Adding an extra MI_STORE_DWORD_IMM to the gpu relocation path for gen3
was good, but still not good enough. To survive 24+ hours under test we
needed to perform not one, not two but three extra store-dw. Doing so
for each GPU relocation was a little unsightly and since we need to
worry about userspace hitting the same issues, we should apply the dummy
store-dw into the EMIT_FLUSH.
Fixes: 7dd4f6729f ("drm/i915: Async GPU relocation processing")
References: 7fa28e146994 ("drm/i915: Write GPU relocs harder with gen3")
Testcase: igt/gem_tiled_fence_blits # blb/pnv
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181207134037.11848-1-chris@chris-wilson.co.uk
(cherry picked from commit a889580c087a9cf91fddb3832ece284174214183)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
We stopped re-applying the GT workarounds after engine reset since commit
59b449d5c8 ("drm/i915: Split out functions for different kinds of
workarounds").
Issue with this is that some of the GT workarounds live in the MMIO space
which gets lost during engine resets. So far the registers in 0x2xxx and
0xbxxx address range have been identified to be affected.
This losing of applied workarounds has obvious negative effects and can
even lead to hard system hangs (see the linked Bugzilla).
Rather than just restoring this re-application, because we have also
observed that it is not safe to just re-write all GT workarounds after
engine resets (GPU might be live and weird hardware states can happen),
we introduce a new class of per-engine workarounds and move only the
affected GT workarounds over.
Using the framework introduced in the previous patch, we therefore after
engine reset, re-apply only the workarounds living in the affected MMIO
address ranges.
v2:
* Move Wa_1406609255:icl to engine workarounds as well.
* Rename API. (Chris Wilson)
* Drop redundant IS_KABYLAKE. (Chris Wilson)
* Re-order engine wa/ init so latest platforms are first. (Rodrigo Vivi)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Bugzilla: https://bugzilla.freedesktop.org/show_bug.cgi?id=107945
Fixes: 59b449d5c8 ("drm/i915: Split out functions for different kinds of workarounds")
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: intel-gfx@lists.freedesktop.org
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20181203133341.10258-1-tvrtko.ursulin@linux.intel.com
(cherry picked from commit 4a15c75c42460252a63d30f03b4766a52945fb47)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
To enable later verification of GT workaround state at various stages of
driver lifetime, we record the list of applicable ones per platforms to a
list, from which they are also applied.
The added data structure is a simple array of register, mask and value
items, which is allocated on demand as workarounds are added to the list.
This is a temporary implementation which later in the series gets fused
with the existing per context workaround list handling. It is separated at
this stage since the following patch fixes a bug which needs to be as easy
to backport as possible.
Also, since in the following patch we will be adding a new class of
workarounds (per engine) which can be applied from interrupt context, we
straight away make the provision for safe read-modify-write cycle.
v2:
* Change dev_priv to i915 along the init path. (Chris Wilson)
* API rename. (Chris Wilson)
v3:
* Remove explicit list size tracking in favour of growing the allocation
in power of two chunks. (Chris Wilson)
v4:
Chris Wilson:
* Change wa_list_finish to early return.
* Copy workarounds using the compiler for static checking.
* Do not bother zeroing unused entries.
* Re-order struct i915_wa_list.
v5:
* kmalloc_array.
* Whitespace cleanup.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20181203133319.10174-1-tvrtko.ursulin@linux.intel.com
(cherry picked from commit 25d140faaa25f728159eb8c304eae53d88a7f14e)
Fixes: 59b449d5c8 ("drm/i915: Split out functions for different kinds of workarounds")
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
If we force a plane update to fix up our half populated plane state
we'll also force on the pipe gamma for the plane (since we always
enable pipe gamma currently). If the BIOS hasn't programmed a sensible
LUT into the hardware this will cause the image to become corrupted.
Typical symptoms are a purple/yellow/etc. flash when the driver loads.
To avoid this let's program something sensible into the LUT when
we do the plane update. In the future I plan to add proper plane
gamma enable readout so this is just a temporary measure.
Cc: Hans de Goede <hdegoede@redhat.com>
Fixes: 516a49cc19 ("drm/i915: Fix assert_plane() warning on bootup with external display")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181120135450.3634-1-ville.syrjala@linux.intel.com
Tested-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit fa6af5145b4e87a30a530be0d80734a9dd40da77)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Under moderate amounts of GPU stress, we can observe on Bearlake and
Pineview (later gen3 models) that we execute the following batch buffer
before the write into the batch is coherent. Adding extra (tested with
upto 32x) MI_FLUSH to either the invalidation, flush or both phases does
not solve the incoherency issue with the relocations, but emitting the
MI_STORE_DWORD_IMM twice does. So be it.
Fixes: 7dd4f6729f ("drm/i915: Async GPU relocation processing")
Testcase: igt/gem_tiled_fence_blits # blb/pnv
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181119154153.15327-1-chris@chris-wilson.co.uk
(cherry picked from commit 7fa28e146994da1e8a4124623d7da97b798ea520)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
I have a Thinkpad X220 Tablet in my hands that is losing vblank
interrupts whenever LP3 watermarks are used.
If I nudge the latency value written to the WM3 register just
by one in either direction the problem disappears. That to me
suggests that the punit will not enter the corrsponding
powersave mode (MPLL shutdown IIRC) unless the latency value
in the register matches exactly what we read from SSKPD. Ie.
it's not really a latency value but rather just a cookie
by which the punit can identify the desired power saving state.
On HSW/BDW this was changed such that we actually just write
the WM level number into those bits, which makes much more
sense given the observed behaviour.
We could try to handle this by disallowing LP3 watermarks
only when vblank interrupts are enabled but we'd first have
to prove that only vblank interrupts are affected, which
seems unlikely. Also we can't grab the wm mutex from the
vblank enable/disable hooks because those are called with
various spinlocks held. Thus we'd have to redesigne the
watermark locking. So to play it safe and keep the code
simple we simply disable LP3 watermarks on all SNB machines.
To do that we simply zero out the latency values for
watermark level 3, and we adjust the watermark computation
to check for that. The behaviour now matches that of the
g4x/vlv/skl wm code in the presence of a zeroed latency
value.
v2: s/USHRT_MAX/U32_MAX/ for consistency with the types (Chris)
Cc: stable@vger.kernel.org
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101269
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103713
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181114173440.6730-1-ville.syrjala@linux.intel.com
(cherry picked from commit 03981c6ebec4fc7056b9b45f847393aeac90d060)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
In my haste to remove irq_port[] I accidentally changed the
way we deal with hpd pins that are shared by multiple encoders
(DP and HDMI for pre-DDI platforms). Previously we would only
handle such pins via ->hpd_pulse(), but now we queue up the
hotplug work for the HDMI encoder directly. Worse yet, we now
count each hpd twice and this increment the hpd storm count
twice as fast. This can lead to spurious storms being detected.
Go back to the old way of doing things, ie. delegate to
->hpd_pulse() for any pin which has an encoder with that hook
implemented. I don't really like the idea of adding irq_port[]
back so let's loop through the encoders first to check if we
have an encoder with ->hpd_pulse() for the pin, and then go
through all the pins and decided on the correct course of action
based on the earlier findings.
I have occasionally toyed with the idea of unifying the pre-DDI
HDMI and DP encoders into a single encoder as well. Besides the
hotplug processing it would have the other benefit of preventing
userspace from trying to enable both encoders at the same time.
That is simply illegal as they share the same clock/data pins.
We have some testcases that will attempt that and thus fail on
many older machines. But for now let's stick to fixing just the
hotplug code.
Cc: stable@vger.kernel.org # 4.19+
Cc: Lyude Paul <lyude@redhat.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Fixes: b6ca3eee18 ("drm/i915: Nuke dev_priv->irq_port[]")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181108200424.28371-1-ville.syrjala@linux.intel.com
Reviewed-by: Lyude Paul <lyude@redhat.com>
(cherry picked from commit 5a3aeca97af1b6b3498d59a7fd4e8bb95814c108)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Exercising the gpu reloc path strenuously revealed an issue where the
updated relocations (from MI_STORE_DWORD_IMM) were not being observed
upon execution. After some experiments with adding pipecontrols (a lot
of pipecontrols (32) as gen4/5 do not have a bit to wait on earlier pipe
controls or even the current on), it was discovered that we merely
needed to delay the EMIT_INVALIDATE by several flushes. It is important
to note that it is the EMIT_INVALIDATE as opposed to the EMIT_FLUSH that
needs the delay as opposed to what one might first expect -- that the
delay is required for the TLB invalidation to take effect (one presumes
to purge any CS buffers) as opposed to a delay after flushing to ensure
the writes have landed before triggering invalidation.
Testcase: igt/gem_tiled_fence_blits
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181105094305.5767-1-chris@chris-wilson.co.uk
(cherry picked from commit 55f99bf2a9c331838c981694bc872cd1ec4070b2)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Chris Wilson
Joonas Lahtinen
Tvrtko Ursulin
Tina Zhang
Ville Syrjälä
Maarten Lankhorst
Mika Kuoppala
Lionel Landwerlin
Imre Deak
Lyude Paul