This PR adds a new PolkaVM-based executor to Substrate.
- The executor can now be used to actually run a PolkaVM-based runtime,
and successfully produces blocks.
- The executor is always compiled-in, but is disabled by default.
- The `SUBSTRATE_ENABLE_POLKAVM` environment variable must be set to `1`
to enable the executor, in which case the node will accept both WASM and
PolkaVM program blobs (otherwise it'll default to WASM-only). This is
deliberately undocumented and not explicitly exposed anywhere (e.g. in
the command line arguments, or in the API) to disincentivize anyone from
enabling it in production. If/when we'll move this into production usage
I'll remove the environment variable and do it "properly".
- I did not use our legacy runtime allocator for the PolkaVM executor,
so currently every allocation inside of the runtime will leak guest
memory until that particular instance is destroyed. The idea here is
that I will work on the https://github.com/polkadot-fellows/RFCs/pull/4
which will remove the need for the legacy allocator under WASM, and that
will also allow us to use a proper non-leaking allocator under PolkaVM.
- I also did some minor cleanups of the WASM executor and deleted some
dead code.
No prdocs included since this is not intended to be an end-user feature,
but an unofficial experiment, and shouldn't affect any current
production user. Once this is production-ready a full Polkadot
Fellowship RFC will be necessary anyway.
This PR fixes a subtle bug in `wasm-builder` first introduced in
https://github.com/paritytech/polkadot-sdk/pull/1851 (sorry, my bad! I
should have caught this during review) where the status code of the
`cargo` subprocess is not properly checked, which results in builds
silently succeeding when they shouldn't (that is: if we successfully
build a runtime blob, and then modify the code so that it won't compile,
and recompile it again, then the build will succeed and silently use the
*old* blob).
cc @athei This is the bug you were seeing.
[edit]Also fixes a similar PolkaVM-specific bug where I accidentally
used the wrong comparison operator.[/edit]
This PR improves compatibility with RISC-V and PolkaVM, allowing more
runtimes to successfully compile.
In particular, it makes the following changes:
- The `sp-mmr-primitives` and `sp-consensus-beefy` crates
unconditionally required an `std`-only dependency; now they only require
those dependencies when the `std` feature is actually enabled. (Our
RISC-V target is, unlike WASM, a true `no_std` target where you can't
accidentally use stuff from `std` anymore.)
- One of our dependencies (the `bitvec` trace) uses a crate called
`radium` which doesn't compile under RISC-V due to incomplete
autodetection logic in their `build.rs` file. The good news is that this
is already fixed in the newest upstream version of `radium`, and the
newest version of `bitvec` uses it. The bad news is that the newest
version of `bitvec` is not currently released on crates.io, so we can't
use it. I've [created an
issue](https://github.com/ferrilab/ferrilab/issues/5) asking for a new
release, but in the meantime I forked the currently used `radium` 0.7,
[fixed the faulty
logic](https://github.com/paritytech/radium-0.7-fork/commit/ed66c8a294b138c67f93499644051d97d4c7fbda)
and used cargo's patching capabilities to use it for the RISC-V runtime
builds. This might be a little hacky, but it is the least intrusive way
to fix the problem, doesn't affect WASM builds at all, and we can
trivially remove it once a new `bitvec` is released.
- The new runtimes are added to the CI to make sure their compilation
doesn't break.
This PR adds initial support for building RISC-V runtimes targeting
PolkaVM.
- Setting the `SUBSTRATE_RUNTIME_TARGET=riscv` environment variable will
now build a RISC-V runtime instead of a WASM runtime.
- This only adds support for *building* runtimes; running them will need
a PolkaVM-based executor, which I will add in a future PR.
- Only building the minimal runtime is supported (building the Polkadot
runtime doesn't work *yet* due to one of the dependencies).
- The builder now sets a `substrate_runtime` cfg flag when building the
runtimes, with the idea being that instead of doing `#[cfg(not(feature =
"std"))]` or `#[cfg(target_arch = "wasm32")]` to detect that we're
building a runtime you'll do `#[cfg(substrate_runtime)]`. (Switching the
whole codebase to use this will be done in a future PR; I deliberately
didn't do this here to keep this PR minimal and reviewable.)
- Further renaming of things (e.g. types, environment variables and proc
macro attributes having "wasm" in their name) to be target-agnostic will
also be done in a future refactoring PR (while keeping backwards
compatibility where it makes sense; I don't intend to break anyone's
workflow or create unnecessary churn).
- This PR also fixes two bugs in the `wasm-builder` crate:
* The `RUSTC` environment variable is now removed when invoking the
compiler. This prevents the toolchain version from being overridden when
called from a `build.rs` script.
* When parsing the `rustup toolchain list` output the `(default)` is now
properly stripped and not treated as part of the version.
- I've also added a minimal CI job that makes sure this doesn't break in
the future. (cc @paritytech/ci)
cc @athei
------
Also, just a fun little tidbit: quickly comparing the size of the built
runtimes it seems that the PolkaVM runtime is slightly smaller than the
WASM one. (`production` build, with the `names` section substracted from
the WASM's size to keep things fair, since for the PolkaVM runtime we're
currently stripping out everything)
- `.wasm`: 625505 bytes
- `.wasm` (after wasm-opt -O3): 563205 bytes
- `.wasm` (after wasm-opt -Os): 562987 bytes
- `.wasm` (after wasm-opt -Oz): 536852 bytes
- `.polkavm`: ~~580338 bytes~~ 550476 bytes (after enabling extra target
features; I'll add those in another PR once we have an executor working)
---------
Co-authored-by: Bastian Köcher <git@kchr.de>
Our executor currently only supports the WASM MVP feature set, however
nowadays when compiling WASM the Rust compiler has more features enabled
by default.
We do set the `-C target-cpu=mvp` flag to make sure that *our* code gets
compiled in a way that is compatible with our executor, however this
doesn't affect Rust's standard library crates (`std`, `core` and
`alloc`) which are by default precompiled and still can make use of
these extra features.
So in this PR we force the compiler to also compile the standard library
crates for us to make sure that they also only use the MVP features.
I've added the `WASM_BUILD_STD` environment variable which can be used
to disable this behavior if set to `0`.
Unfortunately this *will* slow down the compile times when building
runtimes, but there isn't much that we can do about that.
Fixes https://github.com/paritytech/polkadot-sdk/issues/1755
---------
Co-authored-by: Bastian Köcher <git@kchr.de>
* Remove the `Copy` bound on `AssetId`
* Also relax the `Copy` bound in the assets pallet
* Fix warnings on the newest nightly Rust
* Remove some unnecessary `clone()`s
* Try to satisfy clippy
* Remove some more unnecessary `clone()`s
* Add more `.clone()`s for newly merged code
* Also add `clone()`s in the benchmarks
---------
Co-authored-by: parity-processbot <>
* Switch to the `blake2b_simd` crate in `sp-core-hashing`
* ".git/.scripts/commands/bench/bench.sh" pallet dev frame_benchmarking
---------
Co-authored-by: command-bot <>
* Speed up storage iteration from within the runtime
* Move the cached iterator into an `Option`
* Use `RefCell` in no_std
* Simplify the code slightly
* Use `Option::replace`
* Update doc comment for `next_storage_key_slow`
* Remove `Backend::apply_to_key_values_while`
* Add `IterArgs::start_at_exclusive`
* Use `start_at_exclusive` in functions which used `Backend::apply_to_key_values_while`
* Remove `Backend::apply_to_keys_while`
* Remove `for_keys_with_prefix`, `for_key_values_with_prefix` and `for_child_keys_with_prefix`
* Remove unnecessary `to_vec` calls
* Fix unused method warning in no_std
* Remove unnecessary import
* Also check proof sizes in the test
* Iterate over both keys and values in `prove_range_read_with_size` and add a test
* Rework storage iterators
* Make sure storage iteration is also accounted for when benchmarking
* Use `trie-db` from crates.io
* Appease clippy
* Bump `trie-bench` to 0.35.0
* Fix tests' compilation
* Update comment to clarify how `IterArgs::start_at` works
* Add extra tests
* Fix iterators on `Client` so that they behave as before
* Add extra `unwrap`s in tests
* More clippy fixes
* Come on clippy, give me a break already
* Rename `allow_missing` to `stop_on_incomplete_database`
* Add `#[inline]` to `with_recorder_and_cache`
* Use `with_recorder_and_cache` in `with_trie_db`; add doc comment
* Simplify code: use `with_trie_db` in `next_storage_key_from_root`
* Remove `expect`s in the benchmarking CLI
* Add extra doc comments
* Move `RawIter` before `TrieBackendEssence` (no code changes; just cut-paste)
* Remove a TODO in tests
* Update comment for `StorageIterator::was_complete`
* Update `trie-db` to 0.25.1
* Bump `wasmtime` to 4.0.0 (and a few other deps)
* Use `Error::msg` instead of `anyhow!`
* Bump `wasmtime` to 5.0.0
* Update `Cargo.lock`
* Add `wasmtime` feature to `sp-wasm-interface` dependency
* Remove lifetime from `KeyIterator`
* Remove the lifetime from the method too
* update lockfile for {"substrate"}
Co-authored-by: parity-processbot <>
* Have `KeyIterator` clone the `prefix` it receives
* Stream keys in `storage_size` RPC and add a runtime limit
* Update client/rpc/Cargo.toml
Co-authored-by: Bastian Köcher <git@kchr.de>
* Update client/rpc/src/state/utils.rs
Co-authored-by: Bastian Köcher <git@kchr.de>
* Rename the types to signify that the cancellation is due to a timeout
* Move the test into a `mod tests`
* Add a comment regarding `biased` in `tokio::select`
* Make the `clone` explicit when calling `KeyIterator::{new, new_child}`
Co-authored-by: Bastian Köcher <git@kchr.de>
* Remove the `wasmtime` feature flag
* Update `substrate` and `polkadot` to the newest `master`
* Update `substrate` and `polkadot` to the newest `master`
* Remove `sp_tasks::spawn` API and related code
* Remove `RuntimeTasks::{spawn, join}` host functions
* remove unused
* Remove a few more tests that I forgot to remove
Co-authored-by: Shawn Tabrizi <shawntabrizi@gmail.com>
* Force base weights to be the minimum only when the intercept is negative; emit minimum execution times
* Add an `assert` making sure the intercept is zero when it's supposed to be zero
* Fix template
Signed-off-by: Oliver Tale-Yazdi <oliver.tale-yazdi@parity.io>
* ".git/.scripts/bench-bot.sh" pallet dev pallet_assets
* ".git/.scripts/bench-bot.sh" pallet dev pallet_uniques
Signed-off-by: Oliver Tale-Yazdi <oliver.tale-yazdi@parity.io>
Co-authored-by: Oliver Tale-Yazdi <oliver.tale-yazdi@parity.io>
Co-authored-by: command-bot <>
* Support running the pallet benchmarks analysis without running the benchmarks
* Rename `override-results` to `json-input` and update the help comment
* ".git/.scripts/fmt.sh" 1
Co-authored-by: command-bot <>
* Improve base weights consistency and make sure they're never zero
* Switch back to `linregress` crate; add back estimation errors
* Move the test into `mod tests`
* Use all of the samples instead of only the first one
* Use full precision extrinsic base weights and slopes
* Add an extra comment
* ".git/.scripts/bench-bot.sh" pallet dev pallet_contracts
Co-authored-by: parity-processbot <>
Co-authored-by: Shawn Tabrizi <shawntabrizi@gmail.com>
* Restore `wasmtime`'s default stack size limit to 1MB
* Add extra comments
* Enforce different maximum call depth in release mode
* Split the call depth limit in two
* Align to changes in substrate: `sp_io::storage::get` now returns `Bytes`
* Update substrate and polkadot to the newest master
* update lockfile for {"substrate", "polkadot"}
Co-authored-by: parity-processbot <>
* Expose allocation stats in `FreeingBumpHeapAllocator`
* Return allocation stats when calling into the runtime
* Bump `parity-scale-codec` to 3.1.3 (fork)
* Prevent double allocation of the payload when calling `sp_io::storage::get`
* Fix tests
* Remove unnecessary `mut`
* Enable the `bytes` feature for `parity-scale-codec` in `sp-runtime-interface`
* Update client/allocator/src/freeing_bump.rs
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Bump `parity-scale-codec` to 3.1.3
* Fix some of the UI tests
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Prevend unsoundness in environments with broken `madvise(MADV_DONTNEED)`
* Add the `std` feature to `rustix` dependency
Apparently not having this breaks compilation on non-nightly toolchains.
* Autodetect the page size when checking whether `madvise` works
* Only make sure that the madvice check doesn't return `Err`
* Pump the gossip engine while waiting for the BEEFY runtime pallet
This fixes a memory leak when the BEEFY gadget is turned on, but
the runtime doesn't actually use BEEFY.
* Implement `FusedFuture` for `GossipEngine`
* Fuse futures outside of loops
* Run `sc-executor-wasmtime` unit tests for all instantiation strategies
* Adjust maximum memory pages hard limit for the pooling instantiation strategy
* Switch to pooling copy-on-write instantiation strategy for WASM
* Fix benchmark compilation
* Fix `cargo fmt`
* Fix compilation of another benchmark I've missed
* Cleanups according to review comments
* Move `max_memory_size` to `Semantics`
* Set `memory_guaranteed_dense_image_size` to `max_memory_size`
* Rename `wasm_instantiation_strategy` to `wasmtime_instantiation_strategy`
* Update the doc-comments regarding the instantiation strategy
* Extend the integration tests to test every instantiation strategy
* Don't drop the temporary directory until the runtime is dropped in benchmarks
* Don't drop the temporary directory until the runtime is dropped in tests
* When an RPC is rejected because it's RPC-unsafe say why in the error message
* Make wildcard storage change subscriptions RPC-unsafe
* Apply suggestions from code review
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Fix typo
* Fix tests
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Align to changes in Substrate
* Align to the newest changes in substrate
* Update `Cargo.lock`
* Add hwbenches to `parachain-template` too
* update lockfile for {"polkadot"}
Co-authored-by: parity-processbot <>
* Align to changes in Substrate
* Align to the newest changes in substrate
* Rename `--disable-hardware-benchmarks` to `--no-hardware-benchmarks`
* Fix `polkadot-test-service` compilation
* Fix compilation of test parachains
* Add new hardware and software metrics
* Move sysinfo tests into `mod tests`
* Correct a typo in a comment
* Remove unnecessary `nix` dependency
* Fix the version tests
* Add a `--disable-hardware-benchmarks` CLI argument
* Disable hardware benchmarks in the integration tests
* Remove unused import
* Fix benchmarks compilation
* Move code to a new `sc-sysinfo` crate
* Correct `impl_version` comment
* Move `--disable-hardware-benchmarks` to the chain-specific bin crate
* Move printing out of hardware bench results to `sc-sysinfo`
* Move hardware benchmarks to a separate messages; trigger them manually
* Rename some of the fields in the `HwBench` struct
* Revert changes to the telemetry crate; manually send hwbench messages
* Move sysinfo logs into the sysinfo crate
* Move the `TARGET_OS_*` constants into the sysinfo crate
* Minor cleanups
* Move the `HwBench` struct to the sysinfo crate
* Derive `Clone` for `HwBench`
* Fix broken telemetry connection notification stream
* Prevent the telemetry connection notifiers from leaking if they're disconnected
* Turn the telemetry notification failure log into a debug log
* Rename `--disable-hardware-benchmarks` to `--no-hardware-benchmarks`
* Catch panics on the FFI boundary between the runtime and the host for `wasmtime`
* Use an already existing test runtime function
* Merge the tests together
* Update `regex` to 1.5.5
* Update `libc` to 0.2.121
* Update `handlebars` to 4.2.2
* Update `lru` to 0.7.3
* Update `zeroize` to 1.5.4 and `zeroize_derive` to 1.3.2
* update lockfile for {"substrate"}
Co-authored-by: parity-processbot <>
* Rename to BagError
* Additional parameter for 'revert' command
* Set aux revert param to None
* Align to changes in how the WASM executor is configured in `substrate`
* update lockfile for {"substrate"}
* update lockfile for {"substrate"}
* Update substrate
* Update substrate
Co-authored-by: Keith Yeung <kungfukeith11@gmail.com>
Co-authored-by: Davide Galassi <davxy@datawok.net>
Co-authored-by: Shawn Tabrizi <shawntabrizi@gmail.com>
Co-authored-by: parity-processbot <>
* Add extra WASM heap pages when precompiling the runtime blob
* Fix compilation
* Fix rustdoc
* Fix rustdoc for real this time
* Fix benches compilation
* Improve the builder in `sc-executor-wasmtime`'s tests
* Refactor WASM module instantiation; enable WASM instance pooling
* Disable the `uffd` feature on `wasmtime`
* Restore the original behavior regarding the initial WASM memory size
* Adjust error message
* Remove unnecessary import in the benchmarks
* Preinstantiate the WASM runtime for a slight speedup
* Delete the asserts in `convert_memory_import_into_export`
* `return` -> `break`
* Revert WASM instance pooling for now
* Have `convert_memory_import_into_export` return an error instead of panic
* Update the warning when an import is missing
* Rustfmt and clippy fix
* Fix executor benchmarks' compilation without `wasmtime` being enabled
* rustfmt again
* Align to review comments
* Extend tests so that both imported and exported memories are tested
* Increase the number of heap pages for exported memories too
* Fix `decommit_works` test
* Bump `tokio` to 1.17.0
* Bump the `tokio` version in the test service too
* Revert changes to `bridges/relays/client-substrate/Cargo.toml`
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
* Simplify `num_connected_peers`
* Track requested peer counts
* Revert "Track requested peer counts"
This reverts commit 9f1c8704353df6afc17ed7e9f4ab8d8e29466ae4.
* Remove `substrate_sub_libp2p_peerset_num_requested` metric
* Remove two unused functions that I forgot to get rid of in previous commit
* Add a new host function for reporting fatal errors
* Fix one of the wasmtime executor tests
* Have `#[runtime_interface(wasm_only)]` actually mean WASM-only, and not no_std-only
* Print out errors through `Display` instead of `Debug`
* Switch one more trait to require `Error` for its error instead of only `Debug`
* Align to review comments
* Statically register host WASM functions
* Fix `substrate-test-client` compilation
* Move `ExtendedHostFunctions` to `sp-wasm-interface`
* Fix `sp-runtime-interface` tests' compilation
* Fix `sc-executor-wasmtime` tests' compilation
* Use `runtime_interface` macro in `test-runner`
* Fix `sc-executor` tests' compilation
* Reformatting/`rustfmt`
* Add an extra comment regarding the `H` generic arg in `create_runtime`
* Even more `rustfmt`
* Depend on `wasmtime` without default features in `sp-wasm-interface`
* Bump version of `sp-wasm-interface` to 4.0.1
* Bump `sp-wasm-interface` in `Cargo.lock` too
* Bump all of the `sp-wasm-interface` requirements to 4.0.1
Maybe this will appease cargo-unleash?
* Revert "Bump all of the `sp-wasm-interface` requirements to 4.0.1"
This reverts commit 0f7ccf8e0f371542861121b145ab87af6541ac30.
* Make `cargo-unleash` happy (maybe)
* Use `cargo-unleash` to bump the crates' versions
* Align to review comments
* Fix WASM executor without instance reuse; cleanups and refactoring
* Align to review comments
* Move the functions for reading/writing memory to `util.rs`
* Only `#[ignore]` the test in debug builds
* More review comments and minor extra comments
* Add a block production benchmark
* Simplify the block production benchmark
* Cleanups; switch execution strategy to WASM
* Switch WASM execution to `Compiled`
* Reduce the setup cost of the benchmark
Creating all of those extrinsics takes up *a lot* of time, up to the point
where the majority of the time is actually spent *outside* of the code
which we want to benchmark here. So let's only do it once.
* Add a variant of the block production benchmark with proof recording
* sp-sandbox: convert executors into normal `mod`s instead of using `include!`
* sp-sandbox: run `cargo fmt` on `host_executor.rs`
* sp-sandbox: abstract away the executors behind traits
* sp_sandbox: always compile both executors when possible
* sc-executor: make sure all sandbox tests run on both sandbox executors
* sc-executor: fix brainfart: actually call into the sandbox through the trait
* sc-runtime-test: fix cargo fmt
* sc-runtime-test: deduplicate executor-specific sandbox test entrypoints
* sc-executor: test each sandbox executor in a separate test
* cargo fmt (Github's conflict resolving thingy broke indentation)
* Strip out control codes from the logged messages
* Also strip away C1 control codes
* Add extra comments
* Clear the buffer after flushing; rename `write` to `flush`
* Move control code stripping into its own function
* Also strip out control codes from panic messages
* Also strip out Unicode left-to-right/right-to-left control codes
* Update `tracing`-related dependencies
* Enable `parking_lot` feature in `tracing-subscriber`
* Add an asynchronous stderr logger
* Make clippy happy
* Add an integration test for the logger
* Refactor `test_logger_filters`'s subprocess machinery into a separate function
* Use a child process instead of hooking into stderr for the test
* Add a doc comment for `MakeStderrWriter`
* Move the initialization into the `MakeStderrWriter`'s constructor
* Add an extra test case to trigger the logger's emergency flush mechanism
* Use the buffer's mutex for asynchronous flushes
* Remove vestigial `nix` dependency from one of the previous commits