - Drop timeout args, rename function, use mod delayed work (Luben)
v3:
 - s/XE/Xe (Luben)
 - present tense in commit message (Luben)
 - Adjust comment for drm_sched_tdr_queue_imm (Luben)
v4:
 - Adjust commit message (Luben)

Cc: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://lore.kernel.org/r/20231031032439.1558703-6-matthew.brost@intel.com
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Add drm_sched_start_timeout_unlocked helper

Also add a lockdep assert to drm_sched_start_timeout.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://lore.kernel.org/r/20231031032439.1558703-5-matthew.brost@intel.com
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Split free_job into own work item

Rather than call free_job and run_job in same work item have a dedicated
work item for each. This aligns with the design and intended use of work
queues.

v2:
   - Test for DMA_FENCE_FLAG_TIMESTAMP_BIT before setting
     timestamp in free_job() work item (Danilo)
v3:
  - Drop forward dec of drm_sched_select_entity (Boris)
  - Return in drm_sched_run_job_work if entity NULL (Boris)
v4:
  - Replace dequeue with peek and invert logic (Luben)
  - Wrap to 100 lines (Luben)
  - Update comments for *_queue / *_queue_if_ready functions (Luben)
v5:
  - Drop peek argument, blindly reinit idle (Luben)
  - s/drm_sched_free_job_queue_if_ready/drm_sched_free_job_queue_if_done (Luben)
  - Update work_run_job & work_free_job kernel doc (Luben)
v6:
  - Do not move drm_sched_select_entity in file (Luben)

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://lore.kernel.org/r/20231031032439.1558703-4-matthew.brost@intel.com
Reviewed-by: Luben Tuikov <ltuikov89@gmail.com>
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Convert drm scheduler to use a work queue rather than kthread

In Xe, the new Intel GPU driver, a choice has made to have a 1 to 1
mapping between a drm_gpu_scheduler and drm_sched_entity. At first this
seems a bit odd but let us explain the reasoning below.

1. In Xe the submission order from multiple drm_sched_entity is not
guaranteed to be the same completion even if targeting the same hardware
engine. This is because in Xe we have a firmware scheduler, the GuC,
which allowed to reorder, timeslice, and preempt submissions. If a using
shared drm_gpu_scheduler across multiple drm_sched_entity, the TDR falls
apart as the TDR expects submission order == completion order. Using a
dedicated drm_gpu_scheduler per drm_sched_entity solve this problem.

2. In Xe submissions are done via programming a ring buffer (circular
buffer), a drm_gpu_scheduler provides a limit on number of jobs, if the
limit of number jobs is set to RING_SIZE / MAX_SIZE_PER_JOB we get flow
control on the ring for free.

A problem with this design is currently a drm_gpu_scheduler uses a
kthread for submission / job cleanup. This doesn't scale if a large
number of drm_gpu_scheduler are used. To work around the scaling issue,
use a worker rather than kthread for submission / job cleanup.

v2:
  - (Rob Clark) Fix msm build
  - Pass in run work queue
v3:
  - (Boris) don't have loop in worker
v4:
  - (Tvrtko) break out submit ready, stop, start helpers into own patch
v5:
  - (Boris) default to ordered work queue
v6:
  - (Luben / checkpatch) fix alignment in msm_ringbuffer.c
  - (Luben) s/drm_sched_submit_queue/drm_sched_wqueue_enqueue
  - (Luben) Update comment for drm_sched_wqueue_enqueue
  - (Luben) Positive check for submit_wq in drm_sched_init
  - (Luben) s/alloc_submit_wq/own_submit_wq
v7:
  - (Luben) s/drm_sched_wqueue_enqueue/drm_sched_run_job_queue
v8:
  - (Luben) Adjust var names / comments

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://lore.kernel.org/r/20231031032439.1558703-3-matthew.brost@intel.com
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

drm/sched: Add drm_sched_wqueue_* helpers

Add scheduler wqueue ready, stop, and start helpers to hide the
implementation details of the scheduler from the drivers.

v2:
  - s/sched_wqueue/sched_wqueue (Luben)
  - Remove the extra white line after the return-statement (Luben)
  - update drm_sched_wqueue_ready comment (Luben)

Cc: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Link: https://lore.kernel.org/r/20231031032439.1558703-2-matthew.brost@intel.com
Signed-off-by: Luben Tuikov <ltuikov89@gmail.com>

dma-buf: add dma_fence_timestamp helper

When a fence signals there is a very small race window where the timestamp
isn't updated yet. sync_file solves this by busy waiting for the
timestamp to appear, but on other ocassions didn't handled this
correctly.

Provide a dma_fence_timestamp() helper function for this and use it in
all appropriate cases.

Another alternative would be to grab the spinlock when that happens.

v2 by teddy: add a wait parameter to wait for the timestamp to show up, in case
   the accurate timestamp is needed and/or the timestamp is not based on
   ktime (e.g. hw timestamp)
v3 chk: drop the parameter again for unified handling

Signed-off-by: Yunxiang Li <Yunxiang.Li@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Fixes: 1774baa64f93 ("drm/scheduler: Change scheduled fence track v2")
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
CC: stable@vger.kernel.org
Link: https://patchwork.freedesktop.org/patch/msgid/20230929104725.2358-1-christian.koenig@amd.com

accel/ivpu: Rename VPU to NPU in product strings

VPU was rebranded as NPU (Neural Processing Unit) so user facing
strings have to be updated but the code remains as is and the module
is still called intel_vpu.ko.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-9-stanislaw.gruszka@linux.intel.com

accel/ivpu: Simplify MMU SYNC command

CMD_SYNC does not need any args as we poll for completion anyway.

Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-8-stanislaw.gruszka@linux.intel.com

accel/ivpu: Make DMA allocations for MMU600 write combined

Previously using dma_alloc_wc() API we created cache coherent
(mapped as write-back) mappings.

Because we disable MMU600 snooping it was required to do costly
page walk and cache flushes after each page table modification.

With write-combined buffers it's possible to do a single write memory
barrier to flush write-combined buffer to memory which simplifies the
driver and significantly reduce time of map/unmap operations.

Mapping time of 255 MB is reduced from 2.5 ms to 500 us.

Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-7-stanislaw.gruszka@linux.intel.com

accel/ivpu: Print CMDQ errors after consumer timeout

Add checking of error reason bits in IVPU_MMU_CMDQ_CONS
register when waiting for consumer timeout occurred.

Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-6-stanislaw.gruszka@linux.intel.com

accel/ivpu: Abort pending rx ipc on reset

Waking up process, which wait for particular condition, will go to
sleep again on wake_up() if the condition is not met. Add abort flag
to wake up IPC receivers, which will finish with -ECANCELED error.

This is only needed for reset, run time power management prevent to
suspend VPU when there is pending IPC processing or pending job.

Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-5-stanislaw.gruszka@linux.intel.com

accel/ivpu: Stop job_done_thread on suspend

Stop job_done thread when going to suspend. Use kthread_park() instead
of kthread_stop() to avoid memory allocation and potential failure
on resume.

Use separate function as thread wake up condition. Use spin lock to assure
rx_msg_list is properly protected against concurrent access. This avoid
race condition when the rx_msg_list list is modified and read in
ivpu_ipc_recive() at the same time.

Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-4-stanislaw.gruszka@linux.intel.com

accel/ivpu: Assure device is off if power up sequence fail

We should not leave device half enabled if there is failure somewhere
it power up sequence. Fix device init and resume paths.

Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-3-stanislaw.gruszka@linux.intel.com

accel/ivpu/40xx: Allow to change profiling frequency

Profiling freq is a debug firmware feature. It switches default clock
to higher resolution for fine-grained and more accurate firmware task
profiling. We already configure it during boot up of VPU4.

Add debugfs knob and helpers per HW generation that allow to change it.
For vpu37xx the implementation is empty as profiling frequency can only
be changed on VPU4 or newer.

Signed-off-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028155936.1183342-2-stanislaw.gruszka@linux.intel.com

Revert "drm/omapdrm: Annotate dma-fence critical section in commit path"

This reverts commit 250aa22920cd5d956a5d3e9c6a43d671c2bae217.

The DMA-fence annotations cause a lockdep warning (see below). As per
https://patchwork.freedesktop.org/patch/462170/ it sounds like the
annotations don't work correctly.

======================================================
WARNING: possible circular locking dependency detected
6.5.0-rc2+ #2 Not tainted
------------------------------------------------------
kmstest/219 is trying to acquire lock:
c4705838 (&hdmi->lock){+.+.}-{3:3}, at: hdmi5_bridge_mode_set+0x1c/0x50

but task is already holding lock:
c11e1128 (dma_fence_map){++++}-{0:0}, at: omap_atomic_commit_tail+0x14/0xbc

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (dma_fence_map){++++}-{0:0}:
       __dma_fence_might_wait+0x48/0xb4
       dma_resv_lockdep+0x1b8/0x2bc
       do_one_initcall+0x68/0x3b0
       kernel_init_freeable+0x260/0x34c
       kernel_init+0x14/0x140
       ret_from_fork+0x14/0x28

-> #1 (fs_reclaim){+.+.}-{0:0}:
       fs_reclaim_acquire+0x70/0xa8
       __kmem_cache_alloc_node+0x3c/0x368
       kmalloc_trace+0x28/0x58
       _drm_do_get_edid+0x7c/0x35c
       hdmi5_bridge_get_edid+0xc8/0x1ac
       drm_bridge_connector_get_modes+0x64/0xc0
       drm_helper_probe_single_connector_modes+0x170/0x528
       drm_client_modeset_probe+0x208/0x1334
       __drm_fb_helper_initial_config_and_unlock+0x30/0x548
       omap_fbdev_client_hotplug+0x3c/0x6c
       drm_client_register+0x58/0x94
       pdev_probe+0x544/0x6b0
       platform_probe+0x58/0xbc
       really_probe+0xd8/0x3fc
       __driver_probe_device+0x94/0x1f4
       driver_probe_device+0x2c/0xc4
       __device_attach_driver+0xa4/0x11c
       bus_for_each_drv+0x84/0xdc
       __device_attach+0xac/0x20c
       bus_probe_device+0x8c/0x90
       device_add+0x588/0x7e0
       platform_device_add+0x110/0x24c
       platform_device_register_full+0x108/0x15c
       dss_bind+0x90/0xc0
       try_to_bring_up_aggregate_device+0x1e0/0x2c8
       __component_add+0xa4/0x174
       hdmi5_probe+0x1c8/0x270
       platform_probe+0x58/0xbc
       really_probe+0xd8/0x3fc
       __driver_probe_device+0x94/0x1f4
       driver_probe_device+0x2c/0xc4
       __device_attach_driver+0xa4/0x11c
       bus_for_each_drv+0x84/0xdc
       __device_attach+0xac/0x20c
       bus_probe_device+0x8c/0x90
       deferred_probe_work_func+0x8c/0xd8
       process_one_work+0x2ac/0x6e4
       worker_thread+0x30/0x4ec
       kthread+0x100/0x124
       ret_from_fork+0x14/0x28

-> #0 (&hdmi->lock){+.+.}-{3:3}:
       __lock_acquire+0x145c/0x29cc
       lock_acquire.part.0+0xb4/0x258
       __mutex_lock+0x90/0x950
       mutex_lock_nested+0x1c/0x24
       hdmi5_bridge_mode_set+0x1c/0x50
       drm_bridge_chain_mode_set+0x48/0x5c
       crtc_set_mode+0x188/0x1d0
       omap_atomic_commit_tail+0x2c/0xbc
       commit_tail+0x9c/0x188
       drm_atomic_helper_commit+0x158/0x18c
       drm_atomic_commit+0xa4/0xe8
       drm_mode_atomic_ioctl+0x9a4/0xc38
       drm_ioctl+0x210/0x4a8
       sys_ioctl+0x138/0xf00
       ret_fast_syscall+0x0/0x1c

other info that might help us debug this:

Chain exists of:
  &hdmi->lock --> fs_reclaim --> dma_fence_map

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  rlock(dma_fence_map);
                               lock(fs_reclaim);
                               lock(dma_fence_map);
  lock(&hdmi->lock);

 *** DEADLOCK ***

3 locks held by kmstest/219:
 #0: f1011de4 (crtc_ww_class_acquire){+.+.}-{0:0}, at: drm_mode_atomic_ioctl+0xf0/0xc38
 #1: c47059c8 (crtc_ww_class_mutex){+.+.}-{3:3}, at: modeset_lock+0xf8/0x230
 #2: c11e1128 (dma_fence_map){++++}-{0:0}, at: omap_atomic_commit_tail+0x14/0xbc

stack backtrace:
CPU: 1 PID: 219 Comm: kmstest Not tainted 6.5.0-rc2+ #2
Hardware name: Generic DRA74X (Flattened Device Tree)
 unwind_backtrace from show_stack+0x10/0x14
 show_stack from dump_stack_lvl+0x58/0x70
 dump_stack_lvl from check_noncircular+0x164/0x198
 check_noncircular from __lock_acquire+0x145c/0x29cc
 __lock_acquire from lock_acquire.part.0+0xb4/0x258
 lock_acquire.part.0 from __mutex_lock+0x90/0x950
 __mutex_lock from mutex_lock_nested+0x1c/0x24
 mutex_lock_nested from hdmi5_bridge_mode_set+0x1c/0x50
 hdmi5_bridge_mode_set from drm_bridge_chain_mode_set+0x48/0x5c
 drm_bridge_chain_mode_set from crtc_set_mode+0x188/0x1d0
 crtc_set_mode from omap_atomic_commit_tail+0x2c/0xbc
 omap_atomic_commit_tail from commit_tail+0x9c/0x188
 commit_tail from drm_atomic_helper_commit+0x158/0x18c
 drm_atomic_helper_commit from drm_atomic_commit+0xa4/0xe8
 drm_atomic_commit from drm_mode_atomic_ioctl+0x9a4/0xc38
 drm_mode_atomic_ioctl from drm_ioctl+0x210/0x4a8
 drm_ioctl from sys_ioctl+0x138/0xf00
 sys_ioctl from ret_fast_syscall+0x0/0x1c
Exception stack(0xf1011fa8 to 0xf1011ff0)
1fa0:                   00466d58 be9ab510 00000003 c03864bc be9ab510 be9ab4e0
1fc0: 00466d58 be9ab510 c03864bc 00000036 00466ef0 00466fc0 00467020 00466f20
1fe0: b6bc7ef4 be9ab4d0 b6bbbb00 b6cb2cc0

Fixes: 250aa22920cd ("drm/omapdrm: Annotate dma-fence critical section in commit path")
Reviewed-by: Aradhya Bhatia <a-bhatia1@ti.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230920-dma-fence-annotation-revert-v1-2-7ebf6f7f5bf6@ideasonboard.com

Revert "drm/tidss: Annotate dma-fence critical section in commit path"

This reverts commit 4d56a4f08391857ba93465de489707b66adad114.

The DMA-fence annotations cause a lockdep warning (see below). As per
https://patchwork.freedesktop.org/patch/462170/ it sounds like the
annotations don't work correctly.

======================================================
WARNING: possible circular locking dependency detected
6.6.0-rc2+ #1 Not tainted
------------------------------------------------------
kmstest/733 is trying to acquire lock:
ffff8000819377f0 (fs_reclaim){+.+.}-{0:0}, at: __kmem_cache_alloc_node+0x58/0x2d4

but task is already holding lock:
ffff800081a06aa0 (dma_fence_map){++++}-{0:0}, at: tidss_atomic_commit_tail+0x20/0xc0 [tidss]

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 (dma_fence_map){++++}-{0:0}:
       __dma_fence_might_wait+0x5c/0xd0
       dma_resv_lockdep+0x1a4/0x32c
       do_one_initcall+0x84/0x2fc
       kernel_init_freeable+0x28c/0x4c4
       kernel_init+0x24/0x1dc
       ret_from_fork+0x10/0x20

-> #1 (mmu_notifier_invalidate_range_start){+.+.}-{0:0}:
       fs_reclaim_acquire+0x70/0xe4
       __kmem_cache_alloc_node+0x58/0x2d4
       kmalloc_trace+0x38/0x78
       __kthread_create_worker+0x3c/0x150
       kthread_create_worker+0x64/0x8c
       workqueue_init+0x1e8/0x2f0
       kernel_init_freeable+0x11c/0x4c4
       kernel_init+0x24/0x1dc
       ret_from_fork+0x10/0x20

-> #0 (fs_reclaim){+.+.}-{0:0}:
       __lock_acquire+0x1370/0x20d8
       lock_acquire+0x1e8/0x308
       fs_reclaim_acquire+0xd0/0xe4
       __kmem_cache_alloc_node+0x58/0x2d4
       __kmalloc_node_track_caller+0x58/0xf0
       kmemdup+0x34/0x60
       regmap_bulk_write+0x64/0x2c0
       tc358768_bridge_pre_enable+0x8c/0x12d0 [tc358768]
       drm_atomic_bridge_call_pre_enable+0x68/0x80 [drm]
       drm_atomic_bridge_chain_pre_enable+0x50/0x158 [drm]
       drm_atomic_helper_commit_modeset_enables+0x164/0x264 [drm_kms_helper]
       tidss_atomic_commit_tail+0x58/0xc0 [tidss]
       commit_tail+0xa0/0x188 [drm_kms_helper]
       drm_atomic_helper_commit+0x1a8/0x1c0 [drm_kms_helper]
       drm_atomic_commit+0xa8/0xe0 [drm]
       drm_mode_atomic_ioctl+0x9ec/0xc80 [drm]
       drm_ioctl_kernel+0xc4/0x170 [drm]
       drm_ioctl+0x234/0x4b0 [drm]
       drm_compat_ioctl+0x110/0x12c [drm]
       __arm64_compat_sys_ioctl+0x128/0x150
       invoke_syscall+0x48/0x110
       el0_svc_common.constprop.0+0x40/0xe0
       do_el0_svc_compat+0x1c/0x38
       el0_svc_compat+0x48/0xb4
       el0t_32_sync_handler+0xb0/0x138
       el0t_32_sync+0x194/0x198

other info that might help us debug this:

Chain exists of:
  fs_reclaim --> mmu_notifier_invalidate_range_start --> dma_fence_map

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  rlock(dma_fence_map);
                               lock(mmu_notifier_invalidate_range_start);
                               lock(dma_fence_map);
  lock(fs_reclaim);

 *** DEADLOCK ***

3 locks held by kmstest/733:
 #0: ffff800082e5bba0 (crtc_ww_class_acquire){+.+.}-{0:0}, at: drm_mode_atomic_ioctl+0x118/0xc80 [drm]
 #1: ffff000004224c88 (crtc_ww_class_mutex){+.+.}-{3:3}, at: modeset_lock+0xdc/0x1a0 [drm]
 #2: ffff800081a06aa0 (dma_fence_map){++++}-{0:0}, at: tidss_atomic_commit_tail+0x20/0xc0 [tidss]

stack backtrace:
CPU: 0 PID: 733 Comm: kmstest Not tainted 6.6.0-rc2+ #1
Hardware name: Toradex Verdin AM62 on Verdin Development Board (DT)
Call trace:
 dump_backtrace+0x98/0x118
 show_stack+0x18/0x24
 dump_stack_lvl+0x60/0xac
 dump_stack+0x18/0x24
 print_circular_bug+0x288/0x368
 check_noncircular+0x168/0x17c
 __lock_acquire+0x1370/0x20d8
 lock_acquire+0x1e8/0x308
 fs_reclaim_acquire+0xd0/0xe4
 __kmem_cache_alloc_node+0x58/0x2d4
 __kmalloc_node_track_caller+0x58/0xf0
 kmemdup+0x34/0x60
 regmap_bulk_write+0x64/0x2c0
 tc358768_bridge_pre_enable+0x8c/0x12d0 [tc358768]
 drm_atomic_bridge_call_pre_enable+0x68/0x80 [drm]
 drm_atomic_bridge_chain_pre_enable+0x50/0x158 [drm]
 drm_atomic_helper_commit_modeset_enables+0x164/0x264 [drm_kms_helper]
 tidss_atomic_commit_tail+0x58/0xc0 [tidss]
 commit_tail+0xa0/0x188 [drm_kms_helper]
 drm_atomic_helper_commit+0x1a8/0x1c0 [drm_kms_helper]
 drm_atomic_commit+0xa8/0xe0 [drm]
 drm_mode_atomic_ioctl+0x9ec/0xc80 [drm]
 drm_ioctl_kernel+0xc4/0x170 [drm]
 drm_ioctl+0x234/0x4b0 [drm]
 drm_compat_ioctl+0x110/0x12c [drm]
 __arm64_compat_sys_ioctl+0x128/0x150
 invoke_syscall+0x48/0x110
 el0_svc_common.constprop.0+0x40/0xe0
 do_el0_svc_compat+0x1c/0x38
 el0_svc_compat+0x48/0xb4
 el0t_32_sync_handler+0xb0/0x138
 el0t_32_sync+0x194/0x198

Fixes: 4d56a4f08391 ("drm/tidss: Annotate dma-fence critical section in commit path")
Reviewed-by: Aradhya Bhatia <a-bhatia1@ti.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230920-dma-fence-annotation-revert-v1-1-7ebf6f7f5bf6@ideasonboard.com

drm/panfrost: Remove incorrect IS_ERR() check

sg_page_iter_page() doesn't return an error code, so the IS_ERR() check
is wrong and the error path will never be executed. This also allows
simplifying the code to remove the local variable 'page'.

CC: Adrián Larumbe <adrian.larumbe@collabora.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/r/376713ff-9a4f-4ea3-b097-fb5efb685d95@moroto.mountain
Signed-off-by: Steven Price <steven.price@arm.com>
Reviewed-by: Adrián Larumbe <adrian.larumbe@collabora.com>
Tested-by: Adrián Larumbe <adrian.larumbe@collabora.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231020104405.53992-1-steven.price@arm.com

drm/v3d: wait for all jobs to finish before unregistering

Currently, we are only warning the user if the BIN or RENDER jobs don't
finish before we unregister V3D. We must wait for all jobs to finish
before unregistering. Therefore, warn the user if TFU or CSD jobs
are not done by the time the driver is unregistered.

Signed-off-by: Maíra Canal <mcanal@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Maíra Canal <mairacanal@riseup.net>
Link: https://patchwork.freedesktop.org/patch/msgid/20231023105927.101502-1-mcanal@igalia.com

accel/ivpu: Add support for delayed D0i3 entry message

Currently the VPU firmware prepares for D0i3 every time the VPU
is entering D0i2 Idle state. This is not optimal as we might not
enter D0i3 every time we enter D0i2 Idle and this preparation
is quite costly.

This optimization moves D0i3 preparation to a dedicated
message sent from the host driver only when the driver is about
to enter D0i3 - this reduces power consumption and latency for
certain workloads, for example audio workloads that submit
inference every 10 ms.

The VPU needs non zero time to enter IDLE state after responding to
D0i3 entry message. If the driver does not wait for the VPU to enter
IDLE state it could cause warm boot failures.

Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-12-stanislaw.gruszka@linux.intel.com

accel/ivpu/37xx: Print warning when VPUIP is not idle during power down

Print warning if VPUIP is not idle during power down.

Use warn log level also when we fail to enter reset state
as this is not really an error but unexpected behavior.

Reviewed-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-11-stanislaw.gruszka@linux.intel.com

accel/ivpu: Introduce ivpu_ipc_send_receive_active()

Split ivpu_ipc_send_receive() implementation to have a version
that does not call pm_runtime_resume_and_get(). That implementation
can be invoked when device is up and runtime resume is prohibited
(for example at the end of boot sequence).

The new function will be used for D0i3 entry IPC message addition
in the separate change.

Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-10-stanislaw.gruszka@linux.intel.com

accel/ivpu: Pass D0i3 residency time to the VPU firmware

The firmware needs to know the time spent in D0i3/D3 to
calculate telemetry data. The D0i3/D3 residency time is
calculated by the driver and passed to the firmware
in the boot parameters.

The driver also passes VPU perf counter value captured
right before entering D0i3 - this allows the VPU firmware
to generate monotonic timestamps for the logs.

Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-9-stanislaw.gruszka@linux.intel.com

accel/ivpu/40xx: Capture D0i3 entry host and device timestamps

The driver needs to capture the D0i3 entry timestamp to
calculate D0i3 residency time.

The D0i3 residency time and the VPU timestamp are passed
to the firmware at D0i3 exit (warm boot).

Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-8-stanislaw.gruszka@linux.intel.com

accel/ivpu: Change test_mode module param to bitmask

Change meaning of test_mode module parameter from integer value
to bitmask allowing setting different test features with corresponding
bits.

Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-7-stanislaw.gruszka@linux.intel.com

accel/ivpu: Add support for VPU_JOB_FLAGS_NULL_SUBMISSION_MASK

Add test_mode = 3 that add VPU_JOB_FLAGS_NULL_SUBMISSION_MASK
flag to the job send to the VPU device. Then the VPU will process
the job but won't execute commands (except the command to signal
the fence).

This can be used to estimate job processing overhead in the host
software and VPU firmware.

Unlike the null hardware mode, the null submission mode will
still work even if UMD uses VPU fences to track job completion.

Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-6-stanislaw.gruszka@linux.intel.com

accel/ivpu: Remove reset from power up sequence

Setting a non-zero work point resets the IP hence IP_RESET
trigger is redundant.

Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-5-stanislaw.gruszka@linux.intel.com

accel/ivpu: Add dvfs_mode file to debugfs

Add new debugfs file to set dvfs_mode FW boot parameter and restart
the FW to allow experimenting with DVFS (dynamic voltage & frequency
scaling).

Signed-off-by: Tomasz Rusinowicz <tomasz.rusinowicz@intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-4-stanislaw.gruszka@linux.intel.com

accel/ivpu: Remove unneeded drm_driver declaration

Cleanup drm_driver declaration leftover.

Reviewed-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-3-stanislaw.gruszka@linux.intel.com

accel/ivpu: Update FW API

Bump boot API to 4.20
Bump JSM API to 3.15

Signed-off-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231028133415.1169975-2-stanislaw.gruszka@linux.intel.com

accel/ivpu: avoid build failure with CONFIG_PM=n

The usage count of struct dev_pm_info is an implementation detail that
is only available if CONFIG_PM is enabled, so printing it in a debug message
causes a build failure in configurations without PM:

In file included from include/linux/device.h:15,
                 from include/linux/pci.h:37,
                 from drivers/accel/ivpu/ivpu_pm.c:8:
drivers/accel/ivpu/ivpu_pm.c: In function 'ivpu_rpm_get_if_active':
drivers/accel/ivpu/ivpu_pm.c:254:51: error: 'struct dev_pm_info' has no member named 'usage_count'
  254 |                  atomic_read(&vdev->drm.dev->power.usage_count));
      |                                                   ^
include/linux/dev_printk.h:129:48: note: in definition of macro 'dev_printk'
  129 |                 _dev_printk(level, dev, fmt, ##__VA_ARGS__);            \
      |                                                ^~~~~~~~~~~
drivers/accel/ivpu/ivpu_drv.h:75:17: note: in expansion of macro 'dev_dbg'
   75 |                 dev_dbg((vdev)->drm.dev, "[%s] " fmt, #type, ##args);          \
      |                 ^~~~~~~
drivers/accel/ivpu/ivpu_pm.c:253:9: note: in expansion of macro 'ivpu_dbg'
  253 |         ivpu_dbg(vdev, RPM, "rpm_get_if_active count %d\n",
      |         ^~~~~~~~

The print message does not seem essential, so the easiest workaround is
to just remove it.

Fixes: c39dc15191c4 ("accel/ivpu: Read clock rate only if device is up")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231027152633.528490-1-arnd@kernel.org