[Why]
Legacy cursor plane updates from drm helpers go through the full
atomic codepath. A high volume of cursor updates through this slow
code path can cause subsequent page-flips to skip vblank intervals
since each individual update is slow.
This problem is particularly noticeable for the compton compositor.
[How]
A fast path for cursor plane updates is added by using DRM asynchronous
commit support provided by async_check and async_update. These don't do
a full state/flip_done dependency stall and they don't block other
commit work.
However, DC still expects itself to be single-threaded for anything
that can issue register writes. Screen corruption or hangs can occur
if write sequences overlap. Every call that potentially perform
register writes needs to be guarded for asynchronous updates to work.
The dc_lock mutex was added for this.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106175
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The return statement is redundant as there is a return statement
immediately before it so we have dead code that can be removed.
Also remove the unused declaration of ret.
Detected by CoverityScan, CID#1473793 ("Structurally dead code")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Noticed this while working on redoing the reference counting scheme in
the DP MST helpers. Nouveau doesn't attempt to call
drm_dp_mst_topology_mgr_destroy() at all, which leaves it leaking all of
the resources for drm_dp_mst_topology_mgr and it's children mstbs+ports.
Fixes: f479c0ba4a ("drm/nouveau/kms/nv50: initial support for DP 1.2 multi-stream")
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: <stable@vger.kernel.org> # v4.10+
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Should hopefully fix a regression some people have been seeing since EVO
push buffers were moved to VRAM by default on Pascal GPUs.
Fixes: d00ddd9da ("drm/nouveau/kms/nv50-: allocate push buffers in vidmem on pascal")
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Cc: <stable@vger.kernel.org> # 4.19+
The DDB allocation algorithm currently used by the driver grants each
plane a very small minimum allocation of DDB blocks and then divies up
all of the remaining blocks based on the percentage of the total data
rate that the plane makes up. It turns out that this proportional
allocation approach is overly-generous with the larger planes and can
leave very small planes wthout a big enough allocation to even hit their
level 0 watermark requirements (especially on APL, which has a smaller
DDB in general than other gen9 platforms). Or there can be situations
where the smallest planes hit a lower watermark level than they should
have been able to hit with a more equitable division of DDB blocks, thus
limiting the overall system sleep state that can be achieved.
The bspec now describes an alternate algorithm that can be used to
overcome these types of issues. With the new algorithm, we calculate
all plane watermark values for all wm levels first, then go back and
partition a pipe's DDB space second. The DDB allocation will calculate
what the highest watermark level that can be achieved on *all* active
planes, and then grant the blocks necessary to hit that level to each
plane. Any remaining blocks are then divided up proportionally
according to data rate, similar to the old algorithm.
There was a previous attempt to implement this algorithm a couple years
ago in bb9d85f6e9 ("drm/i915/skl: New ddb allocation algorithm"), but
some regressions were reported, the patch was reverted, and nobody
ever got around to figuring out exactly where the bug was in that
version. Our watermark code has evolved significantly in the meantime,
but we're still getting bug reports caused by the unfair proportional
algorithm, so let's give this another shot.
v2:
- Make sure cursor allocation stays constant and fixed at the end of
the pipe allocation.
- Fix some watermark level iterators that weren't handling the max
level.
v3:
- Ensure we don't leave any DDB blocks unused by using DIV_ROUND_UP+min
to calculate the extra blocks for each plane. (Ville)
- Replace a while() loop with a for() loop to be more consistent with
surrounding code. (Ville)
- Clean unattainable watermark levels with memset rather than directly
clearing the member fields. Also do the same for the transition
watermark values if they can't be achieved. (Ville)
- Drop min_disp_buf_needed calculations in skl_compute_plane_wm() since
the results are no longer needed or used. (Ville)
- Drop skl_latency[0] != 0 sanity check; both watermark methods already
account for an invalid 0 latency by returning FP_16_16_MAX. (Ville)
v4:
- Break DDB allocation loop when total_data_rate=0 rather than
alloc_size=0. If total_data_rate has dropped to 0, all remaining
planes are disabled, which isn't true for alloc_size (we might just
have not had any remaining blocks to hand out). Plus
total_data_rate=0 is the case we need to avoid to a prevent a
div-by-0. (Ville)
- s/DIV_ROUND_UP/DIV64_U64_ROUND_UP/ to prevent 32-bit breakage (Ville)
v5:
- Don't forget to move 'start' pointer forward for UV surface when
setting plane DDB boundaries. (Ville)
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105458
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181211173107.11068-2-matthew.d.roper@intel.com
The bspec gives an if/else chain for choosing whether to use "method 1"
or "method 2" for calculating the watermark "Selected Result Blocks"
value for a plane. One of the branches of the if chain is:
"Else If ('plane buffer allocation' is known and (plane buffer
allocation / plane blocks per line) >=1)"
Since our driver currently calculates DDB allocations first and the
actual watermark values second, the plane buffer allocation is known at
this point in our code and we include this test in our driver's logic.
However we plan to soon move to a "watermarks first, ddb allocation
second" sequence where we won't know the DDB allocation at this point.
Let's drop this arm of the if/else statement (effectively considering
the DDB allocation unknown) as an independent patch so that any
regressions can be more accurately bisected to either the different
watermark value (in this patch) or the new DDB allocation (in the next
patch).
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181211173107.11068-1-matthew.d.roper@intel.com
[Why]
These properties aren't being carried over when the atomic state.
This tricks atomic check and commit tail into performing underscan
and scaling operations when they aren't needed.
With the patch that forced scaling/RMX_ASPECT on by default this
results in many unnecessary surface updates and hangs under certain
conditions.
[How]
Duplicate the properties.
Fixes: 91b66c47ba ("drm/amd/display: Set RMX_ASPECT as default")
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
If the "max bpc" isn't explicitly set in the atomic state then it
have a value of 0. This has the correct behavior of limiting a panel
to 8bpc in the case where the panel supports 8bpc. In the case of eDP
panels this isn't a true assumption - there are panels that can only
do 6bpc.
Banding occurs for these displays.
[How]
Initialize the max_bpc when the connector resets to 8bpc. Also carry
over the value when the state is duplicated.
Bugzilla: https://bugs.freedesktop.org/108825
Fixes: 307638884f72 ("drm/amd/display: Support amdgpu "max bpc" connector property")
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit 91b66c47ba.
Forcing RMX_ASPECT as default uses the preferred/native mode's timings
for any mode the user selects and scales the image. This provides a
a consistently nicer result in the case where the selected mode's
refresh rate matches the native mode's refresh but this isn't always
the case.
For example, if the monitor is 1080p@144Hz and the preferred mode is
60Hz then even if the user selects 1080p@144Hz as their selected mode
they'll get 1080p@60Hz.
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In case of msm drm bind failure, pm runtime put sync
is called from dsi driver which issues an asynchronous
put on mdss device. Subsequently when dpu_mdss_destroy
is triggered the change will make sure to put the mdss
device in suspend and clearing pending work if not
scheduled.
Signed-off-by: Jayant Shekhar <jshekhar@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Do some cleanup in the static inline functions defined in
dpu_media_info.h by cleaning up gotos and unneeded local
variables.
v3: Added spaces between operators per Seal Paul and Sam Ravnborg
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Remove more static inline functions that are lightly used and/or
very simple and easy to build into the calling functions.
v3: Fix a nit from Sean Paul
v2: Removed another unused function from dpu_hw_lm.c and add back
dpu_crtc_get_client_type() since there was a question regarding
its usefulness.
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Do some debugfs cleanups from across the DPU driver. The DRM
destroy functions will do a recursive delete on the entire
debugfs node so there is no need to store dentry pointers for
the debugfs files that are persistent for the life of the
driver. This also means that the destroy functions can go
away too.
Also, use standard API functions where applicable instead of
using hand written code.
v3: No changes
v2: Add more code; most of the dpu debugfs files should be
addressed now.
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
dpu_irq.c does some unneeded checks and passes control
to dpu_core_irq.c The simple functions can be defined
in the same file where we use them and the files and
their associated hangers on can be deleted.
Additionally the postinstall hook isn't used even
in dpu_core_irq.c so zap that entire path.
v3: No changes
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Allow the KMS operation 'irq_postinstall' to be optional
so that the target display drivers don't need to define
a dummy function if they don't need one.
v3: No changes
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Outside of superfluous parameter checks the dpu_hw_blk_init()
doesn't have any failure paths. Switch it over to be a void
function and we can remove error handling paths in all the functions
that call it. While we're in those functions remove unneeded
initialization for a static variable.
v3: No changes
v2: Removed a cleanup intended for a different patch
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Remove some unused container_of() helper functions.
v3: No changes
v2: Retained still used helper functions in the name of readability
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
The static inline function dpu_crtc_enabled() is only called once
and the function that calls it in turn is only called once and
the return value can be easily checked in the calling functions
so collapse everything down.
v3: No changes
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
dpu_crtc_get_mixer_height() is only used once and the value it
returns can be easily derived from the calling function.
v3: No changes
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
The functions in dpu_dbg.c aren't used. The two main dump functions
fail after a lookup from dpu_dbg_base.reg_base_list which turns out
to never be populated and once those are removed the rest of the
file doesn't make any sense.
v3: No changes
v2: Moved some unrelated changes to another patch
Reviewed-by: Sean Paul <sean@poorly.run>
Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Each time it's called we're holding the crtc modeset lock, so it's
redundant.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Instead of assigning/clearing the crtc on vblank enable/disable, we can
just assign and clear the crtc on modeset. That allows us to just toggle
the encoder's vblank interrupts on vblank_enable.
So why is this important? Previously the driver was using the legacy
pointers to assign/clear the crtc. Legacy pointers are cleared _after_
disabling the hardware, so the legacy pointer was valid during
vblank_disable, but that's not something we should rely on.
Instead of relying on the core ordering the legacy pointer assignments
just so, we'll assign the crtc in dpu_crtc enable/disable. This is the
only place that mapping can change, so we're covered there.
We're also taking advantage of drm_crtc_vblank_on/off. By using this, we
ensure that vblank_enable/disable can never be called while the crtc is
off (which means the assigned crtc will always be valid). As such, we
don't need to use modeset locks or the crtc_lock in the
vblank_enable/disable routine to be sure state is consistent.
...I think.
Changes in v2:
- Changed crtc check in toggle_vblank to != (Jeykumar)
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
[dpu_crtc.c change needed to be manually applied b/c of the dpu_crtc_reset change]
Signed-off-by: Rob Clark <robdclark@gmail.com>
The drm_crtc_vblank_on/off calls in enable/disable guarantee that we
won't call this function when crtc is not enabled.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
The indirection of registering a callback and opaque pointer isn't reall
useful when there's only one callsite. So instead of having the
vblank_cb registration, just give encoder a crtc and let it directly
call the vblank handler.
In a later patch, we'll make use of this further.
Changes in v2:
- None
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
I think the intention here was to protect the enc->crtc access, but
that's insufficient to avoid enc->crtc changing. Fortunately we're
already holding the modeset lock when this is called (from
atomic_check), so remove the crtc_lock and add a modeset lock check.
While we're at it, use the encoder mask from crtc state instead of
legacy pointer.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
There are 4 times that _dpu_crtc_vblank_enable_no_lock() is called:
1- crtc enable
2- crtc disable
3- crtc vblank enable
4- crtc vblank disable
When we enable or disable the crtc, we call drm_crtc_vblank_on and
drm_crtc_vblank_off respectively. That will gate vblank enables and
disables to only being called when the crtc is active. That means that
we can just enable/disable pm runtime in crtc enable/disable. This will
be beneficial in trying to eliminate blocking calls from the vblank call
chain.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Add modeset lock checks to functions that could be called outside the
core atomic stack.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
It's for legacy drivers, for atomic drivers crtc->state->encoder_mask
should be used to map encoder to crtc.
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
[seanpaul resolved conflict with async param of dpu_encoder_kickoff]
Signed-off-by: Rob Clark <robdclark@gmail.com>
This patch wraps dpu_core_perf_crtc_release_bw() with modeset locks
since it digs into the state objects.
Changes in v2:
- None
Changes in v3:
- Use those nifty new DRM_MODESET_LOCK_ALL_* helpers (Daniel)
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Now that runtime resume is handled in encoder, we don't need to worry
about crtc_lock recursion when calling pm_runtime_(get|put). So drop the
lock drops in _dpu_crtc_vblank_enable_no_lock().
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
The crtc runtime resume doesn't actually operate on the crtc, but rather
its encoders. The problem with this is that we need to inspect the crtc
state to get the currently connected encoders. Since runtime resume
isn't guaranteed to be called while holding the modeset locks (although
it sometimes is), this presents a race condition.
Now that we have ->enabled on the virtual encoders, and a lock to
protect it, just call resume on each encoder and only restore the ones
that are enabled.
Changes in v2:
- None
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
Add a bool to dpu_encoder_virt to track whether the encoder is enabled
or not. Repurpose the enc_lock mutex to ensure that it is consistent
with the hw state.
Changes in v2:
- None
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
It's only used in core_perf, so stick it there (and change the name to
reflect that).
Changes in v2:
- None
Reviewed-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
