Transaction and BlockHeader already updated in nim-eth repo
to support EIP-1559
EIP-1559 header validation and gasLimit validation
already implemented in previous commit
This commit deals with block validation:
- Effective gasPrice per EIP-1559
- new miner reward based on priorityFee
Previously max gas refunded was defined as gas_used div 2.
Here we name the constant 2 as MAX_REFUND_QUOTIENT and
change its value to 5.
The new equation will be: gas_used div MAX_REFUND_QUOTIENT
SSTORE_CLEARS_SCHEDULE or FeeSchedule[RefundsClear] in evm
have initial value of 15_000 when introduced by EIP-2200.
EIP-2200 also set new value for SSTORE_RESET_GAS
from 5000 to to 5000 - COLD_SLOAD_COST
Now with EIP-3529, SSTORE_CLEARS_SCHEDULE beecome
SSTORE_RESET_GAS + ACCESS_LIST_STORAGE_KEY_COST
or 5000 - COLD_SLOAD_COST + ACCESS_LIST_STORAGE_KEY_COST
of 5000 - 2100 + 1900 = 4800
This preparation is needed for subsequent
EIPs included in London.
- Add London to Fork enum
- Block number to fork
- Parsing London fork in chain config
- Prepare gas costs table for London
- Prepare EVM opcode dispatcher for London
- Block rewards for London
- Prepare hive script for London
* continue importing rlp blocks
why:
a chain of blocks to be imported might have legit blocks
after rejected blocks
details:
import loop only stops if the import list is exhausted or if there
was a decoding error. this adds another four to the count of successful
no-hive tests.
* verify DAO marked extra data field in block header
why:
was ignored, scores another two no-hive tests
* verify minimum required difficulty in header validator
why:
two more nohive tests to succeed
details:
* subsumed extended header tests under validateKinship() and renamed it
more appropriately validateHeaderAndKinship()
* enhanced readability of p2p/chain.nim
* cleaned up test_blockchain_json.nim
* verify positive gasUsed unless no transactions
why:
solves another to nohive tests
details:
straightened test_blockchain_json chech so there is no unconditional
rejection anymore (based on the input test scenario)
* Re-adjust canonical head to parent of block to be inserted
why:
of the failing tests that remain to be solved, 30 of those will succeed
if the canonical database chain head is cleverly adjusted -- yes, it
looks like a hack, indeed.
details:
at the moment, this hack works for the non-hive tests only and is
triggered by a boolean argument passed on to the chain.persistBlocks()
method.
* Use parent instead of canonical head for block to be inserted
why:
side chains need to be inserted typically somewhere before the
canonical head.
details:
the previous _hack_ was unnecessary and removed, it was inspired by
some verification in persistBlocks() which explicitly referenced the
canonical head (which now might or might not refer to the newly inserted
header.)
* remove unnecessary code + comment
why:
some handy features were intended to support the unit test from
the clique/clique_test.go source (the other one is from
clique/snapshot_test.go.)
as this test cannot realistically be implemented without the full
api (includes mining support), it is left as that
Proper nested call functionality is being skipped in this iteration of new EVMC
host code to keep it simpler, to allow testing and architecture to be built
around the less complicated non-nested cases first.
Instead, nested calls use the old `Computation` path, and bypass any
third-party EVM that may be loaded. The results are the same, and mixing
different EVMs in this way is actually permitted in the EVMC specification.
This approach also means third-party EVMs we test don't need to support
precompiles and we don't need to specially handle those cases.
(E.g. "evmone" doesn't support precompiles, just EVM1 opcodes).
(These before/after scope actions are approximately copy-pasted from
`nimbus/vm/evmc_host.nim`, making their detailed behaviour "obviously correct".
Of course they are subject to tests as well. The small stack property of
a3c8a5c3 "EVMC: Small stacks when using EVMC, closes#575 (segfaults)" is
carefully retained.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Make the host service `setStorage` calculate the gas refund itself, instead of
depending on EVM internals.
In EVMC, host `setStorage` is responsible for adding the gas refund using the
rules of EIP-1283 (Constantinople), EIP-2200 (Istanbul) or EIP-2929 (Berlin).
It is not obvious that the host must do it from EVMC documentation, but it's
how it has to be. Note, this is very different from the gas _cost_, which the
host does not calculate.
Gas refund doesn't affect computation. It is applied after the whole
transaction is complete, so it can be tracked on the host or EVM side. But
`setStorage` doesn't return enough information for the EVM to calculate the
refund, so the host must do it when `setStorage` is used.
For now, this continues using Nimbus `Computation` just to hold the gas refund,
to fit around existing structures and get new EVMC working. But the host can't
keep using `Computation`, so gas refund will be moved out of it in due course.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
When processing log operations on the EVMC host side, it causes incorrect
`rootHash` results in some tests. This patch fixes the results.
The cause of these results is known: `Computation` is still doing parts of
contract scope entry/exit which need to be moved to the host. For now, as a
temporary workaround, update logs in `Computation` as it did before.
This makes test pass when using Nimbus EVM. (It breaks third-party EVMs when
`LOG*` ops are used, although most other tests pass.)
We can't keep this as it prevents complete host/EVM separation, but it's useful
in the current code, and it's fine to develop other functionality on top.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
When processing self destructs on the EVMC host side, it causes incorrect
`rootHash` results in some tests. This patch fixes the results.
The cause of these results is known: `Computation` is still doing parts of
contract scope entry/exit which need to be moved to the host. For now, as a
temporary workaround, update self destructs in `Computation` as it did before.
This makes test pass when using Nimbus EVM. (It breaks third-party EVMs when
`SELFDESTRUCT` ops are used, although most other tests pass.)
We can't keep this as it prevents complete host/EVM separation, but it's useful
in the current code, and it's fine to develop other functionality on top.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
In the unusual case where log data is zero-length, `data[0].addr` is invalid
and Nim thoughtfully raises `IndexOutOfBounds`, a `Defect` so it's not even
in `CatchableError`.
This is done in the EVMC host services to handle `LOG*` ops, and it made one
of the EVM tests silently fail with no error message. The fix is obvious.
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>
When `show_tx_calls` is manually set to true, show all the calls from the EVM
to the host, including name, arguments and results.
For example this shows each call to `setStorage`, the key, value and storage
result. This output allows the externally-visible activity of an EVM to be
seen, and it's been useful for guessing what went wrong when a test fails.
In theory, if two EVMs show the same activity in this log, they should have the
same effect on account states, gas, etc. and the same final `roothash`
(which is the only value some tests check).
ps. Ideally we'd use `{.push show.}`...`{.pop.}`, just like with `inline`.
But we can't: https://github.com/nim-lang/Nim/issues/12867
Signed-off-by: Jamie Lokier <jamie@shareable.org>
New pragma `{.show.}` on a proc definition turns on tracing for that proc.
Every call to it shows the name, arguments and results, if `show_tx_calls` is
manually set to true. This is to trace calls from EVM to host.
This started as a template which took a block expression, but the closure it
used led to illegal capture errors. It was easier to write a macro.
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>
We'll re-enable endian conversions based on a negotiated run-time option later,
but for now let's remove one complication to testing the new EVMC paths, and
also gain a little performance.
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>
This provides the functions a loadable VM must provide for a host to use it.
The main access to a loaded EVM is `evmc_create_nimbus_evm`, and this meant to
be the only exported function the caller starts with.
That provides access to other functions, also defined in this patch, to
configure the EVM and then the key interesting function is `execute`.
`execute` runs a full computation, here using Nimbus EVM `Computation`.
(Note, even though everything is EVMC binary-compatible, there is a small
dependency on `TransactionHost` in `execute` here, which prevents this being
used by a host that is not Nimbus at the moment. It is necessary for some
tests, and will eventually go away.)
Although this provides the VM-side functionality needed by the host, it does
not contain the glue functions for `Computation` to call the host, which are
already part of the Nimbus EVM in `nimbus/vm/evmc_api.nim`.
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>
This provides "host services", functions provided by the application to an EVM.
They are a key part of EVMC compatibility, but we will switch to using these
with "native" EVM as well.
These are functions like `getStorage`, `setStorage` and `emitLog` for accessing
the account state, because the EVM is not allowed direct access to the database.
This code is adapted from `nimbus/vm/evmc_host.nim` and other places, but there
is more emphasis on being host-side only, no dependency on the EVM or
`Computation` type. It uses `TransactionHost` and types in `host_types`.
These host services have two goals: To be compatible with EVMC, and to be a
good way for the Nimbus EVM to access the data it needs. In our new Nimbus
internal architecture, the EVM will only access the databases and other
application state via these host service functions.
The reason for containing the EVM like this, even "native" EVM, is that having
one good interface to the data makes it a lot easier to change how the database
works, which is on the roadmap.
These functions almost have EVMC signatures, but they are not binary compatible
with EVMC. (Binary compatibility is provided by another module). It would be
fine for Nimbus EVM to call these functions directly when linked directly.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Block validation failure isn't an error, it's correct rejection of a bad block
from the network. All conditions that lead to block rejection return a simple
boolean.
When a block is rejected, most reasons log at `debug` level. Only `stateRoot`
mismatch shouts a loud, highlighted, multi-line error message with big red
`error` alert.
Historically this was to assist EVM development, because it was more likely to
be a Nimbus EVM bug than a real bad block. But now the EVM is in good shape,
has a large and thorough testsuite, and `stateRoot` mismatch is more likely to
be a real bad block that should be rejected with less fuss.
If there's a genuine EVM bug, we'll still get an alert: Consensus failure will
quickly become obvious, and the block where it happens is easily fetched.
So a big, loud error is no longer useful, and it became a problem during tests.
Recently a few hundred tests were added that trigger it, and now successful
test output is filled with attention-grabbing errors which aren't really errors
or particularly useful.
Since it's not really an error, the original motivation is now backwards, and
other reasons warn at `debug` level, make this like the others.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The common forks list was already used, redirected via `vm_forks` for
historical compatibility. Remove the old `vm_forks` now and divert all imports
to the common forks list outside the EVM.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
File `vm_types2` is obsolete. Remove this file and divert all imports to the
common forks list outside the EVM, or in some cases they don't need it anyway.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Fork names were originally capitalized, and were made lower case by @narimiran
in commit 36a7519 to satisfy `parseEnum` in some tests. Restore the
capitalization and make the tests work with it.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Many places outside the EVM use `Fork` and the fork list, and in general we
want progressively fewer dependencies on EVM internal types and files.
This may prove to be a temporary location, especially when we implement
issue #640. But it's a fine temporary location if so.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The current EVM generates its own new contract addresses, and this is why there
are separate `msg.contractAddress` and `msg.codeAddress` fields in the
computation start message.
In EVMC, account updates are only allowed on the host side, including contract
generation, and the start message has one destination field, `msg.destination`.
The EVM cannot select addresses, only use them. It's a sensible design.
The difference makes the current EVM incompatible with EVMC and its message
format, so this patch corrects the difference. It moves contract address
generation to the host side. This simplifies the EVM and its API a little.
(As an API change, this is incompatible with vm2, so it's guarded under
`evmc_enabled` to allow vm2 to continue to build and run at this time. This is
also why there are fewer deletions than would otherwise be expected.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The rationale in EIP-6[1] for changing names to `selfDestruct` applies to code
as much as it does to specs. Also, Ethereum uses the new names consistently,
so it's useful for our code to match the terms used in later EIP specs and
testsuite entries.
This change is straightforward, and is a prerequisite for patches to come that
do things with the `selfDestruct` fields.
[1] https://eips.ethereum.org/EIPS/eip-6
Hudson Jameson, "EIP-6: Renaming SUICIDE opcode," Ethereum Improvement
Proposals, no. 6, November 2015.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This is the main patch which removes Whisper code from `nimbus-eth1` code.
It removes all configuration, help, startup, JSON-RPC calls and most types.
Note, there is still Whisper functionality in `nim-eth`. Also, the "wrapper"
under `wrappers/` isn't dealt with by this change, but it's not built by
default (and might not currently work).
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This is the patch which removes Whisper functionality from `nimbus-eth1`,
even though the code has yet to be removed after.
After this change, enabling Whisper has no effect. Its configuration is
ignored and it won't be started.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This commit turns Whisper off by default, without changing anything else.
So this can be cherry-picked if you just want to disable Whisper without
removing it.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
There's no need for macro `toSymbolName` to convert fork enum values to their
presentation texts (logging etc) then re-parse them back to a fork enum value.
`asFork` is already used in the same function and works without these steps,
so use it consistently.
Same applies to `op.toSymbolName` and `asOp`.
This makes the code simpler, and removes a text pattern-matching requirement.
The patch has been checked to confirm it doesn't change the compiled code.
Motivation: The forks list will be removed from VM because it is used outside
the VM as well. Doing so highlighted vm2's `toSymbolName`. It's not needed,
and it's best if the VM doesn't constrain text strings used outside the VM
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The last caller of `setupComputation` is gone, now that it's been replaced by
the single entry point for all EVM calls, `runComputation`.
With this removal, EVM's `Computation` type should no longer be used anywhere
outside the call module (except in some tests and the EVM itself).
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Simplify transaction validations to use `runComputation`; drop other code.
Getting everything right up to this point to pass all the tests was trickier
than it looks.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Simplify how JSON fixtures tests are run to use `runComputation`.
Drop other code.
These use the `noTransfer` option, which is similar enough to calling
`c.executeOpcodes()` instead of `c.execComputation()`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Add another flag to disable a processing step when a call doesn't come from
a real transaction:
- `noTransfer`: Don't update balances, nonces, code.
This is to support VM fixtures tests which require account balances and nonces
to be unchanged when running the account's code.
These tests call `c.executeOpcodes()`, an internal function of the EVM, instead
of the usual `c.execComputation()`. It goes direct to the bytecode dispatcher,
skipping parts of `Computation` that are normally called.
But we can't keep calling `c.executeOpcodes()` and have a single entry point to
the VM, let alone an EVMC entry point.
`noTransfer` provides similar enough behaviour to calling `c.executeOpcodes()`
that these tests can use the new single entry point like everything else.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Simplify `estimateGas` to use `runComputation`; drop other code.
The RPC/GraphQL `estimateGas` operation is quite different from the `call`
operation. It is much more like ordinary transaction execution than `call`,
though there are still enough differences that tx validation cannot be used.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Simplify `call` to use `runComputation`; drop other code.
The RPC/GraphQL `call` operation differs in many ways from regular transaction
calls. The following flags are set, to disable various steps in processing.
All four are true (disabling the corresponding step) for `call`:
- `noIntrinsic`: Don't charge intrinsic gas.
- `noAccessList`: Don't initialise EIP2929 access list.
- `noGasCharge`: Don't charge sender account for gas.
- `noRefund`: Don't apply gas refund/burn rule.
Interestingly, the RPC/GraphQL `estimateGas` operation doesn't behave so
differently from regular transactions. It might be that not all these steps
should be disabled for `call` either. But until we investigate what
RPC/GraphQL clients are expecting, keep the same behaviour as before.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The following four flags are added, to change various steps in EVM processing
when a call doesn't come from a real transaction:
- `noIntrinsic`: Don't charge intrinsic gas.
- `noAccessList`: Don't initialise EIP2929 access list.
- `noGasCharge`: Don't charge sender account for gas.
- `noRefund`: Don't apply gas refund/burn rule.
This is to support RPC and GraphQL `call` operations, which behave differently
in some ways from regular transaction calls, and to support some test suites.
In EVMC terms, all these alterations can be performed on the host side.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Calculate extra intrinsic gas for an EIP-2930 transaction with access list.
While we're here, do the rest of the intrinsic gas calculation. Make it clear,
explicit and in one place. (Previous code delegated parts of the calculation
to `transaction.nim` but had to do the rest locally due to mismatched types.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
New entry point `runComputation`, for all EVM calls.
(Later the intent is `runComputationAsync`.)
As noted in commit 297d789, there are six entry points calling EVM computation,
with different parameters and expecting different behaviours. Parameters were
dealt with in `setupComputation`. Behaviours are unified in `runComputation`,
with options passed via `CallParams`.
This code performs the steps used when validating a transaction. Options for
non-standard behaviour for RPC, GraphQL and tests to be added as required.
This replaces `setupComputation`, `execComputation` and `executeOpcodes`
(other than its own calls). As a result `Computation` and other EVM types are
no longer referenced in the main program, and many imports can be dropped.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Change fixtures tests to use shared `setupComputation` instead of
their own slightly different variant.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Change RPC/GraphQL calls to the EVM to use shared `setupComputation`
instead of their own special variant.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
There are currently six entry points to running an EVM computation, all with
slightly different parameters, and expecting slightly different EVM behaviours.
First step in merging them is a common `setupComputation` that replaces all
the different `*...SetupComputation` functions.
This uses the `TransactionHost` type because it's a step towards using that
type for all EVM calls using only EVMC. For now an EVMC message is created
then translated to EVM-internal `Message`. It is done this way to build up
the new interface in stages where all tests pass at each stage.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
`TransactionHost` represents the interface to the EVM from the application once
we've fully transitioned to EVMC. It represents a managed EVM call, and the
"EVMC host" side of the host<->EVM boundary.
This holds transaction state which sits outside the EVM, but a pointer to this
is passed around by the EVM as _opaque_ type `evmc_host_context`.
To the EVM, this offers "host services", which manage account state that the
EVM cannot do, such as balance transfers, call costs, storage, logs, gas
refunds, nested calls and new contract addresses. The EVM has no direct access
to the account state database or network; it's all via "host services".
To the application (host side), this object represents a managed EVM call, with
inputs, a method to run, outputs, hidden transaction state, and later async
scheduling. It is to replace `Computation` on the application side, while
`Computation` will remain but just be for the EVM side.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This fixes a bug spotted by @mjfh that was introduced by commit 2a7ccceb:
try:
if not c.continuation.isNil:
(c.continuation)()
c.continuation = nil
c.selectVM(fork)
except CatchableError as e:
...
The call to `(c.continuation)()` was moved by 2a7ccceb inside the `try` so
that, like all the Op functions do already, if the continuation raises, the
interpreter's general catch turns the exception into a an error status result.
But if the continuation raises an exception, `continuation` is not cleared in
the next line, and at the next resumption the continuation is called again.
It may loop doing this.
This doesn't currently happen because the continuations don't really raise, but
it's still a correctness issue.
This fix also allows a continuation to spawn a second continuation, if it
encounters a second suspension point. This also doesn't happen currently,
but the pattern will become useful with async EVM.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
There is no valid `CREATE_CONTRACT_ADDRESS`. Some places on the internet say
account zero means contract creation, but that's not correct.
Transactions to `ZERO_ADDRESS` are legitimate transfers to that account, not
contract creations. They are used to "burn" Eth. People also send Eth to
address zero by accident, unrecoverably, due to poor user interface issues.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Each place in `call_evm` that sets up an EVM call calculates the new contract
address for contract creations. But it's redundant, because `newComputation`
ignores the provided value and does the calculation again.
Remove the unused address calculation.
This is also a step to merging different entry points and EVMC. This change
ends up with the same value in both `msg.contractAddress` and `msg.codeAddress`
for every entry point, and this is good because it matches the EVMC message
structure, where they are replaced by only one value called `msg.destination`
Signed-off-by: Jamie Lokier <jamie@shareable.org>
`c.executeOpcodes` is called by some JSON fixture tests. These tests bypass
some of the setup and return, and because of this call, continuations aren't
processed either. Opcodes that use continuations will behave incorrectly.
The opcodes used in these particular tests don't use continuations currently,
so just add some assertions to verify this remains the case.
This is only used by local tests, and the call to `c.executeOpcodes` will be
replaced by the common entry point (that handles things like this correctly in
all cases) so we don't need to spend more time on this.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
why:
previously, table data were stored with the table iterator. while
loading a table with permuted entries will always reconstruct equivalent
tables (in the sense of `==`), serialisation data are not comparable.
this patch produces always the same serialised data for equivalent
tables.
why:
source-local unit tests would hardly be triggered by github CI as rightly
criticised with the last patch.
details:
source-local unit tests have been moved to tests folder.
this version also contains rlp serialisation code so rlp encode/decode
will apply tranparently. this is not needed in p2p/validate but will be
useful with the clique protocol.
why:
to be used in Clique consensus protocol which suggests 4k cache entries.
the previous implementation used OrderTable[] which has complexity O(n)
for deleting entries.
after EIP2718/EIP2930, we have additional fields:
type AccessTuple {
address: Address!
storageKeys : [Bytes32!]
}
type Transaction {
r: BigInt!
s: BigInt!
v: BigInt!
# Envelope transaction support
type: Int
accessList: [AccessTuple!]
}
close#606
instead of using stdlib/json, now we switch to json_serialization
the result is much tidier code and more robust when parsing
optional fields.
fixes#635
although this is not part of EIP 1767
but the hive test cases derived from besu
test cases contains this.
we add this now to pass more test hive.graphql cases
Move the EVM setup and call in precompile tests to `fixtureCallEvm` in
`call_evm`. Extra return values needed for testing are returned specially, and
the convention for reporting gas used is changed to match `asmCallEvm`.
Although the precompile tests used `execPrecompiles` before, `executeOpcodes`
does perfectly well as a substitute, allowing `fixtureCallEvm` to be shared.
_Significantly, this patch also makes `Computation` more or less an internal
type of the EVM now._
Nothing outside the EVM (except `call_evm`) needs access any more to
`Computation`, `execComputation`, `executeOpcodes` or `execPrecompiles`.
Many imports can be trimmed, some files removed, and EVMC is much closer.
(As a bonus, the functions in `call_evm` reveal what capabilities parts of the
program have needed over time, makes certain bugs and inconsistencies clearer,
and suggests how to refactor into a more useful shared entry point.)
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Allow the fork to be specified consistently through an `option[Fork]` instead
of varying inconsistencies depending on which call. When fork is not
specified, the `BaseVMState` code picks the correct fork by default for the
block number and chain.
This change actually deletes code, because a number of functions (RPC etc) had
redundant code to pick the fork, which always resolved to same as default.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Move the EVM setup and call in fixtures "vm json tests" to new function
`fixtureCallEvm` in `call_evm`. Extra return values needed for testing are
returned specially.
This entry point is different from all other `..CallEvm` type functions,
because it uses `executeOpcodes` instead of `execComputation`, so it doesn't
update the account balance or nonce on entry and exit from the EVM.
The new code is a bit redundant and simplistic intentionally, as the purpose is
to move functionality to `call_evm` with high confidence nothing really
changed. The calls will be jointly refactored afterwards to merge differences.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
In the `text_vm_json` ("fixtures") test code, there is another variant of
`rpcSetupComputation` and `txSetupComputation` with slightly different
paremeters. The similarity is obvious.
It is a special setup for testing, though, as it requires slightly different
parameters.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
usually, there is always a sender around `getRecipient` call.
no need to recalculate sender. and more important, in some of
JSON-RPC/GraphQL call, the sender is come from `rpcCallData`,
not from `tx.getSender`. or in ohter situation when the tx is
an unsigned tx, without `r,s,v` fields to calculate sender.
Move the EVM setup and call in `macro_assembler` (`runVM`) entirely to new
function `asmCallEvm` in `call_evm`. Extra return values needed for
testing are returned specially from `asmCallEvm`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The second `asmSetupComputation looks up state by block number and preceding
block number, modifies the first transaction with code for testing, and uses
some parts of that transaction to setup an an EVM test.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
In the `macro_assembler` test code, `initComputation` is another variant of
`rpcSetupComputation` and `txSetupComputation` with slightly different
paremeters. The similarity is obvious.
It is a special setup for testing, though, as it requires a contract-creation
transaction for parameters, but sets up a `CALL` execution not `CREATE`.
Gather this into `call_evm`: `initComputation` -> `asmSetupComputation`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The point of the `call_vm` exercise is to allow `Computation` to become an
internal type of the EVM, not used as API by the rest of the program. So
`rpcSetupComputation` should be private. It was left exported by mistake.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Split out and move the EVM setup and call in `processTransaction` to
`call_evm`. This is the last part of the main program which calls the EVM
to be moved. (There's still test code.)
While we're here, move the EIP2929 access list setup too, as the similarity
to `rpcInitialAccessListEIP2929` is obvious.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
There's only one call left to `refundGas(Transaction, ...)`, and the
similarity to the tail of `rpcEstimateGas` is obvious.
Gather this into `call_evm`: `refundGas` -> `txRefundGas`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
After recent changes, there's only one call left to `setupComputation`, and
it's just a variant like `rpcSetupComputation` but for transaction processing.
The similarity to `rpcSetupComputation` is obvious.
Gather this into `call_evm`: `setupComputation` -> `txSetupComputation`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
It's possible for `tx.value` in the transaction to have a deliberately
constructed large 256-bit value, such that adding `gasLimit * gasPrice` to it
overflows to a small value.
Prior to this patch, the code would allow such a transaction to pass
validation, even though such a large transfer cannot be valid.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Change `rpcEstimateGas` to setup and execute a computation directly, in a
similar way to `rpcDoCall` and `rpcMakeCall`, instead of constructing a fake
transaction and then validating it.
This patch does not (or should not) change any behaviour.
Although this looks a bit messy as it duplicates parts of `validateTransaction`
and `processTransaction`, proc names have been used to hopefully keep the
meanings clear, and it's just a stepping stone as those transaction functions
will be changed next. Also the behaviour of RPC `estimateGas` may not be
correct (although this patch is careful not to change it), so it's good to make
it explicit so we can see how it differs from other RPCs.
Doing this change exposed some interesting behaviour differences between the
`call` RPC and `estimateGas` RPC, which may be bugs, or may be intentional.
These differences are now obvious and explicit.
The unclear areas are not well documented by any of the clients, even Infura
which says a bit more than the others. So to find out if they are intended,
we'll have to run tests against other Ethereum services.
Guessing, on the face of it, it looks likely that RPC `call` should:
- Setup EIP2929 access lists
- Account for intrinsic gas (maybe not because zero-gas transactions are ok)
And it looks likely that RPC `estimateGas` should:
- Not return zero when an account has insufficient balance
- Maybe use a different gas cost estimate when one isn't supplied in the RPC
Signed-off-by: Jamie Lokier <jamie@shareable.org>
The RPC `estimateGas` behaves differently from RPC `call` in a number of ways.
These differences may be bugs due to `rpcEstimateGas` calling the EVM in a very
different way than `rpcDoCall`, or they may be intentional. To be sure, we'll
need to test behaviour with Geth, Infura etc to find out (their documentation
isn't enough.) For now, though, we'll keep the same behaviour as we always had.
`rpcEstimateGas` cannot use `rpcSetupComputation` as it is, because
`estimateGas` accounts for "intrinsic gas", and `call` does not.
This patch changes `rpcSetupComputation` to accomodate both behaviours.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
`makeCall` used by GraphQL is another way to setup and call the EVM.
Move it to `transaction/call_evm`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
`estimateGas` used by JSON-RPC is another way to setup and call the EVM,
also used by GraphQL. Move it to `transaction/call_evm`.
This function has too much direct knowledge of details that shouldn't be used
outside transaction handling code, details we need to change when changing the
db and transaction memory layer.
Moving this one exposed quite a bit of abstraction leakage, as it calls
directly to the hexary trie db around `processTransaction`.
It looks like the _intended_ functionality of `estimateGas` is similar to
`rpcDoCall` with the only real difference being to not store the final state.
It looks like the extra stuff in `estimateGas` compared with `doCall` is a
messy workaround for computation not exposing the right API ("don't save final
state") for RPC to use.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
`doCall` used by JSON-RPC is another way to setup and call the EVM.
Move it to `transaction/call_evm`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
Start gathering the functions that call the EVM into one place,
`transaction/call_evm.nim`.
This is first of a series of changes to gather all ways the EVM is called to
one place. Duplicate, slightly different setup functions have accumulated over
time, each with some knowledge of EVM internals. When they are brought
together, these methods will be changed to use a single entry point to the EVM,
allowing the entry point to be refactored, EVMC to be completed, and async
concurrency to be implemented on top. This also simplifies the callers.
First, a helper function used by RPC and GraphQL to make EVM calls without
permanently modifying the account state. `setupComputation` ->
`rpcSetupComputation`.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
the `processArguments` now have overloaded proc, one with opt param and one without.
the OptParser now can be passed to `opt` param.
this is useful in scenario where in test code we need to simulate something
without using real command line arguments.
rather than initialize it to 0, those block numbers
are initialized to high(BlockNumber). this will fix
issue when imported genesis.json doesn't contains all
forks' blockNumber.
The account database code is not supposed to raise exceptions in the EVM, and
the behaviour is not well defined if it does. It isn't compliant with EVMC
spec either. But that will be dealt with properly when the account state-cache
is dealt with, as there is some work to be done on it.
Meanwhile, if it raises in code under `chainTo` and then `(continuation)()`,
the behaviour was changed slightly by the stack-shrink patches.
Before those patches, an exception after the recursion-point was converted to
`c.setError` "Opcode Dispatch Error" in `executeOpcodes. After, it would
propagate out, a different behaviour. (It still correctly walked the chain of
`c.dispose()` calls to clean up.)
It's easy to restore the original behaviour just by moving the continuation
call, so let's do that.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
why:
only two public functions left: executeOpcodes() and execCallOrCreate()
where the former one was originally in interpreter_dispatch.nim and
the latter one calls this one.
improves maintainability
overview:
can be verified by running "make check_vm2 X=0" in the nimbus directory
(be patient when running it.) the X=0 flag is necessary if there is a
native NIM compiler which may bail out at some vendor imports.
details:
when compiling state_transaction.nim, the nim flag vm2_enabled must
be set in order to avoid implicit import of native VM definitions.
why:
kludge not needed anymore for oph_handlers.nim sub-sources and sources
that rely on oph_handlers.nim (but not state_transactions.nim which
relies on computation.nim.)
also:
re-integrated stack_defs.nim back into stack.nim
why:
the v2 prefix of the file name was used as a visual aid when
comparing vm2 against vm sources
why:
the v2 prefix of the file name was used as a visual aid when
comparing vm2 against vm sources
details:
all renamed v2*.nim sources compile locally with the -d:kludge:1 flag
set or without (some work with either)
only sources not renamed yet: v2state_transactions.nim
why:
on 32bit windows 7, there seems to be a 64k memory ceiling for the gcc
compiler which was exceeded on some test platform.
details:
compiling VM2 for low memory C compiler can be triggered with
"make ENABLE_VM2LOWMEM". this comes with a ~24% longer execution time
of the test suite against old VM and optimised VM2.
why:
the new implementation lost more then 25% execution time on the test
suite when compared to the original VM. so the handler call and the
surrounding statements have been wrapped in a big case statement similar
to the original VM implementation. on Linux/x64, the execution time of
the new VM2 seems to be on par with the old VM.
details:
on Linux/x64, computed goto works and is activated with the -d:release
flag. here the execution time of the new VM2 was tested short of 0.02%
better than the old VM. without the computed goto, it is short of
0.4% slower than the old VM.
why:
using function stubs made it possible to check the syntax of an op
handler source file by compiling this very file. this was previously
impossible due cyclic import/include mechanism.
details:
only oph_call.nim, oph_create.nim and subsequently op_handlers.nim
still need the -d:kludge:1 flag for syntax check compiling. this flag
also works with interpreter_dispatch.nim which imports op_handlers.nim.
why:
step towards breaking circular dependency
details:
some functions from v2computation.nim have been extracted into
compu_helper.nim which does not explicitly back-import
v2computation.nim. all non recursive op handlers now import this source
file rather than v2computation.nim.
recursive call/create op handler still need to import v2computation.nim.
the executeOpcodes() function from interpreter_dispatch.nim has been
moved to v2computation.nim which allows for <import> rather than
<include> the interpreter_dispatch.nim source.
why:
this allows for passing back information which can eventually be
used for reducing use of exceptions
caveat:
call/create currently needs to un-capture the call-by-reference
(wrapper) argument using the Computation reference inside
why:
the previous approach was replacing the function-lets in
opcode_impl.nim by the particulate table handlers. the test
functions will verify the the handler functions are sort of
correct but not the assignments in the fork tables.
the handler names of old and new for tables are checked here.
caveat:
verifying tables currently takes a while at compile time.
details:
the op handler table is accessible via op_handlers.nim module
op handler function implementations are found in the op_handlers/
sub-directory
kludge:
for development and pre-testing, any new module can be individually
compiled setting the kludge flag using -d:kludge:1. this causes some
proc/func replacements in turn allowing for omitting imports that would
otherwise cause a circular dependency. otherwise individual compilation
would fail.
in order to prove the overall correctness of the code, the
op_handlers.nim is imported by opcodes_impl.nim when compiling all,
nimbus or test.
why:
subsequent development will compile sources as main without setting
the vm2_enabled flag. also, the doc generator would fail an vm2 without
setting the flag for the vm2 files.
why:
generally, there is no role for libbacktrace when docs are generated
for vm2, undo settings of config.nim and provide the "kludge" flag, so
circular import/include dependencies can be taken care of (not only)
for generating docs
why:
new name forks_list.nim file name matches additional documentation
file names.
details:
v2forks.nim remains a hollowed out shell serving as interface file.
why:
activate NIM comments needed re-write. as there is no advantage in using
the macro replacing a few missing op-codes by "Nop##" name symbols, the
macro wrapper has been removed.
details:
when explicitly accessed by numeric value ##, missing Op enum entries
result in a symbol name something like "Op ##".
rather than implicitly using a macro to fix the op-codes list, missing
entries are detected at compile time when a fatal exception is thrown.
the static compile time check verifies that
all op-codes 0 .. 255 are defined
op code name/mnemonic has at least 2 chars and starts with a capital
op code name/mnemonic is not NIM auto-generated (i.e. has a space)
also, original '#' comments are exposed as doc comments '##'
Capitalisation:
- The option is lower case `--logmetrics` but help said `--logMetrics`
- Same for `--logmetricsiterval`
- Same for `--metricsserver` and `--metricsserverport`
Ethereum network selection:
- Moved out into their own, cleaner help section
- Added help for `--mainnet`, `--goerli` and `--kovan`
- Moved `--networkid` and `--customnetwork` to this section as well
Other:
- Reworded or formatted some help lines for clarity and consistency
Changed options:
- Renamed `--metricserver` to `--metrics`
- Renamed `--matricsserverport` to `--metricsport`
- Removed Morden network; this didn't have an option, but could be
selected with `--networkid:2` and then fail to work
Signed-off-by: Jamie Lokier <jamie@shareable.org>
This patch reduces stack space used with EVM in ENABLE_EVMC=1 mode, from 13 MB
worst case to 550 kB, a 24x reduction.
This completes fixing the "stack problem" and closes#575 (`EVM: Different
segmentation faults when running the test suite with EVMC`).
It also closes#256 (`recursive EVM call trigger unrecoverable stack overflow`).
After this patch, it is possible to re-enable the CI targets which had to be
disabled due to #575.
This change is also a required precursor for switching over to "nearly EVMC" as
the clean and focused Nimbus-internal API between EVM and sync/database
processes, and is also key to the use of Chronos `async` in those processes
when calling the EVM.
(The motivation is the internal interface has to be substantially changed
_anyway_ for the parallel sync and database processes, and EVMC turns out to be
well-designed and well-suited for this. It provides good separation between
modules, and suits our needs better than our other current interface. Might as
well use a good one designed by someone else. EVMC is 98% done in Nimbus
thanks to great work done before by @jangko, and we can use Nimbus-specific
extensions where we need flexibility, including for performance. Being aligned
with the ecosystem is a useful bonus feature.)
All tests below were run on Ubuntu 20.04 LTS server, x86-64. This matches one
of the targets that has been disabled for a while in CI in EVMC mode due to
stack overflow crashing the tests, so it's a good choice.
Measurements before
===================
Testing commit `e76e0144 2021-04-22 11:29:42 +0700 add submodules: graphql and
toml-serialization`.
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 16384 # Requires larger stack than default to avoid crash.
$ ./build/all_tests 9 | tee tlog
[Suite] persist block json tests
...
Stack range 38416 depthHigh 3
...
Stack range 13074720 depthHigh 1024
[OK] tests/fixtures/PersistBlockTests/block1431916.json
These tests use 13.07 MB of stack to run, and so crash with the default stack
limit on Ubuntu Server 20.04 (8MB). Exactly 12768 bytes per EVM call stack
frame.
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 16384 # Requires larger stack than default.
$ ./build/all_tests 7 | tee tlog
[Suite] new generalstate json tests
...
Stack range 14384 depthHigh 2
...
Stack range 3495456 depthHigh 457
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest639.json
...
Stack range 3709600 depthHigh 485
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest458.json
...
Stack range 7831600 depthHigh 1024
[OK] tests/fixtures/eth_tests/GeneralStateTests/stCreate2/Create2OnDepth1024.json
These tests use 7.83MB of stack to run. About 7648 bytes per EVM call stack
frame. It _only just_ avoids crashing with the default Ubuntu Server stack
limit of 8 MB. However, it still crashes on Windows x86-64, which is why the
Windows CI EVMC target is currently disabled.
On Linux where this passes, this is so borderline that it affects work and
testing of the complex storage code, because that's called from the EVM.
Also, this greatly exceeds the default thread stack size.
Measurements after
==================
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 600 # Because we can! 600k stack.
$ ./build/all_tests 9 | tee tlog
[Suite] persist block json tests
...
Stack range 1936 depthHigh 3
...
Stack range 556272 depthHigh 1022
Stack range 556512 depthHigh 1023
Stack range 556816 depthHigh 1023
Stack range 557056 depthHigh 1024
Stack range 557360 depthHigh 1024
[OK] tests/fixtures/PersistBlockTests/block1431916.json
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 600 # Because we can! 600k stack.
$ ./build/all_tests 7 | tee tlog
[Suite] new generalstate json tests
...
Stack range 1392 depthHigh 2
...
Stack range 248912 depthHigh 457
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest639.json
...
Stack range 264144 depthHigh 485
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest458.json
...
Stack range 557360 depthHigh 1024
[OK] tests/fixtures/eth_tests/GeneralStateTests/stStaticCall/static_CallRecursiveBombPreCall.json
For both tests, a satisfying *544 bytes* per EVM call stack frame, and EVM
takes less than 600 kB total. With other overheads, both tests run in 600 kB
stack total at maximum EVM depth.
We must add some headroom on this for database activity called from the EVM,
and different compile targets. But it means the EVM itself is no longer a
stack burden.
This is much smaller than the default thread stack size on Linux (2MB), with
plenty of margin. (Just fyi, it isn't smaller than a _small_ thread stack on
Linux from a long time ago (128kB), and some small embedded C targets.)
This size is well suited to running EVMs in threads.
Further reduction
=================
This patch solves the stack problem. Windows and Linux 64-bit EVMC CI targets
can be re-enabled, and there is no longer a problem with stack usage.
We can reduce further to ~340 bytes per frame and 350 kB total, while still
complying with EVMC. But as this involves changing how errors are handled to
comply fully with EVMC, and removing `dispose` calls, it's not worth doing now
while there are other EVMC changes in progress that will have the same effect.
A Nimbus-specific extension will allow us to avoid recursion with EVMC anyway,
bringing bytes per frame to zero. We need the extension anyway, to support
Chronos `async` which parallel transaction processing is built around.
Interop with non-Nimbus over EVMC won't let us avoid recursion, but then we
can't control the stack frame size either. To prevent stack overflow in
interop I anticipate using (this method in Aleth)
[6e96ce34e3/libethereum/ExtVM.cpp (L61)].
Smoke test other versions of GCC and Clang/LLVM
===============================================
As all builds including Windows use GCC or Apple's Clang/LLVM, this is just to
verify we're in the right ballpark on all targets. I've only checked `x86_64`
though, not 32-bit, and not ARM.
It's interesting to see GCC 10 uses less stack. This is because it optimises
`struct` returns better, sometimes skipping an intermediate copy. Here it
benefits the EVMC API, but I found GCC 10 also improves the larger stack usage
of the rest of `nimbus-eth1` as well.
Apple clang 12.0.0 (clang-1200.0.26.2) on MacOS 10.15:
- 544 bytes per EVM call stack frame
GCC 10.3.0 (Ubuntu 10.3.0-1ubuntu1) on Ubuntu 21.04:
- 464 bytes per EVM call stack frame
GCC 10.2.0 (Ubuntu 10.2.0-5ubuntu1~20.04) on Ubuntu 20.04 LTS:
- 464 bytes per EVM call stack frame
GCC 11.0.1 20210417 (experimental; Ubuntu 11-20210417-1ubuntu1) on Ubuntu 21.04:
- 8 bytes per EVM call stack frame
GCC 9.3.0 (Ubuntu 9.3.0-17ubuntu1~20.04) on Ubuntu 20.04 LTS:
- 544 bytes per EVM call stack frame
GCC 8.4.0 (Ubuntu 8.4.0-3ubuntu2) on Ubuntu 20.04 LTS:
- 544 bytes per EVM call stack frame
GCC 7.5.0 (Ubuntu 7.5.0-6ubuntu2) on Ubuntu 20.04 LTS:
- 544 bytes per EVM call stack frame
GCC 9.2.1 20191008 (Ubuntu 9.2.1-9ubuntu2) on Ubuntu 19.10:
- 528 bytes per EVM call stack frame
Signed-off-by: Jamie Lokier <jamie@shareable.org>
instead of using header as input param, now getReceipts using
receiptRoot hash, the intention is clearer and less data passed around
when we only using receiptRoot instead of whole block header.
why:
these files provide part of the externally accessible interface
provided by vm_cpmputation.nim, vm_internals.nim. so the
new filename indicates that the source code belongs to vm2 (rather
than vm).
why:
these files provide part of the externally accessible interface
provided by vm_message.nim, vm_precompile.nim, vm_gas_cost.nim. so the
new filename indicates that the source code belongs to vm2 (rather
than vm).
why:
these files provide part of the externally accessible interface
provided by vm_state*.nim. so the new filename indicates that the
source code belongs to vm2 (rather than vm).
why:
these files provide part of the externally accessible interface
provided by vm_types*.nim. so the new filename indicates that the
source code belongs to vm2 (rather than vm).
why:
making sure that deep links into vm2 sources are configured properly. it
is intended that only the vm_*.nim interface headers are allowed to
source files in vm2. the sentinels just protect from coding errors.
why:
vm2 enabled by ENABLE_VM2=1 behaves as vm without ENABLE_EVMC=1 until
it doesn't in some future fatch set. this leaves some wiggle room
to work on a vm copy without degrading the original implementation.
details:
+ additional make flag ENABLE_VM2=1 (or ENABLE_VM2=0 to explicitely disable)
+ when both flags ENABLE_EVMC=1 and ENABLE_VM2=1 are present, the former
flag ENABLE_EVMC=1 takes precedence, this is implemented at the NIM
compiler level for -d:evmc_enabled and -d:vm2_enabled
There's been a lot of talk about the Nimbus EVM "stack problem". I think we
assumed changing it would require big changes to the interpreter code, touching
a lot of functions.
It turned out to be a low hanging fruit.
This patch solves the stack problem, but hardly touches anything. The change
in EVM stack memory is from 13 MB worst case to just 48 kB, a 250x reduction.
I've been doing work on the database/storage/trie code. While looking at the
API between the EVM and the database/storage/trie, this stack patch stood out
and made itself obvious. As it's tiny, rather than more talk, here it is.
Note: This patch is intentionally small, non-invasive, and hopefully easy to
understand, so that it doesn't conflict with other work done on the EVM, and
can easily be grafted into any other EVM structure.
Motivation
==========
- We run out of space and crash on some targets, unless the stack limit is
raised above its default. Surprise segmentation faults are unhelpful.
- Some CI targets have been disabled for months due to this.
- Because usage borders on the system limits, when working on
database/storage/trie/sync code (called from the EVM), segmentation faults
occur and are misleading. They cause lost time due to thinking there's a
crash bug in the code being worked on, when there's nothing wrong with it.
- Sometimes unrelated, trivial code changes elsewhere trigger CI test failures.
It looks like abrupt termination. A simple, recent patch was crashing in
`make test` even though it was a trivial refactor. Turns out it pushed the
stack over the edge.
- A large stack has to be scanned by the Nim garbage collector sometimes.
Larger stack means slower GC and memory allocation.
- The structure of this small patch suggests how to weave async into the EVM
with almost no changes to the EVM, and no async transformation overhead.
- The patch seemed obvious when working on the API between EVM and storage.
Measurements before
===================
All these tests were run on Ubuntu 20.04 server, x86-64. This is one of the
targets that has been disabled for a while in CI in EVMC mode due to crashing,
and excessive stack usage is the cause.
Testing commit 0c34a8e3 `2021-04-08 17:46:00 +0200 CI: use MSYS2 on Windows`.
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 16384 # Requires larger stack than default to avoid crash.
$ ./build/all_tests 9 | tee tlog
[Suite] persist block json tests
...
Stack range 38496 depthHigh 3
...
Stack range 13140272 depthHigh 1024
[OK] tests/fixtures/PersistBlockTests/block1431916.json
These tests use 13.14 MB of stack to run, and so crash with the default stack
limit on Ubuntu Server 20.04 (8MB). Exactly 12832 bytes per EVM call stack
frame. It's interesting to see some stack frames take a bit more.
$ rm -f build/all_tests && make ENABLE_EVMC=1 test
$ ulimit -S -s 16384 # Requires larger stack than default.
$ ./build/all_tests 7 | tee tlog
[Suite] new generalstate json tests
...
Stack range 15488 depthHigh 2
...
Stack range 3539312 depthHigh 457
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest639.json
...
Stack range 3756144 depthHigh 485
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest458.json
...
Stack range 7929968 depthHigh 1024
[OK] tests/fixtures/eth_tests/GeneralStateTests/stCreate2/Create2OnDepth1024.json
These tests use 7.92MB of stack to run. About 7264 bytes per EVM call stack
frame. It _only just_ avoids crashing with the default Ubuntu Server stack
limit of 8 MB. However, it still crashes on Windows x86-64, which is why the
CI target is currently disabled.
On Linux where this passes, this is so borderline that it affects work and
testing of storage and sync code, because that's called from the EVM. Which
was a motivation for dealing with the stack instead of letting this linger.
Also, this stack greatly exceeds the default thread stack size.
$ rm -f build/all_tests && make ENABLE_EVMC=0 test
$ ulimit -S -s 16384 # Requires larger stack than default to avoid crash.
$ ./build/all_tests 9 | tee tlog
[Suite] persist block json tests
...
Stack range 33216 depthHigh 3
...
Stack range 11338032 depthHigh 1024
[OK] tests/fixtures/PersistBlockTests/block1431916.json
These tests use 11.33 MB stack to run, and so crash with a default stack limit
of 8MB. Exactly 11072 bytes per EVM call stack frame. It's interesting to see
some stack frames take a bit more.
$ rm -f build/all_tests && make ENABLE_EVMC=0 test
$ ulimit -S -s 16384 # Requires larger stack than default.
$ ./build/all_tests 7 | tee tlog
[Suite] new generalstate json tests
...
Stack range 10224 depthHigh 2
...
Stack range 2471760 depthHigh 457
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest639.json
...
Stack range 2623184 depthHigh 485
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest458.json
...
Stack range 5537824 depthHigh 1024
[OK] tests/fixtures/eth_tests/GeneralStateTests/stCreate2/Create2OnDepth1024.json
These tests use 5.54 MB of stack to run, and avoid crashing on with a default
stack limit of 8 MB. About 5408 bytes per EVM call stack frame.
However, this is uncomfortably close to the limit, as the stack frame size is
sensitive to changes in the code.
Also, this stack greatly exceeds the default thread stack size.
Measurements after
==================
(This patch doesn't address EVMC mode, which is not our default. EVMC stack
usage remains about the same. EVMC mode is addressed in another tiny patch.)
$ rm -f build/all_tests && make ENABLE_EVMC=0 test
$ ulimit -S -s 80 # Because we can! 80k stack.
$ ./build/all_tests 9 | tee tlog
[Suite] persist block json tests
...
Stack range 496 depthHigh 3
...
Stack range 49504 depthHigh 1024
[OK] tests/fixtures/PersistBlockTests/block1431916.json
$ rm -f build/all_tests && make ENABLE_EVMC=0 test
$ ulimit -S -s 72 # Because we can! 72k stack.
$ ./build/all_tests 7 | tee tlog
[Suite] new generalstate json tests
...
Stack range 448 depthHigh 2
...
Stack range 22288 depthHigh 457
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest639.json
...
Stack range 23632 depthHigh 485
[OK] tests/fixtures/eth_tests/GeneralStateTests/stRandom2/randomStatetest458.json
...
Stack range 49504 depthHigh 1024
[OK] tests/fixtures/eth_tests/GeneralStateTests/stCreate2/Create2OnDepth1024.json
For both tests, a satisfying *48 bytes* per EVM call stack frame, and EVM takes
not much more than 48 kB. With other overheads, both tests run in 80 kB stack
total at maximum EVM depth.
We must add some headroom on this for database activity called from the EVM,
and different compile targets. But it means the EVM itself is no longer a
stack burden.
This is much smaller than the default thread stack size on Linux (2MB), with
plenty of margin. It's even smaller than Linux from a long time ago (128kB),
and some small embedded C targets. (Just fyi, though, some JVM environments
allocated just 32 kB to thread stacks.)
This size is also well suited to running EVMs in threads, if that's useful.
Subtle exception handling and `dispose`
=======================================
It is important that each `snapshot` has a corresponding `dispose` in the event
of an exception being raised. This code does do that, but in a subtle way.
The pair of functions `execCallOrCreate` and `execCallOrCreateAux` are
equivalent to the following code, where you can see `dispose` more clearly:
proc execCallOrCreate*(c: Computation) =
defer: c.dispose()
if c.beforeExec():
return
c.executeOpcodes()
while not c.continuation.isNil:
c.child.execCallOrCreate()
c.child = nil
(c.continuation)()
c.executeOpcodes()
c.afterExec()
That works fine, but only reduces the stack used to 300-700 kB instead of 48 kB.
To get lower we split the above into separate `execCallOrCreate` and
`execCallOrCreateAux`. Only the outermost has `defer`, and instead of handling
one level, it walks the entire `c.parent` chain calling `dispose` if needed.
The inner one avoids `defer`, which greatly reduces the size of its stackframe.
`c` is a `var` parameter, at each level of recursion. So the outermost proc
sees the temporary changes made by all inner calls. This is why `c` is updated
and the `c.parent` chain is maintained at each step.
Signed-off-by: Jamie Lokier <jamie@shareable.org>
* switch to chronos metrics, remove insecure
See https://github.com/status-im/nimbus-eth2/pull/2468
also fixes pcre linking for real, and adds some random build flags that
help nimbus-eth2 stay afloat
* fix help
* don't omit frame pointers on windows
why:
it was convenient to have relocatable source modules when writing the
vm interface wrappers. this patch moves it back to the standard.
also:
there are no deep links into the vm folder anymore which leaves some
room for manoeuvring inside
why:
the nvm_ prefix was used inside the vm folder to hide them temporarily
from the outside world while writing export wrappers. now all
functionality is accessed via vm_*, rather than vm/* imports.
todo:
at a later stage the import headers of the vm modules need to get fixed
to meet style guide standards (as jacek kindly pointed out.)
why:
relative paths make sources inherently non-relocatable
details:
import base is set to the nimbus directoy, so importing ./stack
from file interpreter.nim becomes vm/stack etc.
caveat:
a file named nimbus/strformat.nim would clash with strformat (but
not with std/strformat)
why:
under win/mingw32 the --docRoot argument value for the NIM doc
generator needs a window-ish path C:\\MinGW\\msys\\1.0\\home\\...
also:
fix clean up for nimdoc.out.css or nimdoc_out.css (varies on
win/posix operating system)
why:
* easy browsing of prototype docs, allows to follow links with
a web browser on the local file system
* some md & png files may contain additional documentation
overview:
* separate nimbus/makefile, try "make -C nimbus" for instructions
* running "make -C nimbus docs" will do the job
* x-ref file in nimbus/docs/theindex.html
* additional md and png files in nimbus/docs/ex/.. subdirectory
details:
* a newer nim compiler provides better referencing when available, in
particular the back link to the indices are not provided by the 1.2.10
nim compiler (automatically handled by makefile)
* make patterns are used to update files only when the timestamp changes
* should provide "discount" markdown generator, otherwise fallback
to <pre/> encapsulated text file
jangko
KonradStaniec
Jordan Hrycaj
Jamie Lokier
Jacek Sieka
kdeme
Jordan Hrycaj