Klaytn Virtual Machine (Previous docs)
The current version of the Klaytn Virtual Machine (KLVM) is derived from the Ethereum Virtual Machine (EVM). The content of this chapter is based primarily on the Ethereum Yellow Paper. KLVM is continuously being improved by the Klaytn team, thus this document could be updated frequently. Please do not regard this document as the final version of the KLVM specification. As described in the Klaytn position paper, the Klaytn team also plans to adopt other virtual machines or execution environments in order to strengthen the capability and performance of the Klaytn platform. This chapter presents a specification of KLVM and the differences between KLVM and EVM.
KLVM is a virtual state machine that formally specifies Klaytn's execution model. The execution model specifies how the system state is altered given a series of bytecode instructions and a small tuple of environmental data. KLVM is a quasi-Turing-complete machine; the quasi qualification stems from the fact that the computation is intrinsically bounded through a parameter, gas, which limits the total amount of computation performed.
KLVM executes Klaytn virtual machine code (or Klaytn bytecode) which consists of a sequence of KLVM instructions. The KLVM code is the programming language used for accounts on the Klaytn blockchain that contain code. The KLVM code associated with an account is executed every time a message is sent to that account; this code has the ability to read/write from/to storage and send messages.
We use the following notations and conventions in this document.
A := B
:=is used to define
- We use the terms "smart contract" and "contract" interchangeably.
The following tables summarize the symbols used in the KLVM specification.
KLVM is a simple stack-based architecture. The word size of the machine (and thus the size of stack items) is 256-bit. This was chosen to facilitate the Keccak-256 hash scheme and the elliptic-curve computations. The memory model is a simple word-addressed byte array. The stack has a maximum size of 1024. The machine also has an independent storage model; this is similar in concept to the memory but rather than a byte array, it is a word-addressable word array. Unlike memory, which is volatile, storage is nonvolatile and is maintained as part of the system state. All locations in both storage and memory are initially well-defined as zero.
The machine does not follow the standard von Neumann architecture. Rather than storing program code in generally accessible memory or storage, code is stored separately in virtual read-only memory and can be interacted with only through specialized instructions.
The machine can execute exception code for several reasons, including stack underflows and invalid instructions. Similar to an out-of-gas exception, these exceptions do not leave state changes intact. Rather, the virtual machine halts immediately and reports the issue to the execution agent (either the transaction processor or, recursively, the spawning execution environment), which will be addressed separately.
Fees (denominated in gas) are charged under three distinct circumstances, all three are prerequisite to operation execution. The first and most common is the fee intrinsic to the computation of the operation. Second, gas may be deducted to form the payment for a subordinate message call or contract creation; this forms part of the payment for
CALLCODE. Finally, gas may be charged due to an increase in memory usage.
Over an account's execution, the total fee payable for memory-usage payable is proportional to the smallest multiple of 32 bytes that are required to include all memory indices (whether for read or write) in the range. This fee is paid on a just-in-time basis; consequently, referencing an area of memory at least 32 bytes greater than any previously indexed memory will result in an additional memory usage fee. Due to this fee, it is highly unlikely that addresses will ever exceed the 32-bit bounds. That said, implementations must be able to manage this eventuality.
Storage fees have a slightly nuanced behavior. To incentivize minimization of the use of storage (which corresponds directly to a larger state database on all nodes), the execution fee for an operation that clears an entry from storage is not only waived but also elicits a qualified refund; in fact, this refund is effectively paid in advance because the initial usage of a storage location costs substantially more than normal usage.
The fee schedule
Gis a tuple of 37 scalar values corresponding to the relative costs, in gas, of a number of abstract operations that a transaction may incur. For other tables such as
accounts, please refer to this document
We define the following subsets of instructions:
The general gas cost function,
C, is defined as follows:
C(S_system, S_machine, I) := C_mem(S_machine,i') - C_mem(S_machine, i) +
C_SSTORE(S_system, S_machine), if
w == SSTORE
(w == EXP) && (S_machine == 0)
G_exp + G_expbyte x (1 + floor(log_256(S_machine,sp))),if
(w == EXP) && (S_machine,sp > 0)
G_verylow + G_copy x ceil(S_machine,sp / 32),if
w == CALLDATACOPY || CODECOPY || RETURNDATACOPY
G_extcode + G_copy x ceil(S_machine,sp / 32),if
w == EXTCODECOPY
G_log + G_logdata x S_machine,sp,if
w == LOG0
G_log + G_logdata x S_machine,sp + G_logtopic,if
w == LOG1
G_log + G_logdata x S_machine,sp + 2 x G_logtopic,if
w == LOG2
G_log + G_logdata x S_machine,sp + 3 x G_logtopic,if
w == LOG3
G_log + G_logdata x S_machine,sp + 4 x G_logtopic,if
w == LOG4
w == CALL || CALLCODE || DELEGATECALL
w == SELFDESTRUCT
w == CREATE
G_sha3 + G_sha3word x ceil(s / 32),if
w == SHA3
w == JUMPDEST
w == SLOAD
w == BALANCE
w == BLOCKHASH
S_machine,pc < length(T_code)
C_mem(a) := G_memory x a + floor(a^2 / 512)
C_SSTOREwhich will be described in the future.
The execution environment consists of the system state
S_system, the remaining gas for computation
G_rem, and the information
Ithat the execution agent provides.
Iis a tuple defined as shown below:
I := (B_header, T_code, T_depth, T_value, T_data, A_tx_sender, A_code_executor, A_code_owner, G_price, P_modify_state)
The execution model defines the function
F_apply, which can compute the resultant state
S_system', the remaining gas
G_rem', the accrued substate
Aand the resultant output
O_resultwhen given these definitions. For the present context, we will define it as follows:
(S_system', G_rem', A, O_result) = F_apply(S_system, G_rem, I)
where we must remember that
A, the accrued substate, is defined as the tuple of the suicides set
Set_suicide, the log series
L, the touched accounts
Set_touched_accountsand the refunds
A := (Set_suicide, L, Set_touched_accounts, G_refund)
In most practical implementations,
F_applywill be modeled as an iterative progression of the pair comprising the full system state
S_systemand the machine state
S_machine. Formally, we define it recursively with a function
Xthat uses an iterator function
O(which defines the result of a single cycle of the state machine) together with functions
Z, which determines if the present state is an exceptional halted machine state, and
H, which specifies the output data of an instruction if and only if the present state is a normal halted machine state.
The empty sequence, denoted as
(), is not equal to the empty set, denoted as
Set_empty; this is important when interpreting the output of
H, which evaluates to
Set_emptywhen execution is to continue but to a series (potentially empty) when execution should halt.
F_apply(S_machine, G_rem, I, T) := (S_system', S_machine,g', A, o)
(S_system', S_machine,g', A, ..., o) := X((S_system, S_machine, A^0, I))
S_machine,g := G_rem
S_machine,pc := 0
S_machine,memory := (0, 0, ...)
S_machine,i := 0
S_machine,stack := ()
S_machine,o := ()
X((S_system, S_machine, A, I)) :=
(Set_empty, S_machine, A^0, I, Set_empty)if
Z(S_system, S_machine, I)
(Set_empty, S_machine', A^0, I, o)if
w = REVERT
O(S_system, S_machine, A, I) · oif
o != Set_empty
X(O(S_system, S_machine, A, I))otherwise
o := H(S_machine, I)
(a, b, c, d) · e := (a, b, c, d, e)
S_machine' := S_machineexcept
S_machine,g' := S_machine,g - C(S_system, S_machine, I)
- This means that when we evaluate
F_apply, weextract the remaining gas
S_machine,g'from theresultant machine state
Xis thus cycled (recursively here, but implementations are generally expected to use a simple iterative loop) until either
Zbecomes true, indicating that the present state is exceptional and that the machine must be halted and any changes are discarded, or until
Hbecomes a series (rather than the empty set), indicating that the machine has reached a controlled halt.
The machine state
S_machineis defined as a tuple
(g, pc, memory, i, stack), which represent the available gas, the program counter
pc(64-bit unsigned integer), the memory contents, the active number of words in memory (counting continuously from position 0), and the stack contents. The memory contents
S_machine,memoryare a series of zeroes of size 2^256.
For ease of reading, the instruction mnemonics written in small-caps (e.g.,
ADD) should be interpreted as their numeric equivalents; the full table of instructions and their specifics is given in the Instruction Set section.
O, we define
was the current operation to be executed:
w := T_code[S_machine,pc]if
S_machine,pc < len(T_code)
NOTE: This section will be filled in the future.
As mentioned earlier, the current KLVM is based on EVM; thus, its specification currently is very similar to that of EVM. Some differences between KLVM and EVM are listed below.
- KLVM uses Klaytn's gas units, such as peb, ston, or KLAY.
- KLVM does not accept a gas price from the user; instead, it uses a platform-defined value as the gas price.
The Klaytn team will try to maintain compatibility between KLVM and EVM, but as Klaytn becomes increasingly implemented and evolves, the KLVM specification will be updated, and there will probably be more differences compared to EVM.
NOTE: This section will be updated in the future.