include/exec/exec-all.h

   1 /*
   2  * internal execution defines for qemu
   3  *
   4  *  Copyright (c) 2003 Fabrice Bellard
   5  *
   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 of the License, or (at your option) any later version.
  10  *
  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.
  15  *
  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/>.
  18  */
  19
  20 #ifndef _EXEC_ALL_H_
  21 #define _EXEC_ALL_H_
  22
  23 #include "qemu-common.h"
  24
  25 /* allow to see translation results - the slowdown should be negligible, so we leave it */
  26 #define DEBUG_DISAS
  27
  28 /* Page tracking code uses ram addresses in system mode, and virtual
  29    addresses in userspace mode.  Define tb_page_addr_t to be an appropriate
  30    type.  */
  31 #if defined(CONFIG_USER_ONLY)
  32 typedef abi_ulong tb_page_addr_t;
  33 #else
  34 typedef ram_addr_t tb_page_addr_t;
  35 #endif
  36
  37 /* is_jmp field values */
  38 #define DISAS_NEXT    0 /* next instruction can be analyzed */
  39 #define DISAS_JUMP    1 /* only pc was modified dynamically */
  40 #define DISAS_UPDATE  2 /* cpu state was modified dynamically */
  41 #define DISAS_TB_JUMP 3 /* only pc was modified statically */
  42
  43 struct TranslationBlock;
  44 typedef struct TranslationBlock TranslationBlock;
  45
  46 /* XXX: make safe guess about sizes */
  47 #define MAX_OP_PER_INSTR 208
  48
  49 #if HOST_LONG_BITS == 32
  50 #define MAX_OPC_PARAM_PER_ARG 2
  51 #else
  52 #define MAX_OPC_PARAM_PER_ARG 1
  53 #endif
  54 #define MAX_OPC_PARAM_IARGS 5
  55 #define MAX_OPC_PARAM_OARGS 1
  56 #define MAX_OPC_PARAM_ARGS (MAX_OPC_PARAM_IARGS + MAX_OPC_PARAM_OARGS)
  57
  58 /* A Call op needs up to 4 + 2N parameters on 32-bit archs,
  59  * and up to 4 + N parameters on 64-bit archs
  60  * (N = number of input arguments + output arguments).  */
  61 #define MAX_OPC_PARAM (4 + (MAX_OPC_PARAM_PER_ARG * MAX_OPC_PARAM_ARGS))
  62 #define OPC_BUF_SIZE 640
  63 #define OPC_MAX_SIZE (OPC_BUF_SIZE - MAX_OP_PER_INSTR)
  64
  65 /* Maximum size a TCG op can expand to.  This is complicated because a
  66    single op may require several host instructions and register reloads.
  67    For now take a wild guess at 192 bytes, which should allow at least
  68    a couple of fixup instructions per argument.  */
  69 #define TCG_MAX_OP_SIZE 192
  70
  71 #define OPPARAM_BUF_SIZE (OPC_BUF_SIZE * MAX_OPC_PARAM)
  72
  73 #include "qemu/log.h"
  74
  75 void gen_intermediate_code(CPUArchState *env, struct TranslationBlock *tb);
  76 void gen_intermediate_code_pc(CPUArchState *env, struct TranslationBlock *tb);
  77 void restore_state_to_opc(CPUArchState *env, struct TranslationBlock *tb,
  78                           int pc_pos);
  79
  80 void cpu_gen_init(void);
  81 int cpu_gen_code(CPUArchState *env, struct TranslationBlock *tb,
  82                  int *gen_code_size_ptr);
  83 bool cpu_restore_state(CPUArchState *env, uintptr_t searched_pc);
  84
  85 void QEMU_NORETURN cpu_resume_from_signal(CPUArchState *env1, void *puc);
  86 void QEMU_NORETURN cpu_io_recompile(CPUArchState *env, uintptr_t retaddr);
  87 TranslationBlock *tb_gen_code(CPUArchState *env,
  88                               target_ulong pc, target_ulong cs_base, int flags,
  89                               int cflags);
  90 void cpu_exec_init(CPUArchState *env);
  91 void QEMU_NORETURN cpu_loop_exit(CPUArchState *env1);
  92 int page_unprotect(target_ulong address, uintptr_t pc, void *puc);
  93 void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
  94                                    int is_cpu_write_access);
  95 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end,
  96                               int is_cpu_write_access);
  97 #if !defined(CONFIG_USER_ONLY)
  98 /* cputlb.c */
  99 void tlb_flush_page(CPUArchState *env, target_ulong addr);
 100 void tlb_flush(CPUArchState *env, int flush_global);
 101 void tlb_set_page(CPUArchState *env, target_ulong vaddr,
 102                   hwaddr paddr, int prot,
 103                   int mmu_idx, target_ulong size);
 104 void tb_invalidate_phys_addr(hwaddr addr);
 105 #else
 106 static inline void tlb_flush_page(CPUArchState *env, target_ulong addr)
 107 {
 108 }
 109
 110 static inline void tlb_flush(CPUArchState *env, int flush_global)
 111 {
 112 }
 113 #endif
 114
 115 #define CODE_GEN_ALIGN           16 /* must be >= of the size of a icache line */
 116
 117 #define CODE_GEN_PHYS_HASH_BITS     15
 118 #define CODE_GEN_PHYS_HASH_SIZE     (1 << CODE_GEN_PHYS_HASH_BITS)
 119
 120 /* estimated block size for TB allocation */
 121 /* XXX: use a per code average code fragment size and modulate it
 122    according to the host CPU */
 123 #if defined(CONFIG_SOFTMMU)
 124 #define CODE_GEN_AVG_BLOCK_SIZE 128
 125 #else
 126 #define CODE_GEN_AVG_BLOCK_SIZE 64
 127 #endif
 128
 129 #if defined(__arm__) || defined(_ARCH_PPC) \
 130     || defined(__x86_64__) || defined(__i386__) \
 131     || defined(__sparc__) || defined(__aarch64__) \
 132     || defined(CONFIG_TCG_INTERPRETER)
 133 #define USE_DIRECT_JUMP
 134 #endif
 135
 136 struct TranslationBlock {
 137     target_ulong pc;   /* simulated PC corresponding to this block (EIP + CS base) */
 138     target_ulong cs_base; /* CS base for this block */
 139     uint64_t flags; /* flags defining in which context the code was generated */
 140     uint16_t size;      /* size of target code for this block (1 <=
 141                            size <= TARGET_PAGE_SIZE) */
 142     uint16_t cflags;    /* compile flags */
 143 #define CF_COUNT_MASK  0x7fff
 144 #define CF_LAST_IO     0x8000 /* Last insn may be an IO access.  */
 145
 146     uint8_t *tc_ptr;    /* pointer to the translated code */
 147     /* next matching tb for physical address. */
 148     struct TranslationBlock *phys_hash_next;
 149     /* first and second physical page containing code. The lower bit
 150        of the pointer tells the index in page_next[] */
 151     struct TranslationBlock *page_next[2];
 152     tb_page_addr_t page_addr[2];
 153
 154     /* the following data are used to directly call another TB from
 155        the code of this one. */
 156     uint16_t tb_next_offset[2]; /* offset of original jump target */
 157 #ifdef USE_DIRECT_JUMP
 158     uint16_t tb_jmp_offset[2]; /* offset of jump instruction */
 159 #else
 160     uintptr_t tb_next[2]; /* address of jump generated code */
 161 #endif
 162     /* list of TBs jumping to this one. This is a circular list using
 163        the two least significant bits of the pointers to tell what is
 164        the next pointer: 0 = jmp_next[0], 1 = jmp_next[1], 2 =
 165        jmp_first */
 166     struct TranslationBlock *jmp_next[2];
 167     struct TranslationBlock *jmp_first;
 168     uint32_t icount;
 169 };
 170
 171 #include "exec/spinlock.h"
 172
 173 typedef struct TBContext TBContext;
 174
 175 struct TBContext {
 176
 177     TranslationBlock *tbs;
 178     TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE];
 179     int nb_tbs;
 180     /* any access to the tbs or the page table must use this lock */
 181     spinlock_t tb_lock;
 182
 183     /* statistics */
 184     int tb_flush_count;
 185     int tb_phys_invalidate_count;
 186
 187     int tb_invalidated_flag;
 188 };
 189
 190 static inline unsigned int tb_jmp_cache_hash_page(target_ulong pc)
 191 {
 192     target_ulong tmp;
 193     tmp = pc ^ (pc >> (TARGET_PAGE_BITS - TB_JMP_PAGE_BITS));
 194     return (tmp >> (TARGET_PAGE_BITS - TB_JMP_PAGE_BITS)) & TB_JMP_PAGE_MASK;
 195 }
 196
 197 static inline unsigned int tb_jmp_cache_hash_func(target_ulong pc)
 198 {
 199     target_ulong tmp;
 200     tmp = pc ^ (pc >> (TARGET_PAGE_BITS - TB_JMP_PAGE_BITS));
 201     return (((tmp >> (TARGET_PAGE_BITS - TB_JMP_PAGE_BITS)) & TB_JMP_PAGE_MASK)
 202             | (tmp & TB_JMP_ADDR_MASK));
 203 }
 204
 205 static inline unsigned int tb_phys_hash_func(tb_page_addr_t pc)
 206 {
 207     return (pc >> 2) & (CODE_GEN_PHYS_HASH_SIZE - 1);
 208 }
 209
 210 void tb_free(TranslationBlock *tb);
 211 void tb_flush(CPUArchState *env);
 212 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr);
 213
 214 #if defined(USE_DIRECT_JUMP)
 215
 216 #if defined(CONFIG_TCG_INTERPRETER)
 217 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
 218 {
 219     /* patch the branch destination */
 220     *(uint32_t *)jmp_addr = addr - (jmp_addr + 4);
 221     /* no need to flush icache explicitly */
 222 }
 223 #elif defined(_ARCH_PPC)
 224 void ppc_tb_set_jmp_target(unsigned long jmp_addr, unsigned long addr);
 225 #define tb_set_jmp_target1 ppc_tb_set_jmp_target
 226 #elif defined(__i386__) || defined(__x86_64__)
 227 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
 228 {
 229     /* patch the branch destination */
 230     *(uint32_t *)jmp_addr = addr - (jmp_addr + 4);
 231     /* no need to flush icache explicitly */
 232 }
 233 #elif defined(__aarch64__)
 234 void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr);
 235 #define tb_set_jmp_target1 aarch64_tb_set_jmp_target
 236 #elif defined(__arm__)
 237 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
 238 {
 239 #if !QEMU_GNUC_PREREQ(4, 1)
 240     register unsigned long _beg __asm ("a1");
 241     register unsigned long _end __asm ("a2");
 242     register unsigned long _flg __asm ("a3");
 243 #endif
 244
 245     /* we could use a ldr pc, [pc, #-4] kind of branch and avoid the flush */
 246     *(uint32_t *)jmp_addr =
 247         (*(uint32_t *)jmp_addr & ~0xffffff)
 248         | (((addr - (jmp_addr + 8)) >> 2) & 0xffffff);
 249
 250 #if QEMU_GNUC_PREREQ(4, 1)
 251     __builtin___clear_cache((char *) jmp_addr, (char *) jmp_addr + 4);
 252 #else
 253     /* flush icache */
 254     _beg = jmp_addr;
 255     _end = jmp_addr + 4;
 256     _flg = 0;
 257     __asm __volatile__ ("swi 0x9f0002" : : "r" (_beg), "r" (_end), "r" (_flg));
 258 #endif
 259 }
 260 #elif defined(__sparc__)
 261 void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr);
 262 #else
 263 #error tb_set_jmp_target1 is missing
 264 #endif
 265
 266 static inline void tb_set_jmp_target(TranslationBlock *tb,
 267                                      int n, uintptr_t addr)
 268 {
 269     uint16_t offset = tb->tb_jmp_offset[n];
 270     tb_set_jmp_target1((uintptr_t)(tb->tc_ptr + offset), addr);
 271 }
 272
 273 #else
 274
 275 /* set the jump target */
 276 static inline void tb_set_jmp_target(TranslationBlock *tb,
 277                                      int n, uintptr_t addr)
 278 {
 279     tb->tb_next[n] = addr;
 280 }
 281
 282 #endif
 283
 284 static inline void tb_add_jump(TranslationBlock *tb, int n,
 285                                TranslationBlock *tb_next)
 286 {
 287     /* NOTE: this test is only needed for thread safety */
 288     if (!tb->jmp_next[n]) {
 289         /* patch the native jump address */
 290         tb_set_jmp_target(tb, n, (uintptr_t)tb_next->tc_ptr);
 291
 292         /* add in TB jmp circular list */
 293         tb->jmp_next[n] = tb_next->jmp_first;
 294         tb_next->jmp_first = (TranslationBlock *)((uintptr_t)(tb) | (n));
 295     }
 296 }
 297
 298 /* The return address may point to the start of the next instruction.
 299    Subtracting one gets us the call instruction itself.  */
 300 #if defined(CONFIG_TCG_INTERPRETER)
 301 extern uintptr_t tci_tb_ptr;
 302 # define GETPC() tci_tb_ptr
 303 #elif defined(__s390__) && !defined(__s390x__)
 304 # define GETPC() \
 305     (((uintptr_t)__builtin_return_address(0) & 0x7fffffffUL) - 1)
 306 #elif defined(__arm__)
 307 /* Thumb return addresses have the low bit set, so we need to subtract two.
 308    This is still safe in ARM mode because instructions are 4 bytes.  */
 309 # define GETPC() ((uintptr_t)__builtin_return_address(0) - 2)
 310 #else
 311 # define GETPC() ((uintptr_t)__builtin_return_address(0) - 1)
 312 #endif
 313
 314 #if defined(CONFIG_QEMU_LDST_OPTIMIZATION) && defined(CONFIG_SOFTMMU)
 315 /* qemu_ld/st optimization split code generation to fast and slow path, thus,
 316    it needs special handling for an MMU helper which is called from the slow
 317    path, to get the fast path's pc without any additional argument.
 318    It uses a tricky solution which embeds the fast path pc into the slow path.
 319
 320    Code flow in slow path:
 321    (1) pre-process
 322    (2) call MMU helper
 323    (3) jump to (5)
 324    (4) fast path information (implementation specific)
 325    (5) post-process (e.g. stack adjust)
 326    (6) jump to corresponding code of the next of fast path
 327  */
 328 # if defined(__i386__) || defined(__x86_64__)
 329 /* To avoid broken disassembling, long jmp is used for embedding fast path pc,
 330    so that the destination is the next code of fast path, though this jmp is
 331    never executed.
 332
 333    call MMU helper
 334    jmp POST_PROC (2byte)    <- GETRA()
 335    jmp NEXT_CODE (5byte)
 336    POST_PROCESS ...         <- GETRA() + 7
 337  */
 338 #  define GETRA() ((uintptr_t)__builtin_return_address(0))
 339 #  define GETPC_LDST() ((uintptr_t)(GETRA() + 7 + \
 340                                     *(int32_t *)((void *)GETRA() + 3) - 1))
 341 # elif defined (_ARCH_PPC) && !defined (_ARCH_PPC64)
 342 #  define GETRA() ((uintptr_t)__builtin_return_address(0))
 343 #  define GETPC_LDST() ((uintptr_t) ((*(int32_t *)(GETRA() - 4)) - 1))
 344 # elif defined(__arm__)
 345 /* We define two insns between the return address and the branch back to
 346    straight-line.  Find and decode that branch insn.  */
 347 #  define GETRA()       ((uintptr_t)__builtin_return_address(0))
 348 #  define GETPC_LDST()  tcg_getpc_ldst(GETRA())
 349 static inline uintptr_t tcg_getpc_ldst(uintptr_t ra)
 350 {
 351     int32_t b;
 352     ra += 8;                    /* skip the two insns */
 353     b = *(int32_t *)ra;         /* load the branch insn */
 354     b = (b << 8) >> (8 - 2);    /* extract the displacement */
 355     ra += 8;                    /* branches are relative to pc+8 */
 356     ra += b;                    /* apply the displacement */
 357     ra -= 4;                    /* return a pointer into the current opcode,
 358                                    not the start of the next opcode  */
 359     return ra;
 360 }
 361 #elif defined(__aarch64__)
 362 #  define GETRA()       ((uintptr_t)__builtin_return_address(0))
 363 #  define GETPC_LDST()  tcg_getpc_ldst(GETRA())
 364 static inline uintptr_t tcg_getpc_ldst(uintptr_t ra)
 365 {
 366     int32_t b;
 367     ra += 4;                    /* skip one instruction */
 368     b = *(int32_t *)ra;         /* load the branch insn */
 369     b = (b << 6) >> (6 - 2);    /* extract the displacement */
 370     ra += b;                    /* apply the displacement  */
 371     ra -= 4;                    /* return a pointer into the current opcode,
 372                                    not the start of the next opcode  */
 373     return ra;
 374 }
 375 # else
 376 #  error "CONFIG_QEMU_LDST_OPTIMIZATION needs GETPC_LDST() implementation!"
 377 # endif
 378 bool is_tcg_gen_code(uintptr_t pc_ptr);
 379 # define GETPC_EXT() (is_tcg_gen_code(GETRA()) ? GETPC_LDST() : GETPC())
 380 #else
 381 # define GETPC_EXT() GETPC()
 382 #endif
 383
 384 #if !defined(CONFIG_USER_ONLY)
 385
 386 struct MemoryRegion *iotlb_to_region(hwaddr index);
 387 bool io_mem_read(struct MemoryRegion *mr, hwaddr addr,
 388                  uint64_t *pvalue, unsigned size);
 389 bool io_mem_write(struct MemoryRegion *mr, hwaddr addr,
 390                   uint64_t value, unsigned size);
 391
 392 void tlb_fill(CPUArchState *env1, target_ulong addr, int is_write, int mmu_idx,
 393               uintptr_t retaddr);
 394
 395 #include "exec/softmmu_defs.h"
 396
 397 #define ACCESS_TYPE (NB_MMU_MODES + 1)
 398 #define MEMSUFFIX _code
 399
 400 #define DATA_SIZE 1
 401 #include "exec/softmmu_header.h"
 402
 403 #define DATA_SIZE 2
 404 #include "exec/softmmu_header.h"
 405
 406 #define DATA_SIZE 4
 407 #include "exec/softmmu_header.h"
 408
 409 #define DATA_SIZE 8
 410 #include "exec/softmmu_header.h"
 411
 412 #undef ACCESS_TYPE
 413 #undef MEMSUFFIX
 414
 415 #endif
 416
 417 #if defined(CONFIG_USER_ONLY)
 418 static inline tb_page_addr_t get_page_addr_code(CPUArchState *env1, target_ulong addr)
 419 {
 420     return addr;
 421 }
 422 #else
 423 /* cputlb.c */
 424 tb_page_addr_t get_page_addr_code(CPUArchState *env1, target_ulong addr);
 425 #endif
 426
 427 typedef void (CPUDebugExcpHandler)(CPUArchState *env);
 428
 429 void cpu_set_debug_excp_handler(CPUDebugExcpHandler *handler);
 430
 431 /* vl.c */
 432 extern int singlestep;
 433
 434 /* cpu-exec.c */
 435 extern volatile sig_atomic_t exit_request;
 436
 437 /* Deterministic execution requires that IO only be performed on the last
 438    instruction of a TB so that interrupts take effect immediately.  */
 439 static inline int can_do_io(CPUArchState *env)
 440 {
 441     CPUState *cpu = ENV_GET_CPU(env);
 442
 443     if (!use_icount) {
 444         return 1;
 445     }
 446     /* If not executing code then assume we are ok.  */
 447     if (cpu->current_tb == NULL) {
 448         return 1;
 449     }
 450     return env->can_do_io != 0;
 451 }
 452
 453 #endif