The implementation is pretty straightforward. Wasm atomic instructions fall
into 5 groups
* atomic read-modify-write
* atomic compare-and-swap
* atomic loads
* atomic stores
* fences
and the implementation mirrors that structure, at both the CLIF and AArch64
levels.
At the CLIF level, there are five new instructions, one for each group. Some
comments about these:
* for those that take addresses (all except fences), the address is contained
entirely in a single `Value`; there is no offset field as there is with
normal loads and stores. Wasm atomics require alignment checks, and
removing the offset makes implementation of those checks a bit simpler.
* atomic loads and stores get their own instructions, rather than reusing the
existing load and store instructions, for two reasons:
- per above comment, makes alignment checking simpler
- reuse of existing loads and stores would require extension of `MemFlags`
to indicate atomicity, which sounds semantically unclean. For example,
then *any* instruction carrying `MemFlags` could be marked as atomic, even
in cases where it is meaningless or ambiguous.
* I tried to specify, in comments, the behaviour of these instructions as
tightly as I could. Unfortunately there is no way (per my limited CLIF
knowledge) to enforce the constraint that they may only be used on I8, I16,
I32 and I64 types, and in particular not on floating point or vector types.
The translation from Wasm to CLIF, in `code_translator.rs` is unremarkable.
At the AArch64 level, there are also five new instructions, one for each
group. All of them except `::Fence` contain multiple real machine
instructions. Atomic r-m-w and atomic c-a-s are emitted as the usual
load-linked store-conditional loops, guarded at both ends by memory fences.
Atomic loads and stores are emitted as a load preceded by a fence, and a store
followed by a fence, respectively. The amount of fencing may be overkill, but
it reflects exactly what the SM Wasm baseline compiler for AArch64 does.
One reason to implement r-m-w and c-a-s as a single insn which is expanded
only at emission time is that we must be very careful what instructions we
allow in between the load-linked and store-conditional. In particular, we
cannot allow *any* extra memory transactions in there, since -- particularly
on low-end hardware -- that might cause the transaction to fail, hence
deadlocking the generated code. That implies that we can't present the LL/SC
loop to the register allocator as its constituent instructions, since it might
insert spills anywhere. Hence we must present it as a single indivisible
unit, as we do here. It also has the benefit of reducing the total amount of
work the RA has to do.
The only other notable feature of the r-m-w and c-a-s translations into
AArch64 code, is that they both need a scratch register internally. Rather
than faking one up by claiming, in `get_regs` that it modifies an extra
scratch register, and having to have a dummy initialisation of it, these new
instructions (`::LLSC` and `::CAS`) simply use fixed registers in the range
x24-x28. We rely on the RA's ability to coalesce V<-->R copies to make the
cost of the resulting extra copies zero or almost zero. x24-x28 are chosen so
as to be call-clobbered, hence their use is less likely to interfere with long
live ranges that span calls.
One subtlety regarding the use of completely fixed input and output registers
is that we must be careful how the surrounding copy from/to of the arg/result
registers is done. In particular, it is not safe to simply emit copies in
some arbitrary order if one of the arg registers is a real reg. For that
reason, the arguments are first moved into virtual regs if they are not
already there, using a new method `<LowerCtx for Lower>::ensure_in_vreg`.
Again, we rely on coalescing to turn them into no-ops in the common case.
There is also a ridealong fix for the AArch64 lowering case for
`Opcode::Trapif | Opcode::Trapff`, which removes a bug in which two trap insns
in a row were generated.
In the patch as submitted there are 6 "FIXME JRS" comments, which mark things
which I believe to be correct, but for which I would appreciate a second
opinion. Unless otherwise directed, I will remove them for the final commit
but leave the associated code/comments unchanged.
* wasmtime: Implement `global.{get,set}` for externref globals
We use libcalls to implement these -- unlike `table.{get,set}`, for which we
create inline JIT fast paths -- because no known toolchain actually uses
externref globals.
Part of #929
* wasmtime: Enable `{extern,func}ref` globals in the API
These instructions have fast, inline JIT paths for the common cases, and only
call out to host VM functions for the slow paths. This required some changes to
`cranelift-wasm`'s `FuncEnvironment`: instead of taking a `FuncCursor` to insert
an instruction sequence within the current basic block,
`FuncEnvironment::translate_table_{get,set}` now take a `&mut FunctionBuilder`
so that they can create whole new basic blocks. This is necessary for
implementing GC read/write barriers that involve branching (e.g. checking for
null, or whether a store buffer is at capacity).
Furthermore, it required that the `load`, `load_complex`, and `store`
instructions handle loading and storing through an `r{32,64}` rather than just
`i{32,64}` addresses. This involved making `r{32,64}` types acceptable
instantiations of the `iAddr` type variable, plus a few new instruction
encodings.
Part of #929
`funcref`s are implemented as `NonNull<VMCallerCheckedAnyfunc>`.
This should be more efficient than using a `VMExternRef` that points at a
`VMCallerCheckedAnyfunc` because it gets rid of an indirection, dynamic
allocation, and some reference counting.
Note that the null function reference is *NOT* a null pointer; it is a
`VMCallerCheckedAnyfunc` that has a null `func_ptr` member.
Part of #929
This serves two purposes:
1. It ensures that we call `get_or_create_table` to ensure that the embedder
already had a chance to create the given table (although this is mostly
redundant due to validation).
2. It allows the embedder to easily get the `ir::TableData` associated with this
table, and more easily emit whatever inline JIT code to translate the table
instruction (rather than falling back to VM calls).
This commit enables `wasmtime_runtime::Table` to internally hold elements of
either `funcref` (all that is currently supported) or `externref` (newly
introduced in this commit).
This commit updates `Table`'s API, but does NOT generally propagate those
changes outwards all the way through the Wasmtime embedding API. It only does
enough to get everything compiling and the current test suite passing. It is
expected that as we implement more of the reference types spec, we will bubble
these changes out and expose them to the embedding API.
For host VM code, we use plain reference counting, where cloning increments
the reference count, and dropping decrements it. We can avoid many of the
on-stack increment/decrement operations that typically plague the
performance of reference counting via Rust's ownership and borrowing system.
Moving a `VMExternRef` avoids mutating its reference count, and borrowing it
either avoids the reference count increment or delays it until if/when the
`VMExternRef` is cloned.
When passing a `VMExternRef` into compiled Wasm code, we don't want to do
reference count mutations for every compiled `local.{get,set}`, nor for
every function call. Therefore, we use a variation of **deferred reference
counting**, where we only mutate reference counts when storing
`VMExternRef`s somewhere that outlives the activation: into a global or
table. Simultaneously, we over-approximate the set of `VMExternRef`s that
are inside Wasm function activations. Periodically, we walk the stack at GC
safe points, and use stack map information to precisely identify the set of
`VMExternRef`s inside Wasm activations. Then we take the difference between
this precise set and our over-approximation, and decrement the reference
count for each of the `VMExternRef`s that are in our over-approximation but
not in the precise set. Finally, the over-approximation is replaced with the
precise set.
The `VMExternRefActivationsTable` implements the over-approximized set of
`VMExternRef`s referenced by Wasm activations. Calling a Wasm function and
passing it a `VMExternRef` moves the `VMExternRef` into the table, and the
compiled Wasm function logically "borrows" the `VMExternRef` from the
table. Similarly, `global.get` and `table.get` operations clone the gotten
`VMExternRef` into the `VMExternRefActivationsTable` and then "borrow" the
reference out of the table.
When a `VMExternRef` is returned to host code from a Wasm function, the host
increments the reference count (because the reference is logically
"borrowed" from the `VMExternRefActivationsTable` and the reference count
from the table will be dropped at the next GC).
For more general information on deferred reference counting, see *An
Examination of Deferred Reference Counting and Cycle Detection* by Quinane:
https://openresearch-repository.anu.edu.au/bitstream/1885/42030/2/hon-thesis.pdf
cc #929Fixes#1804
This is enough to get an `externref -> externref` identity function
passing.
However, `externref`s that are dropped by compiled Wasm code are (safely)
leaked. Follow up work will leverage cranelift's stack maps to resolve this
issue.
About half of the `FuncEnvironment::translate_table_*` methods were using the
`TableIndex` newtype, while the other half were using raw `u32`s. This commit
makes everything use `TableIndex`.
* Implement trap info in Lightbeam
* Start using wasm-reader instead of wasmparser for parsing operators
* Update to use wasm-reader, some reductions in allocation, support source location tracking for traps, start to support multi-value
The only thing that still needs to be supported for multi-value is stack returns, but we need to make it compatible with Cranelift.
* Error when running out of registers (although we'd hope it should be impossible) instead of panicking
* WIP: Update Lightbeam to work with latest Wasmtime
* WIP: Update Lightbeam to use current wasmtime
* WIP: Migrate to new system for builtin functions
* WIP: Update Lightbeam to work with latest Wasmtime
* Remove multi_mut
* Format
* Fix some bugs around arguments, add debuginfo offset tracking
* Complete integration with new Wasmtime
* Remove commented code
* Fix formatting
* Fix warnings, remove unused dependencies
* Fix `iter` if there are too many elements, fix compilation for latest wasmtime
* Fix float arguments on stack
* Remove wasm-reader and trap info work
* Allocate stack space _before_ passing arguments, fail if we can't zero a xmm reg
* Fix stack argument offset calculation
* Fix stack arguments in Lightbeam
* Re-add WASI because it somehow got removed during rebase
* Workaround for apparent `type_alias_impl_trait`-related bug in rustdoc
* Fix breakages caused by rebase, remove module offset info as it is unrelated to wasmtime integration PR and was broken by rebase
* Add TODO comment explaining `lightbeam::ModuleContext` trait
* Implement interrupting wasm code, reimplement stack overflow
This commit is a relatively large change for wasmtime with two main
goals:
* Primarily this enables interrupting executing wasm code with a trap,
preventing infinite loops in wasm code. Note that resumption of the
wasm code is not a goal of this commit.
* Additionally this commit reimplements how we handle stack overflow to
ensure that host functions always have a reasonable amount of stack to
run on. This fixes an issue where we might longjmp out of a host
function, skipping destructors.
Lots of various odds and ends end up falling out in this commit once the
two goals above were implemented. The strategy for implementing this was
also lifted from Spidermonkey and existing functionality inside of
Cranelift. I've tried to write up thorough documentation of how this all
works in `crates/environ/src/cranelift.rs` where gnarly-ish bits are.
A brief summary of how this works is that each function and each loop
header now checks to see if they're interrupted. Interrupts and the
stack overflow check are actually folded into one now, where function
headers check to see if they've run out of stack and the sentinel value
used to indicate an interrupt, checked in loop headers, tricks functions
into thinking they're out of stack. An interrupt is basically just
writing a value to a location which is read by JIT code.
When interrupts are delivered and what triggers them has been left up to
embedders of the `wasmtime` crate. The `wasmtime::Store` type has a
method to acquire an `InterruptHandle`, where `InterruptHandle` is a
`Send` and `Sync` type which can travel to other threads (or perhaps
even a signal handler) to get notified from. It's intended that this
provides a good degree of flexibility when interrupting wasm code. Note
though that this does have a large caveat where interrupts don't work
when you're interrupting host code, so if you've got a host import
blocking for a long time an interrupt won't actually be received until
the wasm starts running again.
Some fallout included from this change is:
* Unix signal handlers are no longer registered with `SA_ONSTACK`.
Instead they run on the native stack the thread was already using.
This is possible since stack overflow isn't handled by hitting the
guard page, but rather it's explicitly checked for in wasm now. Native
stack overflow will continue to abort the process as usual.
* Unix sigaltstack management is now no longer necessary since we don't
use it any more.
* Windows no longer has any need to reset guard pages since we no longer
try to recover from faults on guard pages.
* On all targets probestack intrinsics are disabled since we use a
different mechanism for catching stack overflow.
* The C API has been updated with interrupts handles. An example has
also been added which shows off how to interrupt a module.
Closes#139Closes#860Closes#900
* Update comment about magical interrupt value
* Store stack limit as a global value, not a closure
* Run rustfmt
* Handle review comments
* Add a comment about SA_ONSTACK
* Use `usize` for type of `INTERRUPTED`
* Parse human-readable durations
* Bring back sigaltstack handling
Allows libstd to print out stack overflow on failure still.
* Add parsing and emission of stack limit-via-preamble
* Fix new example for new apis
* Fix host segfault test in release mode
* Fix new doc example
* Compute instance exports on demand.
Instead having instances eagerly compute a Vec of Externs, and bumping
the refcount for each Extern, compute Externs on demand.
This also enables `Instance::get_export` to avoid doing a linear search.
This also means that the closure returned by `get0` and friends now
holds an `InstanceHandle` to dynamically hold the instance live rather
than being scoped to a lifetime.
* Compute module imports and exports on demand too.
And compute Extern::ty on demand too.
* Add a utility function for computing an ExternType.
* Add a utility function for looking up a function's signature.
* Add a utility function for computing the ValType of a Global.
* Rename wasmtime_environ::Export to EntityIndex.
This helps differentiate it from other Export types in the tree, and
describes what it is.
* Fix a typo in a comment.
* Simplify module imports and exports.
* Make `Instance::exports` return the export names.
This significantly simplifies the public API, as it's relatively common
to need the names, and this avoids the need to do a zip with
`Module::exports`.
This also changes `ImportType` and `ExportType` to have public members
instead of private members and accessors, as I find that simplifies the
usage particularly in cases where there are temporary instances.
* Remove `Instance::module`.
This doesn't quite remove `Instance`'s `module` member, it gets a step
closer.
* Use a InstanceHandle utility function.
* Don't consume self in the `Func::get*` methods.
Instead, just create a closure containing the instance handle and the
export for them to call.
* Use `ExactSizeIterator` to avoid needing separate `num_*` methods.
* Rename `Extern::func()` etc. to `into_func()` etc.
* Revise examples to avoid using `nth`.
* Add convenience methods to instance for getting specific extern types.
* Use the convenience functions in more tests and examples.
* Avoid cloning strings for `ImportType` and `ExportType`.
* Remove more obviated clone() calls.
* Simplify `Func`'s closure state.
* Make wasmtime::Export's fields private.
This makes them more consistent with ExportType.
* Fix compilation error.
* Make a lifetime parameter explicit, and use better lifetime names.
Instead of 'me, use 'instance and 'module to make it clear what the
lifetime is.
* More lifetime cleanups.
* Consolidate trap/frame information
This commit removes `TrapRegistry` in favor of consolidating this
information in the `FRAME_INFO` we already have in the `wasmtime` crate.
This allows us to keep information generally in one place and have one
canonical location for "map this PC to some original wasm stuff". The
intent for this is to next update with enough information to go from a
program counter to a position in the original wasm file.
* Expose module offset information in `FrameInfo`
This commit implements functionality for `FrameInfo`, the wasm stack
trace of a `Trap`, to return the module/function offset. This allows
knowing the precise wasm location of each stack frame, instead of only
the main trap itself. The intention here is to provide more visibility
into the wasm source when something traps, so you know precisely where
calls were and where traps were, in order to assist in debugging.
Eventually we might use this information for mapping back to native
source languages as well (given sufficient debug information).
This change makes a previously-optional artifact of compilation always
computed on the cranelift side of things. This `ModuleAddressMap` is
then propagated to the same store of information other frame information
is stored within. This also removes the need for passing a `SourceLoc`
with wasm traps or to wasm trap creation, since the backtrace's wasm
frames will be able to infer their own `SourceLoc` from the relevant
program counters.
* Enable the already-passing `bulk-memoryoperations/imports.wast` test
* Implement support for the `memory.init` instruction and passive data
This adds support for passive data segments and the `memory.init` instruction
from the bulk memory operations proposal. Passive data segments are stored on
the Wasm module and then `memory.init` instructions copy their contents into
memory.
* Implement the `data.drop` instruction
This allows wasm modules to deallocate passive data segments that it doesn't
need anymore. We keep track of which segments have not been dropped on an
`Instance` and when dropping them, remove the entry from the instance's hash
map. The module always needs all of the segments for new instantiations.
* Enable final bulk memory operations spec test
This requires special casing an expected error message for an `assert_trap`,
since the expected error message contains the index of an uninitialized table
element, but our trap implementation doesn't save that diagnostic information
and shepherd it out.
This adds support for the `table.copy` instruction from the bulk memory
proposal. It also supports multiple tables, which were introduced by the
reference types proposal.
Part of #928
* Improve robustness of cache loading/storing
Today wasmtime incorrectly loads compiled compiled modules from the
global cache when toggling settings such as optimizations. For example
if you execute `wasmtime foo.wasm` that will cache globally an
unoptimized version of the wasm module. If you then execute `wasmtime -O
foo.wasm` it would then reload the unoptimized version from cache, not
realizing the compilation settings were different, and use that instead.
This can lead to very surprising behavior naturally!
This commit updates how the cache is managed in an attempt to make it
much more robust against these sorts of issues. This takes a leaf out of
rustc's playbook and models the cache with a function that looks like:
fn load<T: Hash>(
&self,
data: T,
compute: fn(T) -> CacheEntry,
) -> CacheEntry;
The goal here is that it guarantees that all the `data` necessary to
`compute` the result of the cache entry is hashable and stored into the
hash key entry. This was previously open-coded and manually managed
where items were hashed explicitly, but this construction guarantees
that everything reasonable `compute` could use to compile the module is
stored in `data`, which is itself hashable.
This refactoring then resulted in a few workarounds and a few fixes,
including the original issue:
* The `Module` type was split into `Module` and `ModuleLocal` where only
the latter is hashed. The previous hash function for a `Module` left
out items like the `start_func` and didn't hash items like the imports
of the module. Omitting the `start_func` was fine since compilation
didn't actually use it, but omitting imports seemed uncomfortable
because while compilation didn't use the import values it did use the
*number* of imports, which seems like it should then be put into the
cache key. The `ModuleLocal` type now derives `Hash` to guarantee that
all of its contents affect the hash key.
* The `ModuleTranslationState` from `cranelift-wasm` doesn't implement
`Hash` which means that we have a manual wrapper to work around that.
This will be fixed with an upstream implementation, since this state
affects the generated wasm code. Currently this is just a map of
signatures, which is present in `Module` anyway, so we should be good
for the time being.
* Hashing `dyn TargetIsa` was also added, where previously it was not
fully hashed. Previously only the target name was used as part of the
cache key, but crucially the flags of compilation were omitted (for
example the optimization flags). Unfortunately the trait object itself
is not hashable so we still have to manually write a wrapper to hash
it, but we likely want to add upstream some utilities to hash isa
objects into cranelift itself. For now though we can continue to add
hashed fields as necessary.
Overall the goal here was to use the compiler to expose what we're not
hashing, and then make sure we organize data and write the right code to
ensure everything is hashed, and nothing more.
* Update crates/environ/src/module.rs
Co-Authored-By: Peter Huene <peterhuene@protonmail.com>
* Fix lightbeam
* Fix compilation of tests
* Update the expected structure of the cache
* Revert "Update the expected structure of the cache"
This reverts commit 2b53fee426a4e411c313d8c1e424841ba304a9cd.
* Separate the cache dir a bit
* Add a test the cache is busted with opt levels
* rustfmt
Co-authored-by: Peter Huene <peterhuene@protonmail.com>
This commit shrinks the `RelocationTarget` enumeration to remove
intrinsic-related relocations since they are no longer used. Instead
these function calls are done indirectly via a table in the `VMContext`.
This means that all of this is essentially dead code!
* Update cranelift to 0.58.0
* Update `wasmprinter` dep to require 0.2.1
We already had it in the lock file, but this ensures we won't ever go back down.
* Ensure that our error messages match `assert_invalid`'s
The bulk of this work was done in
https://github.com/bytecodealliance/wasmparser/pull/186 but now we can test it
at the `wasmtime` level as well.
Fixes#492
* Stop feeling guilty about not matching `assert_malformed` messages
Remove the "TODO" and stop printing warning messages. These would just be busy
work to implement, and getting all the messages the exact same relies on using
the same structure as the spec interpreter's parser, which means that where you
have a helper function and they don't, then things go wrong, and vice versa. Not
worth it.
Fixes#492
* Enable (but ignore) the reference-types proposal tests
* Match test suite directly, instead of roundabout starts/endswith
* Enable (but ignore) bulk memory operations proposal test suite
* Replace the global-exports mechanism with a caller-vmctx mechanism.
This eliminates the global exports mechanism, and instead adds a
caller-vmctx argument to wasm functions so that WASI can obtain the
memory and other things from the caller rather than looking them up in a
global registry.
This replaces #390.
* Fixup some merge conflicts
* Rustfmt
* Ensure VMContext is aligned to 16 bytes
With the removal of `global_exports` it "just so happens" that this
isn't happening naturally any more.
* Fixup some bugs with double vmctx in wasmtime crate
* Trampoline stub needed adjusting
* Use pointer type instead of always using I64 for caller vmctx
* Don't store `ir::Signature` in `Func` since we don't know the pointer
size at creation time.
* Skip the first 2 arguments in IR signatures since that's the two vmctx
parameters.
* Update cranelift to 0.56.0
* Handle more merge conflicts
* Rustfmt
Co-authored-by: Alex Crichton <alex@alexcrichton.com>
* Update to the latest spec_testsuite and dependencies.
Update to target-lexicon 0.10, cranelift 0.54, wast 0.6, faerie 0.14,
and the latest spec_testsuite.
For wast and cranelift-wasm, update the code for API changes.
* Factor out the code for matching f32, f64, and v128.
This takes the idea from #802 to split out `f32_matches`, `f64_matches`,
and `v128_matches` functions, which better factor out the matching
functionality between scalar and vector.
* Add unimplemented stubs for Cranelift interfaces
Cranelift changes to FuncEnvironment, TargetEnvironment, and GlobalInit (see https://github.com/bytecodealliance/cranelift/pull/1073) require these changes to compile wasmtime.
* Upgrade Cranelift to 0.52.0
* Migrate back to `std::` stylistically
This commit moves away from idioms such as `alloc::` and `core::` as
imports of standard data structures and types. Instead it migrates all
crates to uniformly use `std::` for importing standard data structures
and types. This also removes the `std` and `core` features from all
crates to and removes any conditional checking for `feature = "std"`
All of this support was previously added in #407 in an effort to make
wasmtime/cranelift "`no_std` compatible". Unfortunately though this
change comes at a cost:
* The usage of `alloc` and `core` isn't idiomatic. Especially trying to
dual between types like `HashMap` from `std` as well as from
`hashbrown` causes imports to be surprising in some cases.
* Unfortunately there was no CI check that crates were `no_std`, so none
of them actually were. Many crates still imported from `std` or
depended on crates that used `std`.
It's important to note, however, that **this does not mean that wasmtime
will not run in embedded environments**. The style of the code today and
idioms aren't ready in Rust to support this degree of multiplexing and
makes it somewhat difficult to keep up with the style of `wasmtime`.
Instead it's intended that embedded runtime support will be added as
necessary. Currently only `std` is necessary to build `wasmtime`, and
platforms that natively need to execute `wasmtime` will need to use a
Rust target that supports `std`. Note though that not all of `std` needs
to be supported, but instead much of it could be configured off to
return errors, and `wasmtime` would be configured to gracefully handle
errors.
The goal of this PR is to move `wasmtime` back to idiomatic usage of
features/`std`/imports/etc and help development in the short-term.
Long-term when platform concerns arise (if any) they can be addressed by
moving back to `no_std` crates (but fixing the issues mentioned above)
or ensuring that the target in Rust has `std` available.
* Start filling out platform support doc
