There can be periods in the platform version, so it's not safe to use.
Just run PLATFORM_SDK_VERSION instead.

Bug: 131089111

Change-Id: I72c0af3407f88753cf69b2ba2565b221345d3f3a

some gralloc0 entry points not defined for older api levels

bug: Bug: 131089111

Change-Id: If3f1843ffb2c008052a5dec87b9063d2500298f1

Add validateBufferSize and getTransportSize

Bug: 131089111
Test: build, boot
Change-Id: I9820197865e47832ae209cbf1cac3b544f45e60e

This is not a must and adds confusion between
OpenGL (supported) and Vulkan (not supported)
they are mixed (e.g. AHardwareBuffer_allocate
which goes though OpenGL)

Bug: 131184223
Test: lunch sdk_gphone_x86-userdebug
Test: emulator -wipe-data -no-snapshot -show-kernel -feature Vulkan,GLDirectMem
Test: ./android-cts/tools/cts-tradefed -- run cts -m CtsGraphicsTestCases
Test: emulator -wipe-data -no-snapshot -feature GLESDynamicVersion,PlayStoreImage,KernelDeviceTreeBlobSupport -gpu host
Test: ./android-cts/tools/cts-tradefed -- run cts -m CtsNativeHardwareTestCases
Test: ./android-cts/tools/cts-tradefed -- run cts -m CtsGraphicsTestCases
Test: ./android-cts/tools/cts-tradefed -- run cts -m CtsCameraTestCases
Change-Id: Idb5d5885efd18219c86cc1b76722d41f49d68d03
Merged-In: Ie6f40c717a36ae4a9ef477cc420dc083b63ffbc3
Signed-off-by: Roman Kiryanov <rkir@google.com>

HAL_PIXEL_FORMAT_YCbCr_420_888 does not specify the exact
buffer layout. EmulatedFakeCamera3.cpp uses interleaved
UV planes, while other places place them separately.

Bug: 130295800
Bug: 129974968
Test: run cts -m CtsCameraTestCases \
      -t android.hardware.camera2.cts.ImageReaderTest#teutYUVResolutions
Merged-In: If3a495aa794d9ab09288b7b92be7258a07cc077d
Signed-off-by: Roman Kiryanov <rkir@google.com>
Change-Id: I33f6777b5cca6f4fa0c61ac3db994230d6b6efb0

error: 'class std::vector<char>' has no member named 'data'

Bug: 130246090
Test: make
Change-Id: I1c7146a6eada5be4aa0678c7a3ddb762d080c981
Merged-In: I39eec3b4d6621067387e48e1481de3cd614b79a1
Signed-off-by: Roman Kiryanov <rkir@google.com>

[Goldfish-opengl] Add empty function definition for alloc_device_t::dump() am: 9160f9f2fa am: 7359abbf6d
am: 65576fcaca
BUG: 125540538
Change-Id: I00aeb57a840ca5932011a3ae01d4f089c9bf8a2d
(cherry picked from commit 1a85198d3287a31734f62f3f83b97c1c86f66fa9)

To reduce possible memory leaks.

Bug: 128324105
Test: make
Change-Id: I5e1a3d2a0e0b0063878143f4362fd7ff217e337a
Signed-off-by: Roman Kiryanov <rkir@google.com>

Bug: 128324105
Test: make
Change-Id: I1d5afe38b06cfb296480865dc1211f0d7b21e806
Signed-off-by: Roman Kiryanov <rkir@google.com>

There was "delete cb" without decrementing the DMA counter which
could lead to a leak.

Bug: 128324105
Test: make
Change-Id: I7307c133a84132e76648fbfe1e60e49a5ef477dc
Signed-off-by: Roman Kiryanov <rkir@google.com>

The code never reads this field.

Bug: 128324105
Test: make
Change-Id: I788c07d1a7b84990fa161ecb8c2d324b46d60c63

Bug: 128324105
Test: make
Change-Id: Id57fc7358e071e9775236c7fe596dc6c94923482
Signed-off-by: Roman Kiryanov <rkir@google.com>

Bug: 128324105
Test: make
Change-Id: I42b3afe60ee515fe5db439774c425430812476eb
Signed-off-by: Roman Kiryanov <rkir@google.com>

s_grdma is a global variable and could be set
even if there is no DMA support.

Bug: 128324105
Test: make
Change-Id: I8b963fb3364bbe64779bc7d2c9e90632f3ca6700
Signed-off-by: Roman Kiryanov <rkir@google.com>

Bug: 128324105
Test: make
Change-Id: I06c5d830880711a015024e424e219e797fcb5a50
Signed-off-by: Roman Kiryanov <rkir@google.com>

Not used.

Bug: 128324105
Test: make
Change-Id: I48d078d2df1e87cb2d00eaa201c4c161867f2132
Signed-off-by: Roman Kiryanov <rkir@google.com>

To make further refactoring easier.

Bug: 128324105
Test: make
Change-Id: I1bdbb1d61667899e1a215757b98a6580c8edafed
Signed-off-by: Roman Kiryanov <rkir@google.com>

This change adds rcReadColorBuffer to
gralloc_lock_ycbcr from gralloc_lock, fixes the memory
layout for HAL_PIXEL_FORMAT_YCbCr_420_888 and reduces
copy-paste.

Bug: 128451131
Bug: 127995433
Test: CtsNativeHardwareTestCases \
      android.hardware.nativehardware.cts.AHardwareBufferNativeTests
      #SingleLayer_ColorTest_GpuColorOutputAndSampledImage_Y8Cb8Cr8_420
Signed-off-by: Roman Kiryanov <rkir@google.com>

Change-Id: Ib961cfe6c662173e95fd00e055c3c7bdfb5d027d

Now we moved up the host connection define,
but it's not included in any non GOLDFISH_VULKAN build. Oops

Change-Id: I3053a13436775450489d3a42a1740403ac707b95

* libcutils log is deprecated
* Also fix a typo in a debug format string

Test: m hwcomposer.ranchu
Change-Id: I0a8e6434c5e1081ccf7663e20279774632e015d7

bug: 122080810

It can be possible to create AHardwareBuffers
with a Vulkan external buffer.
Currently, we rely on there being one ColorBuffer
for each such AHardwareBuffer, except
when blob format is involved.

This CL causes a ColorBuffer to be created
even if the format is blob format.

Then, we get a ColorBuffer ID with which to correlate
Vulkan external buffers.

Change-Id: I8109b15d1e750b72d262c9d7aa33c4726244796f

Change-Id: Ib21e305c80d69bbd5f4ef52004c4dd86b9e08462

This reverts commit 1b4bcf1e.

Change-Id: I11947e45ea81b33ec67b48f8262902ee3f76728c

reduce jank

Change-Id: I158dbf5e43e9413c40332c831409431f7f35ae42

RefCount Pipe is a different approach to simplify the Color Buffer
management on host. In guest, the handle returned by gralloc
allocator can be passed around between processes through Binder.
Android system ensures that the handle on both calling and receiving
end willl be properly closed. Thus, when gralloc allocates buffer for
the first time, we establish a Refcount pipe between guest and host,
and assign its fd to the handle. when cloned handle is closed, the
duplicated fd will also be closed. Guest kernel will decide when to
really close the fd and to trigger the pipe's closing on host.
On the host side, we only need to destroy the color buffer immediately
and not execute any reference counting code path.

This approach will not only simplify the logic between guest and host
but also reduce memory leaks.

BUG: 116349069

Signed-off-by: Weilun Du <wdu@google.com>
Change-Id: Iab9fe76f5947b914b11fcdccdb61018957e20e50

This happens only for host side tests where we free and unregister
in the same process. Post-Treble Android usually expects not to do this.

However, this CL should not affect behavior of guests either pre or post
Treble.

Change-Id: I4fac91c6d4c7031ac19e7c75233451663e3d31c4

bug: 116981035

fb device was previously allocated with malloc
and freed with delete, not free.

Uncovered from running goldfish-opengl host side test
wih ASAN enabled.

Change-Id: I46b97d212e0baf47cd3d181f1b75ce27cb955fad

Change-Id: Ib130765d39178f0a4f4a482e5555d26c076e33ff

bug: 116712605

- No libdl needed on host
- Define PAGE_SIZE
- Include gralloc in build

+ Consolidate a few build script conditionals
  in egl Android.mk
+ Remove unused Callstack.h

Change-Id: I2e5f8050971d41e7ae26910ce6bd43dad114e5c5

Change-Id: I1f6a14b7bef2dc390f4ddf905b9ebb728adfc968

- Make the thread id getting platform agnostic
- Make QemuPipeStream run either with a real pipe device or
a host-side one
- Split out host versions of stuff, like goldfish_dma_host.cpp

bug: 116524589

Change-Id: Iac8c3161d68e9e71f0e14adecf42ca982d48628e

For certain guest process which maintains a thread pool and only calls gralloc
library, we should exit the host connection to avoid spawning too many
render threads on host emulator.

Test: run empty test activity for 3000 times and the number of render
thread on host should stay roughly the same.
BUG: 80050618

Change-Id: Idb282aa67138572519edee761f8bc6cb50c1653d
Signed-off-by: Weilun Du <wdu@google.com>
(cherry picked from commit 904b90eb097609762c7c35c321f82b2cc5245190)

Gralloc module may be used by CameraHal(32bit) and Apk(64bit).
Counting offset with int_ptr results in different values.

Bug: 73485161

Test: run cts -m CtsCameraTestCases
Change-Id: I7d8d3f20652107df1fd7812d30c01e4546ac4a22

This change makes fields of fixed length to make safe to pass
goldfish_dma_context between 32 and 64 bit environments.

Test: ~/repo-emu-master/external/qemu/objs/emulator \
Test:     -verbose -show-kernel \
Test:     -kernel /tmp/qemu-kernel-x86_64-ranchu/kernel-qemu \
Test:     -feature GLDMA -wipe-data -no-snapshot
Bug: 110152998
Signed-off-by: Roman Kiryanov <rkir@google.com>
Change-Id: I8e2aaf2d6896f189251099dabdc49d86053b64c6

This reverts commit 20ccec25.

Reason for revert: Break sdk_addon build

Change-Id: I3b0ea58ab44324014c34bf30157843415ffc456e

This commit replaces <cutils/log.h> and <utils/Log.h> with <log/log.h>.

Background:
<cutils/log.h> has been moved to <log/log.h> for a while.  Both
<cutils/log.h> and <utils/Log.h> simply includes <log/log.h> for
backward compatibility.  This commit is a part of the effort to remove
<cutils/log.h> and <utils/Log.h> from the source tree eventually.

Bug: 78370064
Test: lunch aosp_arm-userdebug && make && \
      cd device/generic/goldfish-opengl && mma

Change-Id: Ia80adc525617d7099e9f354cef035386ec510262

For certain guest process which maintains a thread pool and only calls gralloc
library, we should exit the host connection to avoid spawning too many
render threads on host emulator.

Test: run empty test activity for 3000 times and the number of render
thread on host should stay roughly the same.
BUG: 80050618

Change-Id: Idb282aa67138572519edee761f8bc6cb50c1653d
Signed-off-by: Weilun Du <wdu@google.com>
(cherry picked from commit 904b90eb097609762c7c35c321f82b2cc5245190)

Gralloc module may be used by CameraHal(32bit) and Apk(64bit).
Counting offset with int_ptr results in different values.

Bug: 73485161

Test: run cts -m CtsCameraTestCases
Change-Id: I7d8d3f20652107df1fd7812d30c01e4546ac4a22

This change makes fields of fixed length to make safe to pass
goldfish_dma_context between 32 and 64 bit environments.

Test: ~/repo-emu-master/external/qemu/objs/emulator \
Test:     -verbose -show-kernel \
Test:     -kernel /tmp/qemu-kernel-x86_64-ranchu/kernel-qemu \
Test:     -feature GLDMA -wipe-data -no-snapshot
Bug: 110152998
Signed-off-by: Roman Kiryanov <rkir@google.com>
Change-Id: I8e2aaf2d6896f189251099dabdc49d86053b64c6

This is an appropriate fix for HAL_PIXEL_FORMAT_RGBX_8888.

The handling of RGBX_8888 is very subtle. Most of time we want it to
be treated as RGBA_8888, with the exception that alpha is always
ignored and treated as 1. The solution is to create 3 channel RGB
texture on the host while telling the guest to use 4 channel RGBA for
read/write. The host GL will handle the Alpha channel approriately.

BUG: 78602661

This cl does not impact real devices.

Test: atest CtsNativeHardwareTestCases
Change-Id: Ife299bfc55ce327d0213616b9902ba537cfb0d82