1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * BPF JIT compiler for ARM64 4 * 5 * Copyright (C) 2014-2016 Zi Shen Lim <zlim.lnx@gmail.com> 6 */ 7 8 #define pr_fmt(fmt) "bpf_jit: " fmt 9 10 #include <linux/bitfield.h> 11 #include <linux/bpf.h> 12 #include <linux/filter.h> 13 #include <linux/printk.h> 14 #include <linux/slab.h> 15 16 #include <asm/byteorder.h> 17 #include <asm/cacheflush.h> 18 #include <asm/debug-monitors.h> 19 #include <asm/set_memory.h> 20 21 #include "bpf_jit.h" 22 23 #define TMP_REG_1 (MAX_BPF_JIT_REG + 0) 24 #define TMP_REG_2 (MAX_BPF_JIT_REG + 1) 25 #define TCALL_CNT (MAX_BPF_JIT_REG + 2) 26 #define TMP_REG_3 (MAX_BPF_JIT_REG + 3) 27 28 /* Map BPF registers to A64 registers */ 29 static const int bpf2a64[] = { 30 /* return value from in-kernel function, and exit value from eBPF */ 31 [BPF_REG_0] = A64_R(7), 32 /* arguments from eBPF program to in-kernel function */ 33 [BPF_REG_1] = A64_R(0), 34 [BPF_REG_2] = A64_R(1), 35 [BPF_REG_3] = A64_R(2), 36 [BPF_REG_4] = A64_R(3), 37 [BPF_REG_5] = A64_R(4), 38 /* callee saved registers that in-kernel function will preserve */ 39 [BPF_REG_6] = A64_R(19), 40 [BPF_REG_7] = A64_R(20), 41 [BPF_REG_8] = A64_R(21), 42 [BPF_REG_9] = A64_R(22), 43 /* read-only frame pointer to access stack */ 44 [BPF_REG_FP] = A64_R(25), 45 /* temporary registers for internal BPF JIT */ 46 [TMP_REG_1] = A64_R(10), 47 [TMP_REG_2] = A64_R(11), 48 [TMP_REG_3] = A64_R(12), 49 /* tail_call_cnt */ 50 [TCALL_CNT] = A64_R(26), 51 /* temporary register for blinding constants */ 52 [BPF_REG_AX] = A64_R(9), 53 }; 54 55 struct jit_ctx { 56 const struct bpf_prog *prog; 57 int idx; 58 int epilogue_offset; 59 int *offset; 60 int exentry_idx; 61 __le32 *image; 62 u32 stack_size; 63 }; 64 65 static inline void emit(const u32 insn, struct jit_ctx *ctx) 66 { 67 if (ctx->image != NULL) 68 ctx->image[ctx->idx] = cpu_to_le32(insn); 69 70 ctx->idx++; 71 } 72 73 static inline void emit_a64_mov_i(const int is64, const int reg, 74 const s32 val, struct jit_ctx *ctx) 75 { 76 u16 hi = val >> 16; 77 u16 lo = val & 0xffff; 78 79 if (hi & 0x8000) { 80 if (hi == 0xffff) { 81 emit(A64_MOVN(is64, reg, (u16)~lo, 0), ctx); 82 } else { 83 emit(A64_MOVN(is64, reg, (u16)~hi, 16), ctx); 84 if (lo != 0xffff) 85 emit(A64_MOVK(is64, reg, lo, 0), ctx); 86 } 87 } else { 88 emit(A64_MOVZ(is64, reg, lo, 0), ctx); 89 if (hi) 90 emit(A64_MOVK(is64, reg, hi, 16), ctx); 91 } 92 } 93 94 static int i64_i16_blocks(const u64 val, bool inverse) 95 { 96 return (((val >> 0) & 0xffff) != (inverse ? 0xffff : 0x0000)) + 97 (((val >> 16) & 0xffff) != (inverse ? 0xffff : 0x0000)) + 98 (((val >> 32) & 0xffff) != (inverse ? 0xffff : 0x0000)) + 99 (((val >> 48) & 0xffff) != (inverse ? 0xffff : 0x0000)); 100 } 101 102 static inline void emit_a64_mov_i64(const int reg, const u64 val, 103 struct jit_ctx *ctx) 104 { 105 u64 nrm_tmp = val, rev_tmp = ~val; 106 bool inverse; 107 int shift; 108 109 if (!(nrm_tmp >> 32)) 110 return emit_a64_mov_i(0, reg, (u32)val, ctx); 111 112 inverse = i64_i16_blocks(nrm_tmp, true) < i64_i16_blocks(nrm_tmp, false); 113 shift = max(round_down((inverse ? (fls64(rev_tmp) - 1) : 114 (fls64(nrm_tmp) - 1)), 16), 0); 115 if (inverse) 116 emit(A64_MOVN(1, reg, (rev_tmp >> shift) & 0xffff, shift), ctx); 117 else 118 emit(A64_MOVZ(1, reg, (nrm_tmp >> shift) & 0xffff, shift), ctx); 119 shift -= 16; 120 while (shift >= 0) { 121 if (((nrm_tmp >> shift) & 0xffff) != (inverse ? 0xffff : 0x0000)) 122 emit(A64_MOVK(1, reg, (nrm_tmp >> shift) & 0xffff, shift), ctx); 123 shift -= 16; 124 } 125 } 126 127 /* 128 * Kernel addresses in the vmalloc space use at most 48 bits, and the 129 * remaining bits are guaranteed to be 0x1. So we can compose the address 130 * with a fixed length movn/movk/movk sequence. 131 */ 132 static inline void emit_addr_mov_i64(const int reg, const u64 val, 133 struct jit_ctx *ctx) 134 { 135 u64 tmp = val; 136 int shift = 0; 137 138 emit(A64_MOVN(1, reg, ~tmp & 0xffff, shift), ctx); 139 while (shift < 32) { 140 tmp >>= 16; 141 shift += 16; 142 emit(A64_MOVK(1, reg, tmp & 0xffff, shift), ctx); 143 } 144 } 145 146 static inline int bpf2a64_offset(int bpf_insn, int off, 147 const struct jit_ctx *ctx) 148 { 149 /* BPF JMP offset is relative to the next instruction */ 150 bpf_insn++; 151 /* 152 * Whereas arm64 branch instructions encode the offset 153 * from the branch itself, so we must subtract 1 from the 154 * instruction offset. 155 */ 156 return ctx->offset[bpf_insn + off] - (ctx->offset[bpf_insn] - 1); 157 } 158 159 static void jit_fill_hole(void *area, unsigned int size) 160 { 161 __le32 *ptr; 162 /* We are guaranteed to have aligned memory. */ 163 for (ptr = area; size >= sizeof(u32); size -= sizeof(u32)) 164 *ptr++ = cpu_to_le32(AARCH64_BREAK_FAULT); 165 } 166 167 static inline int epilogue_offset(const struct jit_ctx *ctx) 168 { 169 int to = ctx->epilogue_offset; 170 int from = ctx->idx; 171 172 return to - from; 173 } 174 175 static bool is_addsub_imm(u32 imm) 176 { 177 /* Either imm12 or shifted imm12. */ 178 return !(imm & ~0xfff) || !(imm & ~0xfff000); 179 } 180 181 /* Stack must be multiples of 16B */ 182 #define STACK_ALIGN(sz) (((sz) + 15) & ~15) 183 184 /* Tail call offset to jump into */ 185 #if IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) 186 #define PROLOGUE_OFFSET 8 187 #else 188 #define PROLOGUE_OFFSET 7 189 #endif 190 191 static int build_prologue(struct jit_ctx *ctx, bool ebpf_from_cbpf) 192 { 193 const struct bpf_prog *prog = ctx->prog; 194 const u8 r6 = bpf2a64[BPF_REG_6]; 195 const u8 r7 = bpf2a64[BPF_REG_7]; 196 const u8 r8 = bpf2a64[BPF_REG_8]; 197 const u8 r9 = bpf2a64[BPF_REG_9]; 198 const u8 fp = bpf2a64[BPF_REG_FP]; 199 const u8 tcc = bpf2a64[TCALL_CNT]; 200 const int idx0 = ctx->idx; 201 int cur_offset; 202 203 /* 204 * BPF prog stack layout 205 * 206 * high 207 * original A64_SP => 0:+-----+ BPF prologue 208 * |FP/LR| 209 * current A64_FP => -16:+-----+ 210 * | ... | callee saved registers 211 * BPF fp register => -64:+-----+ <= (BPF_FP) 212 * | | 213 * | ... | BPF prog stack 214 * | | 215 * +-----+ <= (BPF_FP - prog->aux->stack_depth) 216 * |RSVD | padding 217 * current A64_SP => +-----+ <= (BPF_FP - ctx->stack_size) 218 * | | 219 * | ... | Function call stack 220 * | | 221 * +-----+ 222 * low 223 * 224 */ 225 226 /* BTI landing pad */ 227 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL)) 228 emit(A64_BTI_C, ctx); 229 230 /* Save FP and LR registers to stay align with ARM64 AAPCS */ 231 emit(A64_PUSH(A64_FP, A64_LR, A64_SP), ctx); 232 emit(A64_MOV(1, A64_FP, A64_SP), ctx); 233 234 /* Save callee-saved registers */ 235 emit(A64_PUSH(r6, r7, A64_SP), ctx); 236 emit(A64_PUSH(r8, r9, A64_SP), ctx); 237 emit(A64_PUSH(fp, tcc, A64_SP), ctx); 238 239 /* Set up BPF prog stack base register */ 240 emit(A64_MOV(1, fp, A64_SP), ctx); 241 242 if (!ebpf_from_cbpf) { 243 /* Initialize tail_call_cnt */ 244 emit(A64_MOVZ(1, tcc, 0, 0), ctx); 245 246 cur_offset = ctx->idx - idx0; 247 if (cur_offset != PROLOGUE_OFFSET) { 248 pr_err_once("PROLOGUE_OFFSET = %d, expected %d!\n", 249 cur_offset, PROLOGUE_OFFSET); 250 return -1; 251 } 252 253 /* BTI landing pad for the tail call, done with a BR */ 254 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL)) 255 emit(A64_BTI_J, ctx); 256 } 257 258 ctx->stack_size = STACK_ALIGN(prog->aux->stack_depth); 259 260 /* Set up function call stack */ 261 emit(A64_SUB_I(1, A64_SP, A64_SP, ctx->stack_size), ctx); 262 return 0; 263 } 264 265 static int out_offset = -1; /* initialized on the first pass of build_body() */ 266 static int emit_bpf_tail_call(struct jit_ctx *ctx) 267 { 268 /* bpf_tail_call(void *prog_ctx, struct bpf_array *array, u64 index) */ 269 const u8 r2 = bpf2a64[BPF_REG_2]; 270 const u8 r3 = bpf2a64[BPF_REG_3]; 271 272 const u8 tmp = bpf2a64[TMP_REG_1]; 273 const u8 prg = bpf2a64[TMP_REG_2]; 274 const u8 tcc = bpf2a64[TCALL_CNT]; 275 const int idx0 = ctx->idx; 276 #define cur_offset (ctx->idx - idx0) 277 #define jmp_offset (out_offset - (cur_offset)) 278 size_t off; 279 280 /* if (index >= array->map.max_entries) 281 * goto out; 282 */ 283 off = offsetof(struct bpf_array, map.max_entries); 284 emit_a64_mov_i64(tmp, off, ctx); 285 emit(A64_LDR32(tmp, r2, tmp), ctx); 286 emit(A64_MOV(0, r3, r3), ctx); 287 emit(A64_CMP(0, r3, tmp), ctx); 288 emit(A64_B_(A64_COND_CS, jmp_offset), ctx); 289 290 /* if (tail_call_cnt > MAX_TAIL_CALL_CNT) 291 * goto out; 292 * tail_call_cnt++; 293 */ 294 emit_a64_mov_i64(tmp, MAX_TAIL_CALL_CNT, ctx); 295 emit(A64_CMP(1, tcc, tmp), ctx); 296 emit(A64_B_(A64_COND_HI, jmp_offset), ctx); 297 emit(A64_ADD_I(1, tcc, tcc, 1), ctx); 298 299 /* prog = array->ptrs[index]; 300 * if (prog == NULL) 301 * goto out; 302 */ 303 off = offsetof(struct bpf_array, ptrs); 304 emit_a64_mov_i64(tmp, off, ctx); 305 emit(A64_ADD(1, tmp, r2, tmp), ctx); 306 emit(A64_LSL(1, prg, r3, 3), ctx); 307 emit(A64_LDR64(prg, tmp, prg), ctx); 308 emit(A64_CBZ(1, prg, jmp_offset), ctx); 309 310 /* goto *(prog->bpf_func + prologue_offset); */ 311 off = offsetof(struct bpf_prog, bpf_func); 312 emit_a64_mov_i64(tmp, off, ctx); 313 emit(A64_LDR64(tmp, prg, tmp), ctx); 314 emit(A64_ADD_I(1, tmp, tmp, sizeof(u32) * PROLOGUE_OFFSET), ctx); 315 emit(A64_ADD_I(1, A64_SP, A64_SP, ctx->stack_size), ctx); 316 emit(A64_BR(tmp), ctx); 317 318 /* out: */ 319 if (out_offset == -1) 320 out_offset = cur_offset; 321 if (cur_offset != out_offset) { 322 pr_err_once("tail_call out_offset = %d, expected %d!\n", 323 cur_offset, out_offset); 324 return -1; 325 } 326 return 0; 327 #undef cur_offset 328 #undef jmp_offset 329 } 330 331 static void build_epilogue(struct jit_ctx *ctx) 332 { 333 const u8 r0 = bpf2a64[BPF_REG_0]; 334 const u8 r6 = bpf2a64[BPF_REG_6]; 335 const u8 r7 = bpf2a64[BPF_REG_7]; 336 const u8 r8 = bpf2a64[BPF_REG_8]; 337 const u8 r9 = bpf2a64[BPF_REG_9]; 338 const u8 fp = bpf2a64[BPF_REG_FP]; 339 340 /* We're done with BPF stack */ 341 emit(A64_ADD_I(1, A64_SP, A64_SP, ctx->stack_size), ctx); 342 343 /* Restore fs (x25) and x26 */ 344 emit(A64_POP(fp, A64_R(26), A64_SP), ctx); 345 346 /* Restore callee-saved register */ 347 emit(A64_POP(r8, r9, A64_SP), ctx); 348 emit(A64_POP(r6, r7, A64_SP), ctx); 349 350 /* Restore FP/LR registers */ 351 emit(A64_POP(A64_FP, A64_LR, A64_SP), ctx); 352 353 /* Set return value */ 354 emit(A64_MOV(1, A64_R(0), r0), ctx); 355 356 emit(A64_RET(A64_LR), ctx); 357 } 358 359 #define BPF_FIXUP_OFFSET_MASK GENMASK(26, 0) 360 #define BPF_FIXUP_REG_MASK GENMASK(31, 27) 361 362 int arm64_bpf_fixup_exception(const struct exception_table_entry *ex, 363 struct pt_regs *regs) 364 { 365 off_t offset = FIELD_GET(BPF_FIXUP_OFFSET_MASK, ex->fixup); 366 int dst_reg = FIELD_GET(BPF_FIXUP_REG_MASK, ex->fixup); 367 368 regs->regs[dst_reg] = 0; 369 regs->pc = (unsigned long)&ex->fixup - offset; 370 return 1; 371 } 372 373 /* For accesses to BTF pointers, add an entry to the exception table */ 374 static int add_exception_handler(const struct bpf_insn *insn, 375 struct jit_ctx *ctx, 376 int dst_reg) 377 { 378 off_t offset; 379 unsigned long pc; 380 struct exception_table_entry *ex; 381 382 if (!ctx->image) 383 /* First pass */ 384 return 0; 385 386 if (BPF_MODE(insn->code) != BPF_PROBE_MEM) 387 return 0; 388 389 if (!ctx->prog->aux->extable || 390 WARN_ON_ONCE(ctx->exentry_idx >= ctx->prog->aux->num_exentries)) 391 return -EINVAL; 392 393 ex = &ctx->prog->aux->extable[ctx->exentry_idx]; 394 pc = (unsigned long)&ctx->image[ctx->idx - 1]; 395 396 offset = pc - (long)&ex->insn; 397 if (WARN_ON_ONCE(offset >= 0 || offset < INT_MIN)) 398 return -ERANGE; 399 ex->insn = offset; 400 401 /* 402 * Since the extable follows the program, the fixup offset is always 403 * negative and limited to BPF_JIT_REGION_SIZE. Store a positive value 404 * to keep things simple, and put the destination register in the upper 405 * bits. We don't need to worry about buildtime or runtime sort 406 * modifying the upper bits because the table is already sorted, and 407 * isn't part of the main exception table. 408 */ 409 offset = (long)&ex->fixup - (pc + AARCH64_INSN_SIZE); 410 if (!FIELD_FIT(BPF_FIXUP_OFFSET_MASK, offset)) 411 return -ERANGE; 412 413 ex->fixup = FIELD_PREP(BPF_FIXUP_OFFSET_MASK, offset) | 414 FIELD_PREP(BPF_FIXUP_REG_MASK, dst_reg); 415 416 ctx->exentry_idx++; 417 return 0; 418 } 419 420 /* JITs an eBPF instruction. 421 * Returns: 422 * 0 - successfully JITed an 8-byte eBPF instruction. 423 * >0 - successfully JITed a 16-byte eBPF instruction. 424 * <0 - failed to JIT. 425 */ 426 static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx, 427 bool extra_pass) 428 { 429 const u8 code = insn->code; 430 const u8 dst = bpf2a64[insn->dst_reg]; 431 const u8 src = bpf2a64[insn->src_reg]; 432 const u8 tmp = bpf2a64[TMP_REG_1]; 433 const u8 tmp2 = bpf2a64[TMP_REG_2]; 434 const u8 tmp3 = bpf2a64[TMP_REG_3]; 435 const s16 off = insn->off; 436 const s32 imm = insn->imm; 437 const int i = insn - ctx->prog->insnsi; 438 const bool is64 = BPF_CLASS(code) == BPF_ALU64 || 439 BPF_CLASS(code) == BPF_JMP; 440 const bool isdw = BPF_SIZE(code) == BPF_DW; 441 u8 jmp_cond, reg; 442 s32 jmp_offset; 443 u32 a64_insn; 444 int ret; 445 446 #define check_imm(bits, imm) do { \ 447 if ((((imm) > 0) && ((imm) >> (bits))) || \ 448 (((imm) < 0) && (~(imm) >> (bits)))) { \ 449 pr_info("[%2d] imm=%d(0x%x) out of range\n", \ 450 i, imm, imm); \ 451 return -EINVAL; \ 452 } \ 453 } while (0) 454 #define check_imm19(imm) check_imm(19, imm) 455 #define check_imm26(imm) check_imm(26, imm) 456 457 switch (code) { 458 /* dst = src */ 459 case BPF_ALU | BPF_MOV | BPF_X: 460 case BPF_ALU64 | BPF_MOV | BPF_X: 461 emit(A64_MOV(is64, dst, src), ctx); 462 break; 463 /* dst = dst OP src */ 464 case BPF_ALU | BPF_ADD | BPF_X: 465 case BPF_ALU64 | BPF_ADD | BPF_X: 466 emit(A64_ADD(is64, dst, dst, src), ctx); 467 break; 468 case BPF_ALU | BPF_SUB | BPF_X: 469 case BPF_ALU64 | BPF_SUB | BPF_X: 470 emit(A64_SUB(is64, dst, dst, src), ctx); 471 break; 472 case BPF_ALU | BPF_AND | BPF_X: 473 case BPF_ALU64 | BPF_AND | BPF_X: 474 emit(A64_AND(is64, dst, dst, src), ctx); 475 break; 476 case BPF_ALU | BPF_OR | BPF_X: 477 case BPF_ALU64 | BPF_OR | BPF_X: 478 emit(A64_ORR(is64, dst, dst, src), ctx); 479 break; 480 case BPF_ALU | BPF_XOR | BPF_X: 481 case BPF_ALU64 | BPF_XOR | BPF_X: 482 emit(A64_EOR(is64, dst, dst, src), ctx); 483 break; 484 case BPF_ALU | BPF_MUL | BPF_X: 485 case BPF_ALU64 | BPF_MUL | BPF_X: 486 emit(A64_MUL(is64, dst, dst, src), ctx); 487 break; 488 case BPF_ALU | BPF_DIV | BPF_X: 489 case BPF_ALU64 | BPF_DIV | BPF_X: 490 case BPF_ALU | BPF_MOD | BPF_X: 491 case BPF_ALU64 | BPF_MOD | BPF_X: 492 switch (BPF_OP(code)) { 493 case BPF_DIV: 494 emit(A64_UDIV(is64, dst, dst, src), ctx); 495 break; 496 case BPF_MOD: 497 emit(A64_UDIV(is64, tmp, dst, src), ctx); 498 emit(A64_MSUB(is64, dst, dst, tmp, src), ctx); 499 break; 500 } 501 break; 502 case BPF_ALU | BPF_LSH | BPF_X: 503 case BPF_ALU64 | BPF_LSH | BPF_X: 504 emit(A64_LSLV(is64, dst, dst, src), ctx); 505 break; 506 case BPF_ALU | BPF_RSH | BPF_X: 507 case BPF_ALU64 | BPF_RSH | BPF_X: 508 emit(A64_LSRV(is64, dst, dst, src), ctx); 509 break; 510 case BPF_ALU | BPF_ARSH | BPF_X: 511 case BPF_ALU64 | BPF_ARSH | BPF_X: 512 emit(A64_ASRV(is64, dst, dst, src), ctx); 513 break; 514 /* dst = -dst */ 515 case BPF_ALU | BPF_NEG: 516 case BPF_ALU64 | BPF_NEG: 517 emit(A64_NEG(is64, dst, dst), ctx); 518 break; 519 /* dst = BSWAP##imm(dst) */ 520 case BPF_ALU | BPF_END | BPF_FROM_LE: 521 case BPF_ALU | BPF_END | BPF_FROM_BE: 522 #ifdef CONFIG_CPU_BIG_ENDIAN 523 if (BPF_SRC(code) == BPF_FROM_BE) 524 goto emit_bswap_uxt; 525 #else /* !CONFIG_CPU_BIG_ENDIAN */ 526 if (BPF_SRC(code) == BPF_FROM_LE) 527 goto emit_bswap_uxt; 528 #endif 529 switch (imm) { 530 case 16: 531 emit(A64_REV16(is64, dst, dst), ctx); 532 /* zero-extend 16 bits into 64 bits */ 533 emit(A64_UXTH(is64, dst, dst), ctx); 534 break; 535 case 32: 536 emit(A64_REV32(is64, dst, dst), ctx); 537 /* upper 32 bits already cleared */ 538 break; 539 case 64: 540 emit(A64_REV64(dst, dst), ctx); 541 break; 542 } 543 break; 544 emit_bswap_uxt: 545 switch (imm) { 546 case 16: 547 /* zero-extend 16 bits into 64 bits */ 548 emit(A64_UXTH(is64, dst, dst), ctx); 549 break; 550 case 32: 551 /* zero-extend 32 bits into 64 bits */ 552 emit(A64_UXTW(is64, dst, dst), ctx); 553 break; 554 case 64: 555 /* nop */ 556 break; 557 } 558 break; 559 /* dst = imm */ 560 case BPF_ALU | BPF_MOV | BPF_K: 561 case BPF_ALU64 | BPF_MOV | BPF_K: 562 emit_a64_mov_i(is64, dst, imm, ctx); 563 break; 564 /* dst = dst OP imm */ 565 case BPF_ALU | BPF_ADD | BPF_K: 566 case BPF_ALU64 | BPF_ADD | BPF_K: 567 if (is_addsub_imm(imm)) { 568 emit(A64_ADD_I(is64, dst, dst, imm), ctx); 569 } else if (is_addsub_imm(-imm)) { 570 emit(A64_SUB_I(is64, dst, dst, -imm), ctx); 571 } else { 572 emit_a64_mov_i(is64, tmp, imm, ctx); 573 emit(A64_ADD(is64, dst, dst, tmp), ctx); 574 } 575 break; 576 case BPF_ALU | BPF_SUB | BPF_K: 577 case BPF_ALU64 | BPF_SUB | BPF_K: 578 if (is_addsub_imm(imm)) { 579 emit(A64_SUB_I(is64, dst, dst, imm), ctx); 580 } else if (is_addsub_imm(-imm)) { 581 emit(A64_ADD_I(is64, dst, dst, -imm), ctx); 582 } else { 583 emit_a64_mov_i(is64, tmp, imm, ctx); 584 emit(A64_SUB(is64, dst, dst, tmp), ctx); 585 } 586 break; 587 case BPF_ALU | BPF_AND | BPF_K: 588 case BPF_ALU64 | BPF_AND | BPF_K: 589 a64_insn = A64_AND_I(is64, dst, dst, imm); 590 if (a64_insn != AARCH64_BREAK_FAULT) { 591 emit(a64_insn, ctx); 592 } else { 593 emit_a64_mov_i(is64, tmp, imm, ctx); 594 emit(A64_AND(is64, dst, dst, tmp), ctx); 595 } 596 break; 597 case BPF_ALU | BPF_OR | BPF_K: 598 case BPF_ALU64 | BPF_OR | BPF_K: 599 a64_insn = A64_ORR_I(is64, dst, dst, imm); 600 if (a64_insn != AARCH64_BREAK_FAULT) { 601 emit(a64_insn, ctx); 602 } else { 603 emit_a64_mov_i(is64, tmp, imm, ctx); 604 emit(A64_ORR(is64, dst, dst, tmp), ctx); 605 } 606 break; 607 case BPF_ALU | BPF_XOR | BPF_K: 608 case BPF_ALU64 | BPF_XOR | BPF_K: 609 a64_insn = A64_EOR_I(is64, dst, dst, imm); 610 if (a64_insn != AARCH64_BREAK_FAULT) { 611 emit(a64_insn, ctx); 612 } else { 613 emit_a64_mov_i(is64, tmp, imm, ctx); 614 emit(A64_EOR(is64, dst, dst, tmp), ctx); 615 } 616 break; 617 case BPF_ALU | BPF_MUL | BPF_K: 618 case BPF_ALU64 | BPF_MUL | BPF_K: 619 emit_a64_mov_i(is64, tmp, imm, ctx); 620 emit(A64_MUL(is64, dst, dst, tmp), ctx); 621 break; 622 case BPF_ALU | BPF_DIV | BPF_K: 623 case BPF_ALU64 | BPF_DIV | BPF_K: 624 emit_a64_mov_i(is64, tmp, imm, ctx); 625 emit(A64_UDIV(is64, dst, dst, tmp), ctx); 626 break; 627 case BPF_ALU | BPF_MOD | BPF_K: 628 case BPF_ALU64 | BPF_MOD | BPF_K: 629 emit_a64_mov_i(is64, tmp2, imm, ctx); 630 emit(A64_UDIV(is64, tmp, dst, tmp2), ctx); 631 emit(A64_MSUB(is64, dst, dst, tmp, tmp2), ctx); 632 break; 633 case BPF_ALU | BPF_LSH | BPF_K: 634 case BPF_ALU64 | BPF_LSH | BPF_K: 635 emit(A64_LSL(is64, dst, dst, imm), ctx); 636 break; 637 case BPF_ALU | BPF_RSH | BPF_K: 638 case BPF_ALU64 | BPF_RSH | BPF_K: 639 emit(A64_LSR(is64, dst, dst, imm), ctx); 640 break; 641 case BPF_ALU | BPF_ARSH | BPF_K: 642 case BPF_ALU64 | BPF_ARSH | BPF_K: 643 emit(A64_ASR(is64, dst, dst, imm), ctx); 644 break; 645 646 /* JUMP off */ 647 case BPF_JMP | BPF_JA: 648 jmp_offset = bpf2a64_offset(i, off, ctx); 649 check_imm26(jmp_offset); 650 emit(A64_B(jmp_offset), ctx); 651 break; 652 /* IF (dst COND src) JUMP off */ 653 case BPF_JMP | BPF_JEQ | BPF_X: 654 case BPF_JMP | BPF_JGT | BPF_X: 655 case BPF_JMP | BPF_JLT | BPF_X: 656 case BPF_JMP | BPF_JGE | BPF_X: 657 case BPF_JMP | BPF_JLE | BPF_X: 658 case BPF_JMP | BPF_JNE | BPF_X: 659 case BPF_JMP | BPF_JSGT | BPF_X: 660 case BPF_JMP | BPF_JSLT | BPF_X: 661 case BPF_JMP | BPF_JSGE | BPF_X: 662 case BPF_JMP | BPF_JSLE | BPF_X: 663 case BPF_JMP32 | BPF_JEQ | BPF_X: 664 case BPF_JMP32 | BPF_JGT | BPF_X: 665 case BPF_JMP32 | BPF_JLT | BPF_X: 666 case BPF_JMP32 | BPF_JGE | BPF_X: 667 case BPF_JMP32 | BPF_JLE | BPF_X: 668 case BPF_JMP32 | BPF_JNE | BPF_X: 669 case BPF_JMP32 | BPF_JSGT | BPF_X: 670 case BPF_JMP32 | BPF_JSLT | BPF_X: 671 case BPF_JMP32 | BPF_JSGE | BPF_X: 672 case BPF_JMP32 | BPF_JSLE | BPF_X: 673 emit(A64_CMP(is64, dst, src), ctx); 674 emit_cond_jmp: 675 jmp_offset = bpf2a64_offset(i, off, ctx); 676 check_imm19(jmp_offset); 677 switch (BPF_OP(code)) { 678 case BPF_JEQ: 679 jmp_cond = A64_COND_EQ; 680 break; 681 case BPF_JGT: 682 jmp_cond = A64_COND_HI; 683 break; 684 case BPF_JLT: 685 jmp_cond = A64_COND_CC; 686 break; 687 case BPF_JGE: 688 jmp_cond = A64_COND_CS; 689 break; 690 case BPF_JLE: 691 jmp_cond = A64_COND_LS; 692 break; 693 case BPF_JSET: 694 case BPF_JNE: 695 jmp_cond = A64_COND_NE; 696 break; 697 case BPF_JSGT: 698 jmp_cond = A64_COND_GT; 699 break; 700 case BPF_JSLT: 701 jmp_cond = A64_COND_LT; 702 break; 703 case BPF_JSGE: 704 jmp_cond = A64_COND_GE; 705 break; 706 case BPF_JSLE: 707 jmp_cond = A64_COND_LE; 708 break; 709 default: 710 return -EFAULT; 711 } 712 emit(A64_B_(jmp_cond, jmp_offset), ctx); 713 break; 714 case BPF_JMP | BPF_JSET | BPF_X: 715 case BPF_JMP32 | BPF_JSET | BPF_X: 716 emit(A64_TST(is64, dst, src), ctx); 717 goto emit_cond_jmp; 718 /* IF (dst COND imm) JUMP off */ 719 case BPF_JMP | BPF_JEQ | BPF_K: 720 case BPF_JMP | BPF_JGT | BPF_K: 721 case BPF_JMP | BPF_JLT | BPF_K: 722 case BPF_JMP | BPF_JGE | BPF_K: 723 case BPF_JMP | BPF_JLE | BPF_K: 724 case BPF_JMP | BPF_JNE | BPF_K: 725 case BPF_JMP | BPF_JSGT | BPF_K: 726 case BPF_JMP | BPF_JSLT | BPF_K: 727 case BPF_JMP | BPF_JSGE | BPF_K: 728 case BPF_JMP | BPF_JSLE | BPF_K: 729 case BPF_JMP32 | BPF_JEQ | BPF_K: 730 case BPF_JMP32 | BPF_JGT | BPF_K: 731 case BPF_JMP32 | BPF_JLT | BPF_K: 732 case BPF_JMP32 | BPF_JGE | BPF_K: 733 case BPF_JMP32 | BPF_JLE | BPF_K: 734 case BPF_JMP32 | BPF_JNE | BPF_K: 735 case BPF_JMP32 | BPF_JSGT | BPF_K: 736 case BPF_JMP32 | BPF_JSLT | BPF_K: 737 case BPF_JMP32 | BPF_JSGE | BPF_K: 738 case BPF_JMP32 | BPF_JSLE | BPF_K: 739 if (is_addsub_imm(imm)) { 740 emit(A64_CMP_I(is64, dst, imm), ctx); 741 } else if (is_addsub_imm(-imm)) { 742 emit(A64_CMN_I(is64, dst, -imm), ctx); 743 } else { 744 emit_a64_mov_i(is64, tmp, imm, ctx); 745 emit(A64_CMP(is64, dst, tmp), ctx); 746 } 747 goto emit_cond_jmp; 748 case BPF_JMP | BPF_JSET | BPF_K: 749 case BPF_JMP32 | BPF_JSET | BPF_K: 750 a64_insn = A64_TST_I(is64, dst, imm); 751 if (a64_insn != AARCH64_BREAK_FAULT) { 752 emit(a64_insn, ctx); 753 } else { 754 emit_a64_mov_i(is64, tmp, imm, ctx); 755 emit(A64_TST(is64, dst, tmp), ctx); 756 } 757 goto emit_cond_jmp; 758 /* function call */ 759 case BPF_JMP | BPF_CALL: 760 { 761 const u8 r0 = bpf2a64[BPF_REG_0]; 762 bool func_addr_fixed; 763 u64 func_addr; 764 765 ret = bpf_jit_get_func_addr(ctx->prog, insn, extra_pass, 766 &func_addr, &func_addr_fixed); 767 if (ret < 0) 768 return ret; 769 emit_addr_mov_i64(tmp, func_addr, ctx); 770 emit(A64_BLR(tmp), ctx); 771 emit(A64_MOV(1, r0, A64_R(0)), ctx); 772 break; 773 } 774 /* tail call */ 775 case BPF_JMP | BPF_TAIL_CALL: 776 if (emit_bpf_tail_call(ctx)) 777 return -EFAULT; 778 break; 779 /* function return */ 780 case BPF_JMP | BPF_EXIT: 781 /* Optimization: when last instruction is EXIT, 782 simply fallthrough to epilogue. */ 783 if (i == ctx->prog->len - 1) 784 break; 785 jmp_offset = epilogue_offset(ctx); 786 check_imm26(jmp_offset); 787 emit(A64_B(jmp_offset), ctx); 788 break; 789 790 /* dst = imm64 */ 791 case BPF_LD | BPF_IMM | BPF_DW: 792 { 793 const struct bpf_insn insn1 = insn[1]; 794 u64 imm64; 795 796 imm64 = (u64)insn1.imm << 32 | (u32)imm; 797 emit_a64_mov_i64(dst, imm64, ctx); 798 799 return 1; 800 } 801 802 /* LDX: dst = *(size *)(src + off) */ 803 case BPF_LDX | BPF_MEM | BPF_W: 804 case BPF_LDX | BPF_MEM | BPF_H: 805 case BPF_LDX | BPF_MEM | BPF_B: 806 case BPF_LDX | BPF_MEM | BPF_DW: 807 case BPF_LDX | BPF_PROBE_MEM | BPF_DW: 808 case BPF_LDX | BPF_PROBE_MEM | BPF_W: 809 case BPF_LDX | BPF_PROBE_MEM | BPF_H: 810 case BPF_LDX | BPF_PROBE_MEM | BPF_B: 811 emit_a64_mov_i(1, tmp, off, ctx); 812 switch (BPF_SIZE(code)) { 813 case BPF_W: 814 emit(A64_LDR32(dst, src, tmp), ctx); 815 break; 816 case BPF_H: 817 emit(A64_LDRH(dst, src, tmp), ctx); 818 break; 819 case BPF_B: 820 emit(A64_LDRB(dst, src, tmp), ctx); 821 break; 822 case BPF_DW: 823 emit(A64_LDR64(dst, src, tmp), ctx); 824 break; 825 } 826 827 ret = add_exception_handler(insn, ctx, dst); 828 if (ret) 829 return ret; 830 break; 831 832 /* ST: *(size *)(dst + off) = imm */ 833 case BPF_ST | BPF_MEM | BPF_W: 834 case BPF_ST | BPF_MEM | BPF_H: 835 case BPF_ST | BPF_MEM | BPF_B: 836 case BPF_ST | BPF_MEM | BPF_DW: 837 /* Load imm to a register then store it */ 838 emit_a64_mov_i(1, tmp2, off, ctx); 839 emit_a64_mov_i(1, tmp, imm, ctx); 840 switch (BPF_SIZE(code)) { 841 case BPF_W: 842 emit(A64_STR32(tmp, dst, tmp2), ctx); 843 break; 844 case BPF_H: 845 emit(A64_STRH(tmp, dst, tmp2), ctx); 846 break; 847 case BPF_B: 848 emit(A64_STRB(tmp, dst, tmp2), ctx); 849 break; 850 case BPF_DW: 851 emit(A64_STR64(tmp, dst, tmp2), ctx); 852 break; 853 } 854 break; 855 856 /* STX: *(size *)(dst + off) = src */ 857 case BPF_STX | BPF_MEM | BPF_W: 858 case BPF_STX | BPF_MEM | BPF_H: 859 case BPF_STX | BPF_MEM | BPF_B: 860 case BPF_STX | BPF_MEM | BPF_DW: 861 emit_a64_mov_i(1, tmp, off, ctx); 862 switch (BPF_SIZE(code)) { 863 case BPF_W: 864 emit(A64_STR32(src, dst, tmp), ctx); 865 break; 866 case BPF_H: 867 emit(A64_STRH(src, dst, tmp), ctx); 868 break; 869 case BPF_B: 870 emit(A64_STRB(src, dst, tmp), ctx); 871 break; 872 case BPF_DW: 873 emit(A64_STR64(src, dst, tmp), ctx); 874 break; 875 } 876 break; 877 878 case BPF_STX | BPF_ATOMIC | BPF_W: 879 case BPF_STX | BPF_ATOMIC | BPF_DW: 880 if (insn->imm != BPF_ADD) { 881 pr_err_once("unknown atomic op code %02x\n", insn->imm); 882 return -EINVAL; 883 } 884 885 /* STX XADD: lock *(u32 *)(dst + off) += src 886 * and 887 * STX XADD: lock *(u64 *)(dst + off) += src 888 */ 889 890 if (!off) { 891 reg = dst; 892 } else { 893 emit_a64_mov_i(1, tmp, off, ctx); 894 emit(A64_ADD(1, tmp, tmp, dst), ctx); 895 reg = tmp; 896 } 897 if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS)) { 898 emit(A64_STADD(isdw, reg, src), ctx); 899 } else { 900 emit(A64_LDXR(isdw, tmp2, reg), ctx); 901 emit(A64_ADD(isdw, tmp2, tmp2, src), ctx); 902 emit(A64_STXR(isdw, tmp2, reg, tmp3), ctx); 903 jmp_offset = -3; 904 check_imm19(jmp_offset); 905 emit(A64_CBNZ(0, tmp3, jmp_offset), ctx); 906 } 907 break; 908 909 default: 910 pr_err_once("unknown opcode %02x\n", code); 911 return -EINVAL; 912 } 913 914 return 0; 915 } 916 917 static int build_body(struct jit_ctx *ctx, bool extra_pass) 918 { 919 const struct bpf_prog *prog = ctx->prog; 920 int i; 921 922 /* 923 * - offset[0] offset of the end of prologue, 924 * start of the 1st instruction. 925 * - offset[1] - offset of the end of 1st instruction, 926 * start of the 2nd instruction 927 * [....] 928 * - offset[3] - offset of the end of 3rd instruction, 929 * start of 4th instruction 930 */ 931 for (i = 0; i < prog->len; i++) { 932 const struct bpf_insn *insn = &prog->insnsi[i]; 933 int ret; 934 935 if (ctx->image == NULL) 936 ctx->offset[i] = ctx->idx; 937 ret = build_insn(insn, ctx, extra_pass); 938 if (ret > 0) { 939 i++; 940 if (ctx->image == NULL) 941 ctx->offset[i] = ctx->idx; 942 continue; 943 } 944 if (ret) 945 return ret; 946 } 947 /* 948 * offset is allocated with prog->len + 1 so fill in 949 * the last element with the offset after the last 950 * instruction (end of program) 951 */ 952 if (ctx->image == NULL) 953 ctx->offset[i] = ctx->idx; 954 955 return 0; 956 } 957 958 static int validate_code(struct jit_ctx *ctx) 959 { 960 int i; 961 962 for (i = 0; i < ctx->idx; i++) { 963 u32 a64_insn = le32_to_cpu(ctx->image[i]); 964 965 if (a64_insn == AARCH64_BREAK_FAULT) 966 return -1; 967 } 968 969 if (WARN_ON_ONCE(ctx->exentry_idx != ctx->prog->aux->num_exentries)) 970 return -1; 971 972 return 0; 973 } 974 975 static inline void bpf_flush_icache(void *start, void *end) 976 { 977 flush_icache_range((unsigned long)start, (unsigned long)end); 978 } 979 980 struct arm64_jit_data { 981 struct bpf_binary_header *header; 982 u8 *image; 983 struct jit_ctx ctx; 984 }; 985 986 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog) 987 { 988 int image_size, prog_size, extable_size; 989 struct bpf_prog *tmp, *orig_prog = prog; 990 struct bpf_binary_header *header; 991 struct arm64_jit_data *jit_data; 992 bool was_classic = bpf_prog_was_classic(prog); 993 bool tmp_blinded = false; 994 bool extra_pass = false; 995 struct jit_ctx ctx; 996 u8 *image_ptr; 997 998 if (!prog->jit_requested) 999 return orig_prog; 1000 1001 tmp = bpf_jit_blind_constants(prog); 1002 /* If blinding was requested and we failed during blinding, 1003 * we must fall back to the interpreter. 1004 */ 1005 if (IS_ERR(tmp)) 1006 return orig_prog; 1007 if (tmp != prog) { 1008 tmp_blinded = true; 1009 prog = tmp; 1010 } 1011 1012 jit_data = prog->aux->jit_data; 1013 if (!jit_data) { 1014 jit_data = kzalloc(sizeof(*jit_data), GFP_KERNEL); 1015 if (!jit_data) { 1016 prog = orig_prog; 1017 goto out; 1018 } 1019 prog->aux->jit_data = jit_data; 1020 } 1021 if (jit_data->ctx.offset) { 1022 ctx = jit_data->ctx; 1023 image_ptr = jit_data->image; 1024 header = jit_data->header; 1025 extra_pass = true; 1026 prog_size = sizeof(u32) * ctx.idx; 1027 goto skip_init_ctx; 1028 } 1029 memset(&ctx, 0, sizeof(ctx)); 1030 ctx.prog = prog; 1031 1032 ctx.offset = kcalloc(prog->len + 1, sizeof(int), GFP_KERNEL); 1033 if (ctx.offset == NULL) { 1034 prog = orig_prog; 1035 goto out_off; 1036 } 1037 1038 /* 1. Initial fake pass to compute ctx->idx. */ 1039 1040 /* Fake pass to fill in ctx->offset. */ 1041 if (build_body(&ctx, extra_pass)) { 1042 prog = orig_prog; 1043 goto out_off; 1044 } 1045 1046 if (build_prologue(&ctx, was_classic)) { 1047 prog = orig_prog; 1048 goto out_off; 1049 } 1050 1051 ctx.epilogue_offset = ctx.idx; 1052 build_epilogue(&ctx); 1053 1054 extable_size = prog->aux->num_exentries * 1055 sizeof(struct exception_table_entry); 1056 1057 /* Now we know the actual image size. */ 1058 prog_size = sizeof(u32) * ctx.idx; 1059 image_size = prog_size + extable_size; 1060 header = bpf_jit_binary_alloc(image_size, &image_ptr, 1061 sizeof(u32), jit_fill_hole); 1062 if (header == NULL) { 1063 prog = orig_prog; 1064 goto out_off; 1065 } 1066 1067 /* 2. Now, the actual pass. */ 1068 1069 ctx.image = (__le32 *)image_ptr; 1070 if (extable_size) 1071 prog->aux->extable = (void *)image_ptr + prog_size; 1072 skip_init_ctx: 1073 ctx.idx = 0; 1074 ctx.exentry_idx = 0; 1075 1076 build_prologue(&ctx, was_classic); 1077 1078 if (build_body(&ctx, extra_pass)) { 1079 bpf_jit_binary_free(header); 1080 prog = orig_prog; 1081 goto out_off; 1082 } 1083 1084 build_epilogue(&ctx); 1085 1086 /* 3. Extra pass to validate JITed code. */ 1087 if (validate_code(&ctx)) { 1088 bpf_jit_binary_free(header); 1089 prog = orig_prog; 1090 goto out_off; 1091 } 1092 1093 /* And we're done. */ 1094 if (bpf_jit_enable > 1) 1095 bpf_jit_dump(prog->len, prog_size, 2, ctx.image); 1096 1097 bpf_flush_icache(header, ctx.image + ctx.idx); 1098 1099 if (!prog->is_func || extra_pass) { 1100 if (extra_pass && ctx.idx != jit_data->ctx.idx) { 1101 pr_err_once("multi-func JIT bug %d != %d\n", 1102 ctx.idx, jit_data->ctx.idx); 1103 bpf_jit_binary_free(header); 1104 prog->bpf_func = NULL; 1105 prog->jited = 0; 1106 goto out_off; 1107 } 1108 bpf_jit_binary_lock_ro(header); 1109 } else { 1110 jit_data->ctx = ctx; 1111 jit_data->image = image_ptr; 1112 jit_data->header = header; 1113 } 1114 prog->bpf_func = (void *)ctx.image; 1115 prog->jited = 1; 1116 prog->jited_len = prog_size; 1117 1118 if (!prog->is_func || extra_pass) { 1119 bpf_prog_fill_jited_linfo(prog, ctx.offset + 1); 1120 out_off: 1121 kfree(ctx.offset); 1122 kfree(jit_data); 1123 prog->aux->jit_data = NULL; 1124 } 1125 out: 1126 if (tmp_blinded) 1127 bpf_jit_prog_release_other(prog, prog == orig_prog ? 1128 tmp : orig_prog); 1129 return prog; 1130 } 1131 1132 void *bpf_jit_alloc_exec(unsigned long size) 1133 { 1134 return __vmalloc_node_range(size, PAGE_SIZE, BPF_JIT_REGION_START, 1135 BPF_JIT_REGION_END, GFP_KERNEL, 1136 PAGE_KERNEL, 0, NUMA_NO_NODE, 1137 __builtin_return_address(0)); 1138 } 1139 1140 void bpf_jit_free_exec(void *addr) 1141 { 1142 return vfree(addr); 1143 } 1144