; =============================================================================
; proc.asm - process table management, task creation, fork/exec (stubs)
; =============================================================================
INCLUDE "include/gbos.inc"

SECTION "proc", ROM0

; -----------------------------------------------------------------------------
; ProcInit - table already zeroed by ClearKernelRAM (PS_FREE=0). Seed pids.
; -----------------------------------------------------------------------------
ProcInit::
    ld a, 1
    ld [wNextPid], a
    ld a, 1                 ; bank 0 is the init task's; allocate from 1 up
    ld [wNextRamBank], a
    ret

; -----------------------------------------------------------------------------
; AllocRamBank - bump-allocate a cart-RAM bank.  out: A=bank, CF set if none.
; -----------------------------------------------------------------------------
AllocRamBank::
    ld a, [wNextRamBank]
    cp NUM_RAM_BANKS
    jr nc, .none
    ld b, a
    inc a
    ld [wNextRamBank], a
    ld a, b
    and a                   ; clear carry
    ret
.none
    scf
    ret

; -----------------------------------------------------------------------------
; FindFreeSlot - scan the proc table for a PS_FREE slot.
;   out: A=slot, DE=&PCB, CF set if the table is full.
; -----------------------------------------------------------------------------
FindFreeSlot::
    ld c, 0
.l
    ld a, c
    call PcbPtr             ; DE = &PCB[c]
    ld a, [de]
    cp PS_FREE
    jr z, .ok
    inc c
    ld a, c
    cp MAX_PROCS
    jr c, .l
    scf
    ret
.ok
    ld a, c
    and a                   ; clear carry
    ret

; -----------------------------------------------------------------------------
; PcbPtr - slot -> PCB pointer.  in: A=slot  out: DE=&PCB.  preserves BC.
; -----------------------------------------------------------------------------
PcbPtr::
    push bc
    ld hl, wProcTable
    ld bc, PROC_SIZE
    and a
    jr z, .done
.mul
    add hl, bc
    dec a
    jr nz, .mul
.done
    ld d, h
    ld e, l
    pop bc
    ret

; -----------------------------------------------------------------------------
; ProcCreate - allocate a PCB and build an initial (runnable) task.
;   in:  HL = task entry, B = cart-RAM bank, C = SVBK u-area bank
;   out: A  = pid, or 0 if the table is full
; -----------------------------------------------------------------------------
ProcCreate::
    ; stash params
    ld a, l
    ld [wTmpEntry], a
    ld a, h
    ld [wTmpEntry+1], a
    ld a, b
    ld [wTmpRamB], a
    ld a, c
    ld [wTmpWramB], a

    ; find a free slot
    ld c, 0
.find
    ld a, c
    call PcbPtr                 ; DE = &PCB[c]
    ld a, [de]
    cp PS_FREE
    jr z, .found
    inc c
    ld a, c
    cp MAX_PROCS
    jr c, .find
    xor a                       ; table full
    ret

.found
    ld a, c
    ld [wTmpSlot], a

    ; fill fields; HL walks the PCB
    ld h, d
    ld l, e
    ld a, PS_READY
    ld [hl+], a                 ; PROC_STATE
    ld a, [wNextPid]
    ld [hl+], a                 ; PROC_PID
    ld [wTmpPid], a
    inc a
    ld [wNextPid], a
    inc hl                      ; skip PROC_SP (filled below)
    inc hl
    ld a, 1
    ld [hl+], a                 ; PROC_ROMB lo (placeholder text bank)
    xor a
    ld [hl+], a                 ; PROC_ROMB hi
    ld a, [wTmpRamB]
    ld [hl+], a                 ; PROC_RAMB
    ld a, [wTmpWramB]
    ld [hl+], a                 ; PROC_WRAMB
    xor a
    ld [hl+], a                 ; PROC_PARENT
    ld [hl], a                  ; PROC_EXIT

    ; build the initial stack frame in the task's RAM bank
    ld a, [wTmpEntry]
    ld l, a
    ld a, [wTmpEntry+1]
    ld h, a                     ; HL = entry
    ld a, [wTmpRamB]
    ld b, a                     ; B = RAM bank
    call ProcSetupStack         ; out: HL = initial SP
    ld a, l
    ld [wTmpSP], a
    ld a, h
    ld [wTmpSP+1], a

    ; store SP into PCB.PROC_SP
    ld a, [wTmpSlot]
    call PcbPtr                 ; DE = &PCB
    ld a, e
    add PROC_SP
    ld l, a
    ld a, d
    adc 0
    ld h, a                     ; HL = &PROC_SP
    ld a, [wTmpSP]
    ld [hl+], a
    ld a, [wTmpSP+1]
    ld [hl], a

    ld a, [wTmpPid]
    ret

; -----------------------------------------------------------------------------
; ProcSetupStack - lay a fresh switch-in frame at the top of a task RAM bank.
;   in:  HL = entry, B = cart-RAM bank
;   out: HL = initial SP (points at the saved-HL slot)
; The frame matches hSwitchTo's restore sequence:
;   [SP+0..1]=hl [+2..3]=de [+4..5]=bc [+6]=f [+7]=a [+8..9]=entry (ret target)
; -----------------------------------------------------------------------------
ProcSetupStack::
    ld a, b
    ld [MBC5_RAMB], a           ; map task's RAM bank into $A000-$BFFF
    push hl                     ; save entry
    ld hl, TASK_STACK_TOP - 9   ; frame base = $BFF6
    xor a
    ld [hl+], a                 ; hl_lo
    ld [hl+], a                 ; hl_hi
    ld [hl+], a                 ; de_lo
    ld [hl+], a                 ; de_hi
    ld [hl+], a                 ; bc_lo
    ld [hl+], a                 ; bc_hi
    ld [hl+], a                 ; f
    ld [hl+], a                 ; a
    pop de                      ; DE = entry
    ld a, e
    ld [hl+], a                 ; entry lo (ret target)
    ld a, d
    ld [hl], a                  ; entry hi
    ld hl, TASK_STACK_TOP - 9   ; initial SP
    ret

; -----------------------------------------------------------------------------
; sys_fork / sys_exec - TODO. Plan:
;   fork: alloc slot, copy parent's 8 KiB RAM bank -> child bank, dup u-area,
;         patch child's saved-A to 0 (child return), parent gets child pid.
;   exec: point PROC_ROMB at a flashed program image, reset RAM-bank layout
;         (copy .data from ROM, zero .bss, fresh heap/stack), longjmp in.
; -----------------------------------------------------------------------------

; =============================================================================
; sys_fork() - duplicate the current process.
;   out: A = child pid in the parent, 0 in the child, $FF on failure.
;
; Runs on the parent's user stack at entry (SP = Uafter, pointing at the rst
; return address).  We switch to a kernel stack in WRAM0 so we're free to swap
; the $A000 cart-RAM window (the parent's user stack lives there), copy the
; parent's 8 KiB bank into a freshly allocated child bank via a WRAM bounce
; buffer, then plant a switch-in frame in the child so it "returns from fork"
; with A=0 to the same user PC.
; =============================================================================
sys_fork::
    ; capture parent user SP (= Uafter)
    ld hl, sp+0
    ld a, l
    ld [wForkUserSP], a
    ld a, h
    ld [wForkUserSP+1], a
    ld sp, KSTACK_TOP2          ; move off the cart-RAM stack

    call FindFreeSlot           ; A=slot, DE=&child PCB
    jp c, .fail
    ld [wForkChildSlot], a
    call AllocRamBank           ; A=child cart-RAM bank
    jp c, .fail
    ld [wForkChildBank], a

    ; remember the parent's bank so we can restore the window afterwards
    ld a, [wCurProc]
    add PROC_RAMB
    ld l, a
    ld a, [wCurProc+1]
    adc 0
    ld h, a
    ld a, [hl]
    ld [wForkParentBank], a

    ; ---- copy parent bank -> child bank, 32 pages of 256 bytes ----
    ld c, 0                     ; page index 0..31
.page
    ld a, [wForkParentBank]
    ld [MBC5_RAMB], a           ; map parent bank
    ld a, $A0
    add c
    ld h, a
    ld l, 0                     ; HL = $A000 + c*256
    ld de, wCopyBuf
    ld b, 0                     ; 256 iterations (b wraps 0->255..->0)
.rd
    ld a, [hl+]
    ld [de], a
    inc de
    dec b
    jr nz, .rd
    ld a, [wForkChildBank]
    ld [MBC5_RAMB], a           ; map child bank
    ld a, $A0
    add c
    ld h, a
    ld l, 0
    ld de, wCopyBuf
    ld b, 0
.wr
    ld a, [de]
    ld [hl+], a
    inc de
    dec b
    jr nz, .wr
    inc c
    ld a, c
    cp 32
    jr c, .page
    ; child bank is currently mapped

    ; ---- plant the child's switch-in frame at Uafter-8 (in child bank) ----
    ; hSwitchTo restore pops hl,de,bc,af then rets; we zero those 8 bytes and
    ; leave the copied return address (retU) at Uafter untouched, so the child
    ; resumes at the post-fork PC with A=0.
    ld a, [wForkUserSP]
    sub 8
    ld l, a
    ld a, [wForkUserSP+1]
    sbc 0
    ld h, a                     ; HL = Uafter-8
    xor a
    ld b, 8
.zframe
    ld [hl+], a
    dec b
    jr nz, .zframe

    ; restore the parent's bank into the window (parent stack valid again)
    ld a, [wForkParentBank]
    ld [MBC5_RAMB], a

    ; ---- fill the child PCB ----
    ld a, [wForkChildSlot]
    call PcbPtr                 ; DE = &child PCB
    ld h, d
    ld l, e                     ; HL walks child PCB
    ld a, PS_READY
    ld [hl+], a                 ; STATE
    ld a, [wNextPid]
    ld [hl+], a                 ; PID
    ld [wForkChildPid], a
    inc a
    ld [wNextPid], a
    ld a, [wForkUserSP]         ; PROC_SP = Uafter-8
    sub 8
    ld [hl+], a
    ld a, [wForkUserSP+1]
    sbc 0
    ld [hl+], a
    ; PROC_ROMB = parent ROMB (shared read-only text)
    ld a, [wCurProc]
    add PROC_ROMB
    ld e, a
    ld a, [wCurProc+1]
    adc 0
    ld d, a                     ; DE = &parent.PROC_ROMB
    ld a, [de]
    ld [hl+], a
    inc de
    ld a, [de]
    ld [hl+], a
    ; PROC_RAMB = child bank
    ld a, [wForkChildBank]
    ld [hl+], a
    ; PROC_WRAMB = parent WRAMB (u-area shared for now)
    ld a, [wCurProc]
    add PROC_WRAMB
    ld e, a
    ld a, [wCurProc+1]
    adc 0
    ld d, a
    ld a, [de]
    ld [hl+], a
    ; PROC_PARENT = parent pid
    ld a, [wCurProc]
    add PROC_PID
    ld e, a
    ld a, [wCurProc+1]
    adc 0
    ld d, a
    ld a, [de]
    ld [hl+], a
    ; PROC_EXIT = 0
    xor a
    ld [hl], a

    ; ---- return to the parent with child pid ----
    ld a, [wForkUserSP]
    ld l, a
    ld a, [wForkUserSP+1]
    ld h, a
    ld sp, hl                   ; back on the parent's user stack
    ld a, [wForkChildPid]
    ret

.fail
    ; restore parent user stack and report failure
    ld a, [wForkUserSP]
    ld l, a
    ld a, [wForkUserSP+1]
    ld h, a
    ld sp, hl
    ld a, $FF
    ret