# The OS/K Team licenses this file to you under the MIT license. # See the LICENSE file in the project root for more information. #------------------------------------------------------------------------------# 1. STACK Stack grows downward. 'rbp' and 'rsp' are both used. The lowest stack frame is marked by having 'rbp' = 0. There is a 128-bytes red zone below 'rsp'. A function's assembly code looks like this: label: enter N ... ... ... leave ret 'N' is the number of local variables used by the function. The above code is equivalent to the following, but much faster: label: mov [rsp-8], rbp lea rbp, [rsp-8] sub rsp, (N+1)*8 ... ... ... lea rsp, [rbp+8] mov rbp, [rbp] ret Between the 'enter' and the 'leave', the stack looks like this: . . . . . (caller's) . |---------------| | saved rip | rbp+8 | saved rbp | rbp+0 | . | rsp+(N*8) \ | . | ... | | . | ... | | . | ... |- Local variables | . | rsp+16 | | . | rsp+8 | | . | rsp+0 / |---------------| . (red zone) . . . . . 'enter' and 'leave' can be omitted if ALL the following hold: the function - uses no local variables (on the stack) - never uses any function that changes 'rbp' nor 'rsp', aside from 'call' and 'ret' You can never omit 'enter' without omitting 'leave', and vice-versa. #------------------------------------------------------------------------------# 2. CALLING CONVENTION No matter whether 'enter' and 'leave' were ommited or not, the following must hold true whenever a function is entered or exited from: - 'rbp' must be 8-bytes aligned - 'rsp' must be 8-bytes aligned - the DF flag must be cleared Aside from the DF flag, a function cannot assume anything about the state of the flags in the FLG register. Passing parameters is done using the following registers, in that order: ax0-ax9, a10-a31 The stack is never used for argument passing, except for variadic functions, cf the next section. If you need to pass large structures of data, pass their address in an appropriate register. Return values are passed in 'rax'. If the return value does not fit and require more registers, use the following registers, in that order: rax, rdx, a16-a31 The following registers are volatile; the calling function cannot assume that they will be left unmodified by the called function: rax, rcx, rdx, rx8-r31, ax0-a31 The following registers are nonvolatile; the called function must preserve them: rbx, rsi, rdi, nx0-n31, rbp, rsp #------------------------------------------------------------------------------# 3. VARIADIC FUNCTIONS To call a variadic function, do this: sub rsp, nargs * 8 mov [rsp], arg0 ... ... ... mov [rsp+(N*8)], argN call variadic_func add rsp, nargs * 8 To the variadic function, argN can be accessed the following way: mov reg, [rbp+N*8+16] For instance: mov rax, [rbp+16] ; arg0 mov rdx, [rbp+24] ; arg1 It is recommended to use the reg+reg*imm16+imm16 memory format: mov rax, [rbp+rcx*8+16] ; accesses arg#rcx The 'va_list' type can be regarded as a pointer to the variadic function's rbp+16 #------------------------------------------------------------------------------# 4. SPECIAL REGISTERS The 'inv' register cannot be referenced by machine code except when specified as an offset register in the [reg+reg(*/+...)] memory formats; in these case, 'inv' can be assumed to be always null. The following registers can only be used by the supervisor: - Fast global variables for the supervisor: sa0-sa7 - Debugging registers for the supervisor: dr0-dr7 The following registers cannot be referenced by machine code at all: px0-px1, fc0-fc2, cr0-cr7 #------------------------------------------------------------------------------#