blob: c427ae36374ab4fa6eb1503c15b07de2926006e6 [file] [log] [blame]
Matt Evans0ca87f02011-07-20 15:51:00 +00001/* 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>
Daniel Borkmann5082dfb2012-11-08 11:41:39 +000016#include <linux/if_vlan.h>
17
Matt Evans0ca87f02011-07-20 15:51:00 +000018#include "bpf_jit.h"
19
20#ifndef __BIG_ENDIAN
21/* There are endianness assumptions herein. */
22#error "Little-endian PPC not supported in BPF compiler"
23#endif
24
25int bpf_jit_enable __read_mostly;
26
27
28static inline void bpf_flush_icache(void *start, void *end)
29{
30 smp_wmb();
31 flush_icache_range((unsigned long)start, (unsigned long)end);
32}
33
34static void bpf_jit_build_prologue(struct sk_filter *fp, u32 *image,
35 struct codegen_context *ctx)
36{
37 int i;
38 const struct sock_filter *filter = fp->insns;
39
40 if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) {
41 /* Make stackframe */
42 if (ctx->seen & SEEN_DATAREF) {
43 /* If we call any helpers (for loads), save LR */
Michael Neulingc75df6f2012-06-25 13:33:10 +000044 EMIT(PPC_INST_MFLR | __PPC_RT(R0));
Matt Evans0ca87f02011-07-20 15:51:00 +000045 PPC_STD(0, 1, 16);
46
47 /* Back up non-volatile regs. */
48 PPC_STD(r_D, 1, -(8*(32-r_D)));
49 PPC_STD(r_HL, 1, -(8*(32-r_HL)));
50 }
51 if (ctx->seen & SEEN_MEM) {
52 /*
53 * Conditionally save regs r15-r31 as some will be used
54 * for M[] data.
55 */
56 for (i = r_M; i < (r_M+16); i++) {
57 if (ctx->seen & (1 << (i-r_M)))
58 PPC_STD(i, 1, -(8*(32-i)));
59 }
60 }
Michael Neulingc75df6f2012-06-25 13:33:10 +000061 EMIT(PPC_INST_STDU | __PPC_RS(R1) | __PPC_RA(R1) |
Matt Evans0ca87f02011-07-20 15:51:00 +000062 (-BPF_PPC_STACKFRAME & 0xfffc));
63 }
64
65 if (ctx->seen & SEEN_DATAREF) {
66 /*
67 * If this filter needs to access skb data,
68 * prepare r_D and r_HL:
69 * r_HL = skb->len - skb->data_len
70 * r_D = skb->data
71 */
72 PPC_LWZ_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff,
73 data_len));
74 PPC_LWZ_OFFS(r_HL, r_skb, offsetof(struct sk_buff, len));
75 PPC_SUB(r_HL, r_HL, r_scratch1);
76 PPC_LD_OFFS(r_D, r_skb, offsetof(struct sk_buff, data));
77 }
78
79 if (ctx->seen & SEEN_XREG) {
80 /*
81 * TODO: Could also detect whether first instr. sets X and
82 * avoid this (as below, with A).
83 */
84 PPC_LI(r_X, 0);
85 }
86
87 switch (filter[0].code) {
88 case BPF_S_RET_K:
89 case BPF_S_LD_W_LEN:
90 case BPF_S_ANC_PROTOCOL:
91 case BPF_S_ANC_IFINDEX:
92 case BPF_S_ANC_MARK:
93 case BPF_S_ANC_RXHASH:
Daniel Borkmann5082dfb2012-11-08 11:41:39 +000094 case BPF_S_ANC_VLAN_TAG:
95 case BPF_S_ANC_VLAN_TAG_PRESENT:
Matt Evans0ca87f02011-07-20 15:51:00 +000096 case BPF_S_ANC_CPU:
97 case BPF_S_ANC_QUEUE:
98 case BPF_S_LD_W_ABS:
99 case BPF_S_LD_H_ABS:
100 case BPF_S_LD_B_ABS:
101 /* first instruction sets A register (or is RET 'constant') */
102 break;
103 default:
104 /* make sure we dont leak kernel information to user */
105 PPC_LI(r_A, 0);
106 }
107}
108
109static void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx)
110{
111 int i;
112
113 if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) {
114 PPC_ADDI(1, 1, BPF_PPC_STACKFRAME);
115 if (ctx->seen & SEEN_DATAREF) {
116 PPC_LD(0, 1, 16);
117 PPC_MTLR(0);
118 PPC_LD(r_D, 1, -(8*(32-r_D)));
119 PPC_LD(r_HL, 1, -(8*(32-r_HL)));
120 }
121 if (ctx->seen & SEEN_MEM) {
122 /* Restore any saved non-vol registers */
123 for (i = r_M; i < (r_M+16); i++) {
124 if (ctx->seen & (1 << (i-r_M)))
125 PPC_LD(i, 1, -(8*(32-i)));
126 }
127 }
128 }
129 /* The RETs have left a return value in R3. */
130
131 PPC_BLR();
132}
133
Jan Seiffert05be1822012-04-29 19:02:19 +0000134#define CHOOSE_LOAD_FUNC(K, func) \
135 ((int)K < 0 ? ((int)K >= SKF_LL_OFF ? func##_negative_offset : func) : func##_positive_offset)
136
Matt Evans0ca87f02011-07-20 15:51:00 +0000137/* Assemble the body code between the prologue & epilogue. */
138static int bpf_jit_build_body(struct sk_filter *fp, u32 *image,
139 struct codegen_context *ctx,
140 unsigned int *addrs)
141{
142 const struct sock_filter *filter = fp->insns;
143 int flen = fp->len;
144 u8 *func;
145 unsigned int true_cond;
146 int i;
147
148 /* Start of epilogue code */
149 unsigned int exit_addr = addrs[flen];
150
151 for (i = 0; i < flen; i++) {
152 unsigned int K = filter[i].k;
153
154 /*
155 * addrs[] maps a BPF bytecode address into a real offset from
156 * the start of the body code.
157 */
158 addrs[i] = ctx->idx * 4;
159
160 switch (filter[i].code) {
161 /*** ALU ops ***/
162 case BPF_S_ALU_ADD_X: /* A += X; */
163 ctx->seen |= SEEN_XREG;
164 PPC_ADD(r_A, r_A, r_X);
165 break;
166 case BPF_S_ALU_ADD_K: /* A += K; */
167 if (!K)
168 break;
169 PPC_ADDI(r_A, r_A, IMM_L(K));
170 if (K >= 32768)
171 PPC_ADDIS(r_A, r_A, IMM_HA(K));
172 break;
173 case BPF_S_ALU_SUB_X: /* A -= X; */
174 ctx->seen |= SEEN_XREG;
175 PPC_SUB(r_A, r_A, r_X);
176 break;
177 case BPF_S_ALU_SUB_K: /* A -= K */
178 if (!K)
179 break;
180 PPC_ADDI(r_A, r_A, IMM_L(-K));
181 if (K >= 32768)
182 PPC_ADDIS(r_A, r_A, IMM_HA(-K));
183 break;
184 case BPF_S_ALU_MUL_X: /* A *= X; */
185 ctx->seen |= SEEN_XREG;
186 PPC_MUL(r_A, r_A, r_X);
187 break;
188 case BPF_S_ALU_MUL_K: /* A *= K */
189 if (K < 32768)
190 PPC_MULI(r_A, r_A, K);
191 else {
192 PPC_LI32(r_scratch1, K);
193 PPC_MUL(r_A, r_A, r_scratch1);
194 }
195 break;
196 case BPF_S_ALU_DIV_X: /* A /= X; */
197 ctx->seen |= SEEN_XREG;
198 PPC_CMPWI(r_X, 0);
199 if (ctx->pc_ret0 != -1) {
200 PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]);
201 } else {
202 /*
203 * Exit, returning 0; first pass hits here
204 * (longer worst-case code size).
205 */
206 PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12);
207 PPC_LI(r_ret, 0);
208 PPC_JMP(exit_addr);
209 }
210 PPC_DIVWU(r_A, r_A, r_X);
211 break;
212 case BPF_S_ALU_DIV_K: /* A = reciprocal_divide(A, K); */
213 PPC_LI32(r_scratch1, K);
214 /* Top 32 bits of 64bit result -> A */
215 PPC_MULHWU(r_A, r_A, r_scratch1);
216 break;
217 case BPF_S_ALU_AND_X:
218 ctx->seen |= SEEN_XREG;
219 PPC_AND(r_A, r_A, r_X);
220 break;
221 case BPF_S_ALU_AND_K:
222 if (!IMM_H(K))
223 PPC_ANDI(r_A, r_A, K);
224 else {
225 PPC_LI32(r_scratch1, K);
226 PPC_AND(r_A, r_A, r_scratch1);
227 }
228 break;
229 case BPF_S_ALU_OR_X:
230 ctx->seen |= SEEN_XREG;
231 PPC_OR(r_A, r_A, r_X);
232 break;
233 case BPF_S_ALU_OR_K:
234 if (IMM_L(K))
235 PPC_ORI(r_A, r_A, IMM_L(K));
236 if (K >= 65536)
237 PPC_ORIS(r_A, r_A, IMM_H(K));
238 break;
Daniel Borkmann02871902012-11-08 11:39:41 +0000239 case BPF_S_ANC_ALU_XOR_X:
240 case BPF_S_ALU_XOR_X: /* A ^= X */
241 ctx->seen |= SEEN_XREG;
242 PPC_XOR(r_A, r_A, r_X);
243 break;
244 case BPF_S_ALU_XOR_K: /* A ^= K */
245 if (IMM_L(K))
246 PPC_XORI(r_A, r_A, IMM_L(K));
247 if (K >= 65536)
248 PPC_XORIS(r_A, r_A, IMM_H(K));
249 break;
Matt Evans0ca87f02011-07-20 15:51:00 +0000250 case BPF_S_ALU_LSH_X: /* A <<= X; */
251 ctx->seen |= SEEN_XREG;
252 PPC_SLW(r_A, r_A, r_X);
253 break;
254 case BPF_S_ALU_LSH_K:
255 if (K == 0)
256 break;
257 else
258 PPC_SLWI(r_A, r_A, K);
259 break;
260 case BPF_S_ALU_RSH_X: /* A >>= X; */
261 ctx->seen |= SEEN_XREG;
262 PPC_SRW(r_A, r_A, r_X);
263 break;
264 case BPF_S_ALU_RSH_K: /* A >>= K; */
265 if (K == 0)
266 break;
267 else
268 PPC_SRWI(r_A, r_A, K);
269 break;
270 case BPF_S_ALU_NEG:
271 PPC_NEG(r_A, r_A);
272 break;
273 case BPF_S_RET_K:
274 PPC_LI32(r_ret, K);
275 if (!K) {
276 if (ctx->pc_ret0 == -1)
277 ctx->pc_ret0 = i;
278 }
279 /*
280 * If this isn't the very last instruction, branch to
281 * the epilogue if we've stuff to clean up. Otherwise,
282 * if there's nothing to tidy, just return. If we /are/
283 * the last instruction, we're about to fall through to
284 * the epilogue to return.
285 */
286 if (i != flen - 1) {
287 /*
288 * Note: 'seen' is properly valid only on pass
289 * #2. Both parts of this conditional are the
290 * same instruction size though, meaning the
291 * first pass will still correctly determine the
292 * code size/addresses.
293 */
294 if (ctx->seen)
295 PPC_JMP(exit_addr);
296 else
297 PPC_BLR();
298 }
299 break;
300 case BPF_S_RET_A:
301 PPC_MR(r_ret, r_A);
302 if (i != flen - 1) {
303 if (ctx->seen)
304 PPC_JMP(exit_addr);
305 else
306 PPC_BLR();
307 }
308 break;
309 case BPF_S_MISC_TAX: /* X = A */
310 PPC_MR(r_X, r_A);
311 break;
312 case BPF_S_MISC_TXA: /* A = X */
313 ctx->seen |= SEEN_XREG;
314 PPC_MR(r_A, r_X);
315 break;
316
317 /*** Constant loads/M[] access ***/
318 case BPF_S_LD_IMM: /* A = K */
319 PPC_LI32(r_A, K);
320 break;
321 case BPF_S_LDX_IMM: /* X = K */
322 PPC_LI32(r_X, K);
323 break;
324 case BPF_S_LD_MEM: /* A = mem[K] */
325 PPC_MR(r_A, r_M + (K & 0xf));
326 ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
327 break;
328 case BPF_S_LDX_MEM: /* X = mem[K] */
329 PPC_MR(r_X, r_M + (K & 0xf));
330 ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
331 break;
332 case BPF_S_ST: /* mem[K] = A */
333 PPC_MR(r_M + (K & 0xf), r_A);
334 ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
335 break;
336 case BPF_S_STX: /* mem[K] = X */
337 PPC_MR(r_M + (K & 0xf), r_X);
338 ctx->seen |= SEEN_XREG | SEEN_MEM | (1<<(K & 0xf));
339 break;
340 case BPF_S_LD_W_LEN: /* A = skb->len; */
341 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
342 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, len));
343 break;
344 case BPF_S_LDX_W_LEN: /* X = skb->len; */
345 PPC_LWZ_OFFS(r_X, r_skb, offsetof(struct sk_buff, len));
346 break;
347
348 /*** Ancillary info loads ***/
349
350 /* None of the BPF_S_ANC* codes appear to be passed by
351 * sk_chk_filter(). The interpreter and the x86 BPF
352 * compiler implement them so we do too -- they may be
353 * planted in future.
354 */
355 case BPF_S_ANC_PROTOCOL: /* A = ntohs(skb->protocol); */
356 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff,
357 protocol) != 2);
358 PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
359 protocol));
360 /* ntohs is a NOP with BE loads. */
361 break;
362 case BPF_S_ANC_IFINDEX:
363 PPC_LD_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff,
364 dev));
365 PPC_CMPDI(r_scratch1, 0);
366 if (ctx->pc_ret0 != -1) {
367 PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]);
368 } else {
369 /* Exit, returning 0; first pass hits here. */
370 PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12);
371 PPC_LI(r_ret, 0);
372 PPC_JMP(exit_addr);
373 }
374 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device,
375 ifindex) != 4);
376 PPC_LWZ_OFFS(r_A, r_scratch1,
377 offsetof(struct net_device, ifindex));
378 break;
379 case BPF_S_ANC_MARK:
380 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
381 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
382 mark));
383 break;
384 case BPF_S_ANC_RXHASH:
385 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, rxhash) != 4);
386 PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
387 rxhash));
388 break;
Daniel Borkmann5082dfb2012-11-08 11:41:39 +0000389 case BPF_S_ANC_VLAN_TAG:
390 case BPF_S_ANC_VLAN_TAG_PRESENT:
391 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
392 PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
393 vlan_tci));
394 if (filter[i].code == BPF_S_ANC_VLAN_TAG)
395 PPC_ANDI(r_A, r_A, VLAN_VID_MASK);
396 else
397 PPC_ANDI(r_A, r_A, VLAN_TAG_PRESENT);
398 break;
Matt Evans0ca87f02011-07-20 15:51:00 +0000399 case BPF_S_ANC_QUEUE:
400 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff,
401 queue_mapping) != 2);
402 PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
403 queue_mapping));
404 break;
405 case BPF_S_ANC_CPU:
406#ifdef CONFIG_SMP
407 /*
408 * PACA ptr is r13:
409 * raw_smp_processor_id() = local_paca->paca_index
410 */
411 BUILD_BUG_ON(FIELD_SIZEOF(struct paca_struct,
412 paca_index) != 2);
413 PPC_LHZ_OFFS(r_A, 13,
414 offsetof(struct paca_struct, paca_index));
415#else
416 PPC_LI(r_A, 0);
417#endif
418 break;
419
420 /*** Absolute loads from packet header/data ***/
421 case BPF_S_LD_W_ABS:
Jan Seiffert05be1822012-04-29 19:02:19 +0000422 func = CHOOSE_LOAD_FUNC(K, sk_load_word);
Matt Evans0ca87f02011-07-20 15:51:00 +0000423 goto common_load;
424 case BPF_S_LD_H_ABS:
Jan Seiffert05be1822012-04-29 19:02:19 +0000425 func = CHOOSE_LOAD_FUNC(K, sk_load_half);
Matt Evans0ca87f02011-07-20 15:51:00 +0000426 goto common_load;
427 case BPF_S_LD_B_ABS:
Jan Seiffert05be1822012-04-29 19:02:19 +0000428 func = CHOOSE_LOAD_FUNC(K, sk_load_byte);
Matt Evans0ca87f02011-07-20 15:51:00 +0000429 common_load:
Jan Seiffert05be1822012-04-29 19:02:19 +0000430 /* Load from [K]. */
Matt Evans0ca87f02011-07-20 15:51:00 +0000431 ctx->seen |= SEEN_DATAREF;
Matt Evans0ca87f02011-07-20 15:51:00 +0000432 PPC_LI64(r_scratch1, func);
433 PPC_MTLR(r_scratch1);
434 PPC_LI32(r_addr, K);
435 PPC_BLRL();
436 /*
437 * Helper returns 'lt' condition on error, and an
438 * appropriate return value in r3
439 */
440 PPC_BCC(COND_LT, exit_addr);
441 break;
442
443 /*** Indirect loads from packet header/data ***/
444 case BPF_S_LD_W_IND:
445 func = sk_load_word;
446 goto common_load_ind;
447 case BPF_S_LD_H_IND:
448 func = sk_load_half;
449 goto common_load_ind;
450 case BPF_S_LD_B_IND:
451 func = sk_load_byte;
452 common_load_ind:
453 /*
454 * Load from [X + K]. Negative offsets are tested for
Jan Seiffert05be1822012-04-29 19:02:19 +0000455 * in the helper functions.
Matt Evans0ca87f02011-07-20 15:51:00 +0000456 */
457 ctx->seen |= SEEN_DATAREF | SEEN_XREG;
458 PPC_LI64(r_scratch1, func);
459 PPC_MTLR(r_scratch1);
460 PPC_ADDI(r_addr, r_X, IMM_L(K));
461 if (K >= 32768)
462 PPC_ADDIS(r_addr, r_addr, IMM_HA(K));
463 PPC_BLRL();
464 /* If error, cr0.LT set */
465 PPC_BCC(COND_LT, exit_addr);
466 break;
467
468 case BPF_S_LDX_B_MSH:
Jan Seiffert05be1822012-04-29 19:02:19 +0000469 func = CHOOSE_LOAD_FUNC(K, sk_load_byte_msh);
Matt Evans0ca87f02011-07-20 15:51:00 +0000470 goto common_load;
471 break;
472
473 /*** Jump and branches ***/
474 case BPF_S_JMP_JA:
475 if (K != 0)
476 PPC_JMP(addrs[i + 1 + K]);
477 break;
478
479 case BPF_S_JMP_JGT_K:
480 case BPF_S_JMP_JGT_X:
481 true_cond = COND_GT;
482 goto cond_branch;
483 case BPF_S_JMP_JGE_K:
484 case BPF_S_JMP_JGE_X:
485 true_cond = COND_GE;
486 goto cond_branch;
487 case BPF_S_JMP_JEQ_K:
488 case BPF_S_JMP_JEQ_X:
489 true_cond = COND_EQ;
490 goto cond_branch;
491 case BPF_S_JMP_JSET_K:
492 case BPF_S_JMP_JSET_X:
493 true_cond = COND_NE;
494 /* Fall through */
495 cond_branch:
496 /* same targets, can avoid doing the test :) */
497 if (filter[i].jt == filter[i].jf) {
498 if (filter[i].jt > 0)
499 PPC_JMP(addrs[i + 1 + filter[i].jt]);
500 break;
501 }
502
503 switch (filter[i].code) {
504 case BPF_S_JMP_JGT_X:
505 case BPF_S_JMP_JGE_X:
506 case BPF_S_JMP_JEQ_X:
507 ctx->seen |= SEEN_XREG;
508 PPC_CMPLW(r_A, r_X);
509 break;
510 case BPF_S_JMP_JSET_X:
511 ctx->seen |= SEEN_XREG;
512 PPC_AND_DOT(r_scratch1, r_A, r_X);
513 break;
514 case BPF_S_JMP_JEQ_K:
515 case BPF_S_JMP_JGT_K:
516 case BPF_S_JMP_JGE_K:
517 if (K < 32768)
518 PPC_CMPLWI(r_A, K);
519 else {
520 PPC_LI32(r_scratch1, K);
521 PPC_CMPLW(r_A, r_scratch1);
522 }
523 break;
524 case BPF_S_JMP_JSET_K:
525 if (K < 32768)
526 /* PPC_ANDI is /only/ dot-form */
527 PPC_ANDI(r_scratch1, r_A, K);
528 else {
529 PPC_LI32(r_scratch1, K);
530 PPC_AND_DOT(r_scratch1, r_A,
531 r_scratch1);
532 }
533 break;
534 }
535 /* Sometimes branches are constructed "backward", with
536 * the false path being the branch and true path being
537 * a fallthrough to the next instruction.
538 */
539 if (filter[i].jt == 0)
540 /* Swap the sense of the branch */
541 PPC_BCC(true_cond ^ COND_CMP_TRUE,
542 addrs[i + 1 + filter[i].jf]);
543 else {
544 PPC_BCC(true_cond, addrs[i + 1 + filter[i].jt]);
545 if (filter[i].jf != 0)
546 PPC_JMP(addrs[i + 1 + filter[i].jf]);
547 }
548 break;
549 default:
550 /* The filter contains something cruel & unusual.
551 * We don't handle it, but also there shouldn't be
552 * anything missing from our list.
553 */
554 if (printk_ratelimit())
555 pr_err("BPF filter opcode %04x (@%d) unsupported\n",
556 filter[i].code, i);
557 return -ENOTSUPP;
558 }
559
560 }
561 /* Set end-of-body-code address for exit. */
562 addrs[i] = ctx->idx * 4;
563
564 return 0;
565}
566
567void bpf_jit_compile(struct sk_filter *fp)
568{
569 unsigned int proglen;
570 unsigned int alloclen;
571 u32 *image = NULL;
572 u32 *code_base;
573 unsigned int *addrs;
574 struct codegen_context cgctx;
575 int pass;
576 int flen = fp->len;
577
578 if (!bpf_jit_enable)
579 return;
580
581 addrs = kzalloc((flen+1) * sizeof(*addrs), GFP_KERNEL);
582 if (addrs == NULL)
583 return;
584
585 /*
586 * There are multiple assembly passes as the generated code will change
587 * size as it settles down, figuring out the max branch offsets/exit
588 * paths required.
589 *
590 * The range of standard conditional branches is +/- 32Kbytes. Since
591 * BPF_MAXINSNS = 4096, we can only jump from (worst case) start to
592 * finish with 8 bytes/instruction. Not feasible, so long jumps are
593 * used, distinct from short branches.
594 *
595 * Current:
596 *
597 * For now, both branch types assemble to 2 words (short branches padded
598 * with a NOP); this is less efficient, but assembly will always complete
599 * after exactly 3 passes:
600 *
601 * First pass: No code buffer; Program is "faux-generated" -- no code
602 * emitted but maximum size of output determined (and addrs[] filled
603 * in). Also, we note whether we use M[], whether we use skb data, etc.
604 * All generation choices assumed to be 'worst-case', e.g. branches all
605 * far (2 instructions), return path code reduction not available, etc.
606 *
607 * Second pass: Code buffer allocated with size determined previously.
608 * Prologue generated to support features we have seen used. Exit paths
609 * determined and addrs[] is filled in again, as code may be slightly
610 * smaller as a result.
611 *
612 * Third pass: Code generated 'for real', and branch destinations
613 * determined from now-accurate addrs[] map.
614 *
615 * Ideal:
616 *
617 * If we optimise this, near branches will be shorter. On the
618 * first assembly pass, we should err on the side of caution and
619 * generate the biggest code. On subsequent passes, branches will be
620 * generated short or long and code size will reduce. With smaller
621 * code, more branches may fall into the short category, and code will
622 * reduce more.
623 *
624 * Finally, if we see one pass generate code the same size as the
625 * previous pass we have converged and should now generate code for
626 * real. Allocating at the end will also save the memory that would
627 * otherwise be wasted by the (small) current code shrinkage.
628 * Preferably, we should do a small number of passes (e.g. 5) and if we
629 * haven't converged by then, get impatient and force code to generate
630 * as-is, even if the odd branch would be left long. The chances of a
631 * long jump are tiny with all but the most enormous of BPF filter
632 * inputs, so we should usually converge on the third pass.
633 */
634
635 cgctx.idx = 0;
636 cgctx.seen = 0;
637 cgctx.pc_ret0 = -1;
638 /* Scouting faux-generate pass 0 */
639 if (bpf_jit_build_body(fp, 0, &cgctx, addrs))
640 /* We hit something illegal or unsupported. */
641 goto out;
642
643 /*
644 * Pretend to build prologue, given the features we've seen. This will
645 * update ctgtx.idx as it pretends to output instructions, then we can
646 * calculate total size from idx.
647 */
648 bpf_jit_build_prologue(fp, 0, &cgctx);
649 bpf_jit_build_epilogue(0, &cgctx);
650
651 proglen = cgctx.idx * 4;
652 alloclen = proglen + FUNCTION_DESCR_SIZE;
653 image = module_alloc(max_t(unsigned int, alloclen,
654 sizeof(struct work_struct)));
655 if (!image)
656 goto out;
657
658 code_base = image + (FUNCTION_DESCR_SIZE/4);
659
660 /* Code generation passes 1-2 */
661 for (pass = 1; pass < 3; pass++) {
662 /* Now build the prologue, body code & epilogue for real. */
663 cgctx.idx = 0;
664 bpf_jit_build_prologue(fp, code_base, &cgctx);
665 bpf_jit_build_body(fp, code_base, &cgctx, addrs);
666 bpf_jit_build_epilogue(code_base, &cgctx);
667
668 if (bpf_jit_enable > 1)
669 pr_info("Pass %d: shrink = %d, seen = 0x%x\n", pass,
670 proglen - (cgctx.idx * 4), cgctx.seen);
671 }
672
673 if (bpf_jit_enable > 1)
Daniel Borkmann79617802013-03-21 22:22:03 +0100674 /* Note that we output the base address of the code_base
675 * rather than image, since opcodes are in code_base.
676 */
677 bpf_jit_dump(flen, proglen, pass, code_base);
Matt Evans0ca87f02011-07-20 15:51:00 +0000678
679 if (image) {
Matt Evans0ca87f02011-07-20 15:51:00 +0000680 bpf_flush_icache(code_base, code_base + (proglen/4));
681 /* Function descriptor nastiness: Address + TOC */
682 ((u64 *)image)[0] = (u64)code_base;
683 ((u64 *)image)[1] = local_paca->kernel_toc;
684 fp->bpf_func = (void *)image;
685 }
686out:
687 kfree(addrs);
688 return;
689}
690
691static void jit_free_defer(struct work_struct *arg)
692{
693 module_free(NULL, arg);
694}
695
696/* run from softirq, we must use a work_struct to call
697 * module_free() from process context
698 */
699void bpf_jit_free(struct sk_filter *fp)
700{
701 if (fp->bpf_func != sk_run_filter) {
702 struct work_struct *work = (struct work_struct *)fp->bpf_func;
703
704 INIT_WORK(work, jit_free_defer);
705 schedule_work(work);
706 }
707}