1 /* bpf_jit_comp.c: BPF JIT compiler for PPC64 2 * 3 * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation 4 * 5 * Based on the x86 BPF compiler, by Eric Dumazet (eric.dumazet@gmail.com) 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; version 2 10 * of the License. 11 */ 12 #include <linux/moduleloader.h> 13 #include <asm/cacheflush.h> 14 #include <linux/netdevice.h> 15 #include <linux/filter.h> 16 #include <linux/if_vlan.h> 17 18 #include "bpf_jit.h" 19 20 int bpf_jit_enable __read_mostly; 21 22 static inline void bpf_flush_icache(void *start, void *end) 23 { 24 smp_wmb(); 25 flush_icache_range((unsigned long)start, (unsigned long)end); 26 } 27 28 static void bpf_jit_build_prologue(struct sk_filter *fp, u32 *image, 29 struct codegen_context *ctx) 30 { 31 int i; 32 const struct sock_filter *filter = fp->insns; 33 34 if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) { 35 /* Make stackframe */ 36 if (ctx->seen & SEEN_DATAREF) { 37 /* If we call any helpers (for loads), save LR */ 38 EMIT(PPC_INST_MFLR | __PPC_RT(R0)); 39 PPC_STD(0, 1, 16); 40 41 /* Back up non-volatile regs. */ 42 PPC_STD(r_D, 1, -(8*(32-r_D))); 43 PPC_STD(r_HL, 1, -(8*(32-r_HL))); 44 } 45 if (ctx->seen & SEEN_MEM) { 46 /* 47 * Conditionally save regs r15-r31 as some will be used 48 * for M[] data. 49 */ 50 for (i = r_M; i < (r_M+16); i++) { 51 if (ctx->seen & (1 << (i-r_M))) 52 PPC_STD(i, 1, -(8*(32-i))); 53 } 54 } 55 EMIT(PPC_INST_STDU | __PPC_RS(R1) | __PPC_RA(R1) | 56 (-BPF_PPC_STACKFRAME & 0xfffc)); 57 } 58 59 if (ctx->seen & SEEN_DATAREF) { 60 /* 61 * If this filter needs to access skb data, 62 * prepare r_D and r_HL: 63 * r_HL = skb->len - skb->data_len 64 * r_D = skb->data 65 */ 66 PPC_LWZ_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff, 67 data_len)); 68 PPC_LWZ_OFFS(r_HL, r_skb, offsetof(struct sk_buff, len)); 69 PPC_SUB(r_HL, r_HL, r_scratch1); 70 PPC_LD_OFFS(r_D, r_skb, offsetof(struct sk_buff, data)); 71 } 72 73 if (ctx->seen & SEEN_XREG) { 74 /* 75 * TODO: Could also detect whether first instr. sets X and 76 * avoid this (as below, with A). 77 */ 78 PPC_LI(r_X, 0); 79 } 80 81 switch (filter[0].code) { 82 case BPF_S_RET_K: 83 case BPF_S_LD_W_LEN: 84 case BPF_S_ANC_PROTOCOL: 85 case BPF_S_ANC_IFINDEX: 86 case BPF_S_ANC_MARK: 87 case BPF_S_ANC_RXHASH: 88 case BPF_S_ANC_VLAN_TAG: 89 case BPF_S_ANC_VLAN_TAG_PRESENT: 90 case BPF_S_ANC_CPU: 91 case BPF_S_ANC_QUEUE: 92 case BPF_S_LD_W_ABS: 93 case BPF_S_LD_H_ABS: 94 case BPF_S_LD_B_ABS: 95 /* first instruction sets A register (or is RET 'constant') */ 96 break; 97 default: 98 /* make sure we dont leak kernel information to user */ 99 PPC_LI(r_A, 0); 100 } 101 } 102 103 static void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx) 104 { 105 int i; 106 107 if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) { 108 PPC_ADDI(1, 1, BPF_PPC_STACKFRAME); 109 if (ctx->seen & SEEN_DATAREF) { 110 PPC_LD(0, 1, 16); 111 PPC_MTLR(0); 112 PPC_LD(r_D, 1, -(8*(32-r_D))); 113 PPC_LD(r_HL, 1, -(8*(32-r_HL))); 114 } 115 if (ctx->seen & SEEN_MEM) { 116 /* Restore any saved non-vol registers */ 117 for (i = r_M; i < (r_M+16); i++) { 118 if (ctx->seen & (1 << (i-r_M))) 119 PPC_LD(i, 1, -(8*(32-i))); 120 } 121 } 122 } 123 /* The RETs have left a return value in R3. */ 124 125 PPC_BLR(); 126 } 127 128 #define CHOOSE_LOAD_FUNC(K, func) \ 129 ((int)K < 0 ? ((int)K >= SKF_LL_OFF ? func##_negative_offset : func) : func##_positive_offset) 130 131 /* Assemble the body code between the prologue & epilogue. */ 132 static int bpf_jit_build_body(struct sk_filter *fp, u32 *image, 133 struct codegen_context *ctx, 134 unsigned int *addrs) 135 { 136 const struct sock_filter *filter = fp->insns; 137 int flen = fp->len; 138 u8 *func; 139 unsigned int true_cond; 140 int i; 141 142 /* Start of epilogue code */ 143 unsigned int exit_addr = addrs[flen]; 144 145 for (i = 0; i < flen; i++) { 146 unsigned int K = filter[i].k; 147 148 /* 149 * addrs[] maps a BPF bytecode address into a real offset from 150 * the start of the body code. 151 */ 152 addrs[i] = ctx->idx * 4; 153 154 switch (filter[i].code) { 155 /*** ALU ops ***/ 156 case BPF_S_ALU_ADD_X: /* A += X; */ 157 ctx->seen |= SEEN_XREG; 158 PPC_ADD(r_A, r_A, r_X); 159 break; 160 case BPF_S_ALU_ADD_K: /* A += K; */ 161 if (!K) 162 break; 163 PPC_ADDI(r_A, r_A, IMM_L(K)); 164 if (K >= 32768) 165 PPC_ADDIS(r_A, r_A, IMM_HA(K)); 166 break; 167 case BPF_S_ALU_SUB_X: /* A -= X; */ 168 ctx->seen |= SEEN_XREG; 169 PPC_SUB(r_A, r_A, r_X); 170 break; 171 case BPF_S_ALU_SUB_K: /* A -= K */ 172 if (!K) 173 break; 174 PPC_ADDI(r_A, r_A, IMM_L(-K)); 175 if (K >= 32768) 176 PPC_ADDIS(r_A, r_A, IMM_HA(-K)); 177 break; 178 case BPF_S_ALU_MUL_X: /* A *= X; */ 179 ctx->seen |= SEEN_XREG; 180 PPC_MUL(r_A, r_A, r_X); 181 break; 182 case BPF_S_ALU_MUL_K: /* A *= K */ 183 if (K < 32768) 184 PPC_MULI(r_A, r_A, K); 185 else { 186 PPC_LI32(r_scratch1, K); 187 PPC_MUL(r_A, r_A, r_scratch1); 188 } 189 break; 190 case BPF_S_ALU_MOD_X: /* A %= X; */ 191 ctx->seen |= SEEN_XREG; 192 PPC_CMPWI(r_X, 0); 193 if (ctx->pc_ret0 != -1) { 194 PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]); 195 } else { 196 PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12); 197 PPC_LI(r_ret, 0); 198 PPC_JMP(exit_addr); 199 } 200 PPC_DIVWU(r_scratch1, r_A, r_X); 201 PPC_MUL(r_scratch1, r_X, r_scratch1); 202 PPC_SUB(r_A, r_A, r_scratch1); 203 break; 204 case BPF_S_ALU_MOD_K: /* A %= K; */ 205 PPC_LI32(r_scratch2, K); 206 PPC_DIVWU(r_scratch1, r_A, r_scratch2); 207 PPC_MUL(r_scratch1, r_scratch2, r_scratch1); 208 PPC_SUB(r_A, r_A, r_scratch1); 209 break; 210 case BPF_S_ALU_DIV_X: /* A /= X; */ 211 ctx->seen |= SEEN_XREG; 212 PPC_CMPWI(r_X, 0); 213 if (ctx->pc_ret0 != -1) { 214 PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]); 215 } else { 216 /* 217 * Exit, returning 0; first pass hits here 218 * (longer worst-case code size). 219 */ 220 PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12); 221 PPC_LI(r_ret, 0); 222 PPC_JMP(exit_addr); 223 } 224 PPC_DIVWU(r_A, r_A, r_X); 225 break; 226 case BPF_S_ALU_DIV_K: /* A = reciprocal_divide(A, K); */ 227 PPC_LI32(r_scratch1, K); 228 /* Top 32 bits of 64bit result -> A */ 229 PPC_MULHWU(r_A, r_A, r_scratch1); 230 break; 231 case BPF_S_ALU_AND_X: 232 ctx->seen |= SEEN_XREG; 233 PPC_AND(r_A, r_A, r_X); 234 break; 235 case BPF_S_ALU_AND_K: 236 if (!IMM_H(K)) 237 PPC_ANDI(r_A, r_A, K); 238 else { 239 PPC_LI32(r_scratch1, K); 240 PPC_AND(r_A, r_A, r_scratch1); 241 } 242 break; 243 case BPF_S_ALU_OR_X: 244 ctx->seen |= SEEN_XREG; 245 PPC_OR(r_A, r_A, r_X); 246 break; 247 case BPF_S_ALU_OR_K: 248 if (IMM_L(K)) 249 PPC_ORI(r_A, r_A, IMM_L(K)); 250 if (K >= 65536) 251 PPC_ORIS(r_A, r_A, IMM_H(K)); 252 break; 253 case BPF_S_ANC_ALU_XOR_X: 254 case BPF_S_ALU_XOR_X: /* A ^= X */ 255 ctx->seen |= SEEN_XREG; 256 PPC_XOR(r_A, r_A, r_X); 257 break; 258 case BPF_S_ALU_XOR_K: /* A ^= K */ 259 if (IMM_L(K)) 260 PPC_XORI(r_A, r_A, IMM_L(K)); 261 if (K >= 65536) 262 PPC_XORIS(r_A, r_A, IMM_H(K)); 263 break; 264 case BPF_S_ALU_LSH_X: /* A <<= X; */ 265 ctx->seen |= SEEN_XREG; 266 PPC_SLW(r_A, r_A, r_X); 267 break; 268 case BPF_S_ALU_LSH_K: 269 if (K == 0) 270 break; 271 else 272 PPC_SLWI(r_A, r_A, K); 273 break; 274 case BPF_S_ALU_RSH_X: /* A >>= X; */ 275 ctx->seen |= SEEN_XREG; 276 PPC_SRW(r_A, r_A, r_X); 277 break; 278 case BPF_S_ALU_RSH_K: /* A >>= K; */ 279 if (K == 0) 280 break; 281 else 282 PPC_SRWI(r_A, r_A, K); 283 break; 284 case BPF_S_ALU_NEG: 285 PPC_NEG(r_A, r_A); 286 break; 287 case BPF_S_RET_K: 288 PPC_LI32(r_ret, K); 289 if (!K) { 290 if (ctx->pc_ret0 == -1) 291 ctx->pc_ret0 = i; 292 } 293 /* 294 * If this isn't the very last instruction, branch to 295 * the epilogue if we've stuff to clean up. Otherwise, 296 * if there's nothing to tidy, just return. If we /are/ 297 * the last instruction, we're about to fall through to 298 * the epilogue to return. 299 */ 300 if (i != flen - 1) { 301 /* 302 * Note: 'seen' is properly valid only on pass 303 * #2. Both parts of this conditional are the 304 * same instruction size though, meaning the 305 * first pass will still correctly determine the 306 * code size/addresses. 307 */ 308 if (ctx->seen) 309 PPC_JMP(exit_addr); 310 else 311 PPC_BLR(); 312 } 313 break; 314 case BPF_S_RET_A: 315 PPC_MR(r_ret, r_A); 316 if (i != flen - 1) { 317 if (ctx->seen) 318 PPC_JMP(exit_addr); 319 else 320 PPC_BLR(); 321 } 322 break; 323 case BPF_S_MISC_TAX: /* X = A */ 324 PPC_MR(r_X, r_A); 325 break; 326 case BPF_S_MISC_TXA: /* A = X */ 327 ctx->seen |= SEEN_XREG; 328 PPC_MR(r_A, r_X); 329 break; 330 331 /*** Constant loads/M[] access ***/ 332 case BPF_S_LD_IMM: /* A = K */ 333 PPC_LI32(r_A, K); 334 break; 335 case BPF_S_LDX_IMM: /* X = K */ 336 PPC_LI32(r_X, K); 337 break; 338 case BPF_S_LD_MEM: /* A = mem[K] */ 339 PPC_MR(r_A, r_M + (K & 0xf)); 340 ctx->seen |= SEEN_MEM | (1<<(K & 0xf)); 341 break; 342 case BPF_S_LDX_MEM: /* X = mem[K] */ 343 PPC_MR(r_X, r_M + (K & 0xf)); 344 ctx->seen |= SEEN_MEM | (1<<(K & 0xf)); 345 break; 346 case BPF_S_ST: /* mem[K] = A */ 347 PPC_MR(r_M + (K & 0xf), r_A); 348 ctx->seen |= SEEN_MEM | (1<<(K & 0xf)); 349 break; 350 case BPF_S_STX: /* mem[K] = X */ 351 PPC_MR(r_M + (K & 0xf), r_X); 352 ctx->seen |= SEEN_XREG | SEEN_MEM | (1<<(K & 0xf)); 353 break; 354 case BPF_S_LD_W_LEN: /* A = skb->len; */ 355 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4); 356 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, len)); 357 break; 358 case BPF_S_LDX_W_LEN: /* X = skb->len; */ 359 PPC_LWZ_OFFS(r_X, r_skb, offsetof(struct sk_buff, len)); 360 break; 361 362 /*** Ancillary info loads ***/ 363 case BPF_S_ANC_PROTOCOL: /* A = ntohs(skb->protocol); */ 364 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, 365 protocol) != 2); 366 PPC_NTOHS_OFFS(r_A, r_skb, offsetof(struct sk_buff, 367 protocol)); 368 break; 369 case BPF_S_ANC_IFINDEX: 370 PPC_LD_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff, 371 dev)); 372 PPC_CMPDI(r_scratch1, 0); 373 if (ctx->pc_ret0 != -1) { 374 PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]); 375 } else { 376 /* Exit, returning 0; first pass hits here. */ 377 PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12); 378 PPC_LI(r_ret, 0); 379 PPC_JMP(exit_addr); 380 } 381 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, 382 ifindex) != 4); 383 PPC_LWZ_OFFS(r_A, r_scratch1, 384 offsetof(struct net_device, ifindex)); 385 break; 386 case BPF_S_ANC_MARK: 387 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); 388 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, 389 mark)); 390 break; 391 case BPF_S_ANC_RXHASH: 392 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, rxhash) != 4); 393 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, 394 rxhash)); 395 break; 396 case BPF_S_ANC_VLAN_TAG: 397 case BPF_S_ANC_VLAN_TAG_PRESENT: 398 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); 399 PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, 400 vlan_tci)); 401 if (filter[i].code == BPF_S_ANC_VLAN_TAG) 402 PPC_ANDI(r_A, r_A, VLAN_VID_MASK); 403 else 404 PPC_ANDI(r_A, r_A, VLAN_TAG_PRESENT); 405 break; 406 case BPF_S_ANC_QUEUE: 407 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, 408 queue_mapping) != 2); 409 PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, 410 queue_mapping)); 411 break; 412 case BPF_S_ANC_CPU: 413 #ifdef CONFIG_SMP 414 /* 415 * PACA ptr is r13: 416 * raw_smp_processor_id() = local_paca->paca_index 417 */ 418 BUILD_BUG_ON(FIELD_SIZEOF(struct paca_struct, 419 paca_index) != 2); 420 PPC_LHZ_OFFS(r_A, 13, 421 offsetof(struct paca_struct, paca_index)); 422 #else 423 PPC_LI(r_A, 0); 424 #endif 425 break; 426 427 /*** Absolute loads from packet header/data ***/ 428 case BPF_S_LD_W_ABS: 429 func = CHOOSE_LOAD_FUNC(K, sk_load_word); 430 goto common_load; 431 case BPF_S_LD_H_ABS: 432 func = CHOOSE_LOAD_FUNC(K, sk_load_half); 433 goto common_load; 434 case BPF_S_LD_B_ABS: 435 func = CHOOSE_LOAD_FUNC(K, sk_load_byte); 436 common_load: 437 /* Load from [K]. */ 438 ctx->seen |= SEEN_DATAREF; 439 PPC_LI64(r_scratch1, func); 440 PPC_MTLR(r_scratch1); 441 PPC_LI32(r_addr, K); 442 PPC_BLRL(); 443 /* 444 * Helper returns 'lt' condition on error, and an 445 * appropriate return value in r3 446 */ 447 PPC_BCC(COND_LT, exit_addr); 448 break; 449 450 /*** Indirect loads from packet header/data ***/ 451 case BPF_S_LD_W_IND: 452 func = sk_load_word; 453 goto common_load_ind; 454 case BPF_S_LD_H_IND: 455 func = sk_load_half; 456 goto common_load_ind; 457 case BPF_S_LD_B_IND: 458 func = sk_load_byte; 459 common_load_ind: 460 /* 461 * Load from [X + K]. Negative offsets are tested for 462 * in the helper functions. 463 */ 464 ctx->seen |= SEEN_DATAREF | SEEN_XREG; 465 PPC_LI64(r_scratch1, func); 466 PPC_MTLR(r_scratch1); 467 PPC_ADDI(r_addr, r_X, IMM_L(K)); 468 if (K >= 32768) 469 PPC_ADDIS(r_addr, r_addr, IMM_HA(K)); 470 PPC_BLRL(); 471 /* If error, cr0.LT set */ 472 PPC_BCC(COND_LT, exit_addr); 473 break; 474 475 case BPF_S_LDX_B_MSH: 476 func = CHOOSE_LOAD_FUNC(K, sk_load_byte_msh); 477 goto common_load; 478 break; 479 480 /*** Jump and branches ***/ 481 case BPF_S_JMP_JA: 482 if (K != 0) 483 PPC_JMP(addrs[i + 1 + K]); 484 break; 485 486 case BPF_S_JMP_JGT_K: 487 case BPF_S_JMP_JGT_X: 488 true_cond = COND_GT; 489 goto cond_branch; 490 case BPF_S_JMP_JGE_K: 491 case BPF_S_JMP_JGE_X: 492 true_cond = COND_GE; 493 goto cond_branch; 494 case BPF_S_JMP_JEQ_K: 495 case BPF_S_JMP_JEQ_X: 496 true_cond = COND_EQ; 497 goto cond_branch; 498 case BPF_S_JMP_JSET_K: 499 case BPF_S_JMP_JSET_X: 500 true_cond = COND_NE; 501 /* Fall through */ 502 cond_branch: 503 /* same targets, can avoid doing the test :) */ 504 if (filter[i].jt == filter[i].jf) { 505 if (filter[i].jt > 0) 506 PPC_JMP(addrs[i + 1 + filter[i].jt]); 507 break; 508 } 509 510 switch (filter[i].code) { 511 case BPF_S_JMP_JGT_X: 512 case BPF_S_JMP_JGE_X: 513 case BPF_S_JMP_JEQ_X: 514 ctx->seen |= SEEN_XREG; 515 PPC_CMPLW(r_A, r_X); 516 break; 517 case BPF_S_JMP_JSET_X: 518 ctx->seen |= SEEN_XREG; 519 PPC_AND_DOT(r_scratch1, r_A, r_X); 520 break; 521 case BPF_S_JMP_JEQ_K: 522 case BPF_S_JMP_JGT_K: 523 case BPF_S_JMP_JGE_K: 524 if (K < 32768) 525 PPC_CMPLWI(r_A, K); 526 else { 527 PPC_LI32(r_scratch1, K); 528 PPC_CMPLW(r_A, r_scratch1); 529 } 530 break; 531 case BPF_S_JMP_JSET_K: 532 if (K < 32768) 533 /* PPC_ANDI is /only/ dot-form */ 534 PPC_ANDI(r_scratch1, r_A, K); 535 else { 536 PPC_LI32(r_scratch1, K); 537 PPC_AND_DOT(r_scratch1, r_A, 538 r_scratch1); 539 } 540 break; 541 } 542 /* Sometimes branches are constructed "backward", with 543 * the false path being the branch and true path being 544 * a fallthrough to the next instruction. 545 */ 546 if (filter[i].jt == 0) 547 /* Swap the sense of the branch */ 548 PPC_BCC(true_cond ^ COND_CMP_TRUE, 549 addrs[i + 1 + filter[i].jf]); 550 else { 551 PPC_BCC(true_cond, addrs[i + 1 + filter[i].jt]); 552 if (filter[i].jf != 0) 553 PPC_JMP(addrs[i + 1 + filter[i].jf]); 554 } 555 break; 556 default: 557 /* The filter contains something cruel & unusual. 558 * We don't handle it, but also there shouldn't be 559 * anything missing from our list. 560 */ 561 if (printk_ratelimit()) 562 pr_err("BPF filter opcode %04x (@%d) unsupported\n", 563 filter[i].code, i); 564 return -ENOTSUPP; 565 } 566 567 } 568 /* Set end-of-body-code address for exit. */ 569 addrs[i] = ctx->idx * 4; 570 571 return 0; 572 } 573 574 void bpf_jit_compile(struct sk_filter *fp) 575 { 576 unsigned int proglen; 577 unsigned int alloclen; 578 u32 *image = NULL; 579 u32 *code_base; 580 unsigned int *addrs; 581 struct codegen_context cgctx; 582 int pass; 583 int flen = fp->len; 584 585 if (!bpf_jit_enable) 586 return; 587 588 addrs = kzalloc((flen+1) * sizeof(*addrs), GFP_KERNEL); 589 if (addrs == NULL) 590 return; 591 592 /* 593 * There are multiple assembly passes as the generated code will change 594 * size as it settles down, figuring out the max branch offsets/exit 595 * paths required. 596 * 597 * The range of standard conditional branches is +/- 32Kbytes. Since 598 * BPF_MAXINSNS = 4096, we can only jump from (worst case) start to 599 * finish with 8 bytes/instruction. Not feasible, so long jumps are 600 * used, distinct from short branches. 601 * 602 * Current: 603 * 604 * For now, both branch types assemble to 2 words (short branches padded 605 * with a NOP); this is less efficient, but assembly will always complete 606 * after exactly 3 passes: 607 * 608 * First pass: No code buffer; Program is "faux-generated" -- no code 609 * emitted but maximum size of output determined (and addrs[] filled 610 * in). Also, we note whether we use M[], whether we use skb data, etc. 611 * All generation choices assumed to be 'worst-case', e.g. branches all 612 * far (2 instructions), return path code reduction not available, etc. 613 * 614 * Second pass: Code buffer allocated with size determined previously. 615 * Prologue generated to support features we have seen used. Exit paths 616 * determined and addrs[] is filled in again, as code may be slightly 617 * smaller as a result. 618 * 619 * Third pass: Code generated 'for real', and branch destinations 620 * determined from now-accurate addrs[] map. 621 * 622 * Ideal: 623 * 624 * If we optimise this, near branches will be shorter. On the 625 * first assembly pass, we should err on the side of caution and 626 * generate the biggest code. On subsequent passes, branches will be 627 * generated short or long and code size will reduce. With smaller 628 * code, more branches may fall into the short category, and code will 629 * reduce more. 630 * 631 * Finally, if we see one pass generate code the same size as the 632 * previous pass we have converged and should now generate code for 633 * real. Allocating at the end will also save the memory that would 634 * otherwise be wasted by the (small) current code shrinkage. 635 * Preferably, we should do a small number of passes (e.g. 5) and if we 636 * haven't converged by then, get impatient and force code to generate 637 * as-is, even if the odd branch would be left long. The chances of a 638 * long jump are tiny with all but the most enormous of BPF filter 639 * inputs, so we should usually converge on the third pass. 640 */ 641 642 cgctx.idx = 0; 643 cgctx.seen = 0; 644 cgctx.pc_ret0 = -1; 645 /* Scouting faux-generate pass 0 */ 646 if (bpf_jit_build_body(fp, 0, &cgctx, addrs)) 647 /* We hit something illegal or unsupported. */ 648 goto out; 649 650 /* 651 * Pretend to build prologue, given the features we've seen. This will 652 * update ctgtx.idx as it pretends to output instructions, then we can 653 * calculate total size from idx. 654 */ 655 bpf_jit_build_prologue(fp, 0, &cgctx); 656 bpf_jit_build_epilogue(0, &cgctx); 657 658 proglen = cgctx.idx * 4; 659 alloclen = proglen + FUNCTION_DESCR_SIZE; 660 image = module_alloc(alloclen); 661 if (!image) 662 goto out; 663 664 code_base = image + (FUNCTION_DESCR_SIZE/4); 665 666 /* Code generation passes 1-2 */ 667 for (pass = 1; pass < 3; pass++) { 668 /* Now build the prologue, body code & epilogue for real. */ 669 cgctx.idx = 0; 670 bpf_jit_build_prologue(fp, code_base, &cgctx); 671 bpf_jit_build_body(fp, code_base, &cgctx, addrs); 672 bpf_jit_build_epilogue(code_base, &cgctx); 673 674 if (bpf_jit_enable > 1) 675 pr_info("Pass %d: shrink = %d, seen = 0x%x\n", pass, 676 proglen - (cgctx.idx * 4), cgctx.seen); 677 } 678 679 if (bpf_jit_enable > 1) 680 /* Note that we output the base address of the code_base 681 * rather than image, since opcodes are in code_base. 682 */ 683 bpf_jit_dump(flen, proglen, pass, code_base); 684 685 if (image) { 686 bpf_flush_icache(code_base, code_base + (proglen/4)); 687 /* Function descriptor nastiness: Address + TOC */ 688 ((u64 *)image)[0] = (u64)code_base; 689 ((u64 *)image)[1] = local_paca->kernel_toc; 690 fp->bpf_func = (void *)image; 691 } 692 out: 693 kfree(addrs); 694 return; 695 } 696 697 void bpf_jit_free(struct sk_filter *fp) 698 { 699 if (fp->bpf_func != sk_run_filter) 700 module_free(NULL, fp->bpf_func); 701 kfree(fp); 702 } 703