target/ppc: Add extra float instructions to POWER5P processors
[qemu.git] / target / ppc / mem_helper.c
blob39945d9ea585e4c9db6967b63afac6bc4f492728
1 /*
2 * PowerPC memory access emulation helpers for QEMU.
4 * Copyright (c) 2003-2007 Jocelyn Mayer
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "cpu.h"
22 #include "exec/exec-all.h"
23 #include "qemu/host-utils.h"
24 #include "qemu/main-loop.h"
25 #include "exec/helper-proto.h"
26 #include "helper_regs.h"
27 #include "exec/cpu_ldst.h"
28 #include "internal.h"
29 #include "qemu/atomic128.h"
31 /* #define DEBUG_OP */
33 static inline bool needs_byteswap(const CPUPPCState *env)
35 #if defined(TARGET_WORDS_BIGENDIAN)
36 return msr_le;
37 #else
38 return !msr_le;
39 #endif
42 /*****************************************************************************/
43 /* Memory load and stores */
45 static inline target_ulong addr_add(CPUPPCState *env, target_ulong addr,
46 target_long arg)
48 #if defined(TARGET_PPC64)
49 if (!msr_is_64bit(env, env->msr)) {
50 return (uint32_t)(addr + arg);
51 } else
52 #endif
54 return addr + arg;
58 static void *probe_contiguous(CPUPPCState *env, target_ulong addr, uint32_t nb,
59 MMUAccessType access_type, int mmu_idx,
60 uintptr_t raddr)
62 void *host1, *host2;
63 uint32_t nb_pg1, nb_pg2;
65 nb_pg1 = -(addr | TARGET_PAGE_MASK);
66 if (likely(nb <= nb_pg1)) {
67 /* The entire operation is on a single page. */
68 return probe_access(env, addr, nb, access_type, mmu_idx, raddr);
71 /* The operation spans two pages. */
72 nb_pg2 = nb - nb_pg1;
73 host1 = probe_access(env, addr, nb_pg1, access_type, mmu_idx, raddr);
74 addr = addr_add(env, addr, nb_pg1);
75 host2 = probe_access(env, addr, nb_pg2, access_type, mmu_idx, raddr);
77 /* If the two host pages are contiguous, optimize. */
78 if (host2 == host1 + nb_pg1) {
79 return host1;
81 return NULL;
84 void helper_lmw(CPUPPCState *env, target_ulong addr, uint32_t reg)
86 uintptr_t raddr = GETPC();
87 int mmu_idx = cpu_mmu_index(env, false);
88 void *host = probe_contiguous(env, addr, (32 - reg) * 4,
89 MMU_DATA_LOAD, mmu_idx, raddr);
91 if (likely(host)) {
92 /* Fast path -- the entire operation is in RAM at host. */
93 for (; reg < 32; reg++) {
94 env->gpr[reg] = (uint32_t)ldl_be_p(host);
95 host += 4;
97 } else {
98 /* Slow path -- at least some of the operation requires i/o. */
99 for (; reg < 32; reg++) {
100 env->gpr[reg] = cpu_ldl_mmuidx_ra(env, addr, mmu_idx, raddr);
101 addr = addr_add(env, addr, 4);
106 void helper_stmw(CPUPPCState *env, target_ulong addr, uint32_t reg)
108 uintptr_t raddr = GETPC();
109 int mmu_idx = cpu_mmu_index(env, false);
110 void *host = probe_contiguous(env, addr, (32 - reg) * 4,
111 MMU_DATA_STORE, mmu_idx, raddr);
113 if (likely(host)) {
114 /* Fast path -- the entire operation is in RAM at host. */
115 for (; reg < 32; reg++) {
116 stl_be_p(host, env->gpr[reg]);
117 host += 4;
119 } else {
120 /* Slow path -- at least some of the operation requires i/o. */
121 for (; reg < 32; reg++) {
122 cpu_stl_mmuidx_ra(env, addr, env->gpr[reg], mmu_idx, raddr);
123 addr = addr_add(env, addr, 4);
128 static void do_lsw(CPUPPCState *env, target_ulong addr, uint32_t nb,
129 uint32_t reg, uintptr_t raddr)
131 int mmu_idx;
132 void *host;
133 uint32_t val;
135 if (unlikely(nb == 0)) {
136 return;
139 mmu_idx = cpu_mmu_index(env, false);
140 host = probe_contiguous(env, addr, nb, MMU_DATA_LOAD, mmu_idx, raddr);
142 if (likely(host)) {
143 /* Fast path -- the entire operation is in RAM at host. */
144 for (; nb > 3; nb -= 4) {
145 env->gpr[reg] = (uint32_t)ldl_be_p(host);
146 reg = (reg + 1) % 32;
147 host += 4;
149 switch (nb) {
150 default:
151 return;
152 case 1:
153 val = ldub_p(host) << 24;
154 break;
155 case 2:
156 val = lduw_be_p(host) << 16;
157 break;
158 case 3:
159 val = (lduw_be_p(host) << 16) | (ldub_p(host + 2) << 8);
160 break;
162 } else {
163 /* Slow path -- at least some of the operation requires i/o. */
164 for (; nb > 3; nb -= 4) {
165 env->gpr[reg] = cpu_ldl_mmuidx_ra(env, addr, mmu_idx, raddr);
166 reg = (reg + 1) % 32;
167 addr = addr_add(env, addr, 4);
169 switch (nb) {
170 default:
171 return;
172 case 1:
173 val = cpu_ldub_mmuidx_ra(env, addr, mmu_idx, raddr) << 24;
174 break;
175 case 2:
176 val = cpu_lduw_mmuidx_ra(env, addr, mmu_idx, raddr) << 16;
177 break;
178 case 3:
179 val = cpu_lduw_mmuidx_ra(env, addr, mmu_idx, raddr) << 16;
180 addr = addr_add(env, addr, 2);
181 val |= cpu_ldub_mmuidx_ra(env, addr, mmu_idx, raddr) << 8;
182 break;
185 env->gpr[reg] = val;
188 void helper_lsw(CPUPPCState *env, target_ulong addr,
189 uint32_t nb, uint32_t reg)
191 do_lsw(env, addr, nb, reg, GETPC());
195 * PPC32 specification says we must generate an exception if rA is in
196 * the range of registers to be loaded. In an other hand, IBM says
197 * this is valid, but rA won't be loaded. For now, I'll follow the
198 * spec...
200 void helper_lswx(CPUPPCState *env, target_ulong addr, uint32_t reg,
201 uint32_t ra, uint32_t rb)
203 if (likely(xer_bc != 0)) {
204 int num_used_regs = DIV_ROUND_UP(xer_bc, 4);
205 if (unlikely((ra != 0 && lsw_reg_in_range(reg, num_used_regs, ra)) ||
206 lsw_reg_in_range(reg, num_used_regs, rb))) {
207 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
208 POWERPC_EXCP_INVAL |
209 POWERPC_EXCP_INVAL_LSWX, GETPC());
210 } else {
211 do_lsw(env, addr, xer_bc, reg, GETPC());
216 void helper_stsw(CPUPPCState *env, target_ulong addr, uint32_t nb,
217 uint32_t reg)
219 uintptr_t raddr = GETPC();
220 int mmu_idx;
221 void *host;
222 uint32_t val;
224 if (unlikely(nb == 0)) {
225 return;
228 mmu_idx = cpu_mmu_index(env, false);
229 host = probe_contiguous(env, addr, nb, MMU_DATA_STORE, mmu_idx, raddr);
231 if (likely(host)) {
232 /* Fast path -- the entire operation is in RAM at host. */
233 for (; nb > 3; nb -= 4) {
234 stl_be_p(host, env->gpr[reg]);
235 reg = (reg + 1) % 32;
236 host += 4;
238 val = env->gpr[reg];
239 switch (nb) {
240 case 1:
241 stb_p(host, val >> 24);
242 break;
243 case 2:
244 stw_be_p(host, val >> 16);
245 break;
246 case 3:
247 stw_be_p(host, val >> 16);
248 stb_p(host + 2, val >> 8);
249 break;
251 } else {
252 for (; nb > 3; nb -= 4) {
253 cpu_stl_mmuidx_ra(env, addr, env->gpr[reg], mmu_idx, raddr);
254 reg = (reg + 1) % 32;
255 addr = addr_add(env, addr, 4);
257 val = env->gpr[reg];
258 switch (nb) {
259 case 1:
260 cpu_stb_mmuidx_ra(env, addr, val >> 24, mmu_idx, raddr);
261 break;
262 case 2:
263 cpu_stw_mmuidx_ra(env, addr, val >> 16, mmu_idx, raddr);
264 break;
265 case 3:
266 cpu_stw_mmuidx_ra(env, addr, val >> 16, mmu_idx, raddr);
267 addr = addr_add(env, addr, 2);
268 cpu_stb_mmuidx_ra(env, addr, val >> 8, mmu_idx, raddr);
269 break;
274 static void dcbz_common(CPUPPCState *env, target_ulong addr,
275 uint32_t opcode, bool epid, uintptr_t retaddr)
277 target_ulong mask, dcbz_size = env->dcache_line_size;
278 uint32_t i;
279 void *haddr;
280 int mmu_idx = epid ? PPC_TLB_EPID_STORE : cpu_mmu_index(env, false);
282 #if defined(TARGET_PPC64)
283 /* Check for dcbz vs dcbzl on 970 */
284 if (env->excp_model == POWERPC_EXCP_970 &&
285 !(opcode & 0x00200000) && ((env->spr[SPR_970_HID5] >> 7) & 0x3) == 1) {
286 dcbz_size = 32;
288 #endif
290 /* Align address */
291 mask = ~(dcbz_size - 1);
292 addr &= mask;
294 /* Check reservation */
295 if ((env->reserve_addr & mask) == addr) {
296 env->reserve_addr = (target_ulong)-1ULL;
299 /* Try fast path translate */
300 haddr = probe_write(env, addr, dcbz_size, mmu_idx, retaddr);
301 if (haddr) {
302 memset(haddr, 0, dcbz_size);
303 } else {
304 /* Slow path */
305 for (i = 0; i < dcbz_size; i += 8) {
306 cpu_stq_mmuidx_ra(env, addr + i, 0, mmu_idx, retaddr);
311 void helper_dcbz(CPUPPCState *env, target_ulong addr, uint32_t opcode)
313 dcbz_common(env, addr, opcode, false, GETPC());
316 void helper_dcbzep(CPUPPCState *env, target_ulong addr, uint32_t opcode)
318 dcbz_common(env, addr, opcode, true, GETPC());
321 void helper_icbi(CPUPPCState *env, target_ulong addr)
323 addr &= ~(env->dcache_line_size - 1);
325 * Invalidate one cache line :
326 * PowerPC specification says this is to be treated like a load
327 * (not a fetch) by the MMU. To be sure it will be so,
328 * do the load "by hand".
330 cpu_ldl_data_ra(env, addr, GETPC());
333 void helper_icbiep(CPUPPCState *env, target_ulong addr)
335 #if !defined(CONFIG_USER_ONLY)
336 /* See comments above */
337 addr &= ~(env->dcache_line_size - 1);
338 cpu_ldl_mmuidx_ra(env, addr, PPC_TLB_EPID_LOAD, GETPC());
339 #endif
342 /* XXX: to be tested */
343 target_ulong helper_lscbx(CPUPPCState *env, target_ulong addr, uint32_t reg,
344 uint32_t ra, uint32_t rb)
346 int i, c, d;
348 d = 24;
349 for (i = 0; i < xer_bc; i++) {
350 c = cpu_ldub_data_ra(env, addr, GETPC());
351 addr = addr_add(env, addr, 1);
352 /* ra (if not 0) and rb are never modified */
353 if (likely(reg != rb && (ra == 0 || reg != ra))) {
354 env->gpr[reg] = (env->gpr[reg] & ~(0xFF << d)) | (c << d);
356 if (unlikely(c == xer_cmp)) {
357 break;
359 if (likely(d != 0)) {
360 d -= 8;
361 } else {
362 d = 24;
363 reg++;
364 reg = reg & 0x1F;
367 return i;
370 #ifdef TARGET_PPC64
371 uint64_t helper_lq_le_parallel(CPUPPCState *env, target_ulong addr,
372 uint32_t opidx)
374 Int128 ret;
376 /* We will have raised EXCP_ATOMIC from the translator. */
377 assert(HAVE_ATOMIC128);
378 ret = cpu_atomic_ldo_le_mmu(env, addr, opidx, GETPC());
379 env->retxh = int128_gethi(ret);
380 return int128_getlo(ret);
383 uint64_t helper_lq_be_parallel(CPUPPCState *env, target_ulong addr,
384 uint32_t opidx)
386 Int128 ret;
388 /* We will have raised EXCP_ATOMIC from the translator. */
389 assert(HAVE_ATOMIC128);
390 ret = cpu_atomic_ldo_be_mmu(env, addr, opidx, GETPC());
391 env->retxh = int128_gethi(ret);
392 return int128_getlo(ret);
395 void helper_stq_le_parallel(CPUPPCState *env, target_ulong addr,
396 uint64_t lo, uint64_t hi, uint32_t opidx)
398 Int128 val;
400 /* We will have raised EXCP_ATOMIC from the translator. */
401 assert(HAVE_ATOMIC128);
402 val = int128_make128(lo, hi);
403 cpu_atomic_sto_le_mmu(env, addr, val, opidx, GETPC());
406 void helper_stq_be_parallel(CPUPPCState *env, target_ulong addr,
407 uint64_t lo, uint64_t hi, uint32_t opidx)
409 Int128 val;
411 /* We will have raised EXCP_ATOMIC from the translator. */
412 assert(HAVE_ATOMIC128);
413 val = int128_make128(lo, hi);
414 cpu_atomic_sto_be_mmu(env, addr, val, opidx, GETPC());
417 uint32_t helper_stqcx_le_parallel(CPUPPCState *env, target_ulong addr,
418 uint64_t new_lo, uint64_t new_hi,
419 uint32_t opidx)
421 bool success = false;
423 /* We will have raised EXCP_ATOMIC from the translator. */
424 assert(HAVE_CMPXCHG128);
426 if (likely(addr == env->reserve_addr)) {
427 Int128 oldv, cmpv, newv;
429 cmpv = int128_make128(env->reserve_val2, env->reserve_val);
430 newv = int128_make128(new_lo, new_hi);
431 oldv = cpu_atomic_cmpxchgo_le_mmu(env, addr, cmpv, newv,
432 opidx, GETPC());
433 success = int128_eq(oldv, cmpv);
435 env->reserve_addr = -1;
436 return env->so + success * CRF_EQ_BIT;
439 uint32_t helper_stqcx_be_parallel(CPUPPCState *env, target_ulong addr,
440 uint64_t new_lo, uint64_t new_hi,
441 uint32_t opidx)
443 bool success = false;
445 /* We will have raised EXCP_ATOMIC from the translator. */
446 assert(HAVE_CMPXCHG128);
448 if (likely(addr == env->reserve_addr)) {
449 Int128 oldv, cmpv, newv;
451 cmpv = int128_make128(env->reserve_val2, env->reserve_val);
452 newv = int128_make128(new_lo, new_hi);
453 oldv = cpu_atomic_cmpxchgo_be_mmu(env, addr, cmpv, newv,
454 opidx, GETPC());
455 success = int128_eq(oldv, cmpv);
457 env->reserve_addr = -1;
458 return env->so + success * CRF_EQ_BIT;
460 #endif
462 /*****************************************************************************/
463 /* Altivec extension helpers */
464 #if defined(HOST_WORDS_BIGENDIAN)
465 #define HI_IDX 0
466 #define LO_IDX 1
467 #else
468 #define HI_IDX 1
469 #define LO_IDX 0
470 #endif
473 * We use msr_le to determine index ordering in a vector. However,
474 * byteswapping is not simply controlled by msr_le. We also need to
475 * take into account endianness of the target. This is done for the
476 * little-endian PPC64 user-mode target.
479 #define LVE(name, access, swap, element) \
480 void helper_##name(CPUPPCState *env, ppc_avr_t *r, \
481 target_ulong addr) \
483 size_t n_elems = ARRAY_SIZE(r->element); \
484 int adjust = HI_IDX * (n_elems - 1); \
485 int sh = sizeof(r->element[0]) >> 1; \
486 int index = (addr & 0xf) >> sh; \
487 if (msr_le) { \
488 index = n_elems - index - 1; \
491 if (needs_byteswap(env)) { \
492 r->element[LO_IDX ? index : (adjust - index)] = \
493 swap(access(env, addr, GETPC())); \
494 } else { \
495 r->element[LO_IDX ? index : (adjust - index)] = \
496 access(env, addr, GETPC()); \
499 #define I(x) (x)
500 LVE(lvebx, cpu_ldub_data_ra, I, u8)
501 LVE(lvehx, cpu_lduw_data_ra, bswap16, u16)
502 LVE(lvewx, cpu_ldl_data_ra, bswap32, u32)
503 #undef I
504 #undef LVE
506 #define STVE(name, access, swap, element) \
507 void helper_##name(CPUPPCState *env, ppc_avr_t *r, \
508 target_ulong addr) \
510 size_t n_elems = ARRAY_SIZE(r->element); \
511 int adjust = HI_IDX * (n_elems - 1); \
512 int sh = sizeof(r->element[0]) >> 1; \
513 int index = (addr & 0xf) >> sh; \
514 if (msr_le) { \
515 index = n_elems - index - 1; \
518 if (needs_byteswap(env)) { \
519 access(env, addr, swap(r->element[LO_IDX ? index : \
520 (adjust - index)]), \
521 GETPC()); \
522 } else { \
523 access(env, addr, r->element[LO_IDX ? index : \
524 (adjust - index)], GETPC()); \
527 #define I(x) (x)
528 STVE(stvebx, cpu_stb_data_ra, I, u8)
529 STVE(stvehx, cpu_stw_data_ra, bswap16, u16)
530 STVE(stvewx, cpu_stl_data_ra, bswap32, u32)
531 #undef I
532 #undef LVE
534 #ifdef TARGET_PPC64
535 #define GET_NB(rb) ((rb >> 56) & 0xFF)
537 #define VSX_LXVL(name, lj) \
538 void helper_##name(CPUPPCState *env, target_ulong addr, \
539 ppc_vsr_t *xt, target_ulong rb) \
541 ppc_vsr_t t; \
542 uint64_t nb = GET_NB(rb); \
543 int i; \
545 t.s128 = int128_zero(); \
546 if (nb) { \
547 nb = (nb >= 16) ? 16 : nb; \
548 if (msr_le && !lj) { \
549 for (i = 16; i > 16 - nb; i--) { \
550 t.VsrB(i - 1) = cpu_ldub_data_ra(env, addr, GETPC()); \
551 addr = addr_add(env, addr, 1); \
553 } else { \
554 for (i = 0; i < nb; i++) { \
555 t.VsrB(i) = cpu_ldub_data_ra(env, addr, GETPC()); \
556 addr = addr_add(env, addr, 1); \
560 *xt = t; \
563 VSX_LXVL(lxvl, 0)
564 VSX_LXVL(lxvll, 1)
565 #undef VSX_LXVL
567 #define VSX_STXVL(name, lj) \
568 void helper_##name(CPUPPCState *env, target_ulong addr, \
569 ppc_vsr_t *xt, target_ulong rb) \
571 target_ulong nb = GET_NB(rb); \
572 int i; \
574 if (!nb) { \
575 return; \
578 nb = (nb >= 16) ? 16 : nb; \
579 if (msr_le && !lj) { \
580 for (i = 16; i > 16 - nb; i--) { \
581 cpu_stb_data_ra(env, addr, xt->VsrB(i - 1), GETPC()); \
582 addr = addr_add(env, addr, 1); \
584 } else { \
585 for (i = 0; i < nb; i++) { \
586 cpu_stb_data_ra(env, addr, xt->VsrB(i), GETPC()); \
587 addr = addr_add(env, addr, 1); \
592 VSX_STXVL(stxvl, 0)
593 VSX_STXVL(stxvll, 1)
594 #undef VSX_STXVL
595 #undef GET_NB
596 #endif /* TARGET_PPC64 */
598 #undef HI_IDX
599 #undef LO_IDX
601 void helper_tbegin(CPUPPCState *env)
604 * As a degenerate implementation, always fail tbegin. The reason
605 * given is "Nesting overflow". The "persistent" bit is set,
606 * providing a hint to the error handler to not retry. The TFIAR
607 * captures the address of the failure, which is this tbegin
608 * instruction. Instruction execution will continue with the next
609 * instruction in memory, which is precisely what we want.
612 env->spr[SPR_TEXASR] =
613 (1ULL << TEXASR_FAILURE_PERSISTENT) |
614 (1ULL << TEXASR_NESTING_OVERFLOW) |
615 (msr_hv << TEXASR_PRIVILEGE_HV) |
616 (msr_pr << TEXASR_PRIVILEGE_PR) |
617 (1ULL << TEXASR_FAILURE_SUMMARY) |
618 (1ULL << TEXASR_TFIAR_EXACT);
619 env->spr[SPR_TFIAR] = env->nip | (msr_hv << 1) | msr_pr;
620 env->spr[SPR_TFHAR] = env->nip + 4;
621 env->crf[0] = 0xB; /* 0b1010 = transaction failure */