drm/xe: Add a missing mutex_destroy to xe_ttm_vram_mgr

Ensure that the mutex is destroyed at fini function.

Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Bommithi Sakeena <bommithi.sakeena@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Ensure mutex are destroyed

Add missing mutex_destroy calls to fini functions or convert to
drmm_mutex_init where fini function is not available.

Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Bommithi Sakeena <bommithi.sakeena@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: do not register to PM if GuC is disabled

When working without GuC (i.e. working with execlists), the flow
attempts to perform suspend operation which is failing due to a
lack of support without GuC.

If PM ops are not supported without GuC we may as well avoid PM
registration rather than returning errors from various PM flows.

Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use vfunc for ggtt pte encoding

Use 2 different functions for encoding the ggtt's pte, assigning them
during initialization. Main difference is that before Xe-LPG, the pte
didn't have the cache bits.

v2: Re-use xelp_ggtt_pte_encode_bo() for the common part with
    xelpg_ggtt_pte_encode_bo() (Matt Roper)

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-11-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use pat_index to encode pde/pte

Change the xelp_pte_encode() and xelp_pde_encode() functions to use the
platform-dependent pat_index.  The same function can be used for all
platforms as they only need to encode the pat_index bits in the same
pte/pde layout. For platforms that don't have the most significant bit,
as long as they don't return a bogus index they should be fine.

v2: Use the same logic to encode pde as it's compatible with previous
    logic, it's more future proof and also fixes the cache setting for
    PVC (Matt Roper)

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-10-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pat: Keep track of relevant indexes

Some of the PAT entries are relevant for internal driver use, which
varies per platform. Let the PAT early initialization set what they
should point to so the rest of the driver can use them where needed.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-9-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pat: Prefer the arch/IP names

Both DG2 and PVC are derived from XeHP, but DG2 should not really
re-use something introduced by PVC, so it's odd to have DG2 re-using the
PVC programming for PAT. Let's prefer using the architecture and/or IP
names.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-8-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/dg2: Fix using wrong PAT table

DG2 should use the MCR variant to program the PAT registers, like PVC,
but shouldn't use the same table as PVC.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-7-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use vfunc to initialize PAT

Split the PAT initialization between SW-only and HW. The _early() only
sets up the ops and data structure that are used later to program the
tables. This allows the PAT to be easily extended to other platforms.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-6-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/migrate: Do not hand-encode pte

Instead of encoding the pte, call a new vfunc from xe_vm to handle that.
The encoding may not be the same on every platform, so keeping it in one
place helps to better support them.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-5-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use vfunc for pte/pde ppgtt encoding

Move the function to encode pte/pde to be vfuncs inside struct xe_vm.
This will allow to easily extend to platforms that don't have a
compatible encoding.

v2: Fix kunit build

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-4-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Remove check for vma == NULL

vma at this point can never be NULL as otherwise it would crash earlier
in the only caller, xe_pt_stage_bind_entry(). Remove the extra check and
avoid adding and removing the bits from the pte.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-3-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Normalize pte/pde encoding

Split functions that do only part of the pde/pte encoding and that can
be called by the different places. This normalizes how pde/pte are
encoded so they can be moved elsewhere in a subsequent change.

xe_pte_encode() was calling __pte_encode() with a NULL vma, which is the
opposite of what xe_pt_stage_bind_entry() does. Stop passing a NULL vma
and just split another function that deals with a vma rather than a bo.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Infer service copy functionality from engine list

On platforms with multiple BCS engines (i.e., PVC and Xe2), not all BCS
engines are created equal.  The BCS0 engine is what the specs refer to
as a "resource copy engine," which supports the platform's full set of
copy/fill instructions.  In contast, the non-BCS0 "service copy" engines
are more streamlined and only support a subset of the GPU instructions
supported by the resource copy engine.  Platforms with both types of
copy engines always support the MEM_COPY and MEM_SET instructions which
can be used for simple copy and fill operations on either type of BCS
engine.  Since the simple MEM_SET instruction meets the needs of Xe's
migrate code (and since the more elaborate XY_FAST_COLOR_BLT instruction
isn't available to use on service copy engines), we always prefer to use
MEM_SET for clearing buffers on our newer platforms.

We've been using a 'has_link_copy_engine' feature flag to keep track of
which platforms should use MEM_SET for fills.  However a feature flag
like this is unnecessary since we can already derive the presence of
service copy engines (and in turn the MEM_SET instruction) just by
looking at the platform's pre-fusing engine list.  Utilizing the engine
list for this also avoids mistakes like we've made on Xe2 where we
forget to set the feature flag in the IP definition.

For clarity, "service copy" is a general term that covers any blitter
engines that support a limited subset of the overall blitter instruction
set (in practice this is any non-BCS0 blitter engine).  The "link copy
engines" introduced on PVC and the "paging copy engine" present in Xe2
are both instances of service copy engines.

v2:
 - Rewrite / expand the commit message.  (Bala)
 - Fix checkpatch whitespace error.

Bspec: 65019
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Reviewed-by: Haridhar Kalvala <haridhar.kalvala@intel.com>
Link: https://lore.kernel.org/r/20230927205143.2695089-2-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/irq: Clear GFX_MSTR_IRQ as part of IRQ reset

Starting with Xe_LP+, GFX_MSTR_IRQ contains status bits that have W1C
behavior. If we do not properly reset them, we would miss delivery of
interrupts if a pending bit is set when enabling IRQs.

As an example, the display part of our probe routine contains paths
where we wait for vblank interrupts. If a display interrupt was already
pending when enabling IRQs, we would time out waiting for the vblank.

That in fact happened recently when modprobing Xe on a Lunar Lake with a
specific configuration; and that's how we found out we were missing this
step in the IRQ enabling logic.

Fix the issue by clearing GFX_MSTR_IRQ as part of the IRQ reset.

v2:
  - Make resetting GFX_MSTR_IRQ be the last step to avoid bit
    re-latching. (Ville)
v3:
  - Swap nesting order: guard loop with the IP version check instead of
    doing the check at each iteration. (Lucas)
v4:
  - Add braces for the "if" statement guarding the loop to make the
    compiler happy. (Gustavo)

BSpec: 50875, 54028, 62357
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230926221914.106843-2-gustavo.sousa@intel.com
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add Wa_18028616096

Drop UGM per set fragment threshold to 3

BSpec: 54833
Signed-off-by: Shekhar Chauhan <shekhar.chauhan@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230925160543.915217-1-shekhar.chauhan@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Align size to PAGE_SIZE

Ensure alignment with PAGE_SIZE for the size parameter
passed to __xe_bo_create_locked()

v2: move size alignment under else condition (Lucas)

Signed-off-by: Pallavi Mishra <pallavi.mishra@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230920213259.3458968-1-pallavi.mishra@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Accept a const xe device

Depending on the context, it's preferred to have a const pointer to make
sure nothing is modified underneath. The assert macros only ever read
data from xe/tile/gt for printing, so they can be made const by default.

Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20230922174320.2372617-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: add msix support

In future devices we will need to support msix interrupts.

Reviewed-by: Ohad Sharabi <osharabi@habana.ai>
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: change old msi irq api to a new one

As a preparation for msix support, changing for new msi irq api
which supports both msi and msix.

Reviewed-by: Ohad Sharabi <osharabi@habana.ai>
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Rebase fixes by Rodrigo]

drm/xe: proper setting of irq enabled flag

IRQ enabled flag should be set only after request irq succeeds.

Reviewed-by: Ohad Sharabi <osharabi@habana.ai>
Signed-off-by: Dani Liberman <dliberman@habana.ai>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Implement fdinfo memory stats printing

Use the newly added drm_print_memory_stats helper to show memory
utilisation of our objects in drm/driver specific fdinfo output.

To collect the stats we walk the per memory regions object lists
and accumulate object size into the respective drm_memory_stats
categories.

Objects with multiple possible placements are reported in multiple
regions for total and shared sizes, while other categories are
counted only for the currently active region.

V4:
  - Remove rcu lock - Auld/Thomas
  - take refcnt only if its non-zero - Auld
  - DMA_RESV_USAGE_BOOKKEEP covers all fences - Auld
  - covert to xe_bo for public objects
V3:
  - dont use xe_bo_get/put, not needed
  - use designated initializer - Jani
  - use list_for_each_entry_rcu
  - Fix Checkpatch err - CI
V2:
  - Use static initializer for mem_type - Himal/Jani

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Account ring buffer and context state storage

Account ring buffers and logical context space against the owning client
memory usage stats.

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Track page table memory usage for client

Account page table memory usage in the owning client
memory usage stats.

V2:
  - Minor tweak to if (vm->pt_root[id]) check - Himal

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Record each drm client with its VM

Enable accounting of indirect client memory usage.

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add tracking support for bos per client

In order to show per client memory consumption, we
need tracking support APIs to add at every bo consumption
and removal. Adding APIs here to add tracking calls at
places wherever it is applicable.

V5:
  - Rebase
V4:
  - remove client bo before vm_put
  - spin_lock_irqsave not required - Auld
V3:
  - update .h to return xe_drm_client_remove_bo void
  - protect xe_drm_client_remove_bo under CONFIG_PROC_FS check - Himal
  - Fixed Checkpatch error - CI
V2:
  - make xe_drm_client_remove_bo return void - Himal

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Interface xe drm client with fdinfo interface

DRM core driver has introduced recently fdinfo interface to
show memory stats of individual drm client. Lets interface
xe drm client to fdinfo interface.

V2:
  - cover call to xe_drm_client_fdinfo under PROC_FS

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add drm-client infrastructure

Add drm-client infrastructure to record stats of consumption
done by individual drm client.

V2:
  - Typo - CI

Reviewed-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add child contexts to the GuC context lookup

The CAT_ERROR message from the GuC provides the guc id of the context
that caused the problem, which can be a child context. We therefore
need to be able to match that id to the exec_queue that owns it, which
we do by adding child context to the context lookup.

While at it, fix the error path of the guc id allocation code to
correctly free the ids allocated for parallel queues.

v2: rebase on s/XE_WARN_ON/xe_assert

Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/590
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/wa: Apply tile workarounds at probe/resume

Although the vast majority of workarounds the driver needs to implement
are either GT-based or display-based, there are occasionally workarounds
that reside outside those parts of the hardware (i.e., in they target
registers in the sgunit/soc); we can consider these to be "tile"
workarounds since there will be instance of these registers per tile.
The registers in question should only lose their values during a
function-level reset, so they only need to be applied during probe and
resume; the registers will not be affected by GT/engine resets.

Tile workarounds are rare (there's only one, 22010954014, that's
relevant to Xe at the moment) so it's probably not worth updating the
xe_rtp design to handle tile-level workarounds yet, although we may want
to consider that in the future if/when more of these show up on future
platforms.

Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://lore.kernel.org/r/20230913231411.291933-13-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Disallow pinning dma-bufs in VRAM

For now only support pinning in TT memory, for two reasons:
1) Avoid pinning in a placement not accessible to some importers.
2) Pinning in VRAM requires PIN accounting which is a to-do.

v2:
- Adjust the dma-buf kunit test accordingly.

Suggested-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20230920095001.5539-1-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Simplify final return from xe_irq_install()

At the end of the function, we will always return err no matter it's
value. Simplify this by just returning the result of
drmm_add_action_or_reset().

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230915220233.59736-1-gustavo.sousa@intel.com
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Reinstate pipelined fence enable_signaling

With the GPUVA conversion, the xe_bo::vmas member became replaced with
drm_gem_object::gpuva.list, however there was a couple of usage instances
left using the old member. Most notably the pipelined fence
enable_signaling.

Remove the xe_bo::vmas member completely, fix usage instances and
also enable this pipelined fence enable_signaling even for faulting
VM:s since we actually wait for bind fences to complete.

v2:
- Rebase.
v3:
- Fix display code build error.

Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230915172606.14436-1-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/uc: Add GuC/HuC firmware path overrides

When testing a new binary and/or debugging binary-related issues, it is
useful to have the option to change which binary is loaded without
having to update and re-compile the kernel. To support this option, this
patch adds 2 new modparams to override the FW path for GuC and HuC. The
HuC modparam can also be set to an empty string to disable HuC loading.

Note that those modparams only take effect on platforms where we already
have a default FW, so we're sure there is support for FW loading and the
kernel isn't going to explode in an undefined path.

v2: simplify comment (John),
    rebase on s/guc_submission_enabled/uc_enabled

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/uc: Fix uC status tracking

The current uC status tracking has a few issues:

1) the HuC is moved to "disabled" instead of "not supported"

2) the status is left uninitialized instead of "disabled" when the
   modparam is used to disable support

3) due to #1, a number of checks are done against "disabled" instead of
   the appropriate status.

Address all of those by making sure to follow the appropriate state
transition and checking against the required state.

v2: rebase on s/guc_submission_enabled/uc_enabled/

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/uc: Rename guc_submission_enabled() to uc_enabled()

The guc_submission_enabled() function is being used as a boolean toggle
for all firmwares and all related features, not just GuC submission. We
could add additional flags/functions to distinguish and allow different
use-cases (e.g. loading HuC but not using GuC submission), but given
that not using GuC is a debug-only scenario having a global switch for
all FWs is enough. However, we want to make it clear that this switch
turns off everything, so rename it to uc_enabled().

v2: rebase on s/XE_WARN_ON/xe_assert

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pmu: Enable PMU interface

There are a set of engine group busyness counters provided by HW which are
perfect fit to be exposed via PMU perf events.

BSPEC: 46559, 46560, 46722, 46729, 52071, 71028

events can be listed using:
perf list
  xe_0000_03_00.0/any-engine-group-busy-gt0/         [Kernel PMU event]
  xe_0000_03_00.0/copy-group-busy-gt0/               [Kernel PMU event]
  xe_0000_03_00.0/interrupts/                        [Kernel PMU event]
  xe_0000_03_00.0/media-group-busy-gt0/              [Kernel PMU event]
  xe_0000_03_00.0/render-group-busy-gt0/             [Kernel PMU event]

and can be read using:

perf stat -e "xe_0000_8c_00.0/render-group-busy-gt0/" -I 1000
           time             counts unit events
     1.001139062                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     2.003294678                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     3.005199582                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     4.007076497                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     5.008553068                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     6.010531563              43520 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     7.012468029              44800 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     8.013463515                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
     9.015300183                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
    10.017233010                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/
    10.971934120                  0 ns  xe_0000_8c_00.0/render-group-busy-gt0/

The pmu base implementation is taken from i915.

v2:
Store last known value when device is awake return that while the GT is
suspended and then update the driver copy when read during awake.

v3:
1. drop init_samples, as storing counters before going to suspend should
be sufficient.
2. ported the "drm/i915/pmu: Make PMU sample array two-dimensional" and
dropped helpers to store and read samples.
3. use xe_device_mem_access_get_if_ongoing to check if device is active
before reading the OA registers.
4. dropped format attr as no longer needed
5. introduce xe_pmu_suspend to call engine_group_busyness_store
6. few other nits.

v4: minor nits.

v5: take forcewake when accessing the OAG registers

v6:
1. drop engine_busyness_sample_type
2. update UAPI documentation

v7:
1. update UAPI documentation
2. drop MEDIA_GT specific change for media busyness counter.

Co-developed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Co-developed-by: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty@linux.intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use spinlock in forcewake instead of mutex

In PMU we need to access certain registers which fall under GT power
domain for which we need to take forcewake. But as PMU being an atomic
context can't expect to have any sleeping calls.

Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Get GT clock to nanosecs

Helper to convert GT clock cycles to nanoseconds.

v2: Use DIV_ROUND_CLOSEST_ULL helper(Ashutosh)
v3: rename xe_gt_clock_interval_to_ns to xe_gt_clock_cycles_to_ns

Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/guc: Switch to major-only GuC FW tracking for MTL

Newer HuC binaries for MTL (8.5.1+) require GuC 70.7 or newer, so we
need to move on from 70.6.4. Given that the MTL GuC uses major-only
version matching in i915, we can do the same here instead of just
bumping the version (and having to push the versioned binaries,
because they're not there already for i915).

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Use Xe assert macros instead of XE_WARN_ON macro

The XE_WARN_ON macro maps to WARN_ON which is not justified
in many cases where only a simple debug check is needed.
Replace the use of the XE_WARN_ON macro with the new xe_assert
macros which relies on drm_*. This takes a struct drm_device
argument, which is one of the main changes in this commit. The
other main change is that the condition is reversed, as with
XE_WARN_ON a message is displayed if the condition is true,
whereas with xe_assert it is if the condition is false.

v2:
- Rebase
- Keep WARN splats in xe_wopcm.c (Matt Roper)

v3:
- Rebase

Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Introduce Xe assert macros

As we are moving away from the controversial XE_BUG_ON macro,
relying just on WARN_ON or drm_err does not cover the cases
where we want to annotate functions with additional detailed
debug checks to assert that all prerequisites are satisfied,
without paying footprint or performance penalty on non-debug
builds, where all misuses introduced during code integration
were already fixed.

Introduce family of Xe assert macros that try to follow classic
assert() utility and can be compiled out on non-debug builds.

Macros are based on drm_WARN, but unlikely to origin, disallow
use in expressions since we will compile that code out.

As we are operating on the xe pointers, we can print additional
information about the device, like tile or GT identifier, that
is not available from generic WARN report:

[ ] xe 0000:00:02.0: [drm] Assertion `true == false` failed!
    platform: 1 subplatform: 1
    graphics: Xe_LP 12.00 step B0
    media: Xe_M 12.00 step B0
    display: enabled step D0
    tile: 0 VRAM 0 B
    GT: 0 type 1

[ ] xe 0000:b3:00.0: [drm] Assertion `true == false` failed!
    platform: 7 subplatform: 3
    graphics: Xe_HPG 12.55 step A1
    media: Xe_HPM 12.55 step A1
    display: disabled step **
    tile: 0 VRAM 14.0 GiB
    GT: 0 type 1

[ ] WARNING: CPU: 0 PID: 2687 at drivers/gpu/drm/xe/xe_device.c:281 xe_device_probe+0x374/0x520 [xe]
[ ] RIP: 0010:xe_device_probe+0x374/0x520 [xe]
[ ] Call Trace:
[ ]  ? __warn+0x7b/0x160
[ ]  ? xe_device_probe+0x374/0x520 [xe]
[ ]  ? report_bug+0x1c3/0x1d0
[ ]  ? handle_bug+0x42/0x70
[ ]  ? exc_invalid_op+0x14/0x70
[ ]  ? asm_exc_invalid_op+0x16/0x20
[ ]  ? xe_device_probe+0x374/0x520 [xe]
[ ]  ? xe_device_probe+0x374/0x520 [xe]
[ ]  xe_pci_probe+0x6e3/0x950 [xe]
[ ]  ? lockdep_hardirqs_on+0xc7/0x140
[ ]  pci_device_probe+0x9e/0x160
[ ]  really_probe+0x19d/0x400

v2: use lowercase names
v3: apply xe coding style
v4: fix non-debug build and improve kernel-doc

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Oded Gabbay <ogabbay@kernel.org>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Replace XE_WARN_ON with drm_warn when just printing a string

Use the generic drm_warn instead of the driver-specific XE_WARN_ON
in cases where XE_WARN_ON is used to unconditionally print a debug
message.

v2: Rebase

Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Fix fence reservation accouting

Both execs and the preempt rebind worker can issue rebinds. Rebinds
require a fence, per tile, inserted into dma-resv slots of the VM and
BO (if external). The fence reservation accouting did not take into
account the number of fences required for rebinds, fix this.

v2: Rebase

Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reported-by: Christopher Snowhill <kode54@gmail.com>
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/518
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Convert remaining instances of ttm_eu_reserve_buffers to drm_exec

The VM_BIND functionality and vma destruction was locking
potentially multiple dma_resv objects using the
ttm_eu_reserve_buffers() function. Rework those to use the drm_exec
helper, taking care that any calls to xe_bo_validate() ends up
inside an unsealed locking transaction.

v4:
- Remove an unbalanced xe_bo_put() (igt and Matthew Brost)
v5:
- Rebase conflict

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-7-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Convert pagefaulting code to use drm_exec

Replace the calls into ttm_eu_reserve_buffers with the drm_exec helpers.
Also reuse some code.

v4:
- Kerneldoc xe_vm_prepare_vma().

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-6-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Rework xe_exec and the VM rebind worker to use the drm_exec helper

Replace the calls to ttm_eu_reserve_buffers() by using the drm_exec
helper instead. Also make sure the locking loop covers any calls to
xe_bo_validate() / ttm_bo_validate() so that these function calls may
easily benefit from being called from within an unsealed locking
transaction and may thus perform blocking dma_resv locks in the future.

For the unlock we remove an assert that the vm->rebind_list is empty
when locks are released. Since if the error path is hit with a partly
locked list, that assert may no longer hold true we chose to remove it.

v3:
- Don't accept duplicate bo locks in the rebind worker.
v5:
- Loop over drm_exec objects in reverse when unlocking.
v6:
- We can't keep the WW ticket when retrying validation on OOM. Fix.

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-5-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/bo: Remove the lock_no_vm()/unlock_no_vm() interface

Apart from asserts, it's essentially the same as
xe_bo_lock()/xe_bo_unlock(), and the usage intentions of this interface
was unclear. Remove it.

v2:
- Update the xe_display subsystem as well.

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-4-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/vm: Simplify and document xe_vm_lock()

The xe_vm_lock() function was unnecessarily using ttm_eu_reserve_buffers().
Simplify and document the interface.

v4:
- Improve on xe_vm_lock() documentation (Matthew Brost)
v5:
- Rebase conflict.

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-3-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/bo: Simplify xe_bo_lock()

xe_bo_lock() was, although it only grabbed a single lock, unnecessarily
using ttm_eu_reserve_buffers(). Simplify and document the interface.

v2:
- Update also the xe_display subsystem.
v4:
- Reinstate a lost dma_resv_reserve_fences().
- Improve on xe_bo_lock() documentation (Matthew Brost)

Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230908091716.36984-2-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/mmio: Account for GSI offset when checking ranges

Change xe_mmio_in_range() to use the same logic to account for the GT's
adj_offset as the read and write functions. This is needed when checking
ranges for the MCR registers if the GT has an offset to adjust.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230908225227.1276610-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/uapi: Remove useless max_page_size

The min_page_size is useful information to ensure alignment and it is
an API actually in use. However max_page_size doesn't bring any useful
information to the userspace hence being not used at all.

So, let's remove and only bring it back if that ever gets used.

Suggested-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Francois Dugast <francois.dugast@intel.com>

drm/xe/uapi: Typo lingo and other small backwards compatible fixes

Fix typos, lingo and other small things identified during uapi
review.

v2: Also fix ALIGNMENT typo at xe_query.c
v3: Do not touch property to get/set. (Francois)

Link: https://lore.kernel.org/all/863bebd0c624d6fc2b38c0a06b63e468b4185128.camel@linux.intel.com/
Suggested-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Francois Dugast <francois.dugast@intel.com>

drm/xe: Fix LRC workarounds

Fix 2 issues when writing LRC workarounds by copying the same handling
done when processing other RTP entries:

For masked registers, it was not correctly setting the upper 16bits.
Differently than i915, the entry itself doesn't set the upper bits
for masked registers: this is done when applying them. Testing on ADL-P:

Before:
[drm:xe_gt_record_default_lrcs [xe]] LRC WA rcs0 save-restore MMIOs
[drm:xe_gt_record_default_lrcs [xe]] REG[0x2580] = 0x00000002
...
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7018] = 0x00002000
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7300] = 0x00000040
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7304] = 0x00000200

After:
[drm:xe_gt_record_default_lrcs [xe]] LRC WA rcs0 save-restore MMIOs
[drm:xe_gt_record_default_lrcs [xe]] REG[0x2580] = 0x00060002
...
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7018] = 0x20002000
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7300] = 0x00400040
[drm:xe_gt_record_default_lrcs [xe]] REG[0x7304] = 0x02000200

All of these registers are masked registers, so writing to them without
the relevant bits in the upper 16b doesn't have any effect.

Also, this adds support to regular registers; previously it was assumed
that LRC entries would only contain masked registers. However this is
not true. 0x6604 is not a masked register, but used in workarounds for
e.g.  ADL-P. See commit 28cf243a341a ("drm/i915/gt: Fix context
workarounds with non-masked regs"). In the same test with ADL-P as
above:

Before:
[drm:xe_gt_record_default_lrcs [xe]] REG[0x6604] = 0xe0000000
After:
[drm:xe_gt_record_default_lrcs [xe]] REG[0x6604] = 0xe0efef6f

As can be seen, now it will read what was in the register rather than
completely overwrite the other bits.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230906012053.1733755-5-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add dbg messages for LRC WAs

Just like the GT and engine workarounds, add debug message with the
final value being written to the register for easy debugging.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230906012053.1733755-4-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/reg_sr: Use xe_gt_dbg

Use xe_gt_dbg() instead of drm_dbg() so the GT is added to the log for
easy identification.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230906012053.1733755-3-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/reg_sr: Simplify check for masked registers

For all RTP actions, clr_bits is a superset of the bits being modified.
That's also why the check for "changing all bits" can be done with
`clr_bits + 1`. So always use clr_bits for setting the upper bits of a
masked register.

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://lore.kernel.org/r/20230906012053.1733755-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pat: Use 0 instead of space on error

Use 0 in format string instead of space so it shows as

[drm] *ERROR* Missing PAT table for platform with graphics version 20.04!

instead of

[drm] *ERROR* Missing PAT table for platform with graphics version 20. 4!

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://lore.kernel.org/r/20230906193009.1912129-1-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/selftests: make eviction test tile centric

The concern here is that we may have platforms with dedicated media GT,
and we anyway allocate the object on the tile, which just means running
the same test twice (i.e primary vs media GT).

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/selftests: consider multi-GT for eviction test

We need to sanitize and reset each GT, since xe_bo_evict_all() will
evict everything regardless of GT, which can leave other GTs in a broken
state.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pm: Use PM functions only if CONFIG_PM_SLEEP is enabled

This fixes the build without CONFIG_PM_SLEEP such as for riscv.

Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Fix array of binds

If multiple bind ops in an array of binds touch the same address range
invalid GPUVA operations are generated as each GPUVA operation is
generated based on the orignal GPUVA state. To fix this, after each
GPUVA operations is generated, commit the GPUVA operation updating the
GPUVA state so subsequent bind ops can see a current GPUVA state.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/gpuva: Add drm_gpuva_for_each_op_reverse

Add a helper to walk op list in reverse. Xe will make use of this when
unwinding GPUVA operations.

v2: (Rodrigo) reword commit message

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Fixup unwind on VM ops errors

Remap ops have 3 parts: unmap, prev, and next. The commit step can fail
on any of these. Add a flag for each to these so the unwind is only done
the steps that have been committed.

v2: (Rodrigo) Use bit macros

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Convert xe_vma_op_flags to BIT macros

Rather than open code the shift for values, use BIT macros.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: fix has_llc on rkl

Matches i915. Assumption going forward is that non-llc + igpu is only a
thing on MTL+ which should have explicit coherency pat_index settings
for COH_NONE, 1WAY and 2WAY.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Pallavi Mishra <pallavi.mishra@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi Mishra <pallavi.mishra@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: nuke GuC on unload

On PVC unloading followed by reloading the module often results in a
completely dead machine (seems to be plaguing CI). Resetting the GuC
like we do at load seems to cure it at least when locally testing this.

v2:
  - Move pc_fini into guc_fini. We want to do the GuC reset just after
    calling pc_fini, otherwise we encounter communication failures. It
    also seems like a good idea to do the reset before we start releasing
    the various other GuC resources. In the case of pc_fini there is an
    explicit stop, but for other stuff like logs, ads, ctb there is not.

References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/542
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/597
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Prevent return with locked vm

Reorder vm_id check after the one for VISIBLE_VRAM. This should
prevent returning with locked vm in error scenario.

Signed-off-by: Pallavi Mishra <pallavi.mishra@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Add patch version on guc firmware init

Add patch version info on GuC firmware init. This is required info for
GuC log decoder.

Signed-off-by: Zhanjun Dong <zhanjun.dong@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://lore.kernel.org/r/20230817213028.838531-1-zhanjun.dong@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pvc: Use fast copy engines as migrate engine on PVC

Some copy hardware engine instances are faster than others on PVC.
Use a virtual engine of these plus the reserved instance for the migrate
engine on PVC. The idea being if a fast instance is available it will be
used and the throughput of kernel copies, clears, and pagefault
servicing will be higher.

v2: Use OOB WA, use all copy engines if no WA is required

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pvc: Force even num engines to use 64B

Wa_16017236439 requires that we update BCS_SWCTRL
(via indirect context batch buffer) to set 64B
transfers when running on an even-numbered BCS
engine and 256B on an odd-numbered BCS engine.

v2: Move WA from engine_was[] to lrc_was[]

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/pvc: Blacklist BCS_SWCTRL register

Wa_16017236439 requires the BCS_SWCTRL to be privileged.

v2: Define and use BCS_SWCTRL()

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/ct: fix resv_space print

Actually print the info.resv_space.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: standardize vm-less kernel submissions

The current only submission in the driver that doesn't use a vm is the
WA setup. We still pass a vm structure (the migration one), but we don't
actually use it at submission time and we instead have an hack to use
GGTT for this particular engine.
Instead of special-casing the WA engine, we can skip providing a VM and
use that as selector for whether to use GGTT or PPGTT. As part of this
change, we can drop the special engine flag for the WA engine and switch
the WA submission to use the standard job functions instead of dedicated
ones.

v2: rebased on s/engine/exec_queue

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://lore.kernel.org/r/20230822173334.1664332-4-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: split kernel vs permanent engine flags

If an engine is only destroyed on driver unload, we can skip its
clean-up steps with the GuC because the GuC is going to be tuned off as
well, so it doesn't matter if we're in sync with it or not. Currently,
we apply this optimization to all engines marked as kernel, but this
stops us to supporting kernel engines that don't stick around until
unload. To remove this limitation, add a separate flag to indicate if
the engine is expected to only be destryed on driver unload and use that
to trigger the optimzation.

While at it, add a small comment to explain what each engine flag
represents.

v2: s/XE_BUG_ON/XE_WARN_ON, s/ENGINE/EXEC_QUEUE
v3: rebased

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://lore.kernel.org/r/20230822173334.1664332-3-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: fix submissions without vm

Kernel queues can submit privileged batches directly in GGTT, so they
don't always need a vm. The submission front-end already supports
creating and submitting jobs without a vm, but some parts of the
back-end assume the vm is always there. Fix this by handling a lack of
vm in the back-end as well.

v2: s/XE_BUG_ON/XE_WARN_ON, s/engine/exec_queue

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://lore.kernel.org/r/20230822173334.1664332-2-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Drop xe_mmio_write64()

The only possible 64-bit register writes in the driver come from the
highly questionable MMIO ioctl.  That ioctl's register write support
only operates for userspace running as root and cannot be used by any
real userspace; it exists solely to support the "xe_reg" debug tool in
IGT.  Since the spec indicates that hardware does not officially support
64-bit register accesses, there's no reason to allow such 64-bit writes,
even for debugging.

Bspec: 60027
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://lore.kernel.org/r/20230823003312.1356779-4-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Avoid 64-bit register reads

Intel hardware officially only supports GTTMMADR register accesses of
32-bits or less (although 64-bit accesses to device memory and PTEs in
the GSM are fine).  Even though we do usually seem to get back
reasonable values when performing readq() operations on registers in
BAR0, we shouldn't rely on this violation of the spec working
consistently.  It's likely that even when we do get proper register
values back the hardware is internally satisfying the request via a
non-atomic sequence of two 32-bit reads, which can be problematic for
timestamps and counters if rollover of the lower bits is not considered.

Replace xe_mmio_read64() with xe_mmio_read64_2x32() that implements
64-bit register reads as two 32-bit reads and attempts to ensure that
the upper dword has stabilized to avoid problematic rollovers for
counter and timestamp registers.

v2:
 - Move function from xe_mmio.h to xe_mmio.c.  (Lucas)
 - Convert comment to kerneldoc and note that it shouldn't be used on
   registers where reads may trigger side effects.  (Lucas)

Bspec: 60027
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://lore.kernel.org/r/20230823003312.1356779-3-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/lnl: Hook up MOCS table

LNL uses the Xe2 MOCS table introduced in an earlier patch.

Bspec: 71582
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/lnl: Add GuC firmware definition

Define the GuC firmware to load on the platform.

Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/lnl: Add LNL platform definition

LNL is an integrated GPU based on the Xe2 architecture.

Bspec: 70821
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Program GuC's MOCS on Xe2 and beyond

As with PVC, Xe2 platforms require that the index of an uncached MOCS
entry be programmed into the GUC_SHIM_CONTROL register.  This will
likely be needed on future platforms as well.

Xe2 also extends the size of the MOCS index register field from two bits
to four bits.  Since these extra bits were unused on PVC, it should be
safe to just increase the size of the mask.

Bspec: 60592
Cc: Haridhar Kalvala <haridhar.kalvala@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Add MOCS table

Additional minor change to remove L4_2_RESERVED, which will never be
required.

v2: Make L3/L4 names consistent for GLOB_MOCS defines (Matt Roper)

Bspec: 71582
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Track VA bits independently of max page table level

Starting with Xe2, a 5-level page table is always used, regardless of
the actual virtual address range supported by the platform.  The two
values need to be tracked separately in the device descriptor since Xe2
platforms only have a 48 bit virtual address range.

Bspec: 59505, 65637, 70817
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Define Xe2_LPM IP features

Xe2_LPM media is represented by GMD_ID value 20.00.
It provides 1 VD + 1 VE + 1 SFC.

Bspec: 70821, 70819
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Define Xe2_LPG IP features

Define a common set of Xe2 graphics feature flags and definitions that
will be used for all platforms in this family.

Several of the feature flags are inherited unchanged from Xe_HP and/or
Xe_HPC platforms:
 - dma_mask_size remains 46   (Bspec 70817)
 - supports_usm=1             (Bspec 59651)
 - has_flatccs=1              (Bspec 58797)
 - has_asid=1                 (Bspec 59654, 59265, 60288)
 - has_range_tlb_invalidate=1 (Bspec 71126)

However some of them still need proper implementation in the driver to
be used, so they are disabled.

Notable Xe2-specific changes:
 - All Xe2 platforms use a five-level page table, regardless of the
   virtual address space for the platform.  (Bspec 59505)

The graphics engine mask represents the Xe2 architecture engines (Bspec
60149), but individual platforms may have a reduced set of usable
engines, as reflected by their fusing.

Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: AuxCCS is no longer used

Starting with Xe2, all platforms (including igpu platforms) use FlatCCS
compression rather than AuxCCS.  Similar to PVC, any future platforms
that don't support FlatCCS should not attempt to fall back to AuxCCS
programming.

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Handle fused-off CCS engines

On Xe2 platforms, availability of the CCS engines is reflected in the
FUSE4 register.

Bspec: 62483
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Update context image layouts

Engine register state layout has changed a bit on Xe2.  We'll also
explicitly define a BCS layout to ensure BLIT_SWCTL and BLIT_CCTL are
included.

Bspec: 65182, 60184, 55793
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Add MCR register steering for media GT

Xe2 media has a few types of MCR registers, but all except for "GPMXMT"
can safely steer to instance 0,0.  GPMXMT follows the same rules that
MTL's OADDRM ranges did, so it can re-use the same enum value.

Bspec: 71186
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Add MCR register steering for primary GT

Xe2 uses the same steering control register and steering semaphore
register as MTL.  As with recent platforms, group/instance 0,0 is
sufficient to target a non-terminated instance for most classes of MCR
registers; the only types of ranges that need to consider platform
fusing to find a non-terminated instance are SLICE/DSS ranges and a new
SQIDI_PSMI type of range.

Note that the range of valid bits in XE2_NODE_ENABLE_MASK may be reduced
for some Xe2 SKUs.  However the lowest bits are always valid and only
the lowest instance is obtained via __ffs(), so there's no need to
complicate the masking with extra platform/subplatform checks.

Also note that Wa_14017387313 suggests skipping MCR lock acquisition
around GAM and GAMWKR registers to prevent MCR register accesses in an
interrupt handler from deadlocking when the steering semaphore is
already held outside the interrupt context.  At this time Xe never
issues MCR accesses from within an interrupt handler so the workaround
is not currently needed.

v2:
  - [0x008700-0x0087FF] range to extend up to 0x887F (Matt Attwood)
  - [0x00EF00-0x00F4FF] -> [0x00F000, 0xFFFF] to follow latest
    bspec version (Bala)

Bspec: 71185
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Add GT topology readout

Xe2 platforms have three DSS fuse registers for both geometry and
compute.

Bspec: 67171, 67537, 67401, 67536
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe/xe2: Update render/compute context image sizes

The render and compute context are significantly smaller on Xe2 than on
previous platforms.

Registers:
 - Render:  3008 dwords -> 12032 bytes -> round to 3 pages
 - Compute: 1424 dwords ->  5696 bytes -> round to 2 pages

We also allocate one additional page for the HWSP, so the total
allocation sizes for render and compute are 4 and 3 pages respectively.

Bspec: 65182, 56578, 55793
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: Stop tracking 4-tile support

The choice of Y-major tiling format (either the legacy "TileY" or the
newer "Tile4") is based on graphics IP version (12.50 and beyond have
Tile4, earlier platforms have TileY).  The tracking in xe was originally
added to allow re-using display from i915.  However as of i915 commit
4ebf43d0488f ("drm/i915: Eliminate has_4tile feature flag"), the display
code determines TileY vs Tile4 itself, so this can be removed from xe.

Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230817230407.909816-10-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: enable idle msg and set hysteresis for GSCCS

On MTL (and only on MTL) the GSCCS defaults with idle messaging
disabled. This means that, once awoken, the GSCCS will never signal its
idleness to the GT. To allow the GT to enter the proper low-power state,
we need therefore to turn idle messaging on. As part of this, we also
need to set a proper hysteresis value for the engine.

v2: use MEDIA_VERSION() and CLR() for the RTP rule and action, add reg
bit define in descending order (Matt)

Bspec: 71496
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817221707.1602873-1-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: don't expose the GSCCS to users

The kernel is the only expected user of the GSCCS, so we don't want to
expose it to userspace.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817201831.1583172-7-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: GSC forcewake support

The ID for the GSC forcewake domain already exists, but we're missing
the register definitions and the domain intialization, so add that in.

v2: move reg definition to be in address order (Matt)

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817201831.1583172-6-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: add GSCCS ring ops

Like the BCS, the GSCCS doesn't have any special HW that needs handling
when emitting commands, so we can re-use the same emit_job code. To make
it clear that this is now a shared low-level function, it has been
renamed to use the "simple" postfix, instead of "copy", to indicate that
it applies to all engines that don't need any additional engine-specific
handling.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817201831.1583172-5-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: add GSCCS irq support

The GSCCS has its own enable and mask registers. The interrupt identity
for the GSCCS shows OTHER_CLASS instance 6.

Bspec: 54029, 54030
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817201831.1583172-4-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/xe: base definitions for the GSCCS

The first step in introducing the GSCCS is to add all the basic defs for
it (name, mmio base, class/instance, lrc size etc).

Bspec: 60149, 60421, 63752
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230817201831.1583172-3-daniele.ceraolospurio@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>