On VLV/CHV enabling sprite0 when sprite1 has already been enabled may
lead to an underrun. This only happens when sprite0 FIFO size is zero
prior to enabling it. Hence an effective workaround is to always
allocate at least one cacheline for sprite0 when sprite1 is active.
I've not observed this sort of failure during any other type of plane
enable/disable sequence.
v2: s/noninverted/raw/ for consistency with other platforms
Testcase: igt/kms_plane_blinker
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-14-ville.syrjala@linux.intel.com
Remove crtc->wm.cxsr_allowed and just rely on crtc_state->disable_cxsr
instead. This was used only by vlv/chv to indicate whether to enable
cxsr in the wm computation. That doesn't really work anymore, and as far
as the optimal watermarks go we'll just consider the number of planes
and the current pipe, and for the intermediate watermarks we'll also
start to consider disable_cxsr which is set appropriately when planes
are being enabled/disabled.
We'll also flip over the crtc_state->wm.need_postvbl_update setup so
that it's the wm code that will set it. Previously the generic code set
it up, and then the wm code cleared it again if it thought it's not
needed after all.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-11-ville.syrjala@linux.intel.com
Since the watermark registers arent double buffered on VLV/CHV, we'll
need to play around with intermediate watermarks same was as we do on
ILK-BDW.
The watermark registers on VLV/CHV contain inverted values, so to find
the intermediate watermark value we just take the minimum of the
active and optimal values. This also means that, unlike ILK-BDW,
there's no chance that we'd fail to find a working intermediate
watermarks. As long as both the active and optimal watermarks are valid
the intermediate watermarks will come out valid as well.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-10-ville.syrjala@linux.intel.com
Check whether anything relevant has actually change when we compute new
watermarks for each plane in the state. If the watermarks for no
primary/sprite planes changed we don't have to recompute the FIFO split
or reprogram the DSBARB registers. And even the cursor watermarks didn't
change we can skip the merge+invert step between all the planes on
the pipe as well.
v2: s/noninverted/raw/ for consistency with other platforms
v3: Drop duplicated vlv_get_fifo_size() call during init
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-9-ville.syrjala@linux.intel.com
Start computing the vlv/chv watermarks the atomic way, from the
.compute_pipe_wm() hook. We'll recompute the actual watermarks
for only planes that are part of the state, the other planes will
keep their watermark from the last time it was computed.
And the actual watermark programming will happen from the
.initial_watermarks() hook. For now we'll just compute the
optimal watermarks, and we'll hook up the intermediate
watermarks properly later.
The DSPARB registers responsible for the FIFO paritioning are
double buffered, so they will be programming from
intel_begin_crtc_commit().
v2: s/noninverted/raw/ for consistency with other platforms
s/vlv_plane_wm_set/vlv_raw_plane_wm_set/ for clarity
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-8-ville.syrjala@linux.intel.com
Let's compute the watermarks first and the FIFO size second. This way we
can make sure the FIFO split is the most accommodating to the watermarks.
Previously we could have potentially computed a FIFO split that couldn't
accommodate the PM2 watermarks simply due to a bad split even if the
total FIFO size would have been sufficient.
It'll also allow us to avoid recomputing the wms for all planes whenever
the FIFO split would change. Thus we don't have to add any extra planes
to the state when the FIFO needs to be repartitioned.
To help with this we'll keep around copies of the non-inverted
watermarks in the crtc state. For now that doesn't help too much, but
once we start to do the watermark computation only for the planes
that change we'll need the non-inverted values around for the other
planes.
v2: s/noninverted/raw/ for consistency with other platforms
Fix the memset() of the "raw" watermarks
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170302171508.1666-7-ville.syrjala@linux.intel.com
In order to make cursor updates actually safe wrt. watermark programming
we have to clear the legacy_cursor_update flag in the atomic state. That
will cause the regular atomic update path to do the necessary vblank
wait after the plane update if needed, otherwise the vblank wait would
be skipped and we'd feed the optimal watermarks to the hardware before
the plane update has actually happened.
To make the slow vs. fast path determination in
intel_legacy_cursor_update() a little simpler we can ignore the actual
visibility of the plane (which can only get computed once we've already
chosen out path) and instead we simply check whether the fb is being
set or cleared by the user. This means a fully clipped but logically
visible cursor will be considered visible as far as watermark
programming is concerned. We can do that for the cursor since it's a
fixed size plane and the clipped size doesn't play a role in the
watermark computation.
This should fix underruns that can occur when the cursor gets
enable/disabled or the size gets changed. Hopefully it's good enough
that only pure cursor movement and flips go through unthrottled.
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Uwe Kleine-König <uwe@kleine-koenig.org>
Reported-by: Uwe Kleine-König <uwe@kleine-koenig.org>
Fixes: f79f26921e ("drm/i915: Add a cursor hack to allow converting legacy page flip to atomic, v3.")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170217150159.11683-1-ville.syrjala@linux.intel.com
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Tested-by: Rafael Ristovski <rafael.ristovski@gmail.com>
Certain Baytrails, namely the 4 cpu core variants, have been
plaqued by spurious system hangs, mostly occurring with light loads.
Multiple bisects by various people point to a commit which changes the
reclocking strategy for Baytrail to follow its bigger brethen:
commit 8fb55197e6 ("drm/i915: Agressive downclocking on Baytrail")
There is also a review comment attached to this commit from Deepak S
on avoiding punit access on Cherryview and thus it was excluded on
common reclocking path. By taking the same approach and omitting
the punit access by not tweaking the thresholds when the hardware
has been asked to move into different frequency, considerable gains
in stability have been observed.
With J1900 box, light render/video load would end up in system hang
in usually less than 12 hours. With this patch applied, the cumulative
uptime has now been 34 days without issues. To provoke system hang,
light loads on both render and bsd engines in parallel have been used:
glxgears >/dev/null 2>/dev/null &
mpv --vo=vaapi --hwdec=vaapi --loop=inf vid.mp4
So far, author has not witnessed system hang with above load
and this patch applied. Reports from the tenacious people at
kernel bugzilla are also promising.
Considering that the punit access frequency with this patch is
considerably less, there is a possibility that this will push
the, still unknown, root cause past the triggering point on most loads.
But as we now can reliably reproduce the hang independently,
we can reduce the pain that users are having and use a
static thresholds until a root cause is found.
v3: don't break debugfs and simplification (Chris Wilson)
References: https://bugzilla.kernel.org/show_bug.cgi?id=109051
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: fritsch@xbmc.org
Cc: miku@iki.fi
Cc: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
CC: Michal Feix <michal@feix.cz>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Deepak S <deepak.s@linux.intel.com>
Cc: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Cc: <stable@vger.kernel.org> # v4.2+
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1487166779-26945-1-git-send-email-mika.kuoppala@intel.com
When the RPS tuning was applied to Baytrail, in commit 8fb55197e6
("drm/i915: Agressive downclocking on Baytrail"), concern was given that
it might cause Cherryview excess wakeups of the common power well.
However, the static thresholds perform poorly for Kodi, and the GPU is
unable to deliver the video frames on time. Enabling the dynamic, finer
thresholds used on all other platforms (including Skylake and Broxton
that also have the same multiple powerwell concerns) allows the GPU to
pick a more appropriate frequency and not drop frames.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170210150348.22146-1-chris@chris-wilson.co.uk
Reviewed-by: Radoslaw Szwichtenberg <radoslaw.szwichtenberg@intel.com>
The current dev_cdclk vs. cdclk vs. atomic_cdclk_freq is quite a mess.
So here I'm introducing the "actual" and "logical" naming for our
cdclk state. "actual" is what we'll bash into the hardware and "logical"
is what everyone should use for state computaion/checking and whatnot.
We'll track both using the intel_cdclk_state as both will need other
differing parameters than just the actual cdclk frequency.
While doing that we can at the same time unify the appearance of the
.modeset_calc_cdclk() implementations a little bit.
v2: Commit dev_priv->cdclk.actual since that already has the
new state by the time .modeset_commit_cdclk() is called.
v3: s/locical/logical/ and improve the docs a bit
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170120182205.8141-9-ville.syrjala@linux.intel.com
Along with GLK it was introduced the .is_lp and IS_GEN9_LP.
So, following the same simplification standard we can
put Skylake and Kabylake under the same bucket for most
of the things.
So let's add the IS_GEN9_BC for "Big Core" (non Atom based
platforms).
The i915_drv.c was let out of this patch on purpose
because that is really a decision per platform, just like
other cases where IS_KABYLAKE is different from IS_SKYLAKE.
v2: fix conflict with IS_LP and 3 new cases for this
big core bucket:
- intel_ddi.c: intel_ddi_get_link_dpll
- intel_fbc.c: find_compression_threshold
- i915_gem_gtt.c: gtt_write_workarounds
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1485196357-30599-2-git-send-email-rodrigo.vivi@intel.com
Directly merge drm-misc into drm-intel since Dave is on vacation and
we need the various drm-misc patches (fb format rework, drm mm fixes,
selftest framework and others). Also pulled back -rc2 in first to
resync with drm-intel-fixes and make sure I can reuse the exact rerere
solutions from drm-tip for safety, and because I'm lazy.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
BSpec says:
"Overlay Clock Gating Must be Disabled: Overlay & L2 Cache clock gating
must be disabled in order to prevent device hangs when turning off overlay.SW
must turn off Ovrunit clock gating (6200h) and L2 Cache clock gating (C8h)."
We only turned off the overlay clock gating (due to lack of docs I
presume). After a bit of experimentation it looks like the the magic
C8h register lives in the PCI config space of device 0, and the magic
bit appears to be bit 2. Or at the very least this eliminates the GPU
death after MI_OVERLAY_OFF.
L2 clock gating seems to save ~80mW, so let's keep it on unless we need
to actually use the overlay.
Also let's move the OVRUNIT clock gating to the same place since we can,
and 845 supposedly doesn't need it.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1481131693-27993-11-git-send-email-ville.syrjala@linux.intel.com
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
The requests conversion introduced a nasty bug where we could generate a
new request in the middle of constructing a request if we needed to idle
the system in order to evict space for a context. The request to idle
would be executed (and waited upon) before the current one, creating a
minor havoc in the seqno accounting, as we will consider the current
request to already be completed (prior to deferred seqno assignment) but
ring->last_retired_head would have been updated and still could allow
us to overwrite the current request before execution.
We also employed two different mechanisms to track the active context
until it was switched out. The legacy method allowed for waiting upon an
active context (it could forcibly evict any vma, including context's),
but the execlists method took a step backwards by pinning the vma for
the entire active lifespan of the context (the only way to evict was to
idle the entire GPU, not individual contexts). However, to circumvent
the tricky issue of locking (i.e. we cannot take struct_mutex at the
time of i915_gem_request_submit(), where we would want to move the
previous context onto the active tracker and unpin it), we take the
execlists approach and keep the contexts pinned until retirement.
The benefit of the execlists approach, more important for execlists than
legacy, was the reduction in work in pinning the context for each
request - as the context was kept pinned until idle, it could short
circuit the pinning for all active contexts.
We introduce new engine vfuncs to pin and unpin the context
respectively. The context is pinned at the start of the request, and
only unpinned when the following request is retired (this ensures that
the context is idle and coherent in main memory before we unpin it). We
move the engine->last_context tracking into the retirement itself
(rather than during request submission) in order to allow the submission
to be reordered or unwound without undue difficultly.
And finally an ulterior motive for unifying context handling was to
prepare for mock requests.
v2: Rename to last_retired_context, split out legacy_context tracking
for MI_SET_CONTEXT.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161218153724.8439-3-chris@chris-wilson.co.uk
commit 848496e590
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date: Wed Jul 13 16:32:03 2016 +0300
drm/i915: Wait up to 3ms for the pcu to ack the cdclk change request on SKL
increased the timeout to match the spec, but we still see a timeout on
at least one SKL. A CDCLK change request following the failed one will
succeed nevertheless.
I could reproduce this problem easily by running kms_pipe_crc_basic in a
loop. In all failure cases _wait_for() was pre-empted for >3ms and so in
the worst case - when the pre-emption happened right after calculating
timeout__ in _wait_for() - we called skl_cdclk_wait_for_pcu_ready() only
once which failed and so _wait_for() timed out. As opposed to this the
spec says to keep retrying the request for at most a 3ms period.
To fix this send the first request explicitly to guarantee that there is
3ms between the first and last request. Though this matches the spec, I
noticed that in rare cases this can still time out if we sent only a few
requests (in the worst case 2) _and_ PCODE is busy for some reason even
after a previous request and a 3ms delay. To work around this retry the
polling with pre-emption disabled to maximize the number of requests.
Also increase the timeout to 10ms to account for interrupts that could
reduce the number of requests. With this change I couldn't trigger
the problem.
v2:
- Use 1ms poll period instead of 10us. (Chris)
v3:
- Poll with pre-emption disabled to increase the number of request
attempts. (Ville, Chris)
- Factor out a helper to poll, it's also needed by the next patch.
v4:
- Pass reply_mask, reply to skl_pcode_request(), instead of assuming the
reply is generic. (Ville)
v5:
- List the request specific timeout values as code comment. (Ville)
v6:
- Try the poll first with preemption enabled.
- Add code comment about first request being queued by PCODE. (Art)
- Add timeout_base_ms argument. (Ville)
v7:
- Clarify code comment about first queued request. (Chris)
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Art Runyan <arthur.j.runyan@intel.com>
Cc: <stable@vger.kernel.org> # v4.2- : 3b2c171 : drm/i915: Wait up to 3ms
Cc: <stable@vger.kernel.org> # v4.2-
Fixes: 5d96d8afcf ("drm/i915/skl: Deinit/init the display at suspend/resume")
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=97929
Testcase: igt/kms_pipe_crc_basic/suspend-read-crc-pipe-B
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1480955258-26311-1-git-send-email-imre.deak@intel.com
first set of fixes for -next.
* tag 'drm-intel-next-fixes-2016-12-07' of git://anongit.freedesktop.org/git/drm-intel:
drm/i915: Move priority bumping for flips earlier
drm/i915: Hold a reference on the request for its fence chain
drm/i915/audio: fix hdmi audio noise issue
drm/i915/debugfs: Increment return value of gt.next_seqno
drm/i915/debugfs: Drop i915_hws_info
drm/i915: Initialize dev_priv->atomic_cdclk_freq at init time
drm/i915: Fix cdclk vs. dev_cdclk mess when not recomputing things
drm/i915: Make skl_write_{plane,cursor}_wm() static
drm/i915: Complete requests in nop_submit_request
drm/i915/gvt: fix lock not released bug for dispatch_workload() err path
drm/i915/gvt: fix getting 64bit bar size error
drm/i915/gvt: fix missing init param.primary
This patch changes Watermak calculation to fixed point calculation.
Problem with current calculation is during plane_blocks_per_line
calculation we divide intermediate blocks with min_scanlines and
takes floor of the result because of integer operation.
hence we end-up assigning less blocks than required. Which leads to
flickers.
Changes since V1:
- Add fixed point data type as per Paulo's review
Changes since V2:
- use fixed_point instead of fp_16_16
Changes since V3:
- rebase
Changes since V4 (from Paulo):
- My original renaming suggestion was misunderstood, so implement it
- Simplify fixed_16_16_to_u32 implementation
- Fix indentation
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Mahesh Kumar <mahesh1.kumar@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161201154940.24446-6-mahesh1.kumar@intel.com
