xref: /openbmc/qemu/target/hexagon/translate.c (revision b83a80e8)
1 /*
2  *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
3  *
4  *  This program is free software; you can redistribute it and/or modify
5  *  it under the terms of the GNU General Public License as published by
6  *  the Free Software Foundation; either version 2 of the License, or
7  *  (at your option) any later version.
8  *
9  *  This program is distributed in the hope that it will be useful,
10  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  *  GNU General Public License for more details.
13  *
14  *  You should have received a copy of the GNU General Public License
15  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
16  */
17 
18 #define QEMU_GENERATE
19 #include "qemu/osdep.h"
20 #include "cpu.h"
21 #include "tcg/tcg-op.h"
22 #include "tcg/tcg-op-gvec.h"
23 #include "exec/cpu_ldst.h"
24 #include "exec/log.h"
25 #include "internal.h"
26 #include "attribs.h"
27 #include "insn.h"
28 #include "decode.h"
29 #include "translate.h"
30 #include "printinsn.h"
31 
32 TCGv hex_gpr[TOTAL_PER_THREAD_REGS];
33 TCGv hex_pred[NUM_PREGS];
34 TCGv hex_next_PC;
35 TCGv hex_this_PC;
36 TCGv hex_slot_cancelled;
37 TCGv hex_branch_taken;
38 TCGv hex_new_value[TOTAL_PER_THREAD_REGS];
39 TCGv hex_reg_written[TOTAL_PER_THREAD_REGS];
40 TCGv hex_new_pred_value[NUM_PREGS];
41 TCGv hex_pred_written;
42 TCGv hex_store_addr[STORES_MAX];
43 TCGv hex_store_width[STORES_MAX];
44 TCGv hex_store_val32[STORES_MAX];
45 TCGv_i64 hex_store_val64[STORES_MAX];
46 TCGv hex_pkt_has_store_s1;
47 TCGv hex_dczero_addr;
48 TCGv hex_llsc_addr;
49 TCGv hex_llsc_val;
50 TCGv_i64 hex_llsc_val_i64;
51 TCGv hex_VRegs_updated;
52 TCGv hex_QRegs_updated;
53 TCGv hex_vstore_addr[VSTORES_MAX];
54 TCGv hex_vstore_size[VSTORES_MAX];
55 TCGv hex_vstore_pending[VSTORES_MAX];
56 
57 static const char * const hexagon_prednames[] = {
58   "p0", "p1", "p2", "p3"
59 };
60 
61 intptr_t ctx_future_vreg_off(DisasContext *ctx, int regnum,
62                           int num, bool alloc_ok)
63 {
64     intptr_t offset;
65 
66     /* See if it is already allocated */
67     for (int i = 0; i < ctx->future_vregs_idx; i++) {
68         if (ctx->future_vregs_num[i] == regnum) {
69             return offsetof(CPUHexagonState, future_VRegs[i]);
70         }
71     }
72 
73     g_assert(alloc_ok);
74     offset = offsetof(CPUHexagonState, future_VRegs[ctx->future_vregs_idx]);
75     for (int i = 0; i < num; i++) {
76         ctx->future_vregs_num[ctx->future_vregs_idx + i] = regnum++;
77     }
78     ctx->future_vregs_idx += num;
79     g_assert(ctx->future_vregs_idx <= VECTOR_TEMPS_MAX);
80     return offset;
81 }
82 
83 intptr_t ctx_tmp_vreg_off(DisasContext *ctx, int regnum,
84                           int num, bool alloc_ok)
85 {
86     intptr_t offset;
87 
88     /* See if it is already allocated */
89     for (int i = 0; i < ctx->tmp_vregs_idx; i++) {
90         if (ctx->tmp_vregs_num[i] == regnum) {
91             return offsetof(CPUHexagonState, tmp_VRegs[i]);
92         }
93     }
94 
95     g_assert(alloc_ok);
96     offset = offsetof(CPUHexagonState, tmp_VRegs[ctx->tmp_vregs_idx]);
97     for (int i = 0; i < num; i++) {
98         ctx->tmp_vregs_num[ctx->tmp_vregs_idx + i] = regnum++;
99     }
100     ctx->tmp_vregs_idx += num;
101     g_assert(ctx->tmp_vregs_idx <= VECTOR_TEMPS_MAX);
102     return offset;
103 }
104 
105 static void gen_exception_raw(int excp)
106 {
107     gen_helper_raise_exception(cpu_env, tcg_constant_i32(excp));
108 }
109 
110 static void gen_exec_counters(DisasContext *ctx)
111 {
112     tcg_gen_addi_tl(hex_gpr[HEX_REG_QEMU_PKT_CNT],
113                     hex_gpr[HEX_REG_QEMU_PKT_CNT], ctx->num_packets);
114     tcg_gen_addi_tl(hex_gpr[HEX_REG_QEMU_INSN_CNT],
115                     hex_gpr[HEX_REG_QEMU_INSN_CNT], ctx->num_insns);
116     tcg_gen_addi_tl(hex_gpr[HEX_REG_QEMU_HVX_CNT],
117                     hex_gpr[HEX_REG_QEMU_HVX_CNT], ctx->num_hvx_insns);
118 }
119 
120 static void gen_end_tb(DisasContext *ctx)
121 {
122     gen_exec_counters(ctx);
123     tcg_gen_mov_tl(hex_gpr[HEX_REG_PC], hex_next_PC);
124     tcg_gen_exit_tb(NULL, 0);
125     ctx->base.is_jmp = DISAS_NORETURN;
126 }
127 
128 static void gen_exception_end_tb(DisasContext *ctx, int excp)
129 {
130     gen_exec_counters(ctx);
131     tcg_gen_mov_tl(hex_gpr[HEX_REG_PC], hex_next_PC);
132     gen_exception_raw(excp);
133     ctx->base.is_jmp = DISAS_NORETURN;
134 
135 }
136 
137 #define PACKET_BUFFER_LEN              1028
138 static void print_pkt(Packet *pkt)
139 {
140     GString *buf = g_string_sized_new(PACKET_BUFFER_LEN);
141     snprint_a_pkt_debug(buf, pkt);
142     HEX_DEBUG_LOG("%s", buf->str);
143     g_string_free(buf, true);
144 }
145 #define HEX_DEBUG_PRINT_PKT(pkt) \
146     do { \
147         if (HEX_DEBUG) { \
148             print_pkt(pkt); \
149         } \
150     } while (0)
151 
152 static int read_packet_words(CPUHexagonState *env, DisasContext *ctx,
153                              uint32_t words[])
154 {
155     bool found_end = false;
156     int nwords, max_words;
157 
158     memset(words, 0, PACKET_WORDS_MAX * sizeof(uint32_t));
159     for (nwords = 0; !found_end && nwords < PACKET_WORDS_MAX; nwords++) {
160         words[nwords] =
161             translator_ldl(env, &ctx->base,
162                            ctx->base.pc_next + nwords * sizeof(uint32_t));
163         found_end = is_packet_end(words[nwords]);
164     }
165     if (!found_end) {
166         /* Read too many words without finding the end */
167         return 0;
168     }
169 
170     /* Check for page boundary crossing */
171     max_words = -(ctx->base.pc_next | TARGET_PAGE_MASK) / sizeof(uint32_t);
172     if (nwords > max_words) {
173         /* We can only cross a page boundary at the beginning of a TB */
174         g_assert(ctx->base.num_insns == 1);
175     }
176 
177     HEX_DEBUG_LOG("decode_packet: pc = 0x%x\n", ctx->base.pc_next);
178     HEX_DEBUG_LOG("    words = { ");
179     for (int i = 0; i < nwords; i++) {
180         HEX_DEBUG_LOG("0x%x, ", words[i]);
181     }
182     HEX_DEBUG_LOG("}\n");
183 
184     return nwords;
185 }
186 
187 static bool check_for_attrib(Packet *pkt, int attrib)
188 {
189     for (int i = 0; i < pkt->num_insns; i++) {
190         if (GET_ATTRIB(pkt->insn[i].opcode, attrib)) {
191             return true;
192         }
193     }
194     return false;
195 }
196 
197 static bool need_pc(Packet *pkt)
198 {
199     return check_for_attrib(pkt, A_IMPLICIT_READS_PC);
200 }
201 
202 static bool need_slot_cancelled(Packet *pkt)
203 {
204     return check_for_attrib(pkt, A_CONDEXEC);
205 }
206 
207 static bool need_pred_written(Packet *pkt)
208 {
209     return check_for_attrib(pkt, A_WRITES_PRED_REG);
210 }
211 
212 static void gen_start_packet(DisasContext *ctx, Packet *pkt)
213 {
214     target_ulong next_PC = ctx->base.pc_next + pkt->encod_pkt_size_in_bytes;
215     int i;
216 
217     /* Clear out the disassembly context */
218     ctx->reg_log_idx = 0;
219     bitmap_zero(ctx->regs_written, TOTAL_PER_THREAD_REGS);
220     ctx->preg_log_idx = 0;
221     bitmap_zero(ctx->pregs_written, NUM_PREGS);
222     ctx->future_vregs_idx = 0;
223     ctx->tmp_vregs_idx = 0;
224     ctx->vreg_log_idx = 0;
225     bitmap_zero(ctx->vregs_updated_tmp, NUM_VREGS);
226     bitmap_zero(ctx->vregs_updated, NUM_VREGS);
227     bitmap_zero(ctx->vregs_select, NUM_VREGS);
228     ctx->qreg_log_idx = 0;
229     for (i = 0; i < STORES_MAX; i++) {
230         ctx->store_width[i] = 0;
231     }
232     tcg_gen_movi_tl(hex_pkt_has_store_s1, pkt->pkt_has_store_s1);
233     ctx->s1_store_processed = false;
234     ctx->pre_commit = true;
235 
236     if (HEX_DEBUG) {
237         /* Handy place to set a breakpoint before the packet executes */
238         gen_helper_debug_start_packet(cpu_env);
239         tcg_gen_movi_tl(hex_this_PC, ctx->base.pc_next);
240     }
241 
242     /* Initialize the runtime state for packet semantics */
243     if (need_pc(pkt)) {
244         tcg_gen_movi_tl(hex_gpr[HEX_REG_PC], ctx->base.pc_next);
245     }
246     if (need_slot_cancelled(pkt)) {
247         tcg_gen_movi_tl(hex_slot_cancelled, 0);
248     }
249     if (pkt->pkt_has_cof) {
250         tcg_gen_movi_tl(hex_branch_taken, 0);
251         tcg_gen_movi_tl(hex_next_PC, next_PC);
252     }
253     if (need_pred_written(pkt)) {
254         tcg_gen_movi_tl(hex_pred_written, 0);
255     }
256 
257     if (pkt->pkt_has_hvx) {
258         tcg_gen_movi_tl(hex_VRegs_updated, 0);
259         tcg_gen_movi_tl(hex_QRegs_updated, 0);
260     }
261 }
262 
263 bool is_gather_store_insn(Insn *insn, Packet *pkt)
264 {
265     if (GET_ATTRIB(insn->opcode, A_CVI_NEW) &&
266         insn->new_value_producer_slot == 1) {
267         /* Look for gather instruction */
268         for (int i = 0; i < pkt->num_insns; i++) {
269             Insn *in = &pkt->insn[i];
270             if (GET_ATTRIB(in->opcode, A_CVI_GATHER) && in->slot == 1) {
271                 return true;
272             }
273         }
274     }
275     return false;
276 }
277 
278 /*
279  * The LOG_*_WRITE macros mark most of the writes in a packet
280  * However, there are some implicit writes marked as attributes
281  * of the applicable instructions.
282  */
283 static void mark_implicit_reg_write(DisasContext *ctx, Insn *insn,
284                                     int attrib, int rnum)
285 {
286     if (GET_ATTRIB(insn->opcode, attrib)) {
287         /*
288          * USR is used to set overflow and FP exceptions,
289          * so treat it as conditional
290          */
291         bool is_predicated = GET_ATTRIB(insn->opcode, A_CONDEXEC) ||
292                              rnum == HEX_REG_USR;
293         if (is_predicated && !is_preloaded(ctx, rnum)) {
294             tcg_gen_mov_tl(hex_new_value[rnum], hex_gpr[rnum]);
295         }
296 
297         ctx_log_reg_write(ctx, rnum);
298     }
299 }
300 
301 static void mark_implicit_pred_write(DisasContext *ctx, Insn *insn,
302                                      int attrib, int pnum)
303 {
304     if (GET_ATTRIB(insn->opcode, attrib)) {
305         ctx_log_pred_write(ctx, pnum);
306     }
307 }
308 
309 static void mark_implicit_reg_writes(DisasContext *ctx, Insn *insn)
310 {
311     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_FP,  HEX_REG_FP);
312     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_SP,  HEX_REG_SP);
313     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_LR,  HEX_REG_LR);
314     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_LC0, HEX_REG_LC0);
315     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_SA0, HEX_REG_SA0);
316     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_LC1, HEX_REG_LC1);
317     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_SA1, HEX_REG_SA1);
318     mark_implicit_reg_write(ctx, insn, A_IMPLICIT_WRITES_USR, HEX_REG_USR);
319     mark_implicit_reg_write(ctx, insn, A_FPOP, HEX_REG_USR);
320 }
321 
322 static void mark_implicit_pred_writes(DisasContext *ctx, Insn *insn)
323 {
324     mark_implicit_pred_write(ctx, insn, A_IMPLICIT_WRITES_P0, 0);
325     mark_implicit_pred_write(ctx, insn, A_IMPLICIT_WRITES_P1, 1);
326     mark_implicit_pred_write(ctx, insn, A_IMPLICIT_WRITES_P2, 2);
327     mark_implicit_pred_write(ctx, insn, A_IMPLICIT_WRITES_P3, 3);
328 }
329 
330 static void gen_insn(CPUHexagonState *env, DisasContext *ctx,
331                      Insn *insn, Packet *pkt)
332 {
333     if (insn->generate) {
334         mark_implicit_reg_writes(ctx, insn);
335         insn->generate(env, ctx, insn, pkt);
336         mark_implicit_pred_writes(ctx, insn);
337     } else {
338         gen_exception_end_tb(ctx, HEX_EXCP_INVALID_OPCODE);
339     }
340 }
341 
342 /*
343  * Helpers for generating the packet commit
344  */
345 static void gen_reg_writes(DisasContext *ctx)
346 {
347     int i;
348 
349     for (i = 0; i < ctx->reg_log_idx; i++) {
350         int reg_num = ctx->reg_log[i];
351 
352         tcg_gen_mov_tl(hex_gpr[reg_num], hex_new_value[reg_num]);
353     }
354 }
355 
356 static void gen_pred_writes(DisasContext *ctx, Packet *pkt)
357 {
358     int i;
359 
360     /* Early exit if the log is empty */
361     if (!ctx->preg_log_idx) {
362         return;
363     }
364 
365     /*
366      * Only endloop instructions will conditionally
367      * write a predicate.  If there are no endloop
368      * instructions, we can use the non-conditional
369      * write of the predicates.
370      */
371     if (pkt->pkt_has_endloop) {
372         TCGv zero = tcg_constant_tl(0);
373         TCGv pred_written = tcg_temp_new();
374         for (i = 0; i < ctx->preg_log_idx; i++) {
375             int pred_num = ctx->preg_log[i];
376 
377             tcg_gen_andi_tl(pred_written, hex_pred_written, 1 << pred_num);
378             tcg_gen_movcond_tl(TCG_COND_NE, hex_pred[pred_num],
379                                pred_written, zero,
380                                hex_new_pred_value[pred_num],
381                                hex_pred[pred_num]);
382         }
383         tcg_temp_free(pred_written);
384     } else {
385         for (i = 0; i < ctx->preg_log_idx; i++) {
386             int pred_num = ctx->preg_log[i];
387             tcg_gen_mov_tl(hex_pred[pred_num], hex_new_pred_value[pred_num]);
388             if (HEX_DEBUG) {
389                 /* Do this so HELPER(debug_commit_end) will know */
390                 tcg_gen_ori_tl(hex_pred_written, hex_pred_written,
391                                1 << pred_num);
392             }
393         }
394     }
395 }
396 
397 static void gen_check_store_width(DisasContext *ctx, int slot_num)
398 {
399     if (HEX_DEBUG) {
400         TCGv slot = tcg_constant_tl(slot_num);
401         TCGv check = tcg_constant_tl(ctx->store_width[slot_num]);
402         gen_helper_debug_check_store_width(cpu_env, slot, check);
403     }
404 }
405 
406 static bool slot_is_predicated(Packet *pkt, int slot_num)
407 {
408     for (int i = 0; i < pkt->num_insns; i++) {
409         if (pkt->insn[i].slot == slot_num) {
410             return GET_ATTRIB(pkt->insn[i].opcode, A_CONDEXEC);
411         }
412     }
413     /* If we get to here, we didn't find an instruction in the requested slot */
414     g_assert_not_reached();
415 }
416 
417 void process_store(DisasContext *ctx, Packet *pkt, int slot_num)
418 {
419     bool is_predicated = slot_is_predicated(pkt, slot_num);
420     TCGLabel *label_end = NULL;
421 
422     /*
423      * We may have already processed this store
424      * See CHECK_NOSHUF in macros.h
425      */
426     if (slot_num == 1 && ctx->s1_store_processed) {
427         return;
428     }
429     ctx->s1_store_processed = true;
430 
431     if (is_predicated) {
432         TCGv cancelled = tcg_temp_new();
433         label_end = gen_new_label();
434 
435         /* Don't do anything if the slot was cancelled */
436         tcg_gen_extract_tl(cancelled, hex_slot_cancelled, slot_num, 1);
437         tcg_gen_brcondi_tl(TCG_COND_NE, cancelled, 0, label_end);
438         tcg_temp_free(cancelled);
439     }
440     {
441         TCGv address = tcg_temp_local_new();
442         tcg_gen_mov_tl(address, hex_store_addr[slot_num]);
443 
444         /*
445          * If we know the width from the DisasContext, we can
446          * generate much cleaner code.
447          * Unfortunately, not all instructions execute the fSTORE
448          * macro during code generation.  Anything that uses the
449          * generic helper will have this problem.  Instructions
450          * that use fWRAP to generate proper TCG code will be OK.
451          */
452         switch (ctx->store_width[slot_num]) {
453         case 1:
454             gen_check_store_width(ctx, slot_num);
455             tcg_gen_qemu_st8(hex_store_val32[slot_num],
456                              hex_store_addr[slot_num],
457                              ctx->mem_idx);
458             break;
459         case 2:
460             gen_check_store_width(ctx, slot_num);
461             tcg_gen_qemu_st16(hex_store_val32[slot_num],
462                               hex_store_addr[slot_num],
463                               ctx->mem_idx);
464             break;
465         case 4:
466             gen_check_store_width(ctx, slot_num);
467             tcg_gen_qemu_st32(hex_store_val32[slot_num],
468                               hex_store_addr[slot_num],
469                               ctx->mem_idx);
470             break;
471         case 8:
472             gen_check_store_width(ctx, slot_num);
473             tcg_gen_qemu_st64(hex_store_val64[slot_num],
474                               hex_store_addr[slot_num],
475                               ctx->mem_idx);
476             break;
477         default:
478             {
479                 /*
480                  * If we get to here, we don't know the width at
481                  * TCG generation time, we'll use a helper to
482                  * avoid branching based on the width at runtime.
483                  */
484                 TCGv slot = tcg_constant_tl(slot_num);
485                 gen_helper_commit_store(cpu_env, slot);
486             }
487         }
488         tcg_temp_free(address);
489     }
490     if (is_predicated) {
491         gen_set_label(label_end);
492     }
493 }
494 
495 static void process_store_log(DisasContext *ctx, Packet *pkt)
496 {
497     /*
498      *  When a packet has two stores, the hardware processes
499      *  slot 1 and then slot 0.  This will be important when
500      *  the memory accesses overlap.
501      */
502     if (pkt->pkt_has_store_s1 && !pkt->pkt_has_dczeroa) {
503         process_store(ctx, pkt, 1);
504     }
505     if (pkt->pkt_has_store_s0 && !pkt->pkt_has_dczeroa) {
506         process_store(ctx, pkt, 0);
507     }
508 }
509 
510 /* Zero out a 32-bit cache line */
511 static void process_dczeroa(DisasContext *ctx, Packet *pkt)
512 {
513     if (pkt->pkt_has_dczeroa) {
514         /* Store 32 bytes of zero starting at (addr & ~0x1f) */
515         TCGv addr = tcg_temp_new();
516         TCGv_i64 zero = tcg_constant_i64(0);
517 
518         tcg_gen_andi_tl(addr, hex_dczero_addr, ~0x1f);
519         tcg_gen_qemu_st64(zero, addr, ctx->mem_idx);
520         tcg_gen_addi_tl(addr, addr, 8);
521         tcg_gen_qemu_st64(zero, addr, ctx->mem_idx);
522         tcg_gen_addi_tl(addr, addr, 8);
523         tcg_gen_qemu_st64(zero, addr, ctx->mem_idx);
524         tcg_gen_addi_tl(addr, addr, 8);
525         tcg_gen_qemu_st64(zero, addr, ctx->mem_idx);
526 
527         tcg_temp_free(addr);
528     }
529 }
530 
531 static bool pkt_has_hvx_store(Packet *pkt)
532 {
533     int i;
534     for (i = 0; i < pkt->num_insns; i++) {
535         int opcode = pkt->insn[i].opcode;
536         if (GET_ATTRIB(opcode, A_CVI) && GET_ATTRIB(opcode, A_STORE)) {
537             return true;
538         }
539     }
540     return false;
541 }
542 
543 static void gen_commit_hvx(DisasContext *ctx, Packet *pkt)
544 {
545     int i;
546 
547     /*
548      *    for (i = 0; i < ctx->vreg_log_idx; i++) {
549      *        int rnum = ctx->vreg_log[i];
550      *        if (ctx->vreg_is_predicated[i]) {
551      *            if (env->VRegs_updated & (1 << rnum)) {
552      *                env->VRegs[rnum] = env->future_VRegs[rnum];
553      *            }
554      *        } else {
555      *            env->VRegs[rnum] = env->future_VRegs[rnum];
556      *        }
557      *    }
558      */
559     for (i = 0; i < ctx->vreg_log_idx; i++) {
560         int rnum = ctx->vreg_log[i];
561         bool is_predicated = ctx->vreg_is_predicated[i];
562         intptr_t dstoff = offsetof(CPUHexagonState, VRegs[rnum]);
563         intptr_t srcoff = ctx_future_vreg_off(ctx, rnum, 1, false);
564         size_t size = sizeof(MMVector);
565 
566         if (is_predicated) {
567             TCGv cmp = tcg_temp_new();
568             TCGLabel *label_skip = gen_new_label();
569 
570             tcg_gen_andi_tl(cmp, hex_VRegs_updated, 1 << rnum);
571             tcg_gen_brcondi_tl(TCG_COND_EQ, cmp, 0, label_skip);
572             tcg_temp_free(cmp);
573             tcg_gen_gvec_mov(MO_64, dstoff, srcoff, size, size);
574             gen_set_label(label_skip);
575         } else {
576             tcg_gen_gvec_mov(MO_64, dstoff, srcoff, size, size);
577         }
578     }
579 
580     /*
581      *    for (i = 0; i < ctx->qreg_log_idx; i++) {
582      *        int rnum = ctx->qreg_log[i];
583      *        if (ctx->qreg_is_predicated[i]) {
584      *            if (env->QRegs_updated) & (1 << rnum)) {
585      *                env->QRegs[rnum] = env->future_QRegs[rnum];
586      *            }
587      *        } else {
588      *            env->QRegs[rnum] = env->future_QRegs[rnum];
589      *        }
590      *    }
591      */
592     for (i = 0; i < ctx->qreg_log_idx; i++) {
593         int rnum = ctx->qreg_log[i];
594         bool is_predicated = ctx->qreg_is_predicated[i];
595         intptr_t dstoff = offsetof(CPUHexagonState, QRegs[rnum]);
596         intptr_t srcoff = offsetof(CPUHexagonState, future_QRegs[rnum]);
597         size_t size = sizeof(MMQReg);
598 
599         if (is_predicated) {
600             TCGv cmp = tcg_temp_new();
601             TCGLabel *label_skip = gen_new_label();
602 
603             tcg_gen_andi_tl(cmp, hex_QRegs_updated, 1 << rnum);
604             tcg_gen_brcondi_tl(TCG_COND_EQ, cmp, 0, label_skip);
605             tcg_temp_free(cmp);
606             tcg_gen_gvec_mov(MO_64, dstoff, srcoff, size, size);
607             gen_set_label(label_skip);
608         } else {
609             tcg_gen_gvec_mov(MO_64, dstoff, srcoff, size, size);
610         }
611     }
612 
613     if (pkt_has_hvx_store(pkt)) {
614         gen_helper_commit_hvx_stores(cpu_env);
615     }
616 }
617 
618 static void update_exec_counters(DisasContext *ctx, Packet *pkt)
619 {
620     int num_insns = pkt->num_insns;
621     int num_real_insns = 0;
622     int num_hvx_insns = 0;
623 
624     for (int i = 0; i < num_insns; i++) {
625         if (!pkt->insn[i].is_endloop &&
626             !pkt->insn[i].part1 &&
627             !GET_ATTRIB(pkt->insn[i].opcode, A_IT_NOP)) {
628             num_real_insns++;
629         }
630         if (GET_ATTRIB(pkt->insn[i].opcode, A_CVI)) {
631             num_hvx_insns++;
632         }
633     }
634 
635     ctx->num_packets++;
636     ctx->num_insns += num_real_insns;
637     ctx->num_hvx_insns += num_hvx_insns;
638 }
639 
640 static void gen_commit_packet(CPUHexagonState *env, DisasContext *ctx,
641                               Packet *pkt)
642 {
643     /*
644      * If there is more than one store in a packet, make sure they are all OK
645      * before proceeding with the rest of the packet commit.
646      *
647      * dczeroa has to be the only store operation in the packet, so we go
648      * ahead and process that first.
649      *
650      * When there is an HVX store, there can also be a scalar store in either
651      * slot 0 or slot1, so we create a mask for the helper to indicate what
652      * work to do.
653      *
654      * When there are two scalar stores, we probe the one in slot 0.
655      *
656      * Note that we don't call the probe helper for packets with only one
657      * store.  Therefore, we call process_store_log before anything else
658      * involved in committing the packet.
659      */
660     bool has_store_s0 = pkt->pkt_has_store_s0;
661     bool has_store_s1 = (pkt->pkt_has_store_s1 && !ctx->s1_store_processed);
662     bool has_hvx_store = pkt_has_hvx_store(pkt);
663     if (pkt->pkt_has_dczeroa) {
664         /*
665          * The dczeroa will be the store in slot 0, check that we don't have
666          * a store in slot 1 or an HVX store.
667          */
668         g_assert(has_store_s0 && !has_store_s1 && !has_hvx_store);
669         process_dczeroa(ctx, pkt);
670     } else if (has_hvx_store) {
671         TCGv mem_idx = tcg_constant_tl(ctx->mem_idx);
672 
673         if (!has_store_s0 && !has_store_s1) {
674             gen_helper_probe_hvx_stores(cpu_env, mem_idx);
675         } else {
676             int mask = 0;
677             TCGv mask_tcgv;
678 
679             if (has_store_s0) {
680                 mask |= (1 << 0);
681             }
682             if (has_store_s1) {
683                 mask |= (1 << 1);
684             }
685             if (has_hvx_store) {
686                 mask |= (1 << 2);
687             }
688             mask_tcgv = tcg_constant_tl(mask);
689             gen_helper_probe_pkt_scalar_hvx_stores(cpu_env, mask_tcgv, mem_idx);
690         }
691     } else if (has_store_s0 && has_store_s1) {
692         /*
693          * process_store_log will execute the slot 1 store first,
694          * so we only have to probe the store in slot 0
695          */
696         TCGv mem_idx = tcg_constant_tl(ctx->mem_idx);
697         gen_helper_probe_pkt_scalar_store_s0(cpu_env, mem_idx);
698     }
699 
700     process_store_log(ctx, pkt);
701 
702     gen_reg_writes(ctx);
703     gen_pred_writes(ctx, pkt);
704     if (pkt->pkt_has_hvx) {
705         gen_commit_hvx(ctx, pkt);
706     }
707     update_exec_counters(ctx, pkt);
708     if (HEX_DEBUG) {
709         TCGv has_st0 =
710             tcg_constant_tl(pkt->pkt_has_store_s0 && !pkt->pkt_has_dczeroa);
711         TCGv has_st1 =
712             tcg_constant_tl(pkt->pkt_has_store_s1 && !pkt->pkt_has_dczeroa);
713 
714         /* Handy place to set a breakpoint at the end of execution */
715         gen_helper_debug_commit_end(cpu_env, has_st0, has_st1);
716     }
717 
718     if (pkt->vhist_insn != NULL) {
719         ctx->pre_commit = false;
720         pkt->vhist_insn->generate(env, ctx, pkt->vhist_insn, pkt);
721     }
722 
723     if (pkt->pkt_has_cof) {
724         gen_end_tb(ctx);
725     }
726 }
727 
728 static void decode_and_translate_packet(CPUHexagonState *env, DisasContext *ctx)
729 {
730     uint32_t words[PACKET_WORDS_MAX];
731     int nwords;
732     Packet pkt;
733     int i;
734 
735     nwords = read_packet_words(env, ctx, words);
736     if (!nwords) {
737         gen_exception_end_tb(ctx, HEX_EXCP_INVALID_PACKET);
738         return;
739     }
740 
741     if (decode_packet(nwords, words, &pkt, false) > 0) {
742         HEX_DEBUG_PRINT_PKT(&pkt);
743         gen_start_packet(ctx, &pkt);
744         for (i = 0; i < pkt.num_insns; i++) {
745             gen_insn(env, ctx, &pkt.insn[i], &pkt);
746         }
747         gen_commit_packet(env, ctx, &pkt);
748         ctx->base.pc_next += pkt.encod_pkt_size_in_bytes;
749     } else {
750         gen_exception_end_tb(ctx, HEX_EXCP_INVALID_PACKET);
751     }
752 }
753 
754 static void hexagon_tr_init_disas_context(DisasContextBase *dcbase,
755                                           CPUState *cs)
756 {
757     DisasContext *ctx = container_of(dcbase, DisasContext, base);
758 
759     ctx->mem_idx = MMU_USER_IDX;
760     ctx->num_packets = 0;
761     ctx->num_insns = 0;
762     ctx->num_hvx_insns = 0;
763 }
764 
765 static void hexagon_tr_tb_start(DisasContextBase *db, CPUState *cpu)
766 {
767 }
768 
769 static void hexagon_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
770 {
771     DisasContext *ctx = container_of(dcbase, DisasContext, base);
772 
773     tcg_gen_insn_start(ctx->base.pc_next);
774 }
775 
776 static bool pkt_crosses_page(CPUHexagonState *env, DisasContext *ctx)
777 {
778     target_ulong page_start = ctx->base.pc_first & TARGET_PAGE_MASK;
779     bool found_end = false;
780     int nwords;
781 
782     for (nwords = 0; !found_end && nwords < PACKET_WORDS_MAX; nwords++) {
783         uint32_t word = cpu_ldl_code(env,
784                             ctx->base.pc_next + nwords * sizeof(uint32_t));
785         found_end = is_packet_end(word);
786     }
787     uint32_t next_ptr =  ctx->base.pc_next + nwords * sizeof(uint32_t);
788     return found_end && next_ptr - page_start >= TARGET_PAGE_SIZE;
789 }
790 
791 static void hexagon_tr_translate_packet(DisasContextBase *dcbase, CPUState *cpu)
792 {
793     DisasContext *ctx = container_of(dcbase, DisasContext, base);
794     CPUHexagonState *env = cpu->env_ptr;
795 
796     decode_and_translate_packet(env, ctx);
797 
798     if (ctx->base.is_jmp == DISAS_NEXT) {
799         target_ulong page_start = ctx->base.pc_first & TARGET_PAGE_MASK;
800         target_ulong bytes_max = PACKET_WORDS_MAX * sizeof(target_ulong);
801 
802         if (ctx->base.pc_next - page_start >= TARGET_PAGE_SIZE ||
803             (ctx->base.pc_next - page_start >= TARGET_PAGE_SIZE - bytes_max &&
804              pkt_crosses_page(env, ctx))) {
805             ctx->base.is_jmp = DISAS_TOO_MANY;
806         }
807 
808         /*
809          * The CPU log is used to compare against LLDB single stepping,
810          * so end the TLB after every packet.
811          */
812         HexagonCPU *hex_cpu = env_archcpu(env);
813         if (hex_cpu->lldb_compat && qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
814             ctx->base.is_jmp = DISAS_TOO_MANY;
815         }
816     }
817 }
818 
819 static void hexagon_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
820 {
821     DisasContext *ctx = container_of(dcbase, DisasContext, base);
822 
823     switch (ctx->base.is_jmp) {
824     case DISAS_TOO_MANY:
825         gen_exec_counters(ctx);
826         tcg_gen_movi_tl(hex_gpr[HEX_REG_PC], ctx->base.pc_next);
827         tcg_gen_exit_tb(NULL, 0);
828         break;
829     case DISAS_NORETURN:
830         break;
831     default:
832         g_assert_not_reached();
833     }
834 }
835 
836 static void hexagon_tr_disas_log(const DisasContextBase *dcbase, CPUState *cpu)
837 {
838     qemu_log("IN: %s\n", lookup_symbol(dcbase->pc_first));
839     log_target_disas(cpu, dcbase->pc_first, dcbase->tb->size);
840 }
841 
842 
843 static const TranslatorOps hexagon_tr_ops = {
844     .init_disas_context = hexagon_tr_init_disas_context,
845     .tb_start           = hexagon_tr_tb_start,
846     .insn_start         = hexagon_tr_insn_start,
847     .translate_insn     = hexagon_tr_translate_packet,
848     .tb_stop            = hexagon_tr_tb_stop,
849     .disas_log          = hexagon_tr_disas_log,
850 };
851 
852 void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int max_insns)
853 {
854     DisasContext ctx;
855 
856     translator_loop(&hexagon_tr_ops, &ctx.base, cs, tb, max_insns);
857 }
858 
859 #define NAME_LEN               64
860 static char new_value_names[TOTAL_PER_THREAD_REGS][NAME_LEN];
861 static char reg_written_names[TOTAL_PER_THREAD_REGS][NAME_LEN];
862 static char new_pred_value_names[NUM_PREGS][NAME_LEN];
863 static char store_addr_names[STORES_MAX][NAME_LEN];
864 static char store_width_names[STORES_MAX][NAME_LEN];
865 static char store_val32_names[STORES_MAX][NAME_LEN];
866 static char store_val64_names[STORES_MAX][NAME_LEN];
867 static char vstore_addr_names[VSTORES_MAX][NAME_LEN];
868 static char vstore_size_names[VSTORES_MAX][NAME_LEN];
869 static char vstore_pending_names[VSTORES_MAX][NAME_LEN];
870 
871 void hexagon_translate_init(void)
872 {
873     int i;
874 
875     opcode_init();
876 
877     if (HEX_DEBUG) {
878         if (!qemu_logfile) {
879             qemu_set_log(qemu_loglevel);
880         }
881     }
882 
883     for (i = 0; i < TOTAL_PER_THREAD_REGS; i++) {
884         hex_gpr[i] = tcg_global_mem_new(cpu_env,
885             offsetof(CPUHexagonState, gpr[i]),
886             hexagon_regnames[i]);
887 
888         snprintf(new_value_names[i], NAME_LEN, "new_%s", hexagon_regnames[i]);
889         hex_new_value[i] = tcg_global_mem_new(cpu_env,
890             offsetof(CPUHexagonState, new_value[i]),
891             new_value_names[i]);
892 
893         if (HEX_DEBUG) {
894             snprintf(reg_written_names[i], NAME_LEN, "reg_written_%s",
895                      hexagon_regnames[i]);
896             hex_reg_written[i] = tcg_global_mem_new(cpu_env,
897                 offsetof(CPUHexagonState, reg_written[i]),
898                 reg_written_names[i]);
899         }
900     }
901     for (i = 0; i < NUM_PREGS; i++) {
902         hex_pred[i] = tcg_global_mem_new(cpu_env,
903             offsetof(CPUHexagonState, pred[i]),
904             hexagon_prednames[i]);
905 
906         snprintf(new_pred_value_names[i], NAME_LEN, "new_pred_%s",
907                  hexagon_prednames[i]);
908         hex_new_pred_value[i] = tcg_global_mem_new(cpu_env,
909             offsetof(CPUHexagonState, new_pred_value[i]),
910             new_pred_value_names[i]);
911     }
912     hex_pred_written = tcg_global_mem_new(cpu_env,
913         offsetof(CPUHexagonState, pred_written), "pred_written");
914     hex_next_PC = tcg_global_mem_new(cpu_env,
915         offsetof(CPUHexagonState, next_PC), "next_PC");
916     hex_this_PC = tcg_global_mem_new(cpu_env,
917         offsetof(CPUHexagonState, this_PC), "this_PC");
918     hex_slot_cancelled = tcg_global_mem_new(cpu_env,
919         offsetof(CPUHexagonState, slot_cancelled), "slot_cancelled");
920     hex_branch_taken = tcg_global_mem_new(cpu_env,
921         offsetof(CPUHexagonState, branch_taken), "branch_taken");
922     hex_pkt_has_store_s1 = tcg_global_mem_new(cpu_env,
923         offsetof(CPUHexagonState, pkt_has_store_s1), "pkt_has_store_s1");
924     hex_dczero_addr = tcg_global_mem_new(cpu_env,
925         offsetof(CPUHexagonState, dczero_addr), "dczero_addr");
926     hex_llsc_addr = tcg_global_mem_new(cpu_env,
927         offsetof(CPUHexagonState, llsc_addr), "llsc_addr");
928     hex_llsc_val = tcg_global_mem_new(cpu_env,
929         offsetof(CPUHexagonState, llsc_val), "llsc_val");
930     hex_llsc_val_i64 = tcg_global_mem_new_i64(cpu_env,
931         offsetof(CPUHexagonState, llsc_val_i64), "llsc_val_i64");
932     hex_VRegs_updated = tcg_global_mem_new(cpu_env,
933         offsetof(CPUHexagonState, VRegs_updated), "VRegs_updated");
934     hex_QRegs_updated = tcg_global_mem_new(cpu_env,
935         offsetof(CPUHexagonState, QRegs_updated), "QRegs_updated");
936     for (i = 0; i < STORES_MAX; i++) {
937         snprintf(store_addr_names[i], NAME_LEN, "store_addr_%d", i);
938         hex_store_addr[i] = tcg_global_mem_new(cpu_env,
939             offsetof(CPUHexagonState, mem_log_stores[i].va),
940             store_addr_names[i]);
941 
942         snprintf(store_width_names[i], NAME_LEN, "store_width_%d", i);
943         hex_store_width[i] = tcg_global_mem_new(cpu_env,
944             offsetof(CPUHexagonState, mem_log_stores[i].width),
945             store_width_names[i]);
946 
947         snprintf(store_val32_names[i], NAME_LEN, "store_val32_%d", i);
948         hex_store_val32[i] = tcg_global_mem_new(cpu_env,
949             offsetof(CPUHexagonState, mem_log_stores[i].data32),
950             store_val32_names[i]);
951 
952         snprintf(store_val64_names[i], NAME_LEN, "store_val64_%d", i);
953         hex_store_val64[i] = tcg_global_mem_new_i64(cpu_env,
954             offsetof(CPUHexagonState, mem_log_stores[i].data64),
955             store_val64_names[i]);
956     }
957     for (int i = 0; i < VSTORES_MAX; i++) {
958         snprintf(vstore_addr_names[i], NAME_LEN, "vstore_addr_%d", i);
959         hex_vstore_addr[i] = tcg_global_mem_new(cpu_env,
960             offsetof(CPUHexagonState, vstore[i].va),
961             vstore_addr_names[i]);
962 
963         snprintf(vstore_size_names[i], NAME_LEN, "vstore_size_%d", i);
964         hex_vstore_size[i] = tcg_global_mem_new(cpu_env,
965             offsetof(CPUHexagonState, vstore[i].size),
966             vstore_size_names[i]);
967 
968         snprintf(vstore_pending_names[i], NAME_LEN, "vstore_pending_%d", i);
969         hex_vstore_pending[i] = tcg_global_mem_new(cpu_env,
970             offsetof(CPUHexagonState, vstore_pending[i]),
971             vstore_pending_names[i]);
972     }
973 }
974