xref: /openbmc/qemu/include/exec/exec-all.h (revision 61a67f71)
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 #include "exec/tb-context.h"
25 
26 /* allow to see translation results - the slowdown should be negligible, so we leave it */
27 #define DEBUG_DISAS
28 
29 /* Page tracking code uses ram addresses in system mode, and virtual
30    addresses in userspace mode.  Define tb_page_addr_t to be an appropriate
31    type.  */
32 #if defined(CONFIG_USER_ONLY)
33 typedef abi_ulong tb_page_addr_t;
34 #else
35 typedef ram_addr_t tb_page_addr_t;
36 #endif
37 
38 /* is_jmp field values */
39 #define DISAS_NEXT    0 /* next instruction can be analyzed */
40 #define DISAS_JUMP    1 /* only pc was modified dynamically */
41 #define DISAS_UPDATE  2 /* cpu state was modified dynamically */
42 #define DISAS_TB_JUMP 3 /* only pc was modified statically */
43 
44 #include "qemu/log.h"
45 
46 void gen_intermediate_code(CPUArchState *env, struct TranslationBlock *tb);
47 void restore_state_to_opc(CPUArchState *env, struct TranslationBlock *tb,
48                           target_ulong *data);
49 
50 void cpu_gen_init(void);
51 bool cpu_restore_state(CPUState *cpu, uintptr_t searched_pc);
52 
53 void QEMU_NORETURN cpu_loop_exit_noexc(CPUState *cpu);
54 void QEMU_NORETURN cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
55 TranslationBlock *tb_gen_code(CPUState *cpu,
56                               target_ulong pc, target_ulong cs_base,
57                               uint32_t flags,
58                               int cflags);
59 
60 void QEMU_NORETURN cpu_loop_exit(CPUState *cpu);
61 void QEMU_NORETURN cpu_loop_exit_restore(CPUState *cpu, uintptr_t pc);
62 void QEMU_NORETURN cpu_loop_exit_atomic(CPUState *cpu, uintptr_t pc);
63 
64 #if !defined(CONFIG_USER_ONLY)
65 void cpu_reloading_memory_map(void);
66 /**
67  * cpu_address_space_init:
68  * @cpu: CPU to add this address space to
69  * @as: address space to add
70  * @asidx: integer index of this address space
71  *
72  * Add the specified address space to the CPU's cpu_ases list.
73  * The address space added with @asidx 0 is the one used for the
74  * convenience pointer cpu->as.
75  * The target-specific code which registers ASes is responsible
76  * for defining what semantics address space 0, 1, 2, etc have.
77  *
78  * Before the first call to this function, the caller must set
79  * cpu->num_ases to the total number of address spaces it needs
80  * to support.
81  *
82  * Note that with KVM only one address space is supported.
83  */
84 void cpu_address_space_init(CPUState *cpu, AddressSpace *as, int asidx);
85 #endif
86 
87 #if !defined(CONFIG_USER_ONLY) && defined(CONFIG_TCG)
88 /* cputlb.c */
89 /**
90  * tlb_flush_page:
91  * @cpu: CPU whose TLB should be flushed
92  * @addr: virtual address of page to be flushed
93  *
94  * Flush one page from the TLB of the specified CPU, for all
95  * MMU indexes.
96  */
97 void tlb_flush_page(CPUState *cpu, target_ulong addr);
98 /**
99  * tlb_flush_page_all_cpus:
100  * @cpu: src CPU of the flush
101  * @addr: virtual address of page to be flushed
102  *
103  * Flush one page from the TLB of the specified CPU, for all
104  * MMU indexes.
105  */
106 void tlb_flush_page_all_cpus(CPUState *src, target_ulong addr);
107 /**
108  * tlb_flush_page_all_cpus_synced:
109  * @cpu: src CPU of the flush
110  * @addr: virtual address of page to be flushed
111  *
112  * Flush one page from the TLB of the specified CPU, for all MMU
113  * indexes like tlb_flush_page_all_cpus except the source vCPUs work
114  * is scheduled as safe work meaning all flushes will be complete once
115  * the source vCPUs safe work is complete. This will depend on when
116  * the guests translation ends the TB.
117  */
118 void tlb_flush_page_all_cpus_synced(CPUState *src, target_ulong addr);
119 /**
120  * tlb_flush:
121  * @cpu: CPU whose TLB should be flushed
122  *
123  * Flush the entire TLB for the specified CPU. Most CPU architectures
124  * allow the implementation to drop entries from the TLB at any time
125  * so this is generally safe. If more selective flushing is required
126  * use one of the other functions for efficiency.
127  */
128 void tlb_flush(CPUState *cpu);
129 /**
130  * tlb_flush_all_cpus:
131  * @cpu: src CPU of the flush
132  */
133 void tlb_flush_all_cpus(CPUState *src_cpu);
134 /**
135  * tlb_flush_all_cpus_synced:
136  * @cpu: src CPU of the flush
137  *
138  * Like tlb_flush_all_cpus except this except the source vCPUs work is
139  * scheduled as safe work meaning all flushes will be complete once
140  * the source vCPUs safe work is complete. This will depend on when
141  * the guests translation ends the TB.
142  */
143 void tlb_flush_all_cpus_synced(CPUState *src_cpu);
144 /**
145  * tlb_flush_page_by_mmuidx:
146  * @cpu: CPU whose TLB should be flushed
147  * @addr: virtual address of page to be flushed
148  * @idxmap: bitmap of MMU indexes to flush
149  *
150  * Flush one page from the TLB of the specified CPU, for the specified
151  * MMU indexes.
152  */
153 void tlb_flush_page_by_mmuidx(CPUState *cpu, target_ulong addr,
154                               uint16_t idxmap);
155 /**
156  * tlb_flush_page_by_mmuidx_all_cpus:
157  * @cpu: Originating CPU of the flush
158  * @addr: virtual address of page to be flushed
159  * @idxmap: bitmap of MMU indexes to flush
160  *
161  * Flush one page from the TLB of all CPUs, for the specified
162  * MMU indexes.
163  */
164 void tlb_flush_page_by_mmuidx_all_cpus(CPUState *cpu, target_ulong addr,
165                                        uint16_t idxmap);
166 /**
167  * tlb_flush_page_by_mmuidx_all_cpus_synced:
168  * @cpu: Originating CPU of the flush
169  * @addr: virtual address of page to be flushed
170  * @idxmap: bitmap of MMU indexes to flush
171  *
172  * Flush one page from the TLB of all CPUs, for the specified MMU
173  * indexes like tlb_flush_page_by_mmuidx_all_cpus except the source
174  * vCPUs work is scheduled as safe work meaning all flushes will be
175  * complete once  the source vCPUs safe work is complete. This will
176  * depend on when the guests translation ends the TB.
177  */
178 void tlb_flush_page_by_mmuidx_all_cpus_synced(CPUState *cpu, target_ulong addr,
179                                               uint16_t idxmap);
180 /**
181  * tlb_flush_by_mmuidx:
182  * @cpu: CPU whose TLB should be flushed
183  * @wait: If true ensure synchronisation by exiting the cpu_loop
184  * @idxmap: bitmap of MMU indexes to flush
185  *
186  * Flush all entries from the TLB of the specified CPU, for the specified
187  * MMU indexes.
188  */
189 void tlb_flush_by_mmuidx(CPUState *cpu, uint16_t idxmap);
190 /**
191  * tlb_flush_by_mmuidx_all_cpus:
192  * @cpu: Originating CPU of the flush
193  * @idxmap: bitmap of MMU indexes to flush
194  *
195  * Flush all entries from all TLBs of all CPUs, for the specified
196  * MMU indexes.
197  */
198 void tlb_flush_by_mmuidx_all_cpus(CPUState *cpu, uint16_t idxmap);
199 /**
200  * tlb_flush_by_mmuidx_all_cpus_synced:
201  * @cpu: Originating CPU of the flush
202  * @idxmap: bitmap of MMU indexes to flush
203  *
204  * Flush all entries from all TLBs of all CPUs, for the specified
205  * MMU indexes like tlb_flush_by_mmuidx_all_cpus except except the source
206  * vCPUs work is scheduled as safe work meaning all flushes will be
207  * complete once  the source vCPUs safe work is complete. This will
208  * depend on when the guests translation ends the TB.
209  */
210 void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *cpu, uint16_t idxmap);
211 /**
212  * tlb_set_page_with_attrs:
213  * @cpu: CPU to add this TLB entry for
214  * @vaddr: virtual address of page to add entry for
215  * @paddr: physical address of the page
216  * @attrs: memory transaction attributes
217  * @prot: access permissions (PAGE_READ/PAGE_WRITE/PAGE_EXEC bits)
218  * @mmu_idx: MMU index to insert TLB entry for
219  * @size: size of the page in bytes
220  *
221  * Add an entry to this CPU's TLB (a mapping from virtual address
222  * @vaddr to physical address @paddr) with the specified memory
223  * transaction attributes. This is generally called by the target CPU
224  * specific code after it has been called through the tlb_fill()
225  * entry point and performed a successful page table walk to find
226  * the physical address and attributes for the virtual address
227  * which provoked the TLB miss.
228  *
229  * At most one entry for a given virtual address is permitted. Only a
230  * single TARGET_PAGE_SIZE region is mapped; the supplied @size is only
231  * used by tlb_flush_page.
232  */
233 void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
234                              hwaddr paddr, MemTxAttrs attrs,
235                              int prot, int mmu_idx, target_ulong size);
236 /* tlb_set_page:
237  *
238  * This function is equivalent to calling tlb_set_page_with_attrs()
239  * with an @attrs argument of MEMTXATTRS_UNSPECIFIED. It's provided
240  * as a convenience for CPUs which don't use memory transaction attributes.
241  */
242 void tlb_set_page(CPUState *cpu, target_ulong vaddr,
243                   hwaddr paddr, int prot,
244                   int mmu_idx, target_ulong size);
245 void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr);
246 void probe_write(CPUArchState *env, target_ulong addr, int mmu_idx,
247                  uintptr_t retaddr);
248 #else
249 static inline void tlb_flush_page(CPUState *cpu, target_ulong addr)
250 {
251 }
252 static inline void tlb_flush_page_all_cpus(CPUState *src, target_ulong addr)
253 {
254 }
255 static inline void tlb_flush_page_all_cpus_synced(CPUState *src,
256                                                   target_ulong addr)
257 {
258 }
259 static inline void tlb_flush(CPUState *cpu)
260 {
261 }
262 static inline void tlb_flush_all_cpus(CPUState *src_cpu)
263 {
264 }
265 static inline void tlb_flush_all_cpus_synced(CPUState *src_cpu)
266 {
267 }
268 static inline void tlb_flush_page_by_mmuidx(CPUState *cpu,
269                                             target_ulong addr, uint16_t idxmap)
270 {
271 }
272 
273 static inline void tlb_flush_by_mmuidx(CPUState *cpu, uint16_t idxmap)
274 {
275 }
276 static inline void tlb_flush_page_by_mmuidx_all_cpus(CPUState *cpu,
277                                                      target_ulong addr,
278                                                      uint16_t idxmap)
279 {
280 }
281 static inline void tlb_flush_page_by_mmuidx_all_cpus_synced(CPUState *cpu,
282                                                             target_ulong addr,
283                                                             uint16_t idxmap)
284 {
285 }
286 static inline void tlb_flush_by_mmuidx_all_cpus(CPUState *cpu, uint16_t idxmap)
287 {
288 }
289 static inline void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *cpu,
290                                                        uint16_t idxmap)
291 {
292 }
293 static inline void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr)
294 {
295 }
296 #endif
297 
298 #define CODE_GEN_ALIGN           16 /* must be >= of the size of a icache line */
299 
300 /* Estimated block size for TB allocation.  */
301 /* ??? The following is based on a 2015 survey of x86_64 host output.
302    Better would seem to be some sort of dynamically sized TB array,
303    adapting to the block sizes actually being produced.  */
304 #if defined(CONFIG_SOFTMMU)
305 #define CODE_GEN_AVG_BLOCK_SIZE 400
306 #else
307 #define CODE_GEN_AVG_BLOCK_SIZE 150
308 #endif
309 
310 #if defined(_ARCH_PPC) \
311     || defined(__x86_64__) || defined(__i386__) \
312     || defined(__sparc__) || defined(__aarch64__) \
313     || defined(__s390x__) || defined(__mips__) \
314     || defined(CONFIG_TCG_INTERPRETER)
315 /* NOTE: Direct jump patching must be atomic to be thread-safe. */
316 #define USE_DIRECT_JUMP
317 #endif
318 
319 struct TranslationBlock {
320     target_ulong pc;   /* simulated PC corresponding to this block (EIP + CS base) */
321     target_ulong cs_base; /* CS base for this block */
322     uint32_t flags; /* flags defining in which context the code was generated */
323     uint16_t size;      /* size of target code for this block (1 <=
324                            size <= TARGET_PAGE_SIZE) */
325     uint16_t icount;
326     uint32_t cflags;    /* compile flags */
327 #define CF_COUNT_MASK  0x7fff
328 #define CF_LAST_IO     0x8000 /* Last insn may be an IO access.  */
329 #define CF_NOCACHE     0x10000 /* To be freed after execution */
330 #define CF_USE_ICOUNT  0x20000
331 #define CF_IGNORE_ICOUNT 0x40000 /* Do not generate icount code */
332 
333     /* Per-vCPU dynamic tracing state used to generate this TB */
334     uint32_t trace_vcpu_dstate;
335 
336     uint16_t invalid;
337 
338     void *tc_ptr;    /* pointer to the translated code */
339     uint8_t *tc_search;  /* pointer to search data */
340     /* original tb when cflags has CF_NOCACHE */
341     struct TranslationBlock *orig_tb;
342     /* first and second physical page containing code. The lower bit
343        of the pointer tells the index in page_next[] */
344     struct TranslationBlock *page_next[2];
345     tb_page_addr_t page_addr[2];
346 
347     /* The following data are used to directly call another TB from
348      * the code of this one. This can be done either by emitting direct or
349      * indirect native jump instructions. These jumps are reset so that the TB
350      * just continue its execution. The TB can be linked to another one by
351      * setting one of the jump targets (or patching the jump instruction). Only
352      * two of such jumps are supported.
353      */
354     uint16_t jmp_reset_offset[2]; /* offset of original jump target */
355 #define TB_JMP_RESET_OFFSET_INVALID 0xffff /* indicates no jump generated */
356 #ifdef USE_DIRECT_JUMP
357     uint16_t jmp_insn_offset[2]; /* offset of native jump instruction */
358 #else
359     uintptr_t jmp_target_addr[2]; /* target address for indirect jump */
360 #endif
361     /* Each TB has an assosiated circular list of TBs jumping to this one.
362      * jmp_list_first points to the first TB jumping to this one.
363      * jmp_list_next is used to point to the next TB in a list.
364      * Since each TB can have two jumps, it can participate in two lists.
365      * jmp_list_first and jmp_list_next are 4-byte aligned pointers to a
366      * TranslationBlock structure, but the two least significant bits of
367      * them are used to encode which data field of the pointed TB should
368      * be used to traverse the list further from that TB:
369      * 0 => jmp_list_next[0], 1 => jmp_list_next[1], 2 => jmp_list_first.
370      * In other words, 0/1 tells which jump is used in the pointed TB,
371      * and 2 means that this is a pointer back to the target TB of this list.
372      */
373     uintptr_t jmp_list_next[2];
374     uintptr_t jmp_list_first;
375 };
376 
377 void tb_free(TranslationBlock *tb);
378 void tb_flush(CPUState *cpu);
379 void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr);
380 TranslationBlock *tb_htable_lookup(CPUState *cpu, target_ulong pc,
381                                    target_ulong cs_base, uint32_t flags);
382 
383 #if defined(USE_DIRECT_JUMP)
384 
385 #if defined(CONFIG_TCG_INTERPRETER)
386 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
387 {
388     /* patch the branch destination */
389     atomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
390     /* no need to flush icache explicitly */
391 }
392 #elif defined(_ARCH_PPC)
393 void ppc_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr);
394 #define tb_set_jmp_target1 ppc_tb_set_jmp_target
395 #elif defined(__i386__) || defined(__x86_64__)
396 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
397 {
398     /* patch the branch destination */
399     atomic_set((int32_t *)jmp_addr, addr - (jmp_addr + 4));
400     /* no need to flush icache explicitly */
401 }
402 #elif defined(__s390x__)
403 static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
404 {
405     /* patch the branch destination */
406     intptr_t disp = addr - (jmp_addr - 2);
407     atomic_set((int32_t *)jmp_addr, disp / 2);
408     /* no need to flush icache explicitly */
409 }
410 #elif defined(__aarch64__)
411 void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr);
412 #define tb_set_jmp_target1 aarch64_tb_set_jmp_target
413 #elif defined(__sparc__) || defined(__mips__)
414 void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr);
415 #else
416 #error tb_set_jmp_target1 is missing
417 #endif
418 
419 static inline void tb_set_jmp_target(TranslationBlock *tb,
420                                      int n, uintptr_t addr)
421 {
422     uint16_t offset = tb->jmp_insn_offset[n];
423     tb_set_jmp_target1((uintptr_t)(tb->tc_ptr + offset), addr);
424 }
425 
426 #else
427 
428 /* set the jump target */
429 static inline void tb_set_jmp_target(TranslationBlock *tb,
430                                      int n, uintptr_t addr)
431 {
432     tb->jmp_target_addr[n] = addr;
433 }
434 
435 #endif
436 
437 /* Called with tb_lock held.  */
438 static inline void tb_add_jump(TranslationBlock *tb, int n,
439                                TranslationBlock *tb_next)
440 {
441     assert(n < ARRAY_SIZE(tb->jmp_list_next));
442     if (tb->jmp_list_next[n]) {
443         /* Another thread has already done this while we were
444          * outside of the lock; nothing to do in this case */
445         return;
446     }
447     qemu_log_mask_and_addr(CPU_LOG_EXEC, tb->pc,
448                            "Linking TBs %p [" TARGET_FMT_lx
449                            "] index %d -> %p [" TARGET_FMT_lx "]\n",
450                            tb->tc_ptr, tb->pc, n,
451                            tb_next->tc_ptr, tb_next->pc);
452 
453     /* patch the native jump address */
454     tb_set_jmp_target(tb, n, (uintptr_t)tb_next->tc_ptr);
455 
456     /* add in TB jmp circular list */
457     tb->jmp_list_next[n] = tb_next->jmp_list_first;
458     tb_next->jmp_list_first = (uintptr_t)tb | n;
459 }
460 
461 /* GETPC is the true target of the return instruction that we'll execute.  */
462 #if defined(CONFIG_TCG_INTERPRETER)
463 extern uintptr_t tci_tb_ptr;
464 # define GETPC() tci_tb_ptr
465 #else
466 # define GETPC() \
467     ((uintptr_t)__builtin_extract_return_addr(__builtin_return_address(0)))
468 #endif
469 
470 /* The true return address will often point to a host insn that is part of
471    the next translated guest insn.  Adjust the address backward to point to
472    the middle of the call insn.  Subtracting one would do the job except for
473    several compressed mode architectures (arm, mips) which set the low bit
474    to indicate the compressed mode; subtracting two works around that.  It
475    is also the case that there are no host isas that contain a call insn
476    smaller than 4 bytes, so we don't worry about special-casing this.  */
477 #define GETPC_ADJ   2
478 
479 void tb_lock(void);
480 void tb_unlock(void);
481 void tb_lock_reset(void);
482 
483 #if !defined(CONFIG_USER_ONLY)
484 
485 struct MemoryRegion *iotlb_to_region(CPUState *cpu,
486                                      hwaddr index, MemTxAttrs attrs);
487 
488 void tlb_fill(CPUState *cpu, target_ulong addr, MMUAccessType access_type,
489               int mmu_idx, uintptr_t retaddr);
490 
491 #endif
492 
493 #if defined(CONFIG_USER_ONLY)
494 void mmap_lock(void);
495 void mmap_unlock(void);
496 bool have_mmap_lock(void);
497 
498 static inline tb_page_addr_t get_page_addr_code(CPUArchState *env1, target_ulong addr)
499 {
500     return addr;
501 }
502 #else
503 static inline void mmap_lock(void) {}
504 static inline void mmap_unlock(void) {}
505 
506 /* cputlb.c */
507 tb_page_addr_t get_page_addr_code(CPUArchState *env1, target_ulong addr);
508 
509 void tlb_reset_dirty(CPUState *cpu, ram_addr_t start1, ram_addr_t length);
510 void tlb_set_dirty(CPUState *cpu, target_ulong vaddr);
511 
512 /* exec.c */
513 void tb_flush_jmp_cache(CPUState *cpu, target_ulong addr);
514 
515 MemoryRegionSection *
516 address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr,
517                                   hwaddr *xlat, hwaddr *plen);
518 hwaddr memory_region_section_get_iotlb(CPUState *cpu,
519                                        MemoryRegionSection *section,
520                                        target_ulong vaddr,
521                                        hwaddr paddr, hwaddr xlat,
522                                        int prot,
523                                        target_ulong *address);
524 bool memory_region_is_unassigned(MemoryRegion *mr);
525 
526 #endif
527 
528 /* vl.c */
529 extern int singlestep;
530 
531 #endif
532