A plugin might care to be immediately notified when a write
is made to a port, but it doesn't care to have the VM stopped
while the plugin calls back to resume the VM.
Unfortunately this means that multiple messages can be queued up in the
pipe and read() together by the plugin API. Protobuf's parsing function
doesn't report how many bytes it read, so I've resorted to having crosvm
prefix every message with a length and then have the plugin lib parse
this number. Impact on performance has not been measured.
BUG=b:143294496
TEST=Local build and run of build_test. Verified that new unit
test was executed, exercised the case where multiple msgs are
received together, and completed successfully.
Change-Id: If6ef463e7b4d2e688e649f832a764fa644bf2d36
Signed-off-by: Matt Delco <delco@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1896376
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Stephen Barber <smbarber@chromium.org>
The stats are typically alive in the plugin library, but we typically
don't care about or use these stats.
BUG=None.
TEST=Compiled and ran test. Verified release binary size got smaller
by 25KB and debug by 50KB.
Change-Id: I2469ff83f91a3aebf86d70807088bba3edce6641
Signed-off-by: Matt Delco <delco@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1835034
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This change tries to improve the performance of a plugin-based VM
by adding a hint API that allows crosvm to proactively push cpu
state to the plugin when certain ports for hypercalls are accessed
by the VM.
BUG=None
TEST=build and run. See performance increase significantly.
Change-Id: I71af24ebc034095ffea42eedb9ffda0afc719cd6
Signed-off-by: Matt Delco <delco@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1873005
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
If a plugin makes a set call on vcpu registers then we
can improve performance by deferring the IPC and instead
conbining the request with the next resume call.
BUG=None
TEST=build and run.
Change-Id: I4eb54a3f6eb30c98971aa2f099e3ea5899767eed
Signed-off-by: Matt Delco <delco@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1825262
Reviewed-by: Zach Reizner <zachr@chromium.org>
Pipes have better performance than sockets, so switch the vcpu
communication over to pipes. The vm communication channels will
continue to use sockets since that communication isn't performance
critical (and those messages sometimes exchange file descriptors, and
that functionality requires sockets).
TEST=local compile and confirmed that my diagnostic plugin is still
happy. The time it takes to run my benchmark plugin has decreased by
20%. This combined with my prior commit results in a net wall-clock
time reduction of 32%.
BUG=None
Change-Id: I44c198d62a3bbe3b539ff6ac79707d02488876e3
Signed-off-by: Matt Delco <delco@google.com>
Reviewed-on: https://chromium-review.googlesource.com/1572873
Commit-Ready: Matt Delco <delco@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This change helps to improve performance in plugin communications by
removing unnecessary communication exchange.
The existing protocol basically requires the plugin to send a request
msg and wait for a reply msg. Prior to this change a plugin had to send
a wait request before it got a wait reply (which typically contains an IO
event notication). Similarly, when the plugin sends a resume request
there's also a resume reply that's sent.
The reply to the resume message serves no worthwhile purpose and can be
removed. In the common case there's also no need for the plugin to send
a wait request message--the prior operation was a resume so both sides
know that the only next legal operation is a wait. Thereforce, crosvm
can send a wait reply message without waiting for the plugin's request.
Another way to look at the situation is that a resume request message is
now answered by a wait reply message, and the overall message exchange
pattern looks less like http and more like async I/O.
The plugin's first call to wait is the one time that a wait request is
sent. This in turn will receive an wait-init reply.
TEST=Ran my diagnostic plugin and confirmed that it still passes (after
working around an 8-byte limitation in crosvm). Run my benchmarking
plugin and observed the time it takes to complete go down by 16.5%.
BUG=None
Change-Id: I9c93ba1d3a8f7814ca952f3dc7239d48675192e2
Signed-off-by: Matt Delco <delco@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1571066
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Matt Delco <delco@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This CL fixes four cases of what I believe are undefined behavior:
- In vhost where the original code allocates a Vec<u8> with 1-byte
alignment and casts the Vec's data pointer to a &mut vhost_memory
which is required to be 8-byte aligned. Underaligned references of
type &T or &mut T are always undefined behavior in Rust.
- Same pattern in x86_64.
- Same pattern in plugin::vcpu.
- Code in crosvm_plugin that dereferences a potentially underaligned
pointer. This is always undefined behavior in Rust.
TEST=bin/clippy
TEST=cargo test sys_util
Change-Id: I926f17b1fe022a798f69d738f9990d548f40c59b
Reviewed-on: https://chromium-review.googlesource.com/1566736
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: David Tolnay <dtolnay@chromium.org>
Before the new borrow checker in the 2018 edition, we sometimes used to
have to manually insert curly braced blocks to limit the scope of
borrows. These are no longer needed.
Details in:
https://doc.rust-lang.org/edition-guide/rust-2018/ownership-and-lifetimes/non-lexical-lifetimes.html
TEST=cargo check --all-features
TEST=local kokoro
Change-Id: I59f9f98dcc03c8790c53e080a527ad9b68c8d6f3
Reviewed-on: https://chromium-review.googlesource.com/1568075
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
In Rust 2018 edition, `extern crate` is no longer required for importing
from other crates. Instead of writing:
extern crate dep;
use dep::Thing;
we write:
use dep::Thing;
In this approach, macros are imported individually from the declaring
crate rather than through #[macro_use]. Before:
#[macro_use]
extern crate sys_util;
After:
use sys_util::{debug, error};
The only place that `extern crate` continues to be required is in
importing the compiler's proc_macro API into a procedural macro crate.
This will hopefully be fixed in a future Rust release.
extern crate proc_macro;
TEST=cargo check
TEST=cargo check --all-features
TEST=cargo check --target aarch64-unknown-linux-gnu
TEST=local kokoro
Change-Id: I0b43768c0d81f2a250b1959fb97ba35cbac56293
Reviewed-on: https://chromium-review.googlesource.com/1565302
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: David Tolnay <dtolnay@chromium.org>
This de-duplicates the two separate build.rs files dealing with proto
compilation. The trunks interface.proto will be exposed under
protos::trunks and the plugin proto will be exposed under protos::plugin.
BUG=none
TEST=cargo check
TEST=cargo check --features tpm
TEST=cargo check --features plugin
TEST=cargo check --features tpm,plugin
TEST=FEATURES=test emerge-nami crosvm
TEST=FEATURES=test USE=crosvm-tpm emerge-nami crosvm
TEST=FEATURES=test USE=crosvm-plugin emerge-nami crosvm
TEST=FEATURES=test USE='crosvm-tpm crosvm-plugin' emerge-nami crosvm
TEST=local kokoro
CQ-DEPEND=CL:1553971
Change-Id: I203b654a38e9d671a508156ae06dfb6f70047c4f
Reviewed-on: https://chromium-review.googlesource.com/1556417
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
This matches the version already used by crostini_client.
The newer protobuf version depends on the tempfile crate rather than
tempdir, the latter being now deprecated. So I replaced our immitation
tempdir crate with one that matches the API of tempfile instead. As a
reminder, we use this crate as a patch to avoid pulling in all of the
rand crate and its many dependencies.
TEST=cargo check --features plugin
CQ-DEPEND=CL:1553971
Change-Id: I28eed3ceadb1013f015400b4c582aaf8dc89eee1
Reviewed-on: https://chromium-review.googlesource.com/1562924
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Separated out of CL:1513058 to make it possible to land parts
individually while the affected crate has no other significant CLs
pending. This avoids repeatedly introducing non-textual conflicts with
new code that adds `use` statements.
TEST=cargo check
TEST=cargo check --all-features
TEST=cargo check --target aarch64-unknown-linux-gnu
Change-Id: I78a8e95199da47843bf75bb9b2f0f3dd9899ac19
Reviewed-on: https://chromium-review.googlesource.com/1519703
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: David Tolnay <dtolnay@chromium.org>
This may help reduce cases of conflicts between independent CLs each
appending a dependency at the bottom of the list, of which I hit two
today rebasing some of my open CLs.
TEST=cargo check --all-features
Change-Id: Ief10bb004cc7b44b107dc3841ce36c6b23632aed
Reviewed-on: https://chromium-review.googlesource.com/1557172
Commit-Ready: David Tolnay <dtolnay@chromium.org>
Tested-by: David Tolnay <dtolnay@chromium.org>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
When compiling unit tests in the pre-cq, the Cargo.lock file is deleted.
Testing a sub-crate without a lock file will cause the protobuf
dependency to resolve to the latest version of protobuf, which varies
based on whatever other cros-rust.eclass based ebuilds have been run
beforehand. This change fixes that source of flake.
BUG=None
TEST=pre-cq
Change-Id: Ief951391c08b0f0cc9ff035437824d89860455e2
Reviewed-on: https://chromium-review.googlesource.com/1534962
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Stephen Barber <smbarber@chromium.org>
Remove the double underscore in front of `__has_extension` in crosvm.h.
Double underscores in identifiers are reserved for the compiler's
internal use and as it so happens, `__has_extension` is a macro that
clang defines for code to determine whether the compiler supports a
given feature.
We shouldn't be using double underscores in any of the variable names in
this header file but for now just fix the problematic one so that the
code can actually compile under clang.
BUG=b:80150167
TEST=Compile one of the test plugins with clang
Change-Id: Ibb59e72c968a7f245bd6cc693da99f9263eedf33
Signed-off-by: Chirantan Ekbote <chirantan@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1341100
Reviewed-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
Now that cargo fmt has landed, run it over everything at once to bring
rust source to the standard formatting.
TEST=cargo test
BUG=None
Change-Id: Ic95a48725e5a40dcbd33ba6d5aef2bd01e91865b
Reviewed-on: https://chromium-review.googlesource.com/1259287
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The Scm object was made to reduce the number of heap allocations in
the hot paths of poll loops, at the cost of some code complexity. As it
turns out, the number of file descriptors being sent or received is
usually just one or limited to a fixed amount that can easily be covered
with a fixed size stack allocated buffer.
This change implements that solution, with heap allocation as a backup
in the rare case that many file descriptors must be sent or received.
This change also moves the msg and cmsg manipulation code out of C and
into pure Rust. The move was necessary to allocate the correct amount
of buffer space at compile time. It also improves safety by reducing the
scope of unsafe code. Deleting the code for building the C library is
also a nice bonus.
Finally, the removal of the commonly used Scm struct required
transitioning existing usage to the ScmSocket trait based methods. This
includes all those changes.
TEST=cargo test
BUG=None
Change-Id: If27ba297f5416dd9b8bc686ce740866912fa0aa0
Reviewed-on: https://chromium-review.googlesource.com/1186146
Commit-Ready: ChromeOS CL Exonerator Bot <chromiumos-cl-exonerator@appspot.gserviceaccount.com>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
KVM_GET_MSRS may return less MSRs that were requested; do not fail but
instead let callers to know how many were fetched.
BUG=None
TEST=cargo test --features plugin
Change-Id: Ie14a3d38b66bfe34f5279543bea9c6c78423527e
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1192232
Reviewed-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Userspace is interested in number of supported MSRs even if supplied
buffer is too small, as then it can intelligently [re]allocate the
buffer and repeat the call instead of doing this blindly. So let's
always populate 'out_count'in crosvm_get_msr_index_list() call.
Obviously if there is hard error we will not be able to supply a
meaningful number, so 0 will be returned, but in case of E2BIG error we
can return the real number.
Also let's do the same for get_supported_cpuids() and
get_emulated_cpuids() calls.
BUG=b:111083877
TEST=cargo test -p kvm; cargo test --features=plugin
Change-Id: I37a8d719103fac44597b88ddecb6b8af2dd54ac8
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/1185293
Reviewed-by: Zach Reizner <zachr@chromium.org>
A simple stat collecting. Uses RAII to gather latency on all exit
paths from function/block. The underscore in "let _u = STATS.u(...)" is
to pacify "unused variable" warning. Using "let _ = " makes compiler
optimize out the call.
Rust makes it particularly hard to convert enums from integers, so I had
to add a hack that stores Enum on every invocation of the STATS.u. Looking
at disassembly, it added one move of constant to the field of STATS.entries;
no heap operations or cloning. A clever alternative using macros was
suggested by semenzato@, but I decided saving an instruction was not
worth the complexity.
The output is currently printed on the destruction of crosvm, so tests
print out stats on exit. We probably should find a better place for it
though.
BUG=None
TEST=cargo test --release --features plugin
Change-Id: I78a8920e9896b717af3aaea14f8ed6013be6b94f
Reviewed-on: https://chromium-review.googlesource.com/1036473
Commit-Ready: Slava Malyugin <slavamn@chromium.org>
Tested-by: Slava Malyugin <slavamn@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This change allows plugin to retrieve and set various VM and VCPU states:
interrupt controller, PIT, LAPIC and MP state.
BUG=b:76083711
TEST=cargo test -p kvm
Change-Id: Ie32a67b0cd4a1f0a19ccd826a6e1c9dc25670f95
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/986511
Reviewed-by: Zach Reizner <zachr@chromium.org>
We have decided that API defined in crosvm.h should signal errors by
returning negative values derived from common errno error codes. To
avoid confusion within the rest of crosvm code that is using positive
erro codes, let's perform the conversion to negative on that crosvm API
boundary, so it is contained.
TEST=cargo test --features plugin
BUG=None
Change-Id: Icb1e719c8e99c95fdc32dce13a30d6ff8d3d9dc7
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/947563
Reviewed-by: Zach Reizner <zachr@chromium.org>
This plumbs calls to KVM_GET_SUPPORTED_CPUID and KVM_GET_EMULATED_CPUID
to be available to plugins.
TEST=cargo test --features plugin; cargo test -p kvm
BUG=chromium:800626
Change-Id: I98879599b5f970c6c2720772658689a505d8abe1
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/938674
Reviewed-by: Zach Reizner <zachr@chromium.org>
We need to convert between protobuf and KVM format of cpuid data in
several places, so let's add helpers to plugin_proto crate.
TEST=cargo test --features plugin; cargo test -p kvm
BUG=chromium:800626
Change-Id: Ida7b59825d2146b0d02711e441f477d90dd4263a
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/939660
Reviewed-by: Zach Reizner <zachr@chromium.org>
The raw_os_error() and errno() return positive values (errno values are
all positive), but the rest of crosvm plugin C API works with negative
return codes, so we need to convert raw_os_error()/errno() into
negatives as well.
TEST=cargo test --features plugin; cargo test -p kvm
BUG=None
Change-Id: I8bd72c2e67cb227a638e5c9478cd2f781f0783d0
Signed-off-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/939865
Reviewed-by: Zach Reizner <zachr@chromium.org>
The guest may need to check for KVM extensions before blindly using
them.
TEST=cargo test --features plugin; cargo test -p kvm; ./build_test
BUG=chromium:800626
Change-Id: If87b928753cd71adeabac4fc7732c3fce7265834
Reviewed-on: https://chromium-review.googlesource.com/906008
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
The guest expects to be able to read the CPUID, so the plugin process
needs to specify what the CPUID for each VCPU will have.
TEST=cargo test --features plugin; ./build_test
BUG=chromium:800626
Change-Id: I9258540ab2501126c3d8cadbd09b7fc01d19f7a9
Reviewed-on: https://chromium-review.googlesource.com/906006
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
Dirty logging is not necessary for every memory region, so the plugin
process should be able to specific exactly which regions it would like
dirty logging enabled for.
TEST=cargo test --features plugin
BUG=chromium:800626
Change-Id: I28b8285357e7de1c8c3a1392bdfdb4853ec5a654
Reviewed-on: https://chromium-review.googlesource.com/900294
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
The MSRs are useful for booting a full operating system that requires
them.
TEST=cargo test --features plugin; cargo test -p kvm; ./build_test
BUG=chromium:800626
Change-Id: I817fbf3e6868c85b373808bd48e568b5b2b458eb
Reviewed-on: https://chromium-review.googlesource.com/897412
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
The debug registers are useful to access for the plugin process in some
cases.
TEST=cargo test --features plugin; cargo test -p kvm; ./build_test
BUG=chromium:800626
Change-Id: I8f3f6c31c6989061a43cef948cf5b4e64bd52d30
Reviewed-on: https://chromium-review.googlesource.com/896945
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This is the shared library used by the plugin process to speak to a
crosvm main process that spawned it.
TEST=cargo build --features plugin
BUG=chromium:800626
Change-Id: I100e7ddfc1099fbdf1462c171785a861e075d5d7
Reviewed-on: https://chromium-review.googlesource.com/869356
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Zach Reizner <zachr@chromium.org>
This header file defines the C API used to interface with crosvm as a
plugin process.
TEST=None
BUG=chromium:800626
Change-Id: Ie06b833e25dab8f31f64d8bc8b4b521b61d1ca04
Reviewed-on: https://chromium-review.googlesource.com/764267
Commit-Ready: Zach Reizner <zachr@chromium.org>
Tested-by: Zach Reizner <zachr@chromium.org>
Reviewed-by: Dmitry Torokhov <dtor@chromium.org>
Reviewed-by: Dylan Reid <dgreid@chromium.org>
