IoBuf is defined in a platform-specific way (it is either iovec on unix
or WSABUF for windows), so it fits into the mission statement of the
base crate, which is meant to be *the* platform abstraction layer in
crosvm.
IoBufMut and VolatileMemory/VolatileSlice are defined in terms of IoBuf,
so they are also moved into base for consistency. Every crate that uses
these types already depended on base, so no extra dependencies are
added, and a few can be removed.
BUG=b:312312646
TEST=tools/dev_container tools/presubmit
Change-Id: I4dddc55d46906dfc55b88e8e6a967d7e1c1922dd
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/5046605
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
`AsRawDescriptor` is preferred over `AsRawFd` everywhere. This commit
replaces `AsRawFd` of the file related helpers in base except functions
which explicitly intend to handle FDs such as `validate_raw_fd` or
`get_fd_flags`.
BUG=None
TEST=./tools/presubmit
Change-Id: Ia0f1b149bf281a97dfe754fd20e3d9d898539df1
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/5020382
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Takaya Saeki <takayas@chromium.org>
Currently, `punch_hole` operation is translated to
`fallocate(PUNCH_HOLE|KEEP_SIZE)`. However, this does not work as
expected if the target file is a block device file such as an LVM
partition block file. It does not succeed and the area will not be
reclaimed, which is causing a problem on ARCVM.
The right way to punch-hole a block device file is calling
`ioctl(BLKDISCARD)`. This change fixes `PollSource`, `UringSource`,
and the default implementation of `PunchHole` trait.
BUG=b:302437024
TEST=./tools/presubmit
TEST=`strace -f -e ioctl crosvm run ... --block \
path=/dev/glinux_20230808/test_lv kernel` and `blkdiscard /dev/vda`
inside, then confirmed BLKDISCARD with epoll and io_uring
Change-Id: Ibedecd2c401bbb9ab25d7f0bfe18a3334f79ea85
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/5019243
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Commit-Queue: Takaya Saeki <takayas@chromium.org>
Currently, cros_async's async IO source provides async `fallocate`
function, and `punch_hole` and `write_zeroes_at` are delegated to
`fallocate`. However, this design has three rooms for improvements.
1. Actually, no one calls async `fallocate`.
2. We have duplicated `fallocate` and `AllocateMode` implementations in
`base::sys`.
3. Windows has to pretend to have `fallocate` to provide `punch_hole`
and `write_zeroes_at`
So, this change removes `fallocate` and `AllocateMode`, which no one
actually calls, from `IoSource` and deletegate it to `base::sys`. With
this, we don't have to maintain `fallocate` implementation in
cros_async.
This will also enable us to call a non-`fallocate` systemcall in
`punch_hole` implementation which will be necessary to correctly support
block files.
BUG=b:302437024
TEST=./tools/presubmit
Change-Id: I8e7fcd94380da980cc52e6f751b134fa325d7b39
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/5019241
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
Commit-Queue: Takaya Saeki <takayas@chromium.org>
Remove pointer casts where the target type is the same as the original.
Change-Id: Ibb7bda2e4afcb5df2cd7c0bd9a075d0288b69f67
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/5005512
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Windows needs N threads to poll to complete N IOs concurrently.
The patch createn 'n' concurrent threads that are passed as DiskOption
Bug: 303249046
Test: Ran tests downstream to verify that N ios are running in parallel
Change-Id: I9c1c101993e6ecbf99213218341cc77544fbd6bc
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4985161
Auto-Submit: Vikram Auradkar <auradkar@google.com>
Reviewed-by: Noah Gold <nkgold@google.com>
Commit-Queue: Vikram Auradkar <auradkar@google.com>
The timeout for all pool shutdowns was set to 100 ms in a previous
change in order to make the tests run faster:
<https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4004299>
However, this can lead to flakiness if the system is under load while
the tests are executing. This change makes some more careful timeout
choices to improve reliability without extending the overall time it
takes to run the tests.
Most of the shutdown_with_timeout(100ms) calls are replaced with the
normal shutdown() function, which uses a 5-second timeout. These
shutdown() calls are expected to return immediately (either because of a
non-timeout error condition or because all pool tasks are already
finished), and we can be sure that this 5-second timeout is never
relevant because all of the cros_async tests execute in under 5 seconds.
This should fix a flaky test that was observed in practice where the
disarm_with_pending_work() test would not finish shutting down in the
(rather short) 100 ms timeout previously allowed.
The shutdown_with_blocked_work timeout is reduced from the original 100
ms to 1 ms, since it should not affect the correctness of the test and
will help it run even faster. This test is also renamed to match the
actual behavior - it used to mention "panic", but the shutdown function
simply returns an error and does not panic.
Finally, the shutdown_in_progress() test now uses two different
timeouts. This test spawns a thread that waits until shutdown is in
progress, then calls the normal shutdown() function to ensure it returns
the correct error; this shutdown() call should return immediately, so
again the 5-second timeout is not relevant. The main thread of the test
now uses a timeout of 200 ms, which should be sufficient to ensure the
other thread has time to start up and reach the shutdown() call while
observing the shutdown_in_progress flag. This extends the test time
somewhat relative to the previous 100 ms timeout, but that is mostly
traded off with the timeout reduction in the blocked work test, so the
overall test time should be about the same as before. This is probably
the most likely test to be flaky going forward, since it relies on OS
thread scheduling behavior; if any problems are observed, we can
consider making the timeout longer, although this will slow down the
tests.
BUG=b:306726068
TEST=tools/dev_container tools/presubmit
TEST=time cargo test -p cros_async
Change-Id: Ic6ec117d616ae7c99a8e3c5a4fe6dd1169c72286
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4961353
Commit-Queue: Dennis Kempin <denniskempin@google.com>
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Updates are made to source and documentation.
This more accurately represents the currently supported platforms of
Android/Linux and Windows, without unexpectedly including other
unix-like operating systems.
Command to reproduce:
$ find . -type f -not -path '*/\.git/*' | xargs -I {} sed -i 's/cfg(unix)/cfg(any(target_os = "android", target_os = "linux"))/g' {}
$ cargo fmt
md files manually updated to fix line lengths.
Renaming `unix` modules to `linux` will be done in a later CL.
Test: ./tools/dev_container ./tools/presubmit
Bug: b/298269162
Change-Id: I42c1bf0abf80b9a0df25551613910293217c7295
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4909059
Commit-Queue: Cody Schuffelen <schuffelen@google.com>
Reviewed-by: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
As usual, some unsorted and grouped imports have appeared.
Change-Id: I79b51e4c52cee38f5b8c238e46dfe3193c753554
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4847980
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
There will be fewer `fctl` and `close` syscalls when performing IO,
which might help performance, but a simple FIO benchmark didn't reveal
any gains. The strace output looks cleaner.
Change-Id: I11b2cde23d7951c209a48c19194fd0de15d43f8f
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4722169
Commit-Queue: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
I'm going to add another field to RegisteredSource and so it won't make
as much sense for it to auto convert to the underlying source type.
Change-Id: I4c135397878ed8e13a5443ad68ccc7ffcf0bf491
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4722171
Reviewed-by: Takaya Saeki <takayas@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Because unblock tasks cannot be canceled, and the ones started for
AsyncTube hold locks, this can cause code trying to convert the
AsyncTube back to a Tube to stall because the lock is held on the
unblock thread and won't ever be released until the global pool is
shutdown.
The right solution is to abandon unblock or make it cancellable.
Those are significant undertakings, so in the interim, we can
avoid starting unblock until the read notifier is triggered.
This way, we won't actually be in unblock, and the lock will
not be held.
BUG=b:294134741
TEST=snapshot no longer stalls on Windows.
Change-Id: I36bfd0ca5546c43f766c431a2fb97d2b87a679ae
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4749226
Reviewed-by: Frederick Mayle <fmayle@google.com>
Commit-Queue: Noah Gold <nkgold@google.com>
Reviewed-by: Richard Zhang <rizhang@google.com>
This allows the tests to run on all executor kinds supported by the
generic io_source test harness (io_uring, epoll, and Windows handle).
readmulti is modified to write different data to each region so the test
can verify the correct part of the file is actually read.
fsync is removed since it is already covered by io_source's fsync test.
readwrite_current_file_position is split into two tests that validate
the async io_source behavior using normal non-async File operations
(read_current_file_position and write_current_file_position).
BUG=b:288967124
TEST=tools/dev_container tools/presubmit
Change-Id: I59ce8dd3934135499062fed4cb7e0f54bf446d09
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4655444
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
Introduce a TimerTrait that is implemented by both Timer and FakeTimer.
This trait can be used to write testable code without needing to
condition use statements on cfg(test).
BUG=None
TEST=CQ
Change-Id: I874ffe19fafe73e5c17fd5bb689e5b571e9f5b12
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4664188
Reviewed-by: Junichi Uekawa <uekawa@chromium.org>
Commit-Queue: David Stevens <stevensd@chromium.org>
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
MemRegionIter allows iteration over a &[MemRegion] list, with additional
helper functions to skip bytes and truncate the list.
In particular, this is useful for the virtio DescriptorChain utilities
(Reader/Writer) that previously needed to call collect() to allocate a
new copy of the regions list when it was necessary to limit the length
or skip part of the beginning of the chain.
This also replaces the functionality of MemRegion::truncate(), which
used Cow<[MemRegion]>, meaning it would sometimes need to allocate and
copy. The new MemRegionIter implementation never needs to copy the list
of regions; it just maintains a couple of internal counters and adjusts
the returned MemRegions on the fly.
BUG=None
TEST=tools/dev_container tools/presubmit
TEST=cargo test -p cros_async
Change-Id: Ie66c7a8c4061a166c986c6d6937e21549ddb205e
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4559367
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Reviewed-by: Frederick Mayle <fmayle@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
The previous method required calling `get_volatile_slice()` twice for
each memory region - once to validate that each region is valid, and
again to convert the MemRegion into an IoBufMut. It also relied on
`get_volatile_slice()` returning the same non-Err value twice to avoid a
panic in the `unwrap()` used in the second pass.
Instead, we can iterate over the regions once and convert the iterator
directly into a pinned boxed slice.
BUG=None
TEST=tools/dev_container tools/presubmit
TEST=cargo test -p cros_async
Change-Id: I16e0bd38998b97e072a2d0fa5a72a2dea2d082f9
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4646722
Reviewed-by: Frederick Mayle <fmayle@google.com>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
The new rust toolchain comes with improved clippy warnings that
catch more unnecessary casts than before.
Some cases have to be whitelisted since libc types have varying size
based on the architecture we are compiling for.
BUG=None
TEST=presubmit
Change-Id: Ia473b0bfe6e74bdb7b0d4ca15d2c9fb410fc353e
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4568144
Commit-Queue: Dennis Kempin <denniskempin@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Currently, cros_async::sync::Mutex is named "Mutex", but it is actually
a read-write lock, which is confusing to readers. As a result, readers
often mistakenly think that the part of the code which should use a
read-write lock is using a Mutex.
This change renames Mutex to RwLock to eliminate the confusion. In
addition, this change updates the documentation to recommend developers
to use the standard `futures:🔒:Mutex` if they just need a Mutex,
not a read-write lock.
BUG=None
TEST=./tools/dev_container ./tools/presubmit
Change-Id: Ic008d2448308066d9c5cabb0b6c499a62994e5e2
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4546967
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Frederick Mayle <fmayle@google.com>
Commit-Queue: Takaya Saeki <takayas@chromium.org>
fd_executor, uring_executor, and handle_executor had a lot of duplicate
code. A new trait `Reactor` is introduced to abstract away the
differences between the IO backends. There is now a single
`RawExecutor`, which is generic over `Reactor`s.
No behavior changes are intended, with some small exceptions:
* A new error variant is used in some cases, e.g. instead of
`Error::PollExecutor` an error might be `Error::Io` now. Nothing seems
to inspect these errors, so it is unlikely to cause an issue.
* There is an extra alloc when the executor is dropped (for the boxed
future).
* The Unix backends had an `AtomicI32` state field but the Windows
backend did not. Now Windows does as well. This might save a syscall
here and there.
Change-Id: Ieeff9d9888ed082cf3e0213e586c55c9d8a1f8bf
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4331554
Reviewed-by: Noah Gold <nkgold@google.com>
Commit-Queue: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
I broke this in https://crrev.com/c/4331553 because I didn't notice the
`mem::replace` call. The first wake would switch the special wake event
op status to "complete" and then a second wake would panic because
nothing should complete twice.
Bug: 282757779
Change-Id: Iff9120edccd9d128f44586f50ac4dacae6223843
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4535106
Commit-Queue: Frederick Mayle <fmayle@google.com>
Auto-Submit: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Currently, virtio-blk translates a FLUSH request to fsync system call.
However, fsync always updates the metadata of a file, which causes extra
journaling for metadata update. Since virtio-blk cares only the data of
the given block image file, we should avoid updating extra metadata.
After this change, virtio-blk now calls fdatasync for a FLUSH request.
If a write operation wrote to a data block that is already allocated
before, fdatasync can avoid triggering jdb2 journaling on an ext4 file
system.
Note there are structs which fall back to fsync for fdatasync. In
addition, io_uring executor does not implement fdatasync yet.
We observed statistically significant play_store_shown_time improvement
on trogdor-arc-r by -1.614% with this change. Including non
statistically significant data, we observed the following improvements.
| Name | delta | Count |
|-----------------------|---------|-------|
| trogdor-arc-r (lazor) | -1.614% | 170 |
| brya-arc-t (primus) | -1.478% | 200 |
| octopus-arc-t (apel) | -0.887% | 190 |
| kukui-arc-r (kakadu) | -0.451% | 170 |
BUG=b:281609112
TEST=run a vm with block and confirmed fdatasync is called by strace
Change-Id: Idc94a3ec169e9a5e04394079967f6d79ff4c32db
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4523029
Commit-Queue: Takaya Saeki <takayas@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
It isn't widely used, probably because it isn't very useful: you need a
reference to the Executor to do IO. The same argument could be made for
`block_on`, but it is more widely used in tests.
Change-Id: I3e7b5783866d42a18103025f85912a9abeab5327
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4513228
Commit-Queue: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
This dovetails with an upcoming refactoring which splits the generic
executor code from the IO "reactor" code. It is also simpler and uses
marginally less resources.
The `RawExecutor::wake` call was removed from `RawExecutor::drop`. It
was previously needed only so that the (now deleted) async function
`notify_task` would break out of its loop (even then, it didn't seem
strictly necessary).
(Could we possibly need the `wake` call for something else? I don't
think so. `wake`'s propose is to wake up the `RawExecutor` on *other*
threads, however, in `drop`, it is impossible for other threads to have
access to `self` because we've got a `&mut`, so there will never be
anything to wake up.)
Change-Id: Ic61e304b360d0e33320b486d811504213c2321fe
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4331553
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
This was implicitly tested before, but seems worth an explicit test.
Change-Id: I0d31051a8422d61f91938fc5975f230346bbce65
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4478450
Commit-Queue: Frederick Mayle <fmayle@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Adds an invocation to run doctests to the health_checks builder.
To make this not run by default on every `tools/presubmit`
invocation, this CL splits default_health_checks out of the
health_checks group.
BUG=b:242635135
TEST=tools/presubmit health_checks
Change-Id: Ic850c461eb79e81c4b8118fc3182614fac907569
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4476090
Reviewed-by: Zihan Chen <zihanchen@google.com>
Commit-Queue: Dennis Kempin <denniskempin@google.com>
wait_for_handle always returned 0.
Timer::next_val's return wasn't used and its meaning was platform
dependent.
Change-Id: I2278859c19505398c3dd01b90e69e20894c38e25
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4409940
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
`wait_for_handle` only makes sense on Windows.
`read_u64` can be implemented in terms of the other read methods. We now
have to allocate a `Vec` which shouldn't technically be needed for the
epoll backend. We do this for all other kinds of IO and timers/events
are generally low throughput, so expect the performance loss won't be
noticable.
Change-Id: I1c1f1d5d1fc84073b940bc71425fe150196c4441
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4405314
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Give the unused futures names to avoid triggering the new clippy
let_underscore_future warning.
No functional change.
BUG=b:276487055
TEST=tools/clippy # with rust 1.68
Change-Id: I61631261e5d8acee1fe123c808a0e76ff24a0a6c
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4391113
Reviewed-by: Dennis Kempin <denniskempin@google.com>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
The block device has a flush timer that is started when a write request
is received. This is intended to flush any outstanding writes to the
disk image if the guest does not issue its own explicit flush request
before the timeout. However, the existing implementation did not
actually disarm the tirm if the guest submitted a flush, so the flush
timer would always fire 60 seconds after the most recent write, even if
all previous writes had already been flushed by guest request.
Eliminate the extra flush by disarming the flush timer when a
VIRTIO_BLK_T_FLUSH request is processed.
BUG=None
TEST=strace crosvm block device while writing and syncing
TEST=dd if=/dev/random of=/dev/vdb conv=fsync bs=4096 count=1
Change-Id: I2cc754c84b4f3f85468381133b40f59c460dd309
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4316953
Reviewed-by: Takaya Saeki <takayas@chromium.org>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
this patch implements the virtio-balloon extensions proposed in
go/roziere - namely adding the Working Set Size vqueues and listener on
the wss vqueue. in addition we add a crosvm_control command which allows
a user to request that the balloon device send a request to the driver
to produce a WSS report, similar to asking for balloon stats.
BUG=b:271757427,b:273212204
TEST=sidecar process, available at https://x20.corp.google.com/users/dr/drmasquatch/socket-pinger
TEST=deploy to DUT, restart termina and arcvm
Change-Id: I45753f523aae97d14bfc5dcbad5ddfcb70d10beb
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4260158
Reviewed-by: T.J. Alumbaugh <talumbau@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Maciek Swiech <drmasquatch@google.com>
This consolidates many of the unit tests for async backends into
io_source.rs to reduce code duplication and to ensure consistent
behavior.
Notes:
* Deleted `read_eventfds` because it is unix specific and we have
separate tests for the `AsyncEvent` interface.
* Moved uring specific tests into uring_source.rs to reduce `#[cfg]`
noise.
* There are more tests that should be consolidated, e.g. the
`IoSource::fallocate` tests.
* I hadn't realize some of the complexity of OverlappedSource until I
had done most of the work, so this initial passes excludes it.
Change-Id: I5b11605246a049398688f57d81fc48579a2287c9
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4391794
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Commit-Queue: Frederick Mayle <fmayle@google.com>
Reviewed-by: Noah Gold <nkgold@google.com>
This is supported since Rust 1.62.
Fixes the new derivable_impls clippy lint, added in Rust 1.67.
BUG=b:276487055
TEST=tools/clippy # with rust 1.68
Change-Id: I3696dd3cf4acbf9b5a2cfa8c2046e7656e13dbbd
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4390735
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
One `async_from` function can handle both `Send` and `!Send` IO sources
now.
Note that, before, when we were using `IoSourceExt`, having two
functions for this wasn't really useful anyways because the returned
trait object was always `!Send`.
Change-Id: I2655657988b30a4fb4ec9c4ee543724d4e594de3
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4391793
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Noah Gold <nkgold@google.com>
Commit-Queue: Frederick Mayle <fmayle@google.com>
This removes two layers of boxing and trait objects. One for the IO
source objects and another for the futures generated by calling methods
on them.
Benefits:
* IO futures will now be `Send` (if the underlying source is `Send`,
which is generally the case).
* It is easier to work with a concrete type than a `Box<dyn ...>`.
* More efficient (less boxing and virtual function calls).
Slightly smaller binary size:
$ cargo build --all-features --profile chromeos && ls -l target/chromeos/crosvm
before: 10,811,576
after: 10,692,792 (1.1% smaller)
Change-Id: I7b52cd24945cfb4ff0a6c377739b5d4ab9cacb0b
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4004980
Reviewed-by: Noah Gold <nkgold@google.com>
The SafeDescriptor type manages the lifetime of the file descriptor,
which simplifies the socket wrappers (each wrapper no longer needs to
manually implement Drop to close the descriptor, for example).
Also remove unneeded RawFd-related functions, replacing them with
RawDescriptor and SafeDescriptor equivalents.
BUG=None
TEST=tools/dev_container tools/presubmit
Change-Id: I634a19922ec24d06071b21247c79761cfc21a79a
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4322266
Reviewed-by: Keiichi Watanabe <keiichiw@chromium.org>
Commit-Queue: Daniel Verkamp <dverkamp@chromium.org>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Following the addition of an overlapped capable executor, here is the
experimental overlapped IO source. This is how we'll request overlapped
IO that runs on the overlapped executor. This isn't ready for production use
quite yet, but it's something we'll want in the future.
BUG=b:272614458
TEST=presubmit
Change-Id: Iaa97cf70924334dbe82d4562f43c0daf8888cb35
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/4328234
Reviewed-by: Vikram Auradkar <auradkar@google.com>
Commit-Queue: Noah Gold <nkgold@google.com>
