There are a lot of changes in this one but these are the high-level
points:
* Both executors now support non-fd futures and it's now possible to
start a poll operation on one thread and then await the result inside
a UringExecutor on another thread. The reverse doesn't work yet but
will once we make UringContext sync.
* A few layers of indirection have been removed so hopefully both the
implementation and the interface are simpler.
* The thread local magic is gone in favor of directly calling methods on
an executor. Executor handles can be cheaply cloned to make this
easier.
* Heap allocations are limited to the spawn and spawn_local methods so
it's possible to completely avoid heap allocations if callers only use
the run_until method.
* The IoSource, Executor, FutureList traits are gone.
BUG=none
TEST=unit tests
Cq-Depend: chromium:2629068
Change-Id: I86053007929c36da66f3b2499cdefce0b5e9a180
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2571401
Commit-Queue: Chirantan Ekbote <chirantan@chromium.org>
Tested-by: Chirantan Ekbote <chirantan@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Use the latest version of cfg-if and paste, which gdbstub crate will require.
BUG=chromium:1141812
TEST=cargo build
Cq-Depend: chromium:2507270
Change-Id: I9187cfc9a880f62b2aa1fcf5e5d47a720e5fbe60
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2499241
Tested-by: Keiichi Watanabe <keiichiw@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Commit-Queue: Keiichi Watanabe <keiichiw@chromium.org>
Instead of creating IoSourceExt from AsRawFd implementers, we've
switched to creating from a marker trait `IntoAsync`. This lets us use
other types like RawDescriptor easily with this crate. By using the
marker, we also provide some type safety by requiring consumers of
IoSourceExt to declare that their type is expected to work with async
operations. This way we can provide stronger guarantees that an async
IO trait object will behave in a reasonable way.
This CL also purges the cros_async -> base and io_uring -> base
references, and provides the base types needed to add new async
primitives to base.
BUG=none
TEST=builds
Change-Id: I0b0ce6ca7938b22ae8e8fb4e604439f0292678f2
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2504481
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Noah Gold <nkgold@google.com>
Reviewed-by: Chirantan Ekbote <chirantan@chromium.org>
This CL makes the following fundamental changes to cros_async:
1. Removes PollOrRing and replaces it with IoSourceExt, and the
subtraits ReadAsync & WriteAsync. The blanket implementation of
IoSourceExt has been dropped, and replaced with source specific
implementations of the trait. Those implementations are where
the code from PollOrRing has been moved.
2. Pinning for IoSource has been dropped from UringSource & the uring
futures. This appears to be safe because the IoSource doesn't contain
any self refs, or perform any operations beyond forwarding to the
RegisteredSource. (The FD is duped before being passed to
RingWakerState by RegisteredSource, so there doesn't seem to be any
data which would require pinning.)
3. U64Source was replaced by EventAsync.
It also switches all Error enums to use thiserror, which reduces
boilerplate.
BUG=None
TEST=cargo test -p cros_async
Cq-Depend: chromium:2421742
Change-Id: Ie1dd958da2e1f8dec1ae1fd8c0b4e754223d330d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2416996
Reviewed-by: Chirantan Ekbote <chirantan@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Noah Gold <nkgold@google.com>
This CL includes several smaller changes to how executors work:
* Replace BTreeMap with Slab, which should give us some small
performance benefits by giving O(1) lookups and reducing memory
allocations when adding new I/O operations. It also gives some
improvements to readability as we no longer have to carry around
"next_*" variables. Slab has no dependencies and we're already
pulling it in via the futures crate.
* WakerToken no longer implements Clone.
* Merge pending_ops and completed_ops in URingExecutor into a single
`ops` Slab and introduce an OpStatus enum that indicates whether an
operation is pending or completed. This also fixes a resource leak
where an operation that was canceled before completion would end up
staying in completed_ops ~forever. Add a test for this leak.
* Add a generation number to RingWakerState and include it in all
RegisteredSources. Since a RegisteredSource can outlive the
RingWakerState that created it, the generation number ensures that
it will only affect the RingWakerState that created it. Add a test
for this.
* Poison RegisteredSource so that it doesn't implement Send or Sync.
Since it's associated with a thread-local executor, sending it
across thread boundaries is not ok.
BUG=none
TEST=unit tests
Change-Id: I43dcfbb8166002995ec8773522c22fab8fb2da9e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2374885
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Chirantan Ekbote <chirantan@chromium.org>
The Rc to BackingMemory that is passed in to memory operations was being
dropped before it is safe, The kernel can still access it until the op
is completed, so keep the reference in the pending_ops map.
Add tests so that this safety guarantees made by the executor are
exercise in unit tests.
Thanks to nkgold for finding the issue via code inspection.
BUG=none
TEST=cargo test dont_drop_backing_mem, cargo run with uring block and
run fio tests.
Change-Id: I8e4efedbbafefcbd57d7d7c340dc91a9b159b38c
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2345377
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
For now, this crate simply re-exports all of sys_util, but it will
be updated to provide new interfaces when needed. This is the
first step to making crosvm not directly depend on sys_util, so
that we can make the interface changes we need without fear of
negatively affecting (i.e. completely breaking) other usages
within chromeos.
BUG=b:162363783
TEST=./build_test
Change-Id: I7d0aa3d8a1f66af1c7fee8fd649723ef17027150
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2325168
Tested-by: Michael Hoyle <mikehoyle@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Michael Hoyle <mikehoyle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Implementing `BackingMemory` signals that `GuestMemory` regions can be
used in uring transactions where the lifetime in which the kernel can
modify the memory is not well defined.
Change-Id: I3541fff4c5dac226062a94483672f570e7adeb18
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2275725
Reviewed-by: Chirantan Ekbote <chirantan@chromium.org>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
Tested-by: Dylan Reid <dgreid@chromium.org>
Move GuestAddress and GuestMemory to a new crate for VM memory. This
will make separating sys_util and crosvm independent making it easier
to use sys_util functions outside of crosvm.
Change-Id: I12e14948ea85754dfa6267b3a3fb32b77ef6796e
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2311251
Auto-Submit: Dylan Reid <dgreid@chromium.org>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
Tested-by: Dylan Reid <dgreid@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
These will be needed by by block.
Note that support for uring fallocate was added in 5.6. The users will
have to handle falling back to synchronous calls if fallocate returns
EINVAL.
BUG=901139
TEST=added fallocate unit tests
Change-Id: I51d635adcf0bb4dd55c5bfe50719f2fde2b88e49
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2274996
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Tested-by: Dylan Reid <dgreid@chromium.org>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
std::task::Waker unconditionally implements Send + Sync so the raw waker
that we provide also must implement those traits. Switch to using an
Arc<AtomicBool>.
This also fixes an inconsistency where the waker was defined to be an
Rc<Cell<bool>> but all the vtable functions were treating as an
Rc<AtomicBool>.
To reduce the vtable boilerplate use the ArcWake trait from the futures
crate.
BUG=none
TEST=unit tests
Change-Id: I3870e4d7f6ce0de9f6ac3313a2f4474ae29018b2
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2287079
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Tested-by: Chirantan Ekbote <chirantan@chromium.org>
Commit-Queue: Chirantan Ekbote <chirantan@chromium.org>
Add an executor, similar to `fd_executor`, except driven by io_uring
instead of `PollContext`. This will allow for queueing actions instead
of waiting for 'ready' from poll. This reduces the number of syscalls
and increases the parallelism possible when doing IO.
Change-Id: Ie2e69cb09ee3ab1086ef31d5ebd0169843a8228d
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/2227083
Commit-Queue: Dylan Reid <dgreid@chromium.org>
Tested-by: Dylan Reid <dgreid@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
This crate will house code using the new async/await features to be used
by other parts of crosvm.
Start the crate with a Future executor that runs tasks in a single
thread and allows futures that block on system file descriptors.
Change-Id: If77778ac056210dabbfc6e6e1e63df1c1b904a7f
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1955045
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Tested-by: Dylan Reid <dgreid@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Commit-Queue: Dylan Reid <dgreid@chromium.org>
