Sunil Khatri [Thu, 3 Oct 2024 07:57:08 +0000 (13:27 +0530)]
drm/amdgpu: move error log from ring write to commit
Move the error message from ring write as an optimization
to avoid printing that message on every write instead
print once during commit if it exceeds write the allocated
size i.e ring->count_dw.
Also we do not want to log the error message in between a
ring write and complete the write as its mostly not harmful
as it will overwrite stale data only as GPU read from ring
is faster than CPU write to ring.
This reduces the size of amdgpu.ko module by around
600 Kb as write is very often used function and hence
the print.
drm/amdgpu: Remove the while loop from amdgpu_job_prepare_job
While loop makes it sound like amdgpu_vmid_grab() potentially needs to be
called multiple times to produce a fence, while in reality all code paths
either return an error, assign a valid job->vmid or assign a vmid which
will be valid once the returned fence signals.
Therefore we can remove the loop to make it clear the call does not need
to be repeated.
Alex Deucher [Fri, 4 Oct 2024 13:34:24 +0000 (09:34 -0400)]
drm/amd/display: disable SG displays on cyan skillfish
These parts were mainly for compute workloads, but they have
a display that was available for the console. These chips
should support SG display, but I don't know that the support
was ever validated on Linux so disable it by default. It can
still be enabled by setting sg_display=1 for those that
want to play with it. These systems also generally had large
carve outs so SG display was less of a factor.
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3356 Reviewed-by: Harry Wentland <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
Tvrtko Ursulin [Mon, 20 May 2024 11:13:38 +0000 (12:13 +0100)]
drm/amdgpu: Use drm_print_memory_stats helper from fdinfo
Convert fdinfo memory stats to use the common drm_print_memory_stats
helper.
This achieves alignment with the common keys as documented in
drm-usage-stats.rst, adding specifically drm-total- key the driver was
missing until now.
Additionally I made the code stop skipping total size for objects which
currently do not have a backing store, and I added resident, active and
purgeable reporting.
Legacy keys have been preserved, with the outlook of only potentially
removing only the drm-memory- when the time gets right.
Tvrtko Ursulin [Tue, 13 Aug 2024 13:57:11 +0000 (14:57 +0100)]
Documentation/gpu: Document the situation with unqualified drm-memory-
Currently it is not well defined what is drm-memory- compared to other
categories.
In practice the only driver which emits these keys is amdgpu and in them
exposes the current resident buffer object memory (including shared).
To prevent any confusion, document that drm-memory- is deprecated and an
alias for drm-resident-memory-.
While at it also clarify that the reserved sub-string 'memory' refers to
the memory region component, and also clarify the intended semantics of
other memory categories.
v2:
* Also mark drm-memory- as deprecated.
* Add some more text describing memory categories. (Alex)
v3:
* Semantics of the amdgpu drm-memory is actually as drm-resident.
Make the current_memory_partition sysfs node read/write for requesting a
new NPS mode. The request is only cached and at a later point a driver
unload/reload is required to switch to the new NPS mode.
drm/amdgpu/gfx10: Apply Isolation Enforcement to GFX & Compute rings
This commit applies isolation enforcement to the GFX and Compute rings
in the gfx_v10_0 module.
The commit sets `amdgpu_gfx_enforce_isolation_ring_begin_use` and
`amdgpu_gfx_enforce_isolation_ring_end_use` as the functions to be
called when a ring begins and ends its use, respectively.
`amdgpu_gfx_enforce_isolation_ring_begin_use` is called when a ring
begins its use. This function cancels any scheduled
`enforce_isolation_work` and, if necessary, signals the Kernel Fusion
Driver (KFD) to stop the runqueue.
`amdgpu_gfx_enforce_isolation_ring_end_use` is called when a ring ends
its use. This function schedules `enforce_isolation_work` to be run
after a delay.
These functions are part of the Enforce Isolation Handler, which
enforces shader isolation on AMD GPUs to prevent data leakage between
different processes.
Hamza Mahfooz [Fri, 4 Oct 2024 19:22:57 +0000 (15:22 -0400)]
drm/amd/display: fix hibernate entry for DCN35+
Since, two suspend-resume cycles are required to enter hibernate and,
since we only need to enable idle optimizations in the first cycle
(which is pretty much equivalent to s2idle). We can check in_s0ix, to
prevent the system from entering idle optimizations before it actually
enters hibernate (from display's perspective). Also, call
dc_set_power_state() before dc_allow_idle_optimizations(), since it's
safer to do so because dc_set_power_state() writes to DMUB.
drm/amd/display: Fetch the EDID from _DDC if available for eDP
Some manufacturers have intentionally put an EDID that differs from
the EDID on the internal panel on laptops.
Attempt to fetch this EDID if it exists and prefer it over the EDID
that is provided by the panel. If a user prefers to use the EDID from
the panel, offer a DC debugging parameter that would disable this.
drm/amd/display: remove redundant freesync parser for DP
When updating connector under drm_edid infrastructure, many calculations
and validations are already done and become redundant inside AMD driver.
Remove those driver-specific code in favor of the DRM common code.
drm/amd/display: switch amdgpu_dm_connector to use struct drm_edid
Replace raw edid handling (struct edid) with the opaque EDID type
(struct drm_edid) on amdgpu_dm_connector for consistency. It may also
prevent mismatch of approaches in different parts of the driver code.
drm/amd/display: Initialize new backlight_level_params structure
[Why]
Initialize the new backlight_level_params structure as part of the ABC
framework, the information in this structure is needed to be passed down
to the DMCUB to identify the backlight control type, to adjust the
backlight of the panel and to perform any required conversions from PWM
to nits or vice versa.
[How]
Created initial framework of the backlight_level_params struct and
modified existing functions to include the new structure.
Alex Hung [Mon, 23 Sep 2024 23:48:47 +0000 (17:48 -0600)]
drm/amd/display: Remove redundant assignments
[WHAT & HOW]
log2_blk_height and log2_blk_width are assigned to 0 and then
immediately are updated to other values. The assignments to zero are
redudant and removed.
This fixes 18 UNUSED_VALUE issues reported by Coverity.
Alex Hung [Mon, 23 Sep 2024 20:24:49 +0000 (14:24 -0600)]
drm/amd/display: Remove unnecessary assignments
[WHAT & HOW]
TimeForFetchingMetaPTE, TimeForFetchingRowInVBlank and
LinesToRequestPrefetchPixelData are local variables. They
are freed when CalculatePrefetchSchedule() ends and need
not clearing explicitly.
This fixes 21 UNUSED_VALUE issues reported by Coverity.
drm/amd/display: Update Interface to Check UCLK DPM
[Why]
Videos using YUV420 format may result in high power being used.
Disabling MPO may result in lower power usage.
Update interface that can be used to check power profile of a dc_state.
[How]
Add helper functions that can be used to determine power level:
- get power profile after a dc_state has undergone full validation
Roman Li [Thu, 26 Sep 2024 20:36:15 +0000 (16:36 -0400)]
drm/amd/display: Align static screen idle worker with IPX mode
[Why]
Idle worker thread serves for periodic detection of HPD while system is in IPS2.
Currently it is used in headless and static screen scenarios.
IPX can be configured not to execute IPS2 for static screen.
In this case idle worker is redundant.
[How]
Only use periodic detection for static screen if IPS is fully enabled.
drm/amd/display: calculate final viewport before TAP optimization
Viewport size excess surface size observed sometime with some timings or
resizing the MPO video window to cause MPO unsupported. Calculate final
viewport size first with a 100x100 dummy viewport to get the max TAP
support and then re-run final viewport calculation if TAP value changed.
Removed obsolete preliminary viewport calculation for TAP validation.
This is caused by lost memory during early init phase. First time driver
is removed, memory is freed but when second time the driver is inserted,
VBIOS dmub is not active, since the PSP policy is to retain the driver
loaded version on subsequent warm boots. Hence, communication with VBIOS
DMUB fails.
Fix this by aborting further communication with vbios dmub and release
the memory immediately.
Josip Pavic [Tue, 24 Sep 2024 21:25:54 +0000 (17:25 -0400)]
drm/amd/display: Clear update flags after update has been applied
[Why]
Since the surface/stream update flags aren't cleared after applying
updates, those same updates may be applied again in a future call to
update surfaces/streams for surfaces/streams that aren't actually part
of that update (i.e. applying an update for one surface/stream can
trigger unintended programming on a different surface/stream).
For example, when an update results in a call to
program_front_end_for_ctx, that function may call program_pipe on all
pipes. If there are surface update flags that were never cleared on the
surface some pipe is attached to, then the same update will be
programmed again.
[How]
Clear the surface and stream update flags after applying the updates.
drm/amd/display: Fix low black values by increasing error
[WHY]
Regamma resolution for the first few black levels can have problems for
calibration.
[HOW]
HW LUT is divided into N power-of-2 regions each with K segments. For
SDR mode we set min point at 2^-10 and increments of 2^-13. It's
generally more than 8-bit SDR needs, but some calibration tools and API
use 12-bit curves.
The fix shifts starting point to 2^-12 and starting increments at 2^-16.
drm/amd/display: Display lost signal on playing video
[Why]
When Source extend the vblank to reach the minimum panel
refresh rate, the vtotal length could have 1 line longer
than the maximum supported vtotal.
The reason is we optimized the vtotal/refresh-rate calculation
to get more accurate vtotal number by rounding the calculation
result. But when the target refresh rate is the minimum
refresh rate, the vtotal result could be round up and over
the maximum supported vtotal.
drm/amd/display: Fix garbage or black screen when resetting otg
[Why]
For some EDP to MIPI panel, disabling OTG when link is alive like boot
case, the converter might output garbage or show no display because our
GPU is not sending required pixel data.
Alos Dig fifo underflow was found which might cause garbage, when
resetting otg for other types of EDP panels.
[How]
Skipping resetting OTG if the dig fifo is on. Make sure that the otg for
the pipe is the one that the dig fifo is selecting via the FE mask.
drm/amd/display: Flip All Planes Under OTG Master When Flip Immediate
[Why]
The MPO plane will receive a flip but
desktop plane may not receive a flip when GSL is enabled.
As a result, system will be stuck waiting for a flip that was never sent.
[How]
Set update address update flag of all flip_immediate planes
if there are multiple planes.
drm/amd/display: skip disable CRTC in seemless bootup case
Resync FIFO is a workaround to write the same value to
DENTIST_DISPCLK_CNTL register after programming OTG_PIXEL_RATE_DIV
register, in case seemless boot, there is no OTG_PIXEL_RATE_DIV register
update, so skip CRTC disable when resync FIFO to avoid random FIFO error
and garbage.
drm/amd/display: Unify blank_phantom and blank_pixel_data
[Why]
dcn32_blank_phantom() does not consider the subVP+ODM case when blanking.
Only one of the pipes will get blanked. Remaining pipes are not blanked.
Will cause underflow in the phantom pipe when enabling the CRTC.
[How]
Use blank_pixel_data() instead of blank_phantom().
remove dcn32_blank_phantom() since logic is identical.
Different DPG dimensions get programmed when blanking phantom pipes.
Previously had phantom pipes use DPG dimensions of the main stream.
Now use DPG dimensions of the phantom streams
The patch is to solve null dereference in 'aux.dev', which is
introduced in recent radeon rework. By having 'late_register',
the connector should be registered after 'drm_dev_register'
automatically, where in before it is the opposite.
Fixes: 90985660ba48 ("drm/radeon: remove load callback from kms_driver") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3665 Tested-by: Hans de Goede <[email protected]> Suggested-by: Christophe Leroy <[email protected]> Signed-off-by: Wu Hoi Pok <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
drm/amdgpu/gfx11: Apply Isolation Enforcement to GFX & Compute rings
This commit applies isolation enforcement to the GFX and Compute rings
in the gfx_v11_0 module.
The commit sets `amdgpu_gfx_enforce_isolation_ring_begin_use` and
`amdgpu_gfx_enforce_isolation_ring_end_use` as the functions to be
called when a ring begins and ends its use, respectively.
`amdgpu_gfx_enforce_isolation_ring_begin_use` is called when a ring
begins its use. This function cancels any scheduled
`enforce_isolation_work` and, if necessary, signals the Kernel Fusion
Driver (KFD) to stop the runqueue.
`amdgpu_gfx_enforce_isolation_ring_end_use` is called when a ring ends
its use. This function schedules `enforce_isolation_work` to be run
after a delay.
These functions are part of the Enforce Isolation Handler, which
enforces shader isolation on AMD GPUs to prevent data leakage between
different processes.
drm/amdgpu/gfx12: Implement cleaner shader support for GFX12 hardware
This patch adds support for the PACKET3_RUN_CLEANER_SHADER packet in the
gfx_v12_0 module. This packet is used to emit the cleaner shader, which
is used to clear GPU memory before it's reused, helping to prevent data
leakage between different processes.
Finally, the patch updates the ring function structures to include the
new gfx_v12_0_ring_emit_cleaner_shader function. This allows the
cleaner shader to be emitted as part of the ring's operations.
Philip Yang [Mon, 29 Jul 2024 18:42:18 +0000 (14:42 -0400)]
drm/amdkfd: Increase SMI event fifo size
SMI event fifo size 1KB was enough to report GPU vm fault or reset
event, but could drop the more frequent SVM migration events. Increase
kfifo size to 8KB to store about 100 migrate events, less chance to drop
the migrate events if lots of migration happened in the short period of
time. Add KFD prefix to the macro name.
Philip Yang [Fri, 16 Feb 2024 16:41:16 +0000 (11:41 -0500)]
drm/amdkfd: Output migrate end event if migrate failed
If page migration failed, also output migrate end event to match with
migrate start event, with failure error_code added to the end of the
migrate message macro. This will not break uAPI because application uses
old message macro sscanf drop and ignore the error_code.
Output GPU page fault restore end event if migration failed.
drm/amdgpu/gfx11: Implement cleaner shader support for GFX11 hardware
The patch modifies the gfx_v11_0_kiq_set_resources function to write
the cleaner shader's memory controller address to the ring buffer. It
also adds a new function, gfx_v11_0_ring_emit_cleaner_shader, which
emits the PACKET3_RUN_CLEANER_SHADER packet to the ring buffer.
This patch adds support for the PACKET3_RUN_CLEANER_SHADER packet in the
gfx_v11_0 module. This packet is used to emit the cleaner shader, which
is used to clear GPU memory before it's reused, helping to prevent data
leakage between different processes.
Finally, the patch updates the ring function structures to include the
new gfx_v11_0_ring_emit_cleaner_shader function. This allows the
cleaner shader to be emitted as part of the ring's operations.
drm/amd/pm: use pm_runtime_get_if_active for debugfs getters
Don't wake up the GPU for reading pm values. Instead, take a runtime
powermanagement ref when trying to read it if and only if the GPU is
already awake.
This avoids spurious wake ups (eg: from applets).
We use pm_runtime_get_if_in_active because we care about "is the GPU awake?"
not about "is the GPU awake and something else prevents suspend?".
Sunil Khatri [Wed, 2 Oct 2024 13:54:25 +0000 (19:24 +0530)]
drm/amdgpu: change the comment from handle to ip_block
htmldoc generation depend upon the input arguments etc
to generate the document. After update of handle to
ip_block then update needs in comments too to fix the
warnings.
drm/amdgpu/gfx10: Implement cleaner shader support for GFX10 hardware
The patch modifies the gfx_v10_0_kiq_set_resources function to write
the cleaner shader's memory controller address to the ring buffer. It
also adds a new function, gfx_v10_0_ring_emit_cleaner_shader, which
emits the PACKET3_RUN_CLEANER_SHADER packet to the ring buffer.
This patch adds support for the PACKET3_RUN_CLEANER_SHADER packet in the
gfx_v10_0 module. This packet is used to emit the cleaner shader, which
is used to clear GPU memory before it's reused, helping to prevent data
leakage between different processes.
Finally, the patch updates the ring function structures to include the
new gfx_v10_0_ring_emit_cleaner_shader function. This allows the
cleaner shader to be emitted as part of the ring's operations.
drm/amd/pm: don't update runpm last_usage on debugfs getter
Reading pm values from the GPU shouldn't prevent it to be suspended
by resetting the last active timestamp (eg: if an background app
monitors GPU sensors every second, it would prevent the autosuspend
sequence to trigger).
drm/amdgpu/gfx9: Add Cleaner Shader Deinitialization in gfx_v9_0 Module
This commit addresses an omission in the previous patch related to the
cleaner shader support for GFX9 hardware. Specifically, it adds the
necessary deinitialization code for the cleaner shader in the
gfx_v9_0_sw_fini function.
The added line amdgpu_gfx_cleaner_shader_sw_fini(adev); ensures that any
allocated resources for the cleaner shader are freed correctly, avoiding
potential memory leaks and ensuring that the GPU state is clean for the
next initialization sequence.
drm/amd/powerplay: Delete unused function and maths library
We start with the function 'atomctrl_calculate_voltage_evv_on_sclk'
which has been unused since 2016's commit e805ed83ba1c ("drm/amd/powerplay: delete useless files.")
Remove it.
It was also the last user of the entire fixed point maths library in
ppevvmath.h.
Some distributions have been patching amdgpu to enable overdrive by
default which may compromise stability. Furthermore when bug reports
are brought upstream it's not obvious that the system has been tampered
with.
When overdrive is enabled taint the kernel and leave a critical message
in the logs for users so that it's obvious in a bug report it's been
tampered with.
drm/amdkfd: Fix kdoc entry for 'get_wave_count()' function parameters
Update kdoc entries to reflect the function's parameters. The descriptor
for the 'queue_cnt' parameter has been added, and the incorrect mentions
of 'wave_cnt' and 'vmid', which are not parameters but local variables,
have been removed.
Fixes the below with gcc W=1:
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c:954: warning: Function parameter or struct member 'queue_cnt' not described in 'get_wave_count'
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c:954: warning: Excess function parameter 'wave_cnt' description in 'get_wave_count'
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c:954: warning: Excess function parameter 'vmid' description in 'get_wave_count'
drm/amdgpu: Flush tlb by VM_INVALIDATION packet in sdma_v5_2
In order for SDMA not to be switched between VM_INVALIDATION
request and ack, use an single VM_INVALIDATION packet in function
sdma_v5_2_ring_emit_vm_flush.
DC 3.2.303 contains some improvements as summarized below:
* Improve brightness control
* Add support for UHBR10 eDP
* OPTC required only for DTBCLK_P for dcn401
* Fix TBT monitor resume issue
* Code cleanup
drm/amd/display: Configure DTBCLK_P with OPTC only for dcn401
[WHY]
DTBCLK_P is used to generate virtual pixel clock, and to drive the HPO
stream encoder clock. Programming the required clock when
enabling/disabling both components can cause issues.
For example, if HPO is being disabled and clock source is changed to
REFCLK, virtual pixel rate will then be wrong, causing issues in CRTC.
[HOW]
Only program the DTBCLK_P when programming CRTC, as its expected it will
be enabled prior to HPO, and disabled after HPO in all valid cases.
drm/amd/display: Add IPS residency capture helpers to dc_dmub_srv
This enables starting and stopping IPS residency measurements
and querying the IPS residency information consisting of residency
percent, entry counter, total time active & inactive, and histograms
for the specified IPS mode.
projects / linux.git / log