This snapshot change adds downstream support
for drm 5.x+(msm_lahaina branch) linux kernel.
Change-Id: Ia691c95da155a00e449c91a2f1a5b20a8e71aed4
Signed-off-by: Narendra Muppalla <NarendraM@codeaurora.org>
The RSCC static wakeup and the bandwidth trigger for
the downvotes are based on the prefill lines. Reduce
the prefill lines based on the panel vertical front
porch to avoid issuing bw downvotes during active
region of the previous frame.
Change-Id: I408209ba308c32e71d9f70c5ed7e60c134877c84
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
The encoder idle work is scheduled during the frame-done event
to be executed after a timeout. During the execution, the check
for any on-going/pending frames is invalid as it checks for > 1.
Fix it to check for any non-zero frame-pending and avoid
power-collapse.
Change-Id: If7715ee56cc9bfa63787811458f4fc91de540013
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Add changes to support avr mode config update during
prepare commit which happens before gpu fence wait
for the input buffers.
Change-Id: Ib2cb5b7e1f10501914c003f6cf066b85048f79d4
Signed-off-by: Kalyan Thota <kalyant@codeaurora.org>
Signed-off-by: Jayaprakash <jmadiset@codeaurora.org>
Switch to watchdog vsync whenever panel jitter is
identified during frame-transfer on command mode display.
This would allow the HW to finish processing the frame
with watchdog vsync source. Switch back to default vsync
source after the frame-transfer is complete. This would
help in the MDP hang issues in panels that generate TEs
with thresholds greater than the projected jitter.
Change-Id: Ic3fa78d90e7f44cb0186857716ac27e72505fd32
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Add null check before accessing pingpong hw block
allocated during mode set to physical encoder.
Change-Id: Ic464e7c7087f280b1198f6b7485bc0763322c532
Signed-off-by: Jayaprakash <jmadiset@codeaurora.org>
During SUI validations and transitions, the encoders
are iterated to find the panel-mode. If concurrent
writeback is enabled, it is assumed as video mode.
Add check to consider only the built-in display
panel mode during sui to avoid wrong interpretation.
Change-Id: I2901eb29a0799571a95f6ab6f9f3f2c6154424c5
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Input handler is only registered in command mode, so add this change
to ensure that the input handler will be only unregistered in command
mode.
Change-Id: I15f652439339eee45b0ed3c011aa2922151f68ba
Signed-off-by: Lei Chen <chenlei@codeaurora.org>
2 Frames transfer pending is possible with posted start.
One ongoing frame and another triggered frame. Current SW
waits for pp_done interrupt if pending frame count is greater
than 1. It is possible that interrupt may be missed for ongoing
frame. In that case, SW should run pp_done wait for one by one
frame instead of two frames together. It allows encoder to
check the ctl scheduler status and trigger the frame done
event on time.
Change-Id: I4817842292d96747890ee70da8a5bdf9b56816ed
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Currently LUTDMA kickoff is immediately followed by CTL flush,
immaterial of the LUTDMA DONE status in command mode. If LUTDMA
kickoff happens too close to the read_ptr in auto-refresh case, it might
cause a race condition between LUTDMA & CTL flush, due to a HW issue.
Serialize LUTDMA & CTL flush by making the LUTDMA kickoff as blocking
to avoid the race condition in auto-refresh case similar to video mode.
Change-Id: I4f4ae90a2c4f1bc8a0686d8fd4f8aa439123c531
Signed-off-by: Raviteja Tamatam <travitej@codeaurora.org>
Signed-off-by: Jayaprakash <jmadiset@codeaurora.org>
Handle release fence/frame-done error signalling for
error case like esd failure, pp_done timeout, interrupt
disable on cpu, etc. It fixes the race condition for
pending_frame count update and also triggers correct
wait function for wr_ptr wait failure.
Change-Id: Iad08f20592c97221a1626bb40e607c398a9812b6
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Input handler is used for early wakeup DSI and MDP clocks from
idle power collapse. DSI clock won't be disabled in video mode
during idle power collapse, so the input hanlder isn't needed
for video mode.
This change avoids the input handler to be registered in
video mode.
Change-Id: Id23bad192b6671126978d707db464e7aaee1c77f
Signed-off-by: Lei Chen <chenlei@codeaurora.org>
Print uidle status fal10 enable bit as part of checking status.
Change-Id: Ibe00216ac22bb31fbe0925db3abc1d5dc4371ad3
Signed-off-by: Nilaan Gunabalachandran <ngunabal@codeaurora.org>
CWB, DP and WB displays tear down sequence must reset
3d_merge, ctl, pingpong_binding, etc. MDP HW
blocks. This change fixes the tear down
sequence register programming. It also moves flush
sw reset before encoder_disable call. That allows
CWB tear down to update the flush configuration
on primary ctl path.
Change-Id: I21c521b39456af4144cf836c65d46a25c985f51d
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
Below sequence can trigger encoder restore failure due
to cur_master unavailability.
-> Primary encoder is attached to crtc-0
-> Primary encoder enters in power collapse state
-> CWB encoder attached to crtc-0
-> Atomic_check updates the crtc encoder_mask with
two encoders.
-> Prepare_commit triggers power restore
-> crtc triggers encoder_restore based on encoder_mask
CWB encoder is not enabled at this stage and it will fail
to restore. This patch avoids the error message in this
valid condition.
Create a data structure to maintain available hardware resources
and track capabilities. This data structure is used to send
the current available resources and caps information to
connector ops get_mode_info, get_modes and validate_mode to
process the display mode.
Change-Id: If38fc628ee5ab4729821f88c0050ab45375187b8
Signed-off-by: Nilaan Gunabalachandran <ngunabal@codeaurora.org>
There is a race condition between sdm thread and display thread, when
sdm thread may be accessing the mode_info field while duplicating the
connector state, display thread may be updating the current mode info.
This would lead to the next commit using invalid mode and could cause
wrong hw configuration.
This change removes copying the mode_info in encoder struct while
atomic check phase and also removes accessing the connector mode_info
during mode_set. The encoder struct is now updated at the mode_set
with the cached connector mode.
Change-Id: I069ed592ec017ce4aa5c9c94b340bf94c5e1ebff
Signed-off-by: Abhijit Kulkarni <kabhijit@codeaurora.org>
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
When a dsc block gets disabled, it is not being
flushed properly because the kms_commit clears
the flush bits before the next vsync.
In a two commit scenario where the first commit
disables both dsc blocks within the same dsc
block pair, and where the second commit enables
only one of the dsc blocks in the pair, only the
second commit's dsc block gets flushed properly
because checks for dirty dsc blocks are currently
only happening if dsc is disabled for the
encoder. Since the first commit's disabled dsc
block was not flushed to hardware, there will be
errors such as pingpong timeouts and fifo
underruns, which can lead to distorted images on
the panel.
This problem manifests itself in suspend/resume
scenarios involving dp dsc connections which only
use one dsc block. During a suspend, the two dsc
blocks that were used by dsi are released due to
a null commit coming from userspace. While
resuming, depending on the mode, one of these
released dsc blocks will be alloted to dp,
leading to the problem described above.
CRs-Fixed: 2471336
Change-Id: Ifd71c980a04d63a87923eb0bece96646d4804533
Signed-off-by: Fuad Hossain <fhossain@codeaurora.org>
wr_ptr_irq timeout signifies that the MDP is stuck
on either the current or previous frame. Handle
ctl reset and fence signalling as part of this
timeout handling. This logic would help to recover
the HW faster in case of posted-start.
Change-Id: I09b3d21772df431f9fc4a58b2fd9b4fcac4a7de7
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
SDE RSC does not need to wait for vsync during
dms. It was removed unintentionally with commit
a74d2cf7fa ("disp: msm: add runtime_pm ops
support in drm driver"). However, Panel mode
switch still needs vsync wait based on
recommendation. This change fixes vsync wait
for both cases.
Change-Id: Ic9119132eb15a8c33f3841ba5df2624189d93395
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>
This change takes a snapshot from 4.14 to 4.19 as of
commit 47d149c31967 ("drm/msm/sde: Add null pointer
sanity checks").
Change-Id: Ib40428c562c3561c8a20d9849f16d13151496005
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
Fixes suspend-resume not working after switching dsi clk for
video mode. While switching the dsi clk, FLAG_SEAMLESS_DYN_CLK
is set leading to enable dsi clks which causes extra refcount.
Add check for command mode.
Change-Id: I814eb9c87daf387b5d57c5a3dddf7ae1e60fe784
Signed-off-by: Ritesh Kumar <riteshk@codeaurora.org>
Currently the dynamic bit clock switch trigger for command mode
is supported via sysfs node. This might lead to unnecessary
race conditions, when dsi driver is enabling the dsi bit clock
as part of commit and at the same time if bit rate change via
sysfs happens. So make the trigger happens via kernel mode set
call as done for video mode.
Change-Id: I17acb408d2b6dbd6fa41994e56262e31e43d088b
Signed-off-by: Ritesh Kumar <riteshk@codeaurora.org>
Signed-off-by: Yujun Zhang <yujunzhang@codeaurora.org>
This change adds support for dynamic switching of dsi clocks
to avoid RF interference issues. Also with dynamic dsi clock
switch feature coming into picture, now populate the supported
refresh rate as list instead of providing a range. Modify the
logic to enumerate all the modes in dsi driver, taking dynamic
bit clocks, resolutions and refresh rates into account.
Change-Id: I5b6e62bc935cf2234bdd96fcb3c7537b4e735fff
Signed-off-by: Sandeep Panda <spanda@codeaurora.org>
Signed-off-by: Ritesh Kumar <riteshk@codeaurora.org>
Signed-off-by: Yujun Zhang <yujunzhang@codeaurora.org>
This change ports the missing changes from 4.14 to 4.19
that were missed. It includes changes up until
commit 0f8fb25421ff ("cnss2: Add device version to
SOC info structure").
Change-Id: Idfdfe891f146e389e3c65cc3fc4c98d93220e789
Signed-off-by: Samantha Tran <samtran@codeaurora.org>
Fix resource and splash buffer handling in resource manager
and sde_kms to support continuous splash to be enabled/disabled
independently in multiple built-in display usecase.
Change-Id: I446ea9b08a794e2b053c37f55b31e51404bbcf71
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Currently dynamic mode-switch is allowed only after
the cont-splash handoff is handled during the first
frame. Remove this restriction for cmd-mode alone as
it can handle the use-case.
Change-Id: I5f9dc758f50a91fec0b9f710c74f2ea78c4e75eb
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
Primary and secondary displays should have first priority
when reserving lms. Static reservation can potentially block
higher resolutions for the required displays. This patch gets
the layer mixer requirement for primary or secondary display
if available. It reserves those layer mixers dynamically
for the respective display when connector is registered.
Change-Id: Id69dac4c72d6b20008049f4aeb71c0f97d0a426b
Signed-off-by: Nilaan Gunabalachandran <ngunabal@codeaurora.org>
SDE driver triggers the frame and waits for the
ctl_start interrupt for command mode display. This interrupt
provides confirmation that hardware has picked up the
frame. Retire fence signaling is associated with
this interrupt and it is sent at the rd_ptr interrupt
after ctl_start. Due to lut dma delay, ctl_start interrupt
may be trigger before rd_ptr or after rd_ptr. SW manages
this complexity and handle retire fence for different cases
with 500us threshold logic.
This change replaces the ctl_start interrupt with wr_ptr
interrupt by programming it to trigger at 1st write line
count. This is guaranteed to come every time and it is close
to rd_ptr interrupt. That allows retire fence trigger at
wr_ptr interrupt and simplifies the SW logic. CRTC commit
thread would be held slightly longer with this change
as the wr_ptr is always close to rd_ptr and after
ctl_start.
Change-Id: Ic47a8f82c854b4aded0d70c95af853b28a68ffd6
Signed-off-by: Veera Sundaram Sankaran <veeras@codeaurora.org>
