The EVM stack is a hot spot in EVM execution and we end up paying a nim
seq tax in several ways, adding up to ~5% of execution time:
* on initial allocation, all bytes get zeroed - this means we have to
choose between allocating a full stack or just a partial one and then
growing it
* pushing and popping introduce additional zeroing
* reallocations on growth copy + zero - expensive again!
* redundant range checking on every operation reducing inlining etc
Here a custom stack using C memory is instroduced:
* no zeroing on allocation
* full stack allocated on EVM startup -> no reallocation during
execution
* fast push/pop - no zeroing again
* 32-byte alignment - this makes it easier for the compiler to use
vector instructions
* no stack allocated for precompiles (these never use it anyway)
Of course, this change also means we have to manage memory manually -
for the EVM, this turns out to be not too bad because we already manage
database transactions the same way (they have to be freed "manually") so
we can simply latch on to this mechanism.
While we're at it, this PR also skips database lookup for known
precompiles by resolving such addresses earlier.
It is common for many accounts to share the same code - at the database
level, code is stored by hash meaning only one copy exists per unique
program but when loaded in memory, a copy is made for each account.
Further, every time we execute the code, it must be scanned for invalid
jump destinations which slows down EVM exeuction.
Finally, the extcodesize call causes code to be loaded even if only the
size is needed.
This PR improves on all these points by introducing a shared
CodeBytesRef type whose code section is immutable and that can be shared
between accounts. Further, a dedicated `len` API call is added so that
the EXTCODESIZE opcode can operate without polluting the GC and code
cache, for cases where only the size is requested - rocksdb will in this
case cache the code itself in the row cache meaning that lookup of the
code itself remains fast when length is asked for first.
With 16k code entries, there's a 90% hit rate which goes up to 99%
during the 2.3M attack - the cache significantly lowers memory
consumption and execution time not only during this event but across the
board.
* Bump nim-eth, nim-web3, nimbus-eth2
- Replace std.Option with results.Opt
- Fields name changes
* More fixes
* Fix Portal stream async raises and portal testnet Opt usage
* Bump eth + nimbus-eth2 + more fixes related to eth_types changes
* Fix in utp test app and nimbus-eth2 bump
* Fix test_blockchain_json rebase conflict
* Fix EVMC block_timestamp conversion plus commentary
---------
Co-authored-by: kdeme <kim.demey@gmail.com>
* Remove exception from evm memory
* Remove exception from gas meter
* Remove exception from stack
* Remove exception from precompiles
* Remove exception from gas_costs
* Remove exception from op handlers
* Remove exception from op dispatcher
* Remove exception from call_evm
* Remove exception from EVM
* Fix tools and tests
* Remove exception from EVMC
* fix evmc
* Fix evmc
* Remove remnants of async evm stuff
* Remove superflous error handling
* Proc to func
* Fix errors detected by CI
* Fix EVM op call stack usage
* REmove exception handling from getVmState
* Better error message instead of just doAssert
* Remove unused validation
* Remove superflous catchRaise
* Use results.expect instead of unsafeValue
Also embed evmc_status_code to computation.error, and make
the tracer produce cleaner output. No more "Revert opcode executed"
error message. We can distinguish error code between REVERT
and FAILURE in a more cleaner way.
* Removed some Windows specific unit test annoyances
details:
+ Short put()/get() cycles on persistent database have a race condition
with vendor rocksdb. On a specific (and slow) qemu/win7 a 50ms `sleep()`
in between will mostly do the job (i.e. unless heavy CPU load.) This
issue was not observed on github/ci.
+ Removed annoyances when qemu/Win7 keeps the rocksdb database files
locked even after closing the db. The problem is solved by strictly
using fresh names for each test. No assumption made to be able to
properly clean up. This issue was not observed on github/ci.
* Silence some compiler gossip -- part 7, misc/non(sync or graphql)
details:
Adding some missing exception annotation
This patch adds:
- Load and use a third-party EVM in a shared library, instead of Nimbus EVM.
- New option `--evm` to specify which library to load.
- The library and this loader conforms to the [EVMC]
(https://evmc.ethereum.org/) 9.x specification.
Any third-party EVM which is compatible with EVMC version 9.x and supports EVM1
contract code will be accepted. The operating system's shared library format
applies. These are `.so*` files on Linux, `.dll` files on Windows and `.dylib`
files on Mac.
The alternative EVM can be selected in two ways:
- Nimbus command line option `--evm:<path>`.
- Environment variable `NIMBUS_EVM=<path>`.
The reason for an environment variable is this allows all the test programs to
run with a third-party EVM as well. Some don't parse command line options.
There are some limitations to be aware of:
- The third-party EVM must use EVMC version 9.x, no other major version.
EVMC 9.x supports EIP-1559 / London fork and older transactions.
- Nested `*CALL` and `CREATE*` operations don't use the third-party EVM yet.
These call the built-in Nimbus EVM. This mixing of different EVMs between
levels is explicitly allowed in specs, so there is no problem doing it.
- The third-party EVM doesn't need to support precompiles, because those are
nested calls, which use the built-in Nimbus EVM.
- Third-party EVMs execute contracts correctly, but fail the final `rootHash`
match. The reason is that some account state changes, which are correct, are
currently inside the Nimbus EVM and need to be moved to EVMC host logic.
*This is a known work in progress*. The EVM execution itself is fine.
Test results using "evmone" third-party EVM:
- [evmone](https://github.com/ethereum/evmone) has been tested. Only on
Linux but it "should" work on Windows and Mac equally well.
- [Version 0.8.1](https://github.com/ethereum/evmone/releases/tag/v0.8.1) was
used because it is compatible with EVMC 9.x, which is required for the
EIP-1559 / London fork, which Nimbus supports. Version 0.8.0 could be used
but it looks like an important bug was fixed in 0.8.1.
- evmone runs fine and the trace output looks good. The calls and arguments
are the same as the built-in Nimbus EVM for tests that have been checked
manually, except evmone skips some calls that can be safely skipped.
- The final `rootHash` is incorrect, due to the *work in progress* mentioned
above which is not part of the evmone execution. Due to this, it's possible
to try evmone and verify expected behaviours, which also validates our own
EVMC implementation, but it can't be used as a full substitute yet.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Perform byte-endian conversion for 256-bit numeric values, but not 256-bit
hashes. These conversions are necessary for EVMC binary compatibility.
In new EVMC, all host-side conversions are explicit, calling `flip256`.
These conversions are performed in the EVMC "glue" code, which deals with the
binary interface, so the host services aren't aware of conversions.
We intend to skip these conversions when Nimbus host calls Nimbus EVM, even
when it's a shared library, using a negotiated EVMC extension. But for now
we're focused on correctness and cross-validation with third party EVMs.
The overhead of endian conversion is not too high because most EVMC host calls
access the database anyway. `getTxContext` does not, so the conversions from
that are cached here. Also, well-optimised EVMs don't call it often.
It is arguable whether endian conversion should occur for storage slots (`key`).
In favour of no conversion: Slot keys are 32-byte blobs, and this is clear in
the EVMC definition where slot keys are `evmc_bytes32` (not `evmc_uint256be`),
meaning treating as a number is _not_ expected by EVMC. Although they are
often small numbers, sometimes they are a hash from the contract code plus a
number. Slot keys are hashed on the host side with Keccak256 before any
database calls, so the host side does not look at them numerically.
In favour of conversion: They are often small numbers and it is helpful to log
them as such, rather than a long string of zero digits with 1-2 non-zero. The
representation in JSON has leading zeros removed, like a number rather than a
32-byte blob. There is also an interesting space optimisation when the keys
are used unhashed in storage.
Nimbus currently treats slot keys on the host side as numbers, and the tests
pass when endian conversion is done. So to remain consistent with other parts
of Nimbus we convert slot keys.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Prior to this patch, top-level EVM executions and nested EVM executions did
their `getStorage` and other requests using a completely different set of host
functions. It was just unfinished, to get top-level "new" EVMC working.
This finishes the job - it stops using the old methods. Effect:
- Functionality added at the EVMC host level will be used by all EVM calls.
(The target here is Beam Sync).
- The old set of functions are no longer used, so they can be removed.
- When EVMC host call tracing is enabled (`showTxCalls = true`), it traces
the calls from nested EVM executions as well as top-level.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
We've been filling a "vtable"-like at run time, but it's not necessary.
The new object is a global `let x = evmc_host_interface(...)`, we assume it's
initialised before the first use, and we take its address with `.unsafeAddr`.
(If we use `ref evmc_host_interface`, Nim decides (correctly) that the
functions which use it aren't GC-safe because it's a global.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This combines two things, a C stack usage change with EVM nested calls
via EVMC, and changes to host call tracing.
Feature-wise, the tracing is improved:
- Storage keys and values are make more sense.
- The message/result/context objects are shown with all relevant fields.
- `call` trace is split into entry/exit, so these can be shown around the
called contract's operations, instead of only showing the `call` parameters
after the nested call is finished.
- Nested calls are indented, which helps to highlight the flow.
- C stack usage considerably reduced in nested calls when more functionality
is enabled (either tracing here, or other things to come).
This will seem like a minor patch, but C stack usage was the real motivation,
after plenty of time in the debugger.
Nobody cares about stack when `showTxCalls` (you can just use a big stack when
debugging). But these subtle changes around the `call` path were found to be
necessary for passing all tests when the EVMC nested call code is completed,
and that's a prerequisite for many things: async EVM, dynamic EVM, Beam Sync,
and to fix https://github.com/status-im/nimbus-eth1/issues/345.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The update for London (EIP-1559) in 1cdb30df ("bump nim-emvc with evmc revision
8.0.0 to 9.0.0") really bumped EVMC ABI version from 7.5 up to 9.
In other words, it skipped Berlin, going direct from Istanbul to London.
That was accompanied by EVMC changes in 05e9b891 ("EIP-3198: add baseFee op
code in nim-evm"), which added the API changes needed for London.
But the missing Berlin functions weren't added in the move to London.
As a result, our EVMC host became incompatible with Berlin, London, and really
all revisions of the ABI, and if a third party EVM was loaded, it crashed.
This commit adds the missing Berlin host support, and makes our ABI
binary-compatible with real EVMC again.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Show calls from the host into the EVM. Shows the call, `evmc_message` fields,
and `evmc_result` fields when the call returns.
(When `show_tx_calls` is manually set to true.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
1. Send all EVM executions through the EVMC `execute` function.
It leads to the same place in the end as calling `Computation` before, but
`execute` is the API function used by all EVMC implementations, and it is
very explicit what data is passed back and forth.
2. As a consequence this starts using the new `host_services` code from EVM, so
this is a significant change to the paths used for account state processing.
3. Because we will have to remove the `newComputation` call on the host side,
anticipating that the contract code is now saved in `host` instead of being
copied around. As it's saved in `host`, there is no need to pass it
separately to `evmcExecComputation`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Even though `evmc_create_nimbus_evm` is called, it fails at link time because
the definition of that function isn't included unless it is pulled in
explicitly.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
1. This provides the necessary type adjustments for host services to be
(optionally) callable via the EVMC binary-compatible interface. This layer
is stashed away in a glue module so the host services continue to use
appropriate Nim types, and are still callable directly.
Inlining is used to ensure there should be no real overhead, including stack
frame size for the `call` function. Note, `import` must be used for
`{.inline.}` to work effectively.
2. This also provides a key call in the other direction, the version of host to
EVM `execute` that is called on the host side.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Jacek Sieka
andri lim
Jordan Hrycaj
Jamie Lokier