/*
 * Copyright 2011 Tilera Corporation. All Rights Reserved.
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation, version 2.
 *
 *   This program is distributed in the hope that it will be useful, but
 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 */

#include <linux/linkage.h>
#include <asm/errno.h>
#include <asm/cache.h>
#include <arch/chip.h>

/* Access user memory, but use MMU to avoid propagating kernel exceptions. */

	.pushsection .fixup,"ax"

get_user_fault:
	{ movei r1, -EFAULT; move r0, zero }
	jrp lr
	ENDPROC(get_user_fault)

put_user_fault:
	{ movei r0, -EFAULT; jrp lr }
	ENDPROC(put_user_fault)

	.popsection

/*
 * __get_user_N functions take a pointer in r0, and return 0 in r1
 * on success, with the value in r0; or else -EFAULT in r1.
 */
#define __get_user_N(bytes, LOAD) \
	STD_ENTRY(__get_user_##bytes); \
1:	{ LOAD r0, r0; move r1, zero }; \
	jrp lr; \
	STD_ENDPROC(__get_user_##bytes); \
	.pushsection __ex_table,"a"; \
	.quad 1b, get_user_fault; \
	.popsection

__get_user_N(1, ld1u)
__get_user_N(2, ld2u)
__get_user_N(4, ld4u)
__get_user_N(8, ld)

/*
 * __put_user_N functions take a value in r0 and a pointer in r1,
 * and return 0 in r0 on success or -EFAULT on failure.
 */
#define __put_user_N(bytes, STORE) \
	STD_ENTRY(__put_user_##bytes); \
1:	{ STORE r1, r0; move r0, zero }; \
	jrp lr; \
	STD_ENDPROC(__put_user_##bytes); \
	.pushsection __ex_table,"a"; \
	.quad 1b, put_user_fault; \
	.popsection

__put_user_N(1, st1)
__put_user_N(2, st2)
__put_user_N(4, st4)
__put_user_N(8, st)

/*
 * strnlen_user_asm takes the pointer in r0, and the length bound in r1.
 * It returns the length, including the terminating NUL, or zero on exception.
 * If length is greater than the bound, returns one plus the bound.
 */
STD_ENTRY(strnlen_user_asm)
	{ beqz r1, 2f; addi r3, r0, -1 }  /* bias down to include NUL */
1:      { ld1u r4, r0; addi r1, r1, -1 }
	beqz r4, 2f
	{ bnezt r1, 1b; addi r0, r0, 1 }
2:      { sub r0, r0, r3; jrp lr }
	STD_ENDPROC(strnlen_user_asm)
	.pushsection .fixup,"ax"
strnlen_user_fault:
	{ move r0, zero; jrp lr }
	ENDPROC(strnlen_user_fault)
	.section __ex_table,"a"
	.quad 1b, strnlen_user_fault
	.popsection

/*
 * strncpy_from_user_asm takes the kernel target pointer in r0,
 * the userspace source pointer in r1, and the length bound (including
 * the trailing NUL) in r2.  On success, it returns the string length
 * (not including the trailing NUL), or -EFAULT on failure.
 */
STD_ENTRY(strncpy_from_user_asm)
	{ beqz r2, 2f; move r3, r0 }
1:      { ld1u r4, r1; addi r1, r1, 1; addi r2, r2, -1 }
	{ st1 r0, r4; addi r0, r0, 1 }
	beqz r2, 2f
	bnezt r4, 1b
	addi r0, r0, -1   /* don't count the trailing NUL */
2:      { sub r0, r0, r3; jrp lr }
	STD_ENDPROC(strncpy_from_user_asm)
	.pushsection .fixup,"ax"
strncpy_from_user_fault:
	{ movei r0, -EFAULT; jrp lr }
	ENDPROC(strncpy_from_user_fault)
	.section __ex_table,"a"
	.quad 1b, strncpy_from_user_fault
	.popsection

/*
 * clear_user_asm takes the user target address in r0 and the
 * number of bytes to zero in r1.
 * It returns the number of uncopiable bytes (hopefully zero) in r0.
 * Note that we don't use a separate .fixup section here since we fall
 * through into the "fixup" code as the last straight-line bundle anyway.
 */
STD_ENTRY(clear_user_asm)
	{ beqz r1, 2f; or r2, r0, r1 }
	andi r2, r2, 7
	beqzt r2, .Lclear_aligned_user_asm
1:      { st1 r0, zero; addi r0, r0, 1; addi r1, r1, -1 }
	bnezt r1, 1b
2:      { move r0, r1; jrp lr }
	.pushsection __ex_table,"a"
	.quad 1b, 2b
	.popsection

.Lclear_aligned_user_asm:
1:      { st r0, zero; addi r0, r0, 8; addi r1, r1, -8 }
	bnezt r1, 1b
2:      { move r0, r1; jrp lr }
	STD_ENDPROC(clear_user_asm)
	.pushsection __ex_table,"a"
	.quad 1b, 2b
	.popsection

/*
 * flush_user_asm takes the user target address in r0 and the
 * number of bytes to flush in r1.
 * It returns the number of unflushable bytes (hopefully zero) in r0.
 */
STD_ENTRY(flush_user_asm)
	beqz r1, 2f
	{ movei r2, L2_CACHE_BYTES; add r1, r0, r1 }
	{ sub r2, zero, r2; addi r1, r1, L2_CACHE_BYTES-1 }
	{ and r0, r0, r2; and r1, r1, r2 }
	{ sub r1, r1, r0 }
1:      { flush r0; addi r1, r1, -CHIP_FLUSH_STRIDE() }
	{ addi r0, r0, CHIP_FLUSH_STRIDE(); bnezt r1, 1b }
2:      { move r0, r1; jrp lr }
	STD_ENDPROC(flush_user_asm)
	.pushsection __ex_table,"a"
	.quad 1b, 2b
	.popsection

/*
 * inv_user_asm takes the user target address in r0 and the
 * number of bytes to invalidate in r1.
 * It returns the number of not inv'able bytes (hopefully zero) in r0.
 */
STD_ENTRY(inv_user_asm)
	beqz r1, 2f
	{ movei r2, L2_CACHE_BYTES; add r1, r0, r1 }
	{ sub r2, zero, r2; addi r1, r1, L2_CACHE_BYTES-1 }
	{ and r0, r0, r2; and r1, r1, r2 }
	{ sub r1, r1, r0 }
1:      { inv r0; addi r1, r1, -CHIP_INV_STRIDE() }
	{ addi r0, r0, CHIP_INV_STRIDE(); bnezt r1, 1b }
2:      { move r0, r1; jrp lr }
	STD_ENDPROC(inv_user_asm)
	.pushsection __ex_table,"a"
	.quad 1b, 2b
	.popsection

/*
 * finv_user_asm takes the user target address in r0 and the
 * number of bytes to flush-invalidate in r1.
 * It returns the number of not finv'able bytes (hopefully zero) in r0.
 */
STD_ENTRY(finv_user_asm)
	beqz r1, 2f
	{ movei r2, L2_CACHE_BYTES; add r1, r0, r1 }
	{ sub r2, zero, r2; addi r1, r1, L2_CACHE_BYTES-1 }
	{ and r0, r0, r2; and r1, r1, r2 }
	{ sub r1, r1, r0 }
1:      { finv r0; addi r1, r1, -CHIP_FINV_STRIDE() }
	{ addi r0, r0, CHIP_FINV_STRIDE(); bnezt r1, 1b }
2:      { move r0, r1; jrp lr }
	STD_ENDPROC(finv_user_asm)
	.pushsection __ex_table,"a"
	.quad 1b, 2b
	.popsection