Subversion Repositories or1k

/*

 * vr4300.S -- CPU specific support routines

 *

 * Copyright (c) 1995,1996 Cygnus Support

 *

 * The authors hereby grant permission to use, copy, modify, distribute,

 * and license this software and its documentation for any purpose, provided

 * that existing copyright notices are retained in all copies and that this

 * notice is included verbatim in any distributions. No written agreement,

 * license, or royalty fee is required for any of the authorized uses.

 * Modifications to this software may be copyrighted by their authors

 * and need not follow the licensing terms described here, provided that

 * the new terms are clearly indicated on the first page of each file where

 * they apply.

 */

#ifndef __mips64

        .set mips3

#endif

#ifdef __mips16

/* This file contains 32 bit assembly code.  */

        .set nomips16

#endif

#include "regs.S"

        .text

        .align  2

        # Taken from "R4300 Preliminary RISC Processor Specification

        # Revision 2.0 January 1995" page 39: "The Count

        # register... increments at a constant rate... at one-half the

        # PClock speed."

        # We can use this fact to provide small polled delays.

        .globl  __cpu_timer_poll

        .ent    __cpu_timer_poll

__cpu_timer_poll:

        .set    noreorder

        # in:   a0 = (unsigned int) number of PClock ticks to wait for

        # out:  void

        # The Vr4300 counter updates at half PClock, so divide by 2 to

        # get counter delta:

        bnezl   a0, 1f          # continue if delta non-zero

        srl     a0, a0, 1       # divide ticks by 2             {DELAY SLOT}

        # perform a quick return to the caller:

        j       ra

        nop                     #                               {DELAY SLOT}

1:

        mfc0    v0, $9          # C0_COUNT:  get current counter value

        nop

        nop

        # We cannot just do the simple test, of adding our delta onto

        # the current value (ignoring overflow) and then checking for

        # equality. The counter is incrementing every two PClocks,

        # which means the counter value can change between

        # instructions, making it hard to sample at the exact value

        # desired.

        # However, we do know that our entry delta value is less than

        # half the number space (since we divide by 2 on entry). This

        # means we can use a difference in signs to indicate timer

        # overflow.

        addu    a0, v0, a0      # unsigned add (ignore overflow)

        # We know have our end value (which will have been

        # sign-extended to fill the 64bit register value).

2:

        # get current counter value:

        mfc0    v0, $9  # C0_COUNT

        nop

        nop

        # This is an unsigned 32bit subtraction:

        subu    v0, a0, v0      # delta = (end - now)           {DELAY SLOT}

        bgtzl   v0, 2b          # looping back is most likely

        nop

        # We have now been delayed (in the foreground) for AT LEAST

        # the required number of counter ticks.

        j       ra              # return to caller

        nop                     #                               {DELAY SLOT}

        .set    reorder

        .end    __cpu_timer_poll

        # Flush the processor caches to memory:

        .globl  __cpu_flush

        .ent    __cpu_flush

__cpu_flush:

        .set    noreorder

        # NOTE: The Vr4300 *CANNOT* have any secondary cache (bit 17

        # of the CONFIG registered is hard-wired to 1). We just

        # provide code to flush the Data and Instruction caches.

        # Even though the Vr4300 has hard-wired cache and cache line

        # sizes, we still interpret the relevant Config register

        # bits. This allows this code to be used for other conforming

        # MIPS architectures if desired.

        # Get the config register

        mfc0    a0, C0_CONFIG

        nop

        nop

        li      a1, 1           # a useful constant

        #

        srl     a2, a0, 9       # bits 11..9 for instruction cache size

        andi    a2, a2, 0x7     # 3bits of information

        add     a2, a2, 12      # get full power-of-2 value

        sllv    a2, a1, a2      # instruction cache size

        #

        srl     a3, a0, 6       # bits 8..6 for data cache size

        andi    a3, a3, 0x7     # 3bits of information

        add     a3, a3, 12      # get full power-of-2 value

        sllv    a3, a1, a3      # data cache size

        #

        li      a1, (1 << 5)    # check IB (instruction cache line size)

        and     a1, a0, a1      # mask against the CONFIG register value

        beqz    a1, 1f          # branch on result of delay slot operation

        nop

        li      a1, 32          # non-zero, then 32bytes

        j       2f              # continue

        nop

1:

        li      a1, 16          # 16bytes

2:

        #

        li      t0, (1 << 4)    # check DB (data cache line size)

        and     a0, a0, t0      # mask against the CONFIG register value

        beqz    a0, 3f          # branch on result of delay slot operation

        nop

        li      a0, 32          # non-zero, then 32bytes

        j       4f              # continue

        nop

3:

        li      a0, 16          # 16bytes

4:

        #

        # a0 = data cache line size

        # a1 = instruction cache line size

        # a2 = instruction cache size

        # a3 = data cache size

        #

        lui     t0, ((K0BASE >> 16) & 0xFFFF)

        ori     t0, t0, (K0BASE & 0xFFFF)

        addu    t1, t0, a2      # end cache address

        subu    t2, a1, 1       # line size mask

        not     t2              # invert the mask

        and     t3, t0, t2      # get start address

        addu    t1, -1

        and     t1, t2          # get end address

5:

        cache   INDEX_INVALIDATE_I,0(t3)

        bne     t3, t1, 5b

        addu    t3, a1

        #

        addu    t1, t0, a3      # end cache address

        subu    t2, a0, 1       # line size mask

        not     t2              # invert the mask

        and     t3, t0, t2      # get start address

        addu    t1, -1

        and     t1, t2          # get end address

6:

        cache   INDEX_WRITEBACK_INVALIDATE_D,0(t3)

        bne     t3, t1, 6b

        addu    t3, a0

        #

        j       ra      # return to the caller

        nop

        .set    reorder

        .end    __cpu_flush

        # NOTE: This variable should *NOT* be addressed relative to

        # the $gp register since this code is executed before $gp is

        # initialised... hence we leave it in the text area. This will

        # cause problems if this routine is ever ROMmed:

        .globl  __buserr_cnt

__buserr_cnt:

        .word   0

        .align  3

__k1_save:

        .word   0

        .word   0

        .align  2

        .ent __buserr

        .globl __buserr

__buserr:

        .set noat

        .set noreorder

        # k0 and k1 available for use:

        mfc0    k0,C0_CAUSE

        nop

        nop

        andi    k0,k0,0x7c

        sub     k0,k0,7 << 2

        beq     k0,$0,__buserr_do

        nop

        # call the previous handler

        la      k0,__previous

        jr      k0

        nop

        #

__buserr_do:

        # TODO: check that the cause is indeed a bus error

        # - if not then just jump to the previous handler

        la      k0,__k1_save

        sd      k1,0(k0)

        #

        la      k1,__buserr_cnt

        lw      k0,0(k1)        # increment counter

        addu    k0,1

        sw      k0,0(k1)

        #

        la      k0,__k1_save

        ld      k1,0(k0)

        #

        mfc0    k0,C0_EPC

        nop

        nop

        addu    k0,k0,4         # skip offending instruction

        mtc0    k0,C0_EPC       # update EPC

        nop

        nop

        eret

#        j       k0

#        rfe

        .set reorder

        .set at

        .end __buserr

__exception_code:

        .set noreorder

        lui     k0,%hi(__buserr)

        daddiu  k0,k0,%lo(__buserr)

        jr      k0

        nop

        .set reorder

__exception_code_end:

        .data

__previous:

        .space  (__exception_code_end - __exception_code)

        # This subtracting two addresses is working

        # but is not garenteed to continue working.

        # The assemble reserves the right to put these

        # two labels into different frags, and then

        # cant take their difference.

        .text

        .ent    __default_buserr_handler

        .globl  __default_buserr_handler

__default_buserr_handler:

        .set noreorder

        # attach our simple bus error handler:

        # in:  void

        # out: void

        mfc0    a0,C0_SR

        nop

        li      a1,SR_BEV

        and     a1,a1,a0

        beq     a1,$0,baseaddr

        lui     a0,0x8000       # delay slot

        lui     a0,0xbfc0

        daddiu  a0,a0,0x0200

baseaddr:

        daddiu  a0,a0,0x0180

        # a0 = base vector table address

        la      a1,__exception_code_end

        la      a2,__exception_code

        subu    a1,a1,a2

        la      a3,__previous

        # there must be a better way of doing this????

copyloop:

        lw      v0,0(a0)

        sw      v0,0(a3)

        lw      v0,0(a2)

        sw      v0,0(a0)

        daddiu  a0,a0,4

        daddiu  a2,a2,4

        daddiu  a3,a3,4

        subu    a1,a1,4

        bne     a1,$0,copyloop

        nop

        la      a0,__buserr_cnt

        sw      $0,0(a0)

        j       ra

        nop

        .set reorder

        .end    __default_buserr_handler

        .ent    __restore_buserr_handler

        .globl  __restore_buserr_handler

__restore_buserr_handler:

        .set noreorder

        # restore original (monitor) bus error handler

        # in:  void

        # out: void

        mfc0    a0,C0_SR

        nop

        li      a1,SR_BEV

        and     a1,a1,a0

        beq     a1,$0,res_baseaddr

        lui     a0,0x8000       # delay slot

        lui     a0,0xbfc0

        daddiu  a0,a0,0x0200

res_baseaddr:

        daddiu  a0,a0,0x0180

        # a0 = base vector table address

        la      a1,__exception_code_end

        la      a3,__exception_code

        subu    a1,a1,a3

        la      a3,__previous

        # there must be a better way of doing this????

res_copyloop:

        lw      v0,0(a3)

        sw      v0,0(a0)

        daddiu  a0,a0,4

        daddiu  a3,a3,4

        subu    a1,a1,4

        bne     a1,$0,res_copyloop

        nop

        j       ra

        nop

        .set reorder

        .end    __restore_buserr_handler

        .ent    __buserr_count

        .globl  __buserr_count

__buserr_count:

        .set noreorder

        # restore original (monitor) bus error handler

        # in:  void

        # out: unsigned int __buserr_cnt

        la      v0,__buserr_cnt

        lw      v0,0(v0)

        j       ra

        nop

        .set reorder

        .end    __buserr_count

/* EOF vr4300.S */