drm/amdgpu: skip ASIC reset for APUs when go to S4

For GC IP v11.0.4/11, PSP TMR need to be reserved
for ASIC mode2 reset. But for S4, when psp suspend,
it will destroy the TMR that fails the ASIC reset.

[  96.006101] amdgpu 0000:62:00.0: amdgpu: MODE2 reset
[  100.409717] amdgpu 0000:62:00.0: amdgpu: SMU: I'm not done with your previous command: SMN_C2PMSG_66:0x00000011 SMN_C2PMSG_82:0x00000002
[  100.411593] amdgpu 0000:62:00.0: amdgpu: Mode2 reset failed!
[  100.412470] amdgpu 0000:62:00.0: PM: pci_pm_freeze(): amdgpu_pmops_freeze+0x0/0x50 [amdgpu] returns -62
[  100.414020] amdgpu 0000:62:00.0: PM: dpm_run_callback(): pci_pm_freeze+0x0/0xd0 returns -62
[  100.415311] amdgpu 0000:62:00.0: PM: pci_pm_freeze+0x0/0xd0 returned -62 after 4623202 usecs
[  100.416608] amdgpu 0000:62:00.0: PM: failed to freeze async: error -62

We can skip the reset on APUs, assuming we can resume them
properly. Verified on some GFX11, GFX10 and old GFX9 APUs.

Signed-off-by: Tim Huang <tim.huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: reposition the gpu reset checking for reuse

Move the amdgpu_acpi_should_gpu_reset out of
CONFIG_SUSPEND to share it with hibernate case.

Signed-off-by: Tim Huang <tim.huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: drop the extra sign extension

amdgpu_bo_gpu_offset_no_check() already calls
amdgpu_gmc_sign_extend() so no need to call it twice.

Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Merge tag 'drm-habanalabs-next-2023-03-20' of https://git./linux/kernel/git/ogabbay/linux into drm-next

This tag contains habanalabs driver and accel changes for v6.4:

- uAPI changes:

  - Add opcodes to the CS ioctl to allow user to stall/resume specific engines
    inside Gaudi2. This is to allow the user to perform power
    testing/measurements when training different topologies.

  - Expose in the INFO ioctl the amount of device memory that the driver
    and f/w reserve for themselves.

  - Expose in the INFO ioctl a bit-mask of the available rotator engines
    in Gaudi2. This is to align with other engines that are already exposed.

  - Expose in the INFO ioctl the register's address of the f/w that should
    be used to trigger interrupts from within the user's code running in the
    compute engines.

  - Add a critical-event bit in the eventfd bitmask so the user will know the
    event that was received was critical, and a reset will now occur

  - Expose in the INFO ioctl two new opcodes to fetch information on h/w and
    f/w events. The events recorded are the events that were reported in the
    eventfd.

- New features and improvements:

  - Add a dedicated interrupt ID in MSI-X in the device to the notification of
    an unexpected user-related event in Gaudi2. Handle it in the driver by
    reporting this event.

  - Allow the user to fetch the device memory current usage even when the
    device is undergoing compute-reset (a reset type that only clears the
    compute engines).

  - Enable graceful reset mechanism for compute-reset. This will give the
    user a few seconds before the device is reset. For example, the user can,
    during that time, perform certain device operations (dump data for debug)
    or close the device in an orderly fashion.

  - Align the decoder with the rest of the engines in regard to notification
    to the user about interrupts and in regard to performing graceful reset
    when needed (instead of immediate reset).

  - Add support for assert interrupt from the TPC engine.

  - Get the reset type that is necessary to perform per event from the
    auto-generated irq_map array.

  - Print the specific reason why a device is still in use when notifying to
    the user about it (after the user closed the device's FD).

  - Move to threaded IRQ when handling interrupts of workload completions.

- Firmware related fixes:

  - Fix RAZWI event handler to match newest f/w version.

  - Read error cause register in dma core events because the f/w doesn't
    do that.

  - Increase maximum time to wait for completion of Gaudi2 reset due to f/w
    bug.

  - Align to the latest firmware specs.

- Enforce the release order of the compute device and dma-buf.
  i.e increment the device file refcount for any dma-buf that was exported
  for that device. This will make sure the compute device release function
  won't be called until the user closes all the FDs of the relevant
  dma-bufs. Without this change, closing the device's FD before/without
  closing the dma-buf's FD would always lead to hard-reset of the device.

- Fix a link in the drm documentation to correctly point to the accel section.

- Compilation warnings cleanups

- Misc bug fixes and code cleanups

Signed-off-by: Dave Airlie <airlied@redhat.com>
# -----BEGIN PGP SIGNATURE-----
#
# iQEzBAABCgAdFiEE7TEboABC71LctBLFZR1NuKta54AFAmQYfcAACgkQZR1NuKta
# 54DB4Af/SuiHZkVXwr+yHPv9El726rz9ZQD7mQtzNmehWGonwAvz15yqocNMUSbF
# JbqE/vrZjvbXrP1Uv5UrlRVdnFHSPV18VnHU4BMS/WOm19SsR6vZ0QOXOoa6/AUb
# w+kF3D//DbFI4/mTGfpH5/pzwu51ti8aVktosPFlHIa8iI8CB4/4IV+ivQ8UW4oK
# HyDRkIvHdRmER7vGOfhwhsr4zdqSlJBYrv3C3Z1dkSYBPW/5ICbiM1UlKycwdYKI
# cajQBSdUQwUCWnI+i8RmSy3kjNO6OE4XRUvTv89F2bQeyK/1rJLG2m2xZR/Ml/o5
# 7Cgvbn0hWZyeqe7OObYiBlSOBSehCA==
# =wclm
# -----END PGP SIGNATURE-----
# gpg: Signature made Tue 21 Mar 2023 01:37:36 AEST
# gpg:                using RSA key ED311BA00042EF52DCB412C5651D4DB8AB5AE780
# gpg: Can't check signature: No public key
From: Oded Gabbay <ogabbay@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20230320154026.GA766126@ogabbay-vm-u20.habana-labs.com

Merge tag 'drm-intel-gt-next-2023-03-16' of git://anongit.freedesktop.org/drm/drm-intel into drm-next

Driver Changes:

- Fix issue #6333: "list_add corruption" and full system lockup from
  performance monitoring (Janusz)
- Give the punit time to settle before fatally failing (Aravind, Chris)
- Don't use stolen memory or BAR for ring buffers on LLC platforms (John)
- Add missing ecodes and correct timeline seqno on GuC error captures (John)
- Make sure DSM size has correct 1MiB granularity on Gen12+ (Nirmoy,
  Lucas)
- Fix potential SSEU max_subslices array-index-out-of-bounds access on Gen11 (Andrea)
- Whitelist COMMON_SLICE_CHICKEN3 for UMD access on Gen12+ (Matt R.)
- Apply Wa_1408615072/Wa_1407596294 correctly on Gen11 (Matt R)
- Apply LNCF/LBCF workarounds correctly on XeHP SDV/PVC/DG2 (Matt R)
- Implement Wa_1606376872 for Xe_LP (Gustavo)
- Consider GSI offset when doing MCR lookups on Meteorlake+ (Matt R.)
- Add engine TLB invalidation for Meteorlake (Matt R.)
- Fix GSC Driver-FLR completion on Meteorlake (Alan)
- Fix GSC races on driver load/unload on Meteorlake+ (Daniele)
- Disable MC6 for MTL A step (Badal)

- Consolidate TLB invalidation flow (Tvrtko)
- Improve debug GuC/HuC debug messages (Michal Wa., John)
- Move fd_install after last use of fence (Rob)
- Initialize the obj flags for shmem objects (Aravind)
- Fix missing debug object activation (Nirmoy)
- Probe lmem before the stolen portion (Matt A)
- Improve clean up of GuC busyness stats worker (John)
- Fix missing return code checks in GuC submission init (John)
- Annotate two more workaround/tuning registers as MCR on PVC (Matt R)
- Fix GEN8_MISCCPCTL definition and remove unused INF_UNIT_LEVEL_CLKGATE (Lucas)
- Use sysfs_emit() and sysfs_emit_at() (Nirmoy)
- Make kobj_type structures constant (Thomas W.)
- make kobj attributes const on gt/ (Jani)
- Remove the unused virtualized start hack on buddy allocator (Matt A)
- Remove redundant check for DG1 (Lucas)
- Move DG2 tuning to the right function (Lucas)
- Rename dev_priv to i915 for private data naming consistency in gt/ (Andi)
- Remove unnecessary whitelisting of CS_CTX_TIMESTAMP on Xe_HP platforms (Matt R.)
-

- Escape wildcard in method names in kerneldoc (Bagas)
- Selftest improvements (Chris, Jonathan, Tvrtko, Anshuman, Tejas)
- Fix sparse warnings (Jani)

[airlied: fix unused variable in intel_workarounds]
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ZBMSb42yjjzczRhj@jlahtine-mobl.ger.corp.intel.com

Merge tag 'drm-misc-next-2023-03-16' of git://anongit.freedesktop.org/drm/drm-misc into drm-next

drm-misc-next for v6.4-rc1:

Cross-subsystem Changes:
- Add drm_bridge.h to drm_bridge maintainers.

Core Changes:
- Assorted fixes to TTM, tests, format-helper, accel.
- Assorted Makefile fixes to drivers and accel.
- Implement fbdev emulation for GEM DMA drivers, and convert a lot of
  drivers to use it.
- Use tgid instead of pid for tracking clients.

Driver Changes:
- Assorted fixes in rockchip, vmwgfx, nouveau, cirrus.
- Add imx25 driver.
- Add Elida KD50T048A, Sony TD4353, Novatek NT36523, STARRY 2081101QFH032011-53G panels.
- Add 4K mode support to rockchip.
- Convert cirrus to use regular atomic helpers, and more cirrus
  improvements.
- Add damage clipping to cirrus, virtio.

[airlied: add drm_bridge.h include to imx]
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/f7b765c7-d49d-edb5-2a6a-4f7a7be16a59@linux.intel.com

accel/habanalabs: remove redundant TODOs

As mmu refactor and nic resume are not relevant anymore, remove
their TODO comments.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: change razwi handle after fw fix

FW had one data route for tpc0 and tpc1 when running in secured mode
and a different one when running without secured mode. After fw fixed
this issue, both mode have the same data path.

Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: add handling for unexpected user event

In order for the user to be aware of unexpected events in Gaudi2 that
aren't assigned to a specific engine, we are adding the handling of
this dedicated interrupt.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix a missing-braces compilation warning

Replace initialization of "struct cpucp_packet" from "{0} to "{}" to
avoid a "missing braces around initializer" compilation warning.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix a maybe-uninitialized compilation warnings

Initialize 'index' in gaudi2_handle_qman_err() and 'offset' in
gaudi2_get_nic_idle_status() to avoid "maybe-uninitialized" compilation
warnings.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix page fault event clear

After getting page fault in gaudi2, we need to clear the valid bit
instead of the address.

Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: expose rotator mask to userspace

All engine masks are exposed to user, make sure user gets the
correct rotator enabled mask in gaudi2.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: regenerate gaudi2 ids_map_extended

Some names of events has been modified/added.

Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel: Link to compute accelerator subsystem intro

Commit 2c204f3d53218d ("accel: add dedicated minor for accelerator
devices") adds link to accelerator nodes section of DRM internals doc
(Documentation/gpu/drm-internals.rst), but the target doesn't exist.
Instead, there is only an introduction doc for computer accelerator
subsytem.

Link to that doc until there is documentation of accelerator internals.

Fixes: 2c204f3d53218d ("accel: add dedicated minor for accelerator devices")
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: expose dram reserved size by kmd

We expose this in order for user applications to know how much dram
is reserved for internal use.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: remove '\n' when passing strings to gaudi2_print_event()

Remove all '\n' from strings which are passed as arguments to
gaudi2_print_event(), because the newline character is added internally
in this function.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: return tlb inv error code upon failure

Now that CQ-completion based jobs do not trigger a reset upon failure,
failure of such jobs (e.g., MMU cache invalidation) should be handled
by the caller itself depending on the error code returned to it.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix field names in hl_info_hw_ip_info

Don't use padX for actual reservedX fields.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Ofir Bitton <obitton@habana.ai>

accel/habanalabs: in {e/p}dma_core events read the err cause reg

Since the err_cause register is unprivileged, we should read it from
the driver instead of using the param that came from the FW.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix use of var reset_sleep_ms

- remove reset_sleep_ms arg from functions that don't use it.
- move the call msleep(reset_sleep_ms) from btm poll to gaudi2_hw_fini
as it is called from there already for other flow.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: in hw_fini return error code if polling timed-out

In hw_fini callback, we use either the cpucp packet method or polling a
register. Currently we return error only in the case of cpucp packet
failure. In this patch we also return error if polling timed out.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: increase reset poll timeout

Due to a firmware bug we need to increase reset poll timeout
or else we will timeout in secured environments.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: do not verify engine modes after being changed

Engines idle state can't always be verified between changes of
engine modes (e.g., stall/halt).
For example, if a CS is inflight when altering engine's mode,
idle state will return NOT idle, always.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: align to latest firmware specs

Copy the most up-to-date interface files to the firmware.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Ofir Bitton <obitton@habana.ai>

accel/habanalabs: Fix spelling mistake "maped" -> "mapped"

There is a spelling mistake in a dev_err message. Fix it.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: make gaudi2_is_device_idle() static

This function is only called inside gaudi2.c file.

Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202303071320.X5ouBlNY-lkp@intel.com/
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Ofir Bitton <obitton@habana.ai>

Merge tag 'amd-drm-next-6.4-2023-03-17' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-6.4-2023-03-17:

amdgpu:
- Misc code cleanups
- Documentation fixes
- Make kobj structures const
- Add thermal throttling adjustments for supported APUs
- UMC RAS fixes
- Display reset fixes
- DCN 3.2 fixes
- Freesync fixes
- DC code reorg
- Generalize dmabuf import to work with KFD
- DC DML fixes
- SRIOV fixes
- UVD code cleanups
- IH 4.4.2 updates
- HDP 4.4.2 updates
- SDMA 4.4.2 updates
- PSP 13.0.6 updates
- Add capped/uncapped workload handling for supported APUs
- DCN 3.1.4 updates
- Re-org DC Kconfig
- USB4 fixes
- Reorg DC plane and stream handling
- Register vga_switcheroo for apple-gmux
- SMU 13.0.6 updates
- Fix error checking in read_mm_registers functions for affected families
- VCN 4.0.4 fix
- Drop redundant pci_enable_pcie_error_reporting() call
- RDNA2 SMU OD suspend/resume fix
- Expose additional memory stats via fdinfo
- RAS fixes
- Misc display fixes
- DP MST fixes
- IOMMU regression fix for KFD

amdkfd:
- Make kobj structures const
- Support for exporting buffers via dmabuf
- Multi-VMA page migration fixes
- NBIO fixes
- Misc code cleanups
- Fix possible double free
- Fix possible UAF

radeon:
- iMac fix

UAPI:
- KFD dmabuf export support.  Required for importing KFD buffers into GEM contexts and for RDMA P2P support.
  Proposed user mode changes: https://github.com/fxkamd/ROCT-Thunk-Interface/commits/fxkamd/dmabuf

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230317164416.138340-1-alexander.deucher@amd.com

drm/nouveau/nvfw/acr: set wpr_generic_header_dump storage-class-specifier to static

gcc with W=1 reports
drivers/gpu/drm/nouveau/nvkm/nvfw/acr.c:49:1: error: no previous
  prototype for ‘wpr_generic_header_dump’ [-Werror=missing-prototypes]
   49 | wpr_generic_header_dump(struct nvkm_subdev *subdev,
      | ^~~~~~~~~~~~~~~~~~~~~~~

wpr_generic_header_dump is only used in acr.c, so it should be static

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230302124819.686469-1-trix@redhat.com

drm/nouveau/fifo: set nvkm_engn_cgrp_get storage-class-specifier to static

smatch reports
drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.c:33:18:
  warning: symbol 'nvkm_engn_cgrp_get' was not declared. Should it be static?

nvkm_engn_cgrp_get is only used in runl.c, so it should be static

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230228221533.3240520-1-trix@redhat.com

drm/nouveau/fifo: set gf100_fifo_nonstall_block_dump storage-class-specifier to static

gcc with W=1 reports
drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c:451:1: error:
  no previous prototype for ‘gf100_fifo_nonstall_block’ [-Werror=missing-prototypes]
  451 | gf100_fifo_nonstall_block(struct nvkm_event *event, int type, int index)
      | ^~~~~~~~~~~~~~~~~~~~~~~~~

gf100_fifo_nonstall_block is only used in gf100.c, so it should be static

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230303132731.1919329-1-trix@redhat.com

drm/amdgpu: Don't resume IOMMU after incomplete init

Check kfd->init_complete in kgd2kfd_iommu_resume, consistent with other
kgd2kfd calls. This should fix IOMMU errors on resume from suspend when
KFD IOMMU initialization failed.

Reported-by: Matt Fagnani <matt.fagnani@bell.net>
Link: https://lore.kernel.org/r/4a3b225c-2ffd-e758-4de1-447375e34cad@bell.net/
Link: https://bugzilla.kernel.org/show_bug.cgi?id=217170
Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2454
Cc: Vasant Hegde <vasant.hegde@amd.com>
Cc: Linux regression tracking (Thorsten Leemhuis) <regressions@leemhuis.info>
Cc: stable@vger.kernel.org
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Tested-by: Matt Fagnani <matt.fagnani@bell.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: fix compilation issue with legacy gcc

This patch is used to fix following compilation issue with legacy gcc

error: ‘for’ loop initial declarations are only allowed in C99 mode

Signed-off-by: bobzhou <bob.zhou@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Retire pcie_gen3_enable function

Not needed since from vi. drop the function so
we don't duplicate code when introduce new asics.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Likun Gao <Likun.Gao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Move to common helper to query soc rev_id

Replace soc15, nv, soc21 get_rev_id callback with common
helper so we don't need to duplicate code when introduce
new asics.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Likun Gao <Likun.Gao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Move to common indirect reg access helper

Replace soc15, nv, soc21 specific callbacks with common
one. so we don't need to duplicate code when introduce
new asics.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Likun Gao <Likun.Gao@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: drop ras check at asic level for new blocks

amdgpu_ras_register_ras_block should always be invoked
by ras_sw_init, where driver needs to check ras caps
at ip level, instead of asic level.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Stanley Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Rework pcie_bif ras sw_init

pcie_bif ras blocks needs to be initialized as early
as possible to handle fatal error detected in hw_init
phase. also align the pcie_bif ras sw_init with other
ras blocks

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Stanley Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Rework xgmi_wafl_pcs ras sw_init

To align with other IP blocks.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Stanley Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Rework mca ras sw_init

To align with other IP blocks

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Stanley Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: Fixed kfd_process cleanup on module exit.

Handle case when module is unloaded (kfd_exit) before a process space
(mm_struct) is released.

v2: Fixed potential race conditions by removing all kfd_process from
the process table first, then working on releasing the resources.

v3: Fixed loop element access / synchronization.  Fixed extra empty lines.

Signed-off-by: David Belanger <david.belanger@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: 3.2.227

This version brings along the following:
- FW Release 0.0.158.0
- Fixes to HDCP, DP MST and more
- Improvements on USB4 links and more
- Code re-architecture on link.h

Reviewed-by: Aric Cyr <aric.cyr@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Aric Cyr <aric.cyr@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: [FW Promotion] Release 0.0.158.0

[Why & How]
Add boot control bit to control dispclk and dppclk deep sleep

Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Anthony Koo <Anthony.Koo@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: fix assert condition

[Why & How]
Reversed assert condition when checking that phy_pix_clk[] is not 0

Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Samson Tam <Samson.Tam@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Clearly states if long or short HPD event in dmesg logs

[Why]
The log "DMUB HPD callback" is crucial to identify when DP tunneling
is been established and driver is notified of this event from DMUB.
Same log is shared for long and short hotplug event and we need to
check trailing DC debug log to distinguish between them two, making
debugging on DPIA related issues a bit more troublesome.

[How]
Clearly states in dmesg logs whether this is a long or short hotplug
event.

Reviewed-by: Hamza Mahfooz <Hamza.Mahfooz@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Stylon Wang <stylon.wang@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Make DCN32 functions available to future DCNs

[Why & How]
Make DCN32 functions available for more DCNs.

Reviewed-by: Chris Park <Chris.Park@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Wesley Chalmers <Wesley.Chalmers@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Init MMVM_CONTEXTS_DISABLE in gmc11 golden setting under SRIOV

[Why]
If disable the mmhub vm contexts(set MMVM_CONTEXTS_DISABLE to 0xffff),
driver loading failed on vf due to fence fallback timer expired on all rings.
FLR cannot reset MMVM_CONTEXTS_DISABLE.
So this vf can not be recovered anymore unless trigger a whole gpu reset.

[How]
Under SRIOV, init MMVM_CONTEXTS_DISABLE in gmc11 golden register setting.

Signed-off-by: Yifan Zha <Yifan.Zha@amd.com>
Reviewed-by: Horace Chen <Horace.Chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: reallocate DET for dual displays with high pixel rate ratio

[Why]
For dual displays where pixel rate is much higher on one display,
we may get underflow when DET is evenly allocated.

[How]
Allocate less DET segments for the lower pixel rate display and
more DET segments for the higher pixel rate display

Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Samson Tam <Samson.Tam@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: disconnect MPCC only on OTG change

[Why]
Framedrops are observed while playing Vp9 and Av1 10 bit
video on 8k resolution using VSR while playback controls
are disappeared/appeared

[How]
Now ODM 2 to 1 is disabled for 5k or greater resolutions on VSR.

Cc: stable@vger.kernel.org
Cc: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Ayush Gupta <ayugupta@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Fix DP MST sinks removal issue

[Why]
In USB4 DP tunneling, it's possible to have this scenario that
the path becomes unavailable and CM tears down the path a little bit late.
So, in this case, the HPD is high but fails to read any DPCD register.
That causes the link connection type to be set to sst.
And not all sinks are removed behind the MST branch.

[How]
Restore the link connection type if it fails to read DPCD register.

Cc: stable@vger.kernel.org
Cc: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Wenjing Liu <Wenjing.Liu@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Cruise Hung <Cruise.Hung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: hpd rx irq not working with eDP interface

[Why]
This is the fix for the defect of commit ab144f0b4ad6
("drm/amd/display: Allow individual control of eDP hotplug support").

[How]
To revise the default eDP hotplug setting and use the enum to git rid
of the magic number for different options.

Fixes: ab144f0b4ad6 ("drm/amd/display: Allow individual control of eDP hotplug support")
Cc: stable@vger.kernel.org
Cc: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Wenjing Liu <Wenjing.Liu@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Robin Chen <robin.chen@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/vmwgfx: Fix src/dst_pitch confusion

The src/dst_pitch got mixed up during the rework of the function, make
sure the offset's refer to the correct one.

Spotted by clang:
Clang warns (or errors with CONFIG_WERROR):

  drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c:509:29: error: variable 'dst_pitch' is uninitialized when used here [-Werror,-Wuninitialized]
          src_offset = ddirty->top * dst_pitch + ddirty->left * stdu->cpp;
                                     ^~~~~~~~~
  drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c:492:26: note: initialize the variable 'dst_pitch' to silence this warning
          s32 src_pitch, dst_pitch;
                                  ^
                                   = 0
  1 error generated.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reported-by: Nathan Chancellor <nathan@kernel.org>
Reported-by: Dave Airlie <airlied@gmail.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/1811
Fixes: 39985eea5a6d ("drm/vmwgfx: Abstract placement selection")
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230314211445.1363828-1-zack@kde.org

drm: Track clients by tgid and not tid

Thread group id (aka pid from userspace point of view) is a more
interesting thing to show as an owner of a DRM fd, so track and show that
instead of the thread id.

In the next patch we will make the owner updated post file descriptor
handover, which will also be tgid based to avoid ping-pong when multiple
threads access the fd.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230314141904.1210824-2-tvrtko.ursulin@linux.intel.com

accel/habanalabs: Drop redundant pci_enable_pcie_error_reporting()

pci_enable_pcie_error_reporting() enables the device to send ERR_*
Messages. Since
commit f26e58bf6f54 ("PCI/AER: Enable error reporting when AER is native"),
the PCI core does this for all devices during enumeration, so the
driver doesn't need to do it itself.

Remove the redundant pci_enable_pcie_error_reporting() call from the
driver.  Also remove the corresponding pci_disable_pcie_error_reporting()
from the driver .remove() path.

Note that this only controls ERR_* Messages from the device.  An ERR_*
Message may cause the Root Port to generate an interrupt, depending on the
AER Root Error Command register managed by the AER service driver.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: postpone mem_mgr IDR destruction to hpriv_release()

The memory manager IDR is currently destroyed when user releases the
file descriptor.
However, at this point the user context might be still held, and memory
buffers might be still in use.
Later on, calls to release those buffers will fail due to not finding
their handles in the IDR, leading to a memory leak.
To avoid this leak, split the IDR destruction from the memory manager
fini, and postpone it to hpriv_release() when there is no user context
and no buffers are used.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: move soft-reset wait to soft-reset execute

We plan to do soft-reset either by mmio or by using cpucp packet
depending on the FW version. We don't want to check FW version in two
different places for that (execute soft-reset and wait to soft-reset)
so move the waiting to gaudi2_execute_soft_reset. This also makes sense
because the cpucp also does the waiting.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: add uapi to stall/resume engine

The user might want to stall/resume engines to perform power testing
for various scenarios. Because our current
HL_CS_FLAGS_ENGINE_CORE_COMMAND command only handles the engines' cores,
we need to add another opcode for handling entire engine and not just
its core.

The user supplies an array, where each entry holds the engine's ID and
the command to send to the engine. The size of the array is limited
by the number of engines in the ASIC (only Gaudi2 is currently
supported).

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: use scnprintf() in print_device_in_use_info()

compose_device_in_use_info() was added to handle the snprintf() return
value in a single place.
However, the buffer size in print_device_in_use_info() is set such that
it would be enough for the max possible print, so
compose_device_in_use_info() is not really needed.
Moreover, scnprintf() can be used instead of snprintf(), to save the
check if the return value larger than the given size.

Cc: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: unify err log of hw-fini failure in dirty state

print more informative message when failing in dirty state

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: use a mutex rather than a spinlock

There are two reasons why mutex is better here:
1. There's a critical section relatively long, where in
certain scenarios (e.g., multiple VM allocations) taking a spinlock
might cause noticeable performance degradation.
2. It will remove the incorrect usage of mutex under
spin_lock (where preemption is disabled).

Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: allow getting HL_INFO_DRAM_USAGE during soft-reset

We can allow userspace to query the dram usage during soft-reset.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix register address on PDMA/EDMA idle check

The PDMA/EDMA is_idle routines didn't check the correct CORE register
in order to get the accurate idle state.
Moreover, it's better to make the is_idle routine more robust by adding
additional checks (IS_HALTED) before announcing that the core is idle.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: remove a useless is_idle TPC flag

Is appears that the flag -
DCORE0_TPC0_CFG_STATUS_VECTOR_PIPE_EMPTY_MASK, has no actual use when
it comes to querying TPC idleness, since this flag's corresponding bit
turns-off after stalling the engine, and turns back on after resuming
it.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: fix few misspelled words in the code

Run spell checker on the code and fix accordingly.

Signed-off-by: farah kassabri <fkassabri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: verify return code after scrubbing ARCs DCCMs

In case the KDMA fails scrubbing the DCCMs (following a soft-reset
upon device release), the driver will only print failure until reset
flow ends, rather than escalating it into a hard-reset.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: use notifications and graceful reset for decoder

Add notifications to user in case of decoder abnormal interrupts, and
use the graceful reset mechanism if reset is required.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: assert return value of hw_fini

Since hw_fini return error code for failure indication, we should
check its return value. Currently it might only fail upon soft-reset
from hl_device_reset. Later patch will add hw_fini failure in case of
polling timeout in hard-reset.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: break is_idle function into per-engine sub-routines

is_idle() was too long, so break it up for readability.
In addition, we can now use the new sub-routines from other places.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: add device id to all threads names

Compute driver threads names will start with hlX-*, when X is the
device id.
This will help distinguish them from the NIC thread names.

Signed-off-by: Sagiv Ozeri <sozeri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: add helper function to get vm hash node

Add a helper function to search the vm hash for a node with a given
virtual address.
As opposed to the current code, this function explicitly returns NULL
when no node is found, instead of basing on the loop cursor object's
value.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: remove unneeded irq_handler variable

'irq_handler' in gaudi2_enable_msix(), is just assigned with a function
name and then used when calling request_threaded_irq().
Remove the variable and use the function name directly as an argument.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: change hw_fini to return int to indicate error

We later use cpucp packet for soft reset which might fail
so we should be able propagate the failure case.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: improve readability of engines idle mask print

Remove leading zeroes when printing the idle mask to make it clearer.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: organize hl_device structure comment

Make the comments align with the order of the fields in the structure

Signed-off-by: Sagiv Ozeri <sozeri@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: set hl_capture_*_err storage-class-specifier to static

smatch reports
drivers/accel/habanalabs/common/device.c:2619:6: warning:
  symbol 'hl_capture_hw_err' was not declared. Should it be static?
drivers/accel/habanalabs/common/device.c:2641:6: warning:
  symbol 'hl_capture_fw_err' was not declared. Should it be static?

both are only used in device.c, so they should be static

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: change unused extern decl of hdev to forward decl of hl_device

Building with clang W=2 has several similar warnings
drivers/accel/habanalabs/common/decoder.c:46:51: error: declaration shadows a variable in the global scope [-Werror,-Wshadow]
static void
dec_error_intr_work(struct hl_device *hdev, u32 base_addr, u32 core_id)
                                                  ^
drivers/accel/habanalabs/common/security.h:13:26: note: previous declaration is here
extern struct hl_device *hdev;
                         ^

There is no global definition of hdev, so the extern is not needed.
Searched with
grep -r '^struct' . | grep hl_dev

Change to an forward decl to resolve these issues
drivers/accel/habanalabs/common/mmu/../security.h:133:40: error: ‘struct hl_device’ declared inside parameter list will not be visible outside of this definition or declaration [-Werror]
  133 |         bool (*skip_block_hook)(struct hl_device *hdev,
      |                                        ^~~~~~~~~

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>

accel/habanalabs: don't trace cpu accessible dma alloc/free

The cpu accessible dma allocations use the gen_pool api which actually
does not allocate new memory from the system but manages memory already
allocated before. When tracing this together with real dma
allocation/free it cause confusing logs like a '0' dma address and
a cpu address appearing twice etc.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: in hl_device_reset small refactor for readabilty

in the out_err flow, combine the two cases of soft-reset since
they have mostly common code. In addition unlock reset_info.lock
after touching reset count.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: in hl_device_reset remove 'hard_instead_of_soft'

Because this field is only used for debug print,
we can do more precise debug directly instead.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: rename security function parameters

To match their description above the function

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: tiny refactor of hl_device_reset for readability

Align assignment of reset_upon_device_release to the convention used
in this function.

Signed-off-by: Dafna Hirschfeld <dhirschfeld@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: remove hl_irq_handler_default()

hl_irq_handler_default() is not used and can be removed.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: fix print in hl_irq_handler_eq()

"eq_base[eq->ci].hdr.ctl" is used directly in a print without a
le32_to_cpu() conversion.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: add support for TPC assert

In order to allow TPC engines to raise an assert, we must expose
the relevant MSIX interrupt to the user so he will configure the engine
correctly. In addition, we implement the corresponding interrupt
handler that will notify the user upon such an event.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: capture interrupt timestamp in handler

In order for interrupt timestamp to be more accurate we should
capture it during the interrupt handling rather than in threaded
irq context.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: change user interrupt to threaded IRQ

We prefer not to handle the user interrupt job inside the interrupt
context. Instead, use threaded IRQ to handle the user interrupts.
This will allow to avoid disabling interrupts when the user process
registers for a new event and to avoid long handling inside an
interrupt.

Signed-off-by: Tal Cohen <talcohen@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: modify events reset policy

The policy file of the events reset has been modified.
This change is reflected in the autogenerated file.

Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: get reset type indication from irq_map

When getting an event, add the ability to deduce the reset type from
the IRQ map table instead of using hard reset regardless.

Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: enable graceful reset mechanism for compute-reset

The graceful reset mechanism is currently enabled only for reset
requests that will end up with hard-reset.
In future, reset requests due to errors in some device engines, are
going to be modified to request compute-reset, as the much longer
hard-reset is not really needed there.
To allow it, enable graceful reset also for compute-reset, and reset
after user releases the device won't be escalated to hard-reset in those
cases.
If watchdog expires and user didn't release the device, hard-reset will
be initiated in any case.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: disable PCI when escalating compute to hard-reset

In case a compute reset has failed or a request for a hard reset has
just arrived, then we escalate current reset procedure from compute
to hard-reset.
In such a case, the FW should be aware of the updated error cause,
and if LKD is the one who performs the reset (rather than the FW),
then we ask the FW to disable PCI access.

We would also like to have relevant debug info and therefore
we print the currently escalating reset type.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: minimize error prints when mem map fails

This commit minimizes the "chain of errors" displayed when memory
mapping fails.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: unsecure CFG_TPC_ID register

Required to allow the TPC compiler to know on which offset of the index
space it works on.

Signed-off-by: Koby Elbaz <kelbaz@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: expose engine core int reg address

In order for engine cores to raise interrupts towards FW, They need
to know which register the event data should be written to.
Hence, we forward the relevant scratchpad register received during
dynamic regs handshake with FW.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: add critical-event bit in notifier

Enhance the existing user notifications by adding a HW and FW critical
event bits to be used when a HW or FW event occur that requires
both SW abort and hard-resetting the chip.

Signed-off-by: Moti Haimovski <mhaimovski@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: enforce release order of compute device and dma-buf

When user closes the compute device file descriptor without closing a
dma-buf file descriptor, the device will be considered as in use,
leading to hard reset and killing the user process, to ensure the
release of the dma-buf.
Same thing will happen if user first releases the compute device file
and only then the dma-buf.

The implication of this is the duration of hard reset, during which the
device cannot be reacquired.
Moreover, this behavior adds a constraint on a user process to follow
this order of release operations.

To avoid killing the user process and to remove this constraint, enforce
the correct order of release operations inside the driver, by
incrementing the device file refcount for any dma-buf until it is
released.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: add info when FD released while device still in use

When user closes the device file descriptor, it is checked whether the
device is still in use, and a message is printed if it is.
To make this message more informative, add to this print also the reason
due to which the device is considered as in use.
The possible reasons which are checked for now are active CS and
exported dma-buf.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: fix address decode RAZWI handling

PSOC RAZWI handling code did not took into account single router that
supports several initiators with different XY coordinates. Also, it
ignored XY_HI coordinate. This caused 2 problems:
1. RAZWI handle ignored some initiators.
2. When getting PSOC RAZWI from some routers, there was a lot of
   possible engines which could have caused the RAZWI.

Fixed the above issue by handling PSOC RAZWI with both low and high
XY coordinates. This way driver supports all initiators and in
the worst case there are not more than 2 possible engines for RAZWI.

Signed-off-by: Dani Liberman <dliberman@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: use memhash_node_export_put() in hl_release_dmabuf()

The same mutex lock/unlock and counter decrementing in
hl_release_dmabuf() is already done in the memhash_node_export_put()
helper function.

Signed-off-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: refactor debugfs init

Make it easier to later add support for accel device.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>

accel/habanalabs: save class in hdev

It is more concise than to pass it to device init. Once we will add the
accel class, then we won't need to change the function signatures.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Tomer Tayar <ttayar@habana.ai>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>