Merge "drm_hwcomposer: Add Dump() to the Compositor interface" into mnc-dev

drm_hwcomposer: Add Dump() to the Compositor interface

Add a Dump() function to the Compositor base class and pipe it
through hwcomposer. This can be invoked from the shell by
running "dumpsys SurfaceFlinger".

Change-Id: I87646f4e926ac6d52a04ba130957abdfab99c491
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Don't compile all importers at once

Choose based on the BoardConfig flags.

I'd like to revert this once drm_gralloc is available
in all working branches.

Change-Id: I171d8f9146bdca0f0671092b8e539031c2f64f0f
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Initialize DrmProperty type_ member

Initialize it to an invalid value instead of INT.

Change-Id: I5d75b9a3696a748d6352d918263b482034a22001
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Fix case on DrmCrtc mode_valid_ member variable

Change-Id: If2a431a4a18ea7908654a4a92a4a6acca8ed92d1
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Move event worker into VSyncWorker

Using the new Worker interface, split out the event worker
now that it just handles vsyncs.

Change-Id: I7ebc6237b10411fdba3b8826935921b8c83c4b6c
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Use hw planes + drm atomic interface

Replace the basic, single overlay, modeset/flip implementation with
a new Compositor instantiation for drm. The new compositor uses the
drm atomic interface to composite layers to multiple hardware planes.

This change also introduces an importer argument in
Compositor::CreateComposition. By specifying the importer, Compositor
instances are responsible for cleaning up buffer objects submitted
via Composition::AddLayer.

Change-Id: Ic292829cd93475d754294b00428de132301092b2
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Add Importer to CreateComposition()

The Compositor will be responsible for releasing buffers that
are successfully added to a composition using AddLayer. To achieve
this, the Importer of the buffers must be provided to Compositor/Composition.

Change-Id: I1f0acd70c245b7a2b6c93649f247e30e7f2e9dc4
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Wrap the importers in a class

This will allow us to move the importer functionality into the
drmcompositor.

Change-Id: I4274ebd1776c4d7879342c54a2b7c4095ebc50f4
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Allow importer to be specified from BoardConfig

Use BOARD_DRM_HWCOMPOSER_BUFFER_IMPORTER from BoardConfig to
determine which importer to use.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Ia081628830b78187ea1f8a3e3e4260188e130324

drm_hwcomposer: add overlapping rectangle seperation algorithm

Change-Id: I40734ce77191ac37fb2a2c156d6010530df0434b

drm_hwcomposer: Add Worker class to abstract thread nasties

This will allow us to spin up threads more easily.

Change-Id: I74add3d20ab14a61f3c646877a3ee17d0bd99ac4
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Set requires_modeset() to false after modeset

Reset requires_modeset on a crtc once the initial modeset is
completed. Without this, we'll just do modesets instead of flips.

Change-Id: I998914c914ccb97346ed194a993c703f387bf93b
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Store displays in std::map

Instead of a fixed length array, use a std::map for
displays.

Change-Id: I77a4893f3051d6af4030e3208d096b5796796c84
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Fix up #include ordering

Change-Id: I9d808389c60b9d77c8e03a09aeaf3f309353cde8
Signed-off-by: Sean Paul <seanpaul@chromium.org>

Merge "drm_hwcomposer: Wrap libdrm ops (minus modeset/flip) in C++ classes" into mnc-dev

drm_hwcomposer: Wrap libdrm ops (minus modeset/flip) in C++ classes

This allows us to compartmentalize a bunch of code/logic from
hwcomposer.cpp into drm classes.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Id3f912126f1fdcd44d32c3eb4fba646f77590278

Merge "drm_hwcomposer: add GLES compositor" into mnc-dev

drm_hwcomposer: add GLES compositor

This patch adds the GLES based compositor. This doesn't actually make
the hwcomposer use it, which will come in a later commit. This
version of the compositor uses syncpoints optimally and has no
fallback for the case where too many (32 currently) layers overlap.

(cherry picked from commit 6996227fe43bcf78c6e692e8098bdddb95bfda6e)

Change-Id: I3503cb5d1b309ae06d1ed2a8d244640ad81eda9e

Add missing include for calloc/free.

Bug: http://b/15193147
Change-Id: Ibb2fb6fef4ca6a23eb2f53111c4b01847fe468b2

Stop using stlport.

This was dead a long time ago.

Bug: http://b/15193147
Change-Id: I9de3ad7fb14f987d36823e6fc4a60a33c92ddf4f
(cherry picked from commit 59368c2ccea1b108a580cab11443ca9630658cb7)

drm_hwcomposer: Update style to conform with Google Style Guide

Change-Id: I9b1a7dbcb9b8c6b0a6eb3c6416b5e4898a5097d8
Signed-off-by: Sean Paul <seanpaul@chromium.org>

Merge "drm_hwcomposer: add a .clang-format for our style"

Merge "drm_hwcomposer: Gate compilation on BOARD_USES_DRM_HWCOMPOSER"

drm_hwcomposer: Gate compilation on BOARD_USES_DRM_HWCOMPOSER

To avoid compiling on unwanted platforms.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Iec8122d32c25ea10fc68fa78e6e051e186ff0efb

drm_hwcomposer: Gate compilation on BOARD_USES_DRM_HWCOMPOSER

To avoid compiling on unwanted platforms.

Change-Id: Iec8122d32c25ea10fc68fa78e6e051e186ff0efb
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: add a .clang-format for our style

Change-Id: If6059a0456376d2aba6ad09af28fbb7add33f0b4
(cherry picked from commit fd425aec49615e749e1edc596384a3502ea0db05)

drm_hwcomposer: add compositor interface header

Change-Id: I4b743e442e140f8fe5452fba3dd718995fe77b1b
(cherry picked from commit af3c4ecc830bef9d538333d0bf200fed39ae3649)

drm_hwcomposer: Fix missing header

Signed-off-by: David Riley <davidriley@google.com>

Merge "drm_hwcomposer: Introduce a new event worker to handle vblanks/flips" into lmp-mr1-dragon-dev

drm_hwcomposer: Introduce a new event worker to handle vblanks/flips

Instead of having a dedicated vsync worker for each display, factor out
the drm fd wait from the set workers into a global event worker. This
worker will dispatch vsync() callbacks to SurfaceFlinger, as well as
unblock the appropriate set worker when flips complete.

Change-Id: I169132b2a93a2c9feabcab116aab4e61861166e1
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: fix new then free error for hwc_context_t

Change-Id: Id987b9eebe818ded689dbdfd1c695118b0a81d5e

drm_hwcomposer: Fix header compilation for C

Change-Id: I07ff3f16f3542071b42cfaa6b294017e54a84c4a
Reviewed-on: https://chrome-internal-review.googlesource.com/203636
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Cache FB and gem in gralloc buffer

Move drm fb object creation to the importer.

In NV importer, use GRALLOC_MODULE_PERFORM_SET_IMPORTER_PRIVATE
and GRALLOC_MODULE_PERFORM_GET_IMPORTER_PRIVATE to cache the hwc_drm_bo
in the gralloc buffer.  This avoids the need to recreate fb objects
again each frame, and also greatly simplifies managing the lifetime of
the gem handles.

The drm_gralloc importer does not support caching at this point.  In an
attempt to keep it somewhat working, add hwc_import_bo_release function
that hwc calls when it's ok to rm fb.  If hwc_import_bo_release returns
true, then hwc will clean up gem handles as it did before.

We should consider doing a follow-up patch that adds fb/gem caching to
drm_gralloc importer also.  Then some of the code that is kept in this
patch for backwards compatibility can be removed.

Change-Id: I92857dcaddf8cea219ebc041e16ef76087a1b696
Reviewed-on: https://chrome-internal-review.googlesource.com/200895
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Initialize timeline_next to 1

Initialize timeline_next to 1, because point 0 will be the very first
set operation. Since we increment every time set() is called,
initializing to 0 would cause an off-by-one error where
surfaceflinger would always composite on the front buffer.

Change-Id: I1e67c36864716595eb489286a7db3f1dcb04f67e
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Reviewed-on: https://chrome-internal-review.googlesource.com/200735
Tested-by: Alexandre Courbot <acourbot@nvidia.com>
Reviewed-by: Alexandre Courbot <acourbot@nvidia.com>
Reviewed-by: Allen Martin <amartin@nvidia.com>
Reviewed-by: Sean Paul <seanpaul@google.com>

drm_hwcomposer: initialize timeline_next

Initialize timeline_next to 0

Change-Id: I5d9f051ac659b06be19f5e5eaaab8a665f10832c
Reviewed-on: https://chrome-internal-review.googlesource.com/200646
Reviewed-by: Sean Paul <seanpaul@google.com>
Commit-Queue: Sean Paul <seanpaul@google.com>
Tested-by: Sean Paul <seanpaul@google.com>

drm_hwcomposer: move AddFb/RmFb to reference gem handles

Move drmModeAddFB2 and drmModeRmFB to ensure a reference to the gem
handles are taken while the buffer is in use.

Change-Id: Ib9afdb91144850480ecaac9d73601d887af5fa12
Reviewed-on: https://chrome-internal-review.googlesource.com/200686
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Check against gem handles for popped buffer

Add check before closing gem handle for most recently popped buffer in
addition to buffers in queue.

Change-Id: Icb096bd4db439b1195a123e28468142829fc61ea
Reviewed-on: https://chrome-internal-review.googlesource.com/200725
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Do not close gem handle if queued

Fix an issue where if the same gem handle was queued twice we would
close the handle when the first one came out of the queue, leading to
a failure when calling drmModeAddFB2() on the second buffer.

Change-Id: I0bd35158215f74b51111d2e259bca3ba5125b203
Reviewed-on: https://chrome-internal-review.googlesource.com/200715
Tested-by: Allen Martin <amartin@nvidia.com>
Reviewed-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Support Nvidia gralloc

When using an Nvidia gralloc implementation, ask Nvidia gralloc to import the
buffer_handle_t to the hwcomposer's drm device.

Arguably, the code to perform the import would more naturally belong to
drm_hwcomposer than Nvidia gralloc.  But letting Nvidia gralloc do it has two
advantages.  First, drm_hwcomposer doesn't need to include vendor headers or
link against vendor libraries.  This means that it's easy to support both
drm_gralloc and Nvidia gralloc from the same default build of drm_hwcomposer.
Second, Nvidia gralloc will take care of properly setting up any vendor
specific metadata (such as tiling parameters) of the imported GEM handle.

Change-Id: I0f77fb301e7e0bff2a8cd7befeae4bc4892220ad
Reviewed-on: https://chrome-internal-review.googlesource.com/195115
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Support other drm nodes than card0

On dragon, card0 = Nouveau, card1 = TegraDRM.

Change-Id: Ie69639e40162ceb4ab99025dd9406ae44b7204c8
Reviewed-on: https://chrome-internal-review.googlesource.com/196021
Reviewed-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Tested-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Fix fence leak

Hwcomposer is responsible for closing the acquire fences passed
in during hwc_set.  Make sure that it does so, even if there is
an error condition.

Change-Id: I5c9096804c2613ec317c306dd361cb5f1a2ba5b2
Signed-off-by: Lauri Peltonen <lpeltonen@nvidia.com>
Reviewed-on: https://chrome-internal-review.googlesource.com/198356
Tested-by: Stéphane Marchesin <marcheu@google.com>
Commit-Queue: Stéphane Marchesin <marcheu@google.com>
Reviewed-by: Stéphane Marchesin <marcheu@google.com>

drm_hwcomposer: Remove drmSetMaster call

Now that gralloc is using a render node, we can remove
this call. This allows us to run as a non-root user as
well.

Change-Id: I46c9c16c7a9a169378e567d4973172f8392d441b
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Use prime to import gem handles

Change-Id: Id5d5f9a61f092f33130fed4d8b8d00fcaf401dea
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Include stdlib.h in the importer

Change-Id: I3fffc5e62c07de9561b6f998700a98862ebd5fc8
Signed-off-by: Sean Paul <seanpaul@chromium.org>

drm_hwcomposer: Set display config on initialization

We shouldn't wait for surfaceflinger to set active config, initialize
the display with a config. For now we just choose the first config in
the list. This should change to a more sane choice, such as the preferred
mode or something similar.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Ib063080c282cec28c9b0218c0ba5151d48b367ce

drm_hwcomposer: Ignore NULL display_contents

surfaceflinger will send in NULL display_contents in prepare/set,
make sure we don't blow up.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Ia5085d4bd5e5928b52d81d14d027d0a162995946

drm_hwcomposer: Don't try to get configs for a disconnected connector

Short circuit getting the configs for connectors which are disconnected.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Ieabbf08f4d4a6c8b03f948399113e3ad04a2d93a

drm_hwcomposer: Store active mode instead of config index

Since the config indices can change out from under us, store the
actual mode instead of an index. This avoids having to
reset active_config when we get a new list from the driver.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I09a1ffbdde16c44ccc8a1d2865f09a5da22c2e26

drm_hwcomposer: Always perform a setcrtc initially

Always do a setcrtc on the first set(), this ensures things
are set up the way we expect them to be.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I4279d4de64339e72f0e7f9b315b2b9695704c78c

drm_hwcomposer: Implement eventControl with vsync worker

Adds a new worker to handle events requested by eventControl.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I6d63bb43b5f112e2efb0c426e7fd59910fc98953

drm_hwcomposer: Implement release and retire fences using sw_sync

Once drm supports passing sync through, we'll move onto that.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I78bd5d380212b4aaec4b41dcba6cc882f1e5cc20

drm_hwcomposer: Reduce time spent in display_lock

This reduces the amount of time set will block on the
worker.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: If5d110ce78e98642265ffa7dcc25710bc227330b

drm_hwcomposer: Use queue buffer for pending sets

Instead of just using a front and back buffer, use a queue buffer
to allow more than one pending set.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I451acbcb08fac5191b76ec2e372fe612d227167c

drm_hwcomposer: Split gralloc out into an importer

Add the concept of an importer so we can plug in different
sources of bo.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I4f741ef4fa7c44e9cb31db61a146aed273854a69

drm_hwcomposer: Add a new set thread to post bo's

Adds a new set thread to wait on acquire fence, flip the bo,
and wait for it to be presented on screen.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: Ia931a7cbe287f4c744068fc80e476f81880bf608

drm_hwcomposer: Initial implementation

This adds a very basic hwcomposer implementation on top of
libdrm.

Signed-off-by: Sean Paul <seanpaul@chromium.org>
Change-Id: I472feb9bc08a4c4cd9849370c849d266ea58da9c

Initial empty repository