/*
 *  Startup Code for MIPS32 XBURST CPU-core
 *
 *  Copyright (c) 2010 Xiangfu Liu <xiangfu@sharism.cc>
 *
 * SPDX-License-Identifier:	GPL-2.0+
 */

#include <config.h>
#include <version.h>
#include <asm/regdef.h>
#include <asm/mipsregs.h>
#include <asm/addrspace.h>
#include <asm/cacheops.h>

	.set noreorder

	.globl _start
	.text
_start:
	/* Initialize $gp */
	bal	1f
	 nop
	.word	_gp
1:
	lw	gp, 0(ra)

	/* Set up temporary stack */
	li	sp, CONFIG_SYS_SDRAM_BASE + CONFIG_SYS_INIT_SP_OFFSET

	la	t9, board_init_f
	jr	t9
	 nop

/*
 * void relocate_code (addr_sp, gd, addr_moni)
 *
 * This "function" does not return, instead it continues in RAM
 * after relocating the monitor code.
 *
 * a0 = addr_sp
 * a1 = gd
 * a2 = destination address
 */
	.globl	relocate_code
	.ent	relocate_code
relocate_code:
	move	sp, a0			# set new stack pointer

	move	s0, a1			# save gd in s0
	move	s2, a2			# save destination address in s2

	li	t0, CONFIG_SYS_MONITOR_BASE
	sub	s1, s2, t0		# s1 <-- relocation offset

	la	t3, in_ram
	lw	t2, -12(t3)		# t2 <-- __image_copy_end
	move	t1, a2

	add	gp, s1			# adjust gp

	/*
	 * t0 = source address
	 * t1 = target address
	 * t2 = source end address
	 */
1:
	lw	t3, 0(t0)
	sw	t3, 0(t1)
	addu	t0, 4
	blt	t0, t2, 1b
	 addu	t1, 4

	/* If caches were enabled, we would have to flush them here. */

	/* flush d-cache */
	li	t0, KSEG0
	addi	t1, t0, CONFIG_SYS_DCACHE_SIZE
2:
	cache	INDEX_WRITEBACK_INV_D, 0(t0)
	bne	t0, t1, 2b
	 addi	t0, CONFIG_SYS_CACHELINE_SIZE

	sync

	/* flush i-cache */
	li	t0, KSEG0
	addi	t1, t0, CONFIG_SYS_ICACHE_SIZE
3:
	cache	INDEX_INVALIDATE_I, 0(t0)
	bne	t0, t1, 3b
	 addi	t0, CONFIG_SYS_CACHELINE_SIZE

	/* Invalidate BTB */
	mfc0	t0, CP0_CONFIG, 7
	nop
	ori	t0, 2
	mtc0	t0, CP0_CONFIG, 7
	nop

	/* Jump to where we've relocated ourselves */
	addi	t0, s2, in_ram - _start
	jr	t0
	 nop

	.word	__rel_dyn_end
	.word	__rel_dyn_start
	.word	__image_copy_end
	.word	_GLOBAL_OFFSET_TABLE_
	.word	num_got_entries

in_ram:
	/*
	 * Now we want to update GOT.
	 *
	 * GOT[0] is reserved. GOT[1] is also reserved for the dynamic object
	 * generated by GNU ld. Skip these reserved entries from relocation.
	 */
	lw	t3, -4(t0)		# t3 <-- num_got_entries
	lw	t8, -8(t0)		# t8 <-- _GLOBAL_OFFSET_TABLE_
	add	t8, s1			# t8 now holds relocated _G_O_T_
	addi	t8, t8, 8		# skipping first two entries
	li	t2, 2
1:
	lw	t1, 0(t8)
	beqz	t1, 2f
	 add	t1, s1
	sw	t1, 0(t8)
2:
	addi	t2, 1
	blt	t2, t3, 1b
	 addi	t8, 4

	/* Update dynamic relocations */
	lw	t1, -16(t0)		# t1 <-- __rel_dyn_start
	lw	t2, -20(t0)		# t2 <-- __rel_dyn_end

	b	2f			# skip first reserved entry
	 addi	t1, 8

1:
	lw	t8, -4(t1)		# t8 <-- relocation info

	li	t3, 3
	bne	t8, t3, 2f		# skip non R_MIPS_REL32 entries
	 nop

	lw	t3, -8(t1)		# t3 <-- location to fix up in FLASH

	lw	t8, 0(t3)		# t8 <-- original pointer
	add	t8, s1			# t8 <-- adjusted pointer

	add	t3, s1			# t3 <-- location to fix up in RAM
	sw	t8, 0(t3)

2:
	blt	t1, t2, 1b
	 addi	t1, 8			# each rel.dyn entry is 8 bytes

	/*
	 * Clear BSS
	 *
	 * GOT is now relocated. Thus __bss_start and __bss_end can be
	 * accessed directly via $gp.
	 */
	la	t1, __bss_start		# t1 <-- __bss_start
	la	t2, __bss_end		# t2 <-- __bss_end

1:
	sw	zero, 0(t1)
	blt	t1, t2, 1b
	 addi	t1, 4

	move	a0, s0			# a0 <-- gd
	la	t9, board_init_r
	jr	t9
	 move	a1, s2

	.end	relocate_code