OSDN Git Service

uclinux-h8/linux.git
9 years agodrm/i915: Skip allocating shadow batch for 0-length batches
Chris Wilson [Fri, 27 Mar 2015 11:02:10 +0000 (11:02 +0000)]
drm/i915: Skip allocating shadow batch for 0-length batches

Since

commit 17cabf571e50677d980e9ab2a43c5f11213003ae
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jan 14 11:20:57 2015 +0000

    drm/i915: Trim the command parser allocations

we may then try to allocate a zero-sized object and attempt to extract
its pages. Understandably this fails.

Testcase: igt/gem_exec_nop #ivb,byt,hsw
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Handle error to get connector state when staging config
Ander Conselvan de Oliveira [Fri, 27 Mar 2015 13:33:51 +0000 (15:33 +0200)]
drm/i915: Handle error to get connector state when staging config

The return value of one of the calls to drm_atomic_get_connector_state()
in intel_modeset_stage_output_state() wasn't checked for errors.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Compare GGTT view structs instead of types
Joonas Lahtinen [Fri, 27 Mar 2015 11:09:22 +0000 (13:09 +0200)]
drm/i915: Compare GGTT view structs instead of types

To allow for views where the view type is not defined by the view type only,
like it is in stereo or rotated 90 degree view, change the semantic to require
the whole view structure for comparison when we match a GGTT view.

This allows including parameters like offset to be included in the view which
is useful for eg. partial views.

v3:
- Rely on ggtt_view type being 0 for non-GGTT vma's, which equals to
  I915_GGTT_VIEW_NORMAL. (Daniel Vetter)
- Do not use potentially slower comparison when we only want to know if
  something is or is not a normal view.
- Rebase on top of rotated view patches. Add rotated view singleton.
- If one view is missing in comparison they're equal only if both are missing.

v4:
- Use comparison helper in obj_to_ggtt_view too. (Tvrtko Ursulin)
- Do WARN_ON if one view is NULL. (Tvrtko Ursulin)

Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: fix simple_return.cocci warnings
kbuild test robot [Fri, 27 Mar 2015 11:26:35 +0000 (19:26 +0800)]
drm/i915: fix simple_return.cocci warnings

drivers/gpu/drm/i915/i915_gem_gtt.c:1349:1-4: WARNING: end returns can be simpified and declaration on line 1347 can be dropped

 Simplify a trivial if-return sequence.  Possibly combine with a
 preceding function call.
Generated by: scripts/coccinelle/misc/simple_return.cocci

CC: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Add module param to test the load detect code
Daniel Vetter [Tue, 3 Mar 2015 17:03:47 +0000 (18:03 +0100)]
drm/i915: Add module param to test the load detect code

This is useful for writing igts to make sure we don't break this,
without being forced to own a one of these dinosaurs.

Suggested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove usage of encoder->new_crtc from clock computations
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:19 +0000 (16:18 +0200)]
drm/i915: Remove usage of encoder->new_crtc from clock computations

Some of the crtc_compute_clock() still depended on encoder->new_crtc
since they didn't use intel_pipe_will_have_type() and used an open
coded version of that function instead. This patch replaces those with
the appropriate code that checks the atomic state intead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Separate the if (!connector) continue to facility easier
extraction of a loop iterator for all of these (there's lots more in
i915 and atomic helpers).]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't look at staged config crtc when changing DRRS state
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:18 +0000 (16:18 +0200)]
drm/i915: Don't look at staged config crtc when changing DRRS state

The function intel_dp_set_drrs_state() would decide which pipe to
downclock based on the staged config for the given connector. However,
the result of that function is immediate, and it uses input values from
crtc->config, so it should be looking at the current crtc instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Convert intel_pipe_will_have_type() to using atomic state
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:17 +0000 (16:18 +0200)]
drm/i915: Convert intel_pipe_will_have_type() to using atomic state

Pass a crtc_state to it and find whether the pipe has an encoder of a
given type by looking at the drm_atomic_state the crtc_state points to.

Until recently i9xx_get_refclk() used to be called indirectly from
vlv_force_pll_on() with a dummy crtc_state. That dummy crtc state is not
converted to be part of a full drm atomic state, so add a WARN in case
someone decides to call that again with a such dummy state. This was
removed in

commit 9cbe40c15a753e02f5da16f6de901decf3276cf1
Author: Vijay Purushothaman <vijay.a.purushothaman@linux.intel.com>
Date:   Thu Mar 5 19:33:08 2015 +0530

    drm/i915: Update prop, int co-eff and gain threshold for CHV

v2: Warn if there is no connectors for a given crtc. (Daniel)
    Replace comment i9xx_get_refclk() with a WARN_ON(). (Ander)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Add commit reference for when i9xx_get_refclk was removed.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Pass an atomic state to modeset_global_resources() functions
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:15 +0000 (16:18 +0200)]
drm/i915: Pass an atomic state to modeset_global_resources() functions

Follow up patches will convert some functions called from there to use
the atomic state, instead of directly accessing the new or current
config. This patch just changes the parameters, but shouldn't have any
functional changes.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Add dynamic page trace events
Michel Thierry [Tue, 24 Mar 2015 15:46:23 +0000 (15:46 +0000)]
drm/i915: Add dynamic page trace events

Traces for page directories and tables allocation and map.

v2: Removed references to teardown.
v3: bitmap_scnprintf has been deprecated.
v4: Replace bitmap_scnprintf with scnprintf correctly, and get right
range lengths. (Mika)

Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Finish gen6/7 dynamic page table allocation
Michel Thierry [Tue, 24 Mar 2015 15:46:22 +0000 (15:46 +0000)]
drm/i915: Finish gen6/7 dynamic page table allocation

This patch continues on the idea from "Track GEN6 page table usage".
From here on, in the steady state, PDEs are all pointing to the scratch
page table (as recommended in the spec). When an object is allocated in
the VA range, the code will determine if we need to allocate a page for
the page table. Similarly when the object is destroyed, we will remove,
and free the page table pointing the PDE back to the scratch page.

Following patches will work to unify the code a bit as we bring in GEN8
support. GEN6 and GEN8 are different enough that I had a hard time to
get to this point with as much common code as I do.

The aliasing PPGTT must pre-allocate all of the page tables. There are a
few reasons for this. Two trivial ones: aliasing ppgtt goes through the
ggtt paths, so it's hard to maintain, we currently do not restore the
default context (assuming the previous force reload is indeed
necessary). Most importantly though, the only way (it seems from
empirical evidence) to invalidate the CS TLBs on non-render ring is to
either use ring sync (which requires actually stopping the rings in
order to synchronize when the sync completes vs. where you are in
execution), or to reload DCLV.  Since without full PPGTT we do not ever
reload the DCLV register, there is no good way to achieve this. The
simplest solution is just to not support dynamic page table
creation/destruction in the aliasing PPGTT.

We could always reload DCLV, but this seems like quite a bit of excess
overhead only to save at most 2MB-4k of memory for the aliasing PPGTT
page tables.

v2: Make the page table bitmap declared inside the function (Chris)
Simplify the way scratching address space works.
Move the alloc/teardown tracepoints up a level in the call stack so that
both all implementations get the trace.

v3: Updated trace event to spit out a name

v4: Aliasing ppgtt is now initialized differently (in setup global gtt)

v5: Rebase to latest code. Also removed unnecessary aliasing ppgtt check
for trace, as it is no longer possible after the PPGTT cleanup patch series
of a couple of months ago (Daniel).

v6: Implement changes from code review (Daniel):
 - allocate/teardown_va_range calls added.
 - Add a scratch page allocation helper (only need the address).
 - Move trace events to a new patch.
 - Use updated mark_tlbs_dirty.
 - Moved pt preallocation for aliasing ppgtt into gen6_ppgtt_init.

v7: teardown_va_range removed (Daniel).
    In init, gen6_ppgtt_clear_range call is only needed for aliasing ppgtt.

v8: Rebase after s/page_tables/page_table/.

v9: Remove unnecessary scratch flag in page_table struct, future patches
can just compare against ppgtt->scratch_pt, and alloc_pt_scratch becomes
redundant. Initialize scratch_pt and pt. (Mika)

v10: Clean up aliasing ppgtt init error path and prevent leaking the
ppgtt obj when init fails. (Mika)
Updated commit author. (Daniel)

Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v4+)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove unnecessary gen6_ppgtt_unmap_pages
Michel Thierry [Tue, 24 Mar 2015 15:46:21 +0000 (15:46 +0000)]
drm/i915: Remove unnecessary gen6_ppgtt_unmap_pages

We are already unmapping them in gen6_ppgtt_free. This function became
redundant since commit 06fda602dbca9c59d87db7da71192e4b54c9f5ff
("drm/i915: Create page table allocators").

Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix i915_dma_map_single positive error code
Michel Thierry [Tue, 24 Mar 2015 17:06:33 +0000 (17:06 +0000)]
drm/i915: Fix i915_dma_map_single positive error code

i915_dma_map_single relies on dma_mapping_error, which returns positive
error codes. Found by static checker.

Introduced by commit 678d96fbb3b5995a2fdff2bca5e1ab4a40b7e968
("drm/i915: Track GEN6 page table usage").

v2: Return negative error code and renamed commit title. (Dan)
v3: Missing reported-by tag (Daniel)

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Prevent out of range pt in gen6_for_each_pde
Michel Thierry [Tue, 24 Mar 2015 15:46:19 +0000 (15:46 +0000)]
drm/i915: Prevent out of range pt in gen6_for_each_pde

Found by static analysis tool, this was harmless as the pt was not
used out of scope though.

Introduced by commit 678d96fbb3b5995a2fdff2bca5e1ab4a40b7e968
("drm/i915: Track GEN6 page table usage").

Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: fix definition of the DRM_IOCTL_I915_GET_SPRITE_COLORKEY ioctl
Tommi Rantala [Thu, 26 Mar 2015 19:47:16 +0000 (21:47 +0200)]
drm/i915: fix definition of the DRM_IOCTL_I915_GET_SPRITE_COLORKEY ioctl

Fix definition of the DRM_IOCTL_I915_GET_SPRITE_COLORKEY ioctl, so that it
is different from the DRM_IOCTL_I915_SET_SPRITE_COLORKEY ioctl.

Note that this is just for accuracy, the ioctl implementation itself is totally
unused and already ripped out.

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
[danvet: Add note that this is a dead ioctl.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Rip out GET_SPRITE_COLORKEY ioctl
Daniel Vetter [Fri, 27 Mar 2015 08:08:04 +0000 (09:08 +0100)]
drm/i915: Rip out GET_SPRITE_COLORKEY ioctl

It's completely unused and Tommi noticed that the #define is borked
since forever. I've done a git search in userspace and only found
broken definitions and no users anywhere.

Cc: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm/i915: Don't use encoder->new_crtc in intel_lvds_compute_config()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:14 +0000 (16:18 +0200)]
drm/i915: Don't use encoder->new_crtc in intel_lvds_compute_config()

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't use staged config in intel_dp_mst_compute_config()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:13 +0000 (16:18 +0200)]
drm/i915: Don't use staged config in intel_dp_mst_compute_config()

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use atomic state in intel_ddi_crtc_get_new_encoder()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:12 +0000 (16:18 +0200)]
drm/i915: Use atomic state in intel_ddi_crtc_get_new_encoder()

Instead of using connector->new_encoder, get the same information from
the pipe_config, thus making the function ready for the atomic
conversion.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't depend on encoder->new_crtc in intel_hdmi_compute_config
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:11 +0000 (16:18 +0200)]
drm/i915: Don't depend on encoder->new_crtc in intel_hdmi_compute_config

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't depend on encoder->new_crtc in intel_dp_compute_config()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:10 +0000 (16:18 +0200)]
drm/i915: Don't depend on encoder->new_crtc in intel_dp_compute_config()

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't use encoder->new_crtc in compute_baseline_pipe_bpp()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:09 +0000 (16:18 +0200)]
drm/i915: Don't use encoder->new_crtc in compute_baseline_pipe_bpp()

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Keep the if (!connector) continue; separate so that it's
easier to eventually extract a for_each_connector_in_state iterator.
And because of the upcast it's also safer.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't use encoder->new_crtc in intel_modeset_pipe_config()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:08 +0000 (16:18 +0200)]
drm/i915: Don't use encoder->new_crtc in intel_modeset_pipe_config()

Move towards atomic by using the legacy modeset's drm_atomic_state
instead.

v2: Move call to drm_atomic_add_affected_connectors() to
    intel_modeset_compute_config(). (Daniel)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Resurrect the ret local variable which I've dropped from an
earlier patch and which is now needed.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Copy the staged connector config to the legacy atomic state
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:07 +0000 (16:18 +0200)]
drm/i915: Copy the staged connector config to the legacy atomic state

With this in place, we can start converting pieces of the modeset code
to look at the connector atomic state instead of the staged config.

v2: Handle the load detect staged config changes too. (Ander)
    Remove unnecessary blank line. (Daniel)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Update dummy connector atomic state with current config
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:05 +0000 (16:18 +0200)]
drm/i915: Update dummy connector atomic state with current config

Keep that state updated so that we can write code that depends on it on
the follow up patches.

v2: Fix BUG due to stale connector_state->crtc value. (Chandra)

v3: Update comment about dummy state connectors. (Chandra)
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Implement connector state duplication
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:06 +0000 (16:18 +0200)]
drm/i915: Implement connector state duplication

So that we can add connector states to the drm_atomic_state used in the
legacy modeset.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: fix semicolon.cocci warnings
kbuild test robot [Thu, 26 Mar 2015 14:30:21 +0000 (22:30 +0800)]
drm/i915/skl: fix semicolon.cocci warnings

drivers/gpu/drm/i915/intel_pm.c:2913:4-5: Unneeded semicolon

 Removes unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

CC: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Allocate a crtc_state also when the crtc is being disabled
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:04 +0000 (16:18 +0200)]
drm/i915: Allocate a crtc_state also when the crtc is being disabled

For consistency, allocate a new crtc_state for a crtc that is being
disabled. Previously only the enabled value of the current state would
change.

v2: Rebase on v5 of previous patch. (Ander)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Resolve rebase conflict.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Allocate a drm_atomic_state for the legacy modeset code
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:03 +0000 (16:18 +0200)]
drm/i915: Allocate a drm_atomic_state for the legacy modeset code

For the atomic conversion, the mode set paths need to be changed to rely
on an atomic state instead of using the staged config. By using an
atomic state for the legacy code, we will be able to convert the code
base in small chunks.

v2: Squash patch that adds stat argument to intel_set_mode(). (Ander)
    Make every caller of intel_set_mode() allocate state. (Daniel)
    Call drm_atomic_state_clear() in set config's error path. (Daniel)

v3: Copy staged config to atomic state in force restore path. (Ander)

v4: Don't update ->new_config for disabled pipes in __intel_set_mode(),
    since it is expected to be NULL in that case. (Ander)

v5: Don't change return type of intel_modeset_pipe_config(). (Chandra)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Remove spurious ret local variable due to changes in v5.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Pass acquire ctx also to intel_release_load_detect_pipe()
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:02 +0000 (16:18 +0200)]
drm/i915: Pass acquire ctx also to intel_release_load_detect_pipe()

For now this is not necessary since intel_set_mode() doesn't acquire any
new locks. However, once that function is converted to atomic, that will
change, since we'll pass an atomic state to it, and that needs to have
the right acquire context set.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Add intel_atomic_get_crtc_state() helper function
Ander Conselvan de Oliveira [Fri, 20 Mar 2015 14:18:01 +0000 (16:18 +0200)]
drm/i915: Add intel_atomic_get_crtc_state() helper function

The pattern of getting the crtc state with drm_atomic_get_crtc_state()
and then converting it to intel_crtc_state will repeat quite often in
the following patches, so add a helper function to save some typing.

v2: Fix upcasting so that crtc_state base field could be moved. (Daniel)
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: always preserve bios swizzling
Daniel Vetter [Thu, 26 Mar 2015 11:42:24 +0000 (12:42 +0100)]
drm/i915: always preserve bios swizzling

Currently we only set preserve_bios_swizzling when the initial fb is
shared and totally miss the single-screen case. Fix this by
consolidating all the logic for both cases.

This seems to go back to when swizzle preservation was originally
merged in

commit d9ceb8163339134bd3ffb9fb87a0db4698283e32
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Thu Oct 9 12:57:43 2014 -0700

    drm/i915: preserve swizzle settings if necessary v4

Cc: Kristian Høgsberg <hoegsberg@gmail.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm/i915: Add initial_ prefix to bios fb takeover code
Daniel Vetter [Thu, 26 Mar 2015 11:17:05 +0000 (12:17 +0100)]
drm/i915: Add initial_ prefix to bios fb takeover code

In spirit with

commit 5724dbd1678e2f573b13f0688277941fad66cb88
Author: Damien Lespiau <damien.lespiau@intel.com>
Date:   Tue Jan 20 12:51:52 2015 +0000

    drm/i915: Rename plane_config to initial_plane_config

to make it clear that this code is all special-purpose for the initial
plane takeover.

Cc: Damien Lespiau <damien.lespiau@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm/i915: Remove duplicated psr.active unset
Rodrigo Vivi [Wed, 25 Mar 2015 19:18:23 +0000 (12:18 -0700)]
drm/i915: Remove duplicated psr.active unset

psr.active is being unset out of the if so this here is useless and
duplicated.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fixup legacy plane->crtc link for initial fb config
Daniel Vetter [Wed, 25 Mar 2015 17:30:38 +0000 (18:30 +0100)]
drm/i915: Fixup legacy plane->crtc link for initial fb config

This is a very similar bug in the load detect code fixed in

commit 9128b040eb774e04bc23777b005ace2b66ab2a85
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Mar 3 17:31:21 2015 +0100

    drm/i915: Fix modeset state confusion in the load detect code

But this time around it was the initial fb code that forgot to update
the plane->crtc pointer. Otherwise it's the exact same bug, with the
exact same restrains (any set_config call/ioctl that doesn't disable
the pipe papers over the bug for free, so fairly hard to hit in normal
testing). So if you want the full explanation just go read that one
over there - it's rather long ...

Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Josh Boyer <jwboyer@fedoraproject.org>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Reported-and-tested-by: Josh Boyer <jwboyer@fedoraproject.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: kill i915.powersave
Rodrigo Vivi [Tue, 24 Mar 2015 19:40:09 +0000 (12:40 -0700)]
drm/i915: kill i915.powersave

This flag was being mostly used as a meta flag in some
cases and not covering other cases.

One of the risks is that it was masking some frontbuffer
trackings without disabling PSR.

So, better to kill this at once and avoid umbrella parameters.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Drop unused out: label to appease gcc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Add fault address to error state for gen8 and gen9
Mika Kuoppala [Tue, 24 Mar 2015 12:54:19 +0000 (14:54 +0200)]
drm/i915: Add fault address to error state for gen8 and gen9

The faulting virtual address is >32bits and has been moved
to different registers. Add to error state and output upper
register first, in the same line for easy reconstruction of
the fault address.

v2: correct gen masking (Michel)
v3: s/TBL/TLB (Ville)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Fix up positive error code
Tvrtko Ursulin [Wed, 25 Mar 2015 10:15:26 +0000 (10:15 +0000)]
drm/i915/skl: Fix up positive error code

It should have been negative since it is returned with ERR_PTR().

Introduced in new code commit:

   commit 50470bb011c4be278097670bea92462f4e8c8945
   Author: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
   Date:   Mon Mar 23 11:10:36 2015 +0000

    drm/i915/skl: Support secondary (rotated) frame buffer mapping

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: make unsupported fb modifier message DRM_DEBUG
Jesse Barnes [Mon, 23 Mar 2015 19:43:50 +0000 (12:43 -0700)]
drm/i915: make unsupported fb modifier message DRM_DEBUG

Or users can just spam the log all they want.

Issue introduced in

commit 9a8f0a1290993c86c4e35756a2624bfe461f9036
Author: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Date:   Fri Feb 27 11:15:24 2015 +0000

    drm/i915/skl: Allow Y (and Yf) frame buffer creation

References: https://bugs.freedesktop.org/show_bug.cgi?id=89628
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Removing the drrs capability enum initialization
Ramalingam C [Mon, 23 Mar 2015 17:42:02 +0000 (23:12 +0530)]
drm/i915: Removing the drrs capability enum initialization

As part of allocation of the drm_i915_private variable, drrs capability
enum is initialized to DRRS_NOT_SUPPORTED. Hence need not initialize at
each connector init.

Moreover initializing this enum at connector init will reset
the successful DRRS initialization of previous connector, as we have
the DRRS support for only one panel at a time.

Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Sivakumar Thulasimani <sivakumar.thulasimani@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: move clearing of RPS interrupt bits from disable to reset time
Imre Deak [Mon, 23 Mar 2015 17:11:35 +0000 (19:11 +0200)]
drm/i915: move clearing of RPS interrupt bits from disable to reset time

The logical place for clearing the RPS latched interrupt bits is when
resetting the RPS interrupts, so move the corresponding part from the RPS
disable function to the reset function. During resetting we already
cleared the IIR bits, so the only thing missing there was clearing pm_iir.

Note that we call gen6_disable_rps_interrupts() also during driver load
and resume time via intel_uncore_sanitize() when i915 interrupts are
still not installed. If there are any pending RPS bits at this point
(which after this patch wouldn't be cleared) they will be cleared by the
reset code via the interrupt preinstall hooks.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: fix race when clearing RPS IIR bits
Imre Deak [Mon, 23 Mar 2015 17:11:34 +0000 (19:11 +0200)]
drm/i915: fix race when clearing RPS IIR bits

When disabling RPS interrupts there is a race where we disable RPS
inerrupts while the interrupt handler is running and the handler has
already latched the pending RPS interrupt from the master IIR register.
Afterwards the disabling path clears the PM IIR bits, making the state
of pending interrupts inconsistent from the interrupt handler's point of
view. This triggers the following warning: "The master control interrupt
lied (PM)!".

To fix this make sure that any running interrupt handler (which may
have already latched the master IIR) finishes before clearing the IIR
bits.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=87347
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Take 90/270 rotation into account in watermark calculations
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:38 +0000 (11:10 +0000)]
drm/i915/skl: Take 90/270 rotation into account in watermark calculations

v2: Pass in rotation info to sprite plane updates as well.

v3: Use helper to determine 90/270 rotation. (Michel Thierry)

v4: Rebased for fb modifiers and atomic changes.

For: VIZ-4546
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v3)
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Query display address through a wrapper
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:37 +0000 (11:10 +0000)]
drm/i915/skl: Query display address through a wrapper

Need to do this in order to support 90/270 rotated display.

v2: Pass in drm_plane instead of plane index to intel_obj_display_address.

v3:
    * Renamed intel_obj_display_address to intel_plane_obj_offset.
      (Chris Wilson)
    * Simplified rotation check to bitwise AND. (Chris Wilson)

v4:
    * Extracted 90/270 rotation check into a helper function. (Michel Thierry)

v5:
    * Rebased for ggtt view changes.

For: VIZ-4545
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Support secondary (rotated) frame buffer mapping
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:36 +0000 (11:10 +0000)]
drm/i915/skl: Support secondary (rotated) frame buffer mapping

90/270 rotated scanout needs a rotated GTT view of the framebuffer.

This is put in a separate VMA with a dedicated ggtt view and wired such that
it is created when a framebuffer is pinned to a 90/270 rotated plane.

Rotation is only possible with Yb/Yf buffers and error is propagated to
user space in case of a mismatch.

Special rotated page view is constructed at the VMA creation time by
borrowing the DMA addresses from obj->pages.

v2:
    * Do not bother with pages for rotated sg list, just populate the DMA
      addresses. (Daniel Vetter)
    * Checkpatch cleanup.

v3:
    * Rebased on top of new plane handling (create rotated mapping when
      setting the rotation property).
    * Unpin rotated VMA on unpinning from display plane.
    * Simplify rotation check using bitwise AND. (Chris Wilson)

v4:
    * Fix unpinning of optional rotated mapping so it is really considered
      to be optional.

v5:
   * Rebased for fb modifier changes.
   * Rebased for atomic commit.
   * Only pin needed view for display. (Ville Syrjälä, Daniel Vetter)

v6:
   * Rebased after preparatory work has been extracted out. (Daniel Vetter)

v7:
   * Slightly simplified tiling geometry calculation.
   * Moved rotated GGTT view implementation into i915_gem_gtt.c (Daniel Vetter)

v8:
   * Do not use i915_gem_obj_size to get object size since that actually
     returns the size of an VMA which may not exist.
   * Rebased for ggtt view changes.

v9:
   * Rebased after code review changes on the preceding patches.
   * Tidy function definitions. (Joonas Lahtinen)

For: VIZ-4726
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v4)
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Helper function to determine GGTT view from plane state
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:35 +0000 (11:10 +0000)]
drm/i915: Helper function to determine GGTT view from plane state

For now only default implementation defaulting to normal view.

v2: Some code review cleanups. (Joonas Lahtinen)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> (v2)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Pass in plane state when (un)pinning frame buffers
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:34 +0000 (11:10 +0000)]
drm/i915: Pass in plane state when (un)pinning frame buffers

Plane state carries the rotation information which is needed for determining
the appropriate GGTT view type.

This just adds the parameter with the actual usage coming in future patches.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use GGTT view when (un)pinning objects to planes
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:33 +0000 (11:10 +0000)]
drm/i915: Use GGTT view when (un)pinning objects to planes

To support frame buffer rotation we need to be able to pass on the information
on what kind of GGTT view is required for display.

This patch just adds the parameter and makes all the callers default to the
normal view.

v2: Rebased for ggtt view changes.
v3: Don't limit PIN_MAPPABLE to normal views just yet. (Joonas Lahtinen)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> (v3)
[danvet: s/BUG/WARN/ in the patch hunk because. At least where the
BUG_ON isn't fatal right away.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Extract tile height code into a helper function
Tvrtko Ursulin [Mon, 23 Mar 2015 11:10:32 +0000 (11:10 +0000)]
drm/i915/skl: Extract tile height code into a helper function

It will be used in a later patch and also convert all height parameters
from int to unsigned int.

v2: Rebased for fb modifiers.
v3: Fixed v2 rebase.
v4:
   * Height should be unsigned int.
   * Make it take pixel_format for consistency and simplicity.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v1)
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> (v4)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use usleep_range() in wait_for()
Ville Syrjälä [Fri, 20 Mar 2015 19:28:08 +0000 (21:28 +0200)]
drm/i915: Use usleep_range() in wait_for()

msleep() can sleep for way too long, so switch wait_for() to use
usleep_range() instead. Following a totally unscientific method
I just picked the range as W-2W.

This cuts the i915 init time on my BSW to almost half:
- initcall i915_init+0x0/0xa8 [i915] returned 0 after 419977 usecs
+ initcall i915_init+0x0/0xa8 [i915] returned 0 after 238419 usecs

Note that I didn't perform any other benchmarks on this so far.

Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Do not leak objects after capturing error state
Michel Thierry [Fri, 20 Mar 2015 09:41:03 +0000 (09:41 +0000)]
drm/i915: Do not leak objects after capturing error state

While running kmemleak chasing a different memleak, I saw that the
capture_error_state function was leaking some objects, for example:

unreferenced object 0xffff8800a9b72148 (size 8192):
  comm "kworker/u16:0", pid 1499, jiffies 4295201243 (age 990.096s)
  hex dump (first 32 bytes):
    00 00 04 00 00 00 00 00 5d f4 ff ff 00 00 00 00  ........].......
    00 30 b0 01 00 00 00 00 37 00 00 00 00 00 00 00  .0......7.......
  backtrace:
    [<ffffffff811e5ae4>] create_object+0x104/0x2c0
    [<ffffffff8178f50a>] kmemleak_alloc+0x7a/0xc0
    [<ffffffff811cde4b>] __kmalloc+0xeb/0x220
    [<ffffffffa038f1d9>] kcalloc.constprop.12+0x2d/0x2f [i915]
    [<ffffffffa0316064>] i915_capture_error_state+0x3f4/0x1660 [i915]
    [<ffffffffa03207df>] i915_handle_error+0x7f/0x660 [i915]
    [<ffffffffa03210f7>] i915_hangcheck_elapsed+0x2e7/0x470 [i915]
    [<ffffffff8108d574>] process_one_work+0x144/0x490
    [<ffffffff8108dfbd>] worker_thread+0x11d/0x530
    [<ffffffff81094079>] kthread+0xc9/0xe0
    [<ffffffff817a2398>] ret_from_fork+0x58/0x90
    [<ffffffffffffffff>] 0xffffffffffffffff

The following objects are allocated in i915_gem_capture_buffers, but not
released in i915_error_state_free:
  - error->active_bo_count
  - error->pinned_bo
  - error->pinned_bo_count
  - error->active_bo[vm_count] (allocated in i915_gem_capture_vm).

The leaks were introduced by
commit 95f5301dd880da2dea2c9a9c29750064536d426a
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Wed Jul 31 17:00:15 2013 -0700

    drm/i915: Update error capture for VMs

v2: Reuse iterator and add culprit commit details (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix SKL sprite disable double buffer register update
Ville Syrjälä [Thu, 19 Mar 2015 15:57:14 +0000 (17:57 +0200)]
drm/i915: Fix SKL sprite disable double buffer register update

Write the PLANE_SURF register instead of PLANE_CTL to arm the double
buffer regisrter update.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Eliminate plane control register RMW from sprite code
Ville Syrjälä [Thu, 19 Mar 2015 15:57:13 +0000 (17:57 +0200)]
drm/i915: Eliminate plane control register RMW from sprite code

Replace the RMW access with explicit initialization of the entire plane
control register, as was done for primary planes in:

 commit f45651bae2ee73ae551699d481f76aa6ad92138f
 Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
 Date:   Fri Aug 8 21:51:10 2014 +0300

    drm/i915: Eliminate rmw from .update_primary_plane()

The automagic primary plane disable is still doing RMWs, but that will
require more work to untangle, so leave it alone for now.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Eliminate the RMW sprite colorkey management
Ville Syrjälä [Thu, 19 Mar 2015 19:18:57 +0000 (21:18 +0200)]
drm/i915: Eliminate the RMW sprite colorkey management

Store the colorkey in intel_plane and kill off all the RMW stuff
handling it.

This is just an intermediate step and eventually the colorkey needs to
be converted into some properties.

v2: Actually update the hardware state in the set_colorkey ioctl (Daniel)

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Move vblank wait determination to 'check' phase
Matt Roper [Wed, 18 Mar 2015 22:04:47 +0000 (15:04 -0700)]
drm/i915: Move vblank wait determination to 'check' phase

Determining whether we'll need to wait for vblanks is something we
should determine during the atomic 'check' phase, not the 'commit'
phase.  Note that we only set these bits in the branch of 'check' where
intel_crtc->active is true so that we don't try to wait on a disabled
CRTC.

The whole 'wait for vblank after update' flag should go away in the
future, once we start handling watermarks in a proper atomic manner.

This regression has been introduced in

commit 2fdd7def16dd7580f297827930126c16b152ec11
Author: Matt Roper <matthew.d.roper@intel.com>
Date:   Wed Mar 4 10:49:04 2015 -0800
    drm/i915: Don't clobber plane state on internal disables

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Root-cause-analysis-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89550
Testcase: igt/pm_rpm/legacy-planes
Testcase: igt/pm_rpm/legacy-planes-dpms
Testcase: igt/pm_rpm/universal-planes
Testcase: igt/pm_rpm/universal-planes-dpms
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/chv: use vlv_PLL_is_optimal in chv_find_best_dpll
Imre Deak [Tue, 17 Mar 2015 09:40:05 +0000 (11:40 +0200)]
drm/i915/chv: use vlv_PLL_is_optimal in chv_find_best_dpll

Prepare chv_find_best_dpll to be used for BXT too, where we want to
consider the error between target and calculated frequency too when
choosing a better PLL configuration.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: check for div-by-zero in vlv_PLL_is_optimal
Imre Deak [Tue, 17 Mar 2015 09:40:04 +0000 (11:40 +0200)]
drm/i915: check for div-by-zero in vlv_PLL_is_optimal

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: factor out vlv_PLL_is_optimal
Imre Deak [Tue, 17 Mar 2015 09:40:03 +0000 (11:40 +0200)]
drm/i915: factor out vlv_PLL_is_optimal

Factor out the logic to decide whether the newly calculated dividers are
better than the best found so far. Do this for clarity and to prepare
for the upcoming BXT helper needing the same.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Kill intel_plane->obj
Ville Syrjälä [Thu, 19 Mar 2015 15:57:11 +0000 (17:57 +0200)]
drm/i915: Kill intel_plane->obj

intel_plane->obj is not used anymore so kill it. Also don't pass both
the fb and obj to the sprite .update_plane() hook, as just passing the fb
is enough.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Initialize all contexts
Ben Widawsky [Mon, 16 Mar 2015 16:00:58 +0000 (16:00 +0000)]
drm/i915: Initialize all contexts

The problem is we're going to switch to a new context, which could be
the default context. The plan was to use restore inhibit, which would be
fine, except if we are using dynamic page tables (which we will). If we
use dynamic page tables and we don't load new page tables, the previous
page tables might go away, and future operations will fault.

CTXA runs.
switch to default, restore inhibit
CTXA dies and has its address space taken away.
Run CTXB, tries to save using the context A's address space - this
fails.

The general solution is to make sure every context has it's own state,
and its own address space. For cases when we must restore inhibit, first
thing we do is load a valid address space. I thought this would be
enough, but apparently there are references within the context itself
which will refer to the old address space - therefore, we also must
reinitialize.

v2: to->ppgtt is only valid in full ppgtt.
v3: Rebased.
v4: Make post PDP update clearer.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Track page table reload need
Ben Widawsky [Thu, 19 Mar 2015 12:53:28 +0000 (12:53 +0000)]
drm/i915: Track page table reload need

This patch was formerly known as, "Force pd restore when PDEs change,
gen6-7." I had to change the name because it is needed for GEN8 too.

The real issue this is trying to solve is when a new object is mapped
into the current address space. The GPU does not snoop the new mapping
so we must do the gen specific action to reload the page tables.

GEN8 and GEN7 do differ in the way they load page tables for the RCS.
GEN8 does so with the context restore, while GEN7 requires the proper
load commands in the command streamer. Non-render is similar for both.

Caveat for GEN7
The docs say you cannot change the PDEs of a currently running context.
We never map new PDEs of a running context, and expect them to be
present - so I think this is okay. (We can unmap, but this should also
be okay since we only unmap unreferenced objects that the GPU shouldn't
be tryingto va->pa xlate.) The MI_SET_CONTEXT command does have a flag
to signal that even if the context is the same, force a reload. It's
unclear exactly what this does, but I have a hunch it's the right thing
to do.

The logic assumes that we always emit a context switch after mapping new
PDEs, and before we submit a batch. This is the case today, and has been
the case since the inception of hardware contexts. A note in the comment
let's the user know.

It's not just for gen8. If the current context has mappings change, we
need a context reload to switch

v2: Rebased after ppgtt clean up patches. Split the warning for aliasing
and true ppgtt options. And do not break aliasing ppgtt, where to->ppgtt
is always null.

v3: Invalidate PPGTT TLBs inside alloc_va_range.

v4: Rename ppgtt_invalidate_tlbs to mark_tlbs_dirty and move
pd_dirty_rings from i915_address_space to i915_hw_ppgtt. Fixes when
neither ctx->ppgtt and aliasing_ppgtt exist.

v5: Removed references to teardown_va_range.

v6: Updated needs_pd_load_pre/post.

v7: Fix pd_dirty_rings check in needs_pd_load_post, and update/move
comment about updated PDEs to object_pin/bind (Mika).

Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Track GEN6 page table usage
Ben Widawsky [Mon, 16 Mar 2015 16:00:56 +0000 (16:00 +0000)]
drm/i915: Track GEN6 page table usage

Instead of implementing the full tracking + dynamic allocation, this
patch does a bit less than half of the work, by tracking and warning on
unexpected conditions. The tracking itself follows which PTEs within a
page table are currently being used for objects. The next patch will
modify this to actually allocate the page tables only when necessary.

With the current patch there isn't much in the way of making a gen
agnostic range allocation function. However, in the next patch we'll add
more specificity which makes having separate functions a bit easier to
manage.

One important change introduced here is that DMA mappings are
created/destroyed at the same page directories/tables are
allocated/deallocated.

Notice that aliasing PPGTT is not managed here. The patch which actually
begins dynamic allocation/teardown explains the reasoning for this.

v2: s/pdp.page_directory/pdp.page_directories
Make a scratch page allocation helper

v3: Rebase and expand commit message.

v4: Allocate required pagetables only when it is needed, _bind_to_vm
instead of bind_vma (Daniel).

v5: Rebased to remove the unnecessary noise in the diff, also:
 - PDE mask is GEN agnostic, renamed GEN6_PDE_MASK to I915_PDE_MASK.
 - Removed unnecessary checks in gen6_alloc_va_range.
 - Changed map/unmap_px_single macros to use dma functions directly and
   be part of a static inline function instead.
 - Moved drm_device plumbing through page tables operation to its own
   patch.
 - Moved allocate/teardown_va_range calls until they are fully
   implemented (in subsequent patch).
 - Merged pt and scratch_pt unmap_and_free path.
 - Moved scratch page allocator helper to the patch that will use it.

v6: Reduce complexity by not tearing down pagetables dynamically, the
same can be achieved while freeing empty vms. (Daniel)

v7: s/i915_dma_map_px_single/i915_dma_map_single
s/gen6_write_pdes/gen6_write_pde
Prevent a NULL case when only GGTT is available. (Mika)

v8: Rebased after s/page_tables/page_table/.

v9: Reworked i915_pte_index and i915_pte_count.
Also exercise bitmap allocation here (gen6_alloc_va_range) and fix
incorrect write_page_range in i915_gem_restore_gtt_mappings (Mika).

Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v3+)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Extract context switch skip and add pd load logic
Ben Widawsky [Mon, 16 Mar 2015 16:00:55 +0000 (16:00 +0000)]
drm/i915: Extract context switch skip and add pd load logic

In Gen8, PDPs are saved and restored with legacy contexts (legacy contexts
only exist on the render ring). So change the ordering of LRI vs MI_SET_CONTEXT
for the initialization of the context. Also the only cases in which we
need to manually update the PDPs are when MI_RESTORE_INHIBIT has been
set in MI_SET_CONTEXT (i.e. when the context is not yet initialized or
it is the default context).

Legacy submission is not available post GEN8, so it isn't necessary to
add extra checks for newer generations.

v2: Use new functions to replace the logic right away (Daniel)
v3: Add missing pd load logic.
v4: Add warning in case pd_load_pre & pd_load_post are true, and add
missing trace_switch_mm. Cleaned up pd_load conditions. Add more
information about when is pd_load_post needed. (Mika)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Michel Thierry <michel.thierry@intel.com> (v2+)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: page table generalizations
Michel Thierry [Mon, 16 Mar 2015 16:00:54 +0000 (16:00 +0000)]
drm/i915: page table generalizations

No functional changes, but will improve code clarity and removed some
duplicated defines.

Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Send out the full AUX address
Ville Syrjälä [Thu, 19 Mar 2015 09:44:06 +0000 (11:44 +0200)]
drm/i915: Send out the full AUX address

AUX addresses are 20 bits long. Send out the entire address instead of
just the low 16 bits.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: kerneldoc for i915_gem_shrinker.c
Daniel Vetter [Wed, 18 Mar 2015 13:47:59 +0000 (14:47 +0100)]
drm/i915: kerneldoc for i915_gem_shrinker.c

And remove one bogus * from i915_gem_gtt.c since that's not a
kerneldoc there.

v2: Review from Chris:
- Clarify memory space to better distinguish from address space.
- Add note that shrink doesn't guarantee the freed memory and that
  users must fall back to shrink_all.
- Explain how pinning ties in with eviction/shrinker.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Extract i915_gem_shrinker.c
Daniel Vetter [Wed, 18 Mar 2015 09:46:04 +0000 (10:46 +0100)]
drm/i915: Extract i915_gem_shrinker.c

Two code changes:
- Extract i915_gem_shrinker_init.
- Inline i915_gem_object_is_purgeable since we open-code it everywhere
  else too.

This already has the benefit of pulling all the shrinker code
together, next patch adds a bit of kerneldoc.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use down ei for manual Baytrail RPS calculations
Chris Wilson [Wed, 18 Mar 2015 09:48:23 +0000 (09:48 +0000)]
drm/i915: Use down ei for manual Baytrail RPS calculations

Use both up/down manual ei calcuations for symmetry and greater
flexibility for reclocking, instead of faking the down interrupt based
on a fixed integer number of up interrupts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Improved w/a for rps on Baytrail
Chris Wilson [Wed, 18 Mar 2015 09:48:22 +0000 (09:48 +0000)]
drm/i915: Improved w/a for rps on Baytrail

Rewrite commit 31685c258e0b0ad6aa486c5ec001382cf8a64212
Author: Deepak S <deepak.s@linux.intel.com>
Date:   Thu Jul 3 17:33:01 2014 -0400

    drm/i915/vlv: WA for Turbo and RC6 to work together.

Other than code clarity, the major improvement is to disable the extra
interrupts generated when idle.  However, the reclocking remains rather
slow under the new manual regime, in particular it fails to downclock as
quickly as desired. The second major improvement is that for certain
workloads, like games, we need to combine render+media activity counters
as the work of displaying the frame is split across the engines and both
need to be taken into account when deciding the global GPU frequency as
memory cycles are shared.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Deepak S <deepak.s@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Relax RPS contraints to allows setting minfreq on idle
Chris Wilson [Wed, 18 Mar 2015 09:48:21 +0000 (09:48 +0000)]
drm/i915: Relax RPS contraints to allows setting minfreq on idle

When we idle, we set the GPU frequency to the hardware minimum (not user
minimum). We introduce a new variable to distinguish between the
different roles, and to allow easy tuning of the idle frequency without
impacting over aspects of RPS. Setting the minimum frequency should be a
safety blanket as the pcu on the GPU should be power gating itself
anyway. However, in order for us to do set the absolute minimum
frequency, we need to relax a few of our assertions that we do not
exceed the user limits.

v2: Add idle_freq
v3: Init idle_freq for vlv and add a bunch of WARNs

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fallback to using CPU relocations for large batch buffers
Chris Wilson [Wed, 14 Jan 2015 11:20:56 +0000 (11:20 +0000)]
drm/i915: Fallback to using CPU relocations for large batch buffers

If the batch buffer is too large to fit into the aperture and we need a
GTT mapping for relocations, we currently fail. This only applies to a
subset of machines for a subset of environments, quite undesirable. We
can simply check after failing to insert the batch into the GTT as to
whether we only need a mappable binding for relocation and, if so, we can
revert to using a non-mappable binding and an alternate relocation
method. However, using relocate_entry_cpu() is excruciatingly slow for
large buffers on non-LLC as the entire buffer requires clflushing before
and after the relocation handling. Alternatively, we can implement a
third relocation method that only clflushes around the relocation entry.
This is still slower than updating through the GTT, so we prefer using
the GTT where possible, but is orders of magnitude faster as we
typically do not have to then clflush the entire buffer.

An alternative idea of using a temporary WC mapping of the backing store
is promising (it should be faster than using the GTT itself), but
requires fairly extensive arch/x86 support - along the lines of
kmap_atomic_prof_pfn() (which is not universally implemented even for
x86).

Testcase: igt/gem_exec_big #pnv,byt
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88392
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add a WARN_ONCE for the impossible reloc case and explain in
a short comment why we want to avoid ping-pong.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Turn on PIN_GLOBAL in i915_gem_object_ggtt_pin
Tvrtko Ursulin [Tue, 17 Mar 2015 15:36:51 +0000 (15:36 +0000)]
drm/i915: Turn on PIN_GLOBAL in i915_gem_object_ggtt_pin

This makes the interface consistent to old i915_gem_obj_ggtt_pin.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: memory leak in __i915_gem_vma_create()
Dan Carpenter [Wed, 18 Mar 2015 08:21:58 +0000 (11:21 +0300)]
drm/i915: memory leak in __i915_gem_vma_create()

In the original code then if WARN_ON(i915_is_ggtt(vm) != !!ggtt_view)
was true then we leak "vma".  Presumably that doesn't happen often but
static checkers complain and this bug is easy to fix.

Fixes: c3bbb6f2825d ('drm/i915: Do not use ggtt_view with (aliasing) PPGTT')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/dp: return number of bytes written for short aux/i2c writes
Jani Nikula [Tue, 17 Mar 2015 15:18:54 +0000 (17:18 +0200)]
drm/i915/dp: return number of bytes written for short aux/i2c writes

Allow for a larger receive data size, and check if the receiver returned
the number of bytes written. Without this, we've basically skipped all
the unwritten bytes for short writes.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Unconfuse DP link rate array names
Ville Syrjälä [Fri, 13 Mar 2015 17:40:31 +0000 (19:40 +0200)]
drm/i915: Unconfuse DP link rate array names

To keep things clear rename the intel_dp->supported_rates[] to
intel_dp->sink_rates[], and rename the supported_rates[] name we used
elsewhere for the intersection of source and sink rates to
common_rates[].

Cc: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Include the sink/source/supported rates in debug output
Ville Syrjälä [Thu, 12 Mar 2015 15:10:39 +0000 (17:10 +0200)]
drm/i915: Include the sink/source/supported rates in debug output

TODO: Is there an actually nice way to print an array of ints?

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Add eDP intermediate frequencies for CHV
Ville Syrjälä [Thu, 12 Mar 2015 15:10:38 +0000 (17:10 +0200)]
drm/i915: Add eDP intermediate frequencies for CHV

"P1273_DPLL_Programming Spreadsheet.xlsm" lists a boatload of
frequencies for eDP. Try to use them all.

For now I've decided not to add hardcoded DPLL dividers for these cases
since chv_find_best_dpll() works just fine.

I've not actually tested any of these since I don't have an eDP 1.4 panel.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Avoid overflowing the DP link rate arrays
Ville Syrjälä [Thu, 12 Mar 2015 15:10:37 +0000 (17:10 +0200)]
drm/i915: Avoid overflowing the DP link rate arrays

Complain loudly if we ever attempt to overflow the the supported_rates[]
array. This should never happen since the sink_rates[] array will always
be smaller or of equal size. But should someone change that we want to
catch it without scribblign over the stack.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix MST link rate handling
Ville Syrjälä [Thu, 12 Mar 2015 15:10:36 +0000 (17:10 +0200)]
drm/i915: Fix MST link rate handling

Now that intel_dp_max_link_bw() no longer considers the source
restrictions we may try to enable MST with 5.4GHz even when the source
doesn't support it. To fix that switch the code over to handle the link
rate in the same way as the SST code handles it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use DP_LINK_RATE_SET whenever possible
Ville Syrjälä [Thu, 12 Mar 2015 15:10:35 +0000 (17:10 +0200)]
drm/i915: Use DP_LINK_RATE_SET whenever possible

Drop the gen9 checks from the code and issue DP_LINK_RATE_SET whenever
the sink reports to support it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix max link rate in intel_dp_mode_valid()
Ville Syrjälä [Thu, 12 Mar 2015 15:10:34 +0000 (17:10 +0200)]
drm/i915: Fix max link rate in intel_dp_mode_valid()

Consider the link rates reported by the sink via
DP_SUPPORTED_LINK_RATES when checking modes against the max link
rate.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Hide the source vs. sink rate handling from intel_dp_compute_config()
Ville Syrjälä [Thu, 12 Mar 2015 15:10:33 +0000 (17:10 +0200)]
drm/i915: Hide the source vs. sink rate handling from intel_dp_compute_config()

intel_dp_compute_config() only really needs to know the rates supported
by both source and sink, so hide the raw source and sink arrays from it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fully separate source vs. sink rates
Ville Syrjälä [Thu, 12 Mar 2015 15:10:32 +0000 (17:10 +0200)]
drm/i915: Fully separate source vs. sink rates

Remove the sink vs. source limit mess from intel_dp_max_link_bw() and
just move the source restriction checks to intel_dp_source_rates().

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
[danvet: Resolve conflict with WaDisableHBR2:skl patch.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove special case from intel_supported_rates()
Ville Syrjälä [Thu, 12 Mar 2015 15:10:31 +0000 (17:10 +0200)]
drm/i915: Remove special case from intel_supported_rates()

Now that both source and sink rates are always filled in there's no need
for any special cases in intel_supported_rates().

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't copy sink rates either
Ville Syrjälä [Thu, 12 Mar 2015 15:10:30 +0000 (17:10 +0200)]
drm/i915: Don't copy sink rates either

Once we've read the rates from the sink we don't have to mess with them,
so the caller can just look at the stored rates without doing extra
copies. If the sink doesn't support the new link rate stuff, we just
point the caller at the default_rates[] array.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't copy the DP source rates arrays
Ville Syrjälä [Thu, 12 Mar 2015 15:10:29 +0000 (17:10 +0200)]
drm/i915: Don't copy the DP source rates arrays

The source rates don't change, so we can just point the caller at the
const arrays.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Store the converted link rates in intel_dp->supported_rates[]
Ville Syrjälä [Thu, 12 Mar 2015 15:10:28 +0000 (17:10 +0200)]
drm/i915: Store the converted link rates in intel_dp->supported_rates[]

No point in converting from hardware format every single time, just
store the rates in the final format under intel_dp.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Make the DP rates int instead of uint32_t
Ville Syrjälä [Thu, 12 Mar 2015 15:10:27 +0000 (17:10 +0200)]
drm/i915: Make the DP rates int instead of uint32_t

No point in using uint32_t here, just plain old int will do.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Sonika Jindal <sonika.jindal@intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Do not use ggtt_view with (aliasing) PPGTT
Joonas Lahtinen [Mon, 16 Mar 2015 12:11:13 +0000 (14:11 +0200)]
drm/i915: Do not use ggtt_view with (aliasing) PPGTT

GGTT views are only applicable when dealing with GGTT. Change the code to
reject ggtt_view where it should not be used and require it when it should
be.

v2:
- Dropped _ppgtt_ infixes, allow both types to be passed
- Disregard other but normal views when no view is specified
- More checks that valid parameters are passed
- More readable error checking

v3:
- Prefer WARN_ONCE over BUG_ON when there is code path for failure

Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
[danvet: Drop unecessary forward decl from earlier patch iterations.]
[danvet: Remove unused variable spotted by Tvrtko.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix sink crc connector iteration
Rodrigo Vivi [Fri, 13 Mar 2015 23:13:59 +0000 (16:13 -0700)]
drm/i915: Fix sink crc connector iteration

Regressed by this commit:

commit 3455454e18ca3f92c565700539e744c620d8276b
Author: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Date:   Tue Mar 3 15:21:56 2015 +0200

    drm/i915: Add a for_each_intel_connector macro

Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agoMerge tag 'drm-intel-fixes-2015-03-19' into drm-intel-next
Daniel Vetter [Fri, 20 Mar 2015 10:43:59 +0000 (11:43 +0100)]
Merge tag 'drm-intel-fixes-2015-03-19' into drm-intel-next

Backmerge because of numerous and interleaving conflicts and git
rerere getting confused a bit too often.

Conflicts:
drivers/gpu/drm/i915/intel_display.c

All conflicts are because of -next patches backported to -fixes, so
just go with the code in -next.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm/i915: Make sure the primary plane is enabled before reading out the fb state
Damien Lespiau [Thu, 5 Feb 2015 19:35:13 +0000 (19:35 +0000)]
drm/i915: Make sure the primary plane is enabled before reading out the fb state

We don't want to end up in a state where we track that the pipe has its
primary plane enabled when primary plane registers are programmed with
values that look possible but the plane actually disabled.

Refuse to read out the fb state when the primary plane isn't enabled.

Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Suggested-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reported-by: Andrey Skvortsov <andrej.skvortzov@gmail.com>
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Reference: http://mid.gmane.org/20150203191507.GA2374@crion86
Tested-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
9 years agodrm/i915: Update DRIVER_DATE to 20150313
Daniel Vetter [Fri, 13 Mar 2015 21:15:40 +0000 (22:15 +0100)]
drm/i915: Update DRIVER_DATE to 20150313

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix vmap_batch page iterator overrun
Mika Kuoppala [Fri, 13 Mar 2015 13:21:53 +0000 (15:21 +0200)]
drm/i915: Fix vmap_batch page iterator overrun

vmap_batch() calculates amount of needed pages for the mapping
we are going to create. And it uses this page count as an
argument for the for_each_sg_pages() macro. The macro takes the number
of sg list entities as an argument, not the page count. So we ended
up iterating through all the pages on the mapped object, corrupting
memory past the smaller pages[] array.

Fix this by bailing out when we have enough pages.

This regression has been introduced in

commit 17cabf571e50677d980e9ab2a43c5f11213003ae
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jan 14 11:20:57 2015 +0000

    drm/i915: Trim the command parser allocations

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Export total subslice and EU counts
Jeff McGee [Mon, 9 Mar 2015 23:06:54 +0000 (16:06 -0700)]
drm/i915: Export total subslice and EU counts

Setup new I915_GETPARAM ioctl entries for subslice total and
EU total. Userspace drivers need these values when constructing
GPGPU commands. This kernel query method is intended to replace
the PCI ID-based tables that userspace drivers currently maintain.
The kernel driver can employ fuse register reads as needed to
ensure the most accurate determination of GT config attributes.
This first became important with Cherryview in which the config
could differ between devices with the same PCI ID.

The kernel detection of these values is device-specific and not
included in this patch. Because zero is not a valid value for any of
these parameters, a value of zero is interpreted as unknown for the
device. Userspace drivers should continue to maintain ID-based tables
for older devices not supported by the new query method.

v2: Increment our I915_GETPARAM indices to fit after REVISION
    which was merged ahead of us.

For: VIZ-4636
Signed-off-by: Jeff McGee <jeff.mcgee@intel.com>
Tested-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Acked-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: redefine WARN_ON_ONCE to include the condition
Jani Nikula [Thu, 12 Mar 2015 11:01:12 +0000 (13:01 +0200)]
drm/i915: redefine WARN_ON_ONCE to include the condition

Same as

commit c883ef1b1c998d2d66866772fd0fc34afa45641e
Author: Mika Kuoppala <miku@iki.fi>
Date:   Tue Oct 28 17:32:30 2014 +0200

    drm/i915: Redefine WARN_ON to include the condition

but for WARN_ON_ONCE. Since the kernel WARN_ON_ONCE actually picks up
*our* version of WARN_ON, we end up with messages like

[  838.285319] WARN_ON(!__warned)

which are not that helpful.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Implement WaDisableHBR2
Damien Lespiau [Wed, 11 Feb 2015 17:43:24 +0000 (17:43 +0000)]
drm/i915/skl: Implement WaDisableHBR2

v2: Use the recently introduced INTEL_REVID() and SKL_REVID defines
    (Nick Hoath)

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Tested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89554
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove the preliminary_hw_support shackles from CHV
Ville Syrjälä [Wed, 11 Mar 2015 20:52:28 +0000 (22:52 +0200)]
drm/i915: Remove the preliminary_hw_support shackles from CHV

CHV should be in a good enough shape now, so let's drop the
.is_preliminary flag.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Read CHV_PLL_DW8 from the correct offset
Ville Syrjälä [Wed, 11 Mar 2015 20:52:08 +0000 (22:52 +0200)]
drm/i915: Read CHV_PLL_DW8 from the correct offset

We accidentally pass 'pipe' instead of 'port' to CHV_PLL_DW8() and
with PIPE_C we end up at register offset 0x8320 which isn't the
0x8020 we wanted. Fix it.

The problem was fortunately caught by the sanity check in vlv_dpio_read():
WARNING: CPU: 1 PID: 238 at ../drivers/gpu/drm/i915/intel_sideband.c:200 vlv_dpio_read+0x77/0x80 [i915]()
DPIO read pipe C reg 0x8320 == 0xffffffff

The problem got introduced with this commit:
 commit 71af07f91f12bbab96335e202c82525d31680960
 Author: Vijay Purushothaman <vijay.a.purushothaman@linux.intel.com>
 Date:   Thu Mar 5 19:33:08 2015 +0530

    drm/i915: Update prop, int co-eff and gain threshold for CHV

Cc: Vijay Purushothaman <vijay.a.purushothaman@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Rewrite IVB FDI bifurcation conflict checks
Ville Syrjälä [Wed, 11 Mar 2015 16:52:31 +0000 (18:52 +0200)]
drm/i915: Rewrite IVB FDI bifurcation conflict checks

Ignore the current state of the pipe and just check crtc_state->enable
and the number of FDI lanes required. This means we don't accidentally
mistake the FDI lanes as being available of one of the pipes just
happens to be disabled at the time of the check. Also we no longer
consider pipe C to require FDI lanes when it's driving the eDP
transcoder.

We also take the opportunity to make the code a bit nicer looking by
hiding the ugly bits in the new pipe_required_fdi_lanes() function.

Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>