Reduce the 3 relocation paths down to the single path that accommodates
all. The primary motivation for this is to guard the relocations with a
natural fence (derived from the i915_request used to write the
relocation from the GPU).
The tradeoff in using async gpu relocations is that it increases latency
over using direct CPU relocations, for the cases where the target is
idle and accessible by the CPU. The benefit is greatly reduced lock
contention and improved concurrency by pipelining.
Note that forcing the async gpu relocations does reveal a few issues
they have. Firstly, is that they are visible as writes to gem_busy,
causing to mark some buffers are being to written to by the GPU even
though userspace only reads. Secondly is that, in combination with the
cmdparser, they can cause priority inversions. This should be the case
where the work is being put into a common workqueue losing our priority
information and so being executed in FIFO from the worker, denying us
the opportunity to reorder the requests afterwards.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200604211457.19696-1-chris@chris-wilson.co.uk
There are a couple places in our driver that loop over transcoders A..D
for gen11+; since RKL only has three pipes/transcoders, this can lead to
unclaimed register reads/writes. We should add checks for transcoder
existence where appropriate.
v2: Move one transcoder check that wound up in the wrong function after
conflict resolution. It belongs in bdw_get_trans_port_sync_config
rather than bxt_get_dsi_transcoder_state.
v3: Switch loops to use for_each_cpu_transcoder_masked() since this
iterator already checks the platform's transcoder mask for us.
(Ville)
Cc: Aditya Swarup <aditya.swarup@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200603211529.3005059-10-matthew.d.roper@intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Previous patch didn't take into account all pipes
but only those in state, which could cause wrong
CDCLK conclcusions and calculations.
Also there was a severe issue with min_cdclk being
assigned to 0 every compare cycle.
Too bad this was found by me only after merge.
This could be also causing the issues in test, however
not clear - anyway marking this as fixing the
"Adjust CDCLK accordingly to our DBuf bw needs".
v2: - s/pipe/crtc->pipe/
- save a bit of instructions by
skipping inactive pipes, without
getting 0 DBuf slice mask for it.
Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Fixes: cd19154608 ("drm/i915: Adjust CDCLK accordingly to our DBuf bw needs")
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200601173058.5084-1-stanislav.lisovskiy@intel.com
RKL uses the DDI A, DDI B, DDI USBC1, DDI USBC2 from the DE point of
view, so all DDI/pipe/transcoder register use these indexes to refer to
them. Combo phy and IO functions follow another namespace that we keep
as "enum phy". The VBT in theory would use the DE point of view, but
that does not happen in practice.
Provide a table to convert the child devices to the "correct" port
numbering we use. Now this is the output we get while reading the VBT:
DDIA:
[drm:intel_bios_port_aux_ch [i915]] using AUX A for port A (VBT)
[drm:intel_dp_init_connector [i915]] Adding DP connector on [ENCODER:275:DDI A]
[drm:intel_hdmi_init_connector [i915]] Adding HDMI connector on [ENCODER:275:DDI A]
[drm:intel_hdmi_init_connector [i915]] Using DDC pin 0x1 for port A (VBT)
DDIB:
[drm:intel_bios_port_aux_ch [i915]] using AUX B for port B (platform default)
[drm:intel_hdmi_init_connector [i915]] Adding HDMI connector on [ENCODER:291:DDI B]
[drm:intel_hdmi_init_connector [i915]] Using DDC pin 0x2 for port B (VBT)
DDI USBC1:
[drm:intel_bios_port_aux_ch [i915]] using AUX D for port D (VBT)
[drm:intel_dp_init_connector [i915]] Adding DP connector on [ENCODER:295:DDI D]
[drm:intel_hdmi_init_connector [i915]] Adding HDMI connector on [ENCODER:295:DDI D]
[drm:intel_hdmi_init_connector [i915]] Using DDC pin 0x3 for port D (VBT)
DDI USBC2:
[drm:intel_bios_port_aux_ch [i915]] using AUX E for port E (VBT)
[drm:intel_dp_init_connector [i915]] Adding DP connector on [ENCODER:306:DDI E]
[drm:intel_hdmi_init_connector [i915]] Adding HDMI connector on [ENCODER:306:DDI E]
[drm:intel_hdmi_init_connector [i915]] Using DDC pin 0x9 for port E (VBT)
Cc: Clinton Taylor <Clinton.A.Taylor@intel.com>
Cc: Aditya Swarup <aditya.swarup@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
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/20200603211529.3005059-7-matthew.d.roper@intel.com
If the execbuf is interrupted after building the cmdparser pipeline, and
before we commit to submitting the request to HW, we would attempt to
clean up the cmdparser early. While we held active references to the vma
being parsed and constructed, we did not hold an active reference for
the buffer pool itself. The result was that an interrupted execbuf could
still have run the cmdparser pipeline, but since the buffer pool was
idle, its target vma could have been recycled.
Note this problem only occurs if the cmdparser is running async due to
pipelined waits on busy fences, and the execbuf is interrupted.
Fixes: 686c7c35ab ("drm/i915/gem: Asynchronous cmdparser")
Fixes: 16e8745967 ("drm/i915/gt: Move the batch buffer pool from the engine to the gt")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200604103751.18816-1-chris@chris-wilson.co.uk
The DP spec says:
"When the combination of the requested pre-emphasis level and
voltage swing exceeds the capability of a DPTX, the DPTX shall
set the pre-emphasis level according to the request and use the
highest voltage swing it can output with the given pre-emphasis level."
and
"When a DPTX reads a request beyond the limits of this Standard,
the DPTX shall set the pre-emphasis level according to the request
and set the highest voltage swing level it can output with the
given pre-emphasis level. If a DPTX is requested for 9.5dB of
pre-emphasis level (may be supported for a DPTX) and cannot support
that level, it shall set the pre-emphasis level to the next
highest level, 6dB."
Ie. we should first validate the pre-emphasis, and then select
the appropriate vswing for it.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200512174145.3186-6-ville.syrjala@linux.intel.com
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Ever noticed that our interrupt handlers are where we spend most of our
time on a busy system? In part this is unavoidable as each interrupt
requires to poll and reset several registers, but we can try and do so as
efficiently as possible.
Function old new delta
ilk_irq_handler 2317 2156 -161
v2: Restore the irqreturn_t ret
Function old new delta
ilk_irq_handler.cold 63 72 +9
ilk_irq_handler 2221 2080 -141
A slight improvement in the baseline overnight as well!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200601140355.20243-1-chris@chris-wilson.co.uk
While the current locking/serialization of the global state
suffices for protecting the obj->state access and the actual
hardware reprogramming, we do have a problem with accessing
the old/new states during nonblocking commits.
The state computation and swap will be protected by the crtc
locks, but the commit_tails can finish out of order, thus also
causing the atomic states to be cleaned up out of order. This
would mean the commit that started first but finished last has
had its new state freed as the no-longer-needed old state by the
other commit.
To fix this let's just refcount the states. obj->state amounts
to one reference, and the intel_atomic_state holds extra references
to both its new and old global obj states.
Fixes: 0ef1905ecf ("drm/i915: Introduce better global state handling")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200527200245.13184-1-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Our forcewake utilisation is split into categories: automatic and
manual. Around bare register reads, we look up the right forcewake
domain and automatically acquire and release [upon a timer] the
forcewake domain. For other access, where we know we require the
forcewake across a group of register reads, we manually acquire the
forcewake domain and release it at the end. Again, this currently arms
the domain timer for a later release.
However, looking at some energy utilisation profiles, we have tried to
avoid using forcewake [and rely on the natural wake up to post register
updates] due to that even keep the fw active for a brief period
contributes to a significant power draw [i.e. when the gpu is sleeping
with rc6 at high clocks]. But as it turns out, not posting the writes
immediately also has unintended consequences, such as not reducing the
clocks and so conserving power while busy.
As a compromise, let us only arm the domain timer for automatic
forcewake usage around bare register access, but immediately release the
forcewake when manually acquired by intel_uncore_forcewake_get/_put.
The corollary to this is that we may instead have to take forcewake more
often, and so incur a latency penalty in doing so. For Sandybridge this
was significant, and even on the latest machines, taking forcewake at
interrupt frequency is a huge impact. [So we don't do that anymore!
Hopefully, this will spare us from still needing the mitigation of the
timer for steady state execution.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200601072446.19548-13-chris@chris-wilson.co.uk
Currently the plane property doesn't have support for YCBCR_BT2020,
which enables the corresponding color conversion mode on plane CSC.
Enabling the plane property for the planes for GLK & ICL+ platforms.
Also as per spec, update the Plane Color CSC from YUV601_TO_RGB709
to YUV601_TO_RGB601.
V2: Enabling support for YCBCT_BT2020 for HDR planes on
platforms GLK & ICL
V3: Refined the condition check to handle GLK & ICL+ HDR planes
Also added BT2020 handling in glk_plane_color_ctl.
V4: Combine If-else into single If
V5: Drop the checking for HDR planes and enable YCBCR_BT2020
for platforms GLK & ICL+.
V6: As per Spec, update PLANE_COLOR_CSC_MODE_YUV601_TO_RGB709
to PLANE_COLOR_CSC_MODE_YUV601_TO_RGB601 as per Ville's
feedback.
V7: Rebased
Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Signed-off-by: Kishore Kadiyala <kishore.kadiyala@intel.com>
Signed-off-by: Uma Shankar <uma.shankar@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200601073544.11291-1-kishore.kadiyala@intel.com
With the advent of preempt-to-busy, a request may still be on the GPU as
we unwind. And in the case of a unpreemptible [due to HW] request, that
request will remain indefinitely on the GPU even though we have
returned it back to our submission queue, and cleared the active bit.
We only run the execution callbacks on transferring the request from our
submission queue to the execution queue, but if this is a bonded request
that the HW is waiting for, we will not submit it (as we wait for a
fresh execution) even though it is still being executed.
As we know that there are always preemption points between requests, we
know that only the currently executing request may be still active even
though we have cleared the flag. However, we do not precisely know which
request is in ELSP[0] due to a delay in processing events, and
furthermore we only store the last request in a context in our state
tracker.
Fixes: 22b7a426bb ("drm/i915/execlists: Preempt-to-busy")
Testcase: igt/gem_exec_balancer/bonded-dual
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/20200529143926.3245-1-chris@chris-wilson.co.uk
Replace the use of mode->private_flags with a truly private bitmaks
in our own crtc state. We also need a copy in the crtc itself so the
vblank code can get at it. We already have scanline_offset in there
for a similar reason, as well as the vblank->hwmode which is assigned
via drm_calc_timestamping_constants(). Fortunately we now have a
nice place for doing the crtc_state->crtc copy in
intel_crtc_update_active_timings() which gets called both for
modesets and init/resume readout.
The one slightly iffy spot is the INHERITED flag which we want to
preserve until userspace/fb_helper does the first proper commit after
actually calling .detecti() on the connectors. Otherwise we don't have
the full sink capabilities (audio,infoframes,etc.) when .compute_config()
gets called and thus we will fail to enable those features when the
first userspace commit happens. The only internal commit we do prior to
that should be from intel_initial_commit() and there we can simply
preserve the INHERITED flag from the readout.
v2: Deal with INHERITED in sanitize_watermarks() as well
CC: Sam Ravnborg <sam@ravnborg.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200429103904.11727-1-ville.syrjala@linux.intel.com
