flamenco, vm: (SIMD-0182) consume requested CUs for VM Interp failures

[ravyu-jump]

WIP: Do not merge until

• Anza creates separate feature gate for this: use new feature gate for cu depletion anza-xyz/agave#3994
• Generate fixtures + fuzz runs Will be done once Anza cuts a release for the feature gate

Implements SIMD-0182.

Agave implementation for reference

[0x0ece]

I think that this changes the semantic? Syscalls return a typed error. If we change it to "syscall error", then we loose the type.

FWIW I think the SIMD is very unclear because they rely on SyscallError which is a rust thing, and it's also very unclear when it's thrown vs not. For example, memory violation inside a syscall... is that syscall error or ebpf!?

If we want to avoid syscalls (and I think we should), I think we need to store another var that says "error_because_of_syscall", not reuse the existing err.

[ravyu-jump]

If we want to avoid syscalls (and I think we should), I think we need to store another var that says "error_because_of_syscall", not reuse the existing err.

This FD_VM_ERR_SIGSYSCALL is a new VM error code specifically for the purpose you mentioned. Do we need a separate variable? Would the error information that's stored in the txn_ctx suffice in that case?

FWIW I think the SIMD is very unclear because they rely on SyscallError which is a rust thing, and it's also very unclear when it's thrown vs not. For example, memory violation inside a syscall... is that syscall error or ebpf!?

Yeah it's a little confusing. I am taking it as a "read the error code output of the VM execution and determine if it was caused by a syscall" situation

[0x0ece]

Makes sense

[topointon-jump]

Yeah, it's about whether or not the error was thrown in the middle of a basic block or not. Syscall errors are thrown at boundaries of basic blocks, so we don't have to set consume all the requested in the failure cases.

For example, memory violation inside a syscall... is that syscall error or ebpf!?

This is classified as a syscall error because of where it was thrown. I do agree that the SIMD can be made clearer.