Amber Lin [Wed, 22 Aug 2018 20:48:50 +0000 (16:48 -0400)]
drm/amdgpu: Merge amdkfd into amdgpu
Since KFD is only supported by single GPU driver, it makes sense to merge
amdgpu and amdkfd into one module. This patch is the initial step: merge
Kconfig and Makefile.
[Why]
The dce_i2c_hw code contained four funtcions that were only
called in one place and did not have a clearly delineated
purpose.
[How]
Inline these functions, keeping the same functionality.
This is not a functional change.
The functions disable_i2c_hw_engine and release_engine_dce_hw were
pulled into their respective callers.
The most interesting part of this change is the acquire functions.
dce_i2c_hw_engine_acquire_engine was pulled into
dce_i2c_engine_acquire_hw, and dce_i2c_engine_acquire_hw was pulled
into acquire_i2c_hw_engine.
Some notes to show that this change is not functional:
-Failure conditions in any function resulted in a cascade of calls that
ended in a 'return NULL'.
Those are replaced with a direct 'return NULL'.
-The variable result is the one from dce_i2c_hw_engine_acquire_engine.
The boolean result used as part of return logic was removed.
-As the second half of dce_i2c_hw_engine_acquire_engine is only executed
if that function is returning true and therefore exiting the do-while
loop in dce_i2c_engine_acquire_hw, those lines were moved outside
of the loop.
Eric Yang [Wed, 15 Aug 2018 21:35:50 +0000 (17:35 -0400)]
drm/amd/display: support 48 MHZ refclk off
[Why]
On PCO and up, whenever SMU receive message to indicate active
display count = 0. SMU will turn off 48MHZ TMDP reference clock
by writing to 1 TMDP_48M_Refclk_Driver_PWDN. Once this clock is
off, no PHY register will respond to register access. This means
our current sequence of notifying display count along with requesting
clock will cause driver to hang when accessing PHY registers after
displays count goes to 0.
[How]
Separate the PPSMC_MSG_SetDisplayCount message from the SMU messages
that request clocks, have display own sequencing of this message so
that we can send it at the appropriate time.
Do not redundantly power off HW when entering S3, S4, since display
should already be called to disable all streams. And ASIC soon be
powered down.
drm/amd/display: Support reading hw state from debugfs file
[Why]
Logging hardware state can be done by triggering a write to the
debugfs file. It would also be useful to be able to read the hardware
state from the debugfs file to be able to generate a clean log without
timestamps.
Threading is an obvious concern when dealing with multiple debugfs
operations and blocking on global state in dm or dc seems unfavorable.
Adding an extra parameter for the debugfs log context state is the
implementation done here. Existing code that made use of DTN_INFO
and its associated macros needed to be refactored to support this.
We don't know the size of the log in advance so it reallocates the
log string dynamically. Once the log has been generated it's copied
into the user supplied buffer for the debugfs. This allows for seeking
support but it's worth nothing that unlike triggering output via
dmesg the hardware state might change in-between reads if your buffer
size is too small.
Ken Chalmers [Fri, 10 Aug 2018 19:51:59 +0000 (15:51 -0400)]
drm/amd/display: eliminate long wait between register polls on Maximus
[Why]
Now that we "scale" time delays correctly on Maximus (as of diags svn
r170115), the forced "35 ms" wait time now becomes 35 ms * 500 = 17.5
seconds, which is far too long. Even having to repeat polling a
register once causes excessive delays on Maximus.
[How]
Just use the regular wait time passed to the generic_reg_wait()
function. This is sufficient for Maximus now, and it also means that
there's one less "Maximus-only" code path in DAL.
Also disable the "REG_WAIT taking a while:" message on Maximus, since
things do take a while longer there and 1-2ms delays are not uncommon
(and nothing to worry about).
drm/amd/display: Add support for hw_state logging via debugfs
[Why]
We have logging methods for printing hardware state for newer ASICs
but no way to trigger the log output.
[How]
Add support for triggering the output via writing to a debugfs file
entry. Log output currently goes into dmesg for convenience, but
accessing via a read should be possible later.
David Francis [Thu, 9 Aug 2018 17:20:04 +0000 (13:20 -0400)]
drm/amd/display: Eliminate i2c hw function pointers
[Why]
The function pointers of the dce_i2c_hw struct were never
accessed from outside dce_i2c_hw.c and had only one version.
As function pointers take up space and make debugging difficult,
and they are not needed in this case, they should be removed.
[How]
Remove the dce_i2c_hw_funcs struct and make static all
functions that were previously a part of it. Reorder
the functions in dce_i2c_hw.c.
Rex Zhu [Wed, 22 Aug 2018 10:54:45 +0000 (18:54 +0800)]
drm/amdgpu: Change kiq ring initialize sequence on gfx9
1. initialize kiq before initialize gfx ring.
2. set kiq ring ready immediately when kiq initialize
successfully.
3. split function gfx_v9_0_kiq_resume into two functions.
gfx_v9_0_kiq_resume is for kiq initialize.
gfx_v9_0_kcq_resume is for kcq initialize.
Rex Zhu [Wed, 22 Aug 2018 09:58:31 +0000 (17:58 +0800)]
drm/amdgpu: Change kiq initialize/reset sequence on gfx8
1. initialize kiq before initialize gfx ring.
2. set kiq ring ready immediately when kiq initialize
successfully.
3. split function gfx_v8_0_kiq_resume into two functions.
gfx_v8_0_kiq_resume is for kiq initialize.
gfx_v8_0_kcq_resume is for kcq initialize.
Rex Zhu [Fri, 17 Aug 2018 08:42:35 +0000 (16:42 +0800)]
drm/amdgpu: Remove duplicate code in gfx_v9_0.c
There are no any logical changes here.
1. if kcq can be enabled via kiq, we don't need to
do kiq ring test.
2. amdgpu_ring_test_ring function can be used to
sync the ring complete, remove the duplicate code.
Rex Zhu [Fri, 17 Aug 2018 05:13:12 +0000 (13:13 +0800)]
drm/amdgpu: Remove duplicate code in gfx_v8_0.c
There are no any logical changes here.
1. if kcq can be enabled via kiq, we don't need to
do kiq ring test.
2. amdgpu_ring_test_ring function can be used to
sync the ring complete, remove the duplicate code.
Rex Zhu [Thu, 23 Aug 2018 07:45:15 +0000 (15:45 +0800)]
drm/amdgpu: Remove dead code in amdgpu_pm.c
As we have unify powergate_uvd/vce/mmhub to set_powergating_by_smu,
and set_powergating_by_smu was supported by both dpm and powerplay.
so remove the else case.
Rex Zhu [Thu, 23 Aug 2018 03:46:13 +0000 (11:46 +0800)]
drm/amdgpu: Remove duplicated power source update
when ac/dc switch, driver will be notified by acpi event.
then the power source will be updated. so don't need to
get power source when set power state.
Felix Kuehling [Tue, 21 Aug 2018 21:14:32 +0000 (17:14 -0400)]
drm/amdgpu: Adjust the VM size based on system memory size v2
Set the VM size based on system memory size between the ASIC-specific
limits given by min_vm_size and max_bits. GFXv9 GPUs will keep their
default VM size of 256TB (48 bit). Only older GPUs will adjust VM size
depending on system memory size.
This makes more VM space available for ROCm applications on GFXv8 GPUs
that want to map all available VRAM and system memory in their SVM
address space.
v2:
* Clarify comment
* Round up memory size before >> 30
* Round up automatic vm_size to power of two
Huang Rui [Mon, 6 Aug 2018 02:57:08 +0000 (10:57 +0800)]
drm/amdgpu: use bulk moves for efficient VM LRU handling (v6)
I continue to work for bulk moving that based on the proposal by Christian.
Background:
amdgpu driver will move all PD/PT and PerVM BOs into idle list. Then move all of
them on the end of LRU list one by one. Thus, that cause so many BOs moved to
the end of the LRU, and impact performance seriously.
Then Christian provided a workaround to not move PD/PT BOs on LRU with below
patch:
Commit 0bbf32026cf5ba41e9922b30e26e1bed1ecd38ae ("drm/amdgpu: band aid
validating VM PTs")
However, the final solution should bulk move all PD/PT and PerVM BOs on the LRU
instead of one by one.
Whenever amdgpu_vm_validate_pt_bos() is called and we have BOs which need to be
validated we move all BOs together to the end of the LRU without dropping the
lock for the LRU.
While doing so we note the beginning and end of this block in the LRU list.
Now when amdgpu_vm_validate_pt_bos() is called and we don't have anything to do,
we don't move every BO one by one, but instead cut the LRU list into pieces so
that we bulk move everything to the end in just one operation.
After test them with above three benchmarks include vulkan and opencl. We can
see the visible improvement than original, and even better than original with
workaround.
v2: move all BOs include idle, relocated, and moved list to the end of LRU and
put them together.
v3: remove unused parameter and use list_for_each_entry instead of the one with
save entry.
v4: move the amdgpu_vm_move_to_lru_tail after command submission, at that time,
all bo will be back on idle list.
v5: remove amdgpu_vm_move_to_lru_tail_by_list(), use bulk_moveable instread of
validated, and move ttm_bo_bulk_move_lru_tail() also into
amdgpu_vm_move_to_lru_tail().
v6: clean up and fix return value.
Huang Rui [Mon, 6 Aug 2018 09:28:35 +0000 (17:28 +0800)]
drm/ttm: add bulk move function on LRU
This function allow us to bulk move a group of BOs to the tail of their LRU.
The positions of group of BOs are stored on the (first, last) bulk_move_pos
structure.
Christian König [Mon, 6 Aug 2018 09:05:30 +0000 (17:05 +0800)]
drm/ttm: revise ttm_bo_move_to_lru_tail to support bulk moves
When move a BO to the end of LRU, it need remember the BO positions.
Make sure all moved bo in between "first" and "last". And they will be bulk
moving together.
drm/amdgpu: amdgpu_kiq_reg_write_reg_wait() can be static
Fixes: d790449835e6 ("drm/amdgpu: use kiq to do invalidate tlb") Reviewed-by: Emily Deng <[email protected]> Signed-off-by: kbuild test robot <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
Wen Yang [Fri, 17 Aug 2018 03:09:48 +0000 (11:09 +0800)]
drm/amd/display: fix a compile warning
Fix comile warning like,
CC [M] drivers/gpu/drm/i915/gvt/execlist.o
CC [M] drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.o
CC [M] drivers/gpu/drm/radeon/btc_dpm.o
CC [M] drivers/isdn/hisax/avm_a1p.o
CC [M] drivers/gpu/drm/amd/amdgpu/../display/dc/dcn10/dcn10_dpp.o
drivers/gpu/drm/amd/amdgpu/../display/dc/dcn10/dcn10_hw_sequencer.c: In function ‘dcn10_update_mpcc’:
drivers/gpu/drm/amd/amdgpu/../display/dc/dcn10/dcn10_hw_sequencer.c:1903:9: warning: missing braces around initializer [-Wmissing-braces]
struct mpcc_blnd_cfg blnd_cfg = {0};
^
drivers/gpu/drm/amd/amdgpu/../display/dc/dcn10/dcn10_hw_sequencer.c:1903:9: warning: (near initialization for ‘blnd_cfg.black_color’) [-Wmissing-braces]
drm/scheduler: Add stopped flag to drm_sched_entity
The flag will prevent another thread from same process to
reinsert the entity queue into scheduler's rq after it was already
removewd from there by another thread during drm_sched_entity_flush.
Jun Lei [Wed, 8 Aug 2018 15:53:39 +0000 (11:53 -0400)]
drm/amd/display: implement DPMS DTN test v2
[why]
Existing DTN infrastructure in driver is hacky. It uses implicit log
names, and also incorrect escape ID.
[how]
- Implement using generic DTN escape ID.
- Move file logging functionality from driver to to script; driver now outputs to string/buffer
- Move HWSS debug functionality to separate c file
- Add debug functionalty for per-block logging as CSV
- Add pretty print in python
Anthony Koo [Tue, 21 Aug 2018 19:28:05 +0000 (14:28 -0500)]
drm/amd/display: move edp fast boot optimization flag to stream
[Why]
During S4/S3 stress test it is possible to resume from S4 without
calling mode set on eDP, meaning high level optimization flag is not
reset. If this is followed by an S3 resume call, driver will see
optimization flag is set and consume it and think backend is powered
on when in fact it is not.
This results in PHY being off in sequence where
S4->Resume->S3->Resume->ApplyOpt->black screen.
[How]
Move optimization flag to stream instead of a DC flag.
David Francis [Thu, 9 Aug 2018 17:15:36 +0000 (13:15 -0400)]
drm/amd/display: Combine dce80 and dce100 i2c hw functions
[Why]
There are two versions of the hw function pointers: one for dce80
and one for all other versions. These paired functions are
nearly identical. dce80 and dce100 should not require
different i2c access functions.
[How]
Combine each pair of functions into a single function. Mostly
the new functions are based on the dce100 versions as those
versions are newer, support more features, and
were more maintained.
David Francis [Mon, 23 Jul 2018 18:12:10 +0000 (14:12 -0400)]
drm/amd/display: Create new i2c resource
[Why]
I2C code did not match dc resource model and was generally
unpleasant
[How]
Move code into new svelte dce_i2c files, replacing various i2c
objects with two structs: dce_i2c_sw and dce_i2c_hw. Fully split
sw and hw code paths. Remove all redundant declarations. Use
address lists to distinguish between versions. Change dce80 code
to newer register access macros.
Paul Menzel [Tue, 31 Jul 2018 16:48:41 +0000 (18:48 +0200)]
drm/radeon: Do not evict VRAM on APUs with disabled HIBERNATE
Improve commit d796d844 (drm/radeon/kms: make hibernate work on IGPs) to
only migrate VRAM objects if the Linux kernel is actually built with
support for hibernation (suspend to disk).
Paul Menzel [Wed, 25 Jul 2018 10:54:19 +0000 (12:54 +0200)]
drm/amdgpu: Do not evict VRAM on APUs with disabled HIBERNATE
Improve commit d796d844 (drm/radeon/kms: make hibernate work on IGPs) to
only migrate VRAM objects if the Linux kernel is actually built with
support for hibernation (suspend to disk).
The better solution is to get the information, if this is suspend or
hibernate, from `amdgpu_device_suspend()`, but that’s more involved, so
apply the simple solution first.
Rex Zhu [Tue, 14 Aug 2018 05:32:30 +0000 (13:32 +0800)]
drm/amdgpu: Refine function name and function args
There are no any logical changes here.
1. change function names:
amdgpu_device_ip_late_set_pg/cg_state to
amdgpu_device_set_pg/cg_state.
2. add a function argument cg/pg_state, so
we can enable/disable cg/pg through those functions
drm/amd/display: HDR dynamic meta should be treated as stream update
[Why]
Recently we fixed HDR static meta using AFMT registers to be treated as
fast stream update.
Dynamic meta is still being treated as (full) surface update because it
touches HUBP and it travels with pipe data.
Here we change it to be (fast) stream update.
Note, originally we also wanted to redesign here a bit, but without OS
level support for true dynamic meta, it's left the same. We are simply
using HW that can do dynamic meta to send HDR static meta, I still prefer
keeping it in one static meta type then defining dynamic meta types to
hold the same info. Once we know how OS interfaces look like, we can
do proper design.
[How]
Move dyn meta update from update_hubp_dpp to commit_planes_do_stream_update
drm/amd/display: Handle HDR meta update as fast update
[Why]
Vesa DPMS tool sends different HDR meta in OS flips without changing output
parameters. We don't properly update HDR info frame:
- we label HDR meta update as fast update
- when updating HW info frame, we only do it if full update
[How]
It should still be fast update, so when doing HW infoframe update,
do it always no matter the update type.
Also, don't request passive flip for HDR meta update only without output
transfer function or color space changed.
We only need to register once OS calls the interrupt control.
Also, if we are entering static screen mode, disable after ramping is done.
Disable shall be done via timer of 2 seconds regardless of ramping
complete or not, just to simplify.
Also, ramp to mid instead of min, due to better flicker performance...
Anthony Koo [Fri, 6 Apr 2018 16:12:06 +0000 (12:12 -0400)]
drm/amd/display: fix LFC tearing at top of screen
Tearing occurred because new VTOTAL MIN/MAX was being programmed
too early.
The flip can happen within the VUPDATE high region, and the new min/max
would take effect immediately. But this means that frame is not variable
anymore, and tearing would occur when the flip actually happens.
The fixed insert duration should be programmed on the first VUPDATE
interrupt instead.
Anthony Koo [Thu, 5 Apr 2018 01:04:42 +0000 (21:04 -0400)]
drm/amd/display: Fix bug that causes black screen
Ignore MSA bit on DP display is usually set during SetTimings, but
there was a case where the module thought refresh rate was not valid
and ignore MSA bit was not set.
Later, a valid refresh rate range was requested but since ignore MSA bit
not set, it caused black screen.
Issue if with how the module checked for VRR support. Fix up that logic.
DM should call new valid_range function to determine if timing is supported.
Anthony Koo [Thu, 5 Apr 2018 01:01:21 +0000 (21:01 -0400)]
drm/amd/display: Fix bug where refresh rate becomes fixed
This issue occurs if refresh rate range is very small and lfc is not used.
When frame spikes occur, refresh rate becomes fixed and will not restore properly
Michel Dänzer [Wed, 15 Aug 2018 10:58:13 +0000 (12:58 +0200)]
drm/doc: Adapt GPU scheduler references for renamed C file
Fixes: "drm/scheduler: rename gpu_scheduler.c to sched_main.c" Reviewed-by: Christian König <[email protected]> Signed-off-by: Michel Dänzer <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
projects / linux.git / log