/* arch/riscv/arch.c — RISC-V implementation of the arch contract.
 *
 * QEMU 'virt' exposes a standard 16550-style UART at 0x10000000.
 * (On the real Lichee RV / D1 this address becomes 0x02500000 — same
 * driver, different base; that's the only line that changes for hardware.) */
#include "arch.h"
#include "console.h"

/* UART base differs by platform. Both are 16550-compatible, so the same
 * driver works; only the MMIO base changes.
 *   QEMU 'virt' : 0x10000000
 *   Allwinner D1 (Lichee RV) UART0 : 0x02500000
 * Select with -DBOARD_D1 at build time (see 'make lichee'). */
#ifdef BOARD_D1
#define UART0_BASE 0x02500000UL
#else
#define UART0_BASE 0x10000000UL
#endif
/* Register numbers (16550). The BYTE offset depends on the platform's
 * register spacing:
 *   QEMU 'virt' : 8-bit registers, 1-byte spacing (reg-shift 0)
 *   Allwinner D1: Synopsys DW 8250, 32-bit registers, 4-byte spacing
 *                 (reg-shift 2) — this is the difference that made the
 *                 first hardware boot silent. */
#define UART_RBR   0 /* receive buffer register (read)  */
#define UART_THR   0 /* transmit holding register (write) */
#define UART_LSR   5 /* line status register      */
#define UART_LSR_DR   0x01 /* data ready (RX byte available) */
#define UART_LSR_THRE 0x20 /* THR empty */

#ifdef BOARD_D1
#define UART_SHIFT 2
static inline unsigned int uart_rd(int reg)
{
	return *(volatile unsigned int *)(UART0_BASE + ((unsigned long)reg << UART_SHIFT));
}
static inline void uart_wr(int reg, unsigned int v)
{
	*(volatile unsigned int *)(UART0_BASE + ((unsigned long)reg << UART_SHIFT)) = v;
}
#else
static inline unsigned int uart_rd(int reg)
{
	return *(volatile unsigned char *)(UART0_BASE + reg);
}
static inline void uart_wr(int reg, unsigned int v)
{
	*(volatile unsigned char *)(UART0_BASE + reg) = (unsigned char)v;
}
#endif

#ifdef BOARD_D1
/* Full UART0 bring-up on the Allwinner D1 (sun20i). The FEL boot ROM does
 * NOT guarantee UART0 is clocked/muxed/baud-set, so we do it ourselves.
 * Register map from the D1 user manual. */
#define CCU_BASE       0x02001000UL
#define CCU_UART_BGR   (CCU_BASE + 0x090C) /* UART bus gating & reset */

#define PIO_BASE       0x02000000UL
#define PB_CFG1        (PIO_BASE + 0x0034) /* port B, pins 8..15, 4 bits each */

#define w32(a, v) (*(volatile unsigned int *)(a) = (unsigned int)(v))
#define r32(a)    (*(volatile unsigned int *)(a))

/* DW-8250 register numbers (reg-shift 2 applied by uart_wr/uart_rd). */
#define UART_DLL 0 /* divisor low  (DLAB=1)  */
#define UART_DLH 1 /* divisor high (DLAB=1)  */
#define UART_FCR 2 /* FIFO control (write)   */
#define UART_LCR 3 /* line control           */
#define UART_MCR 4 /* modem control          */
#define UART_MCR_LOOP 0x10 /* internal loopback (TX->RX) */

static void d1_uart0_init(void)
{
	unsigned int v;

	/* 1. Enable UART0: bus gate (bit0) + deassert reset (bit16). */
	v = r32(CCU_UART_BGR);
	v |= (1u << 16) | (1u << 0);
	w32(CCU_UART_BGR, v);

	/* 2. Mux PB8=UART0_TX, PB9=UART0_RX to function 6. */
	v = r32(PB_CFG1);
	v &= ~0xFFu;          /* clear PB8 [3:0] and PB9 [7:4] */
	v |= (6u << 0) | (6u << 4);
	w32(PB_CFG1, v);

	/* 3. Program 115200 8N1. UART src = 24 MHz OSC ->
	 *    divisor = 24000000 / (16 * 115200) ~= 13. */
	uart_wr(UART_LCR, 0x80);        /* DLAB = 1              */
	uart_wr(UART_DLL, 13);
	uart_wr(UART_DLH, 0);
	uart_wr(UART_LCR, 0x03);        /* 8N1, DLAB = 0         */
	uart_wr(UART_FCR, 0x07);        /* enable + clear RX/TX FIFOs */

	/* Defensive: force normal (non-loopback) operation in case the boot
	 * ROM left the modem-control internal-loopback bit set, which would
	 * feed TX straight back into RX. */
	uart_wr(UART_MCR, 0x00);

	/* Drain any stale bytes sitting in the RX FIFO so the first getc()
	 * waits for a real keypress instead of returning garbage. */
	while (uart_rd(UART_LSR) & UART_LSR_DR)
		(void)uart_rd(UART_RBR);
}
#endif

void arch_early_init(void)
{
#ifdef BOARD_D1
	d1_uart0_init();
#endif
	/* QEMU's UART is usable from reset; nothing to do there. */
}

void arch_uart_putc(char c)
{
	while ((uart_rd(UART_LSR) & UART_LSR_THRE) == 0)
		;
	uart_wr(UART_THR, (unsigned char)c);
}

char arch_uart_getc(void)
{
	while ((uart_rd(UART_LSR) & UART_LSR_DR) == 0)
		;
	return (char)(uart_rd(UART_RBR) & 0xff);
}

int arch_uart_rx_ready(void)
{
	return (uart_rd(UART_LSR) & UART_LSR_DR) ? 1 : 0;
}

/* M1: install a minimal machine trap vector (defined in trap.S). */
extern void trap_entry(void);
void arch_set_trap_vector(void)
{
	unsigned long addr = (unsigned long)trap_entry;
	/* mtvec = base | mode(0=direct) */
	__asm__ volatile("csrw mtvec, %0" :: "r"(addr));
}

const char *arch_name(void)
{
#ifdef BOARD_D1
	return "riscv64 (Allwinner D1 / Lichee RV)";
#else
	return "riscv64 (rv64imac, QEMU virt)";
#endif
}

const char *arch_cpu_mode(void)
{
	/* RISC-V has no register that reports the current privilege level to
	 * software (by design). But mhartid is an M-mode-only CSR: if this
	 * code is running and already read it in boot.S without trapping, we
	 * are in M-mode. (A trap here would prove otherwise.) */
	unsigned long id;
	__asm__ volatile("csrr %0, mhartid" : "=r"(id));
	(void)id;
	return "M-mode";
}

void arch_halt(void)
{
	for (;;)
		__asm__ volatile("wfi");
}