- We now have ::empty() for flags instead of ::none().
- PhysicalDeviceType was changed to vulkan_enum which is not Ord, so we
have to change BTreeMap to HashMap.
- The abstract holders for MemoryType or QueueFamily are gone, so more
index tracking is required. Some constructors take indices now.
- Physical devices are proper objects now, and are enumerated via the
instance. To keep things simple, this CL only replaces with equivalent
logic, but the next one should replace device indices entirely with
device UUIDs.
- Device memory errors were condensed into a single DeviceMemoryError
type. We also need to handle the VulkanError type from
enumerate_physical_devices.
BUG=b:244622199
TEST=presubmit
Change-Id: I1bba7e652ca67ecc5d9c10ddba3e1da35185078b
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3913993
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Idan Raiter <idanr@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Some types have a typo on them. For the FFI case we don't currently use
the definition so it should be safe to rename.
BUG=b:244622199
TEST=presubmit
Change-Id: I9e3c29755f0c4479b54bef206f9173ab22030ab9
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3913992
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Idan Raiter <idanr@google.com>
When sandboxing is disabled, rutabaga.external_blob is unset and blob
resources with RUTABAGA_MEM_HANDLE_TYPE_OPAQUE_FD can be mapped via
rutabaga.map() as VmMemorySource::ExternalMapping. This is already
implicitly allowed for the Virglrenderer component, which doesn't
recognize the OPAQUE_FD handle type upon export_blob() and automatically
falls back on the ExternalMapping path. Mesa CI currently relies on this
for testing Venus in Crosvm with the host lavapipe driver.
When sandboxing is enabled, rutabaga.external_blob is set and opaque
fds MUST be mapped from the hypervisor process via Vulkano as
`VmMemorySource::Vulkan` instead.
This CL makes non-sandboxed OPAQUE_FD mapping explicit, documents the
two mapping paths, and prepares for virglrenderer to support sandboxed
OPAQUE_FD mapping.
For the non-sandboxed configuration, Crosvm exports a RutabagaHandle
with handle of type RUTABAGA_MEM_HANDLE_TYPE_OPAQUE_FD upon blob
creation with virglrenderer. When mapped, the RutabagaHandle is dropped,
closing the cloned opaque_fd, and virgl_renderer_resource_map()
completes the mapping internally by looking up the resource_id and
dispatching to the context type from which it was created.
BUG=b:244591751
TEST=`deqp-vk -n dEQP-VK.memory.mapping.*` using venus with lavapipe host driver
TEST=CQ
Change-Id: I2ff32cf8868d6c9c1eb16480456cde98d376cc65
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3900320
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
We want to hide this variable from platforms that don't use it.
BUG=b:213149288
TEST=presubmit
Change-Id: I0d44d2bc50843b86d82b4084f95c8d1a1f750aa6
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3895401
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Commit-Queue: Pujun Lun <lunpujun@google.com>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Chia-I Wu <olv@google.com>
Apart from being used for arg parsing, GpuParameters is also sent
through a Tube on Windows. Since we've added a deserializer for
the context_mask field, we also need a serializer for it.
BUG=b:233676779
TEST=cargo test -p devices --features=all-linux
virtio::gpu::parameters::tests::context_mask_serialize_deserialize
Change-Id: Ic0a01acc07ac39b5962e3cfad970dae1ae8c8b27
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3906468
Commit-Queue: Pujun Lun <lunpujun@google.com>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
GLES 3.1 is not yet well tested with gfxstream backend on Windows
and we don't enable it by default. We control this via command
line arg for now.
BUG=b:213151426
TEST=presubmit
Change-Id: Ic3f8ac0e10ab988cc7bb4a5ab741ee6542617943
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3911097
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Kaiyi Li <kaiyili@google.com>
Commit-Queue: Pujun Lun <lunpujun@google.com>
If the -ldrm option is added before the virglrenderer library in the
linker command line, the libdrm symbols will not be available, since
they aren't used and will be dropped by the linker. Move the
virglrenderer deps into a helper function to avoid duplication and call
it after the virglrenderer build to get the linker options in the right
order.
BUG=b:244618790
TEST=tools/dev_container tools/presubmit --all # with wl-dmabuf feature
Change-Id: I55cc96a28a1f02e7b6081d3cfa08bc3446b8e6a1
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3905092
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
This reverts commit 84e4fbc2be.
Reason for revert: cts regression in b/247131923
Original change's description:
> rutabaga_gfx: Support for cached mappings on msm
>
> This is a stop gap solution until we have some way for the kernel to
> report mapping info to crosvm. But since on arm devices without FWB,
> the mapping attributes are the more restrictive of the S1 and S2
> mappings, for now we can just map them as cached in the S2 pgtables.
>
> BUG=b:239718180
> TEST=cts-tradefed run cts -m CtsVideoTestCases -t android.video.cts.VideoEncoderDecoderTest#testAvcGoog0Perf0320x0240
>
> Change-Id: Ib3f085fd8da768b58e7a6d8deadf3a16c7cbcc4c
> Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3832355
> Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
> Auto-Submit: Rob Clark <robdclark@chromium.org>
> Commit-Queue: Rob Clark <robdclark@chromium.org>
> Tested-by: Rob Clark <robdclark@chromium.org>
BUG=b:247131923
TEST=BasicVulkanGpuTest
Change-Id: I6566ab033840a5f9bd99dca7cbecfc1f42a5c363
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3903099
Commit-Queue: Yiwei Zhang <zzyiwei@chromium.org>
Auto-Submit: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Virglrenderer must not unmap from any thread other than the virtio-gpu
thread because virglrenderer is not thread-safe and EGL requires that
only one thread own a context concurrently.
For the VirglRenderer RutabagaComponent, we can unmap directly from the
virtio-gpu thread. In fact, we can greatly simplify the use of
ExternalMapping if we eliminate the associated map/unmap callbacks for
all RutabagaComponents.
Now VirtioGpu is responsible for ensuring that the map/unmap occurs and
is registered/unregistered. Also, the shared `map_request:
Arc<Mutex<ExternalMapping>>` spanning the KVM and virtio-gpu threads is
no longer necessary.
BUG=b:244626679
TEST=OpenGL and Vulkan gameplay with sandboxing disabled
TEST=OpenGL and Vulkan gameplay on ChromeOS with sandboxing
Change-Id: I4acaf4aabff3891536ca0439221cfbd021dba123
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3888602
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
When converting the GpuParameters to serde_keyvalue, we moved the custom
parsing code into custom deserializations functions. However, in the
case of the 'wsi' member, it looks like the same result can be obtained
by using kebab-case and aliasing "vk" to the "Vulkan" variant.
BUG=b:218223240
TEST=cargo test --features "gpu" parse_gpu
Change-Id: Ie7b892a9aebd0b0915c089473e1a114ab992202e
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3889323
Reviewed-by: Pujun Lun <lunpujun@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Auto-Submit: Alexandre Courbot <acourbot@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
'GpuParameters' are used for parsing the command-line, but are also
passed through a tube. When that happens, the default serializer turns
the 'wsi' member to "Vulkan", but the custom deserializer function
expects 'vk' in order to build it properly.
Fix this by making RutabagaWsi use the kebab-case for
serializing/deserializing so it gets serialized into 'vulkan', and
making the 'deserialize_wsi' function accept 'vulkan' as a valid option.
BUG=b:218223240
TEST=cargo test --features "gpu" parse_gpu
Change-Id: I831ddc623ff6b5ddd74dbfe82cf7ccc0f7647272
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3889322
Auto-Submit: Alexandre Courbot <acourbot@chromium.org>
Reviewed-by: Pujun Lun <lunpujun@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
While we are tweaking all of the copyright headers, let's take the
opportunity to ensure there is always a blank line after the copyright
header for consistency. (Almost all files already follow this style.)
This includes a slightly ugly regex to allow the end of a C-style
comment block after the end of the copyright:
/*
* Example comment block
*/ <-- this line
Change-Id: Idfd0855861e5ecb3d33afae942fdba908af0dcff
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3892521
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
This search/replace updates all copyright notices to drop the
"All rights reserved", Use "ChromiumOS" instead of "Chromium OS"
and drops the trailing dots.
This fulfills the request from legal and unifies our notices.
./tools/health-check has been updated to only accept this style.
BUG=b:246579983
TEST=./tools/health-check
Change-Id: I87a80701dc651f1baf4820e5cc42469d7c5f5bf7
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3894243
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Dennis Kempin <denniskempin@google.com>
Remove the deprecated timer-based fence polling strategy, leaving the
async callback-based strategy that is now in use by all Rutabaga
components.
BUG=b:175527587
TEST=glxgears & vkcube in ChromeOS VM
TEST=Vulkan & OpenGL games in ChromeOS VM
TEST=GfxBench in ARCVM
TEST=Asphalt 9 in ARCVM
Change-Id: I75714f82e36f1d6291a875fe3dd47d850ff2eb82
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3880586
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Kaiyi Li <kaiyili@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
Tested-by: Ryan Neph <ryanneph@google.com>
On Windows, we don't use import_resource_to_display() + flip_to(),
but the newly added resource_flush(). This applies to both Vulkan
and D3D backend. We would need some refactoring for the GpuDisplay
interface to create a better abstraction.
BUG=b:213149288
TEST=presubmit
Change-Id: If1d3dc543a17aa7fabf591aa2c1b87b226e8a4a6
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3869171
Reviewed-by: Kaiyi Li <kaiyili@google.com>
Commit-Queue: Pujun Lun <lunpujun@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: Pujun Lun <lunpujun@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
The exact internals of the API may change though. For example,
using a device UUID is a less error-prone than using physical
device index.
However, this is sufficient to get gfxstream external memory on
atleast some platforms.
BUG=b:235485545
TEST=compile and run ./deqp-vk with kFeature_ExternalBlob enabled
Change-Id: I10d2aaecfbcd61f383dd326184e60942755db196
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3864029
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Idan Raiter <idanr@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
This is a stop gap solution until we have some way for the kernel to
report mapping info to crosvm. But since on arm devices without FWB,
the mapping attributes are the more restrictive of the S1 and S2
mappings, for now we can just map them as cached in the S2 pgtables.
BUG=b:239718180
TEST=cts-tradefed run cts -m CtsVideoTestCases -t android.video.cts.VideoEncoderDecoderTest#testAvcGoog0Perf0320x0240
Change-Id: Ib3f085fd8da768b58e7a6d8deadf3a16c7cbcc4c
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3832355
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Auto-Submit: Rob Clark <robdclark@chromium.org>
Commit-Queue: Rob Clark <robdclark@chromium.org>
Tested-by: Rob Clark <robdclark@chromium.org>
Replace the command parsing code that used VolatileRef with the new
DataInit::read_from_prefix() function. The code is parsing a normal Rust
&[u8] slice, so it does not need to use volatile loads at all.
BUG=None
TEST=tools/presubmit --all
Change-Id: I1ee57404ae39da3400f455959e7d396f20a0d22b
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3823999
Reviewed-by: Ryan Neph <ryanneph@google.com>
Tested-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
crosvm is switching the import style to use one import per line.
While more verbose, this will greatly reduce the occurence of merge
conflicts going forward.
Note: This is using a nightly feature of rustfmt. So it's a one-off
re-format only. We are considering adding a nightly toolchain to
enable the feature permanently.
BUG=b:239937122
TEST=CQ
Change-Id: Id2dd4dbdc0adfc4f8f3dd1d09da1daafa2a39992
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3784345
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Dennis Kempin <denniskempin@google.com>
Initializing gralloc may spawn threads, so it needs to be done after
sandboxing the wl device. Initializing gralloc requires expanding the
wl device's sandbox. Rather than trying to maintain a new dedicated
minijail configuration for wl, reuse the gpu's configuration. This
should be sufficient, since virglrenderer has to open minigbm within the
sandboxed gpu process.
BUG=None
TEST=ARCVM and crostini GUI on volteer, zorc-arc-r, grunt-arc-r
Change-Id: I291fb59c665a8ba65058a6f55dee959c839bb43c
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3787936
Commit-Queue: David Stevens <stevensd@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Tested-by: David Stevens <stevensd@chromium.org>
Directly allocate dma-bufs within the virtio-wl process and remove the
VmMemoryRequest::AllocateAndRegisterGpuMemory type. This is preparation
for migrating to the SharedMemoryMapper interface.
BUG=b:201745804
TEST=Launch arcvm, launch gedit w/crostini, launch gedit w/vhost-user-wl
Change-Id: I232f1fd3dfdb8d7ed068c6b3c2ea23f35d0ddabc
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3765012
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Tested-by: David Stevens <stevensd@chromium.org>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Commit-Queue: David Stevens <stevensd@chromium.org>
Nobody ever toggles this flag, implying this flag can be removed.
By default, the gfxstream assumes ANDROID_EMU_virtio_gpu_native_sync
is present, which is the desired behavior.
BUG=b:239639751
TEST=compile gfxstream
Change-Id: I64d19c5fc2c0ac59a87ee25fbd1e7b6ed9522410
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3778078
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Auto-Submit: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Pull the declaration of the `EventToken` enum up to the top level of the
base crate, replacing the identical implementations inside sys/windows
and sys/unix.
Use the `EventToken` name consistently throughout the tree to remove the
unix-flavored "poll" nomenclature.
BUG=b:213153157
TEST=tools/dev_container tools/presubmit --all
Change-Id: I0ba42037b533b796797a7a3f6d8d7e71a5592aba
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3642673
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
When creating shared memory, 'name' can have one of two meanings. It
could either be a debugging tag with no semantic meaning, or it could
uniquely identify a shared memory object within some namespace. Linux
memfd uses name in the first meaning, whereas Windows (and Linux shm)
uses it in the second meaning.
Currently, crosvm has no use cases for the named shared memory of the
second type, so it is not supported. Make it clear that the SharedMemory
APIs treats name as a debugging-only name. Remove the "anon" and "named"
constructors, since they had no semantic meaning. Also require a name
when constructing a SharedMemory, since there's no reason not to provide
one to make debugging easier.
The only semantic change is setting the name of GuestMemory's underlying
shmem to "crosvm_guest", which it was until recently. This fixes some
ManaTEE tests which use the name to determine CrOS guest memory usage.
BUG=None
TEST=compiles
Change-Id: I78d5046df04d6f19640abbbc67af6bd433a177b0
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3676695
Commit-Queue: David Stevens <stevensd@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Flip the switch for Virglrenderer's virgl/vrend contexts to stop using
timer-based fence polling and start using async fence handling callback.
BUG=b:175527587
TEST=cargo build --features=virgl_renderer,virgl_renderer_next,gfxstream
TEST=Run glxgears and vkcube in a crosvm guest VM with virglrenderer
Change-Id: I0b309be2086225bbbb7beb0943379a22acb62a07
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3628304
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
minigbm virtgpu_cross_domain backend checks for this capset. Eventually
once this check is removed in minigbm (once everyone is using the
context-types arg to control whether cross-domain is exposed) we can
drop the dummy capset. But we need it for now.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I7c955c365f439806c797d1d38daef2e1c9b3079f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3608703
Auto-Submit: Rob Clark <robdclark@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Add a way to control exposed capsets from cmdline. This will let us
enable certain context types only in certain VMs, with a single build
of crosvm.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I8d32a22c4c198566e984b06d6aa545796752c128
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3636797
Auto-Submit: Rob Clark <robdclark@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Expose support for the new virtgpu drm native contexts.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I068e9b14ac47782c9251d9b67156755f2888d17f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3469494
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Auto-Submit: Rob Clark <robdclark@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
If the context_mask is specified, only advertise those capability sets.
If not specified, for backwards compatibility, use the older behavior.
Eventually, the older beahvior may be deprecated.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I3c37f94dfb5df25831d187deff600027aac2382d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3636798
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Auto-Submit: Rob Clark <robdclark@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
A lot of these flags can be deprecated over time. For example,
specifying the context type and "use_vulkan" is a bit redundant.
Things like "use_syncfd" and "use_guest_angle" can also be removed.
BUG=b:173630595
TEST=compile
Change-Id: I77bff7f02ebfd3b0e65145d144bfc15afa50f386
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3655286
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Commit-Queue: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
A way to control exposed capsets.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I91616405b695be73d6cc872aa5ae2a9dcecd0fb1
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3636799
Commit-Queue: Rob Clark <robdclark@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Auto-Submit: Rob Clark <robdclark@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
These variables will be left untouched if no GPU feature is enabled,
triggering a compiler warning.
BUG=None
TEST=`cargo build` does not display any warning.
Change-Id: Id1c80bd8f21272dce9e3a3f37cd3350727e5bc9a
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3659824
Reviewed-by: Junichi Uekawa <uekawa@chromium.org>
Auto-Submit: Alexandre Courbot <acourbot@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Alexandre Courbot <acourbot@chromium.org>
For cross-domain, virgl_renderer needs to be able to import a dmabuf fd
allocated by another component.
BUG=b:230100768
TEST=gfxbench, android games in arcvm
Change-Id: I9d6f965589ec24f06a83a6d921a9c00db1f057f6
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3501130
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Auto-Submit: Rob Clark <robdclark@chromium.org>
Commit-Queue: Rob Clark <robdclark@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Currently, there are two use cases for the cross domain capset:
(1) Sommelier
(2) minigbm
Niether is used by a production VM yet and it was unconditionally
enabled only to enable testing. To effectively productize, it should
be conditionally enabled, which is done later in this patch series.
BUG=b:173630595
TEST=compile
Change-Id: I736f58065c692d7ee4736f28da30fa28de43f0c8
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3655285
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Permits virglrenderer contexts to make use of per-context/per-ring
fencing.
BUG=b:175527587
BUG=b:193892617
TEST=cargo build --features=virgl_renderer,virgl_renderer_next,gfxstream
TEST=Run glxgears and vkcube in a crosvm guest VM with virglrenderer
Cq-Depend: chromium:3654708
Change-Id: I30028afa5daff2a7a1b0ade13bc9fae2cd548f8d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3628302
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
While we support both the old and new fence handling methods of
b:175527587, we need two copies of the RutabagaFenceHandler (one for
each method).
BUG=b:228051513,b:175527587
TEST=emerge crosvm
TEST=tested by gfxstream team
Change-Id: I516cf135819aebb5128e41b8a9d585e17680c6a9
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3646777
Commit-Queue: Ryan Neph <ryanneph@google.com>
Auto-Submit: Ryan Neph <ryanneph@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Kaiyi Li <kaiyili@google.com>
When the virgl_renderer and gfxstream features are not enabled (e.g. in
a default `cargo build`), the use_timer_based_fence_polling variable is
never written, so the compiler warns about an unused `mut`. Add an allow
to silence this warning (we could use cfg checks to optionally add the
mut, but that would be much more complex for no real gain).
BUG=None
TEST=cargo build # no warnings
Change-Id: I176185328322d9d65cadce602ed09d43ef9eb4e4
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3646451
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
This adds the asynchronous interrupts in crosvm-gpu for gfxstream.
This will allow gfxstream to alternate between the main signalling
method (ASG [1]) and the more traditional interrupts when it
makes sense performance-wise.
gfxstream also requires new write fence callbacks that take into
account the ring_idx and ctx_id where the fence is on.
[1] goto.google.com/address-space-graphics
BUG=b:175527587
BUG=b:192614792
TEST=cargo build --features=virgl_renderer,virgl_renderer_next,gfxstream
TEST=Run glxgears and vkcube in a crosvm guest VM with virglrenderer
Change-Id: I52b8eb0e70671e8b43bae6e9e5264268bfb076df
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3628301
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Kaiyi Li <kaiyili@google.com>
An optional performance optimization that almost completely avoids the
wakeup cost of timer-based fence polling when all RutabagaComponents use
async_fence_cb. There is still a small overhead from the per-loop
virtio_gpu.needs_fence_poll() tests. This overhead will be removed in
later CLs, when timer-based fencing support is removed.
BUG=b:175527587
TEST=cargo build --features=virgl_renderer,virgl_renderer_next,gfxstream
TEST=Run glxgears and vkcube in a crosvm guest VM with virglrenderer
Change-Id: Id9f14bcbdaa4dcdc42b24be12d85cad2ea0efed5
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3628300
Commit-Queue: Ryan Neph <ryanneph@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Adds another method for fencing where a rutabaga component may provide
an asynchronous callback for fence completion. When this method is
enabled, fences no longer need to be polled on the main thread.
When async_fence_cb is enabled, optional polling still occurs for
Rutabaga Components that still rely on it for other purposes (e.g.
virglrenderer for GL query checking).
Both the old and new methods of fence handling are supported with this
CL; e.g. timer-based polling and async callback, but timer-based polling
is the only one active. Follow-up CL(s) will switch to using the async
callback for each RutabagaComponent. Support for the old timer-based
polling will be removed after the async callback method stabilizes in
production.
Also, use a BTreeMap rather a HashMap since we only expect a dozen or so
entries at most. In such cases, a BTreeMap is faster.
* v1 (lfrb@collabora.com): remove all polling + add async_cb
* v2 (ryanneph@google.com): re-introduce optional polling to fix
virglrenderer that relies on it for GL query checking.
* v3 (ryanneph@google.com): replace timer-based polling with
eventfd-based poll() signaling for components that want to use it.
* v4 (ryanneph@google.com): add async fence handling beside timer-based
polling to support both for a short time.
This is a pseudo-reland of commit
a999284da2, but modifications have been
made to support both fencing methods simultaneously.
BUG=b:175527587
TEST=cargo build --features=virgl_renderer,virgl_renderer_next,gfxstream
TEST=Run glxgears and vkcube in a crosvm guest VM with virglrenderer
Change-Id: I28f754aeef01972a01dab7990267415814a36c77
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3628299
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Ryan Neph <ryanneph@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Follow-up to commit 13108989b1 to fix
broken rutabaga ffi after introducing the new context_name parameter
to create_context().
BUG=b:230678191
BUG=b:230788302
TEST=emerge crosvm
Change-Id: If1165662e66dd349ef4c52bece26fe724a861b78
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3621210
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Ryan Neph <ryanneph@google.com>
RutabagaComponents receive an Option<String> with calls to
Rutabaga::create_context(). This is desirable for virglrenderer so that
the guest application name can be used by the host renderer,
particularly to enable application specific GPU driver configs based on
program name matching. It is ignored by other RutabagaComponents for
now.
BUG=b:230678191
TEST=emerge crosvm
TEST=vkcube in VM
Change-Id: Ic08a02040ecc1d2e62795c248a66a88394b268e4
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3612467
Commit-Queue: Ryan Neph <ryanneph@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
