openwrt/staging/blogic.git
5 years agodrm/i915: Include register polling in reg_rw traces
Ville Syrjälä [Mon, 4 Feb 2019 21:16:44 +0000 (23:16 +0200)]
drm/i915: Include register polling in reg_rw traces

We generally omit register polling from the i915_reg_rw tracepoint.
Understandable since polling could generate a lot of noise in the
trace. The downside is that the trace is incomplete. As a compromise
let's trace the final register value observed while polling. That
should be generally sufficient to observe what the code should be
doing next.

I suppose in some cases it might make sense to also trace the initial
register value, and maybe the number of times we polled. But that
would require a separate tracepoint so let's leave it for the future.

The other users of _NOTRACE() are i915_pmu and i2c bitbanging,
which I decided to leave alone.

Next we should do something to claw back the tracepoints for
planes and whatnot which were switched to _FW() a while back.
I guess just new macros for raw_rw+trace. The question is
what to call it?

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190204211644.21967-1-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
5 years agodrm/i915: do not return invalid pointers as a *dentry
Greg Kroah-Hartman [Thu, 31 Jan 2019 13:15:07 +0000 (14:15 +0100)]
drm/i915: do not return invalid pointers as a *dentry

When calling debugfs functions, they can now return error values if
something went wrong.  If that happens, return a NULL as a *dentry to
the relay core instead of passing it an illegal pointer.

The relay core should be able to handle an illegal pointer, but add this
check to be safe.

Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: David Airlie <airlied@linux.ie>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: intel-gfx@lists.freedesktop.org
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190131131507.GA19807@kroah.com
5 years agodrm/i915: Rename HAS_GMCH
Rodrigo Vivi [Mon, 4 Feb 2019 22:25:38 +0000 (14:25 -0800)]
drm/i915: Rename HAS_GMCH

First of all GMCH can be considered a feature by itself
since it is a chip present in some platforms that connects
the IA processor to memory and other components in PC.

Also with the introduction of display block at device info,
we got a redundant definition:

.display.has_gmch_display = 1,

So, let's clean up things a bit and use the standardized
way of has_feature on displays side.

No functional change and no manual interaction to generate
this patch.

It is only:

sed -si -e 's/has_gmch_display/has_gmch/g' \
     -e 's/HAS_GMCH_DISPLAY/HAS_GMCH/g' drivers/gpu/drm/i915/*{c,h}

Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190204222538.15842-1-rodrigo.vivi@intel.com
5 years agodrm/i915: Pull i915_gem_active into the i915_active family
Chris Wilson [Tue, 5 Feb 2019 13:00:05 +0000 (13:00 +0000)]
drm/i915: Pull i915_gem_active into the i915_active family

Looking forward, we need to break the struct_mutex dependency on
i915_gem_active. In the meantime, external use of i915_gem_active is
quite beguiling, little do new users suspect that it implies a barrier
as each request it tracks must be ordered wrt the previous one. As one
of many, it can be used to track activity across multiple timelines, a
shared fence, which fits our unordered request submission much better. We
need to steer external users away from the singular, exclusive fence
imposed by i915_gem_active to i915_active instead. As part of that
process, we move i915_gem_active out of i915_request.c into
i915_active.c to start separating the two concepts, and rename it to
i915_active_request (both to tie it to the concept of tracking just one
request, and to give it a longer, less appealing name).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205130005.2807-5-chris@chris-wilson.co.uk
5 years agodrm/i915: Allocate active tracking nodes from a slabcache
Chris Wilson [Tue, 5 Feb 2019 13:00:04 +0000 (13:00 +0000)]
drm/i915: Allocate active tracking nodes from a slabcache

Wrap the active tracking for a GPU references in a slabcache for faster
allocations, and hopefully better fragmentation reduction.

v3: Nothing device specific left, it's just a slabcache that we can
make global.
v4: Include i915_active.h and don't put the initfunc under DEBUG_GEM

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205130005.2807-4-chris@chris-wilson.co.uk
5 years agodrm/i915: Release the active tracker tree upon idling
Chris Wilson [Tue, 5 Feb 2019 13:00:03 +0000 (13:00 +0000)]
drm/i915: Release the active tracker tree upon idling

As soon as we detect that the active tracker is idle and we prepare to
call the retire callback, release the storage for our tree of
per-timeline nodes. We expect these to be infrequently used and quick
to allocate, so there is little benefit in keeping the tree cached and
we would prefer to return the pages back to the system in a timely
fashion.

This also means that when we finalize the struct as a whole, we know as
the activity tracker must be idle, the tree has already been released.
Indeed we can reduce i915_active_fini() just to the assertions that there
is nothing to do.

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/20190205130005.2807-3-chris@chris-wilson.co.uk
5 years agodrm/i915: Generalise GPU activity tracking
Chris Wilson [Tue, 5 Feb 2019 13:00:02 +0000 (13:00 +0000)]
drm/i915: Generalise GPU activity tracking

We currently track GPU memory usage inside VMA, such that we never
release memory used by the GPU until after it has finished accessing it.
However, we may want to track other resources aside from VMA, or we may
want to split a VMA into multiple independent regions and track each
separately. For this purpose, generalise our request tracking (akin to
struct reservation_object) so that we can embed it into other objects.

v2: Tweak error handling during selftest setup.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205130005.2807-2-chris@chris-wilson.co.uk
5 years agodrm/i915/selftests: Exercise some AB...BA preemption chains
Chris Wilson [Tue, 5 Feb 2019 12:38:35 +0000 (12:38 +0000)]
drm/i915/selftests: Exercise some AB...BA preemption chains

Build a chain using 2 contexts (A, B) then request a preemption such
that a later A request runs before the spinner in B.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205123835.25331-1-chris@chris-wilson.co.uk
5 years agodrm/i915/selftests: Context SSEU reconfiguration tests
Tvrtko Ursulin [Tue, 5 Feb 2019 09:50:32 +0000 (09:50 +0000)]
drm/i915/selftests: Context SSEU reconfiguration tests

Exercise the context image reconfiguration logic for idle and busy
contexts, with the resets thrown into the mix as well.

Free from the uAPI restrictions this test runs on all Gen9+ platforms
with slice power gating.

v2:
 * Rename some helpers for clarity.
 * Include subtest names in error logs.
 * Remove unnecessary function export.

v3:
 * Rebase for RUNTIME_INFO.

v4:
 * Fix incomplete unexport from v2. (Chris Wilson)

v5:
 * Rebased for runtime pm api changes.

v6:
 * Rebased for i915_reset.c.

v7:
 * Tidy checkpatch warnings.
 * Consolidate error checking and logging a bit.
 * Skip idle test phase if something failed before it.

v8:
 (Chris Wilson)
 * Fix i915_request_wait error handling.
 * No need to PIN_HIGH the VMA.
 * Remove pointless GEM_BUG_ON before pointer dereference.

v9:
 * Avoid rq leak if rpcs query fails. (Chris)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> # v6
Link: https://patchwork.freedesktop.org/patch/msgid/20190205095032.22673-5-tvrtko.ursulin@linux.intel.com
5 years agodrm/i915: Expose RPCS (SSEU) configuration to userspace (Gen11 only)
Tvrtko Ursulin [Tue, 5 Feb 2019 09:50:31 +0000 (09:50 +0000)]
drm/i915: Expose RPCS (SSEU) configuration to userspace (Gen11 only)

We want to allow userspace to reconfigure the subslice configuration on a
per context basis.

This is required for the functional requirement of shutting down non-VME
enabled sub-slices on Gen11 parts.

To do so, we expose a context parameter to allow adjustment of the RPCS
register stored within the context image (and currently not accessible via
LRI).

If the context is adjusted before first use or whilst idle, the adjustment
is for "free"; otherwise if the context is active we queue a request to do
so (using the kernel context), following all other activity by that
context, which is also marked as barrier for all following submission
against the same context.

Since the overhead of device re-configuration during context switching can
be significant, especially in multi-context workloads, we limit this new
uAPI to only support the Gen11 VME use case. In this use case either the
device is fully enabled, and exactly one slice and half of the subslices
are enabled.

Example usage:

struct drm_i915_gem_context_param_sseu sseu = { };
struct drm_i915_gem_context_param arg = {
.param = I915_CONTEXT_PARAM_SSEU,
.ctx_id = gem_context_create(fd),
.size = sizeof(sseu),
.value = to_user_pointer(&sseu)
};

/* Query device defaults. */
gem_context_get_param(fd, &arg);

/* Set VME configuration on a 1x6x8 part. */
sseu.slice_mask = 0x1;
sseu.subslice_mask = 0xe0;
gem_context_set_param(fd, &arg);

v2: Fix offset of CTX_R_PWR_CLK_STATE in intel_lr_context_set_sseu()
    (Lionel)

v3: Add ability to program this per engine (Chris)

v4: Move most get_sseu() into i915_gem_context.c (Lionel)

v5: Validate sseu configuration against the device's capabilities (Lionel)

v6: Change context powergating settings through MI_SDM on kernel context
    (Chris)

v7: Synchronize the requests following a powergating setting change using
    a global dependency (Chris)
    Iterate timelines through dev_priv.gt.active_rings (Tvrtko)
    Disable RPCS configuration setting for non capable users
    (Lionel/Tvrtko)

v8: s/union intel_sseu/struct intel_sseu/ (Lionel)
    s/dev_priv/i915/ (Tvrtko)
    Change uapi class/instance fields to u16 (Tvrtko)
    Bump mask fields to 64bits (Lionel)
    Don't return EPERM when dynamic sseu is disabled (Tvrtko)

v9: Import context image into kernel context's ppgtt only when
    reconfiguring powergated slice/subslices (Chris)
    Use aliasing ppgtt when needed (Michel)

Tvrtko Ursulin:

v10:
 * Update for upstream changes.
 * Request submit needs a RPM reference.
 * Reject on !FULL_PPGTT for simplicity.
 * Pull out get/set param to helpers for readability and less indent.
 * Use i915_request_await_dma_fence in add_global_barrier to skip waits
   on the same timeline and avoid GEM_BUG_ON.
 * No need to explicitly assign a NULL pointer to engine in legacy mode.
 * No need to move gen8_make_rpcs up.
 * Factored out global barrier as prep patch.
 * Allow to only CAP_SYS_ADMIN if !Gen11.

v11:
 * Remove engine vfunc in favour of local helper. (Chris Wilson)
 * Stop retiring requests before updates since it is not needed
   (Chris Wilson)
 * Implement direct CPU update path for idle contexts. (Chris Wilson)
 * Left side dependency needs only be on the same context timeline.
   (Chris Wilson)
 * It is sufficient to order the timeline. (Chris Wilson)
 * Reject !RCS configuration attempts with -ENODEV for now.

v12:
 * Rebase for make_rpcs.

v13:
 * Centralize SSEU normalization to make_rpcs.
 * Type width checking (uAPI <-> implementation).
 * Gen11 restrictions uAPI checks.
 * Gen11 subslice count differences handling.
 Chris Wilson:
 * args->size handling fixes.
 * Update context image from GGTT.
 * Postpone context image update to pinning.
 * Use i915_gem_active_raw instead of last_request_on_engine.

v14:
 * Add activity tracker on intel_context to fix the lifetime issues
   and simplify the code. (Chris Wilson)

v15:
 * Fix context pin leak if no space in ring by simplifying the
   context pinning sequence.

v16:
 * Rebase for context get/set param locking changes.
 * Just -ENODEV on !Gen11. (Joonas)

v17:
 * Fix one Gen11 subslice enablement rule.
 * Handle error from i915_sw_fence_await_sw_fence_gfp. (Chris Wilson)

v18:
 * Update commit message. (Joonas)
 * Restrict uAPI to VME use case. (Joonas)

v19:
 * Rebase.

v20:
 * Rebase for ce->active_tracker.

v21:
 * Rebase for IS_GEN changes.

v22:
 * Reserve uAPI for flags straight away. (Chris Wilson)

v23:
 * Rebase for RUNTIME_INFO.

v24:
 * Added some headline docs for the uapi usage. (Joonas/Chris)

v25:
 * Renamed class/instance to engine_class/engine_instance to avoid clash
   with C++ keyword. (Tony Ye)

v26:
 * Rebased for runtime pm api changes.

v27:
 * Rebased for intel_context_init.
 * Wrap commit msg to 75.

v28:
 (Chris Wilson)
 * Use i915_gem_ggtt.
 * Use i915_request_await_dma_fence to show a better example.

v29:
 * i915_timeline_set_barrier can now fail. (Chris Wilson)

v30:
 * Capture some acks.

v31:
 * Drop the WARN_ON from use controllable paths. (Chris Wilson)
 * Use overflows_type for all checks.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100899
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107634
Issue: https://github.com/intel/media-driver/issues/267
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Zhipeng Gong <zhipeng.gong@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tony Ye <tony.ye@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Acked-by: Timo Aaltonen <timo.aaltonen@canonical.com>
Acked-by: Takashi Iwai <tiwai@suse.de>
Acked-by: Stéphane Marchesin <marcheu@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205095032.22673-4-tvrtko.ursulin@linux.intel.com
5 years agodrm/i915: Add timeline barrier support
Tvrtko Ursulin [Tue, 5 Feb 2019 09:50:30 +0000 (09:50 +0000)]
drm/i915: Add timeline barrier support

Timeline barrier allows serialization between different timelines.

After calling i915_timeline_set_barrier with a request, all following
submissions on this timeline will be set up as depending on this request,
or barrier. Once the barrier has been completed it automatically gets
cleared and things continue as normal.

This facility will be used by the upcoming context SSEU code.

v2:
 * Assert barrier has been retired on timeline_fini. (Chris Wilson)
 * Fix mock_timeline.

v3:
 * Improved comment language. (Chris Wilson)

v4:
 * Maintain ordering with previous barriers set on the timeline.

v5:
 * Rebase.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205095032.22673-3-tvrtko.ursulin@linux.intel.com
5 years agodrm/i915/perf: lock powergating configuration to default when active
Lionel Landwerlin [Tue, 5 Feb 2019 09:50:29 +0000 (09:50 +0000)]
drm/i915/perf: lock powergating configuration to default when active

If some of the contexts submitting workloads to the GPU have been
configured to shutdown slices/subslices, we might loose the NOA
configurations written in the NOA muxes.

One possible solution to this problem is to reprogram the NOA muxes
when we switch to a new context. We initially tried this in the
workaround batchbuffer but some concerns where raised about the cost
of reprogramming at every context switch. This solution is also not
without consequences from the userspace point of view. Reprogramming
of the muxes can only happen once the powergating configuration has
changed (which happens after context switch). This means for a window
of time during the recording, counters recorded by the OA unit might
be invalid. This requires userspace dealing with OA reports to discard
the invalid values.

Minimizing the reprogramming could be implemented by tracking of the
last programmed configuration somewhere in GGTT and use MI_PREDICATE
to discard some of the programming commands, but the command streamer
would still have to parse all the MI_LRI instructions in the
workaround batchbuffer.

Another solution, which this change implements, is to simply disregard
the user requested configuration for the period of time when i915/perf
is active.

On most platforms there are no issues with this apart from a performance
penality for some media workloads that benefit from running on a partially
powergated GPU. We already prevent RC6 from affecting the programming so
it doesn't sound completely unreasonable to hold on powergating for the
same reason.

On Icelake however there would a functional problem if the slices not-
containing the VME block were left enabled with a running media workload
which explicitly disabled them. To avoid a GPU hang in this case, on
Icelake we lock the enablement to only slices which contain VME blocks.
Downside is that it means degraded GPU performance when OA is active but
there is no known alternative solution for this.

v2: Leave RPCS programming in intel_lrc.c (Lionel)

v3: Update for s/union intel_sseu/struct intel_sseu/ (Lionel)
    More to_intel_context() (Tvrtko)
    s/dev_priv/i915/ (Tvrtko)

Tvrtko Ursulin:

v4:
 * Rebase for make_rpcs changes.

v5:
 * Apply OA restriction from make_rpcs directly.

v6:
 * Rebase for context image setup changes.

v7:
 * Move stream assignment before metric enable.

v8-9:
 * Rebase.

v10:
 * Squashed with ICL support patch.

Bspec: 21140
Co-developed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> # v9
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205095032.22673-2-tvrtko.ursulin@linux.intel.com
5 years agodrm/i915: Record the sseu configuration per-context & engine
Lionel Landwerlin [Tue, 5 Feb 2019 09:50:28 +0000 (09:50 +0000)]
drm/i915: Record the sseu configuration per-context & engine

We want to expose the ability to reconfigure the slices, subslice and
eu per context and per engine. To facilitate that, store the current
configuration on the context for each engine, which is initially set
to the device default upon creation.

v2: record sseu configuration per context & engine (Chris)

v3: introduce the i915_gem_context_sseu to store powergating
    programming, sseu_dev_info has grown quite a bit (Lionel)

v4: rename i915_gem_sseu into intel_sseu (Chris)
    use to_intel_context() (Chris)

v5: More to_intel_context() (Tvrtko)
    Switch intel_sseu from union to struct (Tvrtko)
    Move context default sseu in existing loop (Chris)

v6: s/intel_sseu_from_device_sseu/intel_device_default_sseu/ (Tvrtko)

Tvrtko Ursulin:

v7:
 * Pass intel_sseu by pointer instead of value to make_rpcs.
 * Rebase for make_rpcs changes.

v8:
 * Rebase for RPCS edit on pin.

v9:
 * Rebase for context image setup changes.

v10:
 * Rename dev_priv to i915. (Chris Wilson)

v11:
 * Rebase.

v12:
 * Rebase for IS_GEN changes.

v13:
 * Rebase for RUNTIME_INFO.

v14:
 * Rebase for intel_context_init.

v15:
 * Rebase for drm-tip changes.

v16:
 * Moved struct intel_sseu definition to i915_gem_context.h.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190205095032.22673-1-tvrtko.ursulin@linux.intel.com
5 years agodrm/i915: Trim NEWCLIENT boosting
Chris Wilson [Mon, 4 Feb 2019 15:01:01 +0000 (15:01 +0000)]
drm/i915: Trim NEWCLIENT boosting

Limit the NEWCLIENT boost to only give its small priority boost to fresh
clients only that have no dependencies.

The idea for using NEWCLIENT boosting, commit b16c765122f9 ("drm/i915:
Priority boost for new clients"), is that short-lived streams are often
interactive and require lower latency -- and that by executing those
ahead of the long running hogs, the short-lived clients do little to
interfere with the system throughput by virtue of their short-lived
nature. However, we were only considering the client's own timeline for
determining whether or not it was a fresh stream. This allowed for
compositors to wake up before their vblank and bump all of its client
streams. However, in testing with media-bench this results in chaining
all cooperating contexts together preventing us from being able to
reorder contexts to reduce bubbles (pipeline stalls), overall increasing
latency, and reducing system throughput. The exact opposite of our
intent. The compromise of applying the NEWCLIENT boost to strictly fresh
clients (that do not wait upon anything else) should maintain the
"real-time response under load" characteristics of FQ_CODEL, without
locking together the long chains of dependencies across the system.

References: b16c765122f9 ("drm/i915: Priority boost for new clients")
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/20190204150101.30759-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Allow normal clients to always preempt idle priority clients
Chris Wilson [Mon, 4 Feb 2019 08:41:05 +0000 (08:41 +0000)]
drm/i915: Allow normal clients to always preempt idle priority clients

When first enabling preemption, we hesitated from making it a free-for-all
where every higher priority client would force a preempt-to-idle cycle
and take over from all lower priority clients. We hesitated because we
were uncertain just how well preemption would work in practice, whether
the preemption latency itself would detract from the latency gains for
higher priority tasks and whether it would work at all. Since
introducing preemption, we have been enabling it for more common tasks,
even giving normal clients a small preemptive boost when they first
start (to aide fairness and improve interactivity). Now lets take one
step further and give permission for all normal (priority:0) clients to
preempt any idle (priority:<0) task so that users running long compute
jobs do not overly impact other jobs (i.e. their desktop) and the system
remains responsive under such idle loads.

References: f6322eddaff7 ("drm/i915/preemption: Allow preemption between submission ports")
References: b16c765122f9 ("drm/i915: Priority boost for new clients")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: "Bloomfield, Jon" <jon.bloomfield@intel.com>
Cc: "Stead, Alan" <alan.stead@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190204084116.3013-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Update DRIVER_DATE to 20190202
Rodrigo Vivi [Sat, 2 Feb 2019 08:14:28 +0000 (00:14 -0800)]
drm/i915: Update DRIVER_DATE to 20190202

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 years agodrm/i915/cfl: Adding another PCI Device ID.
Rodrigo Vivi [Fri, 1 Feb 2019 23:50:49 +0000 (15:50 -0800)]
drm/i915/cfl: Adding another PCI Device ID.

While cross checking PCI IDs from Intel Media SDK
and kernel Dmitry noticed this gap. So we checked the
spec and this new ID had been recently added.

v2: Adding new H_GT1 entry to i915_pci.c (Jose)

Reported-by: Dmitry Rogozhkin<dmitry.v.rogozhkin@intel.com>
Cc: Dmitry Rogozhkin<dmitry.v.rogozhkin@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190201235049.27206-1-rodrigo.vivi@intel.com
5 years agoMerge tag 'gvt-next-2019-02-01' of https://github.com/intel/gvt-linux into drm-intel...
Rodrigo Vivi [Fri, 1 Feb 2019 17:03:23 +0000 (09:03 -0800)]
Merge tag 'gvt-next-2019-02-01' of https://github.com/intel/gvt-linux into drm-intel-next-queued

gvt-next-2019-02-01

- new VFIO EDID region support (Henry)

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
From: Zhenyu Wang <zhenyuw@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190201061523.GE5588@zhen-hp.sh.intel.com
5 years agodrm/i915: Enable fastboot by default on VLV and CHV
Hans de Goede [Tue, 29 Jan 2019 14:22:37 +0000 (15:22 +0100)]
drm/i915: Enable fastboot by default on VLV and CHV

We really want to have fastboot enabled by default to avoid an ugly
modeset during boot.

Currently we are enabling fastboot by default on gen9+ (Skylake and newer).
The intention is to enable it on older generations after it has seen more
testing on gen9+.

VLV and CHV devices are still being sold in stores today, as such it is
desirable to also enable fastboot by default on these now.

I've extensively tested fastboot=1 support on over 50 different
Bay- and Cherry-Trail devices. Testing DSI and eDP panels as well as
HDMI output (and even DP over Type-C on one device).

All 50 devices work fine with fastboot=1. On 2 devices their DSI panel
turns black as soon as the i915 driver loads when fastboot=0, so having
fastboot enabled is required for these 2 to work properly (for lack of
a better fix).

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129142237.8684-1-hdegoede@redhat.com
5 years agodrm/i915/icl: restore WaEnableFloatBlendOptimization
Talha Nassar [Fri, 1 Feb 2019 01:08:44 +0000 (17:08 -0800)]
drm/i915/icl: restore WaEnableFloatBlendOptimization

Enables blend optimization for floating point RTs

This restores the workaround that was reverted in c358514ba8da
("Revert "drm/i915/icl: WaEnableFloatBlendOptimization"").

The revert was due to the register write seemingly not sticking,
but the HW team has confirmed that this is because the
register is WO and that the workaround is indeed required.

Here the wa is added with a mask of 0 since the register is WO.

References: https://hsdes.intel.com/resource/1408134172
References: https://bugs.freedesktop.org/show_bug.cgi?id=107338
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Talha Nassar <talha.nassar@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/1548983324-15344-4-git-send-email-talha.nassar@intel.com
5 years agodrm/i915: Save some lines of source code in workarounds
Tvrtko Ursulin [Fri, 1 Feb 2019 01:08:43 +0000 (17:08 -0800)]
drm/i915: Save some lines of source code in workarounds

No functional or code size change - just notice we can compact the source
by re-using a single helper for adding workarounds.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/1548983324-15344-3-git-send-email-talha.nassar@intel.com
5 years agodrm/i915: Move workaround infrastructure code up
Tvrtko Ursulin [Fri, 1 Feb 2019 01:08:42 +0000 (17:08 -0800)]
drm/i915: Move workaround infrastructure code up

Top comment in intel_workarounds.c says common code should come first so
lets respect that. Also, by moving the common code together opportunities
to reduce duplication will become more obvious.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/1548983324-15344-2-git-send-email-talha.nassar@intel.com
5 years agodrm/i915/icl: Work around broken VBTs for port F detection
Imre Deak [Thu, 20 Dec 2018 15:52:11 +0000 (17:52 +0200)]
drm/i915/icl: Work around broken VBTs for port F detection

VBT may include incorrect information about the presence of port F. Work
around this on SKUs where we know the port is not present.

v2:
- Fix IS_ICL_WITH_PORT_F, so it's useable from any context.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108915
Cc: Mika Kahola <mika.kahola@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181220155211.31456-1-imre.deak@intel.com
5 years agodrm/i915/icl: Add TypeC ports only if VBT is present
Imre Deak [Mon, 28 Jan 2019 11:42:42 +0000 (13:42 +0200)]
drm/i915/icl: Add TypeC ports only if VBT is present

We can't safely probe Type C ports, whether they are a legacy or a
USB/Thunderbolt DP Alternate Type C port. This would require performing
the TypeC connect sequence - as described by the specification - but
that may have unwanted side-effects. These side-effects include at least
- without completeness - timeouts during AUX power well enabling and
subsequent PLL enabling errors.

To safely identify these ports we really need VBT, which has the proper
flag for this (ddi_vbt_port_info::supports_typec_usb, supports_tbt).
Based on the above disable Type C ports if we can't load VBT for some
reason.

v2:
- Notice that we disable TypeC ports completely and simplify accordingly
  (Jose).
- Add code comment explaining why we disabled the ports. (Jani)

Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Jose Roberto de Souza <jose.souza@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128114242.28666-1-imre.deak@intel.com
5 years agodrm/i915: Pick the first unused PLL once again
Ville Syrjälä [Wed, 30 Jan 2019 18:13:59 +0000 (20:13 +0200)]
drm/i915: Pick the first unused PLL once again

commit 5b0bd14dcc6b ("drm/i915/icl: keep track of unused pll while
looping") inadvertently (I presume) changed the code to pick the
last unused dpll rather than the first unused one like we did before.

While there should most likely be no harm in changing the order
let's change back just to avoid a change in the behaviour. At
least it might reduce the confusion when staring at logs (took
me a while to figure out why DPLL1 being picked over DPLL0
when the latter was most definitely available).

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190130181359.20693-1-ville.syrjala@linux.intel.com
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Acked-by: Lucas De Marchi <lucas.demarchi@intel.com>
5 years agodrm/i915: Don't use the second dbuf slice on icl
Ville Syrjälä [Wed, 30 Jan 2019 15:51:10 +0000 (17:51 +0200)]
drm/i915: Don't use the second dbuf slice on icl

The code managing the dbuf slices is borked and needs some
real work to fix. In the meantime let's just stop using the
second slice.

v2: Drop the change to intel_enabled_dbuf_slices_num() (Mahesh)

Cc: Mahesh Kumar <mahesh1.sh.kumar@gmail.com>
Reviewed-by: Imre Deak <imre.deak@intel.com> #v1
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190130155110.12918-1-ville.syrjala@linux.intel.com
Reviewed-by: Mahesh Kumar <mahesh1.sh.kumar@gmail.com>
5 years agodrm/i915/gvt: add VFIO EDID region
Hang Yuan [Wed, 30 Jan 2019 10:25:54 +0000 (18:25 +0800)]
drm/i915/gvt: add VFIO EDID region

Implement VFIO EDID region for vgpu. Support EDID blob update and notify
guest on link state change via hotplug event.

v3: move struct edid_region to kvmgt.c <zhenyu>
v2: add EDID sanity check and size update <zhenyu>

Tested-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Hang Yuan <hang.yuan@linux.intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
5 years agodrm/i915/gvt: add hotplug emulation
Hang Yuan [Wed, 30 Jan 2019 10:25:53 +0000 (18:25 +0800)]
drm/i915/gvt: add hotplug emulation

Add function to emulate hotplug interrupt for SKL/KBL platforms

Signed-off-by: Hang Yuan <hang.yuan@linux.intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
5 years agodrm/i915/gvt: add functions to get default resolution
Hang Yuan [Wed, 30 Jan 2019 10:25:52 +0000 (18:25 +0800)]
drm/i915/gvt: add functions to get default resolution

These functions will get default resolution according to vgpu type.

Signed-off-by: Hang Yuan <hang.yuan@linux.intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
5 years agodrm/i915: Apply LUT validation checks to platforms more accurately (v3)
Matt Roper [Wed, 30 Jan 2019 18:10:22 +0000 (10:10 -0800)]
drm/i915: Apply LUT validation checks to platforms more accurately (v3)

Use of the new DRM_COLOR_LUT_NON_DECREASING test was a bit over-zealous;
it doesn't actually need to be applied to the degamma on "bdw-style"
platforms.  Likewise, we overlooked the fact that CHV should have that
test applied to the gamma LUT as well as the degamma LUT.

Rather than adding more complicated platform checking to
intel_color_check(), let's just store the appropriate set of LUT
validation flags for each platform in the intel_device_info structure.

v2:
 - Shuffle around LUT size tests so that the hardware-specific tests
   won't be applied to legacy gamma tables.  (Ville)
 - Add a debug message so that it will be easier to understand why an
   atomic transaction involving incorrectly-sized LUT's got rejected
   by the driver.

v3:
 - Switch size_t's to int's.  (Ville)

Fixes: 85e2d61e4976 ("drm/i915: Validate userspace-provided color management LUT's (v4)")
References: https://lists.freedesktop.org/archives/intel-gfx/2019-January/187634.html
Cc: Ville Syrjälä <ville.syrjala@linux.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/20190130181022.4291-1-matthew.d.roper@intel.com
5 years agodrm/i915: Force background color to black for gen9+ (v2)
Matt Roper [Wed, 30 Jan 2019 18:51:20 +0000 (10:51 -0800)]
drm/i915: Force background color to black for gen9+ (v2)

We don't yet allow userspace to control the CRTC background color, but
we should manually program the color to black to ensure the BIOS didn't
leave us with some other color.  We should also set the pipe gamma and
pipe CSC bits so that the background color goes through the same color
management transformations that a plane with black pixels would.

v2: Rename register to SKL_BOTTOM_COLOR to more closely follow
    bspec naming.  (Ville)

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/20190130185122.10322-2-matthew.d.roper@intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
5 years agodrm/i915: Update DRIVER_DATE to 20190129
Rodrigo Vivi [Wed, 30 Jan 2019 01:39:13 +0000 (17:39 -0800)]
drm/i915: Update DRIVER_DATE to 20190129

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 years agodrm/i915: Use IS_GEN9_LP() for the linetime w/a check
Ville Syrjälä [Fri, 21 Dec 2018 17:14:36 +0000 (19:14 +0200)]
drm/i915: Use IS_GEN9_LP() for the linetime w/a check

IS_GLK||IS_BXT == IS_GEN9_LP

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-10-ville.syrjala@linux.intel.com
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Drop the pointless linetime==0 check
Ville Syrjälä [Fri, 21 Dec 2018 17:14:35 +0000 (19:14 +0200)]
drm/i915: Drop the pointless linetime==0 check

0*whatever==0 so this check is pointless. Remove it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-9-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Drop the definite article in front of SAGV
Ville Syrjälä [Fri, 21 Dec 2018 17:14:34 +0000 (19:14 +0200)]
drm/i915: Drop the definite article in front of SAGV

The spec doesn't use a definite article in front of SAGV. The
rules regarding articles and initialisms are super fuzzy, but
at least to my ears it sounds much more natural to not have
the article. Perhaps because I tend to pronounce it as
"sag-vee" instead of spelling out the letters one at a time.
Actually I might still prefer to leave out the article if I
did spell them out.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-8-ville.syrjala@linux.intel.com
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 years agodrm/i915: Pass dev_priv to skl_needs_memory_bw_wa()
Ville Syrjälä [Fri, 21 Dec 2018 17:14:33 +0000 (19:14 +0200)]
drm/i915: Pass dev_priv to skl_needs_memory_bw_wa()

skl_needs_memory_bw_wa() doesn't look at the passed in state at all.
Possibly it should, but for now let's make life simpler by just
passing in dev_priv.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-7-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Account for minimum ddb allocation restrictions
Ville Syrjälä [Fri, 21 Dec 2018 17:14:32 +0000 (19:14 +0200)]
drm/i915: Account for minimum ddb allocation restrictions

On icl+ bspec tells us to calculate a separate minimum ddb
allocation from the blocks watermark. Both have to be checked
against the actual ddb allocation, but since we do things the
other way around we'll just calculat the minimum acceptable
ddb allocation by taking the maximum of the two values.

We'll also replace the memcmp() with a full trawl over the
the watermarks so that it'll ignore the min_ddb_alloc
because we can't directly read that out from the hw. I suppose
we could reconstruct it from the other values, but I was
too lazy to do that now.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-6-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Fix > vs >= mismatch in watermark/ddb calculations
Ville Syrjälä [Fri, 21 Dec 2018 17:14:31 +0000 (19:14 +0200)]
drm/i915: Fix > vs >= mismatch in watermark/ddb calculations

Bspec says we have to reject the watermark if it's >= the ddb
allocation. Fix the code to reject the == case as it should.
For transition watermarks we can just use >=, for the rest
we'll do +1 when calculating the minimum ddb allocation size.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-5-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Fix bits vs. bytes mixup in dbuf block size computation
Ville Syrjälä [Fri, 21 Dec 2018 17:14:30 +0000 (19:14 +0200)]
drm/i915: Fix bits vs. bytes mixup in dbuf block size computation

The spec used to say "8bpp" which someone took to mean 8 bytes per
pixel when in fact it was supposed to be 8 bits per pixel. The
spec has been updated to make it more clear now. Fix the code
to match.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-4-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Reinstate an early latency==0 check for skl+
Ville Syrjälä [Fri, 21 Dec 2018 17:14:29 +0000 (19:14 +0200)]
drm/i915: Reinstate an early latency==0 check for skl+

I thought we could remove all the early latency==0 checks
and rely on skl_wm_method{1,2}() checking for it. But
skl_compute_plane_wm() applies a bunch of workarounds to bump
up the latency before calling those guys so clearly it won't
end up doing the right thing. Also not sure if the calculations
based on the method1/2 results are safe agaisnt overflows so
it might not work all that well in any case. Let's put the
early check back.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-3-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Don't ignore level 0 lines watermark for glk+
Ville Syrjälä [Fri, 21 Dec 2018 17:14:28 +0000 (19:14 +0200)]
drm/i915: Don't ignore level 0 lines watermark for glk+

On glk+ the level 0 lines watermark actually matters. Do not ignore it.
And while at it let's change things so that we always program a
consistnet 0 to the register when the lines watermarks is ignored
by the hardware.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181221171436.8218-2-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915/icl: keep track of unused pll while looping
Lucas De Marchi [Fri, 25 Jan 2019 22:24:44 +0000 (14:24 -0800)]
drm/i915/icl: keep track of unused pll while looping

Instead of looping again on the range of plls, just keep track of one
unused one and use it later.

Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125222444.19926-5-lucas.demarchi@intel.com
5 years agodrm/i915/icl: remove dpll from clk_sel
Lucas De Marchi [Fri, 25 Jan 2019 22:24:43 +0000 (14:24 -0800)]
drm/i915/icl: remove dpll from clk_sel

We should not pass DPLL_ID_ICL_DPLL0 or DPLL_ID_ICL_DPLL1 to this
function because the path is only taken for non-combophy ports. Let the
warning trigger if improper value is given.

While at it, rename the function to match the register name we are
trying to program.

v2: fix typo in comment

Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125222444.19926-4-lucas.demarchi@intel.com
5 years agodrm/i915: always return something on DDI clock selection
Lucas De Marchi [Fri, 25 Jan 2019 22:24:42 +0000 (14:24 -0800)]
drm/i915: always return something on DDI clock selection

Even if we don't have the correct clock and get a warning, we should not
skip the return.

v2: improve commit message (from Joonas)

Fixes: 1fa11ee2d9d0 ("drm/i915/icl: start adding the TBT pll")
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: <stable@vger.kernel.org> # v4.19+
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125222444.19926-3-lucas.demarchi@intel.com
5 years agodrm/i915/icl: use tc_port in MG_PLL macros
Lucas De Marchi [Fri, 25 Jan 2019 22:24:41 +0000 (14:24 -0800)]
drm/i915/icl: use tc_port in MG_PLL macros

Fix the TODO leftover in the code by changing the argument in MG_PLL
macros. The MG_PLL ids used to access the register values can be
converted from tc_port rather than port.

All these registers can use the TC port to calculate the right offsets
because they are only available for TC ports. The range (PORT_C onwards)
may not be stable and change from platform to platform. So by using the
TC id directly we avoid having to check for the platform in the "leaf
functions" and thus passing dev_priv around.

The helper functions were also renamed to use "tc" as prefix to make
them more generic.

v2: Improve commit message and fix checkpatch warning (from Paulo)

Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125222444.19926-2-lucas.demarchi@intel.com
5 years agodrm/i915: Drop fake breadcrumb irq
Chris Wilson [Tue, 29 Jan 2019 20:52:30 +0000 (20:52 +0000)]
drm/i915: Drop fake breadcrumb irq

Missed breadcrumb detection is defunct due to the tight coupling with
dma_fence signaling and the myriad ways we may signal fences from
everywhere but from an interrupt, i.e. we frequently signal a fence
before we even see its interrupt. This means that even if we miss an
interrupt for a fence, it still is signaled before our breadcrumb
hangcheck fires, so simplify the breadcrumb hangchecking by moving it
into the GPU hangcheck and forgo fake interrupts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129205230.19056-3-chris@chris-wilson.co.uk
5 years agodrm/i915: Replace global breadcrumbs with per-context interrupt tracking
Chris Wilson [Tue, 29 Jan 2019 20:52:29 +0000 (20:52 +0000)]
drm/i915: Replace global breadcrumbs with per-context interrupt tracking

A few years ago, see commit 688e6c725816 ("drm/i915: Slaughter the
thundering i915_wait_request herd"), the issue of handling multiple
clients waiting in parallel was brought to our attention. The
requirement was that every client should be woken immediately upon its
request being signaled, without incurring any cpu overhead.

To handle certain fragility of our hw meant that we could not do a
simple check inside the irq handler (some generations required almost
unbounded delays before we could be sure of seqno coherency) and so
request completion checking required delegation.

Before commit 688e6c725816, the solution was simple. Every client
waiting on a request would be woken on every interrupt and each would do
a heavyweight check to see if their request was complete. Commit
688e6c725816 introduced an rbtree so that only the earliest waiter on
the global timeline would woken, and would wake the next and so on.
(Along with various complications to handle requests being reordered
along the global timeline, and also a requirement for kthread to provide
a delegate for fence signaling that had no process context.)

The global rbtree depends on knowing the execution timeline (and global
seqno). Without knowing that order, we must instead check all contexts
queued to the HW to see which may have advanced. We trim that list by
only checking queued contexts that are being waited on, but still we
keep a list of all active contexts and their active signalers that we
inspect from inside the irq handler. By moving the waiters onto the fence
signal list, we can combine the client wakeup with the dma_fence
signaling (a dramatic reduction in complexity, but does require the HW
being coherent, the seqno must be visible from the cpu before the
interrupt is raised - we keep a timer backup just in case).

Having previously fixed all the issues with irq-seqno serialisation (by
inserting delays onto the GPU after each request instead of random delays
on the CPU after each interrupt), we can rely on the seqno state to
perfom direct wakeups from the interrupt handler. This allows us to
preserve our single context switch behaviour of the current routine,
with the only downside that we lose the RT priority sorting of wakeups.
In general, direct wakeup latency of multiple clients is about the same
(about 10% better in most cases) with a reduction in total CPU time spent
in the waiter (about 20-50% depending on gen). Average herd behaviour is
improved, but at the cost of not delegating wakeups on task_prio.

v2: Capture fence signaling state for error state and add comments to
warm even the most cold of hearts.
v3: Check if the request is still active before busywaiting
v4: Reduce the amount of pointer misdirection with list_for_each_safe
and using a local i915_request variable inside the loops
v5: Add a missing pluralisation to a purely informative selftest message.

References: 688e6c725816 ("drm/i915: Slaughter the thundering i915_wait_request herd")
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/20190129205230.19056-2-chris@chris-wilson.co.uk
5 years agodrm/i915: Remove the intel_engine_notify tracepoint
Chris Wilson [Tue, 29 Jan 2019 20:52:28 +0000 (20:52 +0000)]
drm/i915: Remove the intel_engine_notify tracepoint

The global seqno is defunct and so we have no meaningful indicator of
forward progress for an engine. You need to listen to the request
signaling tracepoints instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129205230.19056-1-chris@chris-wilson.co.uk
5 years agodrm/i915/tv: Bypass the vertical filter if possible
Ville Syrjälä [Tue, 29 Jan 2019 14:19:13 +0000 (16:19 +0200)]
drm/i915/tv: Bypass the vertical filter if possible

Let's switch the pipe into interlaced mode and switch off
the TV encoder vertical filter if the pipe vdisplay
matches the TV YSIZE exactly.

While I didn't measure it I presume this might reduce
the power consumption a little bit, and the pixel rate
is halved as the pipe will now fetching in interlaced
mode rather than in progressive mode (effectively the
same difference as between IF-ID vs. PF-ID pfit modes
on more modern hardware) so a bit easier on the memory
bandwidth.

Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129141913.5515-2-ville.syrjala@linux.intel.com
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
5 years agodrm/i915/tv: Fix adjusted_mode dotclock for interlaced modes
Ville Syrjälä [Tue, 29 Jan 2019 14:19:12 +0000 (16:19 +0200)]
drm/i915/tv: Fix adjusted_mode dotclock for interlaced modes

intel_tv_mode_to_mode() assumes the pipe will be in progressive
fetch mode, and thus when programming the pipe into interlaced
mode we have to halve the calculated dotclock to get the correct
field duration.

This becomes more important when we start to program the pipe
into interlaced mode on i965gm as we depend on the timestamps
to get accurate frame counter values. Withot halving the clock
our guesstimated frame counter would tick at twice the expected
speed.

Cc: Imre Deak <imre.deak@intel.com>
Fixes: 690157f0a9e7 ("drm/i915/tv: Fix >1024 modes on gen3")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129141913.5515-1-ville.syrjala@linux.intel.com
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
5 years agodrm: Constify drm_color_lut_check()
Ville Syrjälä [Tue, 29 Jan 2019 17:06:09 +0000 (19:06 +0200)]
drm: Constify drm_color_lut_check()

drm_color_lut_check() doens't modify the passed in blob so
let's make it const.

Also s/uint32_t/u32/ while at it.

v2: Reduce line wraps (Sam)

Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129170609.5718-1-ville.syrjala@linux.intel.com
Reviewed-by: Sam Ravnborg <sam@ravnborg.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
5 years agodrm/i915/execlists: Suppress preempting self
Chris Wilson [Tue, 29 Jan 2019 18:54:52 +0000 (18:54 +0000)]
drm/i915/execlists: Suppress preempting self

In order to avoid preempting ourselves, we currently refuse to schedule
the tasklet if we reschedule an inflight context. However, this glosses
over a few issues such as what happens after a CS completion event and
we then preempt the newly executing context with itself, or if something
else causes a tasklet_schedule triggering the same evaluation to
preempt the active context with itself.

However, when we avoid preempting ELSP[0], we still retain the preemption
value as it may match a second preemption request within the same time period
that we need to resolve after the next CS event. However, since we only
store the maximum preemption priority seen, it may not match the
subsequent event and so we should double check whether or not we
actually do need to trigger a preempt-to-idle by comparing the top
priorities from each queue. Later, this gives us a hook for finer
control over deciding whether the preempt-to-idle is justified.

The sequence of events where we end up preempting for no avail is:

1. Queue requests/contexts A, B
2. Priority boost A; no preemption as it is executing, but keep hint
3. After CS switch, B is less than hint, force preempt-to-idle
4. Resubmit B after idling

v2: We can simplify a bunch of tests based on the knowledge that PI will
ensure that earlier requests along the same context will have the highest
priority.
v3: Demonstrate the stale preemption hint with a selftest

References: a2bf92e8cc16 ("drm/i915/execlists: Avoid kicking priority on the current context")
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/20190129185452.20989-4-chris@chris-wilson.co.uk
5 years agodrm/i915: Rename execlists->queue_priority to queue_priority_hint
Chris Wilson [Tue, 29 Jan 2019 18:54:51 +0000 (18:54 +0000)]
drm/i915: Rename execlists->queue_priority to queue_priority_hint

After noticing that we trigger preemption events for currently executing
requests, as well as requests that complete before the preemption and
attempting to suppress those preemption events, it is wise to not
consider the queue_priority to be authoritative. As we only track the
maximum priority seen between dequeue passes, if the maximum priority
request is no longer available for dequeuing (it completed or is even
executing on another engine), we have no knowledge of the previous
queue_priority as it would require us to keep a full history of enqueued
requests -- but we already have that history in the priolists!

Rename the queue_priority to queue_priority_hint so that we do not
confuse it as being exactly the maximum priority in the queue, but merely
an indication that we have seen a new maximum priority value and as such
we should check whether it should preempt the currently running request.

v2: s/preempt_priority_hint/queue_priority_hint/ as preempt implies it
being only used for the singular task of preemption and not the wider
question of waking up due to a change in the queue.

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/20190129185452.20989-3-chris@chris-wilson.co.uk
5 years agodrm/i915: Identify active requests
Chris Wilson [Tue, 29 Jan 2019 18:54:50 +0000 (18:54 +0000)]
drm/i915: Identify active requests

To allow requests to forgo a common execution timeline, one question we
need to be able to answer is "is this request running?". To track
whether a request has started on HW, we can emit a breadcrumb at the
beginning of the request and check its timeline's HWSP to see if the
breadcrumb has advanced past the start of this request. (This is in
contrast to the global timeline where we need only ask if we are on the
global timeline and if the timeline has advanced past the end of the
previous request.)

There is still confusion from a preempted request, which has already
started but relinquished the HW to a high priority request. For the
common case, this discrepancy should be negligible. However, for
identification of hung requests, knowing which one was running at the
time of the hang will be much more important.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129185452.20989-2-chris@chris-wilson.co.uk
5 years agodrm/i915/selftests: Apply a subtest filter
Chris Wilson [Tue, 29 Jan 2019 18:54:49 +0000 (18:54 +0000)]
drm/i915/selftests: Apply a subtest filter

In bringup on simulated HW even rudimentary tests are slow, and so many
may fail that we want to be able to filter out the noise to focus on the
specific problem. Even just the tests groups provided for igt is not
specific enough, and we would like to isolate one particular subtest
(and probably subsubtests!). For simplicity, allow the user to provide a
command line parameter such as

i915.st_filter=i915_timeline_mock_selftests/igt_sync

to restrict ourselves to only running on subtest. The exact name to use
is given during a normal run, highlighted as an error if it failed,
debug otherwise. The test group is optional, and then all subtests are
compared for an exact match with the filter (most subtests have unique
names). The filter can be negated, e.g. i915.st_filter=!igt_sync and
then all tests but those that match will be run. More than one match can
be supplied separated by a comma, e.g.

i915.st_filter=igt_vma_create,igt_vma_pin1

to only run those specified, or

i915.st_filter=!igt_vma_create,!igt_vma_pin1

to run all but those named. Mixing a blacklist and whitelist will only
execute those subtests matching the whitelist so long as they are
previously excluded in the blacklist.

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/20190129185452.20989-1-chris@chris-wilson.co.uk
5 years agoMerge drm/drm-next into drm-intel-next-queued
Rodrigo Vivi [Tue, 29 Jan 2019 17:07:07 +0000 (09:07 -0800)]
Merge drm/drm-next into drm-intel-next-queued

A backmerge to unblock gen8+ semaphores.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
5 years agodrm/i915: Fix skl srckey mask bits
Ville Syrjälä [Fri, 25 Jan 2019 18:38:46 +0000 (20:38 +0200)]
drm/i915: Fix skl srckey mask bits

We're incorrectly masking off the R/V channel enable bit from
KEYMSK. Fix it up.

Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Fixes: b20815255693 ("drm/i915: Add plane alpha blending support, v2.")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125183846.28755-1-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
5 years agodrm/i915: Enable fastboot by default on Skylake and newer
Hans de Goede [Thu, 24 Jan 2019 13:01:14 +0000 (14:01 +0100)]
drm/i915: Enable fastboot by default on Skylake and newer

We really want to have fastboot enabled by default to avoid an ugly
modeset during boot.

Rather then enabling it everywhere, lets start with enabling it on
Skylake and newer.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124130114.3967-1-maarten.lankhorst@linux.intel.com
5 years agodrm/i915: Track active timelines
Chris Wilson [Mon, 28 Jan 2019 18:18:12 +0000 (18:18 +0000)]
drm/i915: Track active timelines

Now that we pin timelines around use, we have a clearly defined lifetime
and convenient points at which we can track only the active timelines.
This allows us to reduce the list iteration to only consider those
active timelines and not all.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-6-chris@chris-wilson.co.uk
5 years agodrm/i915: Track the context's seqno in its own timeline HWSP
Chris Wilson [Mon, 28 Jan 2019 18:18:11 +0000 (18:18 +0000)]
drm/i915: Track the context's seqno in its own timeline HWSP

Now that we have allocated ourselves a cacheline to store a breadcrumb,
we can emit a write from the GPU into the timeline's HWSP of the
per-context seqno as we complete each request. This drops the mirroring
of the per-engine HWSP and allows each context to operate independently.
We do not need to unwind the per-context timeline, and so requests are
always consistent with the timeline breadcrumb, greatly simplifying the
completion checks as we no longer need to be concerned about the
global_seqno changing mid check.

One complication though is that we have to be wary that the request may
outlive the HWSP and so avoid touching the potentially danging pointer
after we have retired the fence. We also have to guard our access of the
HWSP with RCU, the release of the obj->mm.pages should already be RCU-safe.

At this point, we are emitting both per-context and global seqno and
still using the single per-engine execution timeline for resolving
interrupts.

v2: s/fake_complete/mark_complete/

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-5-chris@chris-wilson.co.uk
5 years agodrm/i915: Share per-timeline HWSP using a slab suballocator
Chris Wilson [Mon, 28 Jan 2019 18:18:10 +0000 (18:18 +0000)]
drm/i915: Share per-timeline HWSP using a slab suballocator

If we restrict ourselves to only using a cacheline for each timeline's
HWSP (we could go smaller, but want to avoid needless polluting
cachelines on different engines between different contexts), then we can
suballocate a single 4k page into 64 different timeline HWSP. By
treating each fresh allocation as a slab of 64 entries, we can keep it
around for the next 64 allocation attempts until we need to refresh the
slab cache.

John Harrison noted the issue of fragmentation leading to the same worst
case performance of one page per timeline as before, which can be
mitigated by adopting a freelist.

v2: Keep all partially allocated HWSP on a freelist

This is still without migration, so it is possible for the system to end
up with each timeline in its own page, but we ensure that no new
allocation would needless allocate a fresh page!

v3: Throw a selftest at the allocator to try and catch invalid cacheline
reuse.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-4-chris@chris-wilson.co.uk
5 years agodrm/i915: Allocate a status page for each timeline
Chris Wilson [Mon, 28 Jan 2019 18:18:09 +0000 (18:18 +0000)]
drm/i915: Allocate a status page for each timeline

Allocate a page for use as a status page by a group of timelines, as we
only need a dword of storage for each (rounded up to the cacheline for
safety) we can pack multiple timelines into the same page. Each timeline
will then be able to track its own HW seqno.

v2: Reuse the common per-engine HWSP for the solitary ringbuffer
timeline, so that we do not have to emit (using per-gen specialised
vfuncs) the breadcrumb into the distinct timeline HWSP and instead can
keep on using the common MI_STORE_DWORD_INDEX. However, to maintain the
sleight-of-hand for the global/per-context seqno switchover, we will
store both temporarily (and so use a custom offset for the shared timeline
HWSP until the switch over).

v3: Keep things simple and allocate a page for each timeline, page
sharing comes next.

v4: I was caught repeating the same MI_STORE_DWORD_IMM over and over
again in selftests.

v5: And caught red handed copying create timeline + check.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-3-chris@chris-wilson.co.uk
5 years agodrm/i915: Enlarge vma->pin_count
Chris Wilson [Mon, 28 Jan 2019 18:18:08 +0000 (18:18 +0000)]
drm/i915: Enlarge vma->pin_count

Previously we only accommodated having a vma pinned by a small number of
users, with the maximum being pinned for use by the display engine. As
such, we used a small bitfield only large enough to allow the vma to
be pinned twice (for back/front buffers) in each scanout plane. Keeping
the maximum permissible pin_count small allows us to quickly catch a
potential leak. However, as we want to split a 4096B page into 64
different cachelines and pin each cacheline for use by a different
timeline, we will exceed the current maximum permissible vma->pin_count
and so time has come to enlarge it.

Whilst we are here, try to pull together the similar bits:

Address/layout specification:
 - bias, mappable, zone_4g: address limit specifiers
 - fixed: address override, limits still apply though
 - high: not strictly an address limit, but an address direction to search

Search controls:
 - nonblock, nonfault, noevict

v2: Rewrite the guideline comment on bit consumption.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: John Harrison <john.C.Harrison@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-2-chris@chris-wilson.co.uk
5 years agodrm/i915: Introduce concept of per-timeline (context) HWSP
Chris Wilson [Mon, 28 Jan 2019 18:18:07 +0000 (18:18 +0000)]
drm/i915: Introduce concept of per-timeline (context) HWSP

Supplement the per-engine HWSP with a per-timeline HWSP. That is a
per-request pointer through which we can check a local seqno,
abstracting away the presumption of a global seqno. In this first step,
we point each request back into the engine's HWSP so everything
continues to work with the global timeline.

v2: s/i915_request_hwsp/hwsp_seqno/ to emphasis that this is the current
HW value and that we are accessing it via i915_request merely as a
convenience.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128181812.22804-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Move list of timelines under its own lock
Chris Wilson [Mon, 28 Jan 2019 10:23:56 +0000 (10:23 +0000)]
drm/i915: Move list of timelines under its own lock

Currently, the list of timelines is serialised by the struct_mutex, but
to alleviate difficulties with using that mutex in future, move the
list management under its own dedicated mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128102356.15037-5-chris@chris-wilson.co.uk
5 years agodrm/i915: Always allocate an object/vma for the HWSP
Chris Wilson [Mon, 28 Jan 2019 10:23:55 +0000 (10:23 +0000)]
drm/i915: Always allocate an object/vma for the HWSP

Currently we only allocate an object and vma if we are using a GGTT
virtual HWSP, and a plain struct page for a physical HWSP. For
convenience later on with global timelines, it will be useful to always
have the status page being tracked by a struct i915_vma. Make it so.

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/20190128102356.15037-4-chris@chris-wilson.co.uk
5 years agodrm/i915: Move vma lookup to its own lock
Chris Wilson [Mon, 28 Jan 2019 10:23:54 +0000 (10:23 +0000)]
drm/i915: Move vma lookup to its own lock

Remove the struct_mutex requirement for looking up the vma for an
object.

v2: Highlight how the race for duplicate vma creation is resolved on
reacquiring the lock with a short comment.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128102356.15037-3-chris@chris-wilson.co.uk
5 years agodrm/i915: Pull VM lists under the VM mutex.
Chris Wilson [Mon, 28 Jan 2019 10:23:53 +0000 (10:23 +0000)]
drm/i915: Pull VM lists under the VM mutex.

A starting point to counter the pervasive struct_mutex. For the goal of
avoiding (or at least blocking under them!) global locks during user
request submission, a simple but important step is being able to manage
each clients GTT separately. For which, we want to replace using the
struct_mutex as the guard for all things GTT/VM and switch instead to a
specific mutex inside i915_address_space.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128102356.15037-2-chris@chris-wilson.co.uk
5 years agodrm/i915: Stop tracking MRU activity on VMA
Chris Wilson [Mon, 28 Jan 2019 10:23:52 +0000 (10:23 +0000)]
drm/i915: Stop tracking MRU activity on VMA

Our goal is to remove struct_mutex and replace it with fine grained
locking. One of the thorny issues is our eviction logic for reclaiming
space for an execbuffer (or GTT mmaping, among a few other examples).
While eviction itself is easy to move under a per-VM mutex, performing
the activity tracking is less agreeable. One solution is not to do any
MRU tracking and do a simple coarse evaluation during eviction of
active/inactive, with a loose temporal ordering of last
insertion/evaluation. That keeps all the locking constrained to when we
are manipulating the VM itself, neatly avoiding the tricky handling of
possible recursive locking during execbuf and elsewhere.

Note that discarding the MRU (currently implemented as a pair of lists,
to avoid scanning the active list for a NONBLOCKING search) is unlikely
to impact upon our efficiency to reclaim VM space (where we think a LRU
model is best) as our current strategy is to use random idle replacement
first before doing a search, and over time the use of softpinned 48b
per-ppGTT is growing (thereby eliminating any need to perform any eviction
searches, in theory at least) with the remaining users being found on
much older devices (gen2-gen6).

v2: Changelog and commentary rewritten to elaborate on the duality of a
single list being both an inactive and active list.
v3: Consolidate bool parameters into a single set of flags; don't
comment on the duality of a single variable being a multiplicity of
bits.

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/20190128102356.15037-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Try to sanitize bogus DPLL state left over by broken SNB BIOSen
Ville Syrjälä [Fri, 11 Jan 2019 17:49:50 +0000 (19:49 +0200)]
drm/i915: Try to sanitize bogus DPLL state left over by broken SNB BIOSen

Certain SNB machines (eg. ASUS K53SV) seem to have a broken BIOS
which misprograms the hardware badly when encountering a suitably
high resolution display. The programmed pipe timings are somewhat
bonkers and the DPLL is totally misprogrammed (P divider == 0).
That will result in atomic commit timeouts as apparently the pipe
is sufficiently stuck to not signal vblank interrupts.

IIRC something like this was also observed on some other SNB
machine years ago (might have been a Dell XPS 8300) but a BIOS
update cured it. Sadly looks like this was never fixed for the
ASUS K53SV as the latest BIOS (K53SV.320 11/11/2011) is still
broken.

The quickest way to deal with this seems to be to shut down
the pipe+ports+DPLL. Unfortunately doing this during the
normal sanitization phase isn't quite soon enough as we
already spew several WARNs about the bogus hardware state.
But it's better than hanging the boot for a few dozen seconds.
Since this is limited to a few old machines it doesn't seem
entirely worthwile to try and rework the readout+sanitization
code to handle it more gracefully.

v2: Fix potential NULL deref (kbuild test robot)
    Constify has_bogus_dpll_config()

Cc: stable@vger.kernel.org # v4.20+
Cc: Daniel Kamil Kozar <dkk089@gmail.com>
Reported-by: Daniel Kamil Kozar <dkk089@gmail.com>
Tested-by: Daniel Kamil Kozar <dkk089@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109245
Fixes: 516a49cc1946 ("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/20190111174950.10681-1-ville.syrjala@linux.intel.com
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
5 years agodrm/i915/tv: Use the scanline counter for timestamps on i965gm TV output
Ville Syrjälä [Fri, 25 Jan 2019 18:19:31 +0000 (20:19 +0200)]
drm/i915/tv: Use the scanline counter for timestamps on i965gm TV output

Just like the frame counter, the pixel counter also reads zero
all the time when the TV encoder is used. Fortunately the
scanline counter still works sufficiently well so let's use that
to correct the vblank timestamps. Otherwise the timestamps may
en up out of whack, and since we use them to guesstimate the
vblank counter value that may end up incorrect as well.

Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125181931.19482-2-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Fix return value for intel_tv_compute_config()
Ville Syrjälä [Fri, 25 Jan 2019 18:19:30 +0000 (20:19 +0200)]
drm/i915/tv: Fix return value for intel_tv_compute_config()

Ever since commit 204474a6b859 ("drm/i915: Pass down rc in
intel_encoder->compute_config()") we're supposed to return an
errno from .compute_config(). I failed to notice that when
pushing the TV encoder fixes which were written before said
commmit. Fix up the return value for the error case.

Cc: Imre Deak <imre.deak@intel.com>
Fixes: 690157f0a9e7 ("drm/i915/tv: Fix >1024 modes on gen3")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125181931.19482-1-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915: Wait for a moment before forcibly resetting the device
Chris Wilson [Mon, 28 Jan 2019 01:02:18 +0000 (01:02 +0000)]
drm/i915: Wait for a moment before forcibly resetting the device

During igt, we ask to reset the device if any requests are still
outstanding at the end of a test, as this quickly kills off any
erroneous hanging request streams that may escape a test. However, since
it may take the device a few milliseconds to flush itself after the end
of a normal test, *cough* guc *cough*, we may accidentally tell the
device to reset itself after it idles. If we wait a moment, our usual
I915_IDLE_ENGINES_TIMEOUT of 200ms (seems a bit high, but still better
than umpteen hangchecks!), we can differentiate better between a stuck
engine and a healthy one, and so avoid prematurely forcing the reset and
any extra complications that may entail.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190128010245.20148-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Disable -Wuninitialized
Nathan Chancellor [Sat, 26 Jan 2019 07:11:23 +0000 (00:11 -0700)]
drm/i915: Disable -Wuninitialized

This warning is disabled by default in scripts/Makefile.extrawarn when
W= is not provided but this Makefile adds -Wall after this warning is
disabled so it shows up in the build when it shouldn't:

In file included from drivers/gpu/drm/i915/intel_breadcrumbs.c:895:
drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c:350:34: error:
variable 'wq' is uninitialized when used within its own initialization
[-Werror,-Wuninitialized]
        DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
                                        ^~
./include/linux/wait.h:74:63: note: expanded from macro
'DECLARE_WAIT_QUEUE_HEAD_ONSTACK'
        struct wait_queue_head name = __WAIT_QUEUE_HEAD_INIT_ONSTACK(name)
                               ~~~~                                  ^~~~
./include/linux/wait.h:72:33: note: expanded from macro
'__WAIT_QUEUE_HEAD_INIT_ONSTACK'
        ({ init_waitqueue_head(&name); name; })
                                       ^~~~
1 error generated.

Explicitly disable the warning like commit 46e2068081e9 ("drm/i915:
Disable some extra clang warnings").

Link: https://github.com/ClangBuiltLinux/linux/issues/220
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Nick Desaulniers <nick.desaulniers@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190126071122.24557-1-natechancellor@gmail.com
5 years agodrm/i915: correct the pitch check for NV12 framebuffer
P Raviraj Sitaram [Wed, 19 Dec 2018 08:29:12 +0000 (13:59 +0530)]
drm/i915: correct the pitch check for NV12 framebuffer

framebuffer for NV12 requires the pitch to the multiplier of 4, instead
of the width. This patch corrects it.

For instance, a 480p video, whose width and pitch are 854 and 896
respectively, is excluded for NV12 plane so far.

Changes since v1:
    - Removed check for NV12 buffer dimensions since additional checks
      are done for viewport size in intel_sprite.c

Signed-off-by: Dongseong Hwang <dongseong.hwang@intel.com>
Signed-off-by: P Raviraj Sitaram <raviraj.p.sitaram@intel.com>
Cc: Chandra Konduru <chandra.konduru@intel.com>
Cc: Vidya Srinivas <vidya.srinivas@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1545208152-22658-1-git-send-email-raviraj.p.sitaram@intel.com
5 years agodrm/i915: Clean up intel_plane_atomic_check_with_state()
Ville Syrjälä [Fri, 11 Jan 2019 17:08:11 +0000 (19:08 +0200)]
drm/i915: Clean up intel_plane_atomic_check_with_state()

Rename some of the state variables in
intel_plane_atomic_check_with_state() to make it less confusing.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190111170823.4441-2-ville.syrjala@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
5 years agodrm/i915/tv: Filter out >1024 wide modes that would need vertical scaling on gen3
Ville Syrjälä [Mon, 12 Nov 2018 17:00:00 +0000 (19:00 +0200)]
drm/i915/tv: Filter out >1024 wide modes that would need vertical scaling on gen3

Since gen3 can't handle >1024 wide sources with vertical scaling
let's not advertize such modes in the mode list. Less tempetation
to the user to try out things that won't work.

v2: s/IS_GEN3(dev_priv/IS_GEN(dev_priv, 3)/

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-17-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Fix >1024 modes on gen3
Ville Syrjälä [Mon, 12 Nov 2018 16:59:59 +0000 (18:59 +0200)]
drm/i915/tv: Fix >1024 modes on gen3

On gen3 we must disable the TV encoder vertical filter for >1024
pixel wide sources. Once that's done all we can is try to center
the image on the screen. Naturally the TV mode vertical resolution
must be equal or larger than the user mode vertical resolution
or else we'd have to cut off part of the user mode.

And while we may not be able to respect the user's choice of
top and bottom borders exactly (or we'd have to reject he mode
most likely), we can try to maintain the relative sizes of the
top and bottom border with respect to each orher.

Additionally we must configure the pipe as interlaced if the
TV mode is interlaced.

v2: Make +intel_tv_connector_duplicate_state() static and drop
    the badly copy pasted kerneldoc
    s/IS_GEN3(dev_priv/IS_GEN(dev_priv, 3)/

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-16-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Generate better pipe timings for TV encoder
Ville Syrjälä [Mon, 12 Nov 2018 16:59:58 +0000 (18:59 +0200)]
drm/i915/tv: Generate better pipe timings for TV encoder

To make vblank timestamps work better with the TV encoder let's
scale the pipe timings such that the relationship between the
TV active and TV blanking periods is mirrored in the
corresponding pipe timings.

Note that in reality the pipe runs at a faster speed during the
TV vblank, and correspondigly there are periods when the pipe
is enitrely stopped. We pretend that this isn't the case and
as such we incur some error in the vblank timestamps during
the TV vblank. Further explanation of the issues in a big
comment in the code.

This makes the vblank timestamps good enough to make
i965gm (which doesn't have a working frame counter with
the TV encoder) report correct frame numbers. Previously
you could get all kinds of nonsense which resulted in
eg. glxgears reporting that it's running at twice the
actual framerate in most cases.

v2: s/IS_GEN4(dev_priv)/IS_GEN(dev_priv, 4)/ in the comment
    for consistency

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-15-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Add 1080p30/50/60 TV modes
Ville Syrjälä [Mon, 12 Nov 2018 16:59:57 +0000 (18:59 +0200)]
drm/i915/tv: Add 1080p30/50/60 TV modes

Add the missing 1080p TV modes. On gen4 all of them work just fine,
whereas on gen3 only the 30Hz mode actually works correctly.

v2: s/IS_GEN3(dev_priv)/IS_GEN(dev_priv, 3)/

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-14-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Nuke reported_modes[]
Ville Syrjälä [Mon, 12 Nov 2018 16:59:56 +0000 (18:59 +0200)]
drm/i915/tv: Nuke reported_modes[]

Remove the silly reported_modes[] array. I suppse once upon a time
this actually had something to do with modes we reported to userspace.
Now it is just the placeholder for the mode we use for load detection.
We don't need it even for that, and instead we can just rely on
the fallback mode in intel_get_load_detect_pipe().

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-13-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Make TV mode autoselection actually useable
Ville Syrjälä [Mon, 12 Nov 2018 16:59:55 +0000 (18:59 +0200)]
drm/i915/tv: Make TV mode autoselection actually useable

The current code insists on picking a new TV mode when
switching between component and non-component cables.
That's super annoying. Let's just keep the current TV
mode unless the new cable type actually disagrees with it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-12-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Use drm_mode_set_name() to name TV modes
Ville Syrjälä [Mon, 12 Nov 2018 16:59:54 +0000 (18:59 +0200)]
drm/i915/tv: Use drm_mode_set_name() to name TV modes

No point in storing the mode names in the array. drm_mode_set_name()
will give us the same names without wasting space for these string
constants.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-11-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Deobfuscate preferred mode selection
Ville Syrjälä [Mon, 12 Nov 2018 16:59:53 +0000 (18:59 +0200)]
drm/i915/tv: Deobfuscate preferred mode selection

Rewrite the preferred mode selection to just check
whether the TV modes is HD or SD. For SD TV modes we
favor 480 line modes, for 720p we prefer 720 line modes,
and for 1080i/p we prefer 1080 line modes.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-10-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Nuke silly 0 initialzation of xpos/ypos
Ville Syrjälä [Mon, 12 Nov 2018 16:59:51 +0000 (18:59 +0200)]
drm/i915/tv: Nuke silly 0 initialzation of xpos/ypos

Just assign the margin values directly to xpos/ypos instead
of first initializing to zero and then adding the values.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-8-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Use bools where appropriate
Ville Syrjälä [Mon, 12 Nov 2018 16:59:50 +0000 (18:59 +0200)]
drm/i915/tv: Use bools where appropriate

'component_only' is a bool. Initialize it like a bool.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-7-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Store the TV oversampling factor in the TV mode
Ville Syrjälä [Mon, 12 Nov 2018 16:59:49 +0000 (18:59 +0200)]
drm/i915/tv: Store the TV oversampling factor in the TV mode

Store the oversampling factor as a number in the TV modes. We
shall want to arithmetic with this which is easier if it's
a number we can use directly.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-6-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Fix tv mode clocks
Ville Syrjälä [Mon, 12 Nov 2018 16:59:48 +0000 (18:59 +0200)]
drm/i915/tv: Fix tv mode clocks

The oversample clock is always supposed to be either 108 MHz
or 148.5 MHz. Make it so.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-5-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915/tv: Fix interlaced ysize calculation
Ville Syrjälä [Mon, 12 Nov 2018 16:59:47 +0000 (18:59 +0200)]
drm/i915/tv: Fix interlaced ysize calculation

Fix the calculation of the vertical active period for interlaced
TV modes.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181112170000.27531-4-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915: Don't try to use the hardware frame counter with i965gm TV output
Ville Syrjälä [Tue, 27 Nov 2018 20:05:50 +0000 (22:05 +0200)]
drm/i915: Don't try to use the hardware frame counter with i965gm TV output

On i965gm the hardware frame counter does not work when
the TV encoder is active. So let's not try to consult
the hardware frame counter in that case. Instead we'll
fall back to the timestamp based guesstimation method
used on gen2.

Note that the pipe timings generated by the TV encoder
are also rather peculiar. Apparently the pipe wants to
run at a much higher speed (related to the oversample
clock somehow it seems) but during the vertical active
period the TV encoder stalls the pipe every few lines
to keep its speed in check. But once the vertical
blanking period is reached the pipe gets to run at full
speed. This means our vblank timestamp estimates are
suspect. Fixing all that would require quite a bit
more work. This simple fix at least avoids the nasty
vblank timeouts that are happening currently.

Curiously the frame counter works just fine on i945gm
and gm45. I don't really understand what kind of mishap
occurred with the hardware design on i965gm. Sadly
I wasn't able to find any chicken bits etc. that would
fix the frame counter :(

v2: Move the zero vs. non-zero hw counter value handling
    into i915_get_vblank_counter() (Daniel)
    Use the per-crtc maximum exclusively, leaving the
    per-device maximum at zero
v3: max_vblank_count not populated yet in intel_enable_pipe()
    use intel_crtc_max_vblank_count() instead

Cc: stable@vger.kernel.org
Cc: Daniel Vetter <daniel@ffwll.ch>
Fixes: 51e31d49c890 ("drm/i915: Use generic vblank wait")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93782
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190122125149.GE5527@ideak-desk.fi.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
5 years agodrm/i915: Issue engine resets onto idle engines
Chris Wilson [Fri, 25 Jan 2019 13:22:30 +0000 (13:22 +0000)]
drm/i915: Issue engine resets onto idle engines

Always perform the requested reset, even if we believe the engine is
idle. Presumably there was a reason the caller wanted the reset, and in
the near future we lose the easy tracking for whether the engine is
idle.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-5-chris@chris-wilson.co.uk
5 years agodrm/i915/selftests: Trim struct_mutex duration for set-wedged selftest
Chris Wilson [Fri, 25 Jan 2019 13:22:29 +0000 (13:22 +0000)]
drm/i915/selftests: Trim struct_mutex duration for set-wedged selftest

Trim the struct_mutex hold and exclude the call to i915_gem_set_wedged()
as a reminder that it must be callable without struct_mutex held.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-4-chris@chris-wilson.co.uk
5 years agodrm/i915: Remove GPU reset dependence on struct_mutex
Chris Wilson [Fri, 25 Jan 2019 13:22:28 +0000 (13:22 +0000)]
drm/i915: Remove GPU reset dependence on struct_mutex

Now that the submission backends are controlled via their own spinlocks,
with a wave of a magic wand we can lift the struct_mutex requirement
around GPU reset. That is we allow the submission frontend (userspace)
to keep on submitting while we process the GPU reset as we can suspend
the backend independently.

The major change is around the backoff/handoff strategy for performing
the reset. With no mutex deadlock, we no longer have to coordinate with
any waiter, and just perform the reset immediately.

Testcase: igt/gem_mmap_gtt/hang # regresses
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-3-chris@chris-wilson.co.uk
5 years agodrm/i915/guc: Disable global reset
Chris Wilson [Fri, 25 Jan 2019 13:22:27 +0000 (13:22 +0000)]
drm/i915/guc: Disable global reset

The guc (and huc) currently inexcruitably depend on struct_mutex for
device reinitialisation from inside the reset, and indeed taking any
mutex here is verboten (as we must be able to reset from underneath any
of our mutexes). That makes recovering the guc unviable without, for
example, reserving contiguous vma space and pages for it to use.

The plan to re-enable global reset for the GuC centres around reusing the
WOPM reserved space at the top of the aperture (that we know we can
populate a contiguous range large enough to dma xfer the fw image).

In the meantime, hopefully no one even notices as the device-reset is
only used as a backup to the per-engine resets for handling GPU hangs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-2-chris@chris-wilson.co.uk
5 years agodrm/i915: Make all GPU resets atomic
Chris Wilson [Fri, 25 Jan 2019 13:22:26 +0000 (13:22 +0000)]
drm/i915: Make all GPU resets atomic

In preparation for the next few commits, make resetting the GPU atomic.
Currently, we have prepared gen6+ for atomic resetting of individual
engines, but now there is a requirement to perform the whole device
level reset (just the register poking) from inside an atomic context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Compute the HWS offsets explicitly
Chris Wilson [Fri, 25 Jan 2019 12:00:05 +0000 (12:00 +0000)]
drm/i915: Compute the HWS offsets explicitly

Simplify by using sizeof(u32) to convert from the index inside the HWSP
to the byte offset. This has the advantage of not only being shorter
(and so not upsetting checkpatch!) but that it matches use where we are
writing to byte addresses using other commands than MI_STORE_DWORD_IMM.

v2: Drop the now superfluous MI_STORE_DWORD_INDEX_SHIFT, it appears to
be a local invention so keeping it after the final use does not help to
clarify the GPU instruction.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125120005.25191-2-chris@chris-wilson.co.uk
5 years agodrm/i915: Remove manual breadcumb counting
Chris Wilson [Fri, 25 Jan 2019 12:00:04 +0000 (12:00 +0000)]
drm/i915: Remove manual breadcumb counting

Now that we know we measure the size of the engine->emit_breadcrumb()
correctly, we can remove the previous manual counting.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125120005.25191-1-chris@chris-wilson.co.uk
5 years agodrm/i915: Measure the required reserved size for request emission
Chris Wilson [Fri, 25 Jan 2019 10:05:20 +0000 (10:05 +0000)]
drm/i915: Measure the required reserved size for request emission

Instead of tediously and fragilely counting up the number of dwords
required to emit the breadcrumb to seal a request, fake a request and
measure it automatically once during engine setup.

The downside is that this requires a fair amount of mocking to create a
proper breadcrumb. Still, should be less error prone in future as the
breadcrumb size fluctuates!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125100520.20163-1-chris@chris-wilson.co.uk
5 years agodrm/i915/execlists: Move RPCS setup to context pin
Tvrtko Ursulin [Fri, 25 Jan 2019 02:29:33 +0000 (02:29 +0000)]
drm/i915/execlists: Move RPCS setup to context pin

Configuring RPCS in context image just before pin is sufficient and will
come extra handy in one of the following patches.

v2:
 * Split image setup a bit differently. (Chris Wilson)

v3:
 * Update context image after reset as well - otherwise the application
   of pinned default state clears the RPCS.

v4:
 * Use local variable throughout the function. (Chris Wilson)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125023005.1007-1-chris@chris-wilson.co.uk
5 years agodrm/i915/icl: Define MOCS table for Icelake
Tomasz Lis [Thu, 24 Jan 2019 00:06:04 +0000 (16:06 -0800)]
drm/i915/icl: Define MOCS table for Icelake

The table has been unified across OSes to minimize virtualization overhead.

The MOCS table is now published as part of bspec, and versioned. Entries
are supposed to never be modified, but new ones can be added. Adding
entries increases table version. The patch includes version 1 entries.

Meaning of each entry is now explained in bspec, and user mode clients
are expected to know what each entry means. The 3 entries used for previous
platforms are still compatible with their legacy definitions, but that is
not guaranteed to be true for future platforms.

v2: Fixed SCC values, improved commit comment (Daniele)
v3: Improved MOCS table comment (Daniele)
v4: Moved new entries below gen9 ones. Put common entries into
    definition to be used in multiple arrays. (Lucas)
v5: Made defines for or-ing flags. Renamed macros from MOCS_TABLE
    to MOCS_ENTRIES. Switched LE_CoS to upper case. (Joonas)
v6: Removed definitions of reserved entries. (Michal)
    Increased limit of entries sent to the hardware on gen11+.
v7: Simplify table as done for previou gens (Lucas)
v8: Rebase on cached number of entries per-platform and use new
    MOCS_ENTRY() macro (Lucas)
v9: Update comment (from Tomasz)

BSpec: 34007
BSpec: 560

Signed-off-by: Tomasz Lis <tomasz.lis@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-8-lucas.demarchi@intel.com