Main function, txn processing loop

evm/src/all_stark.rs

@@ -338,7 +338,7 @@ mod tests {
             row.opcode_bits = bits_from_opcode(0x5b);
             row.is_cpu_cycle = F::ONE;
             row.is_kernel_mode = F::ONE;
-            row.program_counter = F::from_canonical_usize(KERNEL.global_labels["route_txn"]);
+            row.program_counter = F::from_canonical_usize(KERNEL.global_labels["main"]);
             cpu_stark.generate(row.borrow_mut());
             cpu_trace_rows.push(row.into());
         }
@@ -377,8 +377,8 @@ mod tests {
             row.is_cpu_cycle = F::ONE;
             row.is_kernel_mode = F::ONE;
 
-            // Since these are the first cycle rows, we must start with PC=route_txn then increment.
-            row.program_counter = F::from_canonical_usize(KERNEL.global_labels["route_txn"] + i);
+            // Since these are the first cycle rows, we must start with PC=main then increment.
+            row.program_counter = F::from_canonical_usize(KERNEL.global_labels["main"] + i);
             row.opcode_bits = bits_from_opcode(
                 if logic_trace[logic::columns::IS_AND].values[i] != F::ZERO {
                     0x16

evm/src/cpu/control_flow.rs

@@ -69,12 +69,12 @@ pub fn eval_packed_generic<P: PackedField>(
     );
 
     // If a non-CPU cycle row is followed by a CPU cycle row, then:
-    //  - the `program_counter` of the CPU cycle row is `route_txn` (the entry point of our kernel),
+    //  - the `program_counter` of the CPU cycle row is `main` (the entry point of our kernel),
     //  - execution is in kernel mode, and
     //  - the stack is empty.
     let is_last_noncpu_cycle = (lv.is_cpu_cycle - P::ONES) * nv.is_cpu_cycle;
     let pc_diff =
-        nv.program_counter - P::Scalar::from_canonical_usize(KERNEL.global_labels["route_txn"]);
+        nv.program_counter - P::Scalar::from_canonical_usize(KERNEL.global_labels["main"]);
     yield_constr.constraint_transition(is_last_noncpu_cycle * pc_diff);
     yield_constr.constraint_transition(is_last_noncpu_cycle * (nv.is_kernel_mode - P::ONES));
     yield_constr.constraint_transition(is_last_noncpu_cycle * nv.stack_len);
@@ -118,18 +118,18 @@ pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
     }
 
     // If a non-CPU cycle row is followed by a CPU cycle row, then:
-    //  - the `program_counter` of the CPU cycle row is `route_txn` (the entry point of our kernel),
+    //  - the `program_counter` of the CPU cycle row is `main` (the entry point of our kernel),
     //  - execution is in kernel mode, and
     //  - the stack is empty.
     {
         let is_last_noncpu_cycle =
             builder.mul_sub_extension(lv.is_cpu_cycle, nv.is_cpu_cycle, nv.is_cpu_cycle);
 
-        // Start at `route_txn`.
-        let route_txn = builder.constant_extension(F::Extension::from_canonical_usize(
-            KERNEL.global_labels["route_txn"],
+        // Start at `main`.
+        let main = builder.constant_extension(F::Extension::from_canonical_usize(
+            KERNEL.global_labels["main"],
         ));
-        let pc_diff = builder.sub_extension(nv.program_counter, route_txn);
+        let pc_diff = builder.sub_extension(nv.program_counter, main);
         let pc_constr = builder.mul_extension(is_last_noncpu_cycle, pc_diff);
         yield_constr.constraint_transition(builder, pc_constr);
 

evm/src/cpu/kernel/aggregator.rs

@@ -33,6 +33,7 @@ pub(crate) fn combined_kernel() -> Kernel {
         include_str!("asm/curve/secp256k1/moddiv.asm"),
         include_str!("asm/exp.asm"),
         include_str!("asm/halt.asm"),
+        include_str!("asm/main.asm"),
         include_str!("asm/memory/core.asm"),
         include_str!("asm/memory/memcpy.asm"),
         include_str!("asm/memory/metadata.asm"),

evm/src/cpu/kernel/asm/keccak.asm

@@ -1,8 +0,0 @@
-// Computes the Keccak256 hash of some arbitrary bytes in memory.
-// The given memory values should be in the range of a byte.
-//
-// Pre stack: ADDR, len, retdest
-// Post stack: hash
-global keccak_general:
-    // stack: ADDR, len
-    // TODO

evm/src/cpu/kernel/asm/main.asm

@@ -0,0 +1,8 @@
+global main:
+    // If the prover has no more txns for us to process, halt.
+    PROVER_INPUT(end_of_txns)
+    %jumpi(halt)
+
+    // Call route_txn, returning to main to continue the loop.
+    PUSH main
+    %jump(route_txn)

evm/src/cpu/kernel/asm/rlp/read_to_memory.asm

@@ -6,7 +6,7 @@
 
 global read_rlp_to_memory:
     // stack: retdest
-    PROVER_INPUT // Read the RLP blob length from the prover tape.
+    PROVER_INPUT(rlp) // Read the RLP blob length from the prover tape.
     // stack: len, retdest
     PUSH 0 // initial position
     // stack: pos, len, retdest
@@ -19,7 +19,7 @@ read_rlp_to_memory_loop:
     // stack: pos == len, pos, len, retdest
     %jumpi(read_rlp_to_memory_finish)
     // stack: pos, len, retdest
-    PROVER_INPUT
+    PROVER_INPUT(rlp)
     // stack: byte, pos, len, retdest
     DUP2
     // stack: pos, byte, pos, len, retdest

evm/src/cpu/kernel/asm/transactions/router.asm

@@ -3,14 +3,14 @@
 // jump to the appropriate transaction parsing method.
 
 global route_txn:
-    // stack: (empty)
+    // stack: retdest
     // First load transaction data into memory, where it will be parsed.
     PUSH read_txn_from_memory
     %jump(read_rlp_to_memory)
 
 // At this point, the raw txn data is in memory.
 read_txn_from_memory:
-    // stack: (empty)
+    // stack: retdest
 
     // We will peak at the first byte to determine what type of transaction this is.
     // Note that type 1 and 2 transactions have a first byte of 1 and 2, respectively.
@@ -20,17 +20,17 @@ read_txn_from_memory:
     PUSH 0
     %mload_current(@SEGMENT_RLP_RAW)
     %eq_const(1)
-    // stack: first_byte == 1
+    // stack: first_byte == 1, retdest
     %jumpi(process_type_1_txn)
-    // stack: (empty)
+    // stack: retdest
 
     PUSH 0
     %mload_current(@SEGMENT_RLP_RAW)
     %eq_const(2)
-    // stack: first_byte == 2
+    // stack: first_byte == 2, retdest
     %jumpi(process_type_2_txn)
-    // stack: (empty)
+    // stack: retdest
 
     // At this point, since it's not a type 1 or 2 transaction,
     // it must be a legacy (aka type 0) transaction.
-    %jump(process_type_2_txn)
+    %jump(process_type_0_txn)

evm/src/cpu/kernel/asm/transactions/type_0.asm

@@ -12,15 +12,15 @@
 //     keccak256(rlp([nonce, gas_price, gas_limit, to, value, data]))
 
 global process_type_0_txn:
-    // stack: (empty)
+    // stack: retdest
     PUSH 0 // initial pos
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_list_len
     // We don't actually need the length.
     %stack (pos, len) -> (pos)
 
     // Decode the nonce and store it.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, nonce) -> (nonce, pos)
     %mstore_txn_field(@TXN_FIELD_NONCE)
@@ -29,38 +29,38 @@ global process_type_0_txn:
     // For legacy transactions, we set both the
     // TXN_FIELD_MAX_PRIORITY_FEE_PER_GAS and TXN_FIELD_MAX_FEE_PER_GAS
     // fields to gas_price.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, gas_price) -> (gas_price, gas_price, pos)
     %mstore_txn_field(@TXN_FIELD_MAX_PRIORITY_FEE_PER_GAS)
     %mstore_txn_field(@TXN_FIELD_MAX_FEE_PER_GAS)
 
     // Decode the gas limit and store it.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, gas_limit) -> (gas_limit, pos)
     %mstore_txn_field(@TXN_FIELD_GAS_LIMIT)
 
     // Decode the "to" field and store it.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, to) -> (to, pos)
     %mstore_txn_field(@TXN_FIELD_TO)
 
     // Decode the value field and store it.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, value) -> (value, pos)
     %mstore_txn_field(@TXN_FIELD_VALUE)
 
     // Decode the data length, store it, and compute new_pos after any data.
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_string_len
     %stack (pos, data_len) -> (data_len, pos, data_len, pos, data_len)
     %mstore_txn_field(@TXN_FIELD_DATA_LEN)
-    // stack: pos, data_len, pos, data_len
+    // stack: pos, data_len, pos, data_len, retdest
     ADD
-    // stack: new_pos, pos, data_len
+    // stack: new_pos, pos, data_len, retdest
 
     // Memcpy the txn data from @SEGMENT_RLP_RAW to @SEGMENT_TXN_DATA.
     PUSH parse_v
@@ -70,62 +70,62 @@ global process_type_0_txn:
     PUSH 0
     PUSH @SEGMENT_TXN_DATA
     GET_CONTEXT
-    // stack: DST, SRC, data_len, parse_v, new_pos
+    // stack: DST, SRC, data_len, parse_v, new_pos, retdest
     %jump(memcpy)
 
 parse_v:
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
-    // stack: pos, v
+    // stack: pos, v, retdest
     SWAP1
-    // stack: v, pos
+    // stack: v, pos, retdest
     DUP1
     %gt_const(28)
-    // stack: v > 28, v, pos
+    // stack: v > 28, v, pos, retdest
     %jumpi(process_v_new_style)
 
     // We have an old style v, so y_parity = v - 27.
     // No chain ID is present, so we can leave TXN_FIELD_CHAIN_ID_PRESENT and
     // TXN_FIELD_CHAIN_ID with their default values of zero.
-    // stack: v, pos
+    // stack: v, pos, retdest
     %sub_const(27)
     %stack (y_parity, pos) -> (y_parity, pos)
     %mstore_txn_field(@TXN_FIELD_Y_PARITY)
 
-    // stack: pos
+    // stack: pos, retdest
     %jump(parse_r)
 
 process_v_new_style:
-    // stack: v, pos
+    // stack: v, pos, retdest
     // We have a new style v, so chain_id_present = 1,
     // chain_id = (v - 35) / 2, and y_parity = (v - 35) % 2.
     %stack (v, pos) -> (1, v, pos)
     %mstore_txn_field(@TXN_FIELD_CHAIN_ID_PRESENT)
 
-    // stack: v, pos
+    // stack: v, pos, retdest
     %sub_const(35)
     DUP1
-    // stack: v - 35, v - 35, pos
+    // stack: v - 35, v - 35, pos, retdest
     %div_const(2)
-    // stack: chain_id, v - 35, pos
+    // stack: chain_id, v - 35, pos, retdest
     %mstore_txn_field(@TXN_FIELD_CHAIN_ID)
 
-    // stack: v - 35, pos
+    // stack: v - 35, pos, retdest
     %mod_const(2)
-    // stack: y_parity, pos
+    // stack: y_parity, pos, retdest
     %mstore_txn_field(@TXN_FIELD_Y_PARITY)
 
 parse_r:
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, r) -> (r, pos)
     %mstore_txn_field(@TXN_FIELD_R)
 
-    // stack: pos
+    // stack: pos, retdest
     %decode_rlp_scalar
     %stack (pos, s) -> (s)
     %mstore_txn_field(@TXN_FIELD_S)
-    // stack: (empty)
+    // stack: retdest
 
     // TODO: Write the signed txn data to memory, where it can be hashed and
     // checked against the signature.

evm/src/cpu/kernel/asm/transactions/type_1.asm

@@ -7,5 +7,5 @@
 //                            data, access_list]))
 
 global process_type_1_txn:
-    // stack: (empty)
+    // stack: retdest
     PANIC // TODO: Unfinished

evm/src/cpu/kernel/asm/transactions/type_2.asm

@@ -8,5 +8,5 @@
 //                            access_list]))
 
 global process_type_2_txn:
-    // stack: (empty)
+    // stack: retdest
     PANIC // TODO: Unfinished

evm/src/cpu/kernel/tests/transaction_parsing/parse_type_0_txn.rs

@@ -12,7 +12,8 @@ fn process_type_0_txn() -> Result<()> {
     let process_type_0_txn = KERNEL.global_labels["process_type_0_txn"];
     let process_normalized_txn = KERNEL.global_labels["process_normalized_txn"];
 
-    let mut interpreter = Interpreter::new_with_kernel(process_type_0_txn, vec![]);
+    let retaddr = 0xDEADBEEFu32.into();
+    let mut interpreter = Interpreter::new_with_kernel(process_type_0_txn, vec![retaddr]);
 
     // When we reach process_normalized_txn, we're done with parsing and normalizing.
     // Processing normalized transactions is outside the scope of this test.

evm/src/generation/mod.rs

@@ -23,6 +23,7 @@ use crate::util::trace_rows_to_poly_values;
 pub(crate) mod memory;
 pub(crate) mod mpt;
 pub(crate) mod prover_input;
+pub(crate) mod rlp;
 pub(crate) mod state;
 
 #[derive(Clone, Debug, Deserialize, Serialize, Default)]

evm/src/generation/prover_input.rs

@@ -24,12 +24,24 @@ impl<F: Field> GenerationState<F> {
     #[allow(unused)] // TODO: Should be used soon.
     pub(crate) fn prover_input(&mut self, stack: &[U256], input_fn: &ProverInputFn) -> U256 {
         match input_fn.0[0].as_str() {
+            "end_of_txns" => self.run_end_of_txns(),
             "ff" => self.run_ff(stack, input_fn),
             "mpt" => self.run_mpt(),
+            "rlp" => self.run_rlp(),
             _ => panic!("Unrecognized prover input function."),
         }
     }
 
+    fn run_end_of_txns(&mut self) -> U256 {
+        let end = self.next_txn_index == self.inputs.signed_txns.len();
+        if end {
+            U256::one()
+        } else {
+            self.next_txn_index += 1;
+            U256::zero()
+        }
+    }
+
     /// Finite field operations.
     fn run_ff(&self, stack: &[U256], input_fn: &ProverInputFn) -> U256 {
         let field = EvmField::from_str(input_fn.0[1].as_str()).unwrap();
@@ -44,6 +56,13 @@ impl<F: Field> GenerationState<F> {
             .pop()
             .unwrap_or_else(|| panic!("Out of MPT data"))
     }
+
+    /// RLP data.
+    fn run_rlp(&mut self) -> U256 {
+        self.rlp_prover_inputs
+            .pop()
+            .unwrap_or_else(|| panic!("Out of RLP data"))
+    }
 }
 
 enum EvmField {

evm/src/generation/rlp.rs

@@ -0,0 +1,18 @@
+use ethereum_types::U256;
+
+pub(crate) fn all_rlp_prover_inputs_reversed(signed_txns: &[Vec<u8>]) -> Vec<U256> {
+    let mut inputs = all_rlp_prover_inputs(signed_txns);
+    inputs.reverse();
+    inputs
+}
+
+fn all_rlp_prover_inputs(signed_txns: &[Vec<u8>]) -> Vec<U256> {
+    let mut prover_inputs = vec![];
+    for txn in signed_txns {
+        prover_inputs.push(txn.len().into());
+        for &byte in txn {
+            prover_inputs.push(byte.into());
+        }
+    }
+    prover_inputs
+}

evm/src/generation/state.rs

@@ -7,6 +7,7 @@ use tiny_keccak::keccakf;
 use crate::cpu::columns::{CpuColumnsView, NUM_CPU_COLUMNS};
 use crate::generation::memory::MemoryState;
 use crate::generation::mpt::all_mpt_prover_inputs_reversed;
+use crate::generation::rlp::all_rlp_prover_inputs_reversed;
 use crate::generation::GenerationInputs;
 use crate::keccak_memory::keccak_memory_stark::KeccakMemoryOp;
 use crate::memory::memory_stark::MemoryOp;
@@ -19,6 +20,7 @@ use crate::{keccak, logic};
 pub(crate) struct GenerationState<F: Field> {
     #[allow(unused)] // TODO: Should be used soon.
     pub(crate) inputs: GenerationInputs,
+    pub(crate) next_txn_index: usize,
     pub(crate) cpu_rows: Vec<[F; NUM_CPU_COLUMNS]>,
     pub(crate) current_cpu_row: CpuColumnsView<F>,
 
@@ -32,14 +34,20 @@ pub(crate) struct GenerationState<F: Field> {
     /// Prover inputs containing MPT data, in reverse order so that the next input can be obtained
     /// via `pop()`.
     pub(crate) mpt_prover_inputs: Vec<U256>,
+
+    /// Prover inputs containing RLP data, in reverse order so that the next input can be obtained
+    /// via `pop()`.
+    pub(crate) rlp_prover_inputs: Vec<U256>,
 }
 
 impl<F: Field> GenerationState<F> {
     pub(crate) fn new(inputs: GenerationInputs) -> Self {
         let mpt_prover_inputs = all_mpt_prover_inputs_reversed(&inputs.tries);
+        let rlp_prover_inputs = all_rlp_prover_inputs_reversed(&inputs.signed_txns);
 
         Self {
             inputs,
+            next_txn_index: 0,
             cpu_rows: vec![],
             current_cpu_row: [F::ZERO; NUM_CPU_COLUMNS].into(),
             current_context: 0,
@@ -48,6 +56,7 @@ impl<F: Field> GenerationState<F> {
             keccak_memory_inputs: vec![],
             logic_ops: vec![],
             mpt_prover_inputs,
+            rlp_prover_inputs,
         }
     }