linux/arch/m32r/kernel/align.c

585 lines
12 KiB
C
Raw Normal View History

/*
* align.c - address exception handler for M32R
*
* Copyright (c) 2003 Hitoshi Yamamoto
*/
#include <asm/ptrace.h>
#include <linux/uaccess.h>
static int get_reg(struct pt_regs *regs, int nr)
{
int val;
if (nr < 4)
val = *(unsigned long *)(&regs->r0 + nr);
else if (nr < 7)
val = *(unsigned long *)(&regs->r4 + (nr - 4));
else if (nr < 13)
val = *(unsigned long *)(&regs->r7 + (nr - 7));
else
val = *(unsigned long *)(&regs->fp + (nr - 13));
return val;
}
static void set_reg(struct pt_regs *regs, int nr, int val)
{
if (nr < 4)
*(unsigned long *)(&regs->r0 + nr) = val;
else if (nr < 7)
*(unsigned long *)(&regs->r4 + (nr - 4)) = val;
else if (nr < 13)
*(unsigned long *)(&regs->r7 + (nr - 7)) = val;
else
*(unsigned long *)(&regs->fp + (nr - 13)) = val;
}
#define REG1(insn) (((insn) & 0x0f00) >> 8)
#define REG2(insn) ((insn) & 0x000f)
#define PSW_BC 0x100
/* O- instruction */
#define ISA_LD1 0x20c0 /* ld Rdest, @Rsrc */
#define ISA_LD2 0x20e0 /* ld Rdest, @Rsrc+ */
#define ISA_LDH 0x20a0 /* ldh Rdest, @Rsrc */
#define ISA_LDUH 0x20b0 /* lduh Rdest, @Rsrc */
#define ISA_ST1 0x2040 /* st Rsrc1, @Rsrc2 */
#define ISA_ST2 0x2060 /* st Rsrc1, @+Rsrc2 */
#define ISA_ST3 0x2070 /* st Rsrc1, @-Rsrc2 */
#define ISA_STH1 0x2020 /* sth Rsrc1, @Rsrc2 */
#define ISA_STH2 0x2030 /* sth Rsrc1, @Rsrc2+ */
#ifdef CONFIG_ISA_DUAL_ISSUE
/* OS instruction */
#define ISA_ADD 0x00a0 /* add Rdest, Rsrc */
#define ISA_ADDI 0x4000 /* addi Rdest, #imm8 */
#define ISA_ADDX 0x0090 /* addx Rdest, Rsrc */
#define ISA_AND 0x00c0 /* and Rdest, Rsrc */
#define ISA_CMP 0x0040 /* cmp Rsrc1, Rsrc2 */
#define ISA_CMPEQ 0x0060 /* cmpeq Rsrc1, Rsrc2 */
#define ISA_CMPU 0x0050 /* cmpu Rsrc1, Rsrc2 */
#define ISA_CMPZ 0x0070 /* cmpz Rsrc */
#define ISA_LDI 0x6000 /* ldi Rdest, #imm8 */
#define ISA_MV 0x1080 /* mv Rdest, Rsrc */
#define ISA_NEG 0x0030 /* neg Rdest, Rsrc */
#define ISA_NOP 0x7000 /* nop */
#define ISA_NOT 0x00b0 /* not Rdest, Rsrc */
#define ISA_OR 0x00e0 /* or Rdest, Rsrc */
#define ISA_SUB 0x0020 /* sub Rdest, Rsrc */
#define ISA_SUBX 0x0010 /* subx Rdest, Rsrc */
#define ISA_XOR 0x00d0 /* xor Rdest, Rsrc */
/* -S instruction */
#define ISA_MUL 0x1060 /* mul Rdest, Rsrc */
#define ISA_MULLO_A0 0x3010 /* mullo Rsrc1, Rsrc2, A0 */
#define ISA_MULLO_A1 0x3090 /* mullo Rsrc1, Rsrc2, A1 */
#define ISA_MVFACMI_A0 0x50f2 /* mvfacmi Rdest, A0 */
#define ISA_MVFACMI_A1 0x50f6 /* mvfacmi Rdest, A1 */
static int emu_addi(unsigned short insn, struct pt_regs *regs)
{
char imm = (char)(insn & 0xff);
int dest = REG1(insn);
int val;
val = get_reg(regs, dest);
val += imm;
set_reg(regs, dest, val);
return 0;
}
static int emu_ldi(unsigned short insn, struct pt_regs *regs)
{
char imm = (char)(insn & 0xff);
set_reg(regs, REG1(insn), (int)imm);
return 0;
}
static int emu_add(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
int src = REG2(insn);
int val;
val = get_reg(regs, dest);
val += get_reg(regs, src);
set_reg(regs, dest, val);
return 0;
}
static int emu_addx(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
unsigned int val, tmp;
val = regs->psw & PSW_BC ? 1 : 0;
tmp = get_reg(regs, dest);
val += tmp;
val += (unsigned int)get_reg(regs, REG2(insn));
set_reg(regs, dest, val);
/* C bit set */
if (val < tmp)
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_and(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
int val;
val = get_reg(regs, dest);
val &= get_reg(regs, REG2(insn));
set_reg(regs, dest, val);
return 0;
}
static int emu_cmp(unsigned short insn, struct pt_regs *regs)
{
if (get_reg(regs, REG1(insn)) < get_reg(regs, REG2(insn)))
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_cmpeq(unsigned short insn, struct pt_regs *regs)
{
if (get_reg(regs, REG1(insn)) == get_reg(regs, REG2(insn)))
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_cmpu(unsigned short insn, struct pt_regs *regs)
{
if ((unsigned int)get_reg(regs, REG1(insn))
< (unsigned int)get_reg(regs, REG2(insn)))
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_cmpz(unsigned short insn, struct pt_regs *regs)
{
if (!get_reg(regs, REG2(insn)))
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_mv(unsigned short insn, struct pt_regs *regs)
{
int val;
val = get_reg(regs, REG2(insn));
set_reg(regs, REG1(insn), val);
return 0;
}
static int emu_neg(unsigned short insn, struct pt_regs *regs)
{
int val;
val = get_reg(regs, REG2(insn));
set_reg(regs, REG1(insn), 0 - val);
return 0;
}
static int emu_not(unsigned short insn, struct pt_regs *regs)
{
int val;
val = get_reg(regs, REG2(insn));
set_reg(regs, REG1(insn), ~val);
return 0;
}
static int emu_or(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
int val;
val = get_reg(regs, dest);
val |= get_reg(regs, REG2(insn));
set_reg(regs, dest, val);
return 0;
}
static int emu_sub(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
int val;
val = get_reg(regs, dest);
val -= get_reg(regs, REG2(insn));
set_reg(regs, dest, val);
return 0;
}
static int emu_subx(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
unsigned int val, tmp;
val = tmp = get_reg(regs, dest);
val -= (unsigned int)get_reg(regs, REG2(insn));
val -= regs->psw & PSW_BC ? 1 : 0;
set_reg(regs, dest, val);
/* C bit set */
if (val > tmp)
regs->psw |= PSW_BC;
else
regs->psw &= ~(PSW_BC);
return 0;
}
static int emu_xor(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
unsigned int val;
val = (unsigned int)get_reg(regs, dest);
val ^= (unsigned int)get_reg(regs, REG2(insn));
set_reg(regs, dest, val);
return 0;
}
static int emu_mul(unsigned short insn, struct pt_regs *regs)
{
int dest = REG1(insn);
int reg1, reg2;
reg1 = get_reg(regs, dest);
reg2 = get_reg(regs, REG2(insn));
__asm__ __volatile__ (
"mul %0, %1; \n\t"
: "+r" (reg1) : "r" (reg2)
);
set_reg(regs, dest, reg1);
return 0;
}
static int emu_mullo_a0(unsigned short insn, struct pt_regs *regs)
{
int reg1, reg2;
reg1 = get_reg(regs, REG1(insn));
reg2 = get_reg(regs, REG2(insn));
__asm__ __volatile__ (
"mullo %0, %1, a0; \n\t"
"mvfachi %0, a0; \n\t"
"mvfaclo %1, a0; \n\t"
: "+r" (reg1), "+r" (reg2)
);
regs->acc0h = reg1;
regs->acc0l = reg2;
return 0;
}
static int emu_mullo_a1(unsigned short insn, struct pt_regs *regs)
{
int reg1, reg2;
reg1 = get_reg(regs, REG1(insn));
reg2 = get_reg(regs, REG2(insn));
__asm__ __volatile__ (
"mullo %0, %1, a0; \n\t"
"mvfachi %0, a0; \n\t"
"mvfaclo %1, a0; \n\t"
: "+r" (reg1), "+r" (reg2)
);
regs->acc1h = reg1;
regs->acc1l = reg2;
return 0;
}
static int emu_mvfacmi_a0(unsigned short insn, struct pt_regs *regs)
{
unsigned long val;
val = (regs->acc0h << 16) | (regs->acc0l >> 16);
set_reg(regs, REG1(insn), (int)val);
return 0;
}
static int emu_mvfacmi_a1(unsigned short insn, struct pt_regs *regs)
{
unsigned long val;
val = (regs->acc1h << 16) | (regs->acc1l >> 16);
set_reg(regs, REG1(insn), (int)val);
return 0;
}
static int emu_m32r2(unsigned short insn, struct pt_regs *regs)
{
int res = -1;
if ((insn & 0x7fff) == ISA_NOP) /* nop */
return 0;
switch(insn & 0x7000) {
case ISA_ADDI: /* addi Rdest, #imm8 */
res = emu_addi(insn, regs);
break;
case ISA_LDI: /* ldi Rdest, #imm8 */
res = emu_ldi(insn, regs);
break;
default:
break;
}
if (!res)
return 0;
switch(insn & 0x70f0) {
case ISA_ADD: /* add Rdest, Rsrc */
res = emu_add(insn, regs);
break;
case ISA_ADDX: /* addx Rdest, Rsrc */
res = emu_addx(insn, regs);
break;
case ISA_AND: /* and Rdest, Rsrc */
res = emu_and(insn, regs);
break;
case ISA_CMP: /* cmp Rsrc1, Rsrc2 */
res = emu_cmp(insn, regs);
break;
case ISA_CMPEQ: /* cmpeq Rsrc1, Rsrc2 */
res = emu_cmpeq(insn, regs);
break;
case ISA_CMPU: /* cmpu Rsrc1, Rsrc2 */
res = emu_cmpu(insn, regs);
break;
case ISA_CMPZ: /* cmpz Rsrc */
res = emu_cmpz(insn, regs);
break;
case ISA_MV: /* mv Rdest, Rsrc */
res = emu_mv(insn, regs);
break;
case ISA_NEG: /* neg Rdest, Rsrc */
res = emu_neg(insn, regs);
break;
case ISA_NOT: /* not Rdest, Rsrc */
res = emu_not(insn, regs);
break;
case ISA_OR: /* or Rdest, Rsrc */
res = emu_or(insn, regs);
break;
case ISA_SUB: /* sub Rdest, Rsrc */
res = emu_sub(insn, regs);
break;
case ISA_SUBX: /* subx Rdest, Rsrc */
res = emu_subx(insn, regs);
break;
case ISA_XOR: /* xor Rdest, Rsrc */
res = emu_xor(insn, regs);
break;
case ISA_MUL: /* mul Rdest, Rsrc */
res = emu_mul(insn, regs);
break;
case ISA_MULLO_A0: /* mullo Rsrc1, Rsrc2 */
res = emu_mullo_a0(insn, regs);
break;
case ISA_MULLO_A1: /* mullo Rsrc1, Rsrc2 */
res = emu_mullo_a1(insn, regs);
break;
default:
break;
}
if (!res)
return 0;
switch(insn & 0x70ff) {
case ISA_MVFACMI_A0: /* mvfacmi Rdest */
res = emu_mvfacmi_a0(insn, regs);
break;
case ISA_MVFACMI_A1: /* mvfacmi Rdest */
res = emu_mvfacmi_a1(insn, regs);
break;
default:
break;
}
return res;
}
#endif /* CONFIG_ISA_DUAL_ISSUE */
/*
* ld : ?010 dest 1100 src
* 0010 dest 1110 src : ld Rdest, @Rsrc+
* ldh : ?010 dest 1010 src
* lduh : ?010 dest 1011 src
* st : ?010 src1 0100 src2
* 0010 src1 0110 src2 : st Rsrc1, @+Rsrc2
* 0010 src1 0111 src2 : st Rsrc1, @-Rsrc2
* sth : ?010 src1 0010 src2
*/
static int insn_check(unsigned long insn, struct pt_regs *regs,
unsigned char **ucp)
{
int res = 0;
/*
* 32bit insn
* ld Rdest, @(disp16, Rsrc)
* st Rdest, @(disp16, Rsrc)
*/
if (insn & 0x80000000) { /* 32bit insn */
*ucp += (short)(insn & 0x0000ffff);
regs->bpc += 4;
} else { /* 16bit insn */
#ifdef CONFIG_ISA_DUAL_ISSUE
/* parallel exec check */
if (!(regs->bpc & 0x2) && insn & 0x8000) {
res = emu_m32r2((unsigned short)insn, regs);
regs->bpc += 4;
} else
#endif /* CONFIG_ISA_DUAL_ISSUE */
regs->bpc += 2;
}
return res;
}
static int emu_ld(unsigned long insn32, struct pt_regs *regs)
{
unsigned char *ucp;
unsigned long val;
unsigned short insn16;
int size, src;
insn16 = insn32 >> 16;
src = REG2(insn16);
ucp = (unsigned char *)get_reg(regs, src);
if (insn_check(insn32, regs, &ucp))
return -1;
size = insn16 & 0x0040 ? 4 : 2;
if (copy_from_user(&val, ucp, size))
return -1;
if (size == 2)
val >>= 16;
/* ldh sign check */
if ((insn16 & 0x00f0) == 0x00a0 && (val & 0x8000))
val |= 0xffff0000;
set_reg(regs, REG1(insn16), val);
/* ld increment check */
if ((insn16 & 0xf0f0) == ISA_LD2) /* ld Rdest, @Rsrc+ */
set_reg(regs, src, (unsigned long)(ucp + 4));
return 0;
}
static int emu_st(unsigned long insn32, struct pt_regs *regs)
{
unsigned char *ucp;
unsigned long val;
unsigned short insn16;
int size, src2;
insn16 = insn32 >> 16;
src2 = REG2(insn16);
ucp = (unsigned char *)get_reg(regs, src2);
if (insn_check(insn32, regs, &ucp))
return -1;
size = insn16 & 0x0040 ? 4 : 2;
val = get_reg(regs, REG1(insn16));
if (size == 2)
val <<= 16;
/* st inc/dec check */
if ((insn16 & 0xf0e0) == 0x2060) {
if (insn16 & 0x0010)
ucp -= 4;
else
ucp += 4;
set_reg(regs, src2, (unsigned long)ucp);
}
if (copy_to_user(ucp, &val, size))
return -1;
/* sth inc check */
if ((insn16 & 0xf0f0) == ISA_STH2) {
ucp += 2;
set_reg(regs, src2, (unsigned long)ucp);
}
return 0;
}
int handle_unaligned_access(unsigned long insn32, struct pt_regs *regs)
{
unsigned short insn16;
int res;
insn16 = insn32 >> 16;
/* ld or st check */
if ((insn16 & 0x7000) != 0x2000)
return -1;
/* insn alignment check */
if ((insn16 & 0x8000) && (regs->bpc & 3))
return -1;
if (insn16 & 0x0080) /* ld */
res = emu_ld(insn32, regs);
else /* st */
res = emu_st(insn32, regs);
return res;
}