xref: /linux/arch/x86/net/bpf_jit_comp.c (revision 6be8e297f9bcea666ea85ac7a6cd9d52d6deaf92)
1 /*
2  * bpf_jit_comp.c: BPF JIT compiler
3  *
4  * Copyright (C) 2011-2013 Eric Dumazet (eric.dumazet@gmail.com)
5  * Internal BPF Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.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/netdevice.h>
13 #include <linux/filter.h>
14 #include <linux/if_vlan.h>
15 #include <linux/bpf.h>
16 
17 #include <asm/set_memory.h>
18 #include <asm/nospec-branch.h>
19 
20 static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
21 {
22 	if (len == 1)
23 		*ptr = bytes;
24 	else if (len == 2)
25 		*(u16 *)ptr = bytes;
26 	else {
27 		*(u32 *)ptr = bytes;
28 		barrier();
29 	}
30 	return ptr + len;
31 }
32 
33 #define EMIT(bytes, len) \
34 	do { prog = emit_code(prog, bytes, len); cnt += len; } while (0)
35 
36 #define EMIT1(b1)		EMIT(b1, 1)
37 #define EMIT2(b1, b2)		EMIT((b1) + ((b2) << 8), 2)
38 #define EMIT3(b1, b2, b3)	EMIT((b1) + ((b2) << 8) + ((b3) << 16), 3)
39 #define EMIT4(b1, b2, b3, b4)   EMIT((b1) + ((b2) << 8) + ((b3) << 16) + ((b4) << 24), 4)
40 
41 #define EMIT1_off32(b1, off) \
42 	do { EMIT1(b1); EMIT(off, 4); } while (0)
43 #define EMIT2_off32(b1, b2, off) \
44 	do { EMIT2(b1, b2); EMIT(off, 4); } while (0)
45 #define EMIT3_off32(b1, b2, b3, off) \
46 	do { EMIT3(b1, b2, b3); EMIT(off, 4); } while (0)
47 #define EMIT4_off32(b1, b2, b3, b4, off) \
48 	do { EMIT4(b1, b2, b3, b4); EMIT(off, 4); } while (0)
49 
50 static bool is_imm8(int value)
51 {
52 	return value <= 127 && value >= -128;
53 }
54 
55 static bool is_simm32(s64 value)
56 {
57 	return value == (s64)(s32)value;
58 }
59 
60 static bool is_uimm32(u64 value)
61 {
62 	return value == (u64)(u32)value;
63 }
64 
65 /* mov dst, src */
66 #define EMIT_mov(DST, SRC)								 \
67 	do {										 \
68 		if (DST != SRC)								 \
69 			EMIT3(add_2mod(0x48, DST, SRC), 0x89, add_2reg(0xC0, DST, SRC)); \
70 	} while (0)
71 
72 static int bpf_size_to_x86_bytes(int bpf_size)
73 {
74 	if (bpf_size == BPF_W)
75 		return 4;
76 	else if (bpf_size == BPF_H)
77 		return 2;
78 	else if (bpf_size == BPF_B)
79 		return 1;
80 	else if (bpf_size == BPF_DW)
81 		return 4; /* imm32 */
82 	else
83 		return 0;
84 }
85 
86 /*
87  * List of x86 cond jumps opcodes (. + s8)
88  * Add 0x10 (and an extra 0x0f) to generate far jumps (. + s32)
89  */
90 #define X86_JB  0x72
91 #define X86_JAE 0x73
92 #define X86_JE  0x74
93 #define X86_JNE 0x75
94 #define X86_JBE 0x76
95 #define X86_JA  0x77
96 #define X86_JL  0x7C
97 #define X86_JGE 0x7D
98 #define X86_JLE 0x7E
99 #define X86_JG  0x7F
100 
101 /* Pick a register outside of BPF range for JIT internal work */
102 #define AUX_REG (MAX_BPF_JIT_REG + 1)
103 
104 /*
105  * The following table maps BPF registers to x86-64 registers.
106  *
107  * x86-64 register R12 is unused, since if used as base address
108  * register in load/store instructions, it always needs an
109  * extra byte of encoding and is callee saved.
110  *
111  * Also x86-64 register R9 is unused. x86-64 register R10 is
112  * used for blinding (if enabled).
113  */
114 static const int reg2hex[] = {
115 	[BPF_REG_0] = 0,  /* RAX */
116 	[BPF_REG_1] = 7,  /* RDI */
117 	[BPF_REG_2] = 6,  /* RSI */
118 	[BPF_REG_3] = 2,  /* RDX */
119 	[BPF_REG_4] = 1,  /* RCX */
120 	[BPF_REG_5] = 0,  /* R8  */
121 	[BPF_REG_6] = 3,  /* RBX callee saved */
122 	[BPF_REG_7] = 5,  /* R13 callee saved */
123 	[BPF_REG_8] = 6,  /* R14 callee saved */
124 	[BPF_REG_9] = 7,  /* R15 callee saved */
125 	[BPF_REG_FP] = 5, /* RBP readonly */
126 	[BPF_REG_AX] = 2, /* R10 temp register */
127 	[AUX_REG] = 3,    /* R11 temp register */
128 };
129 
130 /*
131  * is_ereg() == true if BPF register 'reg' maps to x86-64 r8..r15
132  * which need extra byte of encoding.
133  * rax,rcx,...,rbp have simpler encoding
134  */
135 static bool is_ereg(u32 reg)
136 {
137 	return (1 << reg) & (BIT(BPF_REG_5) |
138 			     BIT(AUX_REG) |
139 			     BIT(BPF_REG_7) |
140 			     BIT(BPF_REG_8) |
141 			     BIT(BPF_REG_9) |
142 			     BIT(BPF_REG_AX));
143 }
144 
145 static bool is_axreg(u32 reg)
146 {
147 	return reg == BPF_REG_0;
148 }
149 
150 /* Add modifiers if 'reg' maps to x86-64 registers R8..R15 */
151 static u8 add_1mod(u8 byte, u32 reg)
152 {
153 	if (is_ereg(reg))
154 		byte |= 1;
155 	return byte;
156 }
157 
158 static u8 add_2mod(u8 byte, u32 r1, u32 r2)
159 {
160 	if (is_ereg(r1))
161 		byte |= 1;
162 	if (is_ereg(r2))
163 		byte |= 4;
164 	return byte;
165 }
166 
167 /* Encode 'dst_reg' register into x86-64 opcode 'byte' */
168 static u8 add_1reg(u8 byte, u32 dst_reg)
169 {
170 	return byte + reg2hex[dst_reg];
171 }
172 
173 /* Encode 'dst_reg' and 'src_reg' registers into x86-64 opcode 'byte' */
174 static u8 add_2reg(u8 byte, u32 dst_reg, u32 src_reg)
175 {
176 	return byte + reg2hex[dst_reg] + (reg2hex[src_reg] << 3);
177 }
178 
179 static void jit_fill_hole(void *area, unsigned int size)
180 {
181 	/* Fill whole space with INT3 instructions */
182 	memset(area, 0xcc, size);
183 }
184 
185 struct jit_context {
186 	int cleanup_addr; /* Epilogue code offset */
187 };
188 
189 /* Maximum number of bytes emitted while JITing one eBPF insn */
190 #define BPF_MAX_INSN_SIZE	128
191 #define BPF_INSN_SAFETY		64
192 
193 #define PROLOGUE_SIZE		20
194 
195 /*
196  * Emit x86-64 prologue code for BPF program and check its size.
197  * bpf_tail_call helper will skip it while jumping into another program
198  */
199 static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf)
200 {
201 	u8 *prog = *pprog;
202 	int cnt = 0;
203 
204 	EMIT1(0x55);             /* push rbp */
205 	EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */
206 	/* sub rsp, rounded_stack_depth */
207 	EMIT3_off32(0x48, 0x81, 0xEC, round_up(stack_depth, 8));
208 	EMIT1(0x53);             /* push rbx */
209 	EMIT2(0x41, 0x55);       /* push r13 */
210 	EMIT2(0x41, 0x56);       /* push r14 */
211 	EMIT2(0x41, 0x57);       /* push r15 */
212 	if (!ebpf_from_cbpf) {
213 		/* zero init tail_call_cnt */
214 		EMIT2(0x6a, 0x00);
215 		BUILD_BUG_ON(cnt != PROLOGUE_SIZE);
216 	}
217 	*pprog = prog;
218 }
219 
220 /*
221  * Generate the following code:
222  *
223  * ... bpf_tail_call(void *ctx, struct bpf_array *array, u64 index) ...
224  *   if (index >= array->map.max_entries)
225  *     goto out;
226  *   if (++tail_call_cnt > MAX_TAIL_CALL_CNT)
227  *     goto out;
228  *   prog = array->ptrs[index];
229  *   if (prog == NULL)
230  *     goto out;
231  *   goto *(prog->bpf_func + prologue_size);
232  * out:
233  */
234 static void emit_bpf_tail_call(u8 **pprog)
235 {
236 	u8 *prog = *pprog;
237 	int label1, label2, label3;
238 	int cnt = 0;
239 
240 	/*
241 	 * rdi - pointer to ctx
242 	 * rsi - pointer to bpf_array
243 	 * rdx - index in bpf_array
244 	 */
245 
246 	/*
247 	 * if (index >= array->map.max_entries)
248 	 *	goto out;
249 	 */
250 	EMIT2(0x89, 0xD2);                        /* mov edx, edx */
251 	EMIT3(0x39, 0x56,                         /* cmp dword ptr [rsi + 16], edx */
252 	      offsetof(struct bpf_array, map.max_entries));
253 #define OFFSET1 (41 + RETPOLINE_RAX_BPF_JIT_SIZE) /* Number of bytes to jump */
254 	EMIT2(X86_JBE, OFFSET1);                  /* jbe out */
255 	label1 = cnt;
256 
257 	/*
258 	 * if (tail_call_cnt > MAX_TAIL_CALL_CNT)
259 	 *	goto out;
260 	 */
261 	EMIT2_off32(0x8B, 0x85, -36 - MAX_BPF_STACK); /* mov eax, dword ptr [rbp - 548] */
262 	EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT);     /* cmp eax, MAX_TAIL_CALL_CNT */
263 #define OFFSET2 (30 + RETPOLINE_RAX_BPF_JIT_SIZE)
264 	EMIT2(X86_JA, OFFSET2);                   /* ja out */
265 	label2 = cnt;
266 	EMIT3(0x83, 0xC0, 0x01);                  /* add eax, 1 */
267 	EMIT2_off32(0x89, 0x85, -36 - MAX_BPF_STACK); /* mov dword ptr [rbp -548], eax */
268 
269 	/* prog = array->ptrs[index]; */
270 	EMIT4_off32(0x48, 0x8B, 0x84, 0xD6,       /* mov rax, [rsi + rdx * 8 + offsetof(...)] */
271 		    offsetof(struct bpf_array, ptrs));
272 
273 	/*
274 	 * if (prog == NULL)
275 	 *	goto out;
276 	 */
277 	EMIT3(0x48, 0x85, 0xC0);		  /* test rax,rax */
278 #define OFFSET3 (8 + RETPOLINE_RAX_BPF_JIT_SIZE)
279 	EMIT2(X86_JE, OFFSET3);                   /* je out */
280 	label3 = cnt;
281 
282 	/* goto *(prog->bpf_func + prologue_size); */
283 	EMIT4(0x48, 0x8B, 0x40,                   /* mov rax, qword ptr [rax + 32] */
284 	      offsetof(struct bpf_prog, bpf_func));
285 	EMIT4(0x48, 0x83, 0xC0, PROLOGUE_SIZE);   /* add rax, prologue_size */
286 
287 	/*
288 	 * Wow we're ready to jump into next BPF program
289 	 * rdi == ctx (1st arg)
290 	 * rax == prog->bpf_func + prologue_size
291 	 */
292 	RETPOLINE_RAX_BPF_JIT();
293 
294 	/* out: */
295 	BUILD_BUG_ON(cnt - label1 != OFFSET1);
296 	BUILD_BUG_ON(cnt - label2 != OFFSET2);
297 	BUILD_BUG_ON(cnt - label3 != OFFSET3);
298 	*pprog = prog;
299 }
300 
301 static void emit_mov_imm32(u8 **pprog, bool sign_propagate,
302 			   u32 dst_reg, const u32 imm32)
303 {
304 	u8 *prog = *pprog;
305 	u8 b1, b2, b3;
306 	int cnt = 0;
307 
308 	/*
309 	 * Optimization: if imm32 is positive, use 'mov %eax, imm32'
310 	 * (which zero-extends imm32) to save 2 bytes.
311 	 */
312 	if (sign_propagate && (s32)imm32 < 0) {
313 		/* 'mov %rax, imm32' sign extends imm32 */
314 		b1 = add_1mod(0x48, dst_reg);
315 		b2 = 0xC7;
316 		b3 = 0xC0;
317 		EMIT3_off32(b1, b2, add_1reg(b3, dst_reg), imm32);
318 		goto done;
319 	}
320 
321 	/*
322 	 * Optimization: if imm32 is zero, use 'xor %eax, %eax'
323 	 * to save 3 bytes.
324 	 */
325 	if (imm32 == 0) {
326 		if (is_ereg(dst_reg))
327 			EMIT1(add_2mod(0x40, dst_reg, dst_reg));
328 		b2 = 0x31; /* xor */
329 		b3 = 0xC0;
330 		EMIT2(b2, add_2reg(b3, dst_reg, dst_reg));
331 		goto done;
332 	}
333 
334 	/* mov %eax, imm32 */
335 	if (is_ereg(dst_reg))
336 		EMIT1(add_1mod(0x40, dst_reg));
337 	EMIT1_off32(add_1reg(0xB8, dst_reg), imm32);
338 done:
339 	*pprog = prog;
340 }
341 
342 static void emit_mov_imm64(u8 **pprog, u32 dst_reg,
343 			   const u32 imm32_hi, const u32 imm32_lo)
344 {
345 	u8 *prog = *pprog;
346 	int cnt = 0;
347 
348 	if (is_uimm32(((u64)imm32_hi << 32) | (u32)imm32_lo)) {
349 		/*
350 		 * For emitting plain u32, where sign bit must not be
351 		 * propagated LLVM tends to load imm64 over mov32
352 		 * directly, so save couple of bytes by just doing
353 		 * 'mov %eax, imm32' instead.
354 		 */
355 		emit_mov_imm32(&prog, false, dst_reg, imm32_lo);
356 	} else {
357 		/* movabsq %rax, imm64 */
358 		EMIT2(add_1mod(0x48, dst_reg), add_1reg(0xB8, dst_reg));
359 		EMIT(imm32_lo, 4);
360 		EMIT(imm32_hi, 4);
361 	}
362 
363 	*pprog = prog;
364 }
365 
366 static void emit_mov_reg(u8 **pprog, bool is64, u32 dst_reg, u32 src_reg)
367 {
368 	u8 *prog = *pprog;
369 	int cnt = 0;
370 
371 	if (is64) {
372 		/* mov dst, src */
373 		EMIT_mov(dst_reg, src_reg);
374 	} else {
375 		/* mov32 dst, src */
376 		if (is_ereg(dst_reg) || is_ereg(src_reg))
377 			EMIT1(add_2mod(0x40, dst_reg, src_reg));
378 		EMIT2(0x89, add_2reg(0xC0, dst_reg, src_reg));
379 	}
380 
381 	*pprog = prog;
382 }
383 
384 static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
385 		  int oldproglen, struct jit_context *ctx)
386 {
387 	struct bpf_insn *insn = bpf_prog->insnsi;
388 	int insn_cnt = bpf_prog->len;
389 	bool seen_exit = false;
390 	u8 temp[BPF_MAX_INSN_SIZE + BPF_INSN_SAFETY];
391 	int i, cnt = 0;
392 	int proglen = 0;
393 	u8 *prog = temp;
394 
395 	emit_prologue(&prog, bpf_prog->aux->stack_depth,
396 		      bpf_prog_was_classic(bpf_prog));
397 
398 	for (i = 0; i < insn_cnt; i++, insn++) {
399 		const s32 imm32 = insn->imm;
400 		u32 dst_reg = insn->dst_reg;
401 		u32 src_reg = insn->src_reg;
402 		u8 b2 = 0, b3 = 0;
403 		s64 jmp_offset;
404 		u8 jmp_cond;
405 		int ilen;
406 		u8 *func;
407 
408 		switch (insn->code) {
409 			/* ALU */
410 		case BPF_ALU | BPF_ADD | BPF_X:
411 		case BPF_ALU | BPF_SUB | BPF_X:
412 		case BPF_ALU | BPF_AND | BPF_X:
413 		case BPF_ALU | BPF_OR | BPF_X:
414 		case BPF_ALU | BPF_XOR | BPF_X:
415 		case BPF_ALU64 | BPF_ADD | BPF_X:
416 		case BPF_ALU64 | BPF_SUB | BPF_X:
417 		case BPF_ALU64 | BPF_AND | BPF_X:
418 		case BPF_ALU64 | BPF_OR | BPF_X:
419 		case BPF_ALU64 | BPF_XOR | BPF_X:
420 			switch (BPF_OP(insn->code)) {
421 			case BPF_ADD: b2 = 0x01; break;
422 			case BPF_SUB: b2 = 0x29; break;
423 			case BPF_AND: b2 = 0x21; break;
424 			case BPF_OR: b2 = 0x09; break;
425 			case BPF_XOR: b2 = 0x31; break;
426 			}
427 			if (BPF_CLASS(insn->code) == BPF_ALU64)
428 				EMIT1(add_2mod(0x48, dst_reg, src_reg));
429 			else if (is_ereg(dst_reg) || is_ereg(src_reg))
430 				EMIT1(add_2mod(0x40, dst_reg, src_reg));
431 			EMIT2(b2, add_2reg(0xC0, dst_reg, src_reg));
432 			break;
433 
434 		case BPF_ALU64 | BPF_MOV | BPF_X:
435 		case BPF_ALU | BPF_MOV | BPF_X:
436 			emit_mov_reg(&prog,
437 				     BPF_CLASS(insn->code) == BPF_ALU64,
438 				     dst_reg, src_reg);
439 			break;
440 
441 			/* neg dst */
442 		case BPF_ALU | BPF_NEG:
443 		case BPF_ALU64 | BPF_NEG:
444 			if (BPF_CLASS(insn->code) == BPF_ALU64)
445 				EMIT1(add_1mod(0x48, dst_reg));
446 			else if (is_ereg(dst_reg))
447 				EMIT1(add_1mod(0x40, dst_reg));
448 			EMIT2(0xF7, add_1reg(0xD8, dst_reg));
449 			break;
450 
451 		case BPF_ALU | BPF_ADD | BPF_K:
452 		case BPF_ALU | BPF_SUB | BPF_K:
453 		case BPF_ALU | BPF_AND | BPF_K:
454 		case BPF_ALU | BPF_OR | BPF_K:
455 		case BPF_ALU | BPF_XOR | BPF_K:
456 		case BPF_ALU64 | BPF_ADD | BPF_K:
457 		case BPF_ALU64 | BPF_SUB | BPF_K:
458 		case BPF_ALU64 | BPF_AND | BPF_K:
459 		case BPF_ALU64 | BPF_OR | BPF_K:
460 		case BPF_ALU64 | BPF_XOR | BPF_K:
461 			if (BPF_CLASS(insn->code) == BPF_ALU64)
462 				EMIT1(add_1mod(0x48, dst_reg));
463 			else if (is_ereg(dst_reg))
464 				EMIT1(add_1mod(0x40, dst_reg));
465 
466 			/*
467 			 * b3 holds 'normal' opcode, b2 short form only valid
468 			 * in case dst is eax/rax.
469 			 */
470 			switch (BPF_OP(insn->code)) {
471 			case BPF_ADD:
472 				b3 = 0xC0;
473 				b2 = 0x05;
474 				break;
475 			case BPF_SUB:
476 				b3 = 0xE8;
477 				b2 = 0x2D;
478 				break;
479 			case BPF_AND:
480 				b3 = 0xE0;
481 				b2 = 0x25;
482 				break;
483 			case BPF_OR:
484 				b3 = 0xC8;
485 				b2 = 0x0D;
486 				break;
487 			case BPF_XOR:
488 				b3 = 0xF0;
489 				b2 = 0x35;
490 				break;
491 			}
492 
493 			if (is_imm8(imm32))
494 				EMIT3(0x83, add_1reg(b3, dst_reg), imm32);
495 			else if (is_axreg(dst_reg))
496 				EMIT1_off32(b2, imm32);
497 			else
498 				EMIT2_off32(0x81, add_1reg(b3, dst_reg), imm32);
499 			break;
500 
501 		case BPF_ALU64 | BPF_MOV | BPF_K:
502 		case BPF_ALU | BPF_MOV | BPF_K:
503 			emit_mov_imm32(&prog, BPF_CLASS(insn->code) == BPF_ALU64,
504 				       dst_reg, imm32);
505 			break;
506 
507 		case BPF_LD | BPF_IMM | BPF_DW:
508 			emit_mov_imm64(&prog, dst_reg, insn[1].imm, insn[0].imm);
509 			insn++;
510 			i++;
511 			break;
512 
513 			/* dst %= src, dst /= src, dst %= imm32, dst /= imm32 */
514 		case BPF_ALU | BPF_MOD | BPF_X:
515 		case BPF_ALU | BPF_DIV | BPF_X:
516 		case BPF_ALU | BPF_MOD | BPF_K:
517 		case BPF_ALU | BPF_DIV | BPF_K:
518 		case BPF_ALU64 | BPF_MOD | BPF_X:
519 		case BPF_ALU64 | BPF_DIV | BPF_X:
520 		case BPF_ALU64 | BPF_MOD | BPF_K:
521 		case BPF_ALU64 | BPF_DIV | BPF_K:
522 			EMIT1(0x50); /* push rax */
523 			EMIT1(0x52); /* push rdx */
524 
525 			if (BPF_SRC(insn->code) == BPF_X)
526 				/* mov r11, src_reg */
527 				EMIT_mov(AUX_REG, src_reg);
528 			else
529 				/* mov r11, imm32 */
530 				EMIT3_off32(0x49, 0xC7, 0xC3, imm32);
531 
532 			/* mov rax, dst_reg */
533 			EMIT_mov(BPF_REG_0, dst_reg);
534 
535 			/*
536 			 * xor edx, edx
537 			 * equivalent to 'xor rdx, rdx', but one byte less
538 			 */
539 			EMIT2(0x31, 0xd2);
540 
541 			if (BPF_CLASS(insn->code) == BPF_ALU64)
542 				/* div r11 */
543 				EMIT3(0x49, 0xF7, 0xF3);
544 			else
545 				/* div r11d */
546 				EMIT3(0x41, 0xF7, 0xF3);
547 
548 			if (BPF_OP(insn->code) == BPF_MOD)
549 				/* mov r11, rdx */
550 				EMIT3(0x49, 0x89, 0xD3);
551 			else
552 				/* mov r11, rax */
553 				EMIT3(0x49, 0x89, 0xC3);
554 
555 			EMIT1(0x5A); /* pop rdx */
556 			EMIT1(0x58); /* pop rax */
557 
558 			/* mov dst_reg, r11 */
559 			EMIT_mov(dst_reg, AUX_REG);
560 			break;
561 
562 		case BPF_ALU | BPF_MUL | BPF_K:
563 		case BPF_ALU | BPF_MUL | BPF_X:
564 		case BPF_ALU64 | BPF_MUL | BPF_K:
565 		case BPF_ALU64 | BPF_MUL | BPF_X:
566 		{
567 			bool is64 = BPF_CLASS(insn->code) == BPF_ALU64;
568 
569 			if (dst_reg != BPF_REG_0)
570 				EMIT1(0x50); /* push rax */
571 			if (dst_reg != BPF_REG_3)
572 				EMIT1(0x52); /* push rdx */
573 
574 			/* mov r11, dst_reg */
575 			EMIT_mov(AUX_REG, dst_reg);
576 
577 			if (BPF_SRC(insn->code) == BPF_X)
578 				emit_mov_reg(&prog, is64, BPF_REG_0, src_reg);
579 			else
580 				emit_mov_imm32(&prog, is64, BPF_REG_0, imm32);
581 
582 			if (is64)
583 				EMIT1(add_1mod(0x48, AUX_REG));
584 			else if (is_ereg(AUX_REG))
585 				EMIT1(add_1mod(0x40, AUX_REG));
586 			/* mul(q) r11 */
587 			EMIT2(0xF7, add_1reg(0xE0, AUX_REG));
588 
589 			if (dst_reg != BPF_REG_3)
590 				EMIT1(0x5A); /* pop rdx */
591 			if (dst_reg != BPF_REG_0) {
592 				/* mov dst_reg, rax */
593 				EMIT_mov(dst_reg, BPF_REG_0);
594 				EMIT1(0x58); /* pop rax */
595 			}
596 			break;
597 		}
598 			/* Shifts */
599 		case BPF_ALU | BPF_LSH | BPF_K:
600 		case BPF_ALU | BPF_RSH | BPF_K:
601 		case BPF_ALU | BPF_ARSH | BPF_K:
602 		case BPF_ALU64 | BPF_LSH | BPF_K:
603 		case BPF_ALU64 | BPF_RSH | BPF_K:
604 		case BPF_ALU64 | BPF_ARSH | BPF_K:
605 			if (BPF_CLASS(insn->code) == BPF_ALU64)
606 				EMIT1(add_1mod(0x48, dst_reg));
607 			else if (is_ereg(dst_reg))
608 				EMIT1(add_1mod(0x40, dst_reg));
609 
610 			switch (BPF_OP(insn->code)) {
611 			case BPF_LSH: b3 = 0xE0; break;
612 			case BPF_RSH: b3 = 0xE8; break;
613 			case BPF_ARSH: b3 = 0xF8; break;
614 			}
615 
616 			if (imm32 == 1)
617 				EMIT2(0xD1, add_1reg(b3, dst_reg));
618 			else
619 				EMIT3(0xC1, add_1reg(b3, dst_reg), imm32);
620 			break;
621 
622 		case BPF_ALU | BPF_LSH | BPF_X:
623 		case BPF_ALU | BPF_RSH | BPF_X:
624 		case BPF_ALU | BPF_ARSH | BPF_X:
625 		case BPF_ALU64 | BPF_LSH | BPF_X:
626 		case BPF_ALU64 | BPF_RSH | BPF_X:
627 		case BPF_ALU64 | BPF_ARSH | BPF_X:
628 
629 			/* Check for bad case when dst_reg == rcx */
630 			if (dst_reg == BPF_REG_4) {
631 				/* mov r11, dst_reg */
632 				EMIT_mov(AUX_REG, dst_reg);
633 				dst_reg = AUX_REG;
634 			}
635 
636 			if (src_reg != BPF_REG_4) { /* common case */
637 				EMIT1(0x51); /* push rcx */
638 
639 				/* mov rcx, src_reg */
640 				EMIT_mov(BPF_REG_4, src_reg);
641 			}
642 
643 			/* shl %rax, %cl | shr %rax, %cl | sar %rax, %cl */
644 			if (BPF_CLASS(insn->code) == BPF_ALU64)
645 				EMIT1(add_1mod(0x48, dst_reg));
646 			else if (is_ereg(dst_reg))
647 				EMIT1(add_1mod(0x40, dst_reg));
648 
649 			switch (BPF_OP(insn->code)) {
650 			case BPF_LSH: b3 = 0xE0; break;
651 			case BPF_RSH: b3 = 0xE8; break;
652 			case BPF_ARSH: b3 = 0xF8; break;
653 			}
654 			EMIT2(0xD3, add_1reg(b3, dst_reg));
655 
656 			if (src_reg != BPF_REG_4)
657 				EMIT1(0x59); /* pop rcx */
658 
659 			if (insn->dst_reg == BPF_REG_4)
660 				/* mov dst_reg, r11 */
661 				EMIT_mov(insn->dst_reg, AUX_REG);
662 			break;
663 
664 		case BPF_ALU | BPF_END | BPF_FROM_BE:
665 			switch (imm32) {
666 			case 16:
667 				/* Emit 'ror %ax, 8' to swap lower 2 bytes */
668 				EMIT1(0x66);
669 				if (is_ereg(dst_reg))
670 					EMIT1(0x41);
671 				EMIT3(0xC1, add_1reg(0xC8, dst_reg), 8);
672 
673 				/* Emit 'movzwl eax, ax' */
674 				if (is_ereg(dst_reg))
675 					EMIT3(0x45, 0x0F, 0xB7);
676 				else
677 					EMIT2(0x0F, 0xB7);
678 				EMIT1(add_2reg(0xC0, dst_reg, dst_reg));
679 				break;
680 			case 32:
681 				/* Emit 'bswap eax' to swap lower 4 bytes */
682 				if (is_ereg(dst_reg))
683 					EMIT2(0x41, 0x0F);
684 				else
685 					EMIT1(0x0F);
686 				EMIT1(add_1reg(0xC8, dst_reg));
687 				break;
688 			case 64:
689 				/* Emit 'bswap rax' to swap 8 bytes */
690 				EMIT3(add_1mod(0x48, dst_reg), 0x0F,
691 				      add_1reg(0xC8, dst_reg));
692 				break;
693 			}
694 			break;
695 
696 		case BPF_ALU | BPF_END | BPF_FROM_LE:
697 			switch (imm32) {
698 			case 16:
699 				/*
700 				 * Emit 'movzwl eax, ax' to zero extend 16-bit
701 				 * into 64 bit
702 				 */
703 				if (is_ereg(dst_reg))
704 					EMIT3(0x45, 0x0F, 0xB7);
705 				else
706 					EMIT2(0x0F, 0xB7);
707 				EMIT1(add_2reg(0xC0, dst_reg, dst_reg));
708 				break;
709 			case 32:
710 				/* Emit 'mov eax, eax' to clear upper 32-bits */
711 				if (is_ereg(dst_reg))
712 					EMIT1(0x45);
713 				EMIT2(0x89, add_2reg(0xC0, dst_reg, dst_reg));
714 				break;
715 			case 64:
716 				/* nop */
717 				break;
718 			}
719 			break;
720 
721 			/* ST: *(u8*)(dst_reg + off) = imm */
722 		case BPF_ST | BPF_MEM | BPF_B:
723 			if (is_ereg(dst_reg))
724 				EMIT2(0x41, 0xC6);
725 			else
726 				EMIT1(0xC6);
727 			goto st;
728 		case BPF_ST | BPF_MEM | BPF_H:
729 			if (is_ereg(dst_reg))
730 				EMIT3(0x66, 0x41, 0xC7);
731 			else
732 				EMIT2(0x66, 0xC7);
733 			goto st;
734 		case BPF_ST | BPF_MEM | BPF_W:
735 			if (is_ereg(dst_reg))
736 				EMIT2(0x41, 0xC7);
737 			else
738 				EMIT1(0xC7);
739 			goto st;
740 		case BPF_ST | BPF_MEM | BPF_DW:
741 			EMIT2(add_1mod(0x48, dst_reg), 0xC7);
742 
743 st:			if (is_imm8(insn->off))
744 				EMIT2(add_1reg(0x40, dst_reg), insn->off);
745 			else
746 				EMIT1_off32(add_1reg(0x80, dst_reg), insn->off);
747 
748 			EMIT(imm32, bpf_size_to_x86_bytes(BPF_SIZE(insn->code)));
749 			break;
750 
751 			/* STX: *(u8*)(dst_reg + off) = src_reg */
752 		case BPF_STX | BPF_MEM | BPF_B:
753 			/* Emit 'mov byte ptr [rax + off], al' */
754 			if (is_ereg(dst_reg) || is_ereg(src_reg) ||
755 			    /* We have to add extra byte for x86 SIL, DIL regs */
756 			    src_reg == BPF_REG_1 || src_reg == BPF_REG_2)
757 				EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x88);
758 			else
759 				EMIT1(0x88);
760 			goto stx;
761 		case BPF_STX | BPF_MEM | BPF_H:
762 			if (is_ereg(dst_reg) || is_ereg(src_reg))
763 				EMIT3(0x66, add_2mod(0x40, dst_reg, src_reg), 0x89);
764 			else
765 				EMIT2(0x66, 0x89);
766 			goto stx;
767 		case BPF_STX | BPF_MEM | BPF_W:
768 			if (is_ereg(dst_reg) || is_ereg(src_reg))
769 				EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x89);
770 			else
771 				EMIT1(0x89);
772 			goto stx;
773 		case BPF_STX | BPF_MEM | BPF_DW:
774 			EMIT2(add_2mod(0x48, dst_reg, src_reg), 0x89);
775 stx:			if (is_imm8(insn->off))
776 				EMIT2(add_2reg(0x40, dst_reg, src_reg), insn->off);
777 			else
778 				EMIT1_off32(add_2reg(0x80, dst_reg, src_reg),
779 					    insn->off);
780 			break;
781 
782 			/* LDX: dst_reg = *(u8*)(src_reg + off) */
783 		case BPF_LDX | BPF_MEM | BPF_B:
784 			/* Emit 'movzx rax, byte ptr [rax + off]' */
785 			EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB6);
786 			goto ldx;
787 		case BPF_LDX | BPF_MEM | BPF_H:
788 			/* Emit 'movzx rax, word ptr [rax + off]' */
789 			EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB7);
790 			goto ldx;
791 		case BPF_LDX | BPF_MEM | BPF_W:
792 			/* Emit 'mov eax, dword ptr [rax+0x14]' */
793 			if (is_ereg(dst_reg) || is_ereg(src_reg))
794 				EMIT2(add_2mod(0x40, src_reg, dst_reg), 0x8B);
795 			else
796 				EMIT1(0x8B);
797 			goto ldx;
798 		case BPF_LDX | BPF_MEM | BPF_DW:
799 			/* Emit 'mov rax, qword ptr [rax+0x14]' */
800 			EMIT2(add_2mod(0x48, src_reg, dst_reg), 0x8B);
801 ldx:			/*
802 			 * If insn->off == 0 we can save one extra byte, but
803 			 * special case of x86 R13 which always needs an offset
804 			 * is not worth the hassle
805 			 */
806 			if (is_imm8(insn->off))
807 				EMIT2(add_2reg(0x40, src_reg, dst_reg), insn->off);
808 			else
809 				EMIT1_off32(add_2reg(0x80, src_reg, dst_reg),
810 					    insn->off);
811 			break;
812 
813 			/* STX XADD: lock *(u32*)(dst_reg + off) += src_reg */
814 		case BPF_STX | BPF_XADD | BPF_W:
815 			/* Emit 'lock add dword ptr [rax + off], eax' */
816 			if (is_ereg(dst_reg) || is_ereg(src_reg))
817 				EMIT3(0xF0, add_2mod(0x40, dst_reg, src_reg), 0x01);
818 			else
819 				EMIT2(0xF0, 0x01);
820 			goto xadd;
821 		case BPF_STX | BPF_XADD | BPF_DW:
822 			EMIT3(0xF0, add_2mod(0x48, dst_reg, src_reg), 0x01);
823 xadd:			if (is_imm8(insn->off))
824 				EMIT2(add_2reg(0x40, dst_reg, src_reg), insn->off);
825 			else
826 				EMIT1_off32(add_2reg(0x80, dst_reg, src_reg),
827 					    insn->off);
828 			break;
829 
830 			/* call */
831 		case BPF_JMP | BPF_CALL:
832 			func = (u8 *) __bpf_call_base + imm32;
833 			jmp_offset = func - (image + addrs[i]);
834 			if (!imm32 || !is_simm32(jmp_offset)) {
835 				pr_err("unsupported BPF func %d addr %p image %p\n",
836 				       imm32, func, image);
837 				return -EINVAL;
838 			}
839 			EMIT1_off32(0xE8, jmp_offset);
840 			break;
841 
842 		case BPF_JMP | BPF_TAIL_CALL:
843 			emit_bpf_tail_call(&prog);
844 			break;
845 
846 			/* cond jump */
847 		case BPF_JMP | BPF_JEQ | BPF_X:
848 		case BPF_JMP | BPF_JNE | BPF_X:
849 		case BPF_JMP | BPF_JGT | BPF_X:
850 		case BPF_JMP | BPF_JLT | BPF_X:
851 		case BPF_JMP | BPF_JGE | BPF_X:
852 		case BPF_JMP | BPF_JLE | BPF_X:
853 		case BPF_JMP | BPF_JSGT | BPF_X:
854 		case BPF_JMP | BPF_JSLT | BPF_X:
855 		case BPF_JMP | BPF_JSGE | BPF_X:
856 		case BPF_JMP | BPF_JSLE | BPF_X:
857 		case BPF_JMP32 | BPF_JEQ | BPF_X:
858 		case BPF_JMP32 | BPF_JNE | BPF_X:
859 		case BPF_JMP32 | BPF_JGT | BPF_X:
860 		case BPF_JMP32 | BPF_JLT | BPF_X:
861 		case BPF_JMP32 | BPF_JGE | BPF_X:
862 		case BPF_JMP32 | BPF_JLE | BPF_X:
863 		case BPF_JMP32 | BPF_JSGT | BPF_X:
864 		case BPF_JMP32 | BPF_JSLT | BPF_X:
865 		case BPF_JMP32 | BPF_JSGE | BPF_X:
866 		case BPF_JMP32 | BPF_JSLE | BPF_X:
867 			/* cmp dst_reg, src_reg */
868 			if (BPF_CLASS(insn->code) == BPF_JMP)
869 				EMIT1(add_2mod(0x48, dst_reg, src_reg));
870 			else if (is_ereg(dst_reg) || is_ereg(src_reg))
871 				EMIT1(add_2mod(0x40, dst_reg, src_reg));
872 			EMIT2(0x39, add_2reg(0xC0, dst_reg, src_reg));
873 			goto emit_cond_jmp;
874 
875 		case BPF_JMP | BPF_JSET | BPF_X:
876 		case BPF_JMP32 | BPF_JSET | BPF_X:
877 			/* test dst_reg, src_reg */
878 			if (BPF_CLASS(insn->code) == BPF_JMP)
879 				EMIT1(add_2mod(0x48, dst_reg, src_reg));
880 			else if (is_ereg(dst_reg) || is_ereg(src_reg))
881 				EMIT1(add_2mod(0x40, dst_reg, src_reg));
882 			EMIT2(0x85, add_2reg(0xC0, dst_reg, src_reg));
883 			goto emit_cond_jmp;
884 
885 		case BPF_JMP | BPF_JSET | BPF_K:
886 		case BPF_JMP32 | BPF_JSET | BPF_K:
887 			/* test dst_reg, imm32 */
888 			if (BPF_CLASS(insn->code) == BPF_JMP)
889 				EMIT1(add_1mod(0x48, dst_reg));
890 			else if (is_ereg(dst_reg))
891 				EMIT1(add_1mod(0x40, dst_reg));
892 			EMIT2_off32(0xF7, add_1reg(0xC0, dst_reg), imm32);
893 			goto emit_cond_jmp;
894 
895 		case BPF_JMP | BPF_JEQ | BPF_K:
896 		case BPF_JMP | BPF_JNE | BPF_K:
897 		case BPF_JMP | BPF_JGT | BPF_K:
898 		case BPF_JMP | BPF_JLT | BPF_K:
899 		case BPF_JMP | BPF_JGE | BPF_K:
900 		case BPF_JMP | BPF_JLE | BPF_K:
901 		case BPF_JMP | BPF_JSGT | BPF_K:
902 		case BPF_JMP | BPF_JSLT | BPF_K:
903 		case BPF_JMP | BPF_JSGE | BPF_K:
904 		case BPF_JMP | BPF_JSLE | BPF_K:
905 		case BPF_JMP32 | BPF_JEQ | BPF_K:
906 		case BPF_JMP32 | BPF_JNE | BPF_K:
907 		case BPF_JMP32 | BPF_JGT | BPF_K:
908 		case BPF_JMP32 | BPF_JLT | BPF_K:
909 		case BPF_JMP32 | BPF_JGE | BPF_K:
910 		case BPF_JMP32 | BPF_JLE | BPF_K:
911 		case BPF_JMP32 | BPF_JSGT | BPF_K:
912 		case BPF_JMP32 | BPF_JSLT | BPF_K:
913 		case BPF_JMP32 | BPF_JSGE | BPF_K:
914 		case BPF_JMP32 | BPF_JSLE | BPF_K:
915 			/* cmp dst_reg, imm8/32 */
916 			if (BPF_CLASS(insn->code) == BPF_JMP)
917 				EMIT1(add_1mod(0x48, dst_reg));
918 			else if (is_ereg(dst_reg))
919 				EMIT1(add_1mod(0x40, dst_reg));
920 
921 			if (is_imm8(imm32))
922 				EMIT3(0x83, add_1reg(0xF8, dst_reg), imm32);
923 			else
924 				EMIT2_off32(0x81, add_1reg(0xF8, dst_reg), imm32);
925 
926 emit_cond_jmp:		/* Convert BPF opcode to x86 */
927 			switch (BPF_OP(insn->code)) {
928 			case BPF_JEQ:
929 				jmp_cond = X86_JE;
930 				break;
931 			case BPF_JSET:
932 			case BPF_JNE:
933 				jmp_cond = X86_JNE;
934 				break;
935 			case BPF_JGT:
936 				/* GT is unsigned '>', JA in x86 */
937 				jmp_cond = X86_JA;
938 				break;
939 			case BPF_JLT:
940 				/* LT is unsigned '<', JB in x86 */
941 				jmp_cond = X86_JB;
942 				break;
943 			case BPF_JGE:
944 				/* GE is unsigned '>=', JAE in x86 */
945 				jmp_cond = X86_JAE;
946 				break;
947 			case BPF_JLE:
948 				/* LE is unsigned '<=', JBE in x86 */
949 				jmp_cond = X86_JBE;
950 				break;
951 			case BPF_JSGT:
952 				/* Signed '>', GT in x86 */
953 				jmp_cond = X86_JG;
954 				break;
955 			case BPF_JSLT:
956 				/* Signed '<', LT in x86 */
957 				jmp_cond = X86_JL;
958 				break;
959 			case BPF_JSGE:
960 				/* Signed '>=', GE in x86 */
961 				jmp_cond = X86_JGE;
962 				break;
963 			case BPF_JSLE:
964 				/* Signed '<=', LE in x86 */
965 				jmp_cond = X86_JLE;
966 				break;
967 			default: /* to silence GCC warning */
968 				return -EFAULT;
969 			}
970 			jmp_offset = addrs[i + insn->off] - addrs[i];
971 			if (is_imm8(jmp_offset)) {
972 				EMIT2(jmp_cond, jmp_offset);
973 			} else if (is_simm32(jmp_offset)) {
974 				EMIT2_off32(0x0F, jmp_cond + 0x10, jmp_offset);
975 			} else {
976 				pr_err("cond_jmp gen bug %llx\n", jmp_offset);
977 				return -EFAULT;
978 			}
979 
980 			break;
981 
982 		case BPF_JMP | BPF_JA:
983 			if (insn->off == -1)
984 				/* -1 jmp instructions will always jump
985 				 * backwards two bytes. Explicitly handling
986 				 * this case avoids wasting too many passes
987 				 * when there are long sequences of replaced
988 				 * dead code.
989 				 */
990 				jmp_offset = -2;
991 			else
992 				jmp_offset = addrs[i + insn->off] - addrs[i];
993 
994 			if (!jmp_offset)
995 				/* Optimize out nop jumps */
996 				break;
997 emit_jmp:
998 			if (is_imm8(jmp_offset)) {
999 				EMIT2(0xEB, jmp_offset);
1000 			} else if (is_simm32(jmp_offset)) {
1001 				EMIT1_off32(0xE9, jmp_offset);
1002 			} else {
1003 				pr_err("jmp gen bug %llx\n", jmp_offset);
1004 				return -EFAULT;
1005 			}
1006 			break;
1007 
1008 		case BPF_JMP | BPF_EXIT:
1009 			if (seen_exit) {
1010 				jmp_offset = ctx->cleanup_addr - addrs[i];
1011 				goto emit_jmp;
1012 			}
1013 			seen_exit = true;
1014 			/* Update cleanup_addr */
1015 			ctx->cleanup_addr = proglen;
1016 			if (!bpf_prog_was_classic(bpf_prog))
1017 				EMIT1(0x5B); /* get rid of tail_call_cnt */
1018 			EMIT2(0x41, 0x5F);   /* pop r15 */
1019 			EMIT2(0x41, 0x5E);   /* pop r14 */
1020 			EMIT2(0x41, 0x5D);   /* pop r13 */
1021 			EMIT1(0x5B);         /* pop rbx */
1022 			EMIT1(0xC9);         /* leave */
1023 			EMIT1(0xC3);         /* ret */
1024 			break;
1025 
1026 		default:
1027 			/*
1028 			 * By design x86-64 JIT should support all BPF instructions.
1029 			 * This error will be seen if new instruction was added
1030 			 * to the interpreter, but not to the JIT, or if there is
1031 			 * junk in bpf_prog.
1032 			 */
1033 			pr_err("bpf_jit: unknown opcode %02x\n", insn->code);
1034 			return -EINVAL;
1035 		}
1036 
1037 		ilen = prog - temp;
1038 		if (ilen > BPF_MAX_INSN_SIZE) {
1039 			pr_err("bpf_jit: fatal insn size error\n");
1040 			return -EFAULT;
1041 		}
1042 
1043 		if (image) {
1044 			if (unlikely(proglen + ilen > oldproglen)) {
1045 				pr_err("bpf_jit: fatal error\n");
1046 				return -EFAULT;
1047 			}
1048 			memcpy(image + proglen, temp, ilen);
1049 		}
1050 		proglen += ilen;
1051 		addrs[i] = proglen;
1052 		prog = temp;
1053 	}
1054 	return proglen;
1055 }
1056 
1057 struct x64_jit_data {
1058 	struct bpf_binary_header *header;
1059 	int *addrs;
1060 	u8 *image;
1061 	int proglen;
1062 	struct jit_context ctx;
1063 };
1064 
1065 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
1066 {
1067 	struct bpf_binary_header *header = NULL;
1068 	struct bpf_prog *tmp, *orig_prog = prog;
1069 	struct x64_jit_data *jit_data;
1070 	int proglen, oldproglen = 0;
1071 	struct jit_context ctx = {};
1072 	bool tmp_blinded = false;
1073 	bool extra_pass = false;
1074 	u8 *image = NULL;
1075 	int *addrs;
1076 	int pass;
1077 	int i;
1078 
1079 	if (!prog->jit_requested)
1080 		return orig_prog;
1081 
1082 	tmp = bpf_jit_blind_constants(prog);
1083 	/*
1084 	 * If blinding was requested and we failed during blinding,
1085 	 * we must fall back to the interpreter.
1086 	 */
1087 	if (IS_ERR(tmp))
1088 		return orig_prog;
1089 	if (tmp != prog) {
1090 		tmp_blinded = true;
1091 		prog = tmp;
1092 	}
1093 
1094 	jit_data = prog->aux->jit_data;
1095 	if (!jit_data) {
1096 		jit_data = kzalloc(sizeof(*jit_data), GFP_KERNEL);
1097 		if (!jit_data) {
1098 			prog = orig_prog;
1099 			goto out;
1100 		}
1101 		prog->aux->jit_data = jit_data;
1102 	}
1103 	addrs = jit_data->addrs;
1104 	if (addrs) {
1105 		ctx = jit_data->ctx;
1106 		oldproglen = jit_data->proglen;
1107 		image = jit_data->image;
1108 		header = jit_data->header;
1109 		extra_pass = true;
1110 		goto skip_init_addrs;
1111 	}
1112 	addrs = kmalloc_array(prog->len, sizeof(*addrs), GFP_KERNEL);
1113 	if (!addrs) {
1114 		prog = orig_prog;
1115 		goto out_addrs;
1116 	}
1117 
1118 	/*
1119 	 * Before first pass, make a rough estimation of addrs[]
1120 	 * each BPF instruction is translated to less than 64 bytes
1121 	 */
1122 	for (proglen = 0, i = 0; i < prog->len; i++) {
1123 		proglen += 64;
1124 		addrs[i] = proglen;
1125 	}
1126 	ctx.cleanup_addr = proglen;
1127 skip_init_addrs:
1128 
1129 	/*
1130 	 * JITed image shrinks with every pass and the loop iterates
1131 	 * until the image stops shrinking. Very large BPF programs
1132 	 * may converge on the last pass. In such case do one more
1133 	 * pass to emit the final image.
1134 	 */
1135 	for (pass = 0; pass < 20 || image; pass++) {
1136 		proglen = do_jit(prog, addrs, image, oldproglen, &ctx);
1137 		if (proglen <= 0) {
1138 out_image:
1139 			image = NULL;
1140 			if (header)
1141 				bpf_jit_binary_free(header);
1142 			prog = orig_prog;
1143 			goto out_addrs;
1144 		}
1145 		if (image) {
1146 			if (proglen != oldproglen) {
1147 				pr_err("bpf_jit: proglen=%d != oldproglen=%d\n",
1148 				       proglen, oldproglen);
1149 				goto out_image;
1150 			}
1151 			break;
1152 		}
1153 		if (proglen == oldproglen) {
1154 			header = bpf_jit_binary_alloc(proglen, &image,
1155 						      1, jit_fill_hole);
1156 			if (!header) {
1157 				prog = orig_prog;
1158 				goto out_addrs;
1159 			}
1160 		}
1161 		oldproglen = proglen;
1162 		cond_resched();
1163 	}
1164 
1165 	if (bpf_jit_enable > 1)
1166 		bpf_jit_dump(prog->len, proglen, pass + 1, image);
1167 
1168 	if (image) {
1169 		if (!prog->is_func || extra_pass) {
1170 			bpf_jit_binary_lock_ro(header);
1171 		} else {
1172 			jit_data->addrs = addrs;
1173 			jit_data->ctx = ctx;
1174 			jit_data->proglen = proglen;
1175 			jit_data->image = image;
1176 			jit_data->header = header;
1177 		}
1178 		prog->bpf_func = (void *)image;
1179 		prog->jited = 1;
1180 		prog->jited_len = proglen;
1181 	} else {
1182 		prog = orig_prog;
1183 	}
1184 
1185 	if (!image || !prog->is_func || extra_pass) {
1186 		if (image)
1187 			bpf_prog_fill_jited_linfo(prog, addrs);
1188 out_addrs:
1189 		kfree(addrs);
1190 		kfree(jit_data);
1191 		prog->aux->jit_data = NULL;
1192 	}
1193 out:
1194 	if (tmp_blinded)
1195 		bpf_jit_prog_release_other(prog, prog == orig_prog ?
1196 					   tmp : orig_prog);
1197 	return prog;
1198 }
1199