feat(vm): Implement call tracing for fast VM #2905

joonazan · 2024-09-17T23:10:40Z

What ❔

Implements call tracing for the fast VM.
Enables call tracing for the batch VM executor (i.e., one used in the state keeper) if it is requested. Compares produced traces in the shadow VM mode.

Why ❔

Call tracing is the last large missing part of the fast VM.

joonazan · 2024-11-15T15:10:06Z

This does not turn enable call tracing in API because of the details of how that happens depend on #2863 so it is better to wait until both PRs are in.

popzxc

This is certainly not the domain I'm comfortable with, but I've looked through the diff and it doesn't look suspicious. Not very helpful, but LGTM 😅

joonazan

Didn't find any serious issues. Two very confusing spots that can probably be fixed with comments.

joonazan · 2025-01-20T10:38:16Z

core/lib/multivm/src/tracers/call_tracer/vm_latest/mod.rs

@@ -191,15 +193,14 @@ impl CallTracer {
            .farcall
            .parent_gas
            .saturating_sub(state.vm_local_state.callstack.current.ergs_remaining as u64);
-
        self.save_output_latest(state, memory, ret_opcode, &mut current_call.farcall);

        // If there is a parent call, push the current call to it
        // Otherwise, push the current call to the stack, because it's the top level call


This comment is outdated.

joonazan · 2025-01-20T10:57:45Z

core/lib/multivm/src/versions/vm_fast/tracers/calls.rs

+                        let output_ptr = FatPointer::from(maybe_output_ptr);
+                        if output_ptr.length == 0 && output_ptr.offset == 0 {
+                            // Trivial pointer, which is formally cannot be dereferenced. This only matters
+                            // when extracting the revert reason; the legacy VM treats the trivial pointer specially.


Does this case require special handling? I'd imagine that reading a length zero pointer produces an empty array anyway.

Nit: extra is.

It seems to cause "Unknown revert reason". So length 0, offset 0 is unknown reason but offset 1 is not?

As you've noted, the reason for special handling is purely corresponding to traces produced by the legacy implementation. Subjectively, this special case makes little sense, but it's easy to implement and I assume that compatibility has higher priority.

So length 0, offset 0 is unknown reason but offset 1 is not?

Yep, this is how it works in the legacy VM:

zksync-era/core/lib/multivm/src/tracers/call_tracer/vm_latest/mod.rs

Lines 145 to 153 in 731b824

if !fat_data_pointer.is_trivial() {

Some(memory.read_unaligned_bytes(

fat_data_pointer.memory_page as usize,

fat_data_pointer.start as usize,

fat_data_pointer.length as usize,

))

} else {

None

}

where is_trivial is defined as length 0, offset 0.

🤖 I have created a release *beep* *boop* --- ## [26.1.0](core-v26.0.0...core-v26.1.0) (2025-01-21) ### Features * update l2 erc20 bridge address in updater as well ([#3500](#3500)) ([fe3c7b2](fe3c7b2)) * **vm:** Implement call tracing for fast VM ([#2905](#2905)) ([731b824](731b824)) ### Bug Fixes * copy special case to fast VM call tracer ([#3509](#3509)) ([995e583](995e583)) * fix execute encoding for transactions ([#3501](#3501)) ([4c381a8](4c381a8)) * **gateway:** erc20 workaround for gateway upgrade ([#3511](#3511)) ([c140ba8](c140ba8)) ### Performance Improvements * optimize get_unsealed_l1_batch_inner ([#3491](#3491)) ([9b121c9](9b121c9)) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). --------- Co-authored-by: zksync-era-bot <zksync-era-bot@users.noreply.github.com>

joonazan force-pushed the jms-call-tracer branch 4 times, most recently from 1cc9ec9 to 6a0c824 Compare September 24, 2024 17:20

joonazan force-pushed the jms-call-tracer branch from 6a0c824 to 52a6a72 Compare September 25, 2024 11:59

joonazan changed the base branch from main to jms-pla-908-account-validation-properly-handle-when-out-of-gas November 15, 2024 13:42

joonazan added 7 commits November 15, 2024 15:08

start implementing vm2 call tracer

c81a71d

pass the calltracer test (it is bad)

dad0f47

fix vm_latest calltracer bug

30c9aa7

add gas calcs and call type

c6ca8a2

test that calltracer always produces the same output

6e45f1e

done except for mysterious calldata diff

120c7ee

update reference file; remove last randomness

1e54c6a

joonazan changed the base branch from jms-pla-908-account-validation-properly-handle-when-out-of-gas to main November 15, 2024 14:54

fmt

7ef9bd9

joonazan force-pushed the jms-call-tracer branch from 9078e12 to 7ef9bd9 Compare November 15, 2024 14:55

joonazan marked this pull request as ready for review November 15, 2024 14:55

joonazan requested review from slowli, popzxc and StanislavBreadless November 15, 2024 14:55

slowli added 9 commits January 13, 2025 13:54

Update from upstream

53c61b7

Refactor call tracer unit tests

bf1cead

Move call_tracer module

fc82bbc

Test transfers and reverted txs

387b393

Test out-of-gas and recursive txs

2c2eff4

Fix create calls detection

7fd7e3c

Integrate call tracer into batch executor

8e00b16

Fix revert reason for out-of-gas

3915065

Test revert reason for out-of-gas

de207b7

Update from upstream

5309586

slowli requested a review from perekopskiy January 14, 2025 13:52

slowli changed the title ~~feat: vm2 call tracer~~ feat(vm): Implement call tracing for fast VM Jan 14, 2025

slowli added 3 commits January 14, 2025 15:59

Fix nits

e09a928

Update from upstream

c23eac0

Refactor DA setup

42b78b2

perekopskiy approved these changes Jan 17, 2025

View reviewed changes

popzxc approved these changes Jan 17, 2025

View reviewed changes

slowli added this pull request to the merge queue Jan 17, 2025

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Jan 17, 2025

slowli added this pull request to the merge queue Jan 20, 2025

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Jan 20, 2025

slowli added this pull request to the merge queue Jan 20, 2025

joonazan commented Jan 20, 2025

View reviewed changes

Merged via the queue into main with commit 731b824 Jan 20, 2025
33 checks passed

slowli deleted the jms-call-tracer branch January 20, 2025 11:42

zksync-era-bot mentioned this pull request Jan 20, 2025

chore(main): release core 26.1.0 #3499

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(vm): Implement call tracing for fast VM #2905

feat(vm): Implement call tracing for fast VM #2905

joonazan commented Sep 17, 2024 •

edited by slowli

Loading

joonazan commented Nov 15, 2024

popzxc left a comment

joonazan left a comment

joonazan Jan 20, 2025

joonazan Jan 20, 2025

joonazan Jan 20, 2025

slowli Jan 20, 2025

	if !fat_data_pointer.is_trivial() {
	Some(memory.read_unaligned_bytes(
	fat_data_pointer.memory_page as usize,
	fat_data_pointer.start as usize,
	fat_data_pointer.length as usize,
	))
	} else {
	None
	}

feat(vm): Implement call tracing for fast VM #2905

feat(vm): Implement call tracing for fast VM #2905

Conversation

joonazan commented Sep 17, 2024 • edited by slowli Loading

What ❔

Why ❔

joonazan commented Nov 15, 2024

popzxc left a comment

Choose a reason for hiding this comment

joonazan left a comment

Choose a reason for hiding this comment

joonazan Jan 20, 2025

Choose a reason for hiding this comment

joonazan Jan 20, 2025

Choose a reason for hiding this comment

joonazan Jan 20, 2025

Choose a reason for hiding this comment

slowli Jan 20, 2025

Choose a reason for hiding this comment

joonazan commented Sep 17, 2024 •

edited by slowli

Loading