On GC 9.4.3, DW 41 in MQD is repurposed as compute_tg_chunk_size
for cooperative dispatch. When it's a AQL queue, set compute_tg_chunk_size
as 1 to spread work groups evenly among XCCs. If it's PM4 queue, unset
compute_tg_chunk_size to disable cooperative mode.
v3: set compute_tg_chunk_size as 1 instead of #CUs per XCC
v2: set compute_tg_chunk_size as #CUs per XCC instead of total wave
slots per XCC
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Sean Keely <Sean.Keely@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
v1:
Check new exception bits in TRAPSTS register
Remove single step exception workaround, now part of
exception bits
v2:
GC 9.4.3 uses ttmp11 to store {1’b0, dispatch index [24:0],
wave_id_in_workgroup[5:0]}, so use ttmp13 instead of ttmp11 to
preserve ib_sts. (Laurent)
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Signed-off-by: Laurent Morichetti <Laurent.Morichetti@amd.com>
Reviewed-by: Laurent Morichetti <laurent.morichetti@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Compiling AMD GPU drivers displays a warning:
drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn314/display_rq_dlg_calc_314.c: In function ‘dml_rq_dlg_get_dlg_params’:
drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn314/display_rq_dlg_calc_314.c:991:14: warning: variable ‘scl_enable’ set but not used [-Wunused-but-set-variable]
Get rid of it by removing the variable 'scl_enable'.
Signed-off-by: Caio Novais <caionovais@usp.br>
Signed-off-by: Hamza Mahfooz <hamza.mahfooz@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Commit 075e2099c32c ("drm/amd/display: Fix race condition in DPIA AUX
transfer") was backported to stable, which I noticed because of git
saying
linux-6.1/.git/rebase-apply/patch:37154: space before tab in indent.
while applying patch-6.1.21. While fixing the code location that issued
that warning, improve in few more places.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Hamza Mahfooz <hamza.mahfooz@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Some dock and mst monitor don't like to receive CLEAR_PAYLOAD_ID_TABLE
when mst_en is set to 0. It doesn't make sense to do so in source
side, either.
[How]
Don't send CLEAR_PAYLOAD_ID_TABLE if mst_en is 0
Reviewed-by: George Shen <George.Shen@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Peichen Huang <PeiChen.Huang@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Description]
If transitioning from an FPO config -> FPO config, we want
to keep cursor P-State force disallowed. Any other transition
from FPO config -> non FPO config should unforce the cursor
P-State disallow
Reviewed-by: Wesley Chalmers <Wesley.Chalmers@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Alvin Lee <Alvin.Lee2@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
The 6.75 GBps link rate is part of the new eDP specification
version 1.5 is going to be supported in the future.
Since this standard is very new and there are no existing 6.75 GBps
panels on the market yet, we should put a condition in the driver
on enabling this feature until we can validate it with real hardware.
[How]
- Add boolean flag support_eDP1_5 in struct dc_debug_options.
- Enable the 6.75 link rate in reduce_link_rate(...) only when
the flag is true.
Reviewed-by: Charlene Liu <Charlene.Liu@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Artem Grishin <Artem.Grishin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
The latest eDP spec version 1.5 defines a new generic link
rate of 6.75 Gbps/Lane, which needs to be supported in the driver.
[How]
Added new element to the dc_link_rate enum
Reviewed-by: Charlene Liu <Charlene.Liu@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Artem Grishin <Artem.Grishin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Observing underflow on dcn30+ system config at 4k144hz
[How]
We set the UCLK hardmax on AC/DC switch if softmax is enabled
and also on boot. While booting up the UCLK Hardmax is set
to softmax before the init sequence and the init sequence
resets the hardmax to UCLK max which enables P-state switching.
Just added a conditional check to avoid setting hardmax on init.
Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Reviewed-by: Martin Leung <Martin.Leung@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Ayush Gupta <ayugupta@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Description]
* Pass in pipe index for FPO cmd to DMCUB
- This change will pass in the pipe index for each stream
that is using FPO
- This change is in preparation to enable FPO + VActive
* Use per pipe P-State force for FPO
- For FPO, instead of using max watermarks value for P-State disallow,
use per pipe p-state force instead
- This is in preparation to enable FPO + VActive
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Alvin Lee <Alvin.Lee2@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
DPP Root clock optimization when combined with 4to1 MPC combine results
in the screen turning black.
This is because the DPPCLK is stopped during the middle of an
optimize_bandwidth sequence during commit_minimal_transition without
going through plane power down/power up.
[How]
The intent of a 0Hz DPP clock through update_clocks is to disable the
DTO. This differs from the behavior of stopping the DPPCLK entirely
(utilizing a 0Hz clock on some ASIC) so it's better to move this logic
to reside next to plane power up/power down where we gate the HUBP/DPP
DOMAIN.
The new sequence should be:
Power down: PG enabled -> RCO on
Power up: RCO off -> PG disabled
Rename power_on_plane to power_on_plane_resources to reflect the
actual operation that's occurring.
Cc: stable@vger.kernel.org
Cc: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Description]
- On high refresh rate DRR displays that support VBLANK naturally,
UCLK could be idling at DPM1 instead of DPM0 since it doesn't use
FPO
- To achieve DPM0, enable FPO on these configs even though it can
support P-State without FPO
- Default disable for now, have debug option to enable
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Alvin Lee <Alvin.Lee2@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Reset(mode1) failed as JPRG IP did not reinit under sriov.
[How]
Add JPEG IP block to sriov reinit function.
Signed-off-by: Yifan Zha <Yifan.Zha@amd.com>
Reviewed-by: Horace Chen <Horace.Chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Hack the mmhub 1.7 reg offset for initial support
v2: squash in header switch, CG funcs (Alex)
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make building fbdev emulation depend on DRM_FBDEV_EMULATION. Also
rename the source file to radeon_fbdev.c to align with other fbdev
files.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Implement fbdev emulation on top of struct drm_client and its helpers.
Replaces ad-hoc interfaces for restoring and closing fbdev emulation with
per-client callbacks for hotplugging, restoring and unregistering.
A single function, radeon_fbdev_setup(), starts fbdev emulation after
the DRM device has been registered. Hence, fbdev acts like a regular
DRM client.
The setup call prepares the fbdev emulation and invokes connector
hotplugging. The first successful hotplug event initializes fbdev
emulation with a framebuffer, device file, etc.
Unregistering depends on the hotplug status. Fully initialized emulation
is cleaned up through drm_fb_helper_unregister_info() and fb_destroy.
For prepared-only setups, unregistering unprepares the emulation and
releases all resources. In both cases, fbdev emulation will be cleaned
up.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Improve the fbdev probing function to fully clean up if it failed.
Allows to remove special cases from fb_destroy as well.
This change is reorders the operations within radeonfb_probe(). It
first allocates a buffer object, then builds a DRM framebuffer for
the object and finally creates the fbdev device. If every step
succeeded, the probe function clears the framebuffer memory. This
is the optimal order to rollback any changes if any of the steps
fails.
No functional changes.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fbdev calls struct fb_ops.fb_destroy after cleaning up the final
reference to an fbdev framebuffer. Move radeon's fbdev cleanup code
there.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the coding style in several places in the fbdev-emulation
code. Also rename functions and structure file by comments. No
functional changes.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Move the helpers for creating and destroying fbdev GEM objects
to the top of the source file. Makes them available to fb_ops
functions. This will allow to implement framebuffer cleanup in
fb_destroy. No functional changes.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The screen_base field comes from the fbdev BO and contains the fbdev
framebuffer base address. We get the BO memory via radeon_bo_kmap(),
which already reports the error -ENOMEM if the buffer could not be
mapped. So remove the later test for screen_base, which will never
be NULL at this point.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Both data fields in struct radeon_fbdev, the framebuffer and the
device, are already available in struct drm_fb_helper. Simplify
radeon by converting all callers and removing struct radeon_fbdev.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
