microblaze: Add support for load/store reversed
Load/store reversed (lwr/swr) are insns that endian translate the sub-word part of the address and byteswap the data lanes. Signed-off-by: Edgar E. Iglesias <edgar.iglesias@petalogix.com>
This commit is contained in:
parent
9ef5535763
commit
9f8beb6636
@ -817,12 +817,35 @@ static inline TCGv *compute_ldst_addr(DisasContext *dc, TCGv *t)
|
||||
return t;
|
||||
}
|
||||
|
||||
static inline void dec_byteswap(DisasContext *dc, TCGv dst, TCGv src, int size)
|
||||
{
|
||||
if (size == 4) {
|
||||
tcg_gen_bswap32_tl(dst, src);
|
||||
} else if (size == 2) {
|
||||
TCGv t = tcg_temp_new();
|
||||
|
||||
/* bswap16 assumes the high bits are zero. */
|
||||
tcg_gen_andi_tl(t, src, 0xffff);
|
||||
tcg_gen_bswap16_tl(dst, t);
|
||||
tcg_temp_free(t);
|
||||
} else {
|
||||
/* Ignore.
|
||||
cpu_abort(dc->env, "Invalid ldst byteswap size %d\n", size);
|
||||
*/
|
||||
}
|
||||
}
|
||||
|
||||
static void dec_load(DisasContext *dc)
|
||||
{
|
||||
TCGv t, *addr;
|
||||
unsigned int size;
|
||||
unsigned int size, rev = 0;
|
||||
|
||||
size = 1 << (dc->opcode & 3);
|
||||
|
||||
if (!dc->type_b) {
|
||||
rev = (dc->ir >> 9) & 1;
|
||||
}
|
||||
|
||||
if (size > 4 && (dc->tb_flags & MSR_EE_FLAG)
|
||||
&& (dc->env->pvr.regs[2] & PVR2_ILL_OPCODE_EXC_MASK)) {
|
||||
tcg_gen_movi_tl(cpu_SR[SR_ESR], ESR_EC_ILLEGAL_OP);
|
||||
@ -830,10 +853,63 @@ static void dec_load(DisasContext *dc)
|
||||
return;
|
||||
}
|
||||
|
||||
LOG_DIS("l %x %d\n", dc->opcode, size);
|
||||
LOG_DIS("l%d%s%s\n", size, dc->type_b ? "i" : "", rev ? "r" : "");
|
||||
|
||||
t_sync_flags(dc);
|
||||
addr = compute_ldst_addr(dc, &t);
|
||||
|
||||
/*
|
||||
* When doing reverse accesses we need to do two things.
|
||||
*
|
||||
* 1. Reverse the address wrt endianess.
|
||||
* 2. Byteswap the data lanes on the way back into the CPU core.
|
||||
*/
|
||||
if (rev && size != 4) {
|
||||
/* Endian reverse the address. t is addr. */
|
||||
switch (size) {
|
||||
case 1:
|
||||
{
|
||||
/* 00 -> 11
|
||||
01 -> 10
|
||||
10 -> 10
|
||||
11 -> 00 */
|
||||
TCGv low = tcg_temp_new();
|
||||
|
||||
/* Force addr into the temp. */
|
||||
if (addr != &t) {
|
||||
t = tcg_temp_new();
|
||||
tcg_gen_mov_tl(t, *addr);
|
||||
addr = &t;
|
||||
}
|
||||
|
||||
tcg_gen_andi_tl(low, t, 3);
|
||||
tcg_gen_sub_tl(low, tcg_const_tl(3), low);
|
||||
tcg_gen_andi_tl(t, t, ~3);
|
||||
tcg_gen_or_tl(t, t, low);
|
||||
tcg_gen_mov_tl(env_debug, low);
|
||||
tcg_gen_mov_tl(env_imm, t);
|
||||
tcg_temp_free(low);
|
||||
break;
|
||||
}
|
||||
|
||||
case 2:
|
||||
/* 00 -> 10
|
||||
10 -> 00. */
|
||||
/* Force addr into the temp. */
|
||||
if (addr != &t) {
|
||||
t = tcg_temp_new();
|
||||
tcg_gen_xori_tl(t, *addr, 2);
|
||||
addr = &t;
|
||||
} else {
|
||||
tcg_gen_xori_tl(t, t, 2);
|
||||
}
|
||||
break;
|
||||
default:
|
||||
cpu_abort(dc->env, "Invalid reverse size\n");
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* If we get a fault on a dslot, the jmpstate better be in sync. */
|
||||
sync_jmpstate(dc);
|
||||
|
||||
@ -852,13 +928,22 @@ static void dec_load(DisasContext *dc)
|
||||
tcg_gen_movi_tl(cpu_SR[SR_PC], dc->pc);
|
||||
gen_helper_memalign(*addr, tcg_const_tl(dc->rd),
|
||||
tcg_const_tl(0), tcg_const_tl(size - 1));
|
||||
if (dc->rd)
|
||||
tcg_gen_mov_tl(cpu_R[dc->rd], v);
|
||||
if (dc->rd) {
|
||||
if (rev) {
|
||||
dec_byteswap(dc, cpu_R[dc->rd], v, size);
|
||||
} else {
|
||||
tcg_gen_mov_tl(cpu_R[dc->rd], v);
|
||||
}
|
||||
}
|
||||
tcg_temp_free(v);
|
||||
} else {
|
||||
if (dc->rd) {
|
||||
gen_load(dc, cpu_R[dc->rd], *addr, size);
|
||||
if (rev) {
|
||||
dec_byteswap(dc, cpu_R[dc->rd], cpu_R[dc->rd], size);
|
||||
}
|
||||
} else {
|
||||
/* We are loading into r0, no need to reverse. */
|
||||
gen_load(dc, env_imm, *addr, size);
|
||||
}
|
||||
}
|
||||
@ -885,9 +970,12 @@ static void gen_store(DisasContext *dc, TCGv addr, TCGv val,
|
||||
static void dec_store(DisasContext *dc)
|
||||
{
|
||||
TCGv t, *addr;
|
||||
unsigned int size;
|
||||
unsigned int size, rev = 0;
|
||||
|
||||
size = 1 << (dc->opcode & 3);
|
||||
if (!dc->type_b) {
|
||||
rev = (dc->ir >> 9) & 1;
|
||||
}
|
||||
|
||||
if (size > 4 && (dc->tb_flags & MSR_EE_FLAG)
|
||||
&& (dc->env->pvr.regs[2] & PVR2_ILL_OPCODE_EXC_MASK)) {
|
||||
@ -896,19 +984,84 @@ static void dec_store(DisasContext *dc)
|
||||
return;
|
||||
}
|
||||
|
||||
LOG_DIS("s%d%s\n", size, dc->type_b ? "i" : "");
|
||||
LOG_DIS("s%d%s%s\n", size, dc->type_b ? "i" : "", rev ? "r" : "");
|
||||
t_sync_flags(dc);
|
||||
/* If we get a fault on a dslot, the jmpstate better be in sync. */
|
||||
sync_jmpstate(dc);
|
||||
addr = compute_ldst_addr(dc, &t);
|
||||
|
||||
gen_store(dc, *addr, cpu_R[dc->rd], size);
|
||||
if (rev && size != 4) {
|
||||
/* Endian reverse the address. t is addr. */
|
||||
switch (size) {
|
||||
case 1:
|
||||
{
|
||||
/* 00 -> 11
|
||||
01 -> 10
|
||||
10 -> 10
|
||||
11 -> 00 */
|
||||
TCGv low = tcg_temp_new();
|
||||
|
||||
/* Force addr into the temp. */
|
||||
if (addr != &t) {
|
||||
t = tcg_temp_new();
|
||||
tcg_gen_mov_tl(t, *addr);
|
||||
addr = &t;
|
||||
}
|
||||
|
||||
tcg_gen_andi_tl(low, t, 3);
|
||||
tcg_gen_sub_tl(low, tcg_const_tl(3), low);
|
||||
tcg_gen_andi_tl(t, t, ~3);
|
||||
tcg_gen_or_tl(t, t, low);
|
||||
tcg_gen_mov_tl(env_debug, low);
|
||||
tcg_gen_mov_tl(env_imm, t);
|
||||
tcg_temp_free(low);
|
||||
break;
|
||||
}
|
||||
|
||||
case 2:
|
||||
/* 00 -> 10
|
||||
10 -> 00. */
|
||||
/* Force addr into the temp. */
|
||||
if (addr != &t) {
|
||||
t = tcg_temp_new();
|
||||
tcg_gen_xori_tl(t, *addr, 2);
|
||||
addr = &t;
|
||||
} else {
|
||||
tcg_gen_xori_tl(t, t, 2);
|
||||
}
|
||||
break;
|
||||
default:
|
||||
cpu_abort(dc->env, "Invalid reverse size\n");
|
||||
break;
|
||||
}
|
||||
|
||||
if (size != 1) {
|
||||
TCGv bs_data = tcg_temp_new();
|
||||
dec_byteswap(dc, bs_data, cpu_R[dc->rd], size);
|
||||
gen_store(dc, *addr, bs_data, size);
|
||||
tcg_temp_free(bs_data);
|
||||
} else {
|
||||
gen_store(dc, *addr, cpu_R[dc->rd], size);
|
||||
}
|
||||
} else {
|
||||
if (rev) {
|
||||
TCGv bs_data = tcg_temp_new();
|
||||
dec_byteswap(dc, bs_data, cpu_R[dc->rd], size);
|
||||
gen_store(dc, *addr, bs_data, size);
|
||||
tcg_temp_free(bs_data);
|
||||
} else {
|
||||
gen_store(dc, *addr, cpu_R[dc->rd], size);
|
||||
}
|
||||
}
|
||||
|
||||
/* Verify alignment if needed. */
|
||||
if ((dc->env->pvr.regs[2] & PVR2_UNALIGNED_EXC_MASK) && size > 1) {
|
||||
tcg_gen_movi_tl(cpu_SR[SR_PC], dc->pc);
|
||||
/* FIXME: if the alignment is wrong, we should restore the value
|
||||
* in memory.
|
||||
* in memory. One possible way to acheive this is to probe
|
||||
* the MMU prior to the memaccess, thay way we could put
|
||||
* the alignment checks in between the probe and the mem
|
||||
* access.
|
||||
*/
|
||||
gen_helper_memalign(*addr, tcg_const_tl(dc->rd),
|
||||
tcg_const_tl(1), tcg_const_tl(size - 1));
|
||||
|
Loading…
Reference in New Issue
Block a user