Have all devices trigger interrupts via the Queue rather than directly.
This way we can skip sending the interrupt entirely when notification
suppression features have been negotiated with the guest kernel.
BUG=none
TEST=crostini.Basic.buster_stable
Change-Id: Ica6f978127aa648fd983f641518940d7a857916f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/3026686
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Chirantan Ekbote <chirantan@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
A completed fence callback is set in Rutabaga. When the callback
is executed, the descriptor associated with the fence is immediately
marked as used and the control queue is signalled.
As of now, the callback is still called in the main worker thread
when poll_fence() is executed, but once we enable the async callback
feature in virglrenderer, the callback might be called from another
thread. This is the reason everything is protected with mutexes.
The completed_fences hash map keeps track of the latest completed
fence for each context to avoid any race condition.
Signed-off-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Change-Id: I6f3f55d21b7f4722721bdc2b16da1b39bae4ff7e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2845984
Reviewed-by: Kaiyi Li <kaiyili@google.com>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Cuttlefish's streaming server, which acts as a Wayland compositor
in order to receive display framebuffers from Crosvm, needs some
mechanism to tell which Wayland surface corresponds to which display
(a "display" is a "scanout" in virtio-gpu terminology).
Wayland object ids can not be directly used for this as all Wayland
objects share a single global id space (so the first created Wayland
wl_surface surface object may have id = 15).
Previously, the case of unchanging displays was handled by enforcing
the creation order of surfaces within Crosvm so that Cuttlefish's
streaming server (which is a Wayland compositor) could assume the
creation order corresponded to the display order. However, this still
experienced issues (b:186580833) when surfaces were destroyed and
later recreated when handling `set_scanout(..., resource_id = 0)`
commands.
There is also an ongoing effort to support adding and removing
displays at runtime in (see aosp/1671968) which experiences the
same issue. When surfaces are arbitrarily created and destroyed,
Cuttlefish's streaming server has no way to determine which Wayland
surface corresponds to which display.
To solve all of this, this change introduces an extension to allow
Wayland clients (Crosvm) to attach additional metadata (scanout_id)
to Wayland objects (surfaces) so that Wayland compositors (Cuttlefish's
streaming server) can exactly determine which surfaces correspond
to which displays. I will attempt to upstream this protocol (tracked
in b:191901112).
BUG=b:188904670
BUG=b:187351899
BUG=b:191901112
TEST=launch Cuttlefish with single display
TEST=launch Cuttlefish with multiple displays
TEST=launch Cuttlefish and hotplug some displays
Change-Id: I2aa4b714a49e4d85b6a3c705ba0d5bc1720b838e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2909903
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@chromium.org>
Commit-Queue: Jason Macnak <natsu@google.com>
Updates virtio-gpu frontend to track multiple
scanouts and the resources set for those scanouts.
Adds a --gpu-display command line arg to avoid having
to create a very complicated --gpu string.
BUG=b:173523402
TEST=launch Cuttlefish w/ multiple displays
TEST=(see change series at aosp/1652814)
Change-Id: I73174c11f35f865b8b67ae77d8169d6812f85535
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2836265
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Reviewed-by: Dennis Kempin <denniskempin@chromium.org>
Commit-Queue: Jason Macnak <natsu@google.com>
It allows Rutabaga library users to set a fence callback on the
component builder or while creating a context. The handler is
called when a fence is signaled as completed and could
be executed from another thread.
Signed-off-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Change-Id: Ia9176bf2f0822a0076bed6a7a09e709b426aa620
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2845982
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Stephen Barber <smbarber@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Added preliminary version of keyboard and pointing device routine
for wayland implementation. The pointing input is wired as a multi
touch device. Due to the fact that wayland client is callback based,
all the necessary incoming events are serialized and stashed in the
temp circular buffer and then processed afterwards from the main event
loop.
Known issues:
1. Mouse input can't be properly wired inside the guest as a mouse
device without pointer locking, but this is not what we want. The
approach emulates it as a multitouch device, but, of course, it
implies limitations in functionality. Limitations include cursor
in the VM that doesn't move in unison with the host cursor.
2. I kept the mouse cursor surface since it's not decided yet which
approach for handling pointing input device will be used (see #1).
Removing the mouse surface in the guest would remove the lagging
guest cursor. The alternatives to the multi-touch device are:
"- Relative mice (e.g. a typical PC mouse). These are relative
devices, meaning they send deltas from the current cursor
position. Some apps like games rely on these events.
- Touchscreens (multitouch, single touch). These are absolute
devices, and are much easier to implement seamless guest/host
input for.
- Touchpads (these are absolute devices). I'm not sure these are
really compelling for any use case." -nkgold@
3. This code is for POC purpose only, so there are still lot of minor
issues and negligence in it.
Looking forward for your comments and proposals.
BUG=b:177939148
TEST=crosvm $ARGS \
--display-window-keyboard \
--display-window-mouse \
--gpu=3d,glx=false,egl=true \
--wayland-sock=/run/user/1000/wayland-0 \
$OTHER_ARGS
Change-Id: If4a9b73b8da4e0cc52fa619bbd6e5588ccdb7874
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2688439
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
There is a desire for Wayland and possibly other display backends to
reasonably handle input. Move the event device logic inside the X11
backend up to the common layer to prevent duplication.
The common layer also keeps track of surfaces and external memory
objects to make this easier. The GpuDisplaySurface/GpuDisplayMemory
traits are introduced in case the common layer needs to perform
compositor specific operation.
BUG=b:173630595
TEST=compile and run with X11
Change-Id: Ied060a7cc216ac6c084030aad1fc839c022a3395
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2852523
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
- Removes duplicated code
- Assumes "u32" as the type of the flexible array
length field, which should be fine.
BUG=b:173630595
TEST=compile and run
Change-Id: Ia2d9f74b823e06d2f2f0b60cebb8a3b999a9cba0
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2848283
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Auto-Submit: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
As new features are about to be added, it only makes sense to
improve error handling first.
Also improve/update naming of errors in other GPU-focused areas.
BUG=b:173630595
TEST=compile and run
Change-Id: If0d4f8b7d548c46f0a15b64699502e0fefeaae3e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2844350
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Auto-Submit: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
In this CL, the decoder gets the buffer modifier via resource_bridge,
and then passes the value to the backend.
BUG=b:79682290
TEST=run media CTS and check the modifier is passed to GAVDA
Cq-Depend: chromium:2820734
Change-Id: I27b845251bf4a12efb74193945a6cb952c8f14af
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2821108
Tested-by: Chih-Yu Huang <akahuang@chromium.org>
Commit-Queue: Chih-Yu Huang <akahuang@chromium.org>
Reviewed-by: David Stevens <stevensd@chromium.org>
Reviewed-by: Alex Lau <alexlau@chromium.org>
Reviewed-by: David Staessens <dstaessens@chromium.org>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Because the default value for mode is GpuMode::ModeVirglRenderer, make
the default value for use_vulkan false as well.
Set use_vulkan to true when mode is set to GpuMode::ModeGfxstream for
backward compatibility.
BUG=b:178104043
TEST=cargo build
Change-Id: Idf1417f04d23999cf5a03b0bf640973b69de93e7
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2823012
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: David Riley <davidriley@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Chia-I Wu <olv@google.com>
Commit-Queue: Chia-I Wu <olv@google.com>
Having a trait for interrupts used by queue and the devices allows for a
slightly different implementation to handle interrupting the guest when
using vhost-user.
Change devices to handle the resample event being optional as it is
handled on the VMM side with vhost_user.
Change-Id: I511d3db66a7986e7a2a8bce5f48285171dee3388
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2795284
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
In multi-process mode, we currently rely on dma-buf mmap() to
map GPU buffers into the guest. We usually have to fix the
Mesa driver, and maybe even the kernel to get to work. That's
"kind of" fine for ChromeOS, which owns the entire stack.
For their Virtual Graphics Interface (VGI) initiative, Android
upstream has requested multi-process mode to work in a
cross-platform, generic way. Using Vulkan is the only option
that meets the rigorous, uncompromising, strict, meticulous and
bone-crushing requirements of Android upstream.
This has possible two benefits:
1) We can enable multi-process mode on Nvidia or other
closed-source drivers, which is nice for Cuttlefish.
2) On open-source drivers, dma-buf memory is pinned to the
GTT (amdgpu), even when ideally it can be moved into faster
vram regions. This atleast gives the implementation the
chance to do the smarter and faster option.
We shouldn't run into any SELinux issues since the main crosvm
process is not sandboxed.
Incidentals:
* Changes vulkano_gralloc to consider integrated GPUs and dGPUs.
Metadata query is preferred done on the integrated GPU.
* Update vulkano_gralloc to match top of tree vulkano.
BUG=b:173630595
TEST=used Vulkano allocator and mapped memory into the guest
Change-Id: I78b069c7478d11b3201397894dcccd13bdc61f2c
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2792042
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This patch adds support for creating udmabufs via a guest provided
sg-list. Ideally, we'd create the udmabuf from a virtio-gpu guest
dedicated heap, but that needs further investigation.
In terms of the protocol, these following prototype items are added:
BLOB_CREATE_GUEST_HANDLE: "create an udmabuf" or an OS-specific
equivalent. This can be used with the guest dedicated heap or system
memory. Right now, only system memory is used.
We also want to associate the udmabuf with any host side metadata. For
example, SET_SCANOUT_BLOB doesn't passthrough the modifiers since
virtio-gpu KMS + modifiers is annoying. Simple solution: just ask the
host for the modifier. This could also enable different caching types
if the guest blob is mappable (for example, the MSM GPU driver currently
only supports WC mappings. We might also want cached mappings for
camera).
Incidentals:
* Add a placeholder for RESOURCE_SYNC
BUG=chromium:892806, b:173630595
TEST=create a bunch of udmabufs from the guest
Change-Id: I4686d9f48938f10e7c265750a931d7d2d2d9611b
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2786291
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Allowing each region to have a separate backing FD will make it possible
to build GuestMemory from the vhost `SET_MEM_TABLE` message that
transmits the memory regions for virtio queues in vhost-user devices.
Change-Id: I6f9bc6136915da9d873ea896823e3b8f426ca69d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2795282
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
Hello everyone ..! After 2.5 years of "on the side" inquiries,
I have finally completed my investigations [1] of the udmabuf!!
udmabuf is a kernel driver that turns memfd pages into dmabufs.
The original hope was it would reduce texture upload costs for
virgl, which it did for larger textures [2]. But no measurable
improvements where seen with real games. In addition, a more
advanced "gfx-streaming" model has since come into the horizon[3][4],
which is more performant, conformant, secure and simpler than
virgl. As such, building more on virgl does not seem to be best
option, but that's another story for another day.
Where does that leave udmabuf, then?!? The investigation was
able to turn up two possible use cases:
1) Intel integrated + dGPU PCI-passthrough resource sharing
When the dGPU is passthroughed into the guest, the dGPU's memory
is not available to the host. Ideally, for zero-copy, we would
like to get the render target out of the guest somehow and then
send to the display. Many approaches have been proposed, such
as Vivek Kasireddy's Vdmabuf driver [5]. The current thinking
is virtgpu guest blobs can be exported, and then imported into
the dGPU -- Vivek is looking into this approach right now ..!!
Sommelier or virtgpu KMS can then share the guest blob with the
host. It's a quite complex use case and requires changes to guest
Mesa GBM to get (such as metadata query) to get the right modifier.
Indeed, some would even say you need a virtgpu context type optimized
for sharing across domain boundaries. minigbm already supports this
for Android upstream's Virtual Graphics Interface (VGI) initiative.
2) Guest VRAM dedicated heap created udmabufs
This use case, proposed by automative virtualization expert Dmitry
Sepp [6], is primarily for automotive hypervisors (such COQOS).
It's typically not easy for such hypervisors to get zero-copy via
BLOB_MEM_HOST3D, and these hypervisors have had their homebrew
versions of udmabuf for many years. It's important to upstream the
workarounds that are currently done for such hypervisors. To increase
security and isolation, a guest dedicated heap is preferred over guest
system memory. We might even need dedicated queues, who knows.
crosvm seems like the most likely upstream target due to it's world
class blob support and open-source nature. As such, this CL adds basic
udmabuf capabilites so these use cases can be developed further via
crosvm.
[1] https://www.youtube.com/watch?v=lp6z3s1Gig0
[2] crrev.com/c/1325515
[3] goto.google.com/address-space-graphics
[4] https://drive.google.com/file/d/19K_6M8QUeOn-x7HVYvoNfnuC6G5vkR8f/view
[5] https://lists.freedesktop.org/archives/dri-devel/2021-February/296177.html
[6] https://gitlab.freedesktop.org/virgl/virglrenderer/-/issues/159
BUG=chromium:892806, b:173630595
TEST=Create a bunch of udmabufs from the guest, with the subsequent
patches
Change-Id: Ia8083c0aa065f303f660ec6875ff5fb76f5d7b4f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2786290
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
This is only needed by udmabuf driver, so key it on yet another
feature flag (called "udmabuf").
BUG=chromium:892806, b:173630595
TEST=cargo test
Change-Id: I434a5d1a35d009af0924440df4f72cc7cc9df0e9
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2786288
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
To be truly OS-agnostic, we need an OS-agnostic Rust wrapper over
the OS-specific handle type. SafeDescriptor seems to be the best
option, and I hope it on crates.io in the future.
This converts virtio_gpu/rutabaga to use the SafeDescriptor handle
when practical. minigbm still uses File in some places, since it
needs to SeekFrom(..), but minigbm is a Linux only thing anyways.
BUG=b:173630595
TEST=boot VM in 2D/3D mode
Change-Id: I18d735844d479f52d82d7976bf9b4e383b2e2252
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2779492
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Commit-Queue: Zach Reizner <zachr@chromium.org>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Michael Hoyle <mikehoyle@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Some judgement calls were made about unnecessary wrapping. Usually they
would get resolved by removing the wrapping or returning a convenient
error, but the ones that returned results for consistency with other
functions were added to the allow list.
The error handling in the usb code had a lot of unit error types which
is now a clippy lint. This was resolved by either removing the result
entirely or returning a convenient error.
The field_reassign_with_default lint is faulty and was added to the list
of supressions. This affected virtio-wayland code.
BUG=b:179277332
TEST=cargo clippy with rustc 1.50+
Change-Id: Ie812cdeaf7c42f4f2b47b1dc87f05a7c87a60f8f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2757510
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Auto-Submit: Zach Reizner <zachr@chromium.org>
Commit-Queue: Dennis Kempin <denniskempin@google.com>
This change is to cleanup some dead_code warnings that appear if certain
features aren't enabled.
This also updates the Cargo.lock when changed due to zeroize being added
to libchromeos-rs.
TEST=cargo check --all-features
BUG=None
Change-Id: I5347b584a7426dc37f3933b1e907b23a71145749
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2753128
Reviewed-by: Allen Webb <allenwebb@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Zach Reizner <zachr@chromium.org>
This is the original CL with one minor exception: we don't bind
mount the camera socket with the GPU device. That was the prior
behavior, and for some reason it really doesn't work with Mali +
SECCOMP[1]. It's not really important for the Wayland prototype,
so we'll let the camera team figure it out if and when they are
so inclined.
Bug: b:146066070
Bug: b:173630595
Bug: b:150239451
Bug: b:180126126
TEST=arc.Boot.vm
[1] audit(1613339319.226:43): auid=4294967295 uid=603 gid=603
ses=4294967295 subj=u:r:cros_camera_algo:s0 pid=17107
comm="cros_camera_alg" exe="/usr/bin/cros_camera_algo" sig=31
arch=40000028 syscall=54 compat=1 ip=0xe86a70b8 code=0x0
This reverts commit 51e1c4ad3e3a71a263501d2566d3b1ea59ba2070.
Change-Id: I74f49ece55656d7a9096900e3f19a528234b4224
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2695550
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Robert Tarasov <tutankhamen@chromium.org>
This reverts commit a9c4b3a749.
Reason for revert: This made ARCVM fail to boot on kukui-arc-r. See http://b/180126126.
Original change's description:
> rutabaga_gfx: cross-domain: a new year's miracle in February
>
> The cross-domain context is specialized for cross domain
> allocation/resource sharing. It takes direct inspiration from
> the pioneering virtio_wl device and tries to incorporate
> similiar functionality into virtio_gpu.
>
> The goal here is just to introduce the building blocks so we
> can continue experimenting. In particular, this change:
>
> * hooked up the RutabagaChannels. This is typically a socket to
> Wayland or Mojo for the camera use case.
>
> * added CROSS_DOMAIN_CMD_INIT and CROSS_DOMAIN_CMD_GET_IMAGE_REQS
> to the cross-domain protocol. Further commands (such as
> CROSS_DOMAIN_SEND) will be needed, but that requires more
> Sommelier refactorings.
>
> * added a path to RutabagaGralloc to allocate via minigbm or shared
> memory.
>
> * Recieves responses via a shared ring buffer of type BLOB_MEM_GUEST.
> The synchronization protocol looks positively primitive compared to
> the revolutionary Address Space Graphics (ASG) algorithm [1], but
> it may be sufficient for the Wayland use case.
>
> [1] https://goto.google.com/address-space-graphics
>
> BUG=b:146066070, b:173630595, b:150239451
> TEST=launch virtual machine with 2D mode
> TEST=launch virtual machine with 3D mode
> TEST=run sommelier with "wl-dmabuf" and "wl-shm"
>
> Change-Id: I46784f17040494ce3a646bdbde516800aa64bd5d
> Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2626488
> Tested-by: kokoro <noreply+kokoro@google.com>
> Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
> Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
> Reviewed-by: Zach Reizner <zachr@chromium.org>
Bug: b:146066070
Bug: b:173630595
Bug: b:150239451
Bug: b:180126126
Change-Id: Ie33442fdcedcf43b6a24d25198fa2d88b5b96919
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2695056
Reviewed-by: Ryo Hashimoto <hashimoto@chromium.org>
Reviewed-by: Hiroki Sato <hirokisato@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Hiroki Sato <hirokisato@chromium.org>
Commit-Queue: Hiroki Sato <hirokisato@chromium.org>
The cross-domain context is specialized for cross domain
allocation/resource sharing. It takes direct inspiration from
the pioneering virtio_wl device and tries to incorporate
similiar functionality into virtio_gpu.
The goal here is just to introduce the building blocks so we
can continue experimenting. In particular, this change:
* hooked up the RutabagaChannels. This is typically a socket to
Wayland or Mojo for the camera use case.
* added CROSS_DOMAIN_CMD_INIT and CROSS_DOMAIN_CMD_GET_IMAGE_REQS
to the cross-domain protocol. Further commands (such as
CROSS_DOMAIN_SEND) will be needed, but that requires more
Sommelier refactorings.
* added a path to RutabagaGralloc to allocate via minigbm or shared
memory.
* Recieves responses via a shared ring buffer of type BLOB_MEM_GUEST.
The synchronization protocol looks positively primitive compared to
the revolutionary Address Space Graphics (ASG) algorithm [1], but
it may be sufficient for the Wayland use case.
[1] https://goto.google.com/address-space-graphics
BUG=b:146066070, b:173630595, b:150239451
TEST=launch virtual machine with 2D mode
TEST=launch virtual machine with 3D mode
TEST=run sommelier with "wl-dmabuf" and "wl-shm"
Change-Id: I46784f17040494ce3a646bdbde516800aa64bd5d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2626488
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
rutabaga_gralloc is a cross-platform, Rust-based buffer
manager.
The rationale for this change is:
1) For the {cross-domain, wayland} context type, we need to
have a good story for the crucial "wl-dmabuf" feature. As
minigbm has been thoroughly tested on ChromeOS and currently
powers the "wl-dmabuf" feature, it only makes sense for us to
have a path to minigbm for the cross-domain prototype. This
will be used by Sommelier.
2) While minigbm allocation works well on Chromebooks, it is
not sufficient for cross-platform purposes. For their Virtual
Graphics Interface (VGI) initiative, Android graphics
virtualization experts have expressed their desire for a Vulkan
based allocator. This will to go alongside cros_gralloc in
minigbm, which is considered by many to be the ""world's
premiere gralloc implementation".
3) Android graphics virtualization experts have expressed their
desire for vkMapMemory(..) to be used when crosvm is in
multi-process mode. Currently, only dma-buf mmap() is supported
for zero-copy blobs in multi-process mode. dma-buf mmap() is not
guaranteed to work on Nvidia (a "must have" for Cuttlefish) or
any other driver for that matter (we *make* it work for ChromeOS).
Possibly only solution: vkMapMemory ;-)
With these goals in mind, here's a summary of the revelant changes:
* Renamed the {gpu_allocator.rs, GpuMemoryAllocator trait} to be
{gralloc.rs, Gralloc trait}.
* Moved all GPU allocation out of the resources crate and into
the rutabaga_gfx crate. This will allow the resources crate to
be focused on managing resources for virtual machines.
* Moved the gpu_buffer crate into the gralloc module in the
rutabaga_gfx crate. The same functionality is now under
"minigbm.rs", "minigbm_bindings.rs" and "rendernode.rs"
* Added an optional dependency on vulkano.rs. vulkano.rs is a safe
Rust wrapper around the Vulkan api [a]. It's emphasis on type
safety makes a good fit for crosvm, though there are other high
quality crates out there (gfx-rs, ash.rs). Though development
has slowed down, it should satisfy goals (2) and (3) quite easily.
* Added a system_gralloc implementation based on memfd. This can be
used when minigbm or Vulkano features are not used, to replicate the
highly useful "wl-shm" feature in Sommelier. Astute observers will
note this can also enable seamless Wayland windowing without GPU
features for Android too. Some minor changes to the base crate were
needed.
* Cut down on the amount of DrmFormats to the subset needed by
Sommelier and cros_gralloc.
* Moved checked arithmetic into it's own file.
* Internally renamed to "wl-dmabuf" feature to be the "minigbm"
feature. This is because "wl-dmabuf" has a dependency on minigbm.
* Small rutabaga_gfx cleanups
[a] https://github.com/vulkano-rs/vulkano/blob/master/DESIGN.md
BUG=b:146066070, b:173630595, b:150239451
TEST=launch virtual machine with 2D mode
TEST=launch virtual machine with 3D mode
TEST=run sommelier with "wl-dmabuf" and "wl-shm"
Change-Id: I693a39cef64cd98e56d843d3c60caa7983d4d6e1
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2626487
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
rutabaga_gfx is a cross platform, Rust-based, Wayland and
Vulkan-centric Virtual Graphics Interface (VGI).
Apologies for the mega-change, but it was hard to do this piece
by piece.
The rationale for this change is:
1) Android graphics virtualization experts have been proposing
for a VGI for many months (years?). Their goal is to boot
Android anywhere, everywhere.
2) For the {wayland, cross-domain} context type prototype,
it's desirable to create a {wayland, camera} connection at the
appropriate time. Details can be found in the code, though the
RutabagaChannels have yet to be hooked up.
There's a high chance neither effort will work. As such,
rutabaga is just a prototype.
However, even (1) and (2) don't end up working, this
refactor/cleanup by itself makes a ton of sense.
Here's a summary of revelant changes:
* Removed auto-generated {p_defines, p_format, virgl_protocol}.
These files were added for tests when bringing up crosvm-gpu,
and AFAICT these tests are not run. There's actually now a
commit queue for virglrenderer changes and container boot tests
that provides excellent coverage.
* Removed command_buffer.rs. Used only for the previously
mentioned tests. It's quite nice, but couldn't determine the right
place to put it. Maybe data_model? But removed it in the interim.
* Removed {write_from_guest_memory, read_to_volatile}. The same
basic functionality has been moved into {transfer_write,
transfer_read} in Rutabaga.
* Removed VirtioResource, Virtio3DResource, Virtio2DResource,
and VirtioGfxStreamResource in favor of VirtioGpuResource and
RutabagaResource. This leads to less duplication and clearer
separation between external library functions and VMM functions.
* Moved display and hypervisor memory management functions to
virtio_gpu.rs. This is because external components do not interface
with this functionality, and there was a lot of duplication (for example
map/unmap blob).
* Added context management between gfxstream and virglrenderer.
* Added separate gfxstream and virglrenderer flags.
* Clearer naming.
* Added simple implementations for context init and multiple timelines.
These changes have no effect since all Google kernels don't pass the
revelant flags, but are useful for theoretical {wayland, cross-domain}
prototype.
* Unify RESOURCE_CREATE_3D and RESOURCE_CREATE_2D handling.
* Better error handling.
BUG=b:146066070, b:173630595, b:150239451
TEST=launch virtual machine with 2D mode
TEST=launch virtual machine with 3D mode
TEST=boot ARCVM
Change-Id: I240b0c134a3b562cbc65981837a41f6db7767c92
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2522452
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Lingfeng Yang <lfy@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Fixes crosvm's failure to forward the "use_external_blob"
state flag to the host renderer (virglrenderer).
BUG=b:174794821
TEST=Start a crostini VM and run `DISPLAY=:0 glxgears`, verify graphical
output.
Change-Id: I1ef616079664224e7741cea00b9125c5061ea69d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2574669
Tested-by: Ryan Neph <ryanneph@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Commit-Queue: Ryan Neph <ryanneph@google.com>
This change adds support for exporting virtio_gpu fences into
virtio_wl.
Without support for sending fences over virtio_wl, ARCVM must wait in
the guest for fences to be signaled before sending the corresponding
buffers to the host compositor. Under normal composition, it is
relatively rare for fences to be unsignaled when they need to be sent
over virtio_wl, although it does still happen at the far tail end. If
ARCVM falls back to client composition, without this change ARCVM needs
to wait on average 5-15ms (depending on how heavyweight the app's
rendering is) for fences to be signaled.
Support for sending fences over virtio_wl allows this extra latency to
be hidden.
BUG=b:136129939
TEST=Run YouTube+Play Store+graphics benchmarks simultaneously
in ARCVM, with and without client composition.
Change-Id: Ic7f57069c4171104c497aee6721c5c93dc6bafa4
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2363512
Tested-by: David Stevens <stevensd@chromium.org>
Commit-Queue: David Stevens <stevensd@chromium.org>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
This CL transitions most structs to RawDescriptor and the
associated traits if possible.
BUG=b:162363783
TEST=./build_test
Change-Id: Iabae6ac212787836d77de2b9ffb5d451421ab0dd
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2530911
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: Michael Hoyle <mikehoyle@google.com>
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
Along with this rename comes updating all usages to the
appropriate RawDescriptor traits. As per usual it touches
many files, but makes no significant changes.
The only remaining instance of "keep_fds" is to call out to
3p lib adhd.
BUG=b:162363783
TEST=./build_test
Change-Id: I6d292cc6404a9f84ffa5bf1320b6545a28842afa
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2488071
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
Tested-by: Michael Hoyle <mikehoyle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
... to allow configuring Gfxstream host renderer based
on whether or not the guest Android is using ANGLE as
its native OpenGL driver.
BUG=b:165022040
TEST=launch_cvd --gpu_mode=gfxstream
Change-Id: I566ed1e7590f7f3db99960440421fd65bba08f7b
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2466604
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Jason Macnak <natsu@google.com>
This involves removing RawFd trait implementations from
Event and Timer, and resolving the echoing dependencies from
there.
Ultimately, this results mostly in changes across ioctl
(new thin layer in base), kvm, msg_on_socket, and a few other
areas. As usual, most changes are negligible.
BUG=b:162363783
TEST=./build_test
Change-Id: I47dec94666bc3430fed627b4767919c9cb4fce6f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2462330
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
This is a layer above PollContext with a more generic interface.
As PollContext is used so widely, this is quite a large change
in order to accomodate the interface update, especially with the
use of RawDescriptor. In some cases this has caused an echo
of updates to RawDescriptor, which is fine because of our eventual
goal to move the whole codebase to it regardless.
Note there are a few instances of forcing the RawDescriptor update
chain to stop, ex. ioctl. This is to keep the scope of this CL
concentrated and avoid changing entire other areas.
Note that this CL leaves out a few additional pieces of work:
- The sole usage of EpollContext over PollContext (event_loop),
which poses a bigger challenge for interface changes
- Full PollToken renaming, which is a tiny change turned difficult
due to the unavailability of type aliases for traits.
- Renaming certain methods which have been updated to use
RawDescriptor such as keep_fds. Some have enough dependencies that
they are worth avoiding to keep this CL pointed, but will be
addressed in future CLs to make sure the whole codebase is on the
fd->descriptor train
BUG=b:162363783
TEST=./build_test
Change-Id: Iff2cfe8f90dea55f1388f8e91bdc698e121a8e43
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2455726
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
All virtio devices pass VIRTIO_F_VERSION_1 explicitly. In preparation
for passing additional reserved feature bits, introduce a free function
so that devices only have to worry about device-specific features.
BUG=None
TEST=./build_test
Change-Id: Ifcf309fa010ea518b17b047b0e95e5ad5c7522bb
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2453558
Reviewed-by: Andrew Walbran <qwandor@google.com>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Will Deacon <willdeacon@google.com>
Note the CL size is large entirely due to the rename,
the changes are mostly negligible.
Also making a few small additional changes in sys_util
areas that don't need much attention in base. This includes
typedefing and adding specific imports for areas that don't
require significant interface changes.
BUG=b:162363783
TEST=./build_test
Change-Id: I4a2c9c4cdce7565806ed338e241c6b8c82c855c7
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2415180
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
When requested with the --gpu=cache-path=/path arg, crosvm
will pass it to Mesa via env var MESA_GLSL_CACHE_DIR. In
addition, the cache-size will also be passed along if
provided.
BUG=b:168540438
TEST=run with --gpu=cache-path=/tmp,cache-size=50M and
confirm that files are created in /tmp/mesa_shader_cache.
Change-Id: I2525597749d654a65373a723cefeab6cf2be62d7
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2415509
Tested-by: John Bates <jbates@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: David Riley <davidriley@chromium.org>
When switching to async usage of descriptor queues, it is more useful to
consume the offsets and lengths of the descriptor buffers instead of the
buffers mapped as `VolatileSlice`s. Having the offset,length pair and
the guest memory it refers to is enough to initiate async ops.
Async versions of non-offset read/write case aren't possible as using
the existing "seek" offset of the file doesn't make sense with async IO.
Adding the new version modifies the existing API. This is necessary as
now `Reader` and `Writer` own `GuestMemory`, which makes async possible.
TEST=unit tests and new test to exercise the async functions.
Change-Id: I0a5e50694f40217e5d6ac40f5fcb82944a4c117e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2306784
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Chirantan Ekbote <chirantan@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
Not strictly necessary to build gfxstream, but the MemSlot abstraction
was recently introduced and it makes sense to use it everywhere.
BUG=b/153580313
TEST=gfxstream builds
Change-Id: I06c2ca15edd39c553eacf1256dd95c2fc72f67dc
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2349394
Tested-by: Gurchetan Singh <gurchetansingh@chromium.org>
Reviewed-by: Lingfeng Yang <lfy@google.com>
Reviewed-by: Jason Macnak <natsu@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Gurchetan Singh <gurchetansingh@chromium.org>
