openwrt/staging/blogic.git
6 years agodrm/amdgpu: fix amdgpu_vm_pasid_fault_credit
Christian König [Tue, 9 Jan 2018 18:18:59 +0000 (19:18 +0100)]
drm/amdgpu: fix amdgpu_vm_pasid_fault_credit

As soon as the lock is dropped the VM pointer can be invalid.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: check the return value of register_shrinker
Roger He [Thu, 11 Jan 2018 04:57:58 +0000 (12:57 +0800)]
drm/ttm: check the return value of register_shrinker

This fixes the build warning:
"ignoring return value of 'register_shrinker', declared with
attribute warn_unused_result [-Wunused-result]"

Signed-off-by: Roger He <Hongbo.He@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/radeon: fix sparse warning: Should it be static?
Dmitry Rozhkov [Tue, 2 Jan 2018 08:02:51 +0000 (10:02 +0200)]
drm/radeon: fix sparse warning: Should it be static?

Impact: make symbol static.

Fix this sparse warning:
  drivers/gpu/drm/radeon/radeon_dp_mst.c:332:38: warning: symbol 'mst_cbs' was not declared. Should it be static?

Signed-off-by: Dmitry Rozhkov <dmitry.rozhkov@linux.intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/pp: Implement get_max_high_clocks for CI/VI
Rex Zhu [Wed, 8 Nov 2017 08:39:00 +0000 (16:39 +0800)]
drm/amd/pp: Implement get_max_high_clocks for CI/VI

v2: add table length check.

DC component expect PP to give max engine clock and
memory clock through pp_get_display_mode_validation_clocks
on DGPU as well.

This patch can fix MultiGPU-Display blank
out with 1 IGPU-4k display and 2 DGPU-two 4K
displays.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoMAINTAINERS: add David (Chunming) Zhou as additional amdgpu maintainer
Christian König [Tue, 9 Jan 2018 11:33:13 +0000 (12:33 +0100)]
MAINTAINERS: add David (Chunming) Zhou as additional amdgpu maintainer

Just another contact for the amdgpu driver.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: fix 64bit BAR detection
Christian König [Sun, 7 Jan 2018 09:18:57 +0000 (10:18 +0100)]
drm/amdgpu: fix 64bit BAR detection

Windows added by the BIOS are not marked as 64bit because they are
usually not changeable anyway.

This fixes large BAR support on my new Ryzen build system.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: optimize moved handling only when vm_debug is inactive
Christian König [Wed, 3 Jan 2018 12:36:22 +0000 (13:36 +0100)]
drm/amdgpu: optimize moved handling only when vm_debug is inactive

Otherwise we would completely circumvent that debugging feature.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: simplify huge page handling
Christian König [Thu, 21 Dec 2017 14:47:28 +0000 (15:47 +0100)]
drm/amdgpu: simplify huge page handling

Update the PDEs after resetting the huge flag.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: update VM PDs after the PTs
Christian König [Fri, 1 Sep 2017 18:37:57 +0000 (20:37 +0200)]
drm/amdgpu: update VM PDs after the PTs

Necessary for the next patch.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: minor optimize VM moved handling v2
Christian König [Fri, 1 Sep 2017 18:34:27 +0000 (20:34 +0200)]
drm/amdgpu: minor optimize VM moved handling v2

Try to lock moved BOs if it's successful we can update the
PTEs directly to the new location.

v2: rebase

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: loosen the criteria for huge pages a bit
Christian König [Thu, 21 Dec 2017 12:30:50 +0000 (13:30 +0100)]
drm/amdgpu: loosen the criteria for huge pages a bit

We can actually handle invalid huge pages perfectly fine now.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/powerplay: set pp_num_states as 0 on error situation
Evan Quan [Thu, 28 Dec 2017 06:37:58 +0000 (14:37 +0800)]
drm/amd/powerplay: set pp_num_states as 0 on error situation

Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: specify DMA_ATTR_NO_WARN for huge page pools
Christian König [Thu, 4 Jan 2018 14:09:39 +0000 (15:09 +0100)]
drm/ttm: specify DMA_ATTR_NO_WARN for huge page pools

Suppress warning messages when allocating huge pages fails since we can
always fall back to normal pages.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: remove ttm_bo_default_io_mem_pfn
Tan Xiaojun [Mon, 25 Dec 2017 03:43:34 +0000 (11:43 +0800)]
drm/ttm: remove ttm_bo_default_io_mem_pfn

No one will use this function except ttm_bo_io_mem_pfn() now, so move
the calculation of ttm_bo_default_io_mem_pfn() into ttm_bo_io_mem_pfn()
and do some cleanup.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agostaging: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:33 +0000 (11:43 +0800)]
staging: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/powerplay: fix memory leakage when reload (v2)
Yintian Tao [Thu, 4 Jan 2018 03:45:23 +0000 (11:45 +0800)]
drm/amd/powerplay: fix memory leakage when reload (v2)

add smu_free_memory when smu fini to prevent memory leakage

v2: squash in typo fix (Yintian) and warning (Harry)

Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu/gfx9: only init the apertures used by KGD (v2)
Alex Deucher [Mon, 1 Jan 2018 21:14:43 +0000 (16:14 -0500)]
drm/amdgpu/gfx9: only init the apertures used by KGD (v2)

Use adev->vm_manager.id_mgr[0].num_ids rather than hardcoded 16.

v2: use AMDGPU_GFXHUB rather than hardcoded 0 (Christian)

Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Noticed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: add atpx quirk handling (v2)
Alex Deucher [Wed, 20 Dec 2017 18:29:58 +0000 (13:29 -0500)]
drm/amdgpu: add atpx quirk handling (v2)

Add quirks for handling PX/HG systems.  In this case, add
a quirk for a weston dGPU that only seems to properly power
down using ATPX power control rather than HG (_PR3).

v2: append a new weston XT

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com> (v2)
Reviewed-and-Tested-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
6 years agodrm/amdgpu: Add dpm quirk for Jet PRO (v2)
Alex Deucher [Tue, 21 Nov 2017 17:10:57 +0000 (12:10 -0500)]
drm/amdgpu: Add dpm quirk for Jet PRO (v2)

Fixes stability issues.

v2: clamp sclk to 600 Mhz

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103370
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
6 years agodrm/radeon: Add dpm quirk for Jet PRO (v2)
Alex Deucher [Tue, 21 Nov 2017 17:09:38 +0000 (12:09 -0500)]
drm/radeon: Add dpm quirk for Jet PRO (v2)

Fixes stability issues.

v2: clamp sclk to 600 Mhz

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103370
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
6 years agodrm/amdgpu: use %pap format string for phys_addr_t
Arnd Bergmann [Mon, 8 Jan 2018 12:53:56 +0000 (13:53 +0100)]
drm/amdgpu: use %pap format string for phys_addr_t

The newly added get_local_mem_info() function prints a phys_addr_t
using 0x%llx, which is wrong on most 32-bit systems, as shown by
this warning:

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c: In function 'get_local_mem_info':
include/linux/kern_levels.h:5:18: error: format '%llx' expects argument of type 'long long unsigned int', but argument 2 has type 'resource_size_t {aka unsigned int}' [-Werror=format=]
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c:297:31: note: format string is defined here
  pr_debug("Address base: 0x%llx limit 0x%llx public 0x%llx private 0x%llx\n",

Passing the address by reference to the special %pap format string will
produce the correct output and avoid the warning.

Fixes: 30f1c0421ec5 ("drm/amdgpu: Implement get_local_mem_info")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoMerge tag 'drm-misc-next-2018-01-08' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Tue, 9 Jan 2018 00:24:17 +0000 (10:24 +1000)]
Merge tag 'drm-misc-next-2018-01-08' of git://anongit.freedesktop.org/drm/drm-misc into drm-next

drm-misc-next for 4.16:

Cross-subsystem Changes:

- some dt-binding changes for Ilitek and sun4i devices

Core Changes:

- panel_orientation_quirks: fix tainted kernel

Driver Changes:

- panel changes
- A83T and LVDS support to sun4i

* tag 'drm-misc-next-2018-01-08' of git://anongit.freedesktop.org/drm/drm-misc:
  drm/panel: lvds: Add support for the power-supply property
  dt-bindings: panel: lvds: Document power-supply property
  drm/sun4i: Add A83T support
  drm/sun4i: Add LVDS support
  drm/sun4i: Create minimal multipliers and dividers
  drm/sun4i: Force the mixer rate at 150MHz
  dt-bindings: display: sun4i-drm: Add A83T pipeline
  dt-bindings: display: sun4i-drm: Add LVDS properties
  drm/tinydrm: add driver for ST7735R panels
  dt-bindings: Add binding for Sitronix ST7735R display panels
  dt-bindings: add jianda vendor prefix
  drm/tinydrm: Update ILI9225 compatible string
  dt-bindings: update compatible string for ILI9225
  dt-bindings: Add "vot" vendor prefix
  drm: fix tainted kernel caused by drm_panel_orientation_quirks.c
  drm/panel: Add Ilitek ILI9322 driver
  drm/panel: Add DT bindings for Ilitek ILI9322

6 years agoMerge branch 'drm-next-4.16' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Tue, 9 Jan 2018 00:09:13 +0000 (10:09 +1000)]
Merge branch 'drm-next-4.16' of git://people.freedesktop.org/~agd5f/linux into drm-next

Last few updates for 4.16:
- Misc fixes for amdgpu
- Enable swapout for reserved BOs during allocation for ttm
- Misc cleanups for ttm

* 'drm-next-4.16' of git://people.freedesktop.org/~agd5f/linux: (24 commits)
  drm/amdgpu: Correct the IB size of bo update mapping.
  drm/ttm: enable swapout for reserved BOs during allocation
  drm/ttm: add new function to check if bo is allowable to evict or swapout
  drm/ttm: use an operation ctx for ttm_tt_bind
  drm/ttm: use an operation ctx for ttm_tt_populate in ttm_bo_driver (v2)
  drm/ttm: use an operation ctx for ttm_mem_global_alloc_page
  drm/ttm: use an operation ctx for ttm_mem_global_alloc
  drm/ttm: call ttm_bo_swapout directly when ttm shrink
  drm/vmwgfx: remove the default io_mem_pfn set
  drm/virtio: remove the default io_mem_pfn set
  drm/radeon: remove the default io_mem_pfn set
  drm/qxl: remove the default io_mem_pfn set
  drm/nouveau: remove the default io_mem_pfn set
  drm/mgag200: remove the default io_mem_pfn set
  drm/cirrus: remove the default io_mem_pfn set
  drm/bochs: remove the default io_mem_pfn set
  drm/ast: remove the default io_mem_pfn set
  drm/ttm: add ttm_bo_io_mem_pfn to check io_mem_pfn
  drm/amdgpu: fix VM faults with per VM BOs
  drm/ttm: drop the spin in delayed delete if the trylock doesn't work
  ...

6 years agodrm/panel: lvds: Add support for the power-supply property
Maxime Ripard [Thu, 21 Dec 2017 11:02:28 +0000 (12:02 +0100)]
drm/panel: lvds: Add support for the power-supply property

A significant number of panels need to power up a regulator in order to
operate properly. Add support for the power-supply property to enable and
disable such a regulator whenever needed.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/0c0819bdf88fa948188df95e57a10820a8a4548d.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodt-bindings: panel: lvds: Document power-supply property
Maxime Ripard [Thu, 21 Dec 2017 11:02:27 +0000 (12:02 +0100)]
dt-bindings: panel: lvds: Document power-supply property

The power-supply property is used by a vast majority of panels, including
panel-simple. Let's document it as a common property

Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/0a6a3abcf1a6b7f0e66a81af8a44c5c0566ce06c.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agoMerge tag 'imx-drm-next-2018-01-02' of git://git.pengutronix.de/git/pza/linux into...
Dave Airlie [Fri, 5 Jan 2018 01:33:24 +0000 (11:33 +1000)]
Merge tag 'imx-drm-next-2018-01-02' of git://git.pengutronix.de/git/pza/linux into drm-next

drm/imx: format modifier support

- Add tiled prefetch support to PRE
- Add format modifier support to PRG and imx-drm-core
- Use runtime PM to control PRG clock
- Allow building ipu-v3 under COMPILE_TEST

* tag 'imx-drm-next-2018-01-02' of git://git.pengutronix.de/git/pza/linux:
  gpu: ipu-v3: allow to build with COMPILE_TEST
  drm/imx: advertise supported plane format modifiers
  drm/imx: add FB modifier support
  gpu: ipu-v3: prg: add modifier support
  gpu: ipu-v3: pre: add tiled prefetch support
  gpu: ipu-v3: prg: switch to runtime PM

6 years agoMerge branch 'drm-armada-devel-4.15' of git://git.armlinux.org.uk/~rmk/linux-arm...
Dave Airlie [Thu, 4 Jan 2018 23:43:46 +0000 (09:43 +1000)]
Merge branch 'drm-armada-devel-4.15' of git://git.armlinux.org.uk/~rmk/linux-arm into drm-next

This series builds upon the set of fixes previously submitted to move
Armada DRM closer to atomic modeset.  We're nowhere near yet, but this
series helps to get us closer by unifying some of the differences
between the primary and overlay planes.

New features added allows userspace to disable the primary plane if
overlay is full screen and there's nothing obscuring the colorkey -
this saves having to fetch an entire buffer containing nothing but
colorkey when displaying full screen video.

[airlied: fixup for atomic plane helper rename:
a01cb8ba3f6282934cff65e89ab36b18b14cbe27
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Wed Nov 1 22:16:19 2017 +0200

    drm: Move drm_plane_helper_check_state() into drm_atomic_helper.c
]

* 'drm-armada-devel-4.15' of git://git.armlinux.org.uk/~rmk/linux-arm: (29 commits)
  drm/armada: expand overlay trace entry
  drm/armada: implement primary plane update
  drm/armada: extract register generation from armada_drm_primary_set()
  drm/armada: wait for previous work when moving overlay window
  drm/armada: move overlay plane register update generation
  drm/armada: re-organise overlay register update generation
  drm/armada: disable planes at next blanking period
  drm/armada: avoid work allocation
  drm/armada: allow armada_drm_plane_work_queue() to silently fail
  drm/armada: use drm_plane_helper_check_state()
  drm/armada: only enable HSMOOTH if scaling horizontally
  drm/armada: move writes of LCD_SPU_SRAM_PARA1 under lock
  drm/armada: move regs into armada_plane_work
  drm/armada: move event sending into armada_plane_work
  drm/armada: move fb retirement into armada_plane_work
  drm/armada: move overlay plane work out from under spinlock
  drm/armada: clear plane enable bit when disabling
  drm/armada: clean up armada_drm_crtc_plane_disable()
  drm/armada: allow the primary plane to be disabled
  drm/armada: wait and cancel any pending frame work at disable
  ...

6 years agoMerge branch 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux into drm...
Dave Airlie [Thu, 4 Jan 2018 23:42:04 +0000 (09:42 +1000)]
Merge branch 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux into drm-next

Highlights this time:
1. Fix for a nasty Kconfig dependency chain issue from Philipp.
2. Occlusion query buffer address added to the cmdstream validator by
Christian.
3. Fixes and cleanups to the job handling from me. This allows us to
turn on the GPU performance profiling added in the last cycle.
It is also prep work for hooking in the DRM GPU scheduler, which I hope
to land for the next cycle.

* 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux: (32 commits)
  drm/etnaviv: use memset32 to init pagetable
  drm/etnaviv: move submit free out of critical section
  drm/etnaviv: re-enable perfmon support
  drm/etnaviv: couple runtime PM management to submit object lifetime
  drm/etnaviv: move GPU active handling to bo pin/unpin
  drm/etnaviv: move cmdbuf into submit object
  drm/etnaviv: use submit exec_state for perfmon sampling
  drm/etnaviv: move exec_state to submit object
  drm/etnaviv: move PMRs to submit object
  drm/etnaviv: refcount the submit object
  drm/etnaviv: move ww_acquire_ctx out of submit object
  drm/etnaviv: move object unpinning to submit cleanup
  drm/etnaviv: attach in fence to submit and move fence wait to fence_sync
  drm/etnaviv: rename submit fence to out_fence
  drm/etnaviv: move object fence attachment to gem_submit path
  drm/etnaviv: simplify submit_create
  drm/etnaviv: add lockdep annotations to buffer manipulation functions
  drm/etnaviv: hold GPU lock while inserting END command
  drm/etnaviv: move workqueue to be per GPU
  drm/etnaviv: remove switch_context member from etnaviv_gpu
  ...

6 years agoMerge tag 'exynos-drm-next-for-v4.16' of git://git.kernel.org/pub/scm/linux/kernel...
Dave Airlie [Thu, 4 Jan 2018 23:29:20 +0000 (09:29 +1000)]
Merge tag 'exynos-drm-next-for-v4.16' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-next

Remove lagacy IPP driver
- This driver isn't used anymore so remove it. Marek is preparing new one
  which includes completely rewritten API so this driver will be replaced
  with the new version[1] later.
And cleanups.

[1] https://patches.linaro.org/cover/118386/

* tag 'exynos-drm-next-for-v4.16' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm/exynos: ipp: Remove Exynos DRM IPP subsystem
  drm/exynos/decon: Add include guard to the Exynos7 header
  drm/exynos/decon: Move headers from global to local place
  drm/exynos: decon5433: Remove unnecessary platform_get_resource() error check

6 years agoMerge tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux...
Dave Airlie [Thu, 4 Jan 2018 23:27:49 +0000 (09:27 +1000)]
Merge tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux into drm-next

- Add CWSR (compute wave save restore) support for GFX8 (Carrizo)
- Fix SDMA user-mode queues support for GFX7 (Kaveri)
- Add SDMA user-mode queues support for GFX8 (Carrizo)
- Allow HWS (hardware scheduling) to schedule multiple processes concurrently
- Add debugfs support
- Simplify process locking and lock dependencies
- Refactoring topology code to prepare for dGPU support + fixes to that code
  - Add option to generate dummy/virtual CRAT table when its missing or deformed
  - Recognize CPUs other then APUs as compute entities
- Various clean ups and bug fixes

I have not yet sent the dGPU topology code because it depends on a patch
for the PCI subsystem that adds PCIe atomics support. Once that patch is
upstreamed we can continue with the rest of the dGPU code.

* tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux: (53 commits)
  drm/amdgpu: Add support for reporting VRAM usage
  drm/amdkfd: Ignore ACPI CRAT for non-APU systems
  drm/amdkfd: Module option to disable CRAT table
  drm/amdkfd: Add AQL Queue Memory flag on topology
  drm/amdkfd: Fixup incorrect info in the CZ CRAT table
  drm/amdkfd: Add perf counters to topology
  drm/amdkfd: Add topology support for dGPUs
  drm/amdkfd: Add topology support for CPUs
  drm/amdkfd: Fix sibling_map[] size
  drm/amdkfd: Simplify counting of memory banks
  drm/amdkfd: Turn verbose topology messages into pr_debug
  drm/amdkfd: sync IOLINK defines to thunk spec
  drm/amdkfd: Support enumerating non-GPU devices
  drm/amdkfd: Decouple CRAT parsing from device list update
  drm/amdkfd: Reorganize CRAT fetching from ACPI
  drm/amdkfd: Group up CRAT related functions
  drm/amdkfd: Fix memory leaks in kfd topology
  drm/amdkfd: Topology: Fix location_id
  drm/amdkfd: Update number of compute unit from KGD
  drm/amd: Remove get_vmem_size from KGD-KFD interface
  ...

6 years agodrm/sun4i: Add A83T support
Maxime Ripard [Thu, 21 Dec 2017 11:02:34 +0000 (12:02 +0100)]
drm/sun4i: Add A83T support

Add support for the A83T display pipeline.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/614b430adf3a67320362a75c01b01bd53013da8a.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodrm/sun4i: Add LVDS support
Maxime Ripard [Thu, 21 Dec 2017 11:02:33 +0000 (12:02 +0100)]
drm/sun4i: Add LVDS support

The TCON supports the LVDS interface to output to a panel or a bridge.
Let's add support for it.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/7fbb85f33ee1d5009fde4f0d7d236e11ca58b114.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodrm/sun4i: Create minimal multipliers and dividers
Maxime Ripard [Thu, 21 Dec 2017 11:02:32 +0000 (12:02 +0100)]
drm/sun4i: Create minimal multipliers and dividers

The various outputs the TCON can provide have different constraints on the
dotclock divider. Let's make them configurable by the various mode_set
functions.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/92ff5881c8f8674056d34458b2f264cd48d4e136.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodrm/sun4i: Force the mixer rate at 150MHz
Maxime Ripard [Thu, 21 Dec 2017 11:02:31 +0000 (12:02 +0100)]
drm/sun4i: Force the mixer rate at 150MHz

It seems like the mixer can only run properly when clocked at 150MHz. In
order to have something more robust than simply a fire-and-forget
assigned-clocks-rate, let's put that in the code.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/f5f05307972ed05250e8094b302d68b9e7e167f6.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodt-bindings: display: sun4i-drm: Add A83T pipeline
Maxime Ripard [Thu, 21 Dec 2017 11:02:30 +0000 (12:02 +0100)]
dt-bindings: display: sun4i-drm: Add A83T pipeline

The A83T has two video pipelines in parallel that looks quite similar to
the other SoCs.

The video planes are handled through a controller called the mixer, and the
video signal is then passed to the timing controller (TCON).

And while there is two instances of the mixers and TCONs, they have a
significant number of differences. The TCONs are quite easy to deal with,
one is supposed to generate TV (in the broader term, so including things
like HDMI) signals, the other one LCD (so RGB, LVDS, DSI) signals. And
while they are called TCON0 and TCON1 in the A83t datasheet, newer SoCs
call them TCON-TV and TCON-LCD, which seems more appropriate.

However, the mixers differ mostly by their capabilities, with some features
being available only in the first one, or the number of planes they expose,
but also through their register layout. And while the capabilities could be
represented as properties, the register layout differences would need to
express all the registers offsets as properties, which is usually quite
bad. Especially since documentation on that hardware block is close to
non-existent and we don't even have the list of all those registers in the
first place.

So let's call them mixer 0 and 1 in our compatibles, even though the name
is pretty bad...

At the moment, we only have tested the code on a board that has a single
display output, so we're leaving the tcon-tv and mixer1 out.

Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/2702a5c1d224af1c51743492ad1b917966f2ad43.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodt-bindings: display: sun4i-drm: Add LVDS properties
Maxime Ripard [Thu, 21 Dec 2017 11:02:29 +0000 (12:02 +0100)]
dt-bindings: display: sun4i-drm: Add LVDS properties

Some clocks and resets supposed to drive the LVDS logic in the display
engine have been overlooked when the driver was first introduced.

Add those additional resources to the binding, and we'll deal with the ABI
stability in the code.

Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ddbde28fe2e4f21412974e4c69fbfe1c5ff9383f.1513854122.git-series.maxime.ripard@free-electrons.com
6 years agodrm/tinydrm: add driver for ST7735R panels
David Lechner [Mon, 1 Jan 2018 19:02:16 +0000 (13:02 -0600)]
drm/tinydrm: add driver for ST7735R panels

This adds a new driver for Sitronix ST7735R display panels.

This has been tested using an Adafruit 1.8" TFT.

Signed-off-by: David Lechner <david@lechnology.com>
Reviewed-by: Noralf Trønnes <noralf@tronnes.org>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1514833336-22564-4-git-send-email-david@lechnology.com
6 years agodt-bindings: Add binding for Sitronix ST7735R display panels
David Lechner [Mon, 1 Jan 2018 19:02:15 +0000 (13:02 -0600)]
dt-bindings: Add binding for Sitronix ST7735R display panels

This adds a new device tree binding for Sitronix ST7735R display panels,
such as the Adafruit 1.8" TFT.

Signed-off-by: David Lechner <david@lechnology.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1514833336-22564-3-git-send-email-david@lechnology.com
6 years agodt-bindings: add jianda vendor prefix
David Lechner [Mon, 1 Jan 2018 19:02:14 +0000 (13:02 -0600)]
dt-bindings: add jianda vendor prefix

This adds a vendor prefix "jianda" for Jiandangjing Technology Co., Ltd.

Signed-off-by: David Lechner <david@lechnology.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1514833336-22564-2-git-send-email-david@lechnology.com
6 years agodrm/tinydrm: Update ILI9225 compatible string
David Lechner [Thu, 21 Dec 2017 18:33:07 +0000 (12:33 -0600)]
drm/tinydrm: Update ILI9225 compatible string

This updates the compatible string for a no-name LCD panel to
"vot,v220hf01a-t", "ilitek,ili9225".

The original bindings were the generic "ilitek,ili9225-2.2in-176x220"
because I could not find a datasheet. However, after some more research,
I finally found one, so the actual vendor and model name are now known.

This previous bindings have not made it to the mainline kernel yet, so
this is not breaking backwards compatibility.

Signed-off-by: David Lechner <david@lechnology.com>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1513881187-3197-4-git-send-email-david@lechnology.com
6 years agodt-bindings: update compatible string for ILI9225
David Lechner [Thu, 21 Dec 2017 18:33:06 +0000 (12:33 -0600)]
dt-bindings: update compatible string for ILI9225

This updates the compatible string for a no-name LCD panel to
"vot,v220hf01a-t", "ilitek,ili9225".

The original bindings [1] were the generic "ilitek,ili9225-2.2in-176x220"
because I could not find a datasheet. However, after some more research,
I finally found one, so the actual vendor and model name are now known.

This previous bindings have not made it to the mainline kernel yet, so
this is not breaking backwards compatibility.

This is also following the precedence of the ILI9322 bindings [2] by using
the pattern "vendor,specific-system-config", "vendor,ip-part";

[1]: https://patchwork.ozlabs.org/patch/839352/
[2]: https://patchwork.ozlabs.org/patch/843576/

Signed-off-by: David Lechner <david@lechnology.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1513881187-3197-3-git-send-email-david@lechnology.com
6 years agodt-bindings: Add "vot" vendor prefix
David Lechner [Thu, 21 Dec 2017 18:33:05 +0000 (12:33 -0600)]
dt-bindings: Add "vot" vendor prefix

This adds a vendor prefix "vot" for Vision Optical Technology Co., Ltd.
They make LCD displays.

Signed-off-by: David Lechner <david@lechnology.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1513881187-3197-2-git-send-email-david@lechnology.com
6 years agodrm/etnaviv: use memset32 to init pagetable
Lucas Stach [Mon, 27 Nov 2017 14:33:28 +0000 (15:33 +0100)]
drm/etnaviv: use memset32 to init pagetable

Now that memset32 is available, the open-coded pagetable initialization
loop can be replaced.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move submit free out of critical section
Lucas Stach [Wed, 29 Nov 2017 13:33:57 +0000 (14:33 +0100)]
drm/etnaviv: move submit free out of critical section

There is no need to hold the GPU lock while freeing the submit
object. Only move the retired submits from the GPU active list to
a temporary retire list under the GPU lock.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: re-enable perfmon support
Lucas Stach [Fri, 24 Nov 2017 11:13:25 +0000 (12:13 +0100)]
drm/etnaviv: re-enable perfmon support

Now that the PMR lifetime issues are solved we can safely re-enable
performance counter profiling support.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: couple runtime PM management to submit object lifetime
Lucas Stach [Fri, 24 Nov 2017 16:56:29 +0000 (17:56 +0100)]
drm/etnaviv: couple runtime PM management to submit object lifetime

As long as there is an active submit, we want the GPU to stay awake. This
is slightly complicated by the fact that we really want to wake the GPU
at the last possible moment to achieve maximum power savings.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move GPU active handling to bo pin/unpin
Lucas Stach [Mon, 27 Nov 2017 16:46:15 +0000 (17:46 +0100)]
drm/etnaviv: move GPU active handling to bo pin/unpin

The active count is used to check if the BO is idle, where idle is defined
as not active on the GPU and all VM mappings and reference counts dropped
to the initial state. As the idling of the mappings and references now only
happens in the submit cleanup, the active state handling must be moved to
the same location in order to keep the userspace semantics.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move cmdbuf into submit object
Lucas Stach [Fri, 24 Nov 2017 15:56:37 +0000 (16:56 +0100)]
drm/etnaviv: move cmdbuf into submit object

Less dynamic allocations and slims down the cmdbuf object to only the
required information, as everything else is already available in the
submit object.

This also simplifies buffer and mappings lifetime management, as they
are now exlusively attached to the submit object and not additionally
to the cmdbuf.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: use submit exec_state for perfmon sampling
Lucas Stach [Fri, 24 Nov 2017 14:19:16 +0000 (15:19 +0100)]
drm/etnaviv: use submit exec_state for perfmon sampling

The GPU exec state may have changed at the time when the perfmon sampling
is done, as it reflects the state of the last submission, not the current
GPU execution state.

So for proper sampling we must use the submit exec_state.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move exec_state to submit object
Lucas Stach [Fri, 24 Nov 2017 14:16:58 +0000 (15:16 +0100)]
drm/etnaviv: move exec_state to submit object

We'll need this in some places where only the submit is available. Also
this is a first step at slimming down the cmdbuf object.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move PMRs to submit object
Lucas Stach [Fri, 24 Nov 2017 11:02:38 +0000 (12:02 +0100)]
drm/etnaviv: move PMRs to submit object

To make them available to the event worker even after the actual
command stream execution has finished.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: refcount the submit object
Lucas Stach [Fri, 24 Nov 2017 10:36:03 +0000 (11:36 +0100)]
drm/etnaviv: refcount the submit object

The submit object lifetime will get extended to the actual GPU
execution. As multiple users will depend on this, add a kref to
properly control destruction of the object.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: move ww_acquire_ctx out of submit object
Lucas Stach [Fri, 24 Nov 2017 13:34:46 +0000 (14:34 +0100)]
drm/etnaviv: move ww_acquire_ctx out of submit object

The acquire_ctx is special in that it needs to be released from the same
thread as has been used to initialize it. This collides with the intention to
extend the submit lifetime beyond the gem_submit function with potentially
other threads doing the final cleanup.

Move the ww_acquire_ctx to the function local stack as suggested in the
documentation.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: move object unpinning to submit cleanup
Lucas Stach [Thu, 23 Nov 2017 17:13:40 +0000 (18:13 +0100)]
drm/etnaviv: move object unpinning to submit cleanup

This is safe to call in all paths, as the BO_PINNED flag tells us if the BO
needs unpinning.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: attach in fence to submit and move fence wait to fence_sync
Lucas Stach [Thu, 23 Nov 2017 17:02:43 +0000 (18:02 +0100)]
drm/etnaviv: attach in fence to submit and move fence wait to fence_sync

Simplifies the cleanup path and moves fence waiting to a central location.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: rename submit fence to out_fence
Lucas Stach [Thu, 23 Nov 2017 16:57:12 +0000 (17:57 +0100)]
drm/etnaviv: rename submit fence to out_fence

This is the fence passed out on a sucessful GPU submit. Make the name
more clear.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: move object fence attachment to gem_submit path
Lucas Stach [Thu, 23 Nov 2017 16:49:59 +0000 (17:49 +0100)]
drm/etnaviv: move object fence attachment to gem_submit path

The object fencing has nothing to do with the actual GPU buffer submit,
so move it to the gem submit path to have a cleaner split.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: simplify submit_create
Lucas Stach [Fri, 24 Nov 2017 09:43:07 +0000 (10:43 +0100)]
drm/etnaviv: simplify submit_create

Use kzalloc so other code doesn't need to worry about uninitialized members.
Drop the non-standard GFP flags, as we really don't want to fail the submit
when under slight memory pressure. Remove one level of indentation by using
an early return if the allocation failed. Also remove the unused drm device
member.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: add lockdep annotations to buffer manipulation functions
Lucas Stach [Fri, 17 Nov 2017 16:51:19 +0000 (17:51 +0100)]
drm/etnaviv: add lockdep annotations to buffer manipulation functions

When manipulating the kernel command buffer the GPU mutex must be held, as
otherwise different callers might try to replace the same part of the
buffer, wreacking havok in the GPU execution.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: hold GPU lock while inserting END command
Lucas Stach [Fri, 17 Nov 2017 16:59:26 +0000 (17:59 +0100)]
drm/etnaviv: hold GPU lock while inserting END command

Inserting the END command when suspending the GPU is changing the
command buffer state, which requires the GPU to be held.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/etnaviv: move workqueue to be per GPU
Lucas Stach [Fri, 17 Nov 2017 16:43:37 +0000 (17:43 +0100)]
drm/etnaviv: move workqueue to be per GPU

While the etnaviv workqueue needs to be ordered, as we rely on work items
being executed in queuing order, this is only true for a single GPU.
Having a shared workqueue for all GPUs in the system limits concurrency
artificially.

Getting each GPU its own ordered workqueue still meets our ordering
expectations and enables retire workers to run concurrently.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: remove switch_context member from etnaviv_gpu
Lucas Stach [Fri, 17 Nov 2017 16:19:50 +0000 (17:19 +0100)]
drm/etnaviv: remove switch_context member from etnaviv_gpu

There is no need to store this in the gpu struct. MMU flushes are triggered
correctly in reaction to MMU maps and unmaps, independent of the current ctx.
Any required pipe switches can be infered from the current and the desired
GPU exec state.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/etnaviv: don't flush workqueue in etnaviv_gpu_wait_obj_inactive
Lucas Stach [Fri, 17 Nov 2017 15:35:32 +0000 (16:35 +0100)]
drm/etnaviv: don't flush workqueue in etnaviv_gpu_wait_obj_inactive

There is no need to synchronize with oustanding retire jobs if the object
has gone idle. Retire jobs only ever change the object state from active to
idle, not the other way around.

The IOVA put race is uncritical, as the GEM_WAIT ioctl itself is holding
a reference to the GEM object, so the retire worker will not pull the
object into the CPU domain, which is the thing we are trying to guard
against with etnaviv_gpu_wait_obj_inactive. The ordering of the various
counts and waits may change a bit, but the userspace visible behavior at
the bounds of the syscall are unchanged.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: remove stale TODO in etnaviv_gpu_submit
Lucas Stach [Fri, 17 Nov 2017 14:09:05 +0000 (15:09 +0100)]
drm/etnaviv: remove stale TODO in etnaviv_gpu_submit

Flush and prefetch are properly handled in the buffer code, data endianess
would need much wider changes than adding something to this single function.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/etnaviv: remove -EAGAIN handling from submit path
Lucas Stach [Fri, 17 Nov 2017 13:24:34 +0000 (14:24 +0100)]
drm/etnaviv: remove -EAGAIN handling from submit path

Now that the userptr BO handling doesn't rely on the userspace restarting
the submit after object population, there is no need to special case the
-EAGAIN return value anymore.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: get rid of userptr worker
Lucas Stach [Fri, 17 Nov 2017 13:16:10 +0000 (14:16 +0100)]
drm/etnaviv: get rid of userptr worker

All code paths which populate userptr BOs are fine with the get_pages
function taking the mmap_sem lock. This allows to get rid of the pretty
involved architecture with a worker being scheduled if the mmap_sem
needs to be taken, but instead call GUP directly and allow it to take
the lock if necessary.

This simplifies the code a lot and removes the possibility of this
function returning -EAGAIN, which complicates object population
handling at the callers.

A notable change in behavior is that we don't allow a process to populate
objects with user pages from a foreign MM anymore. This would have been an
invalid use before, as it breaks the assumptions made in the etnaviv kernel
driver to enfore cache coherence. We now disallow this by rejecting the
request to populate those objects. Well behaving userspace is unaffected by
this change.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrm/etnaviv: change return type of etnaviv_gem_obj_add to void
Lucas Stach [Fri, 17 Nov 2017 11:17:14 +0000 (12:17 +0100)]
drm/etnaviv: change return type of etnaviv_gem_obj_add to void

This function never fails, as it does nothing more than adding the GEM
object to the global device list. Making this explicit through the void
return type allows to drop some unnecessary error handling.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: fold __etnaviv_gem_new into caller
Lucas Stach [Fri, 17 Nov 2017 11:11:25 +0000 (12:11 +0100)]
drm/etnaviv: fold __etnaviv_gem_new into caller

This function has only one caller and it isn't expected that there will
be any more in the future. Folding this function into the caller is
helping the readability.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/etnaviv: add lockdep annotation for userptr object population
Lucas Stach [Fri, 17 Nov 2017 10:23:17 +0000 (11:23 +0100)]
drm/etnaviv: add lockdep annotation for userptr object population

The current userptr page population will defer work to a work item if
needed to avoid ever taking the mmap_sem in the direct call path. With
the more fine-grained locking in etnaviv this isn't needed anymore, so
a future commit will simplify this code.

Add a lockdep annotation to validate the assumption that the mmap_sem
can be taken in the direct call path.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/etnaviv: split obj locks in different classes depending on the obj type
Lucas Stach [Fri, 17 Nov 2017 10:14:58 +0000 (11:14 +0100)]
drm/etnaviv: split obj locks in different classes depending on the obj type

Userptr, prime and shmem buffer objects have different lock ordering
requirements. This is mostly due to the fact that we don't allow to mmap
userptr buffers, so we won't ever end up in our fault handler for those,
so some of the code paths are never called with the mmap_sem held.

To avoid lockdep false positives, split them up into different lock classes.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
6 years agodrm/etnaviv: fix GPU vs sync point race
Lucas Stach [Thu, 19 Oct 2017 11:48:40 +0000 (13:48 +0200)]
drm/etnaviv: fix GPU vs sync point race

If the FE is restarted before the sync point event is cleared, the GPU
might trigger a completion IRQ for the next sync point, corrupting
the state of the currently running worker.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agodrm/exynos: ipp: Remove Exynos DRM IPP subsystem
Marek Szyprowski [Thu, 14 Dec 2017 15:10:15 +0000 (16:10 +0100)]
drm/exynos: ipp: Remove Exynos DRM IPP subsystem

Exynos DRM IPP subsystem is in fact non-functional and frankly speaking
dead-code. This patch clearly marks that Exynos DRM IPP subsystem is
broken and never really functional. It will be replaced by a completely
rewritten API.

Exynos DRM IPP user-space API can be obsoleted for the following
reasons:

1. Exynos DRM IPP user-space API can be optional in Exynos DRM, so
userspace should not rely that it is always available and should have
a software fallback in case it is not there.

2. The only mode which was initially semi-working was memory-to-memory
image processing. The remaining modes (LCD-"writeback" and "output")
were never operational due to missing code (both in mainline and even
vendor kernels).

3. Exynos DRM IPP mainline user-space API compatibility for
memory-to-memory got broken very early by commit 083500baefd5 ("drm:
remove DRM_FORMAT_NV12MT", which removed the support for tiled formats,
the main feature which made this API somehow useful on Exynos platforms
(video codec that time produced only tiled frames, to implement xvideo
or any other video overlay, one has to de-tile them for proper
display).

4. Broken drivers. Especially once support for IOMMU has been added,
it revealed that drivers don't configure DMA operations properly and in
many cases operate outside the provided buffers trashing memory around.

5. Need for external patches. Although IPP user-space API has been used
in some vendor kernels, but in such cases there were additional patches
applied (like reverting mentioned 083500baefd5 patch) what means that
those userspace apps which might use it, still won't work with the
mainline kernel version.

We don't have time machines, so we cannot change it, but Exynos DRM IPP
extension should never have been merged to mainline in that form.

Exynos IPP subsystem and user-space API will be rewritten, so remove
current IPP core code and mark existing drivers as BROKEN.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
6 years agodrm/exynos/decon: Add include guard to the Exynos7 header
Krzysztof Kozlowski [Tue, 26 Dec 2017 13:07:19 +0000 (14:07 +0100)]
drm/exynos/decon: Add include guard to the Exynos7 header

Although header is included only once but still having an include guard
is a good practice.  To avoid confusion, add SoC prefix to existing
Exynos5433 header include guard.

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
6 years agodrm/exynos/decon: Move headers from global to local place
Krzysztof Kozlowski [Tue, 26 Dec 2017 13:07:18 +0000 (14:07 +0100)]
drm/exynos/decon: Move headers from global to local place

The DECON headers contain only defines for registers.  There are no
other drivers using them so this should be put locally to the Exynos DRM
driver.  Keeping headers local helps managing the code.

Suggested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
6 years agodrm/exynos: decon5433: Remove unnecessary platform_get_resource() error check
Fabio Estevam [Mon, 18 Dec 2017 13:02:25 +0000 (11:02 -0200)]
drm/exynos: decon5433: Remove unnecessary platform_get_resource() error check

devm_ioremap_resource() already checks if the resource is NULL, so
remove the unnecessary platform_get_resource() error check.

Cc: Inki Dae <inki.dae@samsung.com>
Signed-off-by: Fabio Estevam <fabio.estevam@nxp.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
6 years agodrm/amdgpu: Correct the IB size of bo update mapping.
Emily Deng [Fri, 29 Dec 2017 05:13:08 +0000 (13:13 +0800)]
drm/amdgpu: Correct the IB size of bo update mapping.

The amdgpu_vm_frag_ptes will call amdgpu_vm_update_ptes, and for buffer
object that has shadow buffer, need twice commands.

Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: enable swapout for reserved BOs during allocation
Roger He [Thu, 21 Dec 2017 09:42:53 +0000 (17:42 +0800)]
drm/ttm: enable swapout for reserved BOs during allocation

if the bo shares same reservation object then not lock it again
at swapout time to make it possible to swap out.

v2: refine the commmit message

Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chuming Zhou <david1.zhou@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: add new function to check if bo is allowable to evict or swapout
Roger He [Thu, 21 Dec 2017 09:42:52 +0000 (17:42 +0800)]
drm/ttm: add new function to check if bo is allowable to evict or swapout

extract a function as ttm_bo_evict_swapout_allowable since eviction and
swapout can share same logic.

v2: modify commit message and add description in the code

Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chuming Zhou <david1.zhou@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: use an operation ctx for ttm_tt_bind
Roger He [Thu, 21 Dec 2017 09:42:51 +0000 (17:42 +0800)]
drm/ttm: use an operation ctx for ttm_tt_bind

forward the operation context to ttm_tt_bind as well,
and the ultimate goal is swapout enablement for reserved BOs.

v2: use common term rather than amd specific

Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chuming Zhou <david1.zhou@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: use an operation ctx for ttm_tt_populate in ttm_bo_driver (v2)
Roger He [Thu, 21 Dec 2017 09:42:50 +0000 (17:42 +0800)]
drm/ttm: use an operation ctx for ttm_tt_populate in ttm_bo_driver (v2)

forward the operation context to ttm_tt_populate as well,
and the ultimate goal is swapout enablement for reserved BOs.

v2: squash in fix for vboxvideo

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agoMerge tag 'drm-intel-next-2017-12-22' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Wed, 27 Dec 2017 19:20:31 +0000 (05:20 +1000)]
Merge tag 'drm-intel-next-2017-12-22' of git://anongit.freedesktop.org/drm/drm-intel into drm-next

- Allow internal page allocation to fail (Chris)
- More improvements on logs, dumps, and trace (Chris, Michal)
- Coffee Lake important fix for stolen memory (Lucas)
- Continue to make GPU reset more robust as well
   improving selftest coverage for it (Chris)
- Unifying debugfs return codes (Michal)
- Using existing helper for testing obj pages (Matthew)
- Organize and improve gem_request tracepoints (Lionel)
- Protect DDI port to DPLL map from theoretical race (Rodrigo)
- ... and consequently fixing the indentation on this DDI clk selection function (Chris)
- ... and consequently properly serializing non-blocking modesets (Ville)
- Add support for horizontal plane flipping on Cannonlake (Joonas)
- Two Cannonlake Workarounds for better stability (Rafael)
- Fix mess around PSR registers (DK)
- More Coffee Lake PCI IDs (Rodrigo)
- Remove CSS modifiers on pipe C of Geminilake (Krisman)
- Disable all planes for load detection (Ville)
- Reorg on i915 display headers (Michal)
- Avoid enabling movntdqa optimization on hypervisor guest (Changbin)

GVT:
- more mmio switch optimization (Weinan)
- cleanup i915_reg_t vs. offset usage (Zhenyu)
- move write protect handler out of mmio handler (Zhenyu)

* tag 'drm-intel-next-2017-12-22' of git://anongit.freedesktop.org/drm/drm-intel: (55 commits)
  drm/i915: Update DRIVER_DATE to 20171222
  drm/i915: Show HWSP in intel_engine_dump()
  drm/i915: Assert that the request is on the execution queue before being removed
  drm/i915/execlists: Show preemption progress in GEM_TRACE
  drm/i915: Put all non-blocking modesets onto an ordered wq
  drm/i915: Disable GMBUS clock gating around GMBUS transfers on gen9+
  drm/i915: Clean up the PNV bit banging vs. GMBUS clock gating w/a
  drm/i915: No need to power up PG2 for GMBUS on BXT
  drm/i915: Disable DC states around GMBUS on GLK
  drm/i915: Do not enable movntdqa optimization in hypervisor guest
  drm/i915: Dump device info at once
  drm/i915: Add pretty printer for runtime part of intel_device_info
  drm/i915: Update intel_device_info_runtime_init() parameter
  drm/i915: Move intel_device_info definitions to its own header
  drm/i915: Move opregion definitions to dedicated intel_opregion.h
  drm/i915: Move display related definitions to dedicated header
  drm/i915: Move some utility functions to i915_util.h
  drm/i915/gvt: move write protect handler out of mmio emulation function
  drm/i915/gvt: cleanup usage for typed mmio reg vs. offset
  drm/i915/gvt: Fix pipe A enable as default for vgpu
  ...

6 years agodrm/ttm: use an operation ctx for ttm_mem_global_alloc_page
Roger He [Fri, 8 Dec 2017 07:21:18 +0000 (15:21 +0800)]
drm/ttm: use an operation ctx for ttm_mem_global_alloc_page

forward the operation context to ttm_mem_global_alloc_page as well,
and the ultimate goal is swapout enablement for reserved BOs.

Here reserved BOs refer to all the BOs which share same reservation object

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: use an operation ctx for ttm_mem_global_alloc
Roger He [Fri, 8 Dec 2017 07:09:50 +0000 (15:09 +0800)]
drm/ttm: use an operation ctx for ttm_mem_global_alloc

forward the operation context to ttm_mem_global_alloc as well, and the
ultimate goal is swapout enablement for reserved BOs

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: call ttm_bo_swapout directly when ttm shrink
Roger He [Mon, 18 Dec 2017 11:50:08 +0000 (19:50 +0800)]
drm/ttm: call ttm_bo_swapout directly when ttm shrink

remove the extra indirection because we have only one implementation anyway

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/vmwgfx: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:32 +0000 (11:43 +0800)]
drm/vmwgfx: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/virtio: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:31 +0000 (11:43 +0800)]
drm/virtio: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/radeon: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:30 +0000 (11:43 +0800)]
drm/radeon: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/qxl: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:29 +0000 (11:43 +0800)]
drm/qxl: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/nouveau: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:28 +0000 (11:43 +0800)]
drm/nouveau: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/mgag200: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:27 +0000 (11:43 +0800)]
drm/mgag200: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/cirrus: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:26 +0000 (11:43 +0800)]
drm/cirrus: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/bochs: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:25 +0000 (11:43 +0800)]
drm/bochs: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ast: remove the default io_mem_pfn set
Tan Xiaojun [Mon, 25 Dec 2017 03:43:24 +0000 (11:43 +0800)]
drm/ast: remove the default io_mem_pfn set

The default interface situation has been taken into the framework, so
remove the default set of each module.

Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: add ttm_bo_io_mem_pfn to check io_mem_pfn
Tan Xiaojun [Mon, 25 Dec 2017 03:43:23 +0000 (11:43 +0800)]
drm/ttm: add ttm_bo_io_mem_pfn to check io_mem_pfn

The io_mem_pfn field was added in commit ea642c3216cb ("drm/ttm: add
io_mem_pfn callback") and is called unconditionally. However, not all
drivers were updated to set it.

Use the ttm_bo_default_io_mem_pfn function if a driver did not set its
own. And add new function ttm_bo_io_mem_pfn() as wrapper.

Signed-off-by: Michal Srb <msrb@suse.com>
Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: fix VM faults with per VM BOs
Christian König [Fri, 22 Dec 2017 16:13:03 +0000 (17:13 +0100)]
drm/amdgpu: fix VM faults with per VM BOs

There was a small window between unreserve and second reserve where the
freshly allocated BO could have been evicted without the VM noticing it.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/ttm: drop the spin in delayed delete if the trylock doesn't work
Christian König [Thu, 21 Dec 2017 18:04:15 +0000 (19:04 +0100)]
drm/ttm: drop the spin in delayed delete if the trylock doesn't work

Thomas actually noticed that, but I didn't realized what he meant until
now.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amd/pp: Get and save CZ/ST smu version
Rex Zhu [Thu, 21 Dec 2017 10:38:47 +0000 (18:38 +0800)]
drm/amd/pp: Get and save CZ/ST smu version

The smu firmware is loaded by the sbios on APUs, so query it
from the smu and save the smu fw version info that is reported
to userspace.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: drop client_id from VM
Christian König [Mon, 18 Dec 2017 16:10:01 +0000 (17:10 +0100)]
drm/amdgpu: drop client_id from VM

Use the fence context from the scheduler entity.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: rename vm_id to vmid
Christian König [Mon, 18 Dec 2017 16:08:25 +0000 (17:08 +0100)]
drm/amdgpu: rename vm_id to vmid

sed -i "s/vm_id/vmid/g" drivers/gpu/drm/amd/amdgpu/*.c
sed -i "s/vm_id/vmid/g" drivers/gpu/drm/amd/amdgpu/*.h

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
6 years agodrm/amdgpu: separate VMID and PASID handling
Christian König [Mon, 18 Dec 2017 15:53:03 +0000 (16:53 +0100)]
drm/amdgpu: separate VMID and PASID handling

Move both into the new files amdgpu_ids.[ch]. No functional change.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>