BPF program consumes the entire budget on panic with jit

[jackcmay]

Problem

When running a local node with jit and executing the panic bpf program the runtime reports that the entire compute budget was consumed when actually only a few thousand were.

This behavior has been observed on mainnet, its not clear there if the issue is only with panic or other program exits as well.

Proposed Solution

This appears to have something to do with how jitt'd programs account for compute budget on exit.

This can be recreated by running a local node with jit enabled, deploying the panic program, and sending it an instruction with no accounts and no instruction data. This should also be recreatable in a unit test but I was not able to observe that.

[jackcmay]

Here is another example, this time the program retuned an error: https://explorer.solana.com/tx/55NUP1eDmu5YnnbVDe7FbvYDFoX1UqSSVLMEmwWsBwDDL6i3Cy9z3EKrC1tMRQPGcSEQVZkWRDM6fMBkPWrop6UY

[Lichtso]

This only happens if JIT is enabled, right?
What happens if you extract the program and run it in the standalone CLI tool?

[jackcmay]

Didn't try that

[jackcmay]

This bug will be exposed further to developers via #18715

[Lichtso]

I just released RBPF v0.2.14 which should fix this.
Can you check if the issues still occurs on this branch?

[jackcmay]

Yup, look good!

[github-actions]

This issue has been automatically locked since there has not been any activity in past 7 days after it was closed. Please open a new issue for related bugs.