summaryrefslogtreecommitdiff
path: root/arch/powerpc/net/bpf_jit_comp.c
blob: 78d61f97371e1b70ecdb6bf19b38f95e53a22707 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
// SPDX-License-Identifier: GPL-2.0-only
/* bpf_jit_comp.c: BPF JIT compiler
 *
 * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation
 *
 * Based on the x86 BPF compiler, by Eric Dumazet (eric.dumazet@gmail.com)
 * Ported to ppc32 by Denis Kirjanov <kda@linux-powerpc.org>
 */
#include <linux/moduleloader.h>
#include <asm/cacheflush.h>
#include <asm/asm-compat.h>
#include <linux/netdevice.h>
#include <linux/filter.h>
#include <linux/if_vlan.h>

#include "bpf_jit32.h"

static inline void bpf_flush_icache(void *start, void *end)
{
	smp_wmb();
	flush_icache_range((unsigned long)start, (unsigned long)end);
}

static void bpf_jit_build_prologue(struct bpf_prog *fp, u32 *image,
				   struct codegen_context *ctx)
{
	int i;
	const struct sock_filter *filter = fp->insns;

	if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) {
		/* Make stackframe */
		if (ctx->seen & SEEN_DATAREF) {
			/* If we call any helpers (for loads), save LR */
			EMIT(PPC_INST_MFLR | __PPC_RT(R0));
			PPC_BPF_STL(0, 1, PPC_LR_STKOFF);

			/* Back up non-volatile regs. */
			PPC_BPF_STL(r_D, 1, -(REG_SZ*(32-r_D)));
			PPC_BPF_STL(r_HL, 1, -(REG_SZ*(32-r_HL)));
		}
		if (ctx->seen & SEEN_MEM) {
			/*
			 * Conditionally save regs r15-r31 as some will be used
			 * for M[] data.
			 */
			for (i = r_M; i < (r_M+16); i++) {
				if (ctx->seen & (1 << (i-r_M)))
					PPC_BPF_STL(i, 1, -(REG_SZ*(32-i)));
			}
		}
		PPC_BPF_STLU(1, 1, -BPF_PPC_STACKFRAME);
	}

	if (ctx->seen & SEEN_DATAREF) {
		/*
		 * If this filter needs to access skb data,
		 * prepare r_D and r_HL:
		 *  r_HL = skb->len - skb->data_len
		 *  r_D	 = skb->data
		 */
		PPC_LWZ_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff,
							 data_len));
		PPC_LWZ_OFFS(r_HL, r_skb, offsetof(struct sk_buff, len));
		EMIT(PPC_RAW_SUB(r_HL, r_HL, r_scratch1));
		PPC_LL_OFFS(r_D, r_skb, offsetof(struct sk_buff, data));
	}

	if (ctx->seen & SEEN_XREG) {
		/*
		 * TODO: Could also detect whether first instr. sets X and
		 * avoid this (as below, with A).
		 */
		EMIT(PPC_RAW_LI(r_X, 0));
	}

	/* make sure we dont leak kernel information to user */
	if (bpf_needs_clear_a(&filter[0]))
		EMIT(PPC_RAW_LI(r_A, 0));
}

static void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx)
{
	int i;

	if (ctx->seen & (SEEN_MEM | SEEN_DATAREF)) {
		EMIT(PPC_RAW_ADDI(1, 1, BPF_PPC_STACKFRAME));
		if (ctx->seen & SEEN_DATAREF) {
			PPC_BPF_LL(0, 1, PPC_LR_STKOFF);
			EMIT(PPC_RAW_MTLR(0));
			PPC_BPF_LL(r_D, 1, -(REG_SZ*(32-r_D)));
			PPC_BPF_LL(r_HL, 1, -(REG_SZ*(32-r_HL)));
		}
		if (ctx->seen & SEEN_MEM) {
			/* Restore any saved non-vol registers */
			for (i = r_M; i < (r_M+16); i++) {
				if (ctx->seen & (1 << (i-r_M)))
					PPC_BPF_LL(i, 1, -(REG_SZ*(32-i)));
			}
		}
	}
	/* The RETs have left a return value in R3. */

	EMIT(PPC_RAW_BLR());
}

#define CHOOSE_LOAD_FUNC(K, func) \
	((int)K < 0 ? ((int)K >= SKF_LL_OFF ? func##_negative_offset : func) : func##_positive_offset)

/* Assemble the body code between the prologue & epilogue. */
static int bpf_jit_build_body(struct bpf_prog *fp, u32 *image,
			      struct codegen_context *ctx,
			      unsigned int *addrs)
{
	const struct sock_filter *filter = fp->insns;
	int flen = fp->len;
	u8 *func;
	unsigned int true_cond;
	int i;

	/* Start of epilogue code */
	unsigned int exit_addr = addrs[flen];

	for (i = 0; i < flen; i++) {
		unsigned int K = filter[i].k;
		u16 code = bpf_anc_helper(&filter[i]);

		/*
		 * addrs[] maps a BPF bytecode address into a real offset from
		 * the start of the body code.
		 */
		addrs[i] = ctx->idx * 4;

		switch (code) {
			/*** ALU ops ***/
		case BPF_ALU | BPF_ADD | BPF_X: /* A += X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_ADD(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_ADD | BPF_K: /* A += K; */
			if (!K)
				break;
			EMIT(PPC_RAW_ADDI(r_A, r_A, IMM_L(K)));
			if (K >= 32768)
				EMIT(PPC_RAW_ADDIS(r_A, r_A, IMM_HA(K)));
			break;
		case BPF_ALU | BPF_SUB | BPF_X: /* A -= X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_SUB(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_SUB | BPF_K: /* A -= K */
			if (!K)
				break;
			EMIT(PPC_RAW_ADDI(r_A, r_A, IMM_L(-K)));
			if (K >= 32768)
				EMIT(PPC_RAW_ADDIS(r_A, r_A, IMM_HA(-K)));
			break;
		case BPF_ALU | BPF_MUL | BPF_X: /* A *= X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_MULW(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_MUL | BPF_K: /* A *= K */
			if (K < 32768)
				EMIT(PPC_RAW_MULI(r_A, r_A, K));
			else {
				PPC_LI32(r_scratch1, K);
				EMIT(PPC_RAW_MULW(r_A, r_A, r_scratch1));
			}
			break;
		case BPF_ALU | BPF_MOD | BPF_X: /* A %= X; */
		case BPF_ALU | BPF_DIV | BPF_X: /* A /= X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_CMPWI(r_X, 0));
			if (ctx->pc_ret0 != -1) {
				PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]);
			} else {
				PPC_BCC_SHORT(COND_NE, (ctx->idx*4)+12);
				EMIT(PPC_RAW_LI(r_ret, 0));
				PPC_JMP(exit_addr);
			}
			if (code == (BPF_ALU | BPF_MOD | BPF_X)) {
				EMIT(PPC_RAW_DIVWU(r_scratch1, r_A, r_X));
				EMIT(PPC_RAW_MULW(r_scratch1, r_X, r_scratch1));
				EMIT(PPC_RAW_SUB(r_A, r_A, r_scratch1));
			} else {
				EMIT(PPC_RAW_DIVWU(r_A, r_A, r_X));
			}
			break;
		case BPF_ALU | BPF_MOD | BPF_K: /* A %= K; */
			PPC_LI32(r_scratch2, K);
			EMIT(PPC_RAW_DIVWU(r_scratch1, r_A, r_scratch2));
			EMIT(PPC_RAW_MULW(r_scratch1, r_scratch2, r_scratch1));
			EMIT(PPC_RAW_SUB(r_A, r_A, r_scratch1));
			break;
		case BPF_ALU | BPF_DIV | BPF_K: /* A /= K */
			if (K == 1)
				break;
			PPC_LI32(r_scratch1, K);
			EMIT(PPC_RAW_DIVWU(r_A, r_A, r_scratch1));
			break;
		case BPF_ALU | BPF_AND | BPF_X:
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_AND(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_AND | BPF_K:
			if (!IMM_H(K))
				EMIT(PPC_RAW_ANDI(r_A, r_A, K));
			else {
				PPC_LI32(r_scratch1, K);
				EMIT(PPC_RAW_AND(r_A, r_A, r_scratch1));
			}
			break;
		case BPF_ALU | BPF_OR | BPF_X:
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_OR(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_OR | BPF_K:
			if (IMM_L(K))
				EMIT(PPC_RAW_ORI(r_A, r_A, IMM_L(K)));
			if (K >= 65536)
				EMIT(PPC_RAW_ORIS(r_A, r_A, IMM_H(K)));
			break;
		case BPF_ANC | SKF_AD_ALU_XOR_X:
		case BPF_ALU | BPF_XOR | BPF_X: /* A ^= X */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_XOR(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_XOR | BPF_K: /* A ^= K */
			if (IMM_L(K))
				EMIT(PPC_RAW_XORI(r_A, r_A, IMM_L(K)));
			if (K >= 65536)
				EMIT(PPC_RAW_XORIS(r_A, r_A, IMM_H(K)));
			break;
		case BPF_ALU | BPF_LSH | BPF_X: /* A <<= X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_SLW(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_LSH | BPF_K:
			if (K == 0)
				break;
			else
				EMIT(PPC_RAW_SLWI(r_A, r_A, K));
			break;
		case BPF_ALU | BPF_RSH | BPF_X: /* A >>= X; */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_SRW(r_A, r_A, r_X));
			break;
		case BPF_ALU | BPF_RSH | BPF_K: /* A >>= K; */
			if (K == 0)
				break;
			else
				EMIT(PPC_RAW_SRWI(r_A, r_A, K));
			break;
		case BPF_ALU | BPF_NEG:
			EMIT(PPC_RAW_NEG(r_A, r_A));
			break;
		case BPF_RET | BPF_K:
			PPC_LI32(r_ret, K);
			if (!K) {
				if (ctx->pc_ret0 == -1)
					ctx->pc_ret0 = i;
			}
			/*
			 * If this isn't the very last instruction, branch to
			 * the epilogue if we've stuff to clean up.  Otherwise,
			 * if there's nothing to tidy, just return.  If we /are/
			 * the last instruction, we're about to fall through to
			 * the epilogue to return.
			 */
			if (i != flen - 1) {
				/*
				 * Note: 'seen' is properly valid only on pass
				 * #2.	Both parts of this conditional are the
				 * same instruction size though, meaning the
				 * first pass will still correctly determine the
				 * code size/addresses.
				 */
				if (ctx->seen)
					PPC_JMP(exit_addr);
				else
					EMIT(PPC_RAW_BLR());
			}
			break;
		case BPF_RET | BPF_A:
			EMIT(PPC_RAW_MR(r_ret, r_A));
			if (i != flen - 1) {
				if (ctx->seen)
					PPC_JMP(exit_addr);
				else
					EMIT(PPC_RAW_BLR());
			}
			break;
		case BPF_MISC | BPF_TAX: /* X = A */
			EMIT(PPC_RAW_MR(r_X, r_A));
			break;
		case BPF_MISC | BPF_TXA: /* A = X */
			ctx->seen |= SEEN_XREG;
			EMIT(PPC_RAW_MR(r_A, r_X));
			break;

			/*** Constant loads/M[] access ***/
		case BPF_LD | BPF_IMM: /* A = K */
			PPC_LI32(r_A, K);
			break;
		case BPF_LDX | BPF_IMM: /* X = K */
			PPC_LI32(r_X, K);
			break;
		case BPF_LD | BPF_MEM: /* A = mem[K] */
			EMIT(PPC_RAW_MR(r_A, r_M + (K & 0xf)));
			ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
			break;
		case BPF_LDX | BPF_MEM: /* X = mem[K] */
			EMIT(PPC_RAW_MR(r_X, r_M + (K & 0xf)));
			ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
			break;
		case BPF_ST: /* mem[K] = A */
			EMIT(PPC_RAW_MR(r_M + (K & 0xf), r_A));
			ctx->seen |= SEEN_MEM | (1<<(K & 0xf));
			break;
		case BPF_STX: /* mem[K] = X */
			EMIT(PPC_RAW_MR(r_M + (K & 0xf), r_X));
			ctx->seen |= SEEN_XREG | SEEN_MEM | (1<<(K & 0xf));
			break;
		case BPF_LD | BPF_W | BPF_LEN: /*	A = skb->len; */
			BUILD_BUG_ON(sizeof_field(struct sk_buff, len) != 4);
			PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff, len));
			break;
		case BPF_LDX | BPF_W | BPF_ABS: /* A = *((u32 *)(seccomp_data + K)); */
			PPC_LWZ_OFFS(r_A, r_skb, K);
			break;
		case BPF_LDX | BPF_W | BPF_LEN: /* X = skb->len; */
			PPC_LWZ_OFFS(r_X, r_skb, offsetof(struct sk_buff, len));
			break;

			/*** Ancillary info loads ***/
		case BPF_ANC | SKF_AD_PROTOCOL: /* A = ntohs(skb->protocol); */
			BUILD_BUG_ON(sizeof_field(struct sk_buff,
						  protocol) != 2);
			PPC_NTOHS_OFFS(r_A, r_skb, offsetof(struct sk_buff,
							    protocol));
			break;
		case BPF_ANC | SKF_AD_IFINDEX:
		case BPF_ANC | SKF_AD_HATYPE:
			BUILD_BUG_ON(sizeof_field(struct net_device,
						ifindex) != 4);
			BUILD_BUG_ON(sizeof_field(struct net_device,
						type) != 2);
			PPC_LL_OFFS(r_scratch1, r_skb, offsetof(struct sk_buff,
								dev));
			EMIT(PPC_RAW_CMPDI(r_scratch1, 0));
			if (ctx->pc_ret0 != -1) {
				PPC_BCC(COND_EQ, addrs[ctx->pc_ret0]);
			} else {
				/* Exit, returning 0; first pass hits here. */
				PPC_BCC_SHORT(COND_NE, ctx->idx * 4 + 12);
				EMIT(PPC_RAW_LI(r_ret, 0));
				PPC_JMP(exit_addr);
			}
			if (code == (BPF_ANC | SKF_AD_IFINDEX)) {
				PPC_LWZ_OFFS(r_A, r_scratch1,
				     offsetof(struct net_device, ifindex));
			} else {
				PPC_LHZ_OFFS(r_A, r_scratch1,
				     offsetof(struct net_device, type));
			}

			break;
		case BPF_ANC | SKF_AD_MARK:
			BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
			PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
							  mark));
			break;
		case BPF_ANC | SKF_AD_RXHASH:
			BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
			PPC_LWZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
							  hash));
			break;
		case BPF_ANC | SKF_AD_VLAN_TAG:
			BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);

			PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
							  vlan_tci));
			break;
		case BPF_ANC | SKF_AD_VLAN_TAG_PRESENT:
			PPC_LBZ_OFFS(r_A, r_skb, PKT_VLAN_PRESENT_OFFSET());
			if (PKT_VLAN_PRESENT_BIT)
				EMIT(PPC_RAW_SRWI(r_A, r_A, PKT_VLAN_PRESENT_BIT));
			if (PKT_VLAN_PRESENT_BIT < 7)
				EMIT(PPC_RAW_ANDI(r_A, r_A, 1));
			break;
		case BPF_ANC | SKF_AD_QUEUE:
			BUILD_BUG_ON(sizeof_field(struct sk_buff,
						  queue_mapping) != 2);
			PPC_LHZ_OFFS(r_A, r_skb, offsetof(struct sk_buff,
							  queue_mapping));
			break;
		case BPF_ANC | SKF_AD_PKTTYPE:
			PPC_LBZ_OFFS(r_A, r_skb, PKT_TYPE_OFFSET());
			EMIT(PPC_RAW_ANDI(r_A, r_A, PKT_TYPE_MAX));
			EMIT(PPC_RAW_SRWI(r_A, r_A, 5));
			break;
		case BPF_ANC | SKF_AD_CPU:
			PPC_BPF_LOAD_CPU(r_A);
			break;
			/*** Absolute loads from packet header/data ***/
		case BPF_LD | BPF_W | BPF_ABS:
			func = CHOOSE_LOAD_FUNC(K, sk_load_word);
			goto common_load;
		case BPF_LD | BPF_H | BPF_ABS:
			func = CHOOSE_LOAD_FUNC(K, sk_load_half);
			goto common_load;
		case BPF_LD | BPF_B | BPF_ABS:
			func = CHOOSE_LOAD_FUNC(K, sk_load_byte);
		common_load:
			/* Load from [K]. */
			ctx->seen |= SEEN_DATAREF;
			PPC_FUNC_ADDR(r_scratch1, func);
			EMIT(PPC_RAW_MTLR(r_scratch1));
			PPC_LI32(r_addr, K);
			EMIT(PPC_RAW_BLRL());
			/*
			 * Helper returns 'lt' condition on error, and an
			 * appropriate return value in r3
			 */
			PPC_BCC(COND_LT, exit_addr);
			break;

			/*** Indirect loads from packet header/data ***/
		case BPF_LD | BPF_W | BPF_IND:
			func = sk_load_word;
			goto common_load_ind;
		case BPF_LD | BPF_H | BPF_IND:
			func = sk_load_half;
			goto common_load_ind;
		case BPF_LD | BPF_B | BPF_IND:
			func = sk_load_byte;
		common_load_ind:
			/*
			 * Load from [X + K].  Negative offsets are tested for
			 * in the helper functions.
			 */
			ctx->seen |= SEEN_DATAREF | SEEN_XREG;
			PPC_FUNC_ADDR(r_scratch1, func);
			EMIT(PPC_RAW_MTLR(r_scratch1));
			EMIT(PPC_RAW_ADDI(r_addr, r_X, IMM_L(K)));
			if (K >= 32768)
				EMIT(PPC_RAW_ADDIS(r_addr, r_addr, IMM_HA(K)));
			EMIT(PPC_RAW_BLRL());
			/* If error, cr0.LT set */
			PPC_BCC(COND_LT, exit_addr);
			break;

		case BPF_LDX | BPF_B | BPF_MSH:
			func = CHOOSE_LOAD_FUNC(K, sk_load_byte_msh);
			goto common_load;
			break;

			/*** Jump and branches ***/
		case BPF_JMP | BPF_JA:
			if (K != 0)
				PPC_JMP(addrs[i + 1 + K]);
			break;

		case BPF_JMP | BPF_JGT | BPF_K:
		case BPF_JMP | BPF_JGT | BPF_X:
			true_cond = COND_GT;
			goto cond_branch;
		case BPF_JMP | BPF_JGE | BPF_K:
		case BPF_JMP | BPF_JGE | BPF_X:
			true_cond = COND_GE;
			goto cond_branch;
		case BPF_JMP | BPF_JEQ | BPF_K:
		case BPF_JMP | BPF_JEQ | BPF_X:
			true_cond = COND_EQ;
			goto cond_branch;
		case BPF_JMP | BPF_JSET | BPF_K:
		case BPF_JMP | BPF_JSET | BPF_X:
			true_cond = COND_NE;
			fallthrough;
		cond_branch:
			/* same targets, can avoid doing the test :) */
			if (filter[i].jt == filter[i].jf) {
				if (filter[i].jt > 0)
					PPC_JMP(addrs[i + 1 + filter[i].jt]);
				break;
			}

			switch (code) {
			case BPF_JMP | BPF_JGT | BPF_X:
			case BPF_JMP | BPF_JGE | BPF_X:
			case BPF_JMP | BPF_JEQ | BPF_X:
				ctx->seen |= SEEN_XREG;
				EMIT(PPC_RAW_CMPLW(r_A, r_X));
				break;
			case BPF_JMP | BPF_JSET | BPF_X:
				ctx->seen |= SEEN_XREG;
				EMIT(PPC_RAW_AND_DOT(r_scratch1, r_A, r_X));
				break;
			case BPF_JMP | BPF_JEQ | BPF_K:
			case BPF_JMP | BPF_JGT | BPF_K:
			case BPF_JMP | BPF_JGE | BPF_K:
				if (K < 32768)
					EMIT(PPC_RAW_CMPLWI(r_A, K));
				else {
					PPC_LI32(r_scratch1, K);
					EMIT(PPC_RAW_CMPLW(r_A, r_scratch1));
				}
				break;
			case BPF_JMP | BPF_JSET | BPF_K:
				if (K < 32768)
					/* PPC_ANDI is /only/ dot-form */
					EMIT(PPC_RAW_ANDI(r_scratch1, r_A, K));
				else {
					PPC_LI32(r_scratch1, K);
					EMIT(PPC_RAW_AND_DOT(r_scratch1, r_A,
						    r_scratch1));
				}
				break;
			}
			/* Sometimes branches are constructed "backward", with
			 * the false path being the branch and true path being
			 * a fallthrough to the next instruction.
			 */
			if (filter[i].jt == 0)
				/* Swap the sense of the branch */
				PPC_BCC(true_cond ^ COND_CMP_TRUE,
					addrs[i + 1 + filter[i].jf]);
			else {
				PPC_BCC(true_cond, addrs[i + 1 + filter[i].jt]);
				if (filter[i].jf != 0)
					PPC_JMP(addrs[i + 1 + filter[i].jf]);
			}
			break;
		default:
			/* The filter contains something cruel & unusual.
			 * We don't handle it, but also there shouldn't be
			 * anything missing from our list.
			 */
			if (printk_ratelimit())
				pr_err("BPF filter opcode %04x (@%d) unsupported\n",
				       filter[i].code, i);
			return -ENOTSUPP;
		}

	}
	/* Set end-of-body-code address for exit. */
	addrs[i] = ctx->idx * 4;

	return 0;
}

void bpf_jit_compile(struct bpf_prog *fp)
{
	unsigned int proglen;
	unsigned int alloclen;
	u32 *image = NULL;
	u32 *code_base;
	unsigned int *addrs;
	struct codegen_context cgctx;
	int pass;
	int flen = fp->len;

	if (!bpf_jit_enable)
		return;

	addrs = kcalloc(flen + 1, sizeof(*addrs), GFP_KERNEL);
	if (addrs == NULL)
		return;

	/*
	 * There are multiple assembly passes as the generated code will change
	 * size as it settles down, figuring out the max branch offsets/exit
	 * paths required.
	 *
	 * The range of standard conditional branches is +/- 32Kbytes.	Since
	 * BPF_MAXINSNS = 4096, we can only jump from (worst case) start to
	 * finish with 8 bytes/instruction.  Not feasible, so long jumps are
	 * used, distinct from short branches.
	 *
	 * Current:
	 *
	 * For now, both branch types assemble to 2 words (short branches padded
	 * with a NOP); this is less efficient, but assembly will always complete
	 * after exactly 3 passes:
	 *
	 * First pass: No code buffer; Program is "faux-generated" -- no code
	 * emitted but maximum size of output determined (and addrs[] filled
	 * in).	 Also, we note whether we use M[], whether we use skb data, etc.
	 * All generation choices assumed to be 'worst-case', e.g. branches all
	 * far (2 instructions), return path code reduction not available, etc.
	 *
	 * Second pass: Code buffer allocated with size determined previously.
	 * Prologue generated to support features we have seen used.  Exit paths
	 * determined and addrs[] is filled in again, as code may be slightly
	 * smaller as a result.
	 *
	 * Third pass: Code generated 'for real', and branch destinations
	 * determined from now-accurate addrs[] map.
	 *
	 * Ideal:
	 *
	 * If we optimise this, near branches will be shorter.	On the
	 * first assembly pass, we should err on the side of caution and
	 * generate the biggest code.  On subsequent passes, branches will be
	 * generated short or long and code size will reduce.  With smaller
	 * code, more branches may fall into the short category, and code will
	 * reduce more.
	 *
	 * Finally, if we see one pass generate code the same size as the
	 * previous pass we have converged and should now generate code for
	 * real.  Allocating at the end will also save the memory that would
	 * otherwise be wasted by the (small) current code shrinkage.
	 * Preferably, we should do a small number of passes (e.g. 5) and if we
	 * haven't converged by then, get impatient and force code to generate
	 * as-is, even if the odd branch would be left long.  The chances of a
	 * long jump are tiny with all but the most enormous of BPF filter
	 * inputs, so we should usually converge on the third pass.
	 */

	cgctx.idx = 0;
	cgctx.seen = 0;
	cgctx.pc_ret0 = -1;
	/* Scouting faux-generate pass 0 */
	if (bpf_jit_build_body(fp, 0, &cgctx, addrs))
		/* We hit something illegal or unsupported. */
		goto out;

	/*
	 * Pretend to build prologue, given the features we've seen.  This will
	 * update ctgtx.idx as it pretends to output instructions, then we can
	 * calculate total size from idx.
	 */
	bpf_jit_build_prologue(fp, 0, &cgctx);
	bpf_jit_build_epilogue(0, &cgctx);

	proglen = cgctx.idx * 4;
	alloclen = proglen + FUNCTION_DESCR_SIZE;
	image = module_alloc(alloclen);
	if (!image)
		goto out;

	code_base = image + (FUNCTION_DESCR_SIZE/4);

	/* Code generation passes 1-2 */
	for (pass = 1; pass < 3; pass++) {
		/* Now build the prologue, body code & epilogue for real. */
		cgctx.idx = 0;
		bpf_jit_build_prologue(fp, code_base, &cgctx);
		bpf_jit_build_body(fp, code_base, &cgctx, addrs);
		bpf_jit_build_epilogue(code_base, &cgctx);

		if (bpf_jit_enable > 1)
			pr_info("Pass %d: shrink = %d, seen = 0x%x\n", pass,
				proglen - (cgctx.idx * 4), cgctx.seen);
	}

	if (bpf_jit_enable > 1)
		/* Note that we output the base address of the code_base
		 * rather than image, since opcodes are in code_base.
		 */
		bpf_jit_dump(flen, proglen, pass, code_base);

	bpf_flush_icache(code_base, code_base + (proglen/4));

#ifdef CONFIG_PPC64
	/* Function descriptor nastiness: Address + TOC */
	((u64 *)image)[0] = (u64)code_base;
	((u64 *)image)[1] = local_paca->kernel_toc;
#endif

	fp->bpf_func = (void *)image;
	fp->jited = 1;

out:
	kfree(addrs);
	return;
}

void bpf_jit_free(struct bpf_prog *fp)
{
	if (fp->jited)
		module_memfree(fp->bpf_func);

	bpf_prog_unlock_free(fp);
}