drm/xe/uc: Use devm to register cleanup that includes exec_queues

Exec_queue cleanup requires HW access, so we need to use devm instead of
drmm for it.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240815230541.3828206-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/uc: Use managed bo for HuC and GSC objects

Drmm actions are not the right ones to clean up BOs and we should use
devm instead. However, we can also instead just allocate the objects
using the managed_bo function, which will internally register the
correct cleanup call and therefore allows us to simplify the code.

While at it, switch to drmm_kzalloc for the GSC proxy allocation to
further simplify the cleanup.

Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240815230541.3828206-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Fix tile fini sequence

Only set tile->mmio.regs to NULL if not the root tile in tile_fini. The
root tile mmio regs is setup ealier in MMIO init thus it should be set
to NULL in mmio_fini.

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809232830.3302251-1-matthew.brost@intel.com

drm/xe: use devm instead of drmm for managed bo

The BO cleanup touches the GGTT and therefore requires the HW to be
available, so we need to use devm instead of drmm.

Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1160
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809231237.1503796-2-daniele.ceraolospurio@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Make exec_queue_kill safe to call twice

An upcoming PXP patch will kill queues at runtime when a PXP
invalidation event occurs, so we need exec_queue_kill to be safe to call
multiple times.

v2: Add documentation (Matt B)

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240814205654.1716586-1-daniele.ceraolospurio@intel.com

drm/xe: fix engine_class bounds check again

This was fixed in commit b7dce525c4fc ("drm/xe/queue: fix engine_class
bounds check"), but then re-introduced in commit 6f20fc09936e ("drm/xe:
Move and export xe_hw_engine lookup.") which should only be simple code
movement of the existing function.

Fixes: 6f20fc09936e ("drm/xe: Move and export xe_hw_engine lookup.")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240812141331.729843-2-matthew.auld@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Define STATELESS_COMPRESSION_CTRL as mcr register

Register STATELESS_COMPRESSION_CTRL should be considered
mcr register which should write to all slices as per
documentation.

Bspec: 71185
Fixes: ecabb5e6ce54 ("drm/xe/xe2: Add performance turning changes")
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Shekhar Chauhan <shekhar.chauhan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240814095614.909774-4-tejas.upadhyay@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Write all slices if its mcr register

Register GAMREQSTRM_CTRL should be considered mcr register
which should write to all slices as per documentation.

Bspec: 71185
Fixes: 01570b446939 ("drm/xe/bmg: implement Wa_16023588340")
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240814095614.909774-3-tejas.upadhyay@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Move enable host l2 VRAM post MCR init

xe_gt_enable_host_l2_vram() is reading the XE2_GAMREQSTRM_CTRL register
that is currently missing the MCR annotation. However, just adding the
annotation doesn't work as this function is called before MCR handling
is initialized in xe_gt_mcr_init().

xe_gt_enable_host_l2_vram() is used to implement WA 16023588340 that
needs to be done as early as possible during initialization in order to
be effective since the MMIO writes impact it. In the failure scenario,
driver would simply not be able to bind successfully.

Moving xe_gt_enable_host_l2_vram() later, after MCR initialization is
done, only incurs a few additional HW accesses, particularly when
loading GuC for hwconfig. Binding/unbinding the driver 100 times in BMG
still works so it should be ok to start handling the WA a little bit
later. This is sufficient to allow adding the MCR annotation to
XE2_GAMREQSTRM_CTRL.

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240814095614.909774-2-tejas.upadhyay@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Rename enable_display module param

The different approach used by xe regarding the initialization of
display HW has been proved a great addition for early driver bring up:
core xe can be tested without having all the bits sorted out on the
display side.

On the other hand, the approach exposed by i915-display is to *actively*
disable the display by programming it if needed, i.e. if it was left
enabled by firmware. It also has its use to make sure the HW is actually
disabled and not wasting power.

However having both the way it is in xe doesn't expose a good interface
wrt module params. From modinfo:

disable_display:Disable display (default: false) (bool)
enable_display:Enable display (bool)

Rename enable_display to probe_display to try to convey the message that
the HW is being touched and improve the module param description. To
avoid confusion, the enable_display is renamed everywhere, not only in
the module param. New description for the parameters:

disable_display:Disable display (default: false) (bool)
probe_display:Probe display HW, otherwise it's left untouched (default: true) (bool)

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240813141931.3141395-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Remove unused xe parameter

Remove the xe parameter from the pde_encode_pat_index and
pte_encode_pat_index functions, as it is no longer used.

Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240813104419.2958046-1-himal.prasad.ghimiray@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Name and document Wa_14019789679

Early in the development of Xe we identified an issue with SVG state
handling on DG2 and MTL (and later on Xe2 as well).  In
commit 72ac304769dd ("drm/xe: Emit SVG state on RCS during driver load
on DG2 and MTL") and commit fb24b858a20d ("drm/xe/xe2: Update SVG state
handling") we implemented our own workaround to prevent SVG state from
leaking from context A to context B in cases where context B never
issues a specific state setting.

The hardware teams have now created official workaround Wa_14019789679
to cover this issue.  The workaround description only requires emitting
3DSTATE_MESH_CONTROL, since they believe that's the only SVG instruction
that would potentially remain unset by a context B, but still cause
notable issues if unwanted values were inherited from context A.
However since we already have a more extensive implementation that emits
the entire SVG state and prevents _any_ SVG state from unintentionally
leaking, we'll stick with our existing implementation just to be safe.

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240812181042.2013508-2-matthew.d.roper@intel.com

drm/xe: add kdev_to_xe_device() helper and use it

There are enough users for kernel device to xe device conversion, add a
helper for it.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/38c80846e70c7e410850530426384e17cff9d031.1723458544.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>

drm/xe: use pdev_to_xe_device() instead of pci_get_drvdata() directly

We have a helper for converting pci device to xe device, use it.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1b87c2e56200e001ce3a5d2f4a93eb26b294df32.1723458544.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>

drm/xe/tests: remove unused leftover xe_call_for_each_device()

xe_call_for_each_device() has been unused since commit 57ecead343e7
("drm/xe/tests: Convert xe_mocs live tests"). Remove it and the related
dev_to_xe_device_fn() and struct kunit_test_data.

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/fa3bb23d005313c9797f557e1211fde09fcb59cc.1723458544.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>

drm/xe/migrate: Parameterize ccs and bo data clear in xe_migrate_clear()

Parameterize clearing ccs and bo data in xe_migrate_clear() which  higher
layers can utilize. This patch will be used later on when doing bo data
clear for igfx as well.

v2: Replace multiple params with flags in xe_migrate_clear (Matt B)
v3: s/CLEAR_BO_DATA_FLAG_*/XE_MIGRATE_CLEAR_FLAG_* and move to
    xe_migrate.h. other nits(Matt B)

Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809220347.25330-1-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe/xe2hpg: Add Wa_14021821874

Wa_14021821874 applies to xe2_hpg

V2(Himal):
  - Use space after define

Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240812134117.813670-1-tejas.upadhyay@intel.com

drm/xe: Add stats for tlb invalidation count

Add stats for tlb invalidation count which can be viewed with per GT
stat debugfs file.

Example output:
cat /sys/kernel/debug/dri/0/gt0/stats
tlb_inval_count: 22

v2: fix #include order(Tejas)

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Sai Gowtham Ch <sai.gowtham.ch@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240810191522.18616-2-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe/gt: Add APIs for printing stats over debugfs

Add skeleton APIs for recording and printing various stats over
debugfs. This currently only added counter types stats which is backed
by atomic_t and wrapped with CONFIG_DRM_XE_STATS so this can be disabled
on production system.

v4: Rebase and other minor fixes (Matt)
v3: s/CONFIG_DRM_XE_STATS/CONFIG_DEBUG_FS(Lucas)
v2: add missing docs
    Add boundary checks for stats id and other improvements (Michal)
    Fix build when CONFIG_DRM_XE_STATS is disabled(Matt)

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Sai Gowtham Ch <sai.gowtham.ch@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240810191522.18616-1-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe: Allow to compile out debugfs

Use a dummy xe_debugfs_register() if debugfs is not enabled and move all
debugfs-related files under `ifeq ($(CONFIG_DEBUG_FS),y)` in the
Makefile. This is similar to what was done for display in
commit 439987f6f471 ("drm/xe: don't build debugfs files when
CONFIG_DEBUG_FS=n").

This removes the following warning while loading xe with
CONFIG_DEUBG_FS=n:

xe 0000:03:00.0: [drm] Create GT directory failed

Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240808171121.2484237-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Allow suspend / resume to be safely called multiple times

Switching modes between LR and dma-fence can result in multiple calls to
suspend / resume. Make these calls safe while still enforcing call
order.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809191929.3138956-6-matthew.brost@intel.com

drm/xe: Only enable scheduling upon resume if needed

No need to enable scheduling in already enabled.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809191929.3138956-5-matthew.brost@intel.com

drm/xe: Add xe_sched_add_msg_locked helper

Will help by allowing callers to own message locking.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809191929.3138956-4-matthew.brost@intel.com

drm/xe: Reinit msg link when processing a message

Will help to avoid adding a static message twice.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809191929.3138956-3-matthew.brost@intel.com

drm/xe: Add xe_sched_msg_lock/unlock helper

Will help callers to own locking when adding messages to scheduler.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809191929.3138956-2-matthew.brost@intel.com

drm/xe: Move VM dma-resv lock from xe_exec_queue_create to __xe_exec_queue_init

The critical section which requires the VM dma-resv is the call
xe_lrc_create in __xe_exec_queue_init. Move this lock to
__xe_exec_queue_init holding it just around xe_lrc_create. Not only is
good practice, this also fixes a locking double of the VM dma-resv in
the error paths of __xe_exec_queue_init as xe_lrc_put tries to acquire
this too resulting in a deadlock.

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240724152831.1848325-1-matthew.brost@intel.com

drm/xe/pf: Fix VF config validation on multi-GT platforms

When validating VF config on the media GT, we may wrongly report
that VF is already partially configured on it, as we consider GGTT
and LMEM provisioning done on the primary GT (since both GGTT and
LMEM are tile-level resources, not a GT-level).

This will cause skipping a VF auto-provisioning on the media-GT and
in result will block a VF from successfully initialize that GT.

Fix that by considering GGTT and LMEM configurations only when
checking if a VF provisioning is complete, and omit GGTT and LMEM
when reporting empty/partial provisioning.

Fixes: 234670cea9a2 ("drm/xe/pf: Skip fair VFs provisioning if already provisioned")
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Piotr Piórkowski <piotr.piorkowski@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240806180516.618-1-michal.wajdeczko@intel.com

drm/xe: fix WA 14018094691

This WA is applied while initializing the media GT, but it a primary
GT WA (because it modifies a register on the primary GT), so the XE_WA
macro is returning false even when the WA should be applied.
Fix this by using the primary GT in the macro.

Note that this WA only applies to PXP and we don't yet support that in
Xe, so there are no negative effects to this bug, which is why we didn't
see any errors in testing.

v2: use the primary GT in the macro instead of marking the WA as
platform-wide (Lucas, Matt).

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240807235333.1370915-1-daniele.ceraolospurio@intel.com

drm/xe/guc: Enable w/a 14022293748 and 22019794406

Enable workarounds for HW bug where render engine reset fails. Given
that we're bumping the minimum required GuC version to 70.29, we're
guaranteed to always have support for this KLV in the GuC.

v2: Enable KLV correctly for either workaround (Lucas)
v4: Add check for minimum supported GuC firmware version. Enable w/a for
hw version 20.01 too. (Daniele)
v5 (Daniele): remove now unneeded fw type and version checks (JohnH)

Signed-off-by: Julia Filipchuk <julia.filipchuk@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240805205435.921921-1-daniele.ceraolospurio@intel.com

drm/xe: Fix access_ok check in user_fence_create

Check size of the data not size of the pointer.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202407300421.IBkAja96-lkp@intel.com/
Fixes: 0fde907da2d5 ("drm/xe: Validate user fence during creation")
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Apoorva Singh <apoorva.singh@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240806110722.28661-1-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe: Add kernel doc for xe_hw_engine_lookup

kerneldoc was missing from earlier commit where
we exported xe_hw_engine_lookup. Add it.

Cc: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Cc: intel-xe@lists.freedesktop.org
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240806153009.1081382-2-mika.kuoppala@linux.intel.com

drm/xe: Export xe_hw_engine's mmio accessors

Export hw engine's mmio accessors. This is in preparation
to use these from eudebug code.

v2: s/hw_engine_mmio/xe_hw_engine_mmio (Matthew)
v3: kernel doc (Matthew)

Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240806153009.1081382-1-mika.kuoppala@linux.intel.com

drm/xe/guc: Define GuC version v70.29.2 for BMG

UAPI version 1.13.4

Signed-off-by: Julia Filipchuk <julia.filipchuk@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240802222129.3976212-4-John.C.Harrison@Intel.com

drm/xe/guc: Bump minimum required GuC version to v70.29.2

The VF API version for this release is 1.13.4.

Bumping the minimum required GuC version just before force-probe
removal allows us to set a baseline for what API features are expected
to be available. I.e., at this point there is no need for any version
checking in the code before using a feature. Of course, if/when the
API is extended in future GuC releases, those new features will need
API version checks in the code.

Bump the recommended GuC versions to match.

Also add numerical comparison helpers to simplify the version number
checks.

v2: Reword commit message and make comparison helpers GuC specific -
review feedback from Daniele, done by JohnH

Signed-off-by: Julia Filipchuk <julia.filipchuk@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240802222129.3976212-3-John.C.Harrison@Intel.com

drm/xe/xe2: Add performance turning changes

Update performance tuning according to the hardware spec.

Bspec: 72161
Signed-off-by: Shekhar Chauhan <shekhar.chauhan@intel.com>
Reviewed-by: Sai Teja Pottumuttu <sai.teja.pottumuttu@intel.com>
Reviewed-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240805053710.877119-1-shekhar.chauhan@intel.com

drm/xe: Only check last fence on user binds

We only set the last fence on user binds, so no need to check last fence
kernel issued binds. Will avoid blowing up last fence lockdep asserts.

Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240805200233.3050325-1-matthew.brost@intel.com

drm/xe/observation: Drop empty sysctl table entry

An empty sysctl table entry was inadvertently left behind for observation
sysctl. The breaks on 6.11 with the following errors:

[  219.654850] sysctl table check failed: dev/xe/(null) procname is null
[  219.654862] sysctl table check failed: dev/xe/(null) No proc_handler

Drop the empty entry.

Fixes: 8169b2097d88 ("drm/xe/uapi: Rename xe perf layer as xe observation layer")
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2419
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240805062057.3547560-1-ashutosh.dixit@intel.com

drm/xe/xe2: Introduce performance changes

Add Compression Performance Improvement Changes in Xe2

v2: Rebase

v3: Rebase, updated as per latest changes on bspec,
    Removed unnecessary default actions (Matt)
    formatting nits (Tejas)

v4: Formatting nits, removed default set action for bit 14 (Matt)

Bspec: 72161
Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/c2dd753fdc55df6a6432026f2df9c2684a0d25c1.1722607628.git.akshata.jahagirdar@intel.com

drm/xe: Faster devcoredump

The current algorithm to read out devcoredump is O(N*N) where N is the
size of coredump due to usage of the drm_coredump_printer in
xe_devcoredump_read. Switch to a O(N) algorithm which prints the
devcoredump into a readable format in snapshot work and update
xe_devcoredump_read to memcpy from the readable format directly.

v2:
 - Fix double free on devcoredump removal (Testing)
 - Set read_data_size after snap work flush
 - Adjust remaining in iterator upon realloc (Testing)
 - Set read_data upon realloc (Testing)
v3:
 - Kernel doc
v4:
 - Two pass algorithm to determine size (Maarten)
v5:
 - Use scope for reading variables (Johnathan)

Reported-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2408
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240801154118.2547543-4-matthew.brost@intel.com

drm/printer: Allow NULL data in devcoredump printer

We want to determine the size of the devcoredump before writing it out.
To that end, we will run the devcoredump printer with NULL data to get
the size, alloc data based on the generated offset, then run the
devcorecump again with a valid data pointer to print.  This necessitates
not writing data to the data pointer on the initial pass, when it is
NULL.

v5:
 - Better commit message (Jonathan)
 - Add kerenl doc with examples (Jani)

Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Acked-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240801154118.2547543-3-matthew.brost@intel.com

drm/xe: Take ref to VM in delayed snapshot

Kernel BO's don't take a ref to the VM, we need the VM for the
delayed snapshot, so take a ref to the VM in delayed snapshot.

v2:
 - Check for lrc_bo before taking a VM ref (CI)
 - Check lrc_bo->vm before taking / dropping a VM ref (CI)
 - Drop VM in xe_lrc_snapshot_free
v5:
 - Fix commit message wording (Johnathan)

Fixes: 47058633d9c5 ("drm/xe: Move lrc snapshot capturing to xe_lrc.c")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240801154118.2547543-2-matthew.brost@intel.com

drm/xe/hwmon: Fix PL1 disable flow in xe_hwmon_power_max_write

In xe_hwmon_power_max_write, for PL1 disable supported case, instead of
returning after PL1 disable, PL1 enable path was also being run.
Fixed it by returning after disable.

v2: Correct typo and grammar in commit message. (Jonathan)

Signed-off-by: Karthik Poosa <karthik.poosa@intel.com>
Fixes: fef6dd12b45a ("drm/xe/hwmon: Protect hwmon rw attributes with hwmon_lock")
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240801112424.1841766-1-karthik.poosa@intel.com

drm/xe/xe2: Enable Priority Mem Read

Enable feature to allow memory reads to take a priority memory path.
This will reduce latency on the read path, but may introduce read after
write (RAW) hazards as read and writes will no longer be ordered.

To avoid RAW hazards, SW can use the MI_MEM_FENCE command or any other
MI command that generates non posted memory writes.  This will ensure
data is coherent in memory prior to execution of commands which read
data from memory. RCS,BCS and CCS support this feature.

No pattern identified in KMD that could lead to a hazard.

v2: Modify commit message, enable priority mem read feature for media,
modify version range, modify bspec detail (Matt Roper)

v3: Rebase, fix cramped line-wrapping (jcavitt)

v4: Rebase

v5: Media does not support Priority Mem Read. Modify commit
to reflect the same.

v6: Rebase

Bspec: 60298, 60237, 60187, 60188

Signed-off-by: Pallavi Mishra <pallavi.mishra@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: José Roberto de Souza <jose.souza@intel.com>
Acked-by: Carl Zhang <carl.zhang@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240731195622.1868401-1-pallavi.mishra@intel.com

drm/xe: Move and export xe_hw_engine lookup.

Move and export xe_hw_engine lookup. This is in preparation
to use this in eudebug code where we want to find active
engine.

v2: s/tile/gt due to uapi changes (Mika)

Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240729130152.100130-1-mika.kuoppala@linux.intel.com

drm/xe: Use dma_fence_chain_free in chain fence unused as a sync

A chain fence is uninitialized if not installed in a drm sync obj. Thus
if xe_sync_entry_cleanup is called and sync->chain_fence is non-NULL the
proper cleanup is dma_fence_chain_free rather than a dma-fence put.

Reported-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2411
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2261
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727012216.2118276-1-matthew.brost@intel.com

drm/xe/oa/uapi: Make bit masks unsigned

When building with gcc-5:

    In function ‘decode_oa_format.isra.26’,
inlined from ‘xe_oa_set_prop_oa_format’ at drivers/gpu/drm/xe/xe_oa.c:1664:6:
    ././include/linux/compiler_types.h:510:38: error: call to ‘__compiletime_assert_1336’ declared with attribute error: FIELD_GET: mask is not constant
    [...]
    ./include/linux/bitfield.h:155:3: note: in expansion of macro ‘__BF_FIELD_CHECK’
       __BF_FIELD_CHECK(_mask, _reg, 0U, "FIELD_GET: "); \
       ^
    drivers/gpu/drm/xe/xe_oa.c:1573:18: note: in expansion of macro ‘FIELD_GET’
      u32 bc_report = FIELD_GET(DRM_XE_OA_FORMAT_MASK_BC_REPORT, fmt);
      ^

Fixes: b6fd51c62119 ("drm/xe/oa/uapi: Define and parse OA stream properties")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240729092634.2227611-1-geert+renesas@glider.be
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/xe2hpg: Introduce performance tuning changes for Xe2_HPG

Add performance tuning changes for Xe2_HPG

Bspec: 72161
Signed-off-by: Sai Teja Pottumuttu <sai.teja.pottumuttu@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240724121521.2347524-1-sai.teja.pottumuttu@intel.com

drm/xe: Migrate OOB WAs to OR rules

Now that rtp has OR rules, it's not needed to extend it to process OOB
WAs. Previously if an entry had no name, it was considered as "a set of
rules OR'ed with the last named entry".

Instead of generating new entries, add OR rules. The syntax for
xe_wa_oob.rules remains the same, with xe_gen_wa_oob generating the
slightly different table. Object sizes delta are negligible, but having
just one logic makes it easier to maintain:

add/remove: 0/0 grow/shrink: 1/2 up/down: 160/-269 (-109)
Function                                     old     new   delta
__compound_literal                          6104    6264    +160
xe_wa_dump                                  1839    1810     -29
oob_was                                      816     576    -240
Total: Before=17257, After=17148, chg -0.63%

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-9-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/rtp: Expand max rules/actions per entry again

Like commit 512660cd1f1a ("drm/xe/rtp: Expand max rules/actions per
entry") did, expand the maximum number of actions/rules. That commit was
too conservative, just incrementing 2. Other than the ugliness of these
macros and additional preprocessor steps when they are used, there are
no downsides on increasing the maximum: the tables in which they are
used use a sentinel to mark the last element.

With rtp processing now supporting OR rules, it's possible to migrate
the extension made for OOB WAs that "entries with name are OR'ed in
previous entry". For that the maximum number of rules needs to be
increased.

Just double it. Hopefully 12 is sufficient for longer than 6 was.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-8-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/rtp: Simplify marking active workarounds

Stop doing the calculation both in rtp_mark_active() and in its caller.
The caller easily knows the number of entries to mark, so just pass it
forward. That also simplifies rtp_mark_active() since now it doesn't
have a special case when handling 1 entry.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-7-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/kunit: Test rtp with no actions

The OOB WAs use xe_rtp_process(), without passing an sr to save result
of the actions since there are none. They are also executed in a gt-only
context, making it harder to share the implementation. Thus, introduce a
new set of tests to check these RTP entries. The only check that can be
done is if the entry was marked as active.

Before commit fd6797ec50c5 ("drm/xe/rtp: Fix off-by-one when processing
rules") several of these tests were failing: the processing of OR'ed
entries would make the subsequent entry to be inadvertently enabled.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-6-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/kunit: Rename rtp test cases

Those tests check the behavior of xe_rtp_process_to_sr(), so name them
accordingly to allow adding tests for xe_rtp_process() later.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-5-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/kunit: Test active rtp entries

Enabling active tracking in the rtp context and check for all the tests
the expected entries become active.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-4-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/kunit: Rename count to count_sr_entries

The RTP tests check both the result of processing the RTP entries and
the outcome saved as SR entries. Rename "count" to be explicit about
what's being counted.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-3-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/kunit: Test WAs for BMG

Add one variant for BMG to make sure the workarounds do not conflict.
This matches the machine with BMG in CI:

        BATTLEMAGE e20b:0000 dgfx:1
gfx:Xe2_LPG / Xe2_HPG (20.01)
media:Xe2_LPM / Xe2_HPM (13.01)
display:yes dma_m_s:46 tc:1 gscfi:0 cscfi:1
        Stepping = (G:A0, M:A1, D:**, B:**)

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240727015907.899192-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/migrate: Future-proof compressed PAT check

Although all current Xe2 platforms support FlatCCS, we probably
shouldn't assume that will be universally true forever.  In the past
we've had platforms like PVC that didn't support compression, and the
same could show up again at some point in the future.  Future-proof the
migration code by adding an explicit check for FlatCCS support to the
condition that decides whether to use a compressed PAT index for
migration.

While we're at it, we can drop the IS_DGFX check since it's redundant
with the src_is_vram check (only dGPUs have VRAM).

Cc: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240726171757.2728819-2-matthew.d.roper@intel.com

drm/xe/rtp: Fix off-by-one when processing rules

Gustavo noticed an odd "+ 2" in rtp_mark_active() while processing
rtp rules and pointed that it should be "+ 1". In fact, while processing
entries without actions (OOB workarounds), if the WA is activated and
has OR rules, it will also inadvertently activate the very next
workaround.

Test in a LNL B0 platform by moving 18024947630 on top of 16020292621,
makes the latter become active:

$ cat /sys/kernel/debug/dri/0/gt0/workarounds
...
OOB Workarounds
18024947630
16020292621
14018094691
16022287689
13011645652
22019338487_display

In future a kunit test will be added to cover the rtp checks for entries
without actions.

Fixes: fe19328b900c ("drm/xe/rtp: Add support for entries with no action")
Cc: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240726064337.797576-6-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Assert G2H outstanding when releasing G2H

Ensure we are managing G2H credits correctly. Extra important now that
this is tied to PM.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240725231801.1958038-1-matthew.brost@intel.com

drm/xe/mmio: Use single logic for waiting functions

The implementations for xe_mmio_wait32() and xe_mmio_wait32_not() are
almost identical. Let us avoid duplication of logic by having them
calling a common __xe_mmio_wait32() function.

Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723120120.5443-2-gustavo.sousa@intel.com

drm/xe: Remove stale declaration of xe_mmio_probe_vram()

The declaration of xe_mmio_probe_vram() became useless since
commit 638d1c79cbf1 ("drm/xe: Promote VRAM initialization function to
own file"). Remove it.

Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240722190524.71548-2-gustavo.sousa@intel.com

drm/xe/huc: Define HuC binary for BMG

Add the unversioned define for the BMG HuC FW.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240725003224.516233-7-daniele.ceraolospurio@intel.com

drm/xe/gsc: Define GSC binary for LNL

As with previous binaries, we match the compatibility version instead of
the build number.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240725003224.516233-6-daniele.ceraolospurio@intel.com

drm/xe/huc: Define HuC binary for LNL

Add the unversioned define for the LNL HuC FW.

All new binaries are GSC-enabled (and even if they weren't the driver
can detect the type of HuC binary), so the new lnl HuC filename doesn't
use the _gsc postfix to avoid confusion with the GSC binary.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240725003224.516233-5-daniele.ceraolospurio@intel.com

drm/xe: Fix possible UAF in guc_exec_queue_process_msg

Store xe_device ahead of processing message as message can be free'd in
some cases.

v2:
 - Including missing local changes
v3:
 - Resend for CI

Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/r/202407231445.rpisd1vA-lkp@intel.com/
Fixes: d930c19fdff3 ("drm/xe: Build PM into GuC CT layer")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240724164341.1848954-1-matthew.brost@intel.com

drm/xe: Delete unused register from xe_regs.h

Register SOFTWARE_FLAGS_SPR33 is unused; therefore, delete it.

Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240715105555.2220444-2-himal.prasad.ghimiray@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Add assert for XE_WA() usage

It's not always safe to call XE_WA() in the driver initialization. Add a
xe_gt_assert() so this doesn't go unnoticed.

While at it, fix typo in kernel-doc about OOB workarounds.

Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240722160815.4085605-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Refactor mmio setup for multi-tile

Extract functions to setup the multi-tile mmio space and extension
space, while better documenting the final memory layout.  No change in
behavior.

Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723045558.288693-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Remove fence check from send_tlb_invalidation

'fence' argument in send_tlb_invalidation cannot be NULL, remove
non-NULL check from send_tlb_invalidation.

Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/r/202407231049.esig0Fkb-lkp@intel.com/
Fixes: 61ac035361ae ("drm/xe: Drop xe_gt_tlb_invalidation_wait")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723190714.1744653-1-matthew.brost@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe: Store process name and pid in xe file

An xe file can outlive the associated process as the GPU cleanup is just
triggered upon file close (process kill) and completes sometime later.
If the file close triggers error conditions (GPU hangs) the process
cannot be safely referenced to retrieve the name and pid for debug
information. Store the process name and pid directly in the xe file to
be safe.

v2:
 - Access file->pid via rcu_access_pointer (Matthew Auld)

Fixes: b10d0c5e9df7 ("drm/xe: Add process name to devcoredump")
Fixes: f6ca930d974e ("drm/xe: Add process name and PID to job timedout message")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723151045.1725417-1-matthew.brost@intel.com

drm/xe: Return -ENOBUFS if a kmalloc fails which is tied to an array of binds

The size of an array of binds is directly tied to several kmalloc in the
KMD, thus making these kmalloc more likely to fail. Return -ENOBUFS in
the case of these failures.

The expected UMD behavior upon returning -ENOBUFS is to split an array
of binds into a series of single binds.

v2:
 - Resend for CI
v3:
 - Resend for CI

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723011702.1684013-1-matthew.brost@intel.com

drm/xe: Fix xe_pt_abort_unbind

When restoring the children PT entries on a bind failure the incorrect
loop index has used resulting in PT entries being leaked. This is shown
by running xe_vm.bind-array-conflict-error-inject on a VRAM device going
into a suspend state after the test completes.

v2:
 - s/childern/children (CI, Matt Auld)

Fixes: a708f6501c69 ("drm/xe: Update PT layer with better error handling")
Cc: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240723010230.1652707-1-matthew.brost@intel.com

drm/xe: Fix warning on unreachable statement

eu_type_to_str() relies on -Wswitch to warn (and -Werror) to make sure
it handles all enum values. However it's perfectly legal to pass an int
to that function so in the end that function may happen to return
nothing. There's too much implicit knowledge about the initialization
of eu_type for a compiler  to notice eu_type is never assigned to
anything other than those values.

Trying to reproduce this issue, none of gcc-9, gcc-10 and gcc-13
triggered for me, but this was reported in a different system with
gcc-10:

drivers/gpu/drm/xe/xe.o: warning: objtool: xe_gt_topology_dump() falls through to next function xe_gt_topology_init()

Also it was reported these warnings when building with clang:

drivers/gpu/drm/xe/xe.o: warning: objtool: xe_gt_topology_dump+0x77: sibling call from callable instruction with modified stack frame
drivers/gpu/drm/xe/xe.o: warning: objtool: xe_gt_topology_dump() falls through to next function xe_dss_mask_group_ffs()
drivers/gpu/drm/xe/xe.o: warning: objtool: xe_gt_topology_dump+0x77: can't find jump dest instruction at .text.xe_gt_topology_dump+0xc0

Since that value is not really possible in real world, just take the
simple approach and return NULL.

Fixes: 7108b4a589cd ("drm/xe/uapi: Expose SIMD16 EU mask in topology query")
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719191534.3845469-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Add NEEDS_2M BO flag

In addition of NEEDS_64K BO flag, add similar one to force 2 MiB
alignment of the buffer objects. Explicitly use this flag during
VF LMEM provisioning as LMTT uses 2 MiB pages and one day we may
drop requirement of allocating pinned objects as contiguous.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240715180538.1418-3-michal.wajdeczko@intel.com

drm/xe: Normalize NEEDS_64K BO flag

In commit 62742d126631 ("drm/xe: Normalize bo flags macros"),
we normalized all BO flags but XE_BO_NEEDS_64K. Do it now.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240715180538.1418-2-michal.wajdeczko@intel.com

drm/xe/tests: Skip xe_mocs live tests on VF device

There is no point to run those tests on VFs devices as they can't
access any of the MOCS registers. Skip testing on the VF device.

  [ ] =================== xe_mocs (1 subtest) ====================
  [ ] ================ xe_live_mocs_kernel_kunit  ================
  [ ] [PASSED] 0000:4d:00.0
  [ ] [SKIPPED] 0000:4d:00.1
  [ ] ============ [PASSED] xe_live_mocs_kernel_kunit ============
  [ ] ===================== [PASSED] xe_mocs =====================

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-8-michal.wajdeczko@intel.com

drm/xe/tests: Convert xe_mocs live tests

Convert xe_mocs live tests to parameterized style.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-7-michal.wajdeczko@intel.com

drm/xe/tests: Convert xe_migrate live tests

Convert xe_migrate live tests to parameterized style.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-6-michal.wajdeczko@intel.com

drm/xe/tests: Convert xe_dma_buf live tests

Convert xe_dma_buf live tests to parameterized style.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-5-michal.wajdeczko@intel.com

drm/xe/tests: Convert xe_bo live tests

Convert xe_bo live tests to parameterized style.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-4-michal.wajdeczko@intel.com

drm/xe/tests: Add helpers for use in live tests

Instead of iterating over available Xe devices within a testcase,
without being able to distinguish potential failures from different
devices on system with many Xe devices, introduce helpers that will
allow to treat each Xe device as a parameter for the testcase like:

static void bar(struct kunit *test)
{
struct xe_device *xe = test->priv;
...
}

struct kunit_case foo_live_tests[] = {
KUNIT_CASE_PARAM(bar, xe_pci_live_device_gen_param),
{}
};

struct kunit_suite foo_suite = {
.name = "foo_live",
.test_cases = foo_live_tests,
.init = xe_kunit_helper_xe_device_live_test_init,
};

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-3-michal.wajdeczko@intel.com

drm/xe: Introduce const cast helper

Typically we want to preserve pointer constness when converting
from one xe pointer to another, but in some rare cases, like kunit
parameter conversions, we might want to discard this constness.
Add a helper that we will use to clearly indicate our intention.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com> #v1
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240720142528.530-2-michal.wajdeczko@intel.com

drm/xe/oa: Don't use hardcoded values

The current implementation uses hardcoded values instead of common defines.

v2:
 - Make the commit a regular commit instead of a fixup commit
 - slightly modify commit message

Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240721071335.101234-1-osharabi@habana.ai

drm/xe: Build PM into GuC CT layer

Take PM ref when any G2H are outstanding, drop when none are
outstanding.

To safely ensure we have PM ref when in the GuC CT layer, a PM ref needs
to be held when scheduler messages are pending too.

v2:
 - Add outer PM protections to xe_file_close (CI)
v3:
 - Only take PM ref 0->1 and drop on 1->0 (Matthew Auld)
v4:
 - Add assert to G2H increment function
v5:
 - Rebase
v6:
 - Declare xe as local variable in xe_file_close (CI)

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719172905.1527927-5-matthew.brost@intel.com

drm/xe: Hold a PM ref when GT TLB invalidations are inflight

Avoid GT TLB invalidation timeouts by holding a PM ref when
invalidations are inflight.

v2:
 - Drop PM ref before signaling fence (CI)
v3:
 - Move invalidation_fence_signal helper in tlb timeout to previous
   patch (Matthew Auld)

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719172905.1527927-4-matthew.brost@intel.com

drm/xe: Drop xe_gt_tlb_invalidation_wait

Having two methods to wait on GT TLB invalidations is not ideal. Remove
xe_gt_tlb_invalidation_wait and only use GT TLB invalidation fences.

In addition to two methods being less than ideal, once GT TLB
invalidations are coalesced the seqno cannot be assigned during
xe_gt_tlb_invalidation_ggtt/range. Thus xe_gt_tlb_invalidation_wait
would not have a seqno to wait one. A fence however can be armed and
later signaled.

v3:
 - Add explaination about coalescing to commit message
v4:
 - Don't put dma fence if defined on stack (CI)
v5:
 - Initialize ret to zero (CI)
v6:
 - Use invalidation_fence_signal helper in tlb timeout (Matthew Auld)

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719172905.1527927-3-matthew.brost@intel.com

drm/xe: Add xe_gt_tlb_invalidation_fence_init helper

Other layers should not be touching struct xe_gt_tlb_invalidation_fence
directly, add helper for initialization.

v2:
 - Add dma_fence_get and list init to xe_gt_tlb_invalidation_fence_init

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719172905.1527927-2-matthew.brost@intel.com

drm/xe/vf: Fix register value lookup

We should use the number of actual entries stored in the runtime
register buffer, not the maximum number of entries that this buffer
can hold, otherwise bsearch() may fail and we may miss the data and
wrongly report unexpected access to some registers.

Fixes: 4edadc41a3a4 ("drm/xe/vf: Use register values obtained from the PF")
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Piotr Piórkowski <piotr.piorkowski@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240718203155.486-1-michal.wajdeczko@intel.com

drm/xe: Fix use after free when client stats are captured

xe_file_close triggers an asynchronous queue cleanup and then frees up
the xef object. Since queue cleanup flushes all pending jobs and the KMD
stores client usage stats into the xef object after jobs are flushed, we
see a use-after-free for the xef object. Resolve this by taking a
reference to xef from xe_exec_queue.

While at it, revert an earlier change that contained a partial work
around for this issue.

v2:
- Take a ref to xef even for the VM bind queue (Matt)
- Squash patches relevant to that fix and work around (Lucas)

v3: Fix typo (Lucas)

Fixes: ce62827bc294 ("drm/xe: Do not access xe file when updating exec queue run_ticks")
Fixes: 6109f24f87d7 ("drm/xe: Add helper to accumulate exec queue runtime")
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/issues/1908
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240718210548.3580382-5-umesh.nerlige.ramappa@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Take a ref to xe file when user creates a VM

Take a reference to xef when user creates the VM and put the reference
when user destroys the VM.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240718210548.3580382-4-umesh.nerlige.ramappa@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Add ref counting for xe_file

Add ref counting for xe_file.

v2:
- Add kernel doc for exported functions (Matt)
- Instead of xe_file_destroy, export the get/put helpers (Lucas)

v3: Fixup the kernel-doc format and description (Matt, Lucas)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240718210548.3580382-3-umesh.nerlige.ramappa@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Move part of xe_file cleanup to a helper

In order to make xe_file ref counted, move destruction of xe_file
members to a helper.

v2: Move xe_vm_close_and_put back into xe_file_close (Matt)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240718210548.3580382-2-umesh.nerlige.ramappa@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe/uapi: Expose SIMD16 EU mask in topology query

PVC, Xe2 and later platforms have 16-wide EUs. We were implicitly
reporting for PVC the number of 16-wide EUs without giving userspace any
hint that they were different than for other platforms. Xe2 and later
also have 16-wide, but in those cases the reported number would
correspond to the 8-wide count.

To avoid confusion and make sure the right number is used by userspace
depending on the platform, add a new item to the topology query and drop
the one that is not available. The new mask reported for both PVC and
Xe2 should now match the numbers reported via hwconfig.

v2: Use a different topo item with EU type in its name to report the
    new mask instead of adding the type itself as the item (Matt Roper)

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: José Roberto de Souza <jose.souza@intel.com>
Acked-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
Acked-by: Wenbin Lu <wenbin.lu@intel.com>
Acked-by: Effie Yu <effie.yu@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240710220446.2169797-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

drm/xe: Remove unused xe_sync_entry_wait

xe_sync_entry_wait is no longer used, remove it.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240717140429.1396820-2-matthew.brost@intel.com

drm/xe: Validate user fence during creation

Fail invalid addresses during user fence creation.

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240717140429.1396820-1-matthew.brost@intel.com

drm/xe/pm: Add trace for pm functions

Add trace for xe pm function for better debuggability.

v2: Fix indentation and add trace for xe_pm_runtime_get_ioctl

Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240717125950.9952-1-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>

drm/xe/fbdev: Limit the usage of stolen for LNL+

As per recommendation in the workarounds:
WA_22019338487

There is an issue with accessing Stolen memory pages due a
hardware limitation. Limit the usage of stolen memory for
fbdev for LNL+. Don't use BIOS FB from stolen on LNL+ and
assign the same from system memory.

v2: Corrected the WA Number, limited WA to LNL and
    Adopted XE_WA framework as suggested by Lucas and Matt.

v3: Introduced the waxxx_display to implement display side
    of WA changes on Lunarlake. Used xe_root_mmio_gt and
    avoid the for loop (Suggested by Lucas)

v4: Fixed some nits (Luca)

Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Uma Shankar <uma.shankar@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240717082252.3875909-1-uma.shankar@intel.com

drm/xe/xe2: Do not run xe_bo_test for xe2+ dgfx

In xe2+ dgfx, we don't need to handle the copying of ccs
metadata during migration. This test validates the ccs data post
clear and copy during evict/restore operation. Thus, we can skip
this test on xe2+ dgfx.

Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/57d9df82ad02e53c9b0d2a7d40bb27acce57b927.1721250309.git.akshata.jahagirdar@intel.com

drm/xe/migrate: Add kunit to test migration functionality for BMG

This part of kunit verifies that
- main data is decompressed and ccs data is clear post bo eviction.
- main data is raw copied and ccs data is clear post bo restore.

v2: Added missing bo_put()/bo_unlock() (Matt Auld)

Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1d36d4377c566508e42b3fb80d3fe4a588fd00ca.1721250309.git.akshata.jahagirdar@intel.com

drm/xe/xe_migrate: Handle migration logic for xe2+ dgfx

During eviction (vram->sysmem), we use compressed -> uncompressed mapping.
During restore (sysmem->vram), we need to use mapping from
uncompressed -> uncompressed.
Handle logic for selecting the compressed identity map for eviction,
and selecting uncompressed map for restore operations.
v2: Move check of xe_migrate_ccs_emit() before calling
xe_migrate_ccs_copy(). (Nirmoy)

Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/79b3a016e686a662ae68c32b5fc7f0f2ac8043e9.1721250309.git.akshata.jahagirdar@intel.com

drm/xe/xe2: Introduce identity map for compressed pat for vram

Xe2+ has unified compression (exactly one compression mode/format),
where compression is now controlled via PAT at PTE level.
This simplifies KMD operations, as it can now decompress freely
without concern for the buffer's original compression format—unlike DG2,
which had multiple compression formats and thus required copying the
raw CCS state during VRAM eviction. In addition mixed VRAM and system
memory buffers were not supported with compression enabled.

On Xe2 dGPU compression is still only supported with VRAM, however we
can now support compression with VRAM and system memory buffers,
with GPU access being seamless underneath. So long as when doing
VRAM -> system memory the KMD uses compressed -> uncompressed,
to decompress it. This also allows CPU access to such buffers,
assuming that userspace first decompress the corresponding
pages being accessed.
If the pages are already in system memory then KMD would have already
decompressed them. When restoring such buffers with sysmem -> VRAM
the KMD can't easily know which pages were originally compressed,
so we always use uncompressed -> uncompressed here.
With this it also means we can drop all the raw CCS handling on such
platforms (including needing to allocate extra CCS storage).

In order to support this we now need to have two different identity
mappings for compressed and uncompressed VRAM.
In this patch, we set up the additional identity map for the VRAM with
compressed pat_index. We then select the appropriate mapping during
migration/clear. During eviction (vram->sysmem), we use the mapping
from compressed -> uncompressed. During restore (sysmem->vram), we need
the mapping from uncompressed -> uncompressed.
Therefore, we need to have two different mappings for compressed and
uncompressed vram. We set up an additional identity map for the vram
with compressed pat_index.
We then select the appropriate mapping during migration/clear.

v2: Formatting nits, Updated code to match recent changes in
    xe_migrate_prepare_vm(). (Matt)

v3: Move identity map loop to a helper function. (Matt Brost)

v4: Split helper function in different patch, and
add asserts and nits. (Matt Brost)

v5: Convert the 2 bool arguments of pte_update_size to flags
argument (Matt Brost)

v6: Formatting nits (Matt Brost)

Signed-off-by: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/b00db5c7267e54260cb6183ba24b15c1e6ae52a3.1721250309.git.akshata.jahagirdar@intel.com