Prefer to get & set exception from exec_env_tls

[paul-hoang]

Exceptions are currently stored in module instance, but if most of them can be stored at the exec_env that would allow using a single module instance in parallel without locking and handle the exception separately.

Using a single module instance in parallel might not be usual, but it's possible from what I understand and tested WAMR, at least for specific use cases (e.g. no writing to globals or wasm memory, only reading from that and writing to wasm stack). Please correct me if I'm missing anything.

This PR proposes a change in how we store and retrieve exception from a given module instance:

• If the exec_env_tls is present, get/set exception from that and not touching the module instance.
• If not, keep getting/setting from the module instance. This is for exceptions that happens before an exec_env is created.

[loganek]

Not sure if we can safely add anything below wasm_stack_u.bottom as this is the region used for the WASM stack frames.

[paul-hoang]

I see, I can move it to anywhere in WASMExecEnv struct. Is after module_inst OK?

[lum1n0us]

would allow using a single module instance in parallel ... Using a single module instance in parallel might not be usual,

I'm not entirely clear on the concept of "parallel" here. Does it mean using a single instance across multiple threads?

[paul-hoang]

I'm not entirely clear on the concept of "parallel" here. Does it mean using a single instance across multiple threads?

Yes, it means use a single instance across multiple threads at the same time without locking.

[yamt]

Using a single module instance in parallel might not be usual, but it's possible from what I understand and tested WAMR, at least for specific use cases (e.g. no writing to globals or wasm memory, only reading from that and writing to wasm stack). Please correct me if I'm missing anything.

currently it isn't possible.
cf. #2512

This PR proposes a change in how we store and retrieve exception from a given module instance:

i don't see what's the point of the change. it just makes it more inconsistent, doesn't it?

[lum1n0us]

Indeed, it is a very unusual scenario and somewhat conflicts with some fundamental principles of WebAssembly, possibly the weak wasm thread model or the threads proposal or others, I can't quite remember which.