Due to the nature of preempt-to-busy the execlists active tracking and
the schedule queue may become temporarily desync'ed (between resubmission
to HW and its ack from HW). This means that we may have unwound a
request and passed it back to the virtual engine, but it is still
inflight on the HW and may even result in a GPU hang. If we detect that
GPU hang and try to reset, the hanging request->engine will no longer
match the current engine, which means that the request is not on the
execlists active list and we should not try to find an older incomplete
request. Given that we have deduced this must be a request on a virtual
engine, it is the single active request in the context and so must be
guilty (as the context is still inflight, it is prevented from being
executed on another engine as we process the reset).
Fixes: 22b7a426bb ("drm/i915/execlists: Preempt-to-busy")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190923152844.8914-2-chris@chris-wilson.co.uk
(cherry picked from commit cb2377a919)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
If we are asked to submit a completed request, just move it onto the
active-list without modifying it's payload. If we try to emit the
modified payload of a completed request, we risk racing with the
ring->head update during retirement which may advance the head past our
breadcrumb and so we generate a warning for the emission being behind
the RING_HEAD.
v2: Commentary for the sneaky, shared responsibility between functions.
v3: Spelling mistakes and bonus assertion
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190923110056.15176-3-chris@chris-wilson.co.uk
(cherry picked from commit c0bb487dc1)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
I'm embarassed to say that even though I've touched
vop_crtc_mode_fixup() twice and I swear I tested it, there's still a
stupid glaring bug in it. Specifically, on veyron_minnie (with all
the latest display timings) we want to be setting our pixel clock to
66,666,666.67 Hz and we tell userspace that's what we set, but we're
actually choosing 66,000,000 Hz. This is confirmed by looking at the
clock tree.
The problem is that in drm_display_mode_from_videomode() we convert
from Hz to kHz with:
dmode->clock = vm->pixelclock / 1000;
...and drm_display_mode_from_videomode() is called from panel-simple
when we have an "override_mode" like we do on veyron_minnie. See
commit 123643e5c4 ("ARM: dts: rockchip: Specify
rk3288-veyron-minnie's display timings").
...so when the device tree specifies a clock of 66666667 for the panel
then DRM translates that to 66666000. The clock framework will always
pick a clock that is _lower_ than the one requested, so it will refuse
to pick 66666667 and we'll end up at 66000000.
While we could try to fix drm_display_mode_from_videomode() to round
to the nearest kHz and it would fix our problem, it wouldn't help if
the clock we actually needed was 60,000,001 Hz. We could
alternatively have DRM always round up, but maybe this would break
someone else who already baked in the assumption that DRM rounds down.
Specifically note that clock drivers are not consistent about whether
they round up or round down when you call clk_set_rate(). We know how
Rockchip's clock driver works, but (for instance) you can see that on
most Qualcomm clocks the default is clk_rcg2_ops which rounds up.
Let's solve this by just adding 999 Hz before calling
clk_round_rate(). This should be safe and work everywhere. As
discussed in more detail in comments in the commit, Rockchip's PLLs
are configured in a way that there shouldn't be another PLL setting
that is only a few kHz off so we won't get mixed up.
NOTE: if this is picked to stable, it's probably easiest to first pick
commit 527e4ca3b6 ("drm/rockchip: Base adjustments of the mode based
on prev adjustments") which shouldn't hurt in stable.
Fixes: b59b8de314 ("drm/rockchip: return a true clock rate to adjusted_mode")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20191003114726.v2.1.Ib233b3e706cf6317858384264d5b0ed35657456e@changeid
Panfrost uses multiple schedulers (one for each slot, so 2 in reality),
and on a timeout has to stop all the schedulers to safely perform a
reset. However more than one scheduler can trigger a timeout at the same
time. This race condition results in jobs being freed while they are
still in use.
When stopping other slots use cancel_delayed_work_sync() to ensure that
any timeout started for that slot has completed. Also use
mutex_trylock() to obtain reset_lock. This means that only one thread
attempts the reset, the other threads will simply complete without doing
anything (the first thread will wait for this in the call to
cancel_delayed_work_sync()).
While we're here and since the function is already dependent on
sched_job not being NULL, let's remove the unnecessary checks.
Fixes: aa20236784 ("drm/panfrost: Prevent concurrent resets")
Tested-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20191009094456.9704-1-steven.price@arm.com
Currently the property docs don't specify whether it's okay for two planes to
have the same zpos value and what user-space should expect in this case.
The unspoken, legacy rule used in the past was to make user-space figure
out the zpos from object IDs. However some drivers break this rule,
that's why the ordering is documented as unspecified in case the zpos
property is missing. User-space should rely on the zpos property only.
There are some cases in which user-space might read identical zpos
values for different planes.
For instance, in case the property is mutable, user-space might set two
planes' zpos to the same value. This is necessary to support user-space
using the legacy DRM API where atomic commits are not possible:
user-space needs to update the planes' zpos one by one.
Because of this, user-space should handle multiple planes with the same
zpos.
While at it, remove the assumption that zpos is only for overlay planes.
Additionally, update the drm_plane_state.zpos docs to clarify that zpos
disambiguation via plane object IDs is a recommendation for drivers, not
something user-space can rely on. In other words, when user-space sets
the same zpos on two planes, drivers should rely on the plane object ID.
v2: clarify drm_plane_state.zpos docs (Daniel)
v3: zpos is for all planes (Marius, Daniel)
v4: completely reword the drm_plane_state.zpos docs to make it clear the
recommendation to use plane IDs is for drivers in case user-space uses
duplicate zpos values (Pekka)
v5: reword commit message (Pekka, James)
v6: remove mention of Arm GPUs having planes which can't overlap,
because this isn't uAPI yet (Daniel)
Signed-off-by: Simon Ser <contact@emersion.fr>
Reviewed-by: Pekka Paalanen <ppaalanen@gmail.com>
Cc: Marius Vlad <marius.vlad@collabora.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: James Qian Wang <james.qian.wang@arm.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/T5nHrvXH0GKOp6ONaFHk-j2cwEb4_4C_sBz9rNw8mmPACuut-DQqC74HMAFKZH3_Q15E8a3YnmKCxap-djKA71VVZv_T-tFxaB0he13O7yA=@emersion.fr
Polaris and vegam use count for the value rather than
level. This looks like a copy paste typo from when
the code was adapted from previous asics.
I'm not sure that the SMU actually uses this value, so
I don't know that it actually is a bug per se.
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=108609
Reported-by: Robert Strube <rstrube@gmail.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
cleanup error handling code and make sure temporary info array
with the handles are freed by amdgpu_bo_list_put() on
idr_replace()'s failure.
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Premature gamma lut prepration and loading which was getting
reflected in first modeset causing different colors on
screen during boot.
Issue: In BIOS, gamma is disabled by default. However, legacy read_luts()
was setting crtc_state->base.gamma_lut and gamma_lut was programmed
with junk values which led to visual artifacts (different
colored screens instead of usual black during boot).
Fix: Calling read_luts() only when gamma is enabled which will happen
after first modeset.
This fix is independent from the revert 1b8588741f ("Revert
"drm/i915/color: Extract icl_read_luts()"") and should fix different colors
on screen in legacy platforms too.
v2:
-Added gamma_enable checks inside read_luts() [Ville/Jani N]
-Corrected gamma enable check for CHV [Ville]
v3:
-Added check in ilk_read_luts() [Ville]
-Simplified gamma enable check for CHV [Ville]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111809
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111885
Tested-by: Jani Saarinen <jani.saarinen@intel.com>
Signed-off-by: Swati Sharma <swati2.sharma@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191009065542.27415-2-swati2.sharma@intel.com
Refer to EDID 1.3 spec, display FEATURE (byte 18h) bit #0 said
"If this bit is set to 1, the display supports timings based on the
GTF standard using default GTF parameter values".
And EDID 1.4 spec shows "If bit 0 is set to 0, then the display
is noncontinuous frequency (multi-mode) and is only specified to accept
the video timing formats that are listed in BASE EDID and certain
EXTENSION Blocks.
When display feature did not support CVT or GFT2 and monitor's EDID version
greater than or equal to "1.2". DRM driver would select GTF as default
for standard timing calculation. It may generated some video timing
that can't display properly by external monitor.
For example. When driver retrieved "0xD1 0xFC" (FHD, 120Hz) and
"0xD1 0xE8" (FHD, 100Hz) from "Standard Timings". GTF formula
would generate video timing like below. It already over monitor's
spec to cause black screen issue.
"1920x1080" 120 368881 1920 2072 2288 2656 1080 1081 1084 1157 0x0 0x6
"1920x1080" 100 301992 1920 2072 2280 2640 1080 1081 1084 1144 0x0 0x6
v2: Just confirm GTF flag and omit the revision check.
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Adam Jackson <ajax@redhat.com>
Cc: Cooper Chiou <cooper.chiou@intel.com>
Signed-off-by: Lee Shawn C <shawn.c.lee@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191007135127.9538-1-shawn.c.lee@intel.com
except mutex mod, mutex mod reg,mutex sof reg,
and mutex sof id will be ddp private data
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
This patch add ovl0/ovl_2l0 usecase
in ovl->ovl_2l0 direct link usecase:
1. the crtc support layer number will 4+2
2. ovl_2l0 background color input select ovl0 when crtc init
and disable it when crtc finish
3. config ovl_2l0 layer, if crtc config layer number is
bigger than ovl0 support layers(max is 4)
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
This patch add background color input select function for ovl/ovl_2l
ovl include 4 DRAM layer and 1 background color layer
ovl_2l include 4 DRAM layer and 1 background color layer
DRAM layer frame buffer data from render hardware, GPU for example.
backgournd color layer is embed in ovl/ovl_2l, we can only set
it color, but not support DRAM frame buffer.
for ovl0->ovl0_2l direct link usecase,
we need set ovl0_2l background color intput select from ovl0
if render send DRAM buffer layer number <=4, all these layer read
by ovl.
layer0 is at the bottom of all layers.
layer3 is at the top of all layers.
if render send DRAM buffer layer numbfer >=4 && <=6
ovl0 read layer0~3
ovl0_2l read layer4~5
layer5 is at the top ot all these layers.
the decision of how to setting ovl0/ovl0_2l read these layer data
is controlled in mtk crtc, which will be another patch
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
This patch add function to background color input select for ovl/ovl_2l
direct link for ovl/ovl_2l direct link usecase, we need set background
color input select for these hardware. This is preparation patch for
ovl/ovl_2l usecase.
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
This patch add layer_nr for ovl private data
ovl_2l almost same with with ovl hardware, except the
layer number for ovl_2l is 2 and ovl is 4.
this patch is a preparation for ovl-2l and
ovl share the same driver.
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
This patch add gmc_bits for ovl private data
GMC register was set RDMA ultra and pre-ultra threshold.
10bit GMC register define is different with other SOC, gmc_thrshd_l not
used.
Signed-off-by: Yongqiang Niu <yongqiang.niu@mediatek.com>
Signed-off-by: CK Hu <ck.hu@mediatek.com>
For historical reasons, the function drm_wait_vblank_ioctl always return
-EINVAL if something gets wrong. This scenario limits the flexibility
for the userspace to make detailed verification of any problem and take
some action. In particular, the validation of “if (!dev->irq_enabled)”
in the drm_wait_vblank_ioctl is responsible for checking if the driver
support vblank or not. If the driver does not support VBlank, the
function drm_wait_vblank_ioctl returns EINVAL, which does not represent
the real issue; this patch changes this behavior by return EOPNOTSUPP.
Additionally, drm_crtc_get_sequence_ioctl and
drm_crtc_queue_sequence_ioctl, also returns EINVAL if vblank is not
supported; this patch also changes the return value to EOPNOTSUPP in
these functions. Lastly, these functions are invoked by libdrm, which is
used by many compositors; because of this, it is important to check if
this change breaks any compositor. In this sense, the following projects
were examined:
* Drm-hwcomposer
* Kwin
* Sway
* Wlroots
* Wayland
* Weston
* Mutter
* Xorg (67 different drivers)
For each repository the verification happened in three steps:
* Update the main branch
* Look for any occurrence of "drmCrtcQueueSequence",
"drmCrtcGetSequence", and "drmWaitVBlank" with the command git grep -n
"STRING".
* Look in the git history of the project with the command
git log -S<STRING>
None of the above projects validate the use of EINVAL when using
drmWaitVBlank(), which make safe, at least for these projects, to change
the return values. On the other hand, mesa and xserver project uses
drmCrtcQueueSequence() and drmCrtcGetSequence(); this change is harmless
for both projects.
Change since V5 (Pekka Paalanen):
- Check if the change also affects Mutter
Change since V4 (Daniel):
- Also return EOPNOTSUPP in drm_crtc_[get|queue]_sequence_ioctl
Change since V3:
- Return EINVAL for _DRM_VBLANK_SIGNAL (Daniel)
Change since V2:
Daniel Vetter and Chris Wilson
- Replace ENOTTY by EOPNOTSUPP
- Return EINVAL if the parameters are wrong
Cc: Keith Packard <keithp@keithp.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Pekka Paalanen <pekka.paalanen@collabora.com>
Signed-off-by: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Acked-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191002140516.adeyj3htylimmlmg@smtp.gmail.com
Configure the display Quality of service (QoS) levels priority if the
optional property node "arm,malidp-aqros-value" is defined in DTS file.
QoS signaling using AQROS and AWQOS AXI interface signals, the AQROS is
driven from the "RQOS" register, so needed to program the RQOS register
to avoid the high resolutions flicker issue on the LS1028A platform.
Signed-off-by: Wen He <wen.he_1@nxp.com>
Signed-off-by: Liviu Dudau <liviu.dudau@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190910075913.17650-2-wen.he_1@nxp.com
