drm/amd/powerplay: add parts of system clock gating support for Fiji. (v2)

Removed fiji_mgcg_cgcg_init that is affected and redundant for new implementation.

v2: re-add mgcg_cgcg init

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amdgpu: add sdma clock gating support for Fiji.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/amdgpu: add gmc clock gating support for Fiji.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/amdgpu: add gfx clock gating support for Fiji.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay/tonga: Add UVD DPM init

Load the UVD DPM state into the SMC.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: fix boolreturn.cocci warnings

drivers/gpu/drm/amd/amdgpu/../powerplay/hwmgr/tonga_hwmgr.c:2653:9-10: WARNING: return of 0/1 in function 'is_pcie_gen2_supported' with return type bool
drivers/gpu/drm/amd/amdgpu/../powerplay/hwmgr/tonga_hwmgr.c:2645:9-10: WARNING: return of 0/1 in function 'is_pcie_gen3_supported' with return type bool

 Return statements in functions returning bool should use
 true/false instead of 1/0.
Generated by: scripts/coccinelle/misc/boolreturn.cocci

CC: yanyang1 <young.yang@amd.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay/fiji: query supported pcie info from cgs (v2)

Rather than hardcode it.

v2: integrate spc fix from Rex

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay/tonga: query supported pcie info from cgs (v2)

Rather than hardcode it.

v2: integrate spc fix from Rex

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/cgs: add sys info query for pcie gen and link width

Needed by powerplay to properly handle pcie dpm switching.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: store pcie gen mask and link width

We'll need this later for pcie dpm.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm: add drm_pcie_get_max_link_width helper (v2)

Add a helper to get the max link width of the port.
Similar to the helper to get the max link speed.

v2: fix typo in commit message

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: extract pcie helpers to common header

These will be used by multiple powerplay drivers and
other IP modules.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay/fiji: enable pcie and mclk forcing for low

When forcing the lowest state also force mclk and pcie.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay/tonga: enable pcie and mclk forcing for low

When forcing the lowest state also force mclk and pcie.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: refine the logic of whether need to update power state.

Better handle power state changes.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amd/powerplay: implement new funcs to check current states for tonga.

Implement the new callbacks for tonga.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amd/powerplay: add and export hwmgr interface to eventmgr to check hw states.

Interface between hwmgr and eventmgr.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu: support per device powerplay enablement (v2)

The amdgu_powerplay variable is global for multiple GPU instances.

v2: fold in Flora's module option change, protect adev reference in
macros

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

drm/amdgpu: enable sysfs interface for powerplay

Same interface exposed in pre-powerplay dpm code.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: export fan control functions to amdgpu

Hook up the amdgpu thermal control callbacks for powerplay.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: implement fan control interface in amd_powerplay_funcs

This adds the interface needed to expose powerplay fan control to sysfs
via hwmon.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: implement thermal control for tonga.

Implement thermal and fan control for tonga.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: enable thermal interrupt task in eventmgr.

Add thermal handling to the event manager.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: add thermal control interface in hwmgr.

Thermal controller interface.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: mv ppinterrupt.h to inc folder to share with other submodule.

Redefine interrupt callback function in accordance with cgs.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: add new function point in hwmgr_funcs for thermal control

Add the interface for fan and thermal control.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Add CG and PG support for tonga

Implement clock and power gating support for tonga.  On Tonga
this is handles by the SMU rather than direct register settings
in the driver.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add new function point in hwmgr_func for CG/PG.

Add callbacks interface for clock and powergating.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: add some definition for other ip block to update cg pg.

Interface for clock and power gating handling.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable powerplay module by default for fiji.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable powerplay module by default for tonga.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: program display gap for tonga.

Implement displaygap programming for tonga.  This is
required for properly mclk switching.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: implement pem_task for display_configuration_change

Add support for display configuration changes to the event manager.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/poweprlay: export program display gap function to eventmgr

This allows the eventmgr to properly update the displaygap on
certain power events.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/powerplay: add function point in hwmgr_funcs for program display gap

Displaygap support is required for proper mclk switching.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: enable powerplay and smc firmware loading for Fiji.

Switch over to handling in the powerplay module.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amdgpu: add amdgpu.powerplay module option

This option can be used to enable the new powerplay implementation,
and it is disabled by default.

Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add Fiji DPM support.

This enabled DPM support for Fiji.  DPM is dynamic
clock and voltage scaling.

v2: rename fiji_hwmgr_early_init to fiji_hwmgr_init
v3: (agd) fold in endian fix, additional function addition

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay: add Fiji SMU support.

Add support for the SMU manager for Fiji. This handles the
firmware loading for other IP blocks (GFX, SDMA, etc.).

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay: update atomctrl for fiji

Add some new functions to support Fiji.  Split out
from the previous patch.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay: add/update headers for Fiji SMU and DPM

New headers for Fiji.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay: add Tonga dpm support (v3)

This implements DPM for tonga.  DPM handles dynamic
clock and voltage scaling.

v2: merge all the patches related with tonga dpm
v3: merge dpm force level fix, cgs display fix, spelling fix

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: yanyang1 <young.yang@amd.com>
Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drm/amd/powerplay: Add Tonga SMU support

The SMU manager handles firmware loading for other IP
blocks (GFX, SDMA, etc.).  This implements it for Tonga.

v3: delete peci sub-module
v2: use cgs interface directly

Signed-off-by: Young Yang <Young.Yang@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add header file for tonga smu and dpm

These headers provide the SMU interface used by the driver.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: yanyang1 <young.yang@amd.com>

drm/amd/powerplay: Move smu7*.h from amdgpu to powerplay.

Move smu7.h, smu7_discrete.h and smu7_fusion.h from amdgpu to powerplay.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: yanyang1 <young.yang@amd.com>

drm/amd/powerplay: Add ixSWRST_COMMAND_1 in bif_5_0_d.h

Add ixSWRST_COMMAND_1 in bif_5_0_d.h.  Required by
new powerplay code for tonga and fiji.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: yanyang1 <young.yang@amd.com>

drm/amd/powerplay: implement functions of amd_powerplay_func

This is the common interface for interacting with the powerplay
module.

v2: squash in fixes

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Acked-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add event manager sub-component

The event manager handles power related driver events.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add CG and PG support for carrizo

This adds clock and powergating support for CZ.

v2: squash in fixes

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add Carrizo dpm support

This patch enables basic DPM support for Carrizo.
DPM handles dynamic clock and voltage scaling.

v3: delete peci sub-module
v2: use cgs interface directly
    correct define SMU_EnabledFeatureScoreboard_SclkDpmOn

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add Carrizo smu support

This implements the SMU firmware manager interface for CZ.
Some header files are moved from amdgpu folder to powerplay as well.

v3: delete peci sub-module.
v2: use cgs interface directly
    add load_mec_firmware function

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add hardware manager sub-component

The hwmgr handles all hardware related calls, including clock/power
gating control, DPM, read and parse PPTable, etc.

v5: squash in fixes
v4: implement acpi's atcs function use cgs interface
v3: fix code style error and add big-endian mode support.
v2: use cgs interface directly in hwmgr sub-module

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add SMU manager sub-component

The SMUMGR is one sub-component of powerplay for SMU firmware support.
The SMU handles firmware loading for other IP blocks (GFX, SDMA, etc.)
on VI parts.  The adds the core powerplay infrastructure to handle that.

v3: direct use printk in powerplay module.
v2: direct use cgs_read/write_register functions in smu-modules

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: export amd_powerplay_func to amdgpu and other ip block

Update amdgpu to deal with the new powerplay module properly.

v2: squash in fixes
v3: squash in Rex's power state reporting fix

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Acked-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: disable legacy path of firmware check if powerplay is enabled

Powerplay will use a different interface once it's integrated.  These
legacy pathes will be removed once powerplay is enabled by default.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add basic powerplay framework

amdgpu_pp_ip_funcs is introduced to handle the two code paths,
the legacy one and the new powerplay implementation.

CONFIG_DRM_AMD_POWERPLAY kernel configuration option is
introduced for the powerplay component.

v4: squash in fixes
v3: register debugfs file when powerplay module enable
v2: add amdgpu_ucode_init_bo in hw init when amdgpu_powerplay enable.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add new cgs interface to get display info (v2)

Add new CGS interfaces to query display info across modules.
This is nedded by the powerplay module for synchronizing with
the display module.

v2: (agd): fold in refresh rate fix, rebase

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: implement cgs interface to query system info

Add a query to get the bus number and function of the
device.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu: implement new cgs interface for acpi function

Add a new driver internal interface for accessing ACPI
methods.  These will be used by various new components
including powerplay.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: mv amdgpu_acpi.h to amd/include/amd_acpi.h

This will be shared with the new powerplay module.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: mv some definition from amdgpu_acpi.c to amdgpu_acpi.h

These will be shared with the new powerplay module.

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: share struct amdgpu_pm_state_type with powerplay module

rename amdgpu_pm_state_type to amd_pm_state_type

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Acked-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: clean up fujitsu quirks

Combine the two quirks.

bug:
https://bugzilla.kernel.org/show_bug.cgi?id=109481

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

drm/amdgpu: Fix off-by-one errors in amdgpu_vm_bo_map

eaddr is sometimes treated as the last address inside the address
range, and sometimes as the first address outside the range. This
was resulting in errors when a test filled up the entire address
space. Make it consistent to always be the last address within the
range.

Signed-off-by: Felix.Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org

drm/radeon: Fix off-by-one errors in radeon_vm_bo_set_addr

eoffset is sometimes treated as the last address inside the address
range, and sometimes as the first address outside the range. This
was resulting in errors when a test filled up the entire address
space. Make it consistent to always be the last address within the
range. Also fixed related errors when checking the VA limit and in
radeon_vm_fence_pts.

Signed-off-by: Felix.Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org

drm/radeon: fix dp link rate selection (v2)

Need to properly handle the max link rate in the dpcd.
This prevents some cases where 5.4 Ghz is selected when
it shouldn't be.

v2: simplify logic, add array bounds check

Reviewed-by: Tom St Denis <tom.stdenis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix dp link rate selection (v2)

Need to properly handle the max link rate in the dpcd.
This prevents some cases where 5.4 Ghz is selected when
it shouldn't be.

v2: simplify logic, add array bounds check

Reviewed-by: Tom St Denis <tom.stdenis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: only increment sync_seq when a fence is really emitted

In the rare situation where the kmalloc fails we're probably screwed anyway,
but let's try to be more robust about it.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Nicolai Hähnle <Nicolai.Haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: fix typo in cik_ring_ib_execute documentation (v2)

v2: agd: clarify commit message, fix "an" as spotted by Michel.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/ttm: fix documentation of ttm_bo_reserve

Previously, the comment was inconsistent. EDEADLK is what the ww_mutex
mechanism really returns.

Signed-off-by: Nicolai Hähnle <Nicolai.Haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Update radeon_get_vblank_counter_kms()

Commit 88e72717c2de ("drm/irq: Use unsigned int pipe in public API")
updated the prototype of this function but not the implementation. This
wasn't noticed even through compile tests because the prototype is part
of the source file that uses it and hence the compiler won't know the
prototype when it compiles the implementation.

The right thing would've been to move the prototype to a header that's
included in radeon_kms.c so that the implementation signature could be
checked against it, but the closest thing would've been radeon_drv.h
and including that results in a lot of build errors, so we'll leave it
as is for now.

Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2)

commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many
vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core
more fragile to drivers which don't update hw vblank counters and
vblank timestamps in sync with firing of the vblank irq and
essentially at leading edge of vblank.

This exposed a problem with radeon-kms/amdgpu-kms which do not
satisfy above requirements:

The vblank irq fires a few scanlines before start of vblank, but
programmed pageflips complete at start of vblank and
vblank timestamps update at start of vblank, whereas the
hw vblank counter increments only later, at start of vsync.

This leads to problems like off by one errors for vblank counter
updates, vblank counters apparently going backwards or vblank
timestamps apparently having time going backwards. The net result
is stuttering of graphics in games, or little hangs, as well as
total failure of timing sensitive applications.

See bug #93147 for an example of the regression on Linux 4.4-rc:

https://bugs.freedesktop.org/show_bug.cgi?id=93147

This patch tries to align all above events better from the
viewpoint of the drm core / of external callers to fix the problem:

1. The apparent start of vblank is shifted a few scanlines earlier,
so the vblank irq now always happens after start of this extended
vblank interval and thereby drm_update_vblank_count() always samples
the updated vblank count and timestamp of the new vblank interval.

To achieve this, the reporting of scanout positions by
radeon_get_crtc_scanoutpos() now operates as if the vblank starts
radeon_crtc->lb_vblank_lead_lines before the real start of the hw
vblank interval. This means that the vblank timestamps which are based
on these scanout positions will now update at this earlier start of
vblank.

2. The driver->get_vblank_counter() function will bump the returned
vblank count as read from the hw by +1 if the query happens after
the shifted earlier start of the vblank, but before the real hw increment
at start of vsync, so the counter appears to increment at start of vblank
in sync with the timestamp update.

3. Calls from vblank irq-context and regular non-irq calls are now
treated identical, always simulating the shifted vblank start, to
avoid inconsistent results for queries happening from vblank irq vs.
happening from drm_vblank_enable() or vblank_disable_fn().

4. The radeon_flip_work_func will delay mmio programming a pageflip until
the start of the real vblank iff it happens to execute inside the shifted
earlier start of the vblank, so pageflips now also appear to execute at
start of the shifted vblank, in sync with vblank counter and timestamp
updates. This to avoid some races between updates of vblank count and
timestamps that are used for swap scheduling and pageflip execution which
could cause pageflips to execute before the scheduled target vblank.

The lb_vblank_lead_lines "fudge" value is calculated as the size of
the display controllers line buffer in scanlines for the given video
mode: Vblank irq's are triggered by the line buffer logic when the line
buffer refill for a video frame ends, ie. when the line buffer source read
position enters the hw vblank. This means that a vblank irq could fire at
most as many scanlines before the current reported scanout position of the
crtc timing generator as the number of scanlines the line buffer can
maximally hold for a given video mode.

This patch has been successfully tested on a RV730 card with DCE-3 display
engine and on a evergreen card with DCE-4 display engine, in single-display
and dual-display configuration, with different video modes.

A similar patch is needed for amdgpu-kms to fix the same problem.

Limitations:

- Line buffer sizes in pixels are hard-coded on < DCE-4 to a value
  i just guessed to be high enough to work ok, lacking info on the true
  sizes atm.

Fixes: fdo#93147

Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Michel Dänzer <michel.daenzer@amd.com>
Cc: Harry Wentland <Harry.Wentland@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
(v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net>

(v2) Refine radeon_flip_work_func() for better efficiency:

     In radeon_flip_work_func, replace the busy waiting udelay(5)
     with event lock held by a more performance and energy efficient
     usleep_range() until at least predicted true start of hw vblank,
     with some slack for scheduler happiness. Release the event lock
     during waits to not delay other outputs in doing their stuff, as
     the waiting can last up to 200 usecs in some cases.

     Retested on DCE-3 and DCE-4 to verify it still works nicely.

(v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Fix "slow" audio over DP on DCE8+

DP audio is derived from the dfs clock.

Signed-off-by: Slava Grigorev <slava.grigorev@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

drm/amdgpu: keep the PTs validation list in the VM v2

This avoids allocating it on the fly.

v2: fix grammar in comment

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

drm/amdgpu: split VM PD and PT handling during CS

This way we avoid the extra allocation for the page directory entry.

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

drm/amdgpu: put VM page tables directly into duplicates list

They share the reservation object with the page directory anyway.

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

drm/amdgpu: restrict the sched jobs number to power of two

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
CC: stable@vger.kernel.org

drm/amdgpu: limit visible vram if it's smaller than the BAR

In some cases the amount of vram may be less than the BAR size,
if so, limit visible vram to the amount of actual vram, not the
BAR size.

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

drm/amdgpu: change default sched jobs to 32

Change the default scheduler queue size from 16 to 32.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: unify AMDGPU_CTX_MAX_CS_PENDING and amdgpu_sched_jobs

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: handle error case for ctx

Properly handle ctx init failure.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add entity only when first job come

umd somtimes will create a context for every ring,
that means some entities wouldn't be used at all.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add more debugging output for driver failures

Add more fine grained debugging output for init/fini/suspend/
resume failures.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: update rev id register for VI

Change-Id: I2ae9bb4a929f7c0c8783e0be563ae04be77596e2
Signed-off-by: Flora Cui <Flora.Cui@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx8: update PA_SC_RASTER_CONFIG:PKR_MAP only

Use default value as a base.

Signed-off-by: Flora Cui <Flora.Cui@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu/gfx8: Enable interrupt on ME1_PIPE3

Otherwise FW cannot see the RLC ACK for the memory clean request
It's for Stoney.

Signed-off-by: Flora Cui <Flora.Cui@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu: add spin lock to protect freed list in vm (v2)

there is a protection fault about freed list when OCL test.
add a spin lock to protect it.

v2: drop changes in vm_fini

Signed-off-by: JimQu <jim.qu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

amdgpu/gfxv8: Remove magic numbers from function gfx_v8_0_tiling_mode_table_init()

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix race condition in amd_sched_entity_push_job

As soon as we leave the spinlock after the job has been added to the job
queue, we can no longer rely on the job's data to be available.

I have seen a null-pointer dereference due to sched == NULL in
amd_sched_wakeup via amd_sched_entity_push_job and
amd_sched_ib_submit_kernel_helper. Since the latter initializes
sched_job->sched with the address of the ring scheduler, which is
guaranteed to be non-NULL, this race appears to be a likely culprit.

Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Bugzilla: https://bugs.freedesktop.org/attachment.cgi?bugid=93079
Reviewed-by: Christian König <christian.koenig@amd.com>

drm/amdgpu: add err check for pin userptr

Missing error check if the operation failed.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

amdgpu/gfxv8: Simplification in gfx_v8_0_enable_gui_idle_interrupt()

Simplified the function by folding the two paths into one.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

amdgpu/gfxv8: Simplification of gfx_v8_0_create_bitmask()

Simplification of the function gfx_v8_0_create_bitmask().

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

amdgpu/gfxv8: Cleanup of gfx_v8_0_tiling_mode_table_init() (v2)

Simplification and LOC reduction of function gfx_v8_0_tiling_mode_table_init()

v2: remove spurious break
bug: https://bugs.freedesktop.org/show_bug.cgi?id=93236

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

amdgpu/gfxv8: Add missing break to switch statement from states init code

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

drm/amd: abstract kernel rq and normal rq to priority of run queue

Allows us to set priorities in the scheduler.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>

drm/amdgpu: add EDC support for CZ (v3)

This adds EDC support for CZ.
EDC = Error Correction and Detection
This code properly initializes the EDC hardware and
resets the error counts.  This is done in late_init
since it requires the IB pool which is not initialized
during hw_init.

v2: fix the IB size as noted by Felix, fix shader pgm
register programming
v3: use the IB for the shaders as suggested by Christian

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

drm/amd: add new gfx8 register definitions for EDC

EDC is a RAS feature for on chip memory.

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

drm/radeon: Use unlocked gem unreferencing

For drm_gem_object_unreference callers are required to hold
dev->struct_mutex, which these paths don't. Enforcing this requirement
has become a bit more strict with

commit ef4c6270bf2867e2f8032e9614d1a8cfc6c71663
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Oct 15 09:36:25 2015 +0200

    drm/gem: Check locking in drm_gem_object_unreference

Cc: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Use unlocked gem unreferencing

For drm_gem_object_unreference callers are required to hold
dev->struct_mutex, which these paths don't. Enforcing this requirement
has become a bit more strict with

commit ef4c6270bf2867e2f8032e9614d1a8cfc6c71663
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Oct 15 09:36:25 2015 +0200

    drm/gem: Check locking in drm_gem_object_unreference

Cc: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Use new read bios from rom callback

Read the vbios directly from the rom.  In some cases,
e.g., virtualization, the rom is not available via
the BAR or other means.  Access it directly.

This is an updated version of Monks original patch which
uses family specific callbacks and unifies some of the
validation checking.

Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>