xref: /openbmc/qemu/include/tcg/tcg.h (revision 7f750efc)
1 /*
2  * Tiny Code Generator for QEMU
3  *
4  * Copyright (c) 2008 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #ifndef TCG_H
26 #define TCG_H
27 
28 #include "exec/memop.h"
29 #include "exec/memopidx.h"
30 #include "qemu/bitops.h"
31 #include "qemu/plugin.h"
32 #include "qemu/queue.h"
33 #include "tcg/tcg-mo.h"
34 #include "tcg-target-reg-bits.h"
35 #include "tcg-target.h"
36 #include "tcg/tcg-cond.h"
37 #include "tcg/debug-assert.h"
38 
39 /* XXX: make safe guess about sizes */
40 #define MAX_OP_PER_INSTR 266
41 
42 #define MAX_CALL_IARGS  7
43 
44 #define CPU_TEMP_BUF_NLONGS 128
45 #define TCG_STATIC_FRAME_SIZE  (CPU_TEMP_BUF_NLONGS * sizeof(long))
46 
47 #if TCG_TARGET_REG_BITS == 32
48 typedef int32_t tcg_target_long;
49 typedef uint32_t tcg_target_ulong;
50 #define TCG_PRIlx PRIx32
51 #define TCG_PRIld PRId32
52 #elif TCG_TARGET_REG_BITS == 64
53 typedef int64_t tcg_target_long;
54 typedef uint64_t tcg_target_ulong;
55 #define TCG_PRIlx PRIx64
56 #define TCG_PRIld PRId64
57 #else
58 #error unsupported
59 #endif
60 
61 #if TCG_TARGET_NB_REGS <= 32
62 typedef uint32_t TCGRegSet;
63 #elif TCG_TARGET_NB_REGS <= 64
64 typedef uint64_t TCGRegSet;
65 #else
66 #error unsupported
67 #endif
68 
69 #if TCG_TARGET_REG_BITS == 32
70 /* Turn some undef macros into false macros.  */
71 #define TCG_TARGET_HAS_extrl_i64_i32    0
72 #define TCG_TARGET_HAS_extrh_i64_i32    0
73 #define TCG_TARGET_HAS_div_i64          0
74 #define TCG_TARGET_HAS_rem_i64          0
75 #define TCG_TARGET_HAS_div2_i64         0
76 #define TCG_TARGET_HAS_rot_i64          0
77 #define TCG_TARGET_HAS_ext8s_i64        0
78 #define TCG_TARGET_HAS_ext16s_i64       0
79 #define TCG_TARGET_HAS_ext32s_i64       0
80 #define TCG_TARGET_HAS_ext8u_i64        0
81 #define TCG_TARGET_HAS_ext16u_i64       0
82 #define TCG_TARGET_HAS_ext32u_i64       0
83 #define TCG_TARGET_HAS_bswap16_i64      0
84 #define TCG_TARGET_HAS_bswap32_i64      0
85 #define TCG_TARGET_HAS_bswap64_i64      0
86 #define TCG_TARGET_HAS_neg_i64          0
87 #define TCG_TARGET_HAS_not_i64          0
88 #define TCG_TARGET_HAS_andc_i64         0
89 #define TCG_TARGET_HAS_orc_i64          0
90 #define TCG_TARGET_HAS_eqv_i64          0
91 #define TCG_TARGET_HAS_nand_i64         0
92 #define TCG_TARGET_HAS_nor_i64          0
93 #define TCG_TARGET_HAS_clz_i64          0
94 #define TCG_TARGET_HAS_ctz_i64          0
95 #define TCG_TARGET_HAS_ctpop_i64        0
96 #define TCG_TARGET_HAS_deposit_i64      0
97 #define TCG_TARGET_HAS_extract_i64      0
98 #define TCG_TARGET_HAS_sextract_i64     0
99 #define TCG_TARGET_HAS_extract2_i64     0
100 #define TCG_TARGET_HAS_movcond_i64      0
101 #define TCG_TARGET_HAS_add2_i64         0
102 #define TCG_TARGET_HAS_sub2_i64         0
103 #define TCG_TARGET_HAS_mulu2_i64        0
104 #define TCG_TARGET_HAS_muls2_i64        0
105 #define TCG_TARGET_HAS_muluh_i64        0
106 #define TCG_TARGET_HAS_mulsh_i64        0
107 /* Turn some undef macros into true macros.  */
108 #define TCG_TARGET_HAS_add2_i32         1
109 #define TCG_TARGET_HAS_sub2_i32         1
110 #endif
111 
112 #ifndef TCG_TARGET_deposit_i32_valid
113 #define TCG_TARGET_deposit_i32_valid(ofs, len) 1
114 #endif
115 #ifndef TCG_TARGET_deposit_i64_valid
116 #define TCG_TARGET_deposit_i64_valid(ofs, len) 1
117 #endif
118 #ifndef TCG_TARGET_extract_i32_valid
119 #define TCG_TARGET_extract_i32_valid(ofs, len) 1
120 #endif
121 #ifndef TCG_TARGET_extract_i64_valid
122 #define TCG_TARGET_extract_i64_valid(ofs, len) 1
123 #endif
124 
125 /* Only one of DIV or DIV2 should be defined.  */
126 #if defined(TCG_TARGET_HAS_div_i32)
127 #define TCG_TARGET_HAS_div2_i32         0
128 #elif defined(TCG_TARGET_HAS_div2_i32)
129 #define TCG_TARGET_HAS_div_i32          0
130 #define TCG_TARGET_HAS_rem_i32          0
131 #endif
132 #if defined(TCG_TARGET_HAS_div_i64)
133 #define TCG_TARGET_HAS_div2_i64         0
134 #elif defined(TCG_TARGET_HAS_div2_i64)
135 #define TCG_TARGET_HAS_div_i64          0
136 #define TCG_TARGET_HAS_rem_i64          0
137 #endif
138 
139 #if !defined(TCG_TARGET_HAS_v64) \
140     && !defined(TCG_TARGET_HAS_v128) \
141     && !defined(TCG_TARGET_HAS_v256)
142 #define TCG_TARGET_MAYBE_vec            0
143 #define TCG_TARGET_HAS_abs_vec          0
144 #define TCG_TARGET_HAS_neg_vec          0
145 #define TCG_TARGET_HAS_not_vec          0
146 #define TCG_TARGET_HAS_andc_vec         0
147 #define TCG_TARGET_HAS_orc_vec          0
148 #define TCG_TARGET_HAS_nand_vec         0
149 #define TCG_TARGET_HAS_nor_vec          0
150 #define TCG_TARGET_HAS_eqv_vec          0
151 #define TCG_TARGET_HAS_roti_vec         0
152 #define TCG_TARGET_HAS_rots_vec         0
153 #define TCG_TARGET_HAS_rotv_vec         0
154 #define TCG_TARGET_HAS_shi_vec          0
155 #define TCG_TARGET_HAS_shs_vec          0
156 #define TCG_TARGET_HAS_shv_vec          0
157 #define TCG_TARGET_HAS_mul_vec          0
158 #define TCG_TARGET_HAS_sat_vec          0
159 #define TCG_TARGET_HAS_minmax_vec       0
160 #define TCG_TARGET_HAS_bitsel_vec       0
161 #define TCG_TARGET_HAS_cmpsel_vec       0
162 #else
163 #define TCG_TARGET_MAYBE_vec            1
164 #endif
165 #ifndef TCG_TARGET_HAS_v64
166 #define TCG_TARGET_HAS_v64              0
167 #endif
168 #ifndef TCG_TARGET_HAS_v128
169 #define TCG_TARGET_HAS_v128             0
170 #endif
171 #ifndef TCG_TARGET_HAS_v256
172 #define TCG_TARGET_HAS_v256             0
173 #endif
174 
175 typedef enum TCGOpcode {
176 #define DEF(name, oargs, iargs, cargs, flags) INDEX_op_ ## name,
177 #include "tcg/tcg-opc.h"
178 #undef DEF
179     NB_OPS,
180 } TCGOpcode;
181 
182 #define tcg_regset_set_reg(d, r)   ((d) |= (TCGRegSet)1 << (r))
183 #define tcg_regset_reset_reg(d, r) ((d) &= ~((TCGRegSet)1 << (r)))
184 #define tcg_regset_test_reg(d, r)  (((d) >> (r)) & 1)
185 
186 #ifndef TCG_TARGET_INSN_UNIT_SIZE
187 # error "Missing TCG_TARGET_INSN_UNIT_SIZE"
188 #elif TCG_TARGET_INSN_UNIT_SIZE == 1
189 typedef uint8_t tcg_insn_unit;
190 #elif TCG_TARGET_INSN_UNIT_SIZE == 2
191 typedef uint16_t tcg_insn_unit;
192 #elif TCG_TARGET_INSN_UNIT_SIZE == 4
193 typedef uint32_t tcg_insn_unit;
194 #elif TCG_TARGET_INSN_UNIT_SIZE == 8
195 typedef uint64_t tcg_insn_unit;
196 #else
197 /* The port better have done this.  */
198 #endif
199 
200 typedef struct TCGRelocation TCGRelocation;
201 struct TCGRelocation {
202     QSIMPLEQ_ENTRY(TCGRelocation) next;
203     tcg_insn_unit *ptr;
204     intptr_t addend;
205     int type;
206 };
207 
208 typedef struct TCGOp TCGOp;
209 typedef struct TCGLabelUse TCGLabelUse;
210 struct TCGLabelUse {
211     QSIMPLEQ_ENTRY(TCGLabelUse) next;
212     TCGOp *op;
213 };
214 
215 typedef struct TCGLabel TCGLabel;
216 struct TCGLabel {
217     bool present;
218     bool has_value;
219     uint16_t id;
220     union {
221         uintptr_t value;
222         const tcg_insn_unit *value_ptr;
223     } u;
224     QSIMPLEQ_HEAD(, TCGLabelUse) branches;
225     QSIMPLEQ_HEAD(, TCGRelocation) relocs;
226     QSIMPLEQ_ENTRY(TCGLabel) next;
227 };
228 
229 typedef struct TCGPool {
230     struct TCGPool *next;
231     int size;
232     uint8_t data[] __attribute__ ((aligned));
233 } TCGPool;
234 
235 #define TCG_POOL_CHUNK_SIZE 32768
236 
237 #define TCG_MAX_TEMPS 512
238 #define TCG_MAX_INSNS 512
239 
240 /* when the size of the arguments of a called function is smaller than
241    this value, they are statically allocated in the TB stack frame */
242 #define TCG_STATIC_CALL_ARGS_SIZE 128
243 
244 typedef enum TCGType {
245     TCG_TYPE_I32,
246     TCG_TYPE_I64,
247     TCG_TYPE_I128,
248 
249     TCG_TYPE_V64,
250     TCG_TYPE_V128,
251     TCG_TYPE_V256,
252 
253     /* Number of different types (integer not enum) */
254 #define TCG_TYPE_COUNT  (TCG_TYPE_V256 + 1)
255 
256     /* An alias for the size of the host register.  */
257 #if TCG_TARGET_REG_BITS == 32
258     TCG_TYPE_REG = TCG_TYPE_I32,
259 #else
260     TCG_TYPE_REG = TCG_TYPE_I64,
261 #endif
262 
263     /* An alias for the size of the native pointer.  */
264 #if UINTPTR_MAX == UINT32_MAX
265     TCG_TYPE_PTR = TCG_TYPE_I32,
266 #else
267     TCG_TYPE_PTR = TCG_TYPE_I64,
268 #endif
269 } TCGType;
270 
271 /**
272  * tcg_type_size
273  * @t: type
274  *
275  * Return the size of the type in bytes.
276  */
277 static inline int tcg_type_size(TCGType t)
278 {
279     unsigned i = t;
280     if (i >= TCG_TYPE_V64) {
281         tcg_debug_assert(i < TCG_TYPE_COUNT);
282         i -= TCG_TYPE_V64 - 1;
283     }
284     return 4 << i;
285 }
286 
287 /**
288  * get_alignment_bits
289  * @memop: MemOp value
290  *
291  * Extract the alignment size from the memop.
292  */
293 static inline unsigned get_alignment_bits(MemOp memop)
294 {
295     unsigned a = memop & MO_AMASK;
296 
297     if (a == MO_UNALN) {
298         /* No alignment required.  */
299         a = 0;
300     } else if (a == MO_ALIGN) {
301         /* A natural alignment requirement.  */
302         a = memop & MO_SIZE;
303     } else {
304         /* A specific alignment requirement.  */
305         a = a >> MO_ASHIFT;
306     }
307     return a;
308 }
309 
310 typedef tcg_target_ulong TCGArg;
311 
312 /* Define type and accessor macros for TCG variables.
313 
314    TCG variables are the inputs and outputs of TCG ops, as described
315    in tcg/README. Target CPU front-end code uses these types to deal
316    with TCG variables as it emits TCG code via the tcg_gen_* functions.
317    They come in several flavours:
318     * TCGv_i32  : 32 bit integer type
319     * TCGv_i64  : 64 bit integer type
320     * TCGv_i128 : 128 bit integer type
321     * TCGv_ptr  : a host pointer type
322     * TCGv_vec  : a host vector type; the exact size is not exposed
323                   to the CPU front-end code.
324     * TCGv      : an integer type the same size as target_ulong
325                   (an alias for either TCGv_i32 or TCGv_i64)
326    The compiler's type checking will complain if you mix them
327    up and pass the wrong sized TCGv to a function.
328 
329    Users of tcg_gen_* don't need to know about any of the internal
330    details of these, and should treat them as opaque types.
331    You won't be able to look inside them in a debugger either.
332 
333    Internal implementation details follow:
334 
335    Note that there is no definition of the structs TCGv_i32_d etc anywhere.
336    This is deliberate, because the values we store in variables of type
337    TCGv_i32 are not really pointers-to-structures. They're just small
338    integers, but keeping them in pointer types like this means that the
339    compiler will complain if you accidentally pass a TCGv_i32 to a
340    function which takes a TCGv_i64, and so on. Only the internals of
341    TCG need to care about the actual contents of the types.  */
342 
343 typedef struct TCGv_i32_d *TCGv_i32;
344 typedef struct TCGv_i64_d *TCGv_i64;
345 typedef struct TCGv_i128_d *TCGv_i128;
346 typedef struct TCGv_ptr_d *TCGv_ptr;
347 typedef struct TCGv_vec_d *TCGv_vec;
348 typedef TCGv_ptr TCGv_env;
349 
350 /* call flags */
351 /* Helper does not read globals (either directly or through an exception). It
352    implies TCG_CALL_NO_WRITE_GLOBALS. */
353 #define TCG_CALL_NO_READ_GLOBALS    0x0001
354 /* Helper does not write globals */
355 #define TCG_CALL_NO_WRITE_GLOBALS   0x0002
356 /* Helper can be safely suppressed if the return value is not used. */
357 #define TCG_CALL_NO_SIDE_EFFECTS    0x0004
358 /* Helper is G_NORETURN.  */
359 #define TCG_CALL_NO_RETURN          0x0008
360 /* Helper is part of Plugins.  */
361 #define TCG_CALL_PLUGIN             0x0010
362 
363 /* convenience version of most used call flags */
364 #define TCG_CALL_NO_RWG         TCG_CALL_NO_READ_GLOBALS
365 #define TCG_CALL_NO_WG          TCG_CALL_NO_WRITE_GLOBALS
366 #define TCG_CALL_NO_SE          TCG_CALL_NO_SIDE_EFFECTS
367 #define TCG_CALL_NO_RWG_SE      (TCG_CALL_NO_RWG | TCG_CALL_NO_SE)
368 #define TCG_CALL_NO_WG_SE       (TCG_CALL_NO_WG | TCG_CALL_NO_SE)
369 
370 /*
371  * Flags for the bswap opcodes.
372  * If IZ, the input is zero-extended, otherwise unknown.
373  * If OZ or OS, the output is zero- or sign-extended respectively,
374  * otherwise the high bits are undefined.
375  */
376 enum {
377     TCG_BSWAP_IZ = 1,
378     TCG_BSWAP_OZ = 2,
379     TCG_BSWAP_OS = 4,
380 };
381 
382 typedef enum TCGTempVal {
383     TEMP_VAL_DEAD,
384     TEMP_VAL_REG,
385     TEMP_VAL_MEM,
386     TEMP_VAL_CONST,
387 } TCGTempVal;
388 
389 typedef enum TCGTempKind {
390     /*
391      * Temp is dead at the end of the extended basic block (EBB),
392      * the single-entry multiple-exit region that falls through
393      * conditional branches.
394      */
395     TEMP_EBB,
396     /* Temp is live across the entire translation block, but dead at end. */
397     TEMP_TB,
398     /* Temp is live across the entire translation block, and between them. */
399     TEMP_GLOBAL,
400     /* Temp is in a fixed register. */
401     TEMP_FIXED,
402     /* Temp is a fixed constant. */
403     TEMP_CONST,
404 } TCGTempKind;
405 
406 typedef struct TCGTemp {
407     TCGReg reg:8;
408     TCGTempVal val_type:8;
409     TCGType base_type:8;
410     TCGType type:8;
411     TCGTempKind kind:3;
412     unsigned int indirect_reg:1;
413     unsigned int indirect_base:1;
414     unsigned int mem_coherent:1;
415     unsigned int mem_allocated:1;
416     unsigned int temp_allocated:1;
417     unsigned int temp_subindex:1;
418 
419     int64_t val;
420     struct TCGTemp *mem_base;
421     intptr_t mem_offset;
422     const char *name;
423 
424     /* Pass-specific information that can be stored for a temporary.
425        One word worth of integer data, and one pointer to data
426        allocated separately.  */
427     uintptr_t state;
428     void *state_ptr;
429 } TCGTemp;
430 
431 typedef struct TCGContext TCGContext;
432 
433 typedef struct TCGTempSet {
434     unsigned long l[BITS_TO_LONGS(TCG_MAX_TEMPS)];
435 } TCGTempSet;
436 
437 /*
438  * With 1 128-bit output, a 32-bit host requires 4 output parameters,
439  * which leaves a maximum of 28 other slots.  Which is enough for 7
440  * 128-bit operands.
441  */
442 #define DEAD_ARG  (1 << 4)
443 #define SYNC_ARG  (1 << 0)
444 typedef uint32_t TCGLifeData;
445 
446 struct TCGOp {
447     TCGOpcode opc   : 8;
448     unsigned nargs  : 8;
449 
450     /* Parameters for this opcode.  See below.  */
451     unsigned param1 : 8;
452     unsigned param2 : 8;
453 
454     /* Lifetime data of the operands.  */
455     TCGLifeData life;
456 
457     /* Next and previous opcodes.  */
458     QTAILQ_ENTRY(TCGOp) link;
459 
460     /* Register preferences for the output(s).  */
461     TCGRegSet output_pref[2];
462 
463     /* Arguments for the opcode.  */
464     TCGArg args[];
465 };
466 
467 #define TCGOP_CALLI(X)    (X)->param1
468 #define TCGOP_CALLO(X)    (X)->param2
469 
470 #define TCGOP_VECL(X)     (X)->param1
471 #define TCGOP_VECE(X)     (X)->param2
472 
473 /* Make sure operands fit in the bitfields above.  */
474 QEMU_BUILD_BUG_ON(NB_OPS > (1 << 8));
475 
476 static inline TCGRegSet output_pref(const TCGOp *op, unsigned i)
477 {
478     return i < ARRAY_SIZE(op->output_pref) ? op->output_pref[i] : 0;
479 }
480 
481 typedef struct TCGProfile {
482     int64_t cpu_exec_time;
483     int64_t tb_count1;
484     int64_t tb_count;
485     int64_t op_count; /* total insn count */
486     int op_count_max; /* max insn per TB */
487     int temp_count_max;
488     int64_t temp_count;
489     int64_t del_op_count;
490     int64_t code_in_len;
491     int64_t code_out_len;
492     int64_t search_out_len;
493     int64_t interm_time;
494     int64_t code_time;
495     int64_t la_time;
496     int64_t opt_time;
497     int64_t restore_count;
498     int64_t restore_time;
499     int64_t table_op_count[NB_OPS];
500 } TCGProfile;
501 
502 struct TCGContext {
503     uint8_t *pool_cur, *pool_end;
504     TCGPool *pool_first, *pool_current, *pool_first_large;
505     int nb_labels;
506     int nb_globals;
507     int nb_temps;
508     int nb_indirects;
509     int nb_ops;
510     TCGType addr_type;            /* TCG_TYPE_I32 or TCG_TYPE_I64 */
511 
512 #ifdef CONFIG_SOFTMMU
513     int tlb_fast_offset;
514     int page_mask;
515     uint8_t page_bits;
516     uint8_t tlb_dyn_max_bits;
517 #endif
518     uint8_t insn_start_words;
519     TCGBar guest_mo;
520 
521     TCGRegSet reserved_regs;
522     intptr_t current_frame_offset;
523     intptr_t frame_start;
524     intptr_t frame_end;
525     TCGTemp *frame_temp;
526 
527     TranslationBlock *gen_tb;     /* tb for which code is being generated */
528     tcg_insn_unit *code_buf;      /* pointer for start of tb */
529     tcg_insn_unit *code_ptr;      /* pointer for running end of tb */
530 
531 #ifdef CONFIG_PROFILER
532     TCGProfile prof;
533 #endif
534 
535 #ifdef CONFIG_DEBUG_TCG
536     int goto_tb_issue_mask;
537     const TCGOpcode *vecop_list;
538 #endif
539 
540     /* Code generation.  Note that we specifically do not use tcg_insn_unit
541        here, because there's too much arithmetic throughout that relies
542        on addition and subtraction working on bytes.  Rely on the GCC
543        extension that allows arithmetic on void*.  */
544     void *code_gen_buffer;
545     size_t code_gen_buffer_size;
546     void *code_gen_ptr;
547     void *data_gen_ptr;
548 
549     /* Threshold to flush the translated code buffer.  */
550     void *code_gen_highwater;
551 
552     /* Track which vCPU triggers events */
553     CPUState *cpu;                      /* *_trans */
554 
555     /* These structures are private to tcg-target.c.inc.  */
556 #ifdef TCG_TARGET_NEED_LDST_LABELS
557     QSIMPLEQ_HEAD(, TCGLabelQemuLdst) ldst_labels;
558 #endif
559 #ifdef TCG_TARGET_NEED_POOL_LABELS
560     struct TCGLabelPoolData *pool_labels;
561 #endif
562 
563     TCGLabel *exitreq_label;
564 
565 #ifdef CONFIG_PLUGIN
566     /*
567      * We keep one plugin_tb struct per TCGContext. Note that on every TB
568      * translation we clear but do not free its contents; this way we
569      * avoid a lot of malloc/free churn, since after a few TB's it's
570      * unlikely that we'll need to allocate either more instructions or more
571      * space for instructions (for variable-instruction-length ISAs).
572      */
573     struct qemu_plugin_tb *plugin_tb;
574 
575     /* descriptor of the instruction being translated */
576     struct qemu_plugin_insn *plugin_insn;
577 #endif
578 
579     GHashTable *const_table[TCG_TYPE_COUNT];
580     TCGTempSet free_temps[TCG_TYPE_COUNT];
581     TCGTemp temps[TCG_MAX_TEMPS]; /* globals first, temps after */
582 
583     QTAILQ_HEAD(, TCGOp) ops, free_ops;
584     QSIMPLEQ_HEAD(, TCGLabel) labels;
585 
586     /* Tells which temporary holds a given register.
587        It does not take into account fixed registers */
588     TCGTemp *reg_to_temp[TCG_TARGET_NB_REGS];
589 
590     uint16_t gen_insn_end_off[TCG_MAX_INSNS];
591     uint64_t *gen_insn_data;
592 
593     /* Exit to translator on overflow. */
594     sigjmp_buf jmp_trans;
595 };
596 
597 static inline bool temp_readonly(TCGTemp *ts)
598 {
599     return ts->kind >= TEMP_FIXED;
600 }
601 
602 extern __thread TCGContext *tcg_ctx;
603 extern const void *tcg_code_gen_epilogue;
604 extern uintptr_t tcg_splitwx_diff;
605 extern TCGv_env cpu_env;
606 
607 bool in_code_gen_buffer(const void *p);
608 
609 #ifdef CONFIG_DEBUG_TCG
610 const void *tcg_splitwx_to_rx(void *rw);
611 void *tcg_splitwx_to_rw(const void *rx);
612 #else
613 static inline const void *tcg_splitwx_to_rx(void *rw)
614 {
615     return rw ? rw + tcg_splitwx_diff : NULL;
616 }
617 
618 static inline void *tcg_splitwx_to_rw(const void *rx)
619 {
620     return rx ? (void *)rx - tcg_splitwx_diff : NULL;
621 }
622 #endif
623 
624 static inline TCGArg temp_arg(TCGTemp *ts)
625 {
626     return (uintptr_t)ts;
627 }
628 
629 static inline TCGTemp *arg_temp(TCGArg a)
630 {
631     return (TCGTemp *)(uintptr_t)a;
632 }
633 
634 #ifdef CONFIG_DEBUG_TCG
635 size_t temp_idx(TCGTemp *ts);
636 TCGTemp *tcgv_i32_temp(TCGv_i32 v);
637 #else
638 static inline size_t temp_idx(TCGTemp *ts)
639 {
640     return ts - tcg_ctx->temps;
641 }
642 
643 /*
644  * Using the offset of a temporary, relative to TCGContext, rather than
645  * its index means that we don't use 0.  That leaves offset 0 free for
646  * a NULL representation without having to leave index 0 unused.
647  */
648 static inline TCGTemp *tcgv_i32_temp(TCGv_i32 v)
649 {
650     return (void *)tcg_ctx + (uintptr_t)v;
651 }
652 #endif
653 
654 static inline TCGTemp *tcgv_i64_temp(TCGv_i64 v)
655 {
656     return tcgv_i32_temp((TCGv_i32)v);
657 }
658 
659 static inline TCGTemp *tcgv_i128_temp(TCGv_i128 v)
660 {
661     return tcgv_i32_temp((TCGv_i32)v);
662 }
663 
664 static inline TCGTemp *tcgv_ptr_temp(TCGv_ptr v)
665 {
666     return tcgv_i32_temp((TCGv_i32)v);
667 }
668 
669 static inline TCGTemp *tcgv_vec_temp(TCGv_vec v)
670 {
671     return tcgv_i32_temp((TCGv_i32)v);
672 }
673 
674 static inline TCGArg tcgv_i32_arg(TCGv_i32 v)
675 {
676     return temp_arg(tcgv_i32_temp(v));
677 }
678 
679 static inline TCGArg tcgv_i64_arg(TCGv_i64 v)
680 {
681     return temp_arg(tcgv_i64_temp(v));
682 }
683 
684 static inline TCGArg tcgv_i128_arg(TCGv_i128 v)
685 {
686     return temp_arg(tcgv_i128_temp(v));
687 }
688 
689 static inline TCGArg tcgv_ptr_arg(TCGv_ptr v)
690 {
691     return temp_arg(tcgv_ptr_temp(v));
692 }
693 
694 static inline TCGArg tcgv_vec_arg(TCGv_vec v)
695 {
696     return temp_arg(tcgv_vec_temp(v));
697 }
698 
699 static inline TCGv_i32 temp_tcgv_i32(TCGTemp *t)
700 {
701     (void)temp_idx(t); /* trigger embedded assert */
702     return (TCGv_i32)((void *)t - (void *)tcg_ctx);
703 }
704 
705 static inline TCGv_i64 temp_tcgv_i64(TCGTemp *t)
706 {
707     return (TCGv_i64)temp_tcgv_i32(t);
708 }
709 
710 static inline TCGv_i128 temp_tcgv_i128(TCGTemp *t)
711 {
712     return (TCGv_i128)temp_tcgv_i32(t);
713 }
714 
715 static inline TCGv_ptr temp_tcgv_ptr(TCGTemp *t)
716 {
717     return (TCGv_ptr)temp_tcgv_i32(t);
718 }
719 
720 static inline TCGv_vec temp_tcgv_vec(TCGTemp *t)
721 {
722     return (TCGv_vec)temp_tcgv_i32(t);
723 }
724 
725 static inline TCGArg tcg_get_insn_param(TCGOp *op, int arg)
726 {
727     return op->args[arg];
728 }
729 
730 static inline void tcg_set_insn_param(TCGOp *op, int arg, TCGArg v)
731 {
732     op->args[arg] = v;
733 }
734 
735 static inline uint64_t tcg_get_insn_start_param(TCGOp *op, int arg)
736 {
737     if (TCG_TARGET_REG_BITS == 64) {
738         return tcg_get_insn_param(op, arg);
739     } else {
740         return deposit64(tcg_get_insn_param(op, arg * 2), 32, 32,
741                          tcg_get_insn_param(op, arg * 2 + 1));
742     }
743 }
744 
745 static inline void tcg_set_insn_start_param(TCGOp *op, int arg, uint64_t v)
746 {
747     if (TCG_TARGET_REG_BITS == 64) {
748         tcg_set_insn_param(op, arg, v);
749     } else {
750         tcg_set_insn_param(op, arg * 2, v);
751         tcg_set_insn_param(op, arg * 2 + 1, v >> 32);
752     }
753 }
754 
755 /* The last op that was emitted.  */
756 static inline TCGOp *tcg_last_op(void)
757 {
758     return QTAILQ_LAST(&tcg_ctx->ops);
759 }
760 
761 /* Test for whether to terminate the TB for using too many opcodes.  */
762 static inline bool tcg_op_buf_full(void)
763 {
764     /* This is not a hard limit, it merely stops translation when
765      * we have produced "enough" opcodes.  We want to limit TB size
766      * such that a RISC host can reasonably use a 16-bit signed
767      * branch within the TB.  We also need to be mindful of the
768      * 16-bit unsigned offsets, TranslationBlock.jmp_reset_offset[]
769      * and TCGContext.gen_insn_end_off[].
770      */
771     return tcg_ctx->nb_ops >= 4000;
772 }
773 
774 /* pool based memory allocation */
775 
776 /* user-mode: mmap_lock must be held for tcg_malloc_internal. */
777 void *tcg_malloc_internal(TCGContext *s, int size);
778 void tcg_pool_reset(TCGContext *s);
779 TranslationBlock *tcg_tb_alloc(TCGContext *s);
780 
781 void tcg_region_reset_all(void);
782 
783 size_t tcg_code_size(void);
784 size_t tcg_code_capacity(void);
785 
786 void tcg_tb_insert(TranslationBlock *tb);
787 void tcg_tb_remove(TranslationBlock *tb);
788 TranslationBlock *tcg_tb_lookup(uintptr_t tc_ptr);
789 void tcg_tb_foreach(GTraverseFunc func, gpointer user_data);
790 size_t tcg_nb_tbs(void);
791 
792 /* user-mode: Called with mmap_lock held.  */
793 static inline void *tcg_malloc(int size)
794 {
795     TCGContext *s = tcg_ctx;
796     uint8_t *ptr, *ptr_end;
797 
798     /* ??? This is a weak placeholder for minimum malloc alignment.  */
799     size = QEMU_ALIGN_UP(size, 8);
800 
801     ptr = s->pool_cur;
802     ptr_end = ptr + size;
803     if (unlikely(ptr_end > s->pool_end)) {
804         return tcg_malloc_internal(tcg_ctx, size);
805     } else {
806         s->pool_cur = ptr_end;
807         return ptr;
808     }
809 }
810 
811 void tcg_init(size_t tb_size, int splitwx, unsigned max_cpus);
812 void tcg_register_thread(void);
813 void tcg_prologue_init(TCGContext *s);
814 void tcg_func_start(TCGContext *s);
815 
816 int tcg_gen_code(TCGContext *s, TranslationBlock *tb, uint64_t pc_start);
817 
818 void tb_target_set_jmp_target(const TranslationBlock *, int,
819                               uintptr_t, uintptr_t);
820 
821 void tcg_set_frame(TCGContext *s, TCGReg reg, intptr_t start, intptr_t size);
822 
823 TCGTemp *tcg_global_mem_new_internal(TCGType, TCGv_ptr,
824                                      intptr_t, const char *);
825 TCGTemp *tcg_temp_new_internal(TCGType, TCGTempKind);
826 TCGv_vec tcg_temp_new_vec(TCGType type);
827 TCGv_vec tcg_temp_new_vec_matching(TCGv_vec match);
828 
829 static inline TCGv_i32 tcg_global_mem_new_i32(TCGv_ptr reg, intptr_t offset,
830                                               const char *name)
831 {
832     TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_I32, reg, offset, name);
833     return temp_tcgv_i32(t);
834 }
835 
836 static inline TCGv_i32 tcg_temp_new_i32(void)
837 {
838     TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I32, TEMP_TB);
839     return temp_tcgv_i32(t);
840 }
841 
842 static inline TCGv_i64 tcg_global_mem_new_i64(TCGv_ptr reg, intptr_t offset,
843                                               const char *name)
844 {
845     TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_I64, reg, offset, name);
846     return temp_tcgv_i64(t);
847 }
848 
849 static inline TCGv_i64 tcg_temp_new_i64(void)
850 {
851     TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I64, TEMP_TB);
852     return temp_tcgv_i64(t);
853 }
854 
855 static inline TCGv_i128 tcg_temp_new_i128(void)
856 {
857     TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I128, TEMP_TB);
858     return temp_tcgv_i128(t);
859 }
860 
861 static inline TCGv_ptr tcg_global_mem_new_ptr(TCGv_ptr reg, intptr_t offset,
862                                               const char *name)
863 {
864     TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_PTR, reg, offset, name);
865     return temp_tcgv_ptr(t);
866 }
867 
868 static inline TCGv_ptr tcg_temp_new_ptr(void)
869 {
870     TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_PTR, TEMP_TB);
871     return temp_tcgv_ptr(t);
872 }
873 
874 int64_t tcg_cpu_exec_time(void);
875 void tcg_dump_info(GString *buf);
876 void tcg_dump_op_count(GString *buf);
877 
878 #define TCG_CT_CONST  1 /* any constant of register size */
879 
880 typedef struct TCGArgConstraint {
881     unsigned ct : 16;
882     unsigned alias_index : 4;
883     unsigned sort_index : 4;
884     unsigned pair_index : 4;
885     unsigned pair : 2;  /* 0: none, 1: first, 2: second, 3: second alias */
886     bool oalias : 1;
887     bool ialias : 1;
888     bool newreg : 1;
889     TCGRegSet regs;
890 } TCGArgConstraint;
891 
892 #define TCG_MAX_OP_ARGS 16
893 
894 /* Bits for TCGOpDef->flags, 8 bits available, all used.  */
895 enum {
896     /* Instruction exits the translation block.  */
897     TCG_OPF_BB_EXIT      = 0x01,
898     /* Instruction defines the end of a basic block.  */
899     TCG_OPF_BB_END       = 0x02,
900     /* Instruction clobbers call registers and potentially update globals.  */
901     TCG_OPF_CALL_CLOBBER = 0x04,
902     /* Instruction has side effects: it cannot be removed if its outputs
903        are not used, and might trigger exceptions.  */
904     TCG_OPF_SIDE_EFFECTS = 0x08,
905     /* Instruction operands are 64-bits (otherwise 32-bits).  */
906     TCG_OPF_64BIT        = 0x10,
907     /* Instruction is optional and not implemented by the host, or insn
908        is generic and should not be implemened by the host.  */
909     TCG_OPF_NOT_PRESENT  = 0x20,
910     /* Instruction operands are vectors.  */
911     TCG_OPF_VECTOR       = 0x40,
912     /* Instruction is a conditional branch. */
913     TCG_OPF_COND_BRANCH  = 0x80
914 };
915 
916 typedef struct TCGOpDef {
917     const char *name;
918     uint8_t nb_oargs, nb_iargs, nb_cargs, nb_args;
919     uint8_t flags;
920     TCGArgConstraint *args_ct;
921 } TCGOpDef;
922 
923 extern TCGOpDef tcg_op_defs[];
924 extern const size_t tcg_op_defs_max;
925 
926 typedef struct TCGTargetOpDef {
927     TCGOpcode op;
928     const char *args_ct_str[TCG_MAX_OP_ARGS];
929 } TCGTargetOpDef;
930 
931 bool tcg_op_supported(TCGOpcode op);
932 
933 void tcg_gen_call0(TCGHelperInfo *, TCGTemp *ret);
934 void tcg_gen_call1(TCGHelperInfo *, TCGTemp *ret, TCGTemp *);
935 void tcg_gen_call2(TCGHelperInfo *, TCGTemp *ret, TCGTemp *, TCGTemp *);
936 void tcg_gen_call3(TCGHelperInfo *, TCGTemp *ret, TCGTemp *,
937                    TCGTemp *, TCGTemp *);
938 void tcg_gen_call4(TCGHelperInfo *, TCGTemp *ret, TCGTemp *, TCGTemp *,
939                    TCGTemp *, TCGTemp *);
940 void tcg_gen_call5(TCGHelperInfo *, TCGTemp *ret, TCGTemp *, TCGTemp *,
941                    TCGTemp *, TCGTemp *, TCGTemp *);
942 void tcg_gen_call6(TCGHelperInfo *, TCGTemp *ret, TCGTemp *, TCGTemp *,
943                    TCGTemp *, TCGTemp *, TCGTemp *, TCGTemp *);
944 void tcg_gen_call7(TCGHelperInfo *, TCGTemp *ret, TCGTemp *, TCGTemp *,
945                    TCGTemp *, TCGTemp *, TCGTemp *, TCGTemp *, TCGTemp *);
946 
947 TCGOp *tcg_emit_op(TCGOpcode opc, unsigned nargs);
948 void tcg_op_remove(TCGContext *s, TCGOp *op);
949 TCGOp *tcg_op_insert_before(TCGContext *s, TCGOp *op,
950                             TCGOpcode opc, unsigned nargs);
951 TCGOp *tcg_op_insert_after(TCGContext *s, TCGOp *op,
952                            TCGOpcode opc, unsigned nargs);
953 
954 /**
955  * tcg_remove_ops_after:
956  * @op: target operation
957  *
958  * Discard any opcodes emitted since @op.  Expected usage is to save
959  * a starting point with tcg_last_op(), speculatively emit opcodes,
960  * then decide whether or not to keep those opcodes after the fact.
961  */
962 void tcg_remove_ops_after(TCGOp *op);
963 
964 void tcg_optimize(TCGContext *s);
965 
966 /*
967  * Locate or create a read-only temporary that is a constant.
968  * This kind of temporary need not be freed, but for convenience
969  * will be silently ignored by tcg_temp_free_*.
970  */
971 TCGTemp *tcg_constant_internal(TCGType type, int64_t val);
972 
973 static inline TCGv_i32 tcg_constant_i32(int32_t val)
974 {
975     return temp_tcgv_i32(tcg_constant_internal(TCG_TYPE_I32, val));
976 }
977 
978 static inline TCGv_i64 tcg_constant_i64(int64_t val)
979 {
980     return temp_tcgv_i64(tcg_constant_internal(TCG_TYPE_I64, val));
981 }
982 
983 TCGv_vec tcg_constant_vec(TCGType type, unsigned vece, int64_t val);
984 TCGv_vec tcg_constant_vec_matching(TCGv_vec match, unsigned vece, int64_t val);
985 
986 #if UINTPTR_MAX == UINT32_MAX
987 # define tcg_constant_ptr(x)     ((TCGv_ptr)tcg_constant_i32((intptr_t)(x)))
988 #else
989 # define tcg_constant_ptr(x)     ((TCGv_ptr)tcg_constant_i64((intptr_t)(x)))
990 #endif
991 
992 TCGLabel *gen_new_label(void);
993 
994 /**
995  * label_arg
996  * @l: label
997  *
998  * Encode a label for storage in the TCG opcode stream.
999  */
1000 
1001 static inline TCGArg label_arg(TCGLabel *l)
1002 {
1003     return (uintptr_t)l;
1004 }
1005 
1006 /**
1007  * arg_label
1008  * @i: value
1009  *
1010  * The opposite of label_arg.  Retrieve a label from the
1011  * encoding of the TCG opcode stream.
1012  */
1013 
1014 static inline TCGLabel *arg_label(TCGArg i)
1015 {
1016     return (TCGLabel *)(uintptr_t)i;
1017 }
1018 
1019 /**
1020  * tcg_ptr_byte_diff
1021  * @a, @b: addresses to be differenced
1022  *
1023  * There are many places within the TCG backends where we need a byte
1024  * difference between two pointers.  While this can be accomplished
1025  * with local casting, it's easy to get wrong -- especially if one is
1026  * concerned with the signedness of the result.
1027  *
1028  * This version relies on GCC's void pointer arithmetic to get the
1029  * correct result.
1030  */
1031 
1032 static inline ptrdiff_t tcg_ptr_byte_diff(const void *a, const void *b)
1033 {
1034     return a - b;
1035 }
1036 
1037 /**
1038  * tcg_pcrel_diff
1039  * @s: the tcg context
1040  * @target: address of the target
1041  *
1042  * Produce a pc-relative difference, from the current code_ptr
1043  * to the destination address.
1044  */
1045 
1046 static inline ptrdiff_t tcg_pcrel_diff(TCGContext *s, const void *target)
1047 {
1048     return tcg_ptr_byte_diff(target, tcg_splitwx_to_rx(s->code_ptr));
1049 }
1050 
1051 /**
1052  * tcg_tbrel_diff
1053  * @s: the tcg context
1054  * @target: address of the target
1055  *
1056  * Produce a difference, from the beginning of the current TB code
1057  * to the destination address.
1058  */
1059 static inline ptrdiff_t tcg_tbrel_diff(TCGContext *s, const void *target)
1060 {
1061     return tcg_ptr_byte_diff(target, tcg_splitwx_to_rx(s->code_buf));
1062 }
1063 
1064 /**
1065  * tcg_current_code_size
1066  * @s: the tcg context
1067  *
1068  * Compute the current code size within the translation block.
1069  * This is used to fill in qemu's data structures for goto_tb.
1070  */
1071 
1072 static inline size_t tcg_current_code_size(TCGContext *s)
1073 {
1074     return tcg_ptr_byte_diff(s->code_ptr, s->code_buf);
1075 }
1076 
1077 /**
1078  * tcg_qemu_tb_exec:
1079  * @env: pointer to CPUArchState for the CPU
1080  * @tb_ptr: address of generated code for the TB to execute
1081  *
1082  * Start executing code from a given translation block.
1083  * Where translation blocks have been linked, execution
1084  * may proceed from the given TB into successive ones.
1085  * Control eventually returns only when some action is needed
1086  * from the top-level loop: either control must pass to a TB
1087  * which has not yet been directly linked, or an asynchronous
1088  * event such as an interrupt needs handling.
1089  *
1090  * Return: The return value is the value passed to the corresponding
1091  * tcg_gen_exit_tb() at translation time of the last TB attempted to execute.
1092  * The value is either zero or a 4-byte aligned pointer to that TB combined
1093  * with additional information in its two least significant bits. The
1094  * additional information is encoded as follows:
1095  *  0, 1: the link between this TB and the next is via the specified
1096  *        TB index (0 or 1). That is, we left the TB via (the equivalent
1097  *        of) "goto_tb <index>". The main loop uses this to determine
1098  *        how to link the TB just executed to the next.
1099  *  2:    we are using instruction counting code generation, and we
1100  *        did not start executing this TB because the instruction counter
1101  *        would hit zero midway through it. In this case the pointer
1102  *        returned is the TB we were about to execute, and the caller must
1103  *        arrange to execute the remaining count of instructions.
1104  *  3:    we stopped because the CPU's exit_request flag was set
1105  *        (usually meaning that there is an interrupt that needs to be
1106  *        handled). The pointer returned is the TB we were about to execute
1107  *        when we noticed the pending exit request.
1108  *
1109  * If the bottom two bits indicate an exit-via-index then the CPU
1110  * state is correctly synchronised and ready for execution of the next
1111  * TB (and in particular the guest PC is the address to execute next).
1112  * Otherwise, we gave up on execution of this TB before it started, and
1113  * the caller must fix up the CPU state by calling the CPU's
1114  * synchronize_from_tb() method with the TB pointer we return (falling
1115  * back to calling the CPU's set_pc method with tb->pb if no
1116  * synchronize_from_tb() method exists).
1117  *
1118  * Note that TCG targets may use a different definition of tcg_qemu_tb_exec
1119  * to this default (which just calls the prologue.code emitted by
1120  * tcg_target_qemu_prologue()).
1121  */
1122 #define TB_EXIT_MASK      3
1123 #define TB_EXIT_IDX0      0
1124 #define TB_EXIT_IDX1      1
1125 #define TB_EXIT_IDXMAX    1
1126 #define TB_EXIT_REQUESTED 3
1127 
1128 #ifdef CONFIG_TCG_INTERPRETER
1129 uintptr_t tcg_qemu_tb_exec(CPUArchState *env, const void *tb_ptr);
1130 #else
1131 typedef uintptr_t tcg_prologue_fn(CPUArchState *env, const void *tb_ptr);
1132 extern tcg_prologue_fn *tcg_qemu_tb_exec;
1133 #endif
1134 
1135 void tcg_register_jit(const void *buf, size_t buf_size);
1136 
1137 #if TCG_TARGET_MAYBE_vec
1138 /* Return zero if the tuple (opc, type, vece) is unsupportable;
1139    return > 0 if it is directly supportable;
1140    return < 0 if we must call tcg_expand_vec_op.  */
1141 int tcg_can_emit_vec_op(TCGOpcode, TCGType, unsigned);
1142 #else
1143 static inline int tcg_can_emit_vec_op(TCGOpcode o, TCGType t, unsigned ve)
1144 {
1145     return 0;
1146 }
1147 #endif
1148 
1149 /* Expand the tuple (opc, type, vece) on the given arguments.  */
1150 void tcg_expand_vec_op(TCGOpcode, TCGType, unsigned, TCGArg, ...);
1151 
1152 /* Replicate a constant C accoring to the log2 of the element size.  */
1153 uint64_t dup_const(unsigned vece, uint64_t c);
1154 
1155 #define dup_const(VECE, C)                                         \
1156     (__builtin_constant_p(VECE)                                    \
1157      ? (  (VECE) == MO_8  ? 0x0101010101010101ull * (uint8_t)(C)   \
1158         : (VECE) == MO_16 ? 0x0001000100010001ull * (uint16_t)(C)  \
1159         : (VECE) == MO_32 ? 0x0000000100000001ull * (uint32_t)(C)  \
1160         : (VECE) == MO_64 ? (uint64_t)(C)                          \
1161         : (qemu_build_not_reached_always(), 0))                    \
1162      : dup_const(VECE, C))
1163 
1164 #ifdef CONFIG_DEBUG_TCG
1165 void tcg_assert_listed_vecop(TCGOpcode);
1166 #else
1167 static inline void tcg_assert_listed_vecop(TCGOpcode op) { }
1168 #endif
1169 
1170 static inline const TCGOpcode *tcg_swap_vecop_list(const TCGOpcode *n)
1171 {
1172 #ifdef CONFIG_DEBUG_TCG
1173     const TCGOpcode *o = tcg_ctx->vecop_list;
1174     tcg_ctx->vecop_list = n;
1175     return o;
1176 #else
1177     return NULL;
1178 #endif
1179 }
1180 
1181 bool tcg_can_emit_vecop_list(const TCGOpcode *, TCGType, unsigned);
1182 
1183 #endif /* TCG_H */
1184