include/rseq/rseq.h

   1 /* SPDX-License-Identifier: MIT */
   2 /* SPDX-FileCopyrightText: 2016-2022 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> */
   3
   4 /*
   5  * rseq.h
   6  */
   7
   8 #ifndef RSEQ_H
   9 #define RSEQ_H
  10
  11 #include <stdint.h>
  12 #include <stdbool.h>
  13 #include <pthread.h>
  14 #include <signal.h>
  15 #include <sched.h>
  16 #include <errno.h>
  17 #include <stdio.h>
  18 #include <stdlib.h>
  19 #include <stddef.h>
  20 #include <assert.h>
  21 #include <rseq/rseq-abi.h>
  22 #include <rseq/compiler.h>
  23
  24 #ifndef rseq_sizeof_field
  25 #define rseq_sizeof_field(TYPE, MEMBER) sizeof((((TYPE *)0)->MEMBER))
  26 #endif
  27
  28 #ifndef rseq_offsetofend
  29 #define rseq_offsetofend(TYPE, MEMBER) \
  30         (offsetof(TYPE, MEMBER) + rseq_sizeof_field(TYPE, MEMBER))
  31 #endif
  32
  33 /*
  34  * Empty code injection macros, override when testing.
  35  * It is important to consider that the ASM injection macros need to be
  36  * fully reentrant (e.g. do not modify the stack).
  37  */
  38 #ifndef RSEQ_INJECT_ASM
  39 #define RSEQ_INJECT_ASM(n)
  40 #endif
  41
  42 #ifndef RSEQ_INJECT_C
  43 #define RSEQ_INJECT_C(n)
  44 #endif
  45
  46 #ifndef RSEQ_INJECT_INPUT
  47 #define RSEQ_INJECT_INPUT
  48 #endif
  49
  50 #ifndef RSEQ_INJECT_CLOBBER
  51 #define RSEQ_INJECT_CLOBBER
  52 #endif
  53
  54 #ifndef RSEQ_INJECT_FAILED
  55 #define RSEQ_INJECT_FAILED
  56 #endif
  57
  58 /*
  59  * User code can define RSEQ_GET_ABI_OVERRIDE to override the
  60  * rseq_get_abi() implementation, for instance to use glibc's symbols
  61  * directly.
  62  */
  63 #ifndef RSEQ_GET_ABI_OVERRIDE
  64
  65 # include <rseq/rseq-thread-pointer.h>
  66
  67 # ifdef __cplusplus
  68 extern "C" {
  69 # endif
  70
  71 /* Offset from the thread pointer to the rseq area. */
  72 extern ptrdiff_t rseq_offset;
  73
  74 /*
  75  * Size of the registered rseq area. 0 if the registration was
  76  * unsuccessful.
  77  */
  78 extern unsigned int rseq_size;
  79
  80 /* Flags used during rseq registration. */
  81 extern unsigned int rseq_flags;
  82
  83 /*
  84  * rseq feature size supported by the kernel. 0 if the registration was
  85  * unsuccessful.
  86  */
  87 extern unsigned int rseq_feature_size;
  88
  89 static inline struct rseq_abi *rseq_get_abi(void)
  90 {
  91         return (struct rseq_abi *) ((uintptr_t) rseq_thread_pointer() + rseq_offset);
  92 }
  93
  94 # ifdef __cplusplus
  95 }
  96 # endif
  97
  98 #endif /* RSEQ_GET_ABI_OVERRIDE */
  99
 100 enum rseq_mo {
 101         RSEQ_MO_RELAXED = 0,
 102         RSEQ_MO_CONSUME = 1,    /* Unused */
 103         RSEQ_MO_ACQUIRE = 2,    /* Unused */
 104         RSEQ_MO_RELEASE = 3,
 105         RSEQ_MO_ACQ_REL = 4,    /* Unused */
 106         RSEQ_MO_SEQ_CST = 5,    /* Unused */
 107 };
 108
 109 enum rseq_percpu_mode {
 110         RSEQ_PERCPU_CPU_ID = 0,
 111         RSEQ_PERCPU_MM_CID = 1,
 112 };
 113
 114 #define rseq_likely(x)          __builtin_expect(!!(x), 1)
 115 #define rseq_unlikely(x)        __builtin_expect(!!(x), 0)
 116 #define rseq_barrier()          __asm__ __volatile__("" : : : "memory")
 117
 118 #define RSEQ_ACCESS_ONCE(x)     (*(__volatile__  __typeof__(x) *)&(x))
 119 #define RSEQ_WRITE_ONCE(x, v)   __extension__ ({ RSEQ_ACCESS_ONCE(x) = (v); })
 120 #define RSEQ_READ_ONCE(x)       RSEQ_ACCESS_ONCE(x)
 121
 122 #define __rseq_str_1(x) #x
 123 #define __rseq_str(x)           __rseq_str_1(x)
 124
 125 #define rseq_log(fmt, ...)                                                     \
 126         fprintf(stderr, fmt "(in %s() at " __FILE__ ":" __rseq_str(__LINE__)"\n", \
 127                 ## __VA_ARGS__, __func__)
 128
 129 #define rseq_bug(fmt, ...)              \
 130         do {                            \
 131                 rseq_log(fmt, ## __VA_ARGS__); \
 132                 abort();                \
 133         } while (0)
 134
 135 #if defined(__x86_64__) || defined(__i386__)
 136 #include <rseq/rseq-x86.h>
 137 #elif defined(__ARMEL__) || defined(__ARMEB__)
 138 #include <rseq/rseq-arm.h>
 139 #elif defined (__AARCH64EL__)
 140 #include <rseq/rseq-arm64.h>
 141 #elif defined(__PPC__)
 142 #include <rseq/rseq-ppc.h>
 143 #elif defined(__mips__)
 144 #include <rseq/rseq-mips.h>
 145 #elif defined(__s390__)
 146 #include <rseq/rseq-s390.h>
 147 #elif defined(__riscv)
 148 #include <rseq/rseq-riscv.h>
 149 #else
 150 #error unsupported target
 151 #endif
 152
 153 #ifdef __cplusplus
 154 extern "C" {
 155 #endif
 156
 157 /*
 158  * Register rseq for the current thread. This needs to be called once
 159  * by any thread which uses restartable sequences, before they start
 160  * using restartable sequences, to ensure restartable sequences
 161  * succeed. A restartable sequence executed from a non-registered
 162  * thread will always fail.
 163  */
 164 int rseq_register_current_thread(void);
 165
 166 /*
 167  * Unregister rseq for current thread.
 168  */
 169 int rseq_unregister_current_thread(void);
 170
 171 /*
 172  * Restartable sequence fallback for reading the current CPU number.
 173  */
 174 int32_t rseq_fallback_current_cpu(void);
 175
 176 /*
 177  * Restartable sequence fallback for reading the current node number.
 178  */
 179 int32_t rseq_fallback_current_node(void);
 180
 181 enum rseq_available_query {
 182         RSEQ_AVAILABLE_QUERY_KERNEL = 0,
 183         RSEQ_AVAILABLE_QUERY_LIBC = 1,
 184 };
 185
 186 /*
 187  * Returns true if rseq is supported.
 188  */
 189 bool rseq_available(unsigned int query);
 190
 191 /*
 192  * Values returned can be either the current CPU number, -1 (rseq is
 193  * uninitialized), or -2 (rseq initialization has failed).
 194  */
 195 static inline int32_t rseq_current_cpu_raw(void)
 196 {
 197         return RSEQ_READ_ONCE(rseq_get_abi()->cpu_id);
 198 }
 199
 200 /*
 201  * Returns a possible CPU number, which is typically the current CPU.
 202  * The returned CPU number can be used to prepare for an rseq critical
 203  * section, which will confirm whether the cpu number is indeed the
 204  * current one, and whether rseq is initialized.
 205  *
 206  * The CPU number returned by rseq_cpu_start should always be validated
 207  * by passing it to a rseq asm sequence, or by comparing it to the
 208  * return value of rseq_current_cpu_raw() if the rseq asm sequence
 209  * does not need to be invoked.
 210  */
 211 static inline uint32_t rseq_cpu_start(void)
 212 {
 213         return RSEQ_READ_ONCE(rseq_get_abi()->cpu_id_start);
 214 }
 215
 216 static inline uint32_t rseq_current_cpu(void)
 217 {
 218         int32_t cpu;
 219
 220         cpu = rseq_current_cpu_raw();
 221         if (rseq_unlikely(cpu < 0))
 222                 cpu = rseq_fallback_current_cpu();
 223         return cpu;
 224 }
 225
 226 static inline bool rseq_node_id_available(void)
 227 {
 228         return (int) rseq_feature_size >= (int) rseq_offsetofend(struct rseq_abi, node_id);
 229 }
 230
 231 /*
 232  * Current NUMA node number.
 233  */
 234 static inline uint32_t rseq_current_node_id(void)
 235 {
 236         assert(rseq_node_id_available());
 237         return RSEQ_READ_ONCE(rseq_get_abi()->node_id);
 238 }
 239
 240 static inline bool rseq_mm_cid_available(void)
 241 {
 242         return (int) rseq_feature_size >= (int) rseq_offsetofend(struct rseq_abi, mm_cid);
 243 }
 244
 245 static inline uint32_t rseq_current_mm_cid(void)
 246 {
 247         return RSEQ_READ_ONCE(rseq_get_abi()->mm_cid);
 248 }
 249
 250 static inline void rseq_clear_rseq_cs(void)
 251 {
 252         RSEQ_WRITE_ONCE(rseq_get_abi()->rseq_cs.arch.ptr, 0);
 253 }
 254
 255 /*
 256  * rseq_prepare_unload() should be invoked by each thread executing a rseq
 257  * critical section at least once between their last critical section and
 258  * library unload of the library defining the rseq critical section (struct
 259  * rseq_cs) or the code referred to by the struct rseq_cs start_ip and
 260  * post_commit_offset fields. This also applies to use of rseq in code
 261  * generated by JIT: rseq_prepare_unload() should be invoked at least once by
 262  * each thread executing a rseq critical section before reclaim of the memory
 263  * holding the struct rseq_cs or reclaim of the code pointed to by struct
 264  * rseq_cs start_ip and post_commit_offset fields.
 265  */
 266 static inline void rseq_prepare_unload(void)
 267 {
 268         rseq_clear_rseq_cs();
 269 }
 270
 271 static inline __attribute__((always_inline))
 272 int rseq_cmpeqv_storev(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 273                        intptr_t *v, intptr_t expect,
 274                        intptr_t newv, int cpu)
 275 {
 276         if (rseq_mo != RSEQ_MO_RELAXED)
 277                 return -1;
 278         switch (percpu_mode) {
 279         case RSEQ_PERCPU_CPU_ID:
 280                 return rseq_cmpeqv_storev_relaxed_cpu_id(v, expect, newv, cpu);
 281         case RSEQ_PERCPU_MM_CID:
 282                 return rseq_cmpeqv_storev_relaxed_mm_cid(v, expect, newv, cpu);
 283         default:
 284                 return -1;
 285         }
 286 }
 287
 288 /*
 289  * Compare @v against @expectnot. When it does _not_ match, load @v
 290  * into @load, and store the content of *@v + voffp into @v.
 291  */
 292 static inline __attribute__((always_inline))
 293 int rseq_cmpnev_storeoffp_load(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 294                                intptr_t *v, intptr_t expectnot, long voffp, intptr_t *load,
 295                                int cpu)
 296 {
 297         if (rseq_mo != RSEQ_MO_RELAXED)
 298                 return -1;
 299         switch (percpu_mode) {
 300         case RSEQ_PERCPU_CPU_ID:
 301                 return rseq_cmpnev_storeoffp_load_relaxed_cpu_id(v, expectnot, voffp, load, cpu);
 302         case RSEQ_PERCPU_MM_CID:
 303                 return rseq_cmpnev_storeoffp_load_relaxed_mm_cid(v, expectnot, voffp, load, cpu);
 304         default:
 305                 return -1;
 306         }
 307 }
 308
 309 static inline __attribute__((always_inline))
 310 int rseq_addv(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 311               intptr_t *v, intptr_t count, int cpu)
 312 {
 313         if (rseq_mo != RSEQ_MO_RELAXED)
 314                 return -1;
 315         switch (percpu_mode) {
 316         case RSEQ_PERCPU_CPU_ID:
 317                 return rseq_addv_relaxed_cpu_id(v, count, cpu);
 318         case RSEQ_PERCPU_MM_CID:
 319                 return rseq_addv_relaxed_mm_cid(v, count, cpu);
 320         default:
 321                 return -1;
 322         }
 323 }
 324
 325 #ifdef RSEQ_ARCH_HAS_OFFSET_DEREF_ADDV
 326 /*
 327  *   pval = *(ptr+off)
 328  *  *pval += inc;
 329  */
 330 static inline __attribute__((always_inline))
 331 int rseq_offset_deref_addv(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 332                            intptr_t *ptr, long off, intptr_t inc, int cpu)
 333 {
 334         if (rseq_mo != RSEQ_MO_RELAXED)
 335                 return -1;
 336         switch (percpu_mode) {
 337         case RSEQ_PERCPU_CPU_ID:
 338                 return rseq_offset_deref_addv_relaxed_cpu_id(ptr, off, inc, cpu);
 339         case RSEQ_PERCPU_MM_CID:
 340                 return rseq_offset_deref_addv_relaxed_mm_cid(ptr, off, inc, cpu);
 341         default:
 342                 return -1;
 343         }
 344 }
 345 #endif
 346
 347 static inline __attribute__((always_inline))
 348 int rseq_cmpeqv_trystorev_storev(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 349                                  intptr_t *v, intptr_t expect,
 350                                  intptr_t *v2, intptr_t newv2,
 351                                  intptr_t newv, int cpu)
 352 {
 353         switch (rseq_mo) {
 354         case RSEQ_MO_RELAXED:
 355                 switch (percpu_mode) {
 356                 case RSEQ_PERCPU_CPU_ID:
 357                         return rseq_cmpeqv_trystorev_storev_relaxed_cpu_id(v, expect, v2, newv2, newv, cpu);
 358                 case RSEQ_PERCPU_MM_CID:
 359                         return rseq_cmpeqv_trystorev_storev_relaxed_mm_cid(v, expect, v2, newv2, newv, cpu);
 360                 default:
 361                         return -1;
 362                 }
 363         case RSEQ_MO_RELEASE:
 364                 switch (percpu_mode) {
 365                 case RSEQ_PERCPU_CPU_ID:
 366                         return rseq_cmpeqv_trystorev_storev_release_cpu_id(v, expect, v2, newv2, newv, cpu);
 367                 case RSEQ_PERCPU_MM_CID:
 368                         return rseq_cmpeqv_trystorev_storev_release_mm_cid(v, expect, v2, newv2, newv, cpu);
 369                 default:
 370                         return -1;
 371                 }
 372         case RSEQ_MO_ACQUIRE:   /* Fallthrough */
 373         case RSEQ_MO_ACQ_REL:   /* Fallthrough */
 374         case RSEQ_MO_CONSUME:   /* Fallthrough */
 375         case RSEQ_MO_SEQ_CST:   /* Fallthrough */
 376         default:
 377                 return -1;
 378         }
 379 }
 380
 381 static inline __attribute__((always_inline))
 382 int rseq_cmpeqv_cmpeqv_storev(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 383                               intptr_t *v, intptr_t expect,
 384                               intptr_t *v2, intptr_t expect2,
 385                               intptr_t newv, int cpu)
 386 {
 387         if (rseq_mo != RSEQ_MO_RELAXED)
 388                 return -1;
 389         switch (percpu_mode) {
 390         case RSEQ_PERCPU_CPU_ID:
 391                 return rseq_cmpeqv_cmpeqv_storev_relaxed_cpu_id(v, expect, v2, expect2, newv, cpu);
 392         case RSEQ_PERCPU_MM_CID:
 393                 return rseq_cmpeqv_cmpeqv_storev_relaxed_mm_cid(v, expect, v2, expect2, newv, cpu);
 394         default:
 395                 return -1;
 396         }
 397 }
 398
 399 static inline __attribute__((always_inline))
 400 int rseq_cmpeqv_trymemcpy_storev(enum rseq_mo rseq_mo, enum rseq_percpu_mode percpu_mode,
 401                                  intptr_t *v, intptr_t expect,
 402                                  void *dst, void *src, size_t len,
 403                                  intptr_t newv, int cpu)
 404 {
 405         switch (rseq_mo) {
 406         case RSEQ_MO_RELAXED:
 407                 switch (percpu_mode) {
 408                 case RSEQ_PERCPU_CPU_ID:
 409                         return rseq_cmpeqv_trymemcpy_storev_relaxed_cpu_id(v, expect, dst, src, len, newv, cpu);
 410                 case RSEQ_PERCPU_MM_CID:
 411                         return rseq_cmpeqv_trymemcpy_storev_relaxed_mm_cid(v, expect, dst, src, len, newv, cpu);
 412                 default:
 413                         return -1;
 414                 }
 415         case RSEQ_MO_RELEASE:
 416                 switch (percpu_mode) {
 417                 case RSEQ_PERCPU_CPU_ID:
 418                         return rseq_cmpeqv_trymemcpy_storev_release_cpu_id(v, expect, dst, src, len, newv, cpu);
 419                 case RSEQ_PERCPU_MM_CID:
 420                         return rseq_cmpeqv_trymemcpy_storev_release_mm_cid(v, expect, dst, src, len, newv, cpu);
 421                 default:
 422                         return -1;
 423                 }
 424         case RSEQ_MO_ACQUIRE:   /* Fallthrough */
 425         case RSEQ_MO_ACQ_REL:   /* Fallthrough */
 426         case RSEQ_MO_CONSUME:   /* Fallthrough */
 427         case RSEQ_MO_SEQ_CST:   /* Fallthrough */
 428         default:
 429                 return -1;
 430         }
 431 }
 432
 433 #ifdef __cplusplus
 434 }
 435 #endif
 436
 437 #endif  /* RSEQ_H_ */