/*
 * Copyright (C) 2009 Michal Simek <monstr@monstr.eu>
 * Copyright (C) 2009 PetaLogix
 * Copyright (C) 2007 LynuxWorks, Inc.
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 */

#include <linux/errno.h>
#include <linux/linkage.h>
#include <asm/page.h>

/* Loop unrolling for __copy_tofrom_user */
#define COPY(offset)    \
1:      lwi     r4 , r6, 0x0000 + offset;       \
2:      lwi     r19, r6, 0x0004 + offset;       \
3:      lwi     r20, r6, 0x0008 + offset;       \
4:      lwi     r21, r6, 0x000C + offset;       \
5:      lwi     r22, r6, 0x0010 + offset;       \
6:      lwi     r23, r6, 0x0014 + offset;       \
7:      lwi     r24, r6, 0x0018 + offset;       \
8:      lwi     r25, r6, 0x001C + offset;       \
9:      swi     r4 , r5, 0x0000 + offset;       \
10:     swi     r19, r5, 0x0004 + offset;       \
11:     swi     r20, r5, 0x0008 + offset;       \
12:     swi     r21, r5, 0x000C + offset;       \
13:     swi     r22, r5, 0x0010 + offset;       \
14:     swi     r23, r5, 0x0014 + offset;       \
15:     swi     r24, r5, 0x0018 + offset;       \
16:     swi     r25, r5, 0x001C + offset;       \
        .section __ex_table,"a";                \
        .word   1b, 33f;                        \
        .word   2b, 33f;                        \
        .word   3b, 33f;                        \
        .word   4b, 33f;                        \
        .word   5b, 33f;                        \
        .word   6b, 33f;                        \
        .word   7b, 33f;                        \
        .word   8b, 33f;                        \
        .word   9b, 33f;                        \
        .word   10b, 33f;                       \
        .word   11b, 33f;                       \
        .word   12b, 33f;                       \
        .word   13b, 33f;                       \
        .word   14b, 33f;                       \
        .word   15b, 33f;                       \
        .word   16b, 33f;                       \
        .text

#define COPY_80(offset) \
        COPY(0x00 + offset);\
        COPY(0x20 + offset);\
        COPY(0x40 + offset);\
        COPY(0x60 + offset);

/*
 * int __copy_tofrom_user(char *to, char *from, int len)
 * Return:
 *   0 on success
 *   number of not copied bytes on error
 */
        .text
.globl __copy_tofrom_user;
.type  __copy_tofrom_user, @function
.align 4;
__copy_tofrom_user:
        /*
         * r5 - to
         * r6 - from
         * r7, r3 - count
         * r4 - tempval
         */
        beqid   r7, 0f /* zero size is not likely */
        or      r3, r5, r6 /* find if is any to/from unaligned */
        or      r3, r3, r7 /* find if count is unaligned */
        andi    r3, r3, 0x3 /* mask last 3 bits */
        bneid   r3, bu1 /* if r3 is not zero then byte copying */
        or      r3, r0, r0

        rsubi   r3, r7, PAGE_SIZE /* detect PAGE_SIZE */
        beqid   r3, page;
        or      r3, r0, r0

w1:     lw      r4, r6, r3 /* at least one 4 byte copy */
w2:     sw      r4, r5, r3
        addik   r7, r7, -4
        bneid   r7, w1
        addik   r3, r3, 4
        addik   r3, r7, 0
        rtsd    r15, 8
        nop

        .section        __ex_table,"a"
        .word   w1, 0f;
        .word   w2, 0f;
        .text

.align 4 /* Alignment is important to keep icache happy */
page:   /* Create room on stack and save registers for storing values */
        addik   r1, r1, -40
        swi     r5, r1, 0
        swi     r6, r1, 4
        swi     r7, r1, 8
        swi     r19, r1, 12
        swi     r20, r1, 16
        swi     r21, r1, 20
        swi     r22, r1, 24
        swi     r23, r1, 28
        swi     r24, r1, 32
        swi     r25, r1, 36
loop:   /* r4, r19, r20, r21, r22, r23, r24, r25 are used for storing values */
        /* Loop unrolling to get performance boost */
        COPY_80(0x000);
        COPY_80(0x080);
        COPY_80(0x100);
        COPY_80(0x180);
        /* copy loop */
        addik   r6, r6, 0x200
        addik   r7, r7, -0x200
        bneid   r7, loop
        addik   r5, r5, 0x200

        /* Restore register content */
        lwi     r5, r1, 0
        lwi     r6, r1, 4
        lwi     r7, r1, 8
        lwi     r19, r1, 12
        lwi     r20, r1, 16
        lwi     r21, r1, 20
        lwi     r22, r1, 24
        lwi     r23, r1, 28
        lwi     r24, r1, 32
        lwi     r25, r1, 36
        addik   r1, r1, 40
        /* return back */
        addik   r3, r0, 0
        rtsd    r15, 8
        nop

/* Fault case - return temp count */
33:
        addik   r3, r7, 0
        /* Restore register content */
        lwi     r5, r1, 0
        lwi     r6, r1, 4
        lwi     r7, r1, 8
        lwi     r19, r1, 12
        lwi     r20, r1, 16
        lwi     r21, r1, 20
        lwi     r22, r1, 24
        lwi     r23, r1, 28
        lwi     r24, r1, 32
        lwi     r25, r1, 36
        addik   r1, r1, 40
        /* return back */
        rtsd    r15, 8
        nop

.align 4 /* Alignment is important to keep icache happy */
bu1:    lbu     r4,r6,r3
bu2:    sb      r4,r5,r3
        addik   r7,r7,-1
        bneid   r7,bu1
        addik   r3,r3,1         /* delay slot */
0:
        addik   r3,r7,0
        rtsd    r15,8
        nop
        .size   __copy_tofrom_user, . - __copy_tofrom_user

        .section        __ex_table,"a"
        .word   bu1, 0b;
        .word   bu2, 0b;
        .text