At runtime, ZKVM should inform the guest of its MethodId

[risc0]

At runtime, ZKVM should inform the guest of its MethodId

[flaub]

A replacement for MethodIDs are close to being complete. This will be an ImageID which is the root hash of a merkle tree where the entries are themselves hashes of a page of memory. When an ELF is produced, an ImageID may also be computed. The initial memory image that is supplied to the proof system will include this merkle tree. The root entry will reside in a well known location.

My proposal for this issue is to add a new ecall which can be used by the guest to return this hash. We need an ecall because guest code is not allowed to access system memory which is where the ImageID resides.

[SchmErik]

I think this can pose a chicken and egg problem but I'm not exactly sure... I'll write down my thoughts...

The MethodID of a given guest method can be thought of as a collection of ID's where each method ID represents a different po2 execution cycles. The ID is computed by looking at the parts of the trace that is static in particular it heavily relies on the values in the trace during the program load phase. After the program load phase, we pad the rest of the po2 execution cycles as if it were in the BODY. Meaning that for each po2 cycles, we will generate different Merkle roots.

The confusion comes in when we're thinking about loading the methodID to memory. How can we compute the methodID's if the program load phase is accounted for during the creation of the methodID? It seems like loading the memory with method ID's would have to be excluded from generating the method ID itself.

[SchmErik]

Oh from reading @intoverflow's proposed fix, we can write the set of method ID's after the load phase, meaning that the initialization of the read only memory region wouldn't be accounted for when generating the method ID.

[SchmErik]

Another point: given an execution trace and a set of method ID's, how do we know that the correct method ID's were loaded to a read only memory region? I guess we could add this as a step during verification?