xref: /openbmc/linux/arch/x86/net/bpf_jit_comp.c (revision 3ce7547e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * BPF JIT compiler
4  *
5  * Copyright (C) 2011-2013 Eric Dumazet (eric.dumazet@gmail.com)
6  * Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
7  */
8 #include <linux/netdevice.h>
9 #include <linux/filter.h>
10 #include <linux/if_vlan.h>
11 #include <linux/bpf.h>
12 #include <linux/memory.h>
13 #include <linux/sort.h>
14 #include <asm/extable.h>
15 #include <asm/set_memory.h>
16 #include <asm/nospec-branch.h>
17 #include <asm/text-patching.h>
18 
19 static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
20 {
21 	if (len == 1)
22 		*ptr = bytes;
23 	else if (len == 2)
24 		*(u16 *)ptr = bytes;
25 	else {
26 		*(u32 *)ptr = bytes;
27 		barrier();
28 	}
29 	return ptr + len;
30 }
31 
32 #define EMIT(bytes, len) \
33 	do { prog = emit_code(prog, bytes, len); } while (0)
34 
35 #define EMIT1(b1)		EMIT(b1, 1)
36 #define EMIT2(b1, b2)		EMIT((b1) + ((b2) << 8), 2)
37 #define EMIT3(b1, b2, b3)	EMIT((b1) + ((b2) << 8) + ((b3) << 16), 3)
38 #define EMIT4(b1, b2, b3, b4)   EMIT((b1) + ((b2) << 8) + ((b3) << 16) + ((b4) << 24), 4)
39 
40 #define EMIT1_off32(b1, off) \
41 	do { EMIT1(b1); EMIT(off, 4); } while (0)
42 #define EMIT2_off32(b1, b2, off) \
43 	do { EMIT2(b1, b2); EMIT(off, 4); } while (0)
44 #define EMIT3_off32(b1, b2, b3, off) \
45 	do { EMIT3(b1, b2, b3); EMIT(off, 4); } while (0)
46 #define EMIT4_off32(b1, b2, b3, b4, off) \
47 	do { EMIT4(b1, b2, b3, b4); EMIT(off, 4); } while (0)
48 
49 #ifdef CONFIG_X86_KERNEL_IBT
50 #define EMIT_ENDBR()	EMIT(gen_endbr(), 4)
51 #else
52 #define EMIT_ENDBR()
53 #endif
54 
55 static bool is_imm8(int value)
56 {
57 	return value <= 127 && value >= -128;
58 }
59 
60 static bool is_simm32(s64 value)
61 {
62 	return value == (s64)(s32)value;
63 }
64 
65 static bool is_uimm32(u64 value)
66 {
67 	return value == (u64)(u32)value;
68 }
69 
70 /* mov dst, src */
71 #define EMIT_mov(DST, SRC)								 \
72 	do {										 \
73 		if (DST != SRC)								 \
74 			EMIT3(add_2mod(0x48, DST, SRC), 0x89, add_2reg(0xC0, DST, SRC)); \
75 	} while (0)
76 
77 static int bpf_size_to_x86_bytes(int bpf_size)
78 {
79 	if (bpf_size == BPF_W)
80 		return 4;
81 	else if (bpf_size == BPF_H)
82 		return 2;
83 	else if (bpf_size == BPF_B)
84 		return 1;
85 	else if (bpf_size == BPF_DW)
86 		return 4; /* imm32 */
87 	else
88 		return 0;
89 }
90 
91 /*
92  * List of x86 cond jumps opcodes (. + s8)
93  * Add 0x10 (and an extra 0x0f) to generate far jumps (. + s32)
94  */
95 #define X86_JB  0x72
96 #define X86_JAE 0x73
97 #define X86_JE  0x74
98 #define X86_JNE 0x75
99 #define X86_JBE 0x76
100 #define X86_JA  0x77
101 #define X86_JL  0x7C
102 #define X86_JGE 0x7D
103 #define X86_JLE 0x7E
104 #define X86_JG  0x7F
105 
106 /* Pick a register outside of BPF range for JIT internal work */
107 #define AUX_REG (MAX_BPF_JIT_REG + 1)
108 #define X86_REG_R9 (MAX_BPF_JIT_REG + 2)
109 
110 /*
111  * The following table maps BPF registers to x86-64 registers.
112  *
113  * x86-64 register R12 is unused, since if used as base address
114  * register in load/store instructions, it always needs an
115  * extra byte of encoding and is callee saved.
116  *
117  * x86-64 register R9 is not used by BPF programs, but can be used by BPF
118  * trampoline. x86-64 register R10 is used for blinding (if enabled).
119  */
120 static const int reg2hex[] = {
121 	[BPF_REG_0] = 0,  /* RAX */
122 	[BPF_REG_1] = 7,  /* RDI */
123 	[BPF_REG_2] = 6,  /* RSI */
124 	[BPF_REG_3] = 2,  /* RDX */
125 	[BPF_REG_4] = 1,  /* RCX */
126 	[BPF_REG_5] = 0,  /* R8  */
127 	[BPF_REG_6] = 3,  /* RBX callee saved */
128 	[BPF_REG_7] = 5,  /* R13 callee saved */
129 	[BPF_REG_8] = 6,  /* R14 callee saved */
130 	[BPF_REG_9] = 7,  /* R15 callee saved */
131 	[BPF_REG_FP] = 5, /* RBP readonly */
132 	[BPF_REG_AX] = 2, /* R10 temp register */
133 	[AUX_REG] = 3,    /* R11 temp register */
134 	[X86_REG_R9] = 1, /* R9 register, 6th function argument */
135 };
136 
137 static const int reg2pt_regs[] = {
138 	[BPF_REG_0] = offsetof(struct pt_regs, ax),
139 	[BPF_REG_1] = offsetof(struct pt_regs, di),
140 	[BPF_REG_2] = offsetof(struct pt_regs, si),
141 	[BPF_REG_3] = offsetof(struct pt_regs, dx),
142 	[BPF_REG_4] = offsetof(struct pt_regs, cx),
143 	[BPF_REG_5] = offsetof(struct pt_regs, r8),
144 	[BPF_REG_6] = offsetof(struct pt_regs, bx),
145 	[BPF_REG_7] = offsetof(struct pt_regs, r13),
146 	[BPF_REG_8] = offsetof(struct pt_regs, r14),
147 	[BPF_REG_9] = offsetof(struct pt_regs, r15),
148 };
149 
150 /*
151  * is_ereg() == true if BPF register 'reg' maps to x86-64 r8..r15
152  * which need extra byte of encoding.
153  * rax,rcx,...,rbp have simpler encoding
154  */
155 static bool is_ereg(u32 reg)
156 {
157 	return (1 << reg) & (BIT(BPF_REG_5) |
158 			     BIT(AUX_REG) |
159 			     BIT(BPF_REG_7) |
160 			     BIT(BPF_REG_8) |
161 			     BIT(BPF_REG_9) |
162 			     BIT(X86_REG_R9) |
163 			     BIT(BPF_REG_AX));
164 }
165 
166 /*
167  * is_ereg_8l() == true if BPF register 'reg' is mapped to access x86-64
168  * lower 8-bit registers dil,sil,bpl,spl,r8b..r15b, which need extra byte
169  * of encoding. al,cl,dl,bl have simpler encoding.
170  */
171 static bool is_ereg_8l(u32 reg)
172 {
173 	return is_ereg(reg) ||
174 	    (1 << reg) & (BIT(BPF_REG_1) |
175 			  BIT(BPF_REG_2) |
176 			  BIT(BPF_REG_FP));
177 }
178 
179 static bool is_axreg(u32 reg)
180 {
181 	return reg == BPF_REG_0;
182 }
183 
184 /* Add modifiers if 'reg' maps to x86-64 registers R8..R15 */
185 static u8 add_1mod(u8 byte, u32 reg)
186 {
187 	if (is_ereg(reg))
188 		byte |= 1;
189 	return byte;
190 }
191 
192 static u8 add_2mod(u8 byte, u32 r1, u32 r2)
193 {
194 	if (is_ereg(r1))
195 		byte |= 1;
196 	if (is_ereg(r2))
197 		byte |= 4;
198 	return byte;
199 }
200 
201 /* Encode 'dst_reg' register into x86-64 opcode 'byte' */
202 static u8 add_1reg(u8 byte, u32 dst_reg)
203 {
204 	return byte + reg2hex[dst_reg];
205 }
206 
207 /* Encode 'dst_reg' and 'src_reg' registers into x86-64 opcode 'byte' */
208 static u8 add_2reg(u8 byte, u32 dst_reg, u32 src_reg)
209 {
210 	return byte + reg2hex[dst_reg] + (reg2hex[src_reg] << 3);
211 }
212 
213 /* Some 1-byte opcodes for binary ALU operations */
214 static u8 simple_alu_opcodes[] = {
215 	[BPF_ADD] = 0x01,
216 	[BPF_SUB] = 0x29,
217 	[BPF_AND] = 0x21,
218 	[BPF_OR] = 0x09,
219 	[BPF_XOR] = 0x31,
220 	[BPF_LSH] = 0xE0,
221 	[BPF_RSH] = 0xE8,
222 	[BPF_ARSH] = 0xF8,
223 };
224 
225 static void jit_fill_hole(void *area, unsigned int size)
226 {
227 	/* Fill whole space with INT3 instructions */
228 	memset(area, 0xcc, size);
229 }
230 
231 int bpf_arch_text_invalidate(void *dst, size_t len)
232 {
233 	return IS_ERR_OR_NULL(text_poke_set(dst, 0xcc, len));
234 }
235 
236 struct jit_context {
237 	int cleanup_addr; /* Epilogue code offset */
238 
239 	/*
240 	 * Program specific offsets of labels in the code; these rely on the
241 	 * JIT doing at least 2 passes, recording the position on the first
242 	 * pass, only to generate the correct offset on the second pass.
243 	 */
244 	int tail_call_direct_label;
245 	int tail_call_indirect_label;
246 };
247 
248 /* Maximum number of bytes emitted while JITing one eBPF insn */
249 #define BPF_MAX_INSN_SIZE	128
250 #define BPF_INSN_SAFETY		64
251 
252 /* Number of bytes emit_patch() needs to generate instructions */
253 #define X86_PATCH_SIZE		5
254 /* Number of bytes that will be skipped on tailcall */
255 #define X86_TAIL_CALL_OFFSET	(11 + ENDBR_INSN_SIZE)
256 
257 static void push_callee_regs(u8 **pprog, bool *callee_regs_used)
258 {
259 	u8 *prog = *pprog;
260 
261 	if (callee_regs_used[0])
262 		EMIT1(0x53);         /* push rbx */
263 	if (callee_regs_used[1])
264 		EMIT2(0x41, 0x55);   /* push r13 */
265 	if (callee_regs_used[2])
266 		EMIT2(0x41, 0x56);   /* push r14 */
267 	if (callee_regs_used[3])
268 		EMIT2(0x41, 0x57);   /* push r15 */
269 	*pprog = prog;
270 }
271 
272 static void pop_callee_regs(u8 **pprog, bool *callee_regs_used)
273 {
274 	u8 *prog = *pprog;
275 
276 	if (callee_regs_used[3])
277 		EMIT2(0x41, 0x5F);   /* pop r15 */
278 	if (callee_regs_used[2])
279 		EMIT2(0x41, 0x5E);   /* pop r14 */
280 	if (callee_regs_used[1])
281 		EMIT2(0x41, 0x5D);   /* pop r13 */
282 	if (callee_regs_used[0])
283 		EMIT1(0x5B);         /* pop rbx */
284 	*pprog = prog;
285 }
286 
287 /*
288  * Emit x86-64 prologue code for BPF program.
289  * bpf_tail_call helper will skip the first X86_TAIL_CALL_OFFSET bytes
290  * while jumping to another program
291  */
292 static void emit_prologue(u8 **pprog, u32 stack_depth, bool ebpf_from_cbpf,
293 			  bool tail_call_reachable, bool is_subprog)
294 {
295 	u8 *prog = *pprog;
296 
297 	/* BPF trampoline can be made to work without these nops,
298 	 * but let's waste 5 bytes for now and optimize later
299 	 */
300 	EMIT_ENDBR();
301 	memcpy(prog, x86_nops[5], X86_PATCH_SIZE);
302 	prog += X86_PATCH_SIZE;
303 	if (!ebpf_from_cbpf) {
304 		if (tail_call_reachable && !is_subprog)
305 			EMIT2(0x31, 0xC0); /* xor eax, eax */
306 		else
307 			EMIT2(0x66, 0x90); /* nop2 */
308 	}
309 	EMIT1(0x55);             /* push rbp */
310 	EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */
311 
312 	/* X86_TAIL_CALL_OFFSET is here */
313 	EMIT_ENDBR();
314 
315 	/* sub rsp, rounded_stack_depth */
316 	if (stack_depth)
317 		EMIT3_off32(0x48, 0x81, 0xEC, round_up(stack_depth, 8));
318 	if (tail_call_reachable)
319 		EMIT1(0x50);         /* push rax */
320 	*pprog = prog;
321 }
322 
323 static int emit_patch(u8 **pprog, void *func, void *ip, u8 opcode)
324 {
325 	u8 *prog = *pprog;
326 	s64 offset;
327 
328 	offset = func - (ip + X86_PATCH_SIZE);
329 	if (!is_simm32(offset)) {
330 		pr_err("Target call %p is out of range\n", func);
331 		return -ERANGE;
332 	}
333 	EMIT1_off32(opcode, offset);
334 	*pprog = prog;
335 	return 0;
336 }
337 
338 static int emit_call(u8 **pprog, void *func, void *ip)
339 {
340 	return emit_patch(pprog, func, ip, 0xE8);
341 }
342 
343 static int emit_jump(u8 **pprog, void *func, void *ip)
344 {
345 	return emit_patch(pprog, func, ip, 0xE9);
346 }
347 
348 static int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
349 				void *old_addr, void *new_addr)
350 {
351 	const u8 *nop_insn = x86_nops[5];
352 	u8 old_insn[X86_PATCH_SIZE];
353 	u8 new_insn[X86_PATCH_SIZE];
354 	u8 *prog;
355 	int ret;
356 
357 	memcpy(old_insn, nop_insn, X86_PATCH_SIZE);
358 	if (old_addr) {
359 		prog = old_insn;
360 		ret = t == BPF_MOD_CALL ?
361 		      emit_call(&prog, old_addr, ip) :
362 		      emit_jump(&prog, old_addr, ip);
363 		if (ret)
364 			return ret;
365 	}
366 
367 	memcpy(new_insn, nop_insn, X86_PATCH_SIZE);
368 	if (new_addr) {
369 		prog = new_insn;
370 		ret = t == BPF_MOD_CALL ?
371 		      emit_call(&prog, new_addr, ip) :
372 		      emit_jump(&prog, new_addr, ip);
373 		if (ret)
374 			return ret;
375 	}
376 
377 	ret = -EBUSY;
378 	mutex_lock(&text_mutex);
379 	if (memcmp(ip, old_insn, X86_PATCH_SIZE))
380 		goto out;
381 	ret = 1;
382 	if (memcmp(ip, new_insn, X86_PATCH_SIZE)) {
383 		text_poke_bp(ip, new_insn, X86_PATCH_SIZE, NULL);
384 		ret = 0;
385 	}
386 out:
387 	mutex_unlock(&text_mutex);
388 	return ret;
389 }
390 
391 int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t,
392 		       void *old_addr, void *new_addr)
393 {
394 	if (!is_kernel_text((long)ip) &&
395 	    !is_bpf_text_address((long)ip))
396 		/* BPF poking in modules is not supported */
397 		return -EINVAL;
398 
399 	/*
400 	 * See emit_prologue(), for IBT builds the trampoline hook is preceded
401 	 * with an ENDBR instruction.
402 	 */
403 	if (is_endbr(*(u32 *)ip))
404 		ip += ENDBR_INSN_SIZE;
405 
406 	return __bpf_arch_text_poke(ip, t, old_addr, new_addr);
407 }
408 
409 #define EMIT_LFENCE()	EMIT3(0x0F, 0xAE, 0xE8)
410 
411 static void emit_indirect_jump(u8 **pprog, int reg, u8 *ip)
412 {
413 	u8 *prog = *pprog;
414 
415 #ifdef CONFIG_RETPOLINE
416 	if (cpu_feature_enabled(X86_FEATURE_RETPOLINE_LFENCE)) {
417 		EMIT_LFENCE();
418 		EMIT2(0xFF, 0xE0 + reg);
419 	} else if (cpu_feature_enabled(X86_FEATURE_RETPOLINE)) {
420 		OPTIMIZER_HIDE_VAR(reg);
421 		emit_jump(&prog, &__x86_indirect_thunk_array[reg], ip);
422 	} else
423 #endif
424 	EMIT2(0xFF, 0xE0 + reg);
425 
426 	*pprog = prog;
427 }
428 
429 /*
430  * Generate the following code:
431  *
432  * ... bpf_tail_call(void *ctx, struct bpf_array *array, u64 index) ...
433  *   if (index >= array->map.max_entries)
434  *     goto out;
435  *   if (tail_call_cnt++ >= MAX_TAIL_CALL_CNT)
436  *     goto out;
437  *   prog = array->ptrs[index];
438  *   if (prog == NULL)
439  *     goto out;
440  *   goto *(prog->bpf_func + prologue_size);
441  * out:
442  */
443 static void emit_bpf_tail_call_indirect(u8 **pprog, bool *callee_regs_used,
444 					u32 stack_depth, u8 *ip,
445 					struct jit_context *ctx)
446 {
447 	int tcc_off = -4 - round_up(stack_depth, 8);
448 	u8 *prog = *pprog, *start = *pprog;
449 	int offset;
450 
451 	/*
452 	 * rdi - pointer to ctx
453 	 * rsi - pointer to bpf_array
454 	 * rdx - index in bpf_array
455 	 */
456 
457 	/*
458 	 * if (index >= array->map.max_entries)
459 	 *	goto out;
460 	 */
461 	EMIT2(0x89, 0xD2);                        /* mov edx, edx */
462 	EMIT3(0x39, 0x56,                         /* cmp dword ptr [rsi + 16], edx */
463 	      offsetof(struct bpf_array, map.max_entries));
464 
465 	offset = ctx->tail_call_indirect_label - (prog + 2 - start);
466 	EMIT2(X86_JBE, offset);                   /* jbe out */
467 
468 	/*
469 	 * if (tail_call_cnt++ >= MAX_TAIL_CALL_CNT)
470 	 *	goto out;
471 	 */
472 	EMIT2_off32(0x8B, 0x85, tcc_off);         /* mov eax, dword ptr [rbp - tcc_off] */
473 	EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT);     /* cmp eax, MAX_TAIL_CALL_CNT */
474 
475 	offset = ctx->tail_call_indirect_label - (prog + 2 - start);
476 	EMIT2(X86_JAE, offset);                   /* jae out */
477 	EMIT3(0x83, 0xC0, 0x01);                  /* add eax, 1 */
478 	EMIT2_off32(0x89, 0x85, tcc_off);         /* mov dword ptr [rbp - tcc_off], eax */
479 
480 	/* prog = array->ptrs[index]; */
481 	EMIT4_off32(0x48, 0x8B, 0x8C, 0xD6,       /* mov rcx, [rsi + rdx * 8 + offsetof(...)] */
482 		    offsetof(struct bpf_array, ptrs));
483 
484 	/*
485 	 * if (prog == NULL)
486 	 *	goto out;
487 	 */
488 	EMIT3(0x48, 0x85, 0xC9);                  /* test rcx,rcx */
489 
490 	offset = ctx->tail_call_indirect_label - (prog + 2 - start);
491 	EMIT2(X86_JE, offset);                    /* je out */
492 
493 	pop_callee_regs(&prog, callee_regs_used);
494 
495 	EMIT1(0x58);                              /* pop rax */
496 	if (stack_depth)
497 		EMIT3_off32(0x48, 0x81, 0xC4,     /* add rsp, sd */
498 			    round_up(stack_depth, 8));
499 
500 	/* goto *(prog->bpf_func + X86_TAIL_CALL_OFFSET); */
501 	EMIT4(0x48, 0x8B, 0x49,                   /* mov rcx, qword ptr [rcx + 32] */
502 	      offsetof(struct bpf_prog, bpf_func));
503 	EMIT4(0x48, 0x83, 0xC1,                   /* add rcx, X86_TAIL_CALL_OFFSET */
504 	      X86_TAIL_CALL_OFFSET);
505 	/*
506 	 * Now we're ready to jump into next BPF program
507 	 * rdi == ctx (1st arg)
508 	 * rcx == prog->bpf_func + X86_TAIL_CALL_OFFSET
509 	 */
510 	emit_indirect_jump(&prog, 1 /* rcx */, ip + (prog - start));
511 
512 	/* out: */
513 	ctx->tail_call_indirect_label = prog - start;
514 	*pprog = prog;
515 }
516 
517 static void emit_bpf_tail_call_direct(struct bpf_jit_poke_descriptor *poke,
518 				      u8 **pprog, u8 *ip,
519 				      bool *callee_regs_used, u32 stack_depth,
520 				      struct jit_context *ctx)
521 {
522 	int tcc_off = -4 - round_up(stack_depth, 8);
523 	u8 *prog = *pprog, *start = *pprog;
524 	int offset;
525 
526 	/*
527 	 * if (tail_call_cnt++ >= MAX_TAIL_CALL_CNT)
528 	 *	goto out;
529 	 */
530 	EMIT2_off32(0x8B, 0x85, tcc_off);             /* mov eax, dword ptr [rbp - tcc_off] */
531 	EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT);         /* cmp eax, MAX_TAIL_CALL_CNT */
532 
533 	offset = ctx->tail_call_direct_label - (prog + 2 - start);
534 	EMIT2(X86_JAE, offset);                       /* jae out */
535 	EMIT3(0x83, 0xC0, 0x01);                      /* add eax, 1 */
536 	EMIT2_off32(0x89, 0x85, tcc_off);             /* mov dword ptr [rbp - tcc_off], eax */
537 
538 	poke->tailcall_bypass = ip + (prog - start);
539 	poke->adj_off = X86_TAIL_CALL_OFFSET;
540 	poke->tailcall_target = ip + ctx->tail_call_direct_label - X86_PATCH_SIZE;
541 	poke->bypass_addr = (u8 *)poke->tailcall_target + X86_PATCH_SIZE;
542 
543 	emit_jump(&prog, (u8 *)poke->tailcall_target + X86_PATCH_SIZE,
544 		  poke->tailcall_bypass);
545 
546 	pop_callee_regs(&prog, callee_regs_used);
547 	EMIT1(0x58);                                  /* pop rax */
548 	if (stack_depth)
549 		EMIT3_off32(0x48, 0x81, 0xC4, round_up(stack_depth, 8));
550 
551 	memcpy(prog, x86_nops[5], X86_PATCH_SIZE);
552 	prog += X86_PATCH_SIZE;
553 
554 	/* out: */
555 	ctx->tail_call_direct_label = prog - start;
556 
557 	*pprog = prog;
558 }
559 
560 static void bpf_tail_call_direct_fixup(struct bpf_prog *prog)
561 {
562 	struct bpf_jit_poke_descriptor *poke;
563 	struct bpf_array *array;
564 	struct bpf_prog *target;
565 	int i, ret;
566 
567 	for (i = 0; i < prog->aux->size_poke_tab; i++) {
568 		poke = &prog->aux->poke_tab[i];
569 		if (poke->aux && poke->aux != prog->aux)
570 			continue;
571 
572 		WARN_ON_ONCE(READ_ONCE(poke->tailcall_target_stable));
573 
574 		if (poke->reason != BPF_POKE_REASON_TAIL_CALL)
575 			continue;
576 
577 		array = container_of(poke->tail_call.map, struct bpf_array, map);
578 		mutex_lock(&array->aux->poke_mutex);
579 		target = array->ptrs[poke->tail_call.key];
580 		if (target) {
581 			ret = __bpf_arch_text_poke(poke->tailcall_target,
582 						   BPF_MOD_JUMP, NULL,
583 						   (u8 *)target->bpf_func +
584 						   poke->adj_off);
585 			BUG_ON(ret < 0);
586 			ret = __bpf_arch_text_poke(poke->tailcall_bypass,
587 						   BPF_MOD_JUMP,
588 						   (u8 *)poke->tailcall_target +
589 						   X86_PATCH_SIZE, NULL);
590 			BUG_ON(ret < 0);
591 		}
592 		WRITE_ONCE(poke->tailcall_target_stable, true);
593 		mutex_unlock(&array->aux->poke_mutex);
594 	}
595 }
596 
597 static void emit_mov_imm32(u8 **pprog, bool sign_propagate,
598 			   u32 dst_reg, const u32 imm32)
599 {
600 	u8 *prog = *pprog;
601 	u8 b1, b2, b3;
602 
603 	/*
604 	 * Optimization: if imm32 is positive, use 'mov %eax, imm32'
605 	 * (which zero-extends imm32) to save 2 bytes.
606 	 */
607 	if (sign_propagate && (s32)imm32 < 0) {
608 		/* 'mov %rax, imm32' sign extends imm32 */
609 		b1 = add_1mod(0x48, dst_reg);
610 		b2 = 0xC7;
611 		b3 = 0xC0;
612 		EMIT3_off32(b1, b2, add_1reg(b3, dst_reg), imm32);
613 		goto done;
614 	}
615 
616 	/*
617 	 * Optimization: if imm32 is zero, use 'xor %eax, %eax'
618 	 * to save 3 bytes.
619 	 */
620 	if (imm32 == 0) {
621 		if (is_ereg(dst_reg))
622 			EMIT1(add_2mod(0x40, dst_reg, dst_reg));
623 		b2 = 0x31; /* xor */
624 		b3 = 0xC0;
625 		EMIT2(b2, add_2reg(b3, dst_reg, dst_reg));
626 		goto done;
627 	}
628 
629 	/* mov %eax, imm32 */
630 	if (is_ereg(dst_reg))
631 		EMIT1(add_1mod(0x40, dst_reg));
632 	EMIT1_off32(add_1reg(0xB8, dst_reg), imm32);
633 done:
634 	*pprog = prog;
635 }
636 
637 static void emit_mov_imm64(u8 **pprog, u32 dst_reg,
638 			   const u32 imm32_hi, const u32 imm32_lo)
639 {
640 	u8 *prog = *pprog;
641 
642 	if (is_uimm32(((u64)imm32_hi << 32) | (u32)imm32_lo)) {
643 		/*
644 		 * For emitting plain u32, where sign bit must not be
645 		 * propagated LLVM tends to load imm64 over mov32
646 		 * directly, so save couple of bytes by just doing
647 		 * 'mov %eax, imm32' instead.
648 		 */
649 		emit_mov_imm32(&prog, false, dst_reg, imm32_lo);
650 	} else {
651 		/* movabsq %rax, imm64 */
652 		EMIT2(add_1mod(0x48, dst_reg), add_1reg(0xB8, dst_reg));
653 		EMIT(imm32_lo, 4);
654 		EMIT(imm32_hi, 4);
655 	}
656 
657 	*pprog = prog;
658 }
659 
660 static void emit_mov_reg(u8 **pprog, bool is64, u32 dst_reg, u32 src_reg)
661 {
662 	u8 *prog = *pprog;
663 
664 	if (is64) {
665 		/* mov dst, src */
666 		EMIT_mov(dst_reg, src_reg);
667 	} else {
668 		/* mov32 dst, src */
669 		if (is_ereg(dst_reg) || is_ereg(src_reg))
670 			EMIT1(add_2mod(0x40, dst_reg, src_reg));
671 		EMIT2(0x89, add_2reg(0xC0, dst_reg, src_reg));
672 	}
673 
674 	*pprog = prog;
675 }
676 
677 /* Emit the suffix (ModR/M etc) for addressing *(ptr_reg + off) and val_reg */
678 static void emit_insn_suffix(u8 **pprog, u32 ptr_reg, u32 val_reg, int off)
679 {
680 	u8 *prog = *pprog;
681 
682 	if (is_imm8(off)) {
683 		/* 1-byte signed displacement.
684 		 *
685 		 * If off == 0 we could skip this and save one extra byte, but
686 		 * special case of x86 R13 which always needs an offset is not
687 		 * worth the hassle
688 		 */
689 		EMIT2(add_2reg(0x40, ptr_reg, val_reg), off);
690 	} else {
691 		/* 4-byte signed displacement */
692 		EMIT1_off32(add_2reg(0x80, ptr_reg, val_reg), off);
693 	}
694 	*pprog = prog;
695 }
696 
697 /*
698  * Emit a REX byte if it will be necessary to address these registers
699  */
700 static void maybe_emit_mod(u8 **pprog, u32 dst_reg, u32 src_reg, bool is64)
701 {
702 	u8 *prog = *pprog;
703 
704 	if (is64)
705 		EMIT1(add_2mod(0x48, dst_reg, src_reg));
706 	else if (is_ereg(dst_reg) || is_ereg(src_reg))
707 		EMIT1(add_2mod(0x40, dst_reg, src_reg));
708 	*pprog = prog;
709 }
710 
711 /*
712  * Similar version of maybe_emit_mod() for a single register
713  */
714 static void maybe_emit_1mod(u8 **pprog, u32 reg, bool is64)
715 {
716 	u8 *prog = *pprog;
717 
718 	if (is64)
719 		EMIT1(add_1mod(0x48, reg));
720 	else if (is_ereg(reg))
721 		EMIT1(add_1mod(0x40, reg));
722 	*pprog = prog;
723 }
724 
725 /* LDX: dst_reg = *(u8*)(src_reg + off) */
726 static void emit_ldx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off)
727 {
728 	u8 *prog = *pprog;
729 
730 	switch (size) {
731 	case BPF_B:
732 		/* Emit 'movzx rax, byte ptr [rax + off]' */
733 		EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB6);
734 		break;
735 	case BPF_H:
736 		/* Emit 'movzx rax, word ptr [rax + off]' */
737 		EMIT3(add_2mod(0x48, src_reg, dst_reg), 0x0F, 0xB7);
738 		break;
739 	case BPF_W:
740 		/* Emit 'mov eax, dword ptr [rax+0x14]' */
741 		if (is_ereg(dst_reg) || is_ereg(src_reg))
742 			EMIT2(add_2mod(0x40, src_reg, dst_reg), 0x8B);
743 		else
744 			EMIT1(0x8B);
745 		break;
746 	case BPF_DW:
747 		/* Emit 'mov rax, qword ptr [rax+0x14]' */
748 		EMIT2(add_2mod(0x48, src_reg, dst_reg), 0x8B);
749 		break;
750 	}
751 	emit_insn_suffix(&prog, src_reg, dst_reg, off);
752 	*pprog = prog;
753 }
754 
755 /* STX: *(u8*)(dst_reg + off) = src_reg */
756 static void emit_stx(u8 **pprog, u32 size, u32 dst_reg, u32 src_reg, int off)
757 {
758 	u8 *prog = *pprog;
759 
760 	switch (size) {
761 	case BPF_B:
762 		/* Emit 'mov byte ptr [rax + off], al' */
763 		if (is_ereg(dst_reg) || is_ereg_8l(src_reg))
764 			/* Add extra byte for eregs or SIL,DIL,BPL in src_reg */
765 			EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x88);
766 		else
767 			EMIT1(0x88);
768 		break;
769 	case BPF_H:
770 		if (is_ereg(dst_reg) || is_ereg(src_reg))
771 			EMIT3(0x66, add_2mod(0x40, dst_reg, src_reg), 0x89);
772 		else
773 			EMIT2(0x66, 0x89);
774 		break;
775 	case BPF_W:
776 		if (is_ereg(dst_reg) || is_ereg(src_reg))
777 			EMIT2(add_2mod(0x40, dst_reg, src_reg), 0x89);
778 		else
779 			EMIT1(0x89);
780 		break;
781 	case BPF_DW:
782 		EMIT2(add_2mod(0x48, dst_reg, src_reg), 0x89);
783 		break;
784 	}
785 	emit_insn_suffix(&prog, dst_reg, src_reg, off);
786 	*pprog = prog;
787 }
788 
789 static int emit_atomic(u8 **pprog, u8 atomic_op,
790 		       u32 dst_reg, u32 src_reg, s16 off, u8 bpf_size)
791 {
792 	u8 *prog = *pprog;
793 
794 	EMIT1(0xF0); /* lock prefix */
795 
796 	maybe_emit_mod(&prog, dst_reg, src_reg, bpf_size == BPF_DW);
797 
798 	/* emit opcode */
799 	switch (atomic_op) {
800 	case BPF_ADD:
801 	case BPF_AND:
802 	case BPF_OR:
803 	case BPF_XOR:
804 		/* lock *(u32/u64*)(dst_reg + off) <op>= src_reg */
805 		EMIT1(simple_alu_opcodes[atomic_op]);
806 		break;
807 	case BPF_ADD | BPF_FETCH:
808 		/* src_reg = atomic_fetch_add(dst_reg + off, src_reg); */
809 		EMIT2(0x0F, 0xC1);
810 		break;
811 	case BPF_XCHG:
812 		/* src_reg = atomic_xchg(dst_reg + off, src_reg); */
813 		EMIT1(0x87);
814 		break;
815 	case BPF_CMPXCHG:
816 		/* r0 = atomic_cmpxchg(dst_reg + off, r0, src_reg); */
817 		EMIT2(0x0F, 0xB1);
818 		break;
819 	default:
820 		pr_err("bpf_jit: unknown atomic opcode %02x\n", atomic_op);
821 		return -EFAULT;
822 	}
823 
824 	emit_insn_suffix(&prog, dst_reg, src_reg, off);
825 
826 	*pprog = prog;
827 	return 0;
828 }
829 
830 bool ex_handler_bpf(const struct exception_table_entry *x, struct pt_regs *regs)
831 {
832 	u32 reg = x->fixup >> 8;
833 
834 	/* jump over faulting load and clear dest register */
835 	*(unsigned long *)((void *)regs + reg) = 0;
836 	regs->ip += x->fixup & 0xff;
837 	return true;
838 }
839 
840 static void detect_reg_usage(struct bpf_insn *insn, int insn_cnt,
841 			     bool *regs_used, bool *tail_call_seen)
842 {
843 	int i;
844 
845 	for (i = 1; i <= insn_cnt; i++, insn++) {
846 		if (insn->code == (BPF_JMP | BPF_TAIL_CALL))
847 			*tail_call_seen = true;
848 		if (insn->dst_reg == BPF_REG_6 || insn->src_reg == BPF_REG_6)
849 			regs_used[0] = true;
850 		if (insn->dst_reg == BPF_REG_7 || insn->src_reg == BPF_REG_7)
851 			regs_used[1] = true;
852 		if (insn->dst_reg == BPF_REG_8 || insn->src_reg == BPF_REG_8)
853 			regs_used[2] = true;
854 		if (insn->dst_reg == BPF_REG_9 || insn->src_reg == BPF_REG_9)
855 			regs_used[3] = true;
856 	}
857 }
858 
859 static void emit_nops(u8 **pprog, int len)
860 {
861 	u8 *prog = *pprog;
862 	int i, noplen;
863 
864 	while (len > 0) {
865 		noplen = len;
866 
867 		if (noplen > ASM_NOP_MAX)
868 			noplen = ASM_NOP_MAX;
869 
870 		for (i = 0; i < noplen; i++)
871 			EMIT1(x86_nops[noplen][i]);
872 		len -= noplen;
873 	}
874 
875 	*pprog = prog;
876 }
877 
878 #define INSN_SZ_DIFF (((addrs[i] - addrs[i - 1]) - (prog - temp)))
879 
880 static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image, u8 *rw_image,
881 		  int oldproglen, struct jit_context *ctx, bool jmp_padding)
882 {
883 	bool tail_call_reachable = bpf_prog->aux->tail_call_reachable;
884 	struct bpf_insn *insn = bpf_prog->insnsi;
885 	bool callee_regs_used[4] = {};
886 	int insn_cnt = bpf_prog->len;
887 	bool tail_call_seen = false;
888 	bool seen_exit = false;
889 	u8 temp[BPF_MAX_INSN_SIZE + BPF_INSN_SAFETY];
890 	int i, excnt = 0;
891 	int ilen, proglen = 0;
892 	u8 *prog = temp;
893 	int err;
894 
895 	detect_reg_usage(insn, insn_cnt, callee_regs_used,
896 			 &tail_call_seen);
897 
898 	/* tail call's presence in current prog implies it is reachable */
899 	tail_call_reachable |= tail_call_seen;
900 
901 	emit_prologue(&prog, bpf_prog->aux->stack_depth,
902 		      bpf_prog_was_classic(bpf_prog), tail_call_reachable,
903 		      bpf_prog->aux->func_idx != 0);
904 	push_callee_regs(&prog, callee_regs_used);
905 
906 	ilen = prog - temp;
907 	if (rw_image)
908 		memcpy(rw_image + proglen, temp, ilen);
909 	proglen += ilen;
910 	addrs[0] = proglen;
911 	prog = temp;
912 
913 	for (i = 1; i <= insn_cnt; i++, insn++) {
914 		const s32 imm32 = insn->imm;
915 		u32 dst_reg = insn->dst_reg;
916 		u32 src_reg = insn->src_reg;
917 		u8 b2 = 0, b3 = 0;
918 		u8 *start_of_ldx;
919 		s64 jmp_offset;
920 		u8 jmp_cond;
921 		u8 *func;
922 		int nops;
923 
924 		switch (insn->code) {
925 			/* ALU */
926 		case BPF_ALU | BPF_ADD | BPF_X:
927 		case BPF_ALU | BPF_SUB | BPF_X:
928 		case BPF_ALU | BPF_AND | BPF_X:
929 		case BPF_ALU | BPF_OR | BPF_X:
930 		case BPF_ALU | BPF_XOR | BPF_X:
931 		case BPF_ALU64 | BPF_ADD | BPF_X:
932 		case BPF_ALU64 | BPF_SUB | BPF_X:
933 		case BPF_ALU64 | BPF_AND | BPF_X:
934 		case BPF_ALU64 | BPF_OR | BPF_X:
935 		case BPF_ALU64 | BPF_XOR | BPF_X:
936 			maybe_emit_mod(&prog, dst_reg, src_reg,
937 				       BPF_CLASS(insn->code) == BPF_ALU64);
938 			b2 = simple_alu_opcodes[BPF_OP(insn->code)];
939 			EMIT2(b2, add_2reg(0xC0, dst_reg, src_reg));
940 			break;
941 
942 		case BPF_ALU64 | BPF_MOV | BPF_X:
943 		case BPF_ALU | BPF_MOV | BPF_X:
944 			emit_mov_reg(&prog,
945 				     BPF_CLASS(insn->code) == BPF_ALU64,
946 				     dst_reg, src_reg);
947 			break;
948 
949 			/* neg dst */
950 		case BPF_ALU | BPF_NEG:
951 		case BPF_ALU64 | BPF_NEG:
952 			maybe_emit_1mod(&prog, dst_reg,
953 					BPF_CLASS(insn->code) == BPF_ALU64);
954 			EMIT2(0xF7, add_1reg(0xD8, dst_reg));
955 			break;
956 
957 		case BPF_ALU | BPF_ADD | BPF_K:
958 		case BPF_ALU | BPF_SUB | BPF_K:
959 		case BPF_ALU | BPF_AND | BPF_K:
960 		case BPF_ALU | BPF_OR | BPF_K:
961 		case BPF_ALU | BPF_XOR | BPF_K:
962 		case BPF_ALU64 | BPF_ADD | BPF_K:
963 		case BPF_ALU64 | BPF_SUB | BPF_K:
964 		case BPF_ALU64 | BPF_AND | BPF_K:
965 		case BPF_ALU64 | BPF_OR | BPF_K:
966 		case BPF_ALU64 | BPF_XOR | BPF_K:
967 			maybe_emit_1mod(&prog, dst_reg,
968 					BPF_CLASS(insn->code) == BPF_ALU64);
969 
970 			/*
971 			 * b3 holds 'normal' opcode, b2 short form only valid
972 			 * in case dst is eax/rax.
973 			 */
974 			switch (BPF_OP(insn->code)) {
975 			case BPF_ADD:
976 				b3 = 0xC0;
977 				b2 = 0x05;
978 				break;
979 			case BPF_SUB:
980 				b3 = 0xE8;
981 				b2 = 0x2D;
982 				break;
983 			case BPF_AND:
984 				b3 = 0xE0;
985 				b2 = 0x25;
986 				break;
987 			case BPF_OR:
988 				b3 = 0xC8;
989 				b2 = 0x0D;
990 				break;
991 			case BPF_XOR:
992 				b3 = 0xF0;
993 				b2 = 0x35;
994 				break;
995 			}
996 
997 			if (is_imm8(imm32))
998 				EMIT3(0x83, add_1reg(b3, dst_reg), imm32);
999 			else if (is_axreg(dst_reg))
1000 				EMIT1_off32(b2, imm32);
1001 			else
1002 				EMIT2_off32(0x81, add_1reg(b3, dst_reg), imm32);
1003 			break;
1004 
1005 		case BPF_ALU64 | BPF_MOV | BPF_K:
1006 		case BPF_ALU | BPF_MOV | BPF_K:
1007 			emit_mov_imm32(&prog, BPF_CLASS(insn->code) == BPF_ALU64,
1008 				       dst_reg, imm32);
1009 			break;
1010 
1011 		case BPF_LD | BPF_IMM | BPF_DW:
1012 			emit_mov_imm64(&prog, dst_reg, insn[1].imm, insn[0].imm);
1013 			insn++;
1014 			i++;
1015 			break;
1016 
1017 			/* dst %= src, dst /= src, dst %= imm32, dst /= imm32 */
1018 		case BPF_ALU | BPF_MOD | BPF_X:
1019 		case BPF_ALU | BPF_DIV | BPF_X:
1020 		case BPF_ALU | BPF_MOD | BPF_K:
1021 		case BPF_ALU | BPF_DIV | BPF_K:
1022 		case BPF_ALU64 | BPF_MOD | BPF_X:
1023 		case BPF_ALU64 | BPF_DIV | BPF_X:
1024 		case BPF_ALU64 | BPF_MOD | BPF_K:
1025 		case BPF_ALU64 | BPF_DIV | BPF_K: {
1026 			bool is64 = BPF_CLASS(insn->code) == BPF_ALU64;
1027 
1028 			if (dst_reg != BPF_REG_0)
1029 				EMIT1(0x50); /* push rax */
1030 			if (dst_reg != BPF_REG_3)
1031 				EMIT1(0x52); /* push rdx */
1032 
1033 			if (BPF_SRC(insn->code) == BPF_X) {
1034 				if (src_reg == BPF_REG_0 ||
1035 				    src_reg == BPF_REG_3) {
1036 					/* mov r11, src_reg */
1037 					EMIT_mov(AUX_REG, src_reg);
1038 					src_reg = AUX_REG;
1039 				}
1040 			} else {
1041 				/* mov r11, imm32 */
1042 				EMIT3_off32(0x49, 0xC7, 0xC3, imm32);
1043 				src_reg = AUX_REG;
1044 			}
1045 
1046 			if (dst_reg != BPF_REG_0)
1047 				/* mov rax, dst_reg */
1048 				emit_mov_reg(&prog, is64, BPF_REG_0, dst_reg);
1049 
1050 			/*
1051 			 * xor edx, edx
1052 			 * equivalent to 'xor rdx, rdx', but one byte less
1053 			 */
1054 			EMIT2(0x31, 0xd2);
1055 
1056 			/* div src_reg */
1057 			maybe_emit_1mod(&prog, src_reg, is64);
1058 			EMIT2(0xF7, add_1reg(0xF0, src_reg));
1059 
1060 			if (BPF_OP(insn->code) == BPF_MOD &&
1061 			    dst_reg != BPF_REG_3)
1062 				/* mov dst_reg, rdx */
1063 				emit_mov_reg(&prog, is64, dst_reg, BPF_REG_3);
1064 			else if (BPF_OP(insn->code) == BPF_DIV &&
1065 				 dst_reg != BPF_REG_0)
1066 				/* mov dst_reg, rax */
1067 				emit_mov_reg(&prog, is64, dst_reg, BPF_REG_0);
1068 
1069 			if (dst_reg != BPF_REG_3)
1070 				EMIT1(0x5A); /* pop rdx */
1071 			if (dst_reg != BPF_REG_0)
1072 				EMIT1(0x58); /* pop rax */
1073 			break;
1074 		}
1075 
1076 		case BPF_ALU | BPF_MUL | BPF_K:
1077 		case BPF_ALU64 | BPF_MUL | BPF_K:
1078 			maybe_emit_mod(&prog, dst_reg, dst_reg,
1079 				       BPF_CLASS(insn->code) == BPF_ALU64);
1080 
1081 			if (is_imm8(imm32))
1082 				/* imul dst_reg, dst_reg, imm8 */
1083 				EMIT3(0x6B, add_2reg(0xC0, dst_reg, dst_reg),
1084 				      imm32);
1085 			else
1086 				/* imul dst_reg, dst_reg, imm32 */
1087 				EMIT2_off32(0x69,
1088 					    add_2reg(0xC0, dst_reg, dst_reg),
1089 					    imm32);
1090 			break;
1091 
1092 		case BPF_ALU | BPF_MUL | BPF_X:
1093 		case BPF_ALU64 | BPF_MUL | BPF_X:
1094 			maybe_emit_mod(&prog, src_reg, dst_reg,
1095 				       BPF_CLASS(insn->code) == BPF_ALU64);
1096 
1097 			/* imul dst_reg, src_reg */
1098 			EMIT3(0x0F, 0xAF, add_2reg(0xC0, src_reg, dst_reg));
1099 			break;
1100 
1101 			/* Shifts */
1102 		case BPF_ALU | BPF_LSH | BPF_K:
1103 		case BPF_ALU | BPF_RSH | BPF_K:
1104 		case BPF_ALU | BPF_ARSH | BPF_K:
1105 		case BPF_ALU64 | BPF_LSH | BPF_K:
1106 		case BPF_ALU64 | BPF_RSH | BPF_K:
1107 		case BPF_ALU64 | BPF_ARSH | BPF_K:
1108 			maybe_emit_1mod(&prog, dst_reg,
1109 					BPF_CLASS(insn->code) == BPF_ALU64);
1110 
1111 			b3 = simple_alu_opcodes[BPF_OP(insn->code)];
1112 			if (imm32 == 1)
1113 				EMIT2(0xD1, add_1reg(b3, dst_reg));
1114 			else
1115 				EMIT3(0xC1, add_1reg(b3, dst_reg), imm32);
1116 			break;
1117 
1118 		case BPF_ALU | BPF_LSH | BPF_X:
1119 		case BPF_ALU | BPF_RSH | BPF_X:
1120 		case BPF_ALU | BPF_ARSH | BPF_X:
1121 		case BPF_ALU64 | BPF_LSH | BPF_X:
1122 		case BPF_ALU64 | BPF_RSH | BPF_X:
1123 		case BPF_ALU64 | BPF_ARSH | BPF_X:
1124 
1125 			/* Check for bad case when dst_reg == rcx */
1126 			if (dst_reg == BPF_REG_4) {
1127 				/* mov r11, dst_reg */
1128 				EMIT_mov(AUX_REG, dst_reg);
1129 				dst_reg = AUX_REG;
1130 			}
1131 
1132 			if (src_reg != BPF_REG_4) { /* common case */
1133 				EMIT1(0x51); /* push rcx */
1134 
1135 				/* mov rcx, src_reg */
1136 				EMIT_mov(BPF_REG_4, src_reg);
1137 			}
1138 
1139 			/* shl %rax, %cl | shr %rax, %cl | sar %rax, %cl */
1140 			maybe_emit_1mod(&prog, dst_reg,
1141 					BPF_CLASS(insn->code) == BPF_ALU64);
1142 
1143 			b3 = simple_alu_opcodes[BPF_OP(insn->code)];
1144 			EMIT2(0xD3, add_1reg(b3, dst_reg));
1145 
1146 			if (src_reg != BPF_REG_4)
1147 				EMIT1(0x59); /* pop rcx */
1148 
1149 			if (insn->dst_reg == BPF_REG_4)
1150 				/* mov dst_reg, r11 */
1151 				EMIT_mov(insn->dst_reg, AUX_REG);
1152 			break;
1153 
1154 		case BPF_ALU | BPF_END | BPF_FROM_BE:
1155 			switch (imm32) {
1156 			case 16:
1157 				/* Emit 'ror %ax, 8' to swap lower 2 bytes */
1158 				EMIT1(0x66);
1159 				if (is_ereg(dst_reg))
1160 					EMIT1(0x41);
1161 				EMIT3(0xC1, add_1reg(0xC8, dst_reg), 8);
1162 
1163 				/* Emit 'movzwl eax, ax' */
1164 				if (is_ereg(dst_reg))
1165 					EMIT3(0x45, 0x0F, 0xB7);
1166 				else
1167 					EMIT2(0x0F, 0xB7);
1168 				EMIT1(add_2reg(0xC0, dst_reg, dst_reg));
1169 				break;
1170 			case 32:
1171 				/* Emit 'bswap eax' to swap lower 4 bytes */
1172 				if (is_ereg(dst_reg))
1173 					EMIT2(0x41, 0x0F);
1174 				else
1175 					EMIT1(0x0F);
1176 				EMIT1(add_1reg(0xC8, dst_reg));
1177 				break;
1178 			case 64:
1179 				/* Emit 'bswap rax' to swap 8 bytes */
1180 				EMIT3(add_1mod(0x48, dst_reg), 0x0F,
1181 				      add_1reg(0xC8, dst_reg));
1182 				break;
1183 			}
1184 			break;
1185 
1186 		case BPF_ALU | BPF_END | BPF_FROM_LE:
1187 			switch (imm32) {
1188 			case 16:
1189 				/*
1190 				 * Emit 'movzwl eax, ax' to zero extend 16-bit
1191 				 * into 64 bit
1192 				 */
1193 				if (is_ereg(dst_reg))
1194 					EMIT3(0x45, 0x0F, 0xB7);
1195 				else
1196 					EMIT2(0x0F, 0xB7);
1197 				EMIT1(add_2reg(0xC0, dst_reg, dst_reg));
1198 				break;
1199 			case 32:
1200 				/* Emit 'mov eax, eax' to clear upper 32-bits */
1201 				if (is_ereg(dst_reg))
1202 					EMIT1(0x45);
1203 				EMIT2(0x89, add_2reg(0xC0, dst_reg, dst_reg));
1204 				break;
1205 			case 64:
1206 				/* nop */
1207 				break;
1208 			}
1209 			break;
1210 
1211 			/* speculation barrier */
1212 		case BPF_ST | BPF_NOSPEC:
1213 			if (boot_cpu_has(X86_FEATURE_XMM2))
1214 				EMIT_LFENCE();
1215 			break;
1216 
1217 			/* ST: *(u8*)(dst_reg + off) = imm */
1218 		case BPF_ST | BPF_MEM | BPF_B:
1219 			if (is_ereg(dst_reg))
1220 				EMIT2(0x41, 0xC6);
1221 			else
1222 				EMIT1(0xC6);
1223 			goto st;
1224 		case BPF_ST | BPF_MEM | BPF_H:
1225 			if (is_ereg(dst_reg))
1226 				EMIT3(0x66, 0x41, 0xC7);
1227 			else
1228 				EMIT2(0x66, 0xC7);
1229 			goto st;
1230 		case BPF_ST | BPF_MEM | BPF_W:
1231 			if (is_ereg(dst_reg))
1232 				EMIT2(0x41, 0xC7);
1233 			else
1234 				EMIT1(0xC7);
1235 			goto st;
1236 		case BPF_ST | BPF_MEM | BPF_DW:
1237 			EMIT2(add_1mod(0x48, dst_reg), 0xC7);
1238 
1239 st:			if (is_imm8(insn->off))
1240 				EMIT2(add_1reg(0x40, dst_reg), insn->off);
1241 			else
1242 				EMIT1_off32(add_1reg(0x80, dst_reg), insn->off);
1243 
1244 			EMIT(imm32, bpf_size_to_x86_bytes(BPF_SIZE(insn->code)));
1245 			break;
1246 
1247 			/* STX: *(u8*)(dst_reg + off) = src_reg */
1248 		case BPF_STX | BPF_MEM | BPF_B:
1249 		case BPF_STX | BPF_MEM | BPF_H:
1250 		case BPF_STX | BPF_MEM | BPF_W:
1251 		case BPF_STX | BPF_MEM | BPF_DW:
1252 			emit_stx(&prog, BPF_SIZE(insn->code), dst_reg, src_reg, insn->off);
1253 			break;
1254 
1255 			/* LDX: dst_reg = *(u8*)(src_reg + off) */
1256 		case BPF_LDX | BPF_MEM | BPF_B:
1257 		case BPF_LDX | BPF_PROBE_MEM | BPF_B:
1258 		case BPF_LDX | BPF_MEM | BPF_H:
1259 		case BPF_LDX | BPF_PROBE_MEM | BPF_H:
1260 		case BPF_LDX | BPF_MEM | BPF_W:
1261 		case BPF_LDX | BPF_PROBE_MEM | BPF_W:
1262 		case BPF_LDX | BPF_MEM | BPF_DW:
1263 		case BPF_LDX | BPF_PROBE_MEM | BPF_DW:
1264 			if (BPF_MODE(insn->code) == BPF_PROBE_MEM) {
1265 				/* Though the verifier prevents negative insn->off in BPF_PROBE_MEM
1266 				 * add abs(insn->off) to the limit to make sure that negative
1267 				 * offset won't be an issue.
1268 				 * insn->off is s16, so it won't affect valid pointers.
1269 				 */
1270 				u64 limit = TASK_SIZE_MAX + PAGE_SIZE + abs(insn->off);
1271 				u8 *end_of_jmp1, *end_of_jmp2;
1272 
1273 				/* Conservatively check that src_reg + insn->off is a kernel address:
1274 				 * 1. src_reg + insn->off >= limit
1275 				 * 2. src_reg + insn->off doesn't become small positive.
1276 				 * Cannot do src_reg + insn->off >= limit in one branch,
1277 				 * since it needs two spare registers, but JIT has only one.
1278 				 */
1279 
1280 				/* movabsq r11, limit */
1281 				EMIT2(add_1mod(0x48, AUX_REG), add_1reg(0xB8, AUX_REG));
1282 				EMIT((u32)limit, 4);
1283 				EMIT(limit >> 32, 4);
1284 				/* cmp src_reg, r11 */
1285 				maybe_emit_mod(&prog, src_reg, AUX_REG, true);
1286 				EMIT2(0x39, add_2reg(0xC0, src_reg, AUX_REG));
1287 				/* if unsigned '<' goto end_of_jmp2 */
1288 				EMIT2(X86_JB, 0);
1289 				end_of_jmp1 = prog;
1290 
1291 				/* mov r11, src_reg */
1292 				emit_mov_reg(&prog, true, AUX_REG, src_reg);
1293 				/* add r11, insn->off */
1294 				maybe_emit_1mod(&prog, AUX_REG, true);
1295 				EMIT2_off32(0x81, add_1reg(0xC0, AUX_REG), insn->off);
1296 				/* jmp if not carry to start_of_ldx
1297 				 * Otherwise ERR_PTR(-EINVAL) + 128 will be the user addr
1298 				 * that has to be rejected.
1299 				 */
1300 				EMIT2(0x73 /* JNC */, 0);
1301 				end_of_jmp2 = prog;
1302 
1303 				/* xor dst_reg, dst_reg */
1304 				emit_mov_imm32(&prog, false, dst_reg, 0);
1305 				/* jmp byte_after_ldx */
1306 				EMIT2(0xEB, 0);
1307 
1308 				/* populate jmp_offset for JB above to jump to xor dst_reg */
1309 				end_of_jmp1[-1] = end_of_jmp2 - end_of_jmp1;
1310 				/* populate jmp_offset for JNC above to jump to start_of_ldx */
1311 				start_of_ldx = prog;
1312 				end_of_jmp2[-1] = start_of_ldx - end_of_jmp2;
1313 			}
1314 			emit_ldx(&prog, BPF_SIZE(insn->code), dst_reg, src_reg, insn->off);
1315 			if (BPF_MODE(insn->code) == BPF_PROBE_MEM) {
1316 				struct exception_table_entry *ex;
1317 				u8 *_insn = image + proglen + (start_of_ldx - temp);
1318 				s64 delta;
1319 
1320 				/* populate jmp_offset for JMP above */
1321 				start_of_ldx[-1] = prog - start_of_ldx;
1322 
1323 				if (!bpf_prog->aux->extable)
1324 					break;
1325 
1326 				if (excnt >= bpf_prog->aux->num_exentries) {
1327 					pr_err("ex gen bug\n");
1328 					return -EFAULT;
1329 				}
1330 				ex = &bpf_prog->aux->extable[excnt++];
1331 
1332 				delta = _insn - (u8 *)&ex->insn;
1333 				if (!is_simm32(delta)) {
1334 					pr_err("extable->insn doesn't fit into 32-bit\n");
1335 					return -EFAULT;
1336 				}
1337 				/* switch ex to rw buffer for writes */
1338 				ex = (void *)rw_image + ((void *)ex - (void *)image);
1339 
1340 				ex->insn = delta;
1341 
1342 				ex->data = EX_TYPE_BPF;
1343 
1344 				if (dst_reg > BPF_REG_9) {
1345 					pr_err("verifier error\n");
1346 					return -EFAULT;
1347 				}
1348 				/*
1349 				 * Compute size of x86 insn and its target dest x86 register.
1350 				 * ex_handler_bpf() will use lower 8 bits to adjust
1351 				 * pt_regs->ip to jump over this x86 instruction
1352 				 * and upper bits to figure out which pt_regs to zero out.
1353 				 * End result: x86 insn "mov rbx, qword ptr [rax+0x14]"
1354 				 * of 4 bytes will be ignored and rbx will be zero inited.
1355 				 */
1356 				ex->fixup = (prog - start_of_ldx) | (reg2pt_regs[dst_reg] << 8);
1357 			}
1358 			break;
1359 
1360 		case BPF_STX | BPF_ATOMIC | BPF_W:
1361 		case BPF_STX | BPF_ATOMIC | BPF_DW:
1362 			if (insn->imm == (BPF_AND | BPF_FETCH) ||
1363 			    insn->imm == (BPF_OR | BPF_FETCH) ||
1364 			    insn->imm == (BPF_XOR | BPF_FETCH)) {
1365 				bool is64 = BPF_SIZE(insn->code) == BPF_DW;
1366 				u32 real_src_reg = src_reg;
1367 				u32 real_dst_reg = dst_reg;
1368 				u8 *branch_target;
1369 
1370 				/*
1371 				 * Can't be implemented with a single x86 insn.
1372 				 * Need to do a CMPXCHG loop.
1373 				 */
1374 
1375 				/* Will need RAX as a CMPXCHG operand so save R0 */
1376 				emit_mov_reg(&prog, true, BPF_REG_AX, BPF_REG_0);
1377 				if (src_reg == BPF_REG_0)
1378 					real_src_reg = BPF_REG_AX;
1379 				if (dst_reg == BPF_REG_0)
1380 					real_dst_reg = BPF_REG_AX;
1381 
1382 				branch_target = prog;
1383 				/* Load old value */
1384 				emit_ldx(&prog, BPF_SIZE(insn->code),
1385 					 BPF_REG_0, real_dst_reg, insn->off);
1386 				/*
1387 				 * Perform the (commutative) operation locally,
1388 				 * put the result in the AUX_REG.
1389 				 */
1390 				emit_mov_reg(&prog, is64, AUX_REG, BPF_REG_0);
1391 				maybe_emit_mod(&prog, AUX_REG, real_src_reg, is64);
1392 				EMIT2(simple_alu_opcodes[BPF_OP(insn->imm)],
1393 				      add_2reg(0xC0, AUX_REG, real_src_reg));
1394 				/* Attempt to swap in new value */
1395 				err = emit_atomic(&prog, BPF_CMPXCHG,
1396 						  real_dst_reg, AUX_REG,
1397 						  insn->off,
1398 						  BPF_SIZE(insn->code));
1399 				if (WARN_ON(err))
1400 					return err;
1401 				/*
1402 				 * ZF tells us whether we won the race. If it's
1403 				 * cleared we need to try again.
1404 				 */
1405 				EMIT2(X86_JNE, -(prog - branch_target) - 2);
1406 				/* Return the pre-modification value */
1407 				emit_mov_reg(&prog, is64, real_src_reg, BPF_REG_0);
1408 				/* Restore R0 after clobbering RAX */
1409 				emit_mov_reg(&prog, true, BPF_REG_0, BPF_REG_AX);
1410 				break;
1411 			}
1412 
1413 			err = emit_atomic(&prog, insn->imm, dst_reg, src_reg,
1414 					  insn->off, BPF_SIZE(insn->code));
1415 			if (err)
1416 				return err;
1417 			break;
1418 
1419 			/* call */
1420 		case BPF_JMP | BPF_CALL:
1421 			func = (u8 *) __bpf_call_base + imm32;
1422 			if (tail_call_reachable) {
1423 				/* mov rax, qword ptr [rbp - rounded_stack_depth - 8] */
1424 				EMIT3_off32(0x48, 0x8B, 0x85,
1425 					    -round_up(bpf_prog->aux->stack_depth, 8) - 8);
1426 				if (!imm32 || emit_call(&prog, func, image + addrs[i - 1] + 7))
1427 					return -EINVAL;
1428 			} else {
1429 				if (!imm32 || emit_call(&prog, func, image + addrs[i - 1]))
1430 					return -EINVAL;
1431 			}
1432 			break;
1433 
1434 		case BPF_JMP | BPF_TAIL_CALL:
1435 			if (imm32)
1436 				emit_bpf_tail_call_direct(&bpf_prog->aux->poke_tab[imm32 - 1],
1437 							  &prog, image + addrs[i - 1],
1438 							  callee_regs_used,
1439 							  bpf_prog->aux->stack_depth,
1440 							  ctx);
1441 			else
1442 				emit_bpf_tail_call_indirect(&prog,
1443 							    callee_regs_used,
1444 							    bpf_prog->aux->stack_depth,
1445 							    image + addrs[i - 1],
1446 							    ctx);
1447 			break;
1448 
1449 			/* cond jump */
1450 		case BPF_JMP | BPF_JEQ | BPF_X:
1451 		case BPF_JMP | BPF_JNE | BPF_X:
1452 		case BPF_JMP | BPF_JGT | BPF_X:
1453 		case BPF_JMP | BPF_JLT | BPF_X:
1454 		case BPF_JMP | BPF_JGE | BPF_X:
1455 		case BPF_JMP | BPF_JLE | BPF_X:
1456 		case BPF_JMP | BPF_JSGT | BPF_X:
1457 		case BPF_JMP | BPF_JSLT | BPF_X:
1458 		case BPF_JMP | BPF_JSGE | BPF_X:
1459 		case BPF_JMP | BPF_JSLE | BPF_X:
1460 		case BPF_JMP32 | BPF_JEQ | BPF_X:
1461 		case BPF_JMP32 | BPF_JNE | BPF_X:
1462 		case BPF_JMP32 | BPF_JGT | BPF_X:
1463 		case BPF_JMP32 | BPF_JLT | BPF_X:
1464 		case BPF_JMP32 | BPF_JGE | BPF_X:
1465 		case BPF_JMP32 | BPF_JLE | BPF_X:
1466 		case BPF_JMP32 | BPF_JSGT | BPF_X:
1467 		case BPF_JMP32 | BPF_JSLT | BPF_X:
1468 		case BPF_JMP32 | BPF_JSGE | BPF_X:
1469 		case BPF_JMP32 | BPF_JSLE | BPF_X:
1470 			/* cmp dst_reg, src_reg */
1471 			maybe_emit_mod(&prog, dst_reg, src_reg,
1472 				       BPF_CLASS(insn->code) == BPF_JMP);
1473 			EMIT2(0x39, add_2reg(0xC0, dst_reg, src_reg));
1474 			goto emit_cond_jmp;
1475 
1476 		case BPF_JMP | BPF_JSET | BPF_X:
1477 		case BPF_JMP32 | BPF_JSET | BPF_X:
1478 			/* test dst_reg, src_reg */
1479 			maybe_emit_mod(&prog, dst_reg, src_reg,
1480 				       BPF_CLASS(insn->code) == BPF_JMP);
1481 			EMIT2(0x85, add_2reg(0xC0, dst_reg, src_reg));
1482 			goto emit_cond_jmp;
1483 
1484 		case BPF_JMP | BPF_JSET | BPF_K:
1485 		case BPF_JMP32 | BPF_JSET | BPF_K:
1486 			/* test dst_reg, imm32 */
1487 			maybe_emit_1mod(&prog, dst_reg,
1488 					BPF_CLASS(insn->code) == BPF_JMP);
1489 			EMIT2_off32(0xF7, add_1reg(0xC0, dst_reg), imm32);
1490 			goto emit_cond_jmp;
1491 
1492 		case BPF_JMP | BPF_JEQ | BPF_K:
1493 		case BPF_JMP | BPF_JNE | BPF_K:
1494 		case BPF_JMP | BPF_JGT | BPF_K:
1495 		case BPF_JMP | BPF_JLT | BPF_K:
1496 		case BPF_JMP | BPF_JGE | BPF_K:
1497 		case BPF_JMP | BPF_JLE | BPF_K:
1498 		case BPF_JMP | BPF_JSGT | BPF_K:
1499 		case BPF_JMP | BPF_JSLT | BPF_K:
1500 		case BPF_JMP | BPF_JSGE | BPF_K:
1501 		case BPF_JMP | BPF_JSLE | BPF_K:
1502 		case BPF_JMP32 | BPF_JEQ | BPF_K:
1503 		case BPF_JMP32 | BPF_JNE | BPF_K:
1504 		case BPF_JMP32 | BPF_JGT | BPF_K:
1505 		case BPF_JMP32 | BPF_JLT | BPF_K:
1506 		case BPF_JMP32 | BPF_JGE | BPF_K:
1507 		case BPF_JMP32 | BPF_JLE | BPF_K:
1508 		case BPF_JMP32 | BPF_JSGT | BPF_K:
1509 		case BPF_JMP32 | BPF_JSLT | BPF_K:
1510 		case BPF_JMP32 | BPF_JSGE | BPF_K:
1511 		case BPF_JMP32 | BPF_JSLE | BPF_K:
1512 			/* test dst_reg, dst_reg to save one extra byte */
1513 			if (imm32 == 0) {
1514 				maybe_emit_mod(&prog, dst_reg, dst_reg,
1515 					       BPF_CLASS(insn->code) == BPF_JMP);
1516 				EMIT2(0x85, add_2reg(0xC0, dst_reg, dst_reg));
1517 				goto emit_cond_jmp;
1518 			}
1519 
1520 			/* cmp dst_reg, imm8/32 */
1521 			maybe_emit_1mod(&prog, dst_reg,
1522 					BPF_CLASS(insn->code) == BPF_JMP);
1523 
1524 			if (is_imm8(imm32))
1525 				EMIT3(0x83, add_1reg(0xF8, dst_reg), imm32);
1526 			else
1527 				EMIT2_off32(0x81, add_1reg(0xF8, dst_reg), imm32);
1528 
1529 emit_cond_jmp:		/* Convert BPF opcode to x86 */
1530 			switch (BPF_OP(insn->code)) {
1531 			case BPF_JEQ:
1532 				jmp_cond = X86_JE;
1533 				break;
1534 			case BPF_JSET:
1535 			case BPF_JNE:
1536 				jmp_cond = X86_JNE;
1537 				break;
1538 			case BPF_JGT:
1539 				/* GT is unsigned '>', JA in x86 */
1540 				jmp_cond = X86_JA;
1541 				break;
1542 			case BPF_JLT:
1543 				/* LT is unsigned '<', JB in x86 */
1544 				jmp_cond = X86_JB;
1545 				break;
1546 			case BPF_JGE:
1547 				/* GE is unsigned '>=', JAE in x86 */
1548 				jmp_cond = X86_JAE;
1549 				break;
1550 			case BPF_JLE:
1551 				/* LE is unsigned '<=', JBE in x86 */
1552 				jmp_cond = X86_JBE;
1553 				break;
1554 			case BPF_JSGT:
1555 				/* Signed '>', GT in x86 */
1556 				jmp_cond = X86_JG;
1557 				break;
1558 			case BPF_JSLT:
1559 				/* Signed '<', LT in x86 */
1560 				jmp_cond = X86_JL;
1561 				break;
1562 			case BPF_JSGE:
1563 				/* Signed '>=', GE in x86 */
1564 				jmp_cond = X86_JGE;
1565 				break;
1566 			case BPF_JSLE:
1567 				/* Signed '<=', LE in x86 */
1568 				jmp_cond = X86_JLE;
1569 				break;
1570 			default: /* to silence GCC warning */
1571 				return -EFAULT;
1572 			}
1573 			jmp_offset = addrs[i + insn->off] - addrs[i];
1574 			if (is_imm8(jmp_offset)) {
1575 				if (jmp_padding) {
1576 					/* To keep the jmp_offset valid, the extra bytes are
1577 					 * padded before the jump insn, so we subtract the
1578 					 * 2 bytes of jmp_cond insn from INSN_SZ_DIFF.
1579 					 *
1580 					 * If the previous pass already emits an imm8
1581 					 * jmp_cond, then this BPF insn won't shrink, so
1582 					 * "nops" is 0.
1583 					 *
1584 					 * On the other hand, if the previous pass emits an
1585 					 * imm32 jmp_cond, the extra 4 bytes(*) is padded to
1586 					 * keep the image from shrinking further.
1587 					 *
1588 					 * (*) imm32 jmp_cond is 6 bytes, and imm8 jmp_cond
1589 					 *     is 2 bytes, so the size difference is 4 bytes.
1590 					 */
1591 					nops = INSN_SZ_DIFF - 2;
1592 					if (nops != 0 && nops != 4) {
1593 						pr_err("unexpected jmp_cond padding: %d bytes\n",
1594 						       nops);
1595 						return -EFAULT;
1596 					}
1597 					emit_nops(&prog, nops);
1598 				}
1599 				EMIT2(jmp_cond, jmp_offset);
1600 			} else if (is_simm32(jmp_offset)) {
1601 				EMIT2_off32(0x0F, jmp_cond + 0x10, jmp_offset);
1602 			} else {
1603 				pr_err("cond_jmp gen bug %llx\n", jmp_offset);
1604 				return -EFAULT;
1605 			}
1606 
1607 			break;
1608 
1609 		case BPF_JMP | BPF_JA:
1610 			if (insn->off == -1)
1611 				/* -1 jmp instructions will always jump
1612 				 * backwards two bytes. Explicitly handling
1613 				 * this case avoids wasting too many passes
1614 				 * when there are long sequences of replaced
1615 				 * dead code.
1616 				 */
1617 				jmp_offset = -2;
1618 			else
1619 				jmp_offset = addrs[i + insn->off] - addrs[i];
1620 
1621 			if (!jmp_offset) {
1622 				/*
1623 				 * If jmp_padding is enabled, the extra nops will
1624 				 * be inserted. Otherwise, optimize out nop jumps.
1625 				 */
1626 				if (jmp_padding) {
1627 					/* There are 3 possible conditions.
1628 					 * (1) This BPF_JA is already optimized out in
1629 					 *     the previous run, so there is no need
1630 					 *     to pad any extra byte (0 byte).
1631 					 * (2) The previous pass emits an imm8 jmp,
1632 					 *     so we pad 2 bytes to match the previous
1633 					 *     insn size.
1634 					 * (3) Similarly, the previous pass emits an
1635 					 *     imm32 jmp, and 5 bytes is padded.
1636 					 */
1637 					nops = INSN_SZ_DIFF;
1638 					if (nops != 0 && nops != 2 && nops != 5) {
1639 						pr_err("unexpected nop jump padding: %d bytes\n",
1640 						       nops);
1641 						return -EFAULT;
1642 					}
1643 					emit_nops(&prog, nops);
1644 				}
1645 				break;
1646 			}
1647 emit_jmp:
1648 			if (is_imm8(jmp_offset)) {
1649 				if (jmp_padding) {
1650 					/* To avoid breaking jmp_offset, the extra bytes
1651 					 * are padded before the actual jmp insn, so
1652 					 * 2 bytes is subtracted from INSN_SZ_DIFF.
1653 					 *
1654 					 * If the previous pass already emits an imm8
1655 					 * jmp, there is nothing to pad (0 byte).
1656 					 *
1657 					 * If it emits an imm32 jmp (5 bytes) previously
1658 					 * and now an imm8 jmp (2 bytes), then we pad
1659 					 * (5 - 2 = 3) bytes to stop the image from
1660 					 * shrinking further.
1661 					 */
1662 					nops = INSN_SZ_DIFF - 2;
1663 					if (nops != 0 && nops != 3) {
1664 						pr_err("unexpected jump padding: %d bytes\n",
1665 						       nops);
1666 						return -EFAULT;
1667 					}
1668 					emit_nops(&prog, INSN_SZ_DIFF - 2);
1669 				}
1670 				EMIT2(0xEB, jmp_offset);
1671 			} else if (is_simm32(jmp_offset)) {
1672 				EMIT1_off32(0xE9, jmp_offset);
1673 			} else {
1674 				pr_err("jmp gen bug %llx\n", jmp_offset);
1675 				return -EFAULT;
1676 			}
1677 			break;
1678 
1679 		case BPF_JMP | BPF_EXIT:
1680 			if (seen_exit) {
1681 				jmp_offset = ctx->cleanup_addr - addrs[i];
1682 				goto emit_jmp;
1683 			}
1684 			seen_exit = true;
1685 			/* Update cleanup_addr */
1686 			ctx->cleanup_addr = proglen;
1687 			pop_callee_regs(&prog, callee_regs_used);
1688 			EMIT1(0xC9);         /* leave */
1689 			EMIT1(0xC3);         /* ret */
1690 			break;
1691 
1692 		default:
1693 			/*
1694 			 * By design x86-64 JIT should support all BPF instructions.
1695 			 * This error will be seen if new instruction was added
1696 			 * to the interpreter, but not to the JIT, or if there is
1697 			 * junk in bpf_prog.
1698 			 */
1699 			pr_err("bpf_jit: unknown opcode %02x\n", insn->code);
1700 			return -EINVAL;
1701 		}
1702 
1703 		ilen = prog - temp;
1704 		if (ilen > BPF_MAX_INSN_SIZE) {
1705 			pr_err("bpf_jit: fatal insn size error\n");
1706 			return -EFAULT;
1707 		}
1708 
1709 		if (image) {
1710 			/*
1711 			 * When populating the image, assert that:
1712 			 *
1713 			 *  i) We do not write beyond the allocated space, and
1714 			 * ii) addrs[i] did not change from the prior run, in order
1715 			 *     to validate assumptions made for computing branch
1716 			 *     displacements.
1717 			 */
1718 			if (unlikely(proglen + ilen > oldproglen ||
1719 				     proglen + ilen != addrs[i])) {
1720 				pr_err("bpf_jit: fatal error\n");
1721 				return -EFAULT;
1722 			}
1723 			memcpy(rw_image + proglen, temp, ilen);
1724 		}
1725 		proglen += ilen;
1726 		addrs[i] = proglen;
1727 		prog = temp;
1728 	}
1729 
1730 	if (image && excnt != bpf_prog->aux->num_exentries) {
1731 		pr_err("extable is not populated\n");
1732 		return -EFAULT;
1733 	}
1734 	return proglen;
1735 }
1736 
1737 static void save_regs(const struct btf_func_model *m, u8 **prog, int nr_args,
1738 		      int stack_size)
1739 {
1740 	int i;
1741 	/* Store function arguments to stack.
1742 	 * For a function that accepts two pointers the sequence will be:
1743 	 * mov QWORD PTR [rbp-0x10],rdi
1744 	 * mov QWORD PTR [rbp-0x8],rsi
1745 	 */
1746 	for (i = 0; i < min(nr_args, 6); i++)
1747 		emit_stx(prog, bytes_to_bpf_size(m->arg_size[i]),
1748 			 BPF_REG_FP,
1749 			 i == 5 ? X86_REG_R9 : BPF_REG_1 + i,
1750 			 -(stack_size - i * 8));
1751 }
1752 
1753 static void restore_regs(const struct btf_func_model *m, u8 **prog, int nr_args,
1754 			 int stack_size)
1755 {
1756 	int i;
1757 
1758 	/* Restore function arguments from stack.
1759 	 * For a function that accepts two pointers the sequence will be:
1760 	 * EMIT4(0x48, 0x8B, 0x7D, 0xF0); mov rdi,QWORD PTR [rbp-0x10]
1761 	 * EMIT4(0x48, 0x8B, 0x75, 0xF8); mov rsi,QWORD PTR [rbp-0x8]
1762 	 */
1763 	for (i = 0; i < min(nr_args, 6); i++)
1764 		emit_ldx(prog, bytes_to_bpf_size(m->arg_size[i]),
1765 			 i == 5 ? X86_REG_R9 : BPF_REG_1 + i,
1766 			 BPF_REG_FP,
1767 			 -(stack_size - i * 8));
1768 }
1769 
1770 static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
1771 			   struct bpf_tramp_link *l, int stack_size,
1772 			   int run_ctx_off, bool save_ret)
1773 {
1774 	u8 *prog = *pprog;
1775 	u8 *jmp_insn;
1776 	int ctx_cookie_off = offsetof(struct bpf_tramp_run_ctx, bpf_cookie);
1777 	struct bpf_prog *p = l->link.prog;
1778 	u64 cookie = l->cookie;
1779 
1780 	/* mov rdi, cookie */
1781 	emit_mov_imm64(&prog, BPF_REG_1, (long) cookie >> 32, (u32) (long) cookie);
1782 
1783 	/* Prepare struct bpf_tramp_run_ctx.
1784 	 *
1785 	 * bpf_tramp_run_ctx is already preserved by
1786 	 * arch_prepare_bpf_trampoline().
1787 	 *
1788 	 * mov QWORD PTR [rbp - run_ctx_off + ctx_cookie_off], rdi
1789 	 */
1790 	emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_1, -run_ctx_off + ctx_cookie_off);
1791 
1792 	/* arg1: mov rdi, progs[i] */
1793 	emit_mov_imm64(&prog, BPF_REG_1, (long) p >> 32, (u32) (long) p);
1794 	/* arg2: lea rsi, [rbp - ctx_cookie_off] */
1795 	EMIT4(0x48, 0x8D, 0x75, -run_ctx_off);
1796 
1797 	if (emit_call(&prog,
1798 		      p->aux->sleepable ? __bpf_prog_enter_sleepable :
1799 		      __bpf_prog_enter, prog))
1800 			return -EINVAL;
1801 	/* remember prog start time returned by __bpf_prog_enter */
1802 	emit_mov_reg(&prog, true, BPF_REG_6, BPF_REG_0);
1803 
1804 	/* if (__bpf_prog_enter*(prog) == 0)
1805 	 *	goto skip_exec_of_prog;
1806 	 */
1807 	EMIT3(0x48, 0x85, 0xC0);  /* test rax,rax */
1808 	/* emit 2 nops that will be replaced with JE insn */
1809 	jmp_insn = prog;
1810 	emit_nops(&prog, 2);
1811 
1812 	/* arg1: lea rdi, [rbp - stack_size] */
1813 	EMIT4(0x48, 0x8D, 0x7D, -stack_size);
1814 	/* arg2: progs[i]->insnsi for interpreter */
1815 	if (!p->jited)
1816 		emit_mov_imm64(&prog, BPF_REG_2,
1817 			       (long) p->insnsi >> 32,
1818 			       (u32) (long) p->insnsi);
1819 	/* call JITed bpf program or interpreter */
1820 	if (emit_call(&prog, p->bpf_func, prog))
1821 		return -EINVAL;
1822 
1823 	/*
1824 	 * BPF_TRAMP_MODIFY_RETURN trampolines can modify the return
1825 	 * of the previous call which is then passed on the stack to
1826 	 * the next BPF program.
1827 	 *
1828 	 * BPF_TRAMP_FENTRY trampoline may need to return the return
1829 	 * value of BPF_PROG_TYPE_STRUCT_OPS prog.
1830 	 */
1831 	if (save_ret)
1832 		emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8);
1833 
1834 	/* replace 2 nops with JE insn, since jmp target is known */
1835 	jmp_insn[0] = X86_JE;
1836 	jmp_insn[1] = prog - jmp_insn - 2;
1837 
1838 	/* arg1: mov rdi, progs[i] */
1839 	emit_mov_imm64(&prog, BPF_REG_1, (long) p >> 32, (u32) (long) p);
1840 	/* arg2: mov rsi, rbx <- start time in nsec */
1841 	emit_mov_reg(&prog, true, BPF_REG_2, BPF_REG_6);
1842 	/* arg3: lea rdx, [rbp - run_ctx_off] */
1843 	EMIT4(0x48, 0x8D, 0x55, -run_ctx_off);
1844 	if (emit_call(&prog,
1845 		      p->aux->sleepable ? __bpf_prog_exit_sleepable :
1846 		      __bpf_prog_exit, prog))
1847 			return -EINVAL;
1848 
1849 	*pprog = prog;
1850 	return 0;
1851 }
1852 
1853 static void emit_align(u8 **pprog, u32 align)
1854 {
1855 	u8 *target, *prog = *pprog;
1856 
1857 	target = PTR_ALIGN(prog, align);
1858 	if (target != prog)
1859 		emit_nops(&prog, target - prog);
1860 
1861 	*pprog = prog;
1862 }
1863 
1864 static int emit_cond_near_jump(u8 **pprog, void *func, void *ip, u8 jmp_cond)
1865 {
1866 	u8 *prog = *pprog;
1867 	s64 offset;
1868 
1869 	offset = func - (ip + 2 + 4);
1870 	if (!is_simm32(offset)) {
1871 		pr_err("Target %p is out of range\n", func);
1872 		return -EINVAL;
1873 	}
1874 	EMIT2_off32(0x0F, jmp_cond + 0x10, offset);
1875 	*pprog = prog;
1876 	return 0;
1877 }
1878 
1879 static int invoke_bpf(const struct btf_func_model *m, u8 **pprog,
1880 		      struct bpf_tramp_links *tl, int stack_size,
1881 		      int run_ctx_off, bool save_ret)
1882 {
1883 	int i;
1884 	u8 *prog = *pprog;
1885 
1886 	for (i = 0; i < tl->nr_links; i++) {
1887 		if (invoke_bpf_prog(m, &prog, tl->links[i], stack_size,
1888 				    run_ctx_off, save_ret))
1889 			return -EINVAL;
1890 	}
1891 	*pprog = prog;
1892 	return 0;
1893 }
1894 
1895 static int invoke_bpf_mod_ret(const struct btf_func_model *m, u8 **pprog,
1896 			      struct bpf_tramp_links *tl, int stack_size,
1897 			      int run_ctx_off, u8 **branches)
1898 {
1899 	u8 *prog = *pprog;
1900 	int i;
1901 
1902 	/* The first fmod_ret program will receive a garbage return value.
1903 	 * Set this to 0 to avoid confusing the program.
1904 	 */
1905 	emit_mov_imm32(&prog, false, BPF_REG_0, 0);
1906 	emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8);
1907 	for (i = 0; i < tl->nr_links; i++) {
1908 		if (invoke_bpf_prog(m, &prog, tl->links[i], stack_size, run_ctx_off, true))
1909 			return -EINVAL;
1910 
1911 		/* mod_ret prog stored return value into [rbp - 8]. Emit:
1912 		 * if (*(u64 *)(rbp - 8) !=  0)
1913 		 *	goto do_fexit;
1914 		 */
1915 		/* cmp QWORD PTR [rbp - 0x8], 0x0 */
1916 		EMIT4(0x48, 0x83, 0x7d, 0xf8); EMIT1(0x00);
1917 
1918 		/* Save the location of the branch and Generate 6 nops
1919 		 * (4 bytes for an offset and 2 bytes for the jump) These nops
1920 		 * are replaced with a conditional jump once do_fexit (i.e. the
1921 		 * start of the fexit invocation) is finalized.
1922 		 */
1923 		branches[i] = prog;
1924 		emit_nops(&prog, 4 + 2);
1925 	}
1926 
1927 	*pprog = prog;
1928 	return 0;
1929 }
1930 
1931 static bool is_valid_bpf_tramp_flags(unsigned int flags)
1932 {
1933 	if ((flags & BPF_TRAMP_F_RESTORE_REGS) &&
1934 	    (flags & BPF_TRAMP_F_SKIP_FRAME))
1935 		return false;
1936 
1937 	/*
1938 	 * BPF_TRAMP_F_RET_FENTRY_RET is only used by bpf_struct_ops,
1939 	 * and it must be used alone.
1940 	 */
1941 	if ((flags & BPF_TRAMP_F_RET_FENTRY_RET) &&
1942 	    (flags & ~BPF_TRAMP_F_RET_FENTRY_RET))
1943 		return false;
1944 
1945 	return true;
1946 }
1947 
1948 /* Example:
1949  * __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev);
1950  * its 'struct btf_func_model' will be nr_args=2
1951  * The assembly code when eth_type_trans is executing after trampoline:
1952  *
1953  * push rbp
1954  * mov rbp, rsp
1955  * sub rsp, 16                     // space for skb and dev
1956  * push rbx                        // temp regs to pass start time
1957  * mov qword ptr [rbp - 16], rdi   // save skb pointer to stack
1958  * mov qword ptr [rbp - 8], rsi    // save dev pointer to stack
1959  * call __bpf_prog_enter           // rcu_read_lock and preempt_disable
1960  * mov rbx, rax                    // remember start time in bpf stats are enabled
1961  * lea rdi, [rbp - 16]             // R1==ctx of bpf prog
1962  * call addr_of_jited_FENTRY_prog
1963  * movabsq rdi, 64bit_addr_of_struct_bpf_prog  // unused if bpf stats are off
1964  * mov rsi, rbx                    // prog start time
1965  * call __bpf_prog_exit            // rcu_read_unlock, preempt_enable and stats math
1966  * mov rdi, qword ptr [rbp - 16]   // restore skb pointer from stack
1967  * mov rsi, qword ptr [rbp - 8]    // restore dev pointer from stack
1968  * pop rbx
1969  * leave
1970  * ret
1971  *
1972  * eth_type_trans has 5 byte nop at the beginning. These 5 bytes will be
1973  * replaced with 'call generated_bpf_trampoline'. When it returns
1974  * eth_type_trans will continue executing with original skb and dev pointers.
1975  *
1976  * The assembly code when eth_type_trans is called from trampoline:
1977  *
1978  * push rbp
1979  * mov rbp, rsp
1980  * sub rsp, 24                     // space for skb, dev, return value
1981  * push rbx                        // temp regs to pass start time
1982  * mov qword ptr [rbp - 24], rdi   // save skb pointer to stack
1983  * mov qword ptr [rbp - 16], rsi   // save dev pointer to stack
1984  * call __bpf_prog_enter           // rcu_read_lock and preempt_disable
1985  * mov rbx, rax                    // remember start time if bpf stats are enabled
1986  * lea rdi, [rbp - 24]             // R1==ctx of bpf prog
1987  * call addr_of_jited_FENTRY_prog  // bpf prog can access skb and dev
1988  * movabsq rdi, 64bit_addr_of_struct_bpf_prog  // unused if bpf stats are off
1989  * mov rsi, rbx                    // prog start time
1990  * call __bpf_prog_exit            // rcu_read_unlock, preempt_enable and stats math
1991  * mov rdi, qword ptr [rbp - 24]   // restore skb pointer from stack
1992  * mov rsi, qword ptr [rbp - 16]   // restore dev pointer from stack
1993  * call eth_type_trans+5           // execute body of eth_type_trans
1994  * mov qword ptr [rbp - 8], rax    // save return value
1995  * call __bpf_prog_enter           // rcu_read_lock and preempt_disable
1996  * mov rbx, rax                    // remember start time in bpf stats are enabled
1997  * lea rdi, [rbp - 24]             // R1==ctx of bpf prog
1998  * call addr_of_jited_FEXIT_prog   // bpf prog can access skb, dev, return value
1999  * movabsq rdi, 64bit_addr_of_struct_bpf_prog  // unused if bpf stats are off
2000  * mov rsi, rbx                    // prog start time
2001  * call __bpf_prog_exit            // rcu_read_unlock, preempt_enable and stats math
2002  * mov rax, qword ptr [rbp - 8]    // restore eth_type_trans's return value
2003  * pop rbx
2004  * leave
2005  * add rsp, 8                      // skip eth_type_trans's frame
2006  * ret                             // return to its caller
2007  */
2008 int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *image_end,
2009 				const struct btf_func_model *m, u32 flags,
2010 				struct bpf_tramp_links *tlinks,
2011 				void *orig_call)
2012 {
2013 	int ret, i, nr_args = m->nr_args;
2014 	int regs_off, ip_off, args_off, stack_size = nr_args * 8, run_ctx_off;
2015 	struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
2016 	struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
2017 	struct bpf_tramp_links *fmod_ret = &tlinks[BPF_TRAMP_MODIFY_RETURN];
2018 	u8 **branches = NULL;
2019 	u8 *prog;
2020 	bool save_ret;
2021 
2022 	/* x86-64 supports up to 6 arguments. 7+ can be added in the future */
2023 	if (nr_args > 6)
2024 		return -ENOTSUPP;
2025 
2026 	if (!is_valid_bpf_tramp_flags(flags))
2027 		return -EINVAL;
2028 
2029 	/* Generated trampoline stack layout:
2030 	 *
2031 	 * RBP + 8         [ return address  ]
2032 	 * RBP + 0         [ RBP             ]
2033 	 *
2034 	 * RBP - 8         [ return value    ]  BPF_TRAMP_F_CALL_ORIG or
2035 	 *                                      BPF_TRAMP_F_RET_FENTRY_RET flags
2036 	 *
2037 	 *                 [ reg_argN        ]  always
2038 	 *                 [ ...             ]
2039 	 * RBP - regs_off  [ reg_arg1        ]  program's ctx pointer
2040 	 *
2041 	 * RBP - args_off  [ args count      ]  always
2042 	 *
2043 	 * RBP - ip_off    [ traced function ]  BPF_TRAMP_F_IP_ARG flag
2044 	 *
2045 	 * RBP - run_ctx_off [ bpf_tramp_run_ctx ]
2046 	 */
2047 
2048 	/* room for return value of orig_call or fentry prog */
2049 	save_ret = flags & (BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_RET_FENTRY_RET);
2050 	if (save_ret)
2051 		stack_size += 8;
2052 
2053 	regs_off = stack_size;
2054 
2055 	/* args count  */
2056 	stack_size += 8;
2057 	args_off = stack_size;
2058 
2059 	if (flags & BPF_TRAMP_F_IP_ARG)
2060 		stack_size += 8; /* room for IP address argument */
2061 
2062 	ip_off = stack_size;
2063 
2064 	stack_size += (sizeof(struct bpf_tramp_run_ctx) + 7) & ~0x7;
2065 	run_ctx_off = stack_size;
2066 
2067 	if (flags & BPF_TRAMP_F_SKIP_FRAME) {
2068 		/* skip patched call instruction and point orig_call to actual
2069 		 * body of the kernel function.
2070 		 */
2071 		if (is_endbr(*(u32 *)orig_call))
2072 			orig_call += ENDBR_INSN_SIZE;
2073 		orig_call += X86_PATCH_SIZE;
2074 	}
2075 
2076 	prog = image;
2077 
2078 	EMIT_ENDBR();
2079 	EMIT1(0x55);		 /* push rbp */
2080 	EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */
2081 	EMIT4(0x48, 0x83, 0xEC, stack_size); /* sub rsp, stack_size */
2082 	EMIT1(0x53);		 /* push rbx */
2083 
2084 	/* Store number of arguments of the traced function:
2085 	 *   mov rax, nr_args
2086 	 *   mov QWORD PTR [rbp - args_off], rax
2087 	 */
2088 	emit_mov_imm64(&prog, BPF_REG_0, 0, (u32) nr_args);
2089 	emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -args_off);
2090 
2091 	if (flags & BPF_TRAMP_F_IP_ARG) {
2092 		/* Store IP address of the traced function:
2093 		 * mov rax, QWORD PTR [rbp + 8]
2094 		 * sub rax, X86_PATCH_SIZE
2095 		 * mov QWORD PTR [rbp - ip_off], rax
2096 		 */
2097 		emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, 8);
2098 		EMIT4(0x48, 0x83, 0xe8, X86_PATCH_SIZE);
2099 		emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -ip_off);
2100 	}
2101 
2102 	save_regs(m, &prog, nr_args, regs_off);
2103 
2104 	if (flags & BPF_TRAMP_F_CALL_ORIG) {
2105 		/* arg1: mov rdi, im */
2106 		emit_mov_imm64(&prog, BPF_REG_1, (long) im >> 32, (u32) (long) im);
2107 		if (emit_call(&prog, __bpf_tramp_enter, prog)) {
2108 			ret = -EINVAL;
2109 			goto cleanup;
2110 		}
2111 	}
2112 
2113 	if (fentry->nr_links)
2114 		if (invoke_bpf(m, &prog, fentry, regs_off, run_ctx_off,
2115 			       flags & BPF_TRAMP_F_RET_FENTRY_RET))
2116 			return -EINVAL;
2117 
2118 	if (fmod_ret->nr_links) {
2119 		branches = kcalloc(fmod_ret->nr_links, sizeof(u8 *),
2120 				   GFP_KERNEL);
2121 		if (!branches)
2122 			return -ENOMEM;
2123 
2124 		if (invoke_bpf_mod_ret(m, &prog, fmod_ret, regs_off,
2125 				       run_ctx_off, branches)) {
2126 			ret = -EINVAL;
2127 			goto cleanup;
2128 		}
2129 	}
2130 
2131 	if (flags & BPF_TRAMP_F_CALL_ORIG) {
2132 		restore_regs(m, &prog, nr_args, regs_off);
2133 
2134 		/* call original function */
2135 		if (emit_call(&prog, orig_call, prog)) {
2136 			ret = -EINVAL;
2137 			goto cleanup;
2138 		}
2139 		/* remember return value in a stack for bpf prog to access */
2140 		emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -8);
2141 		im->ip_after_call = prog;
2142 		memcpy(prog, x86_nops[5], X86_PATCH_SIZE);
2143 		prog += X86_PATCH_SIZE;
2144 	}
2145 
2146 	if (fmod_ret->nr_links) {
2147 		/* From Intel 64 and IA-32 Architectures Optimization
2148 		 * Reference Manual, 3.4.1.4 Code Alignment, Assembly/Compiler
2149 		 * Coding Rule 11: All branch targets should be 16-byte
2150 		 * aligned.
2151 		 */
2152 		emit_align(&prog, 16);
2153 		/* Update the branches saved in invoke_bpf_mod_ret with the
2154 		 * aligned address of do_fexit.
2155 		 */
2156 		for (i = 0; i < fmod_ret->nr_links; i++)
2157 			emit_cond_near_jump(&branches[i], prog, branches[i],
2158 					    X86_JNE);
2159 	}
2160 
2161 	if (fexit->nr_links)
2162 		if (invoke_bpf(m, &prog, fexit, regs_off, run_ctx_off, false)) {
2163 			ret = -EINVAL;
2164 			goto cleanup;
2165 		}
2166 
2167 	if (flags & BPF_TRAMP_F_RESTORE_REGS)
2168 		restore_regs(m, &prog, nr_args, regs_off);
2169 
2170 	/* This needs to be done regardless. If there were fmod_ret programs,
2171 	 * the return value is only updated on the stack and still needs to be
2172 	 * restored to R0.
2173 	 */
2174 	if (flags & BPF_TRAMP_F_CALL_ORIG) {
2175 		im->ip_epilogue = prog;
2176 		/* arg1: mov rdi, im */
2177 		emit_mov_imm64(&prog, BPF_REG_1, (long) im >> 32, (u32) (long) im);
2178 		if (emit_call(&prog, __bpf_tramp_exit, prog)) {
2179 			ret = -EINVAL;
2180 			goto cleanup;
2181 		}
2182 	}
2183 	/* restore return value of orig_call or fentry prog back into RAX */
2184 	if (save_ret)
2185 		emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, -8);
2186 
2187 	EMIT1(0x5B); /* pop rbx */
2188 	EMIT1(0xC9); /* leave */
2189 	if (flags & BPF_TRAMP_F_SKIP_FRAME)
2190 		/* skip our return address and return to parent */
2191 		EMIT4(0x48, 0x83, 0xC4, 8); /* add rsp, 8 */
2192 	EMIT1(0xC3); /* ret */
2193 	/* Make sure the trampoline generation logic doesn't overflow */
2194 	if (WARN_ON_ONCE(prog > (u8 *)image_end - BPF_INSN_SAFETY)) {
2195 		ret = -EFAULT;
2196 		goto cleanup;
2197 	}
2198 	ret = prog - (u8 *)image;
2199 
2200 cleanup:
2201 	kfree(branches);
2202 	return ret;
2203 }
2204 
2205 static int emit_bpf_dispatcher(u8 **pprog, int a, int b, s64 *progs)
2206 {
2207 	u8 *jg_reloc, *prog = *pprog;
2208 	int pivot, err, jg_bytes = 1;
2209 	s64 jg_offset;
2210 
2211 	if (a == b) {
2212 		/* Leaf node of recursion, i.e. not a range of indices
2213 		 * anymore.
2214 		 */
2215 		EMIT1(add_1mod(0x48, BPF_REG_3));	/* cmp rdx,func */
2216 		if (!is_simm32(progs[a]))
2217 			return -1;
2218 		EMIT2_off32(0x81, add_1reg(0xF8, BPF_REG_3),
2219 			    progs[a]);
2220 		err = emit_cond_near_jump(&prog,	/* je func */
2221 					  (void *)progs[a], prog,
2222 					  X86_JE);
2223 		if (err)
2224 			return err;
2225 
2226 		emit_indirect_jump(&prog, 2 /* rdx */, prog);
2227 
2228 		*pprog = prog;
2229 		return 0;
2230 	}
2231 
2232 	/* Not a leaf node, so we pivot, and recursively descend into
2233 	 * the lower and upper ranges.
2234 	 */
2235 	pivot = (b - a) / 2;
2236 	EMIT1(add_1mod(0x48, BPF_REG_3));		/* cmp rdx,func */
2237 	if (!is_simm32(progs[a + pivot]))
2238 		return -1;
2239 	EMIT2_off32(0x81, add_1reg(0xF8, BPF_REG_3), progs[a + pivot]);
2240 
2241 	if (pivot > 2) {				/* jg upper_part */
2242 		/* Require near jump. */
2243 		jg_bytes = 4;
2244 		EMIT2_off32(0x0F, X86_JG + 0x10, 0);
2245 	} else {
2246 		EMIT2(X86_JG, 0);
2247 	}
2248 	jg_reloc = prog;
2249 
2250 	err = emit_bpf_dispatcher(&prog, a, a + pivot,	/* emit lower_part */
2251 				  progs);
2252 	if (err)
2253 		return err;
2254 
2255 	/* From Intel 64 and IA-32 Architectures Optimization
2256 	 * Reference Manual, 3.4.1.4 Code Alignment, Assembly/Compiler
2257 	 * Coding Rule 11: All branch targets should be 16-byte
2258 	 * aligned.
2259 	 */
2260 	emit_align(&prog, 16);
2261 	jg_offset = prog - jg_reloc;
2262 	emit_code(jg_reloc - jg_bytes, jg_offset, jg_bytes);
2263 
2264 	err = emit_bpf_dispatcher(&prog, a + pivot + 1,	/* emit upper_part */
2265 				  b, progs);
2266 	if (err)
2267 		return err;
2268 
2269 	*pprog = prog;
2270 	return 0;
2271 }
2272 
2273 static int cmp_ips(const void *a, const void *b)
2274 {
2275 	const s64 *ipa = a;
2276 	const s64 *ipb = b;
2277 
2278 	if (*ipa > *ipb)
2279 		return 1;
2280 	if (*ipa < *ipb)
2281 		return -1;
2282 	return 0;
2283 }
2284 
2285 int arch_prepare_bpf_dispatcher(void *image, s64 *funcs, int num_funcs)
2286 {
2287 	u8 *prog = image;
2288 
2289 	sort(funcs, num_funcs, sizeof(funcs[0]), cmp_ips, NULL);
2290 	return emit_bpf_dispatcher(&prog, 0, num_funcs - 1, funcs);
2291 }
2292 
2293 struct x64_jit_data {
2294 	struct bpf_binary_header *rw_header;
2295 	struct bpf_binary_header *header;
2296 	int *addrs;
2297 	u8 *image;
2298 	int proglen;
2299 	struct jit_context ctx;
2300 };
2301 
2302 #define MAX_PASSES 20
2303 #define PADDING_PASSES (MAX_PASSES - 5)
2304 
2305 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
2306 {
2307 	struct bpf_binary_header *rw_header = NULL;
2308 	struct bpf_binary_header *header = NULL;
2309 	struct bpf_prog *tmp, *orig_prog = prog;
2310 	struct x64_jit_data *jit_data;
2311 	int proglen, oldproglen = 0;
2312 	struct jit_context ctx = {};
2313 	bool tmp_blinded = false;
2314 	bool extra_pass = false;
2315 	bool padding = false;
2316 	u8 *rw_image = NULL;
2317 	u8 *image = NULL;
2318 	int *addrs;
2319 	int pass;
2320 	int i;
2321 
2322 	if (!prog->jit_requested)
2323 		return orig_prog;
2324 
2325 	tmp = bpf_jit_blind_constants(prog);
2326 	/*
2327 	 * If blinding was requested and we failed during blinding,
2328 	 * we must fall back to the interpreter.
2329 	 */
2330 	if (IS_ERR(tmp))
2331 		return orig_prog;
2332 	if (tmp != prog) {
2333 		tmp_blinded = true;
2334 		prog = tmp;
2335 	}
2336 
2337 	jit_data = prog->aux->jit_data;
2338 	if (!jit_data) {
2339 		jit_data = kzalloc(sizeof(*jit_data), GFP_KERNEL);
2340 		if (!jit_data) {
2341 			prog = orig_prog;
2342 			goto out;
2343 		}
2344 		prog->aux->jit_data = jit_data;
2345 	}
2346 	addrs = jit_data->addrs;
2347 	if (addrs) {
2348 		ctx = jit_data->ctx;
2349 		oldproglen = jit_data->proglen;
2350 		image = jit_data->image;
2351 		header = jit_data->header;
2352 		rw_header = jit_data->rw_header;
2353 		rw_image = (void *)rw_header + ((void *)image - (void *)header);
2354 		extra_pass = true;
2355 		padding = true;
2356 		goto skip_init_addrs;
2357 	}
2358 	addrs = kvmalloc_array(prog->len + 1, sizeof(*addrs), GFP_KERNEL);
2359 	if (!addrs) {
2360 		prog = orig_prog;
2361 		goto out_addrs;
2362 	}
2363 
2364 	/*
2365 	 * Before first pass, make a rough estimation of addrs[]
2366 	 * each BPF instruction is translated to less than 64 bytes
2367 	 */
2368 	for (proglen = 0, i = 0; i <= prog->len; i++) {
2369 		proglen += 64;
2370 		addrs[i] = proglen;
2371 	}
2372 	ctx.cleanup_addr = proglen;
2373 skip_init_addrs:
2374 
2375 	/*
2376 	 * JITed image shrinks with every pass and the loop iterates
2377 	 * until the image stops shrinking. Very large BPF programs
2378 	 * may converge on the last pass. In such case do one more
2379 	 * pass to emit the final image.
2380 	 */
2381 	for (pass = 0; pass < MAX_PASSES || image; pass++) {
2382 		if (!padding && pass >= PADDING_PASSES)
2383 			padding = true;
2384 		proglen = do_jit(prog, addrs, image, rw_image, oldproglen, &ctx, padding);
2385 		if (proglen <= 0) {
2386 out_image:
2387 			image = NULL;
2388 			if (header) {
2389 				bpf_arch_text_copy(&header->size, &rw_header->size,
2390 						   sizeof(rw_header->size));
2391 				bpf_jit_binary_pack_free(header, rw_header);
2392 			}
2393 			/* Fall back to interpreter mode */
2394 			prog = orig_prog;
2395 			if (extra_pass) {
2396 				prog->bpf_func = NULL;
2397 				prog->jited = 0;
2398 				prog->jited_len = 0;
2399 			}
2400 			goto out_addrs;
2401 		}
2402 		if (image) {
2403 			if (proglen != oldproglen) {
2404 				pr_err("bpf_jit: proglen=%d != oldproglen=%d\n",
2405 				       proglen, oldproglen);
2406 				goto out_image;
2407 			}
2408 			break;
2409 		}
2410 		if (proglen == oldproglen) {
2411 			/*
2412 			 * The number of entries in extable is the number of BPF_LDX
2413 			 * insns that access kernel memory via "pointer to BTF type".
2414 			 * The verifier changed their opcode from LDX|MEM|size
2415 			 * to LDX|PROBE_MEM|size to make JITing easier.
2416 			 */
2417 			u32 align = __alignof__(struct exception_table_entry);
2418 			u32 extable_size = prog->aux->num_exentries *
2419 				sizeof(struct exception_table_entry);
2420 
2421 			/* allocate module memory for x86 insns and extable */
2422 			header = bpf_jit_binary_pack_alloc(roundup(proglen, align) + extable_size,
2423 							   &image, align, &rw_header, &rw_image,
2424 							   jit_fill_hole);
2425 			if (!header) {
2426 				prog = orig_prog;
2427 				goto out_addrs;
2428 			}
2429 			prog->aux->extable = (void *) image + roundup(proglen, align);
2430 		}
2431 		oldproglen = proglen;
2432 		cond_resched();
2433 	}
2434 
2435 	if (bpf_jit_enable > 1)
2436 		bpf_jit_dump(prog->len, proglen, pass + 1, image);
2437 
2438 	if (image) {
2439 		if (!prog->is_func || extra_pass) {
2440 			/*
2441 			 * bpf_jit_binary_pack_finalize fails in two scenarios:
2442 			 *   1) header is not pointing to proper module memory;
2443 			 *   2) the arch doesn't support bpf_arch_text_copy().
2444 			 *
2445 			 * Both cases are serious bugs and justify WARN_ON.
2446 			 */
2447 			if (WARN_ON(bpf_jit_binary_pack_finalize(prog, header, rw_header))) {
2448 				/* header has been freed */
2449 				header = NULL;
2450 				goto out_image;
2451 			}
2452 
2453 			bpf_tail_call_direct_fixup(prog);
2454 		} else {
2455 			jit_data->addrs = addrs;
2456 			jit_data->ctx = ctx;
2457 			jit_data->proglen = proglen;
2458 			jit_data->image = image;
2459 			jit_data->header = header;
2460 			jit_data->rw_header = rw_header;
2461 		}
2462 		prog->bpf_func = (void *)image;
2463 		prog->jited = 1;
2464 		prog->jited_len = proglen;
2465 	} else {
2466 		prog = orig_prog;
2467 	}
2468 
2469 	if (!image || !prog->is_func || extra_pass) {
2470 		if (image)
2471 			bpf_prog_fill_jited_linfo(prog, addrs + 1);
2472 out_addrs:
2473 		kvfree(addrs);
2474 		kfree(jit_data);
2475 		prog->aux->jit_data = NULL;
2476 	}
2477 out:
2478 	if (tmp_blinded)
2479 		bpf_jit_prog_release_other(prog, prog == orig_prog ?
2480 					   tmp : orig_prog);
2481 	return prog;
2482 }
2483 
2484 bool bpf_jit_supports_kfunc_call(void)
2485 {
2486 	return true;
2487 }
2488 
2489 void *bpf_arch_text_copy(void *dst, void *src, size_t len)
2490 {
2491 	if (text_poke_copy(dst, src, len) == NULL)
2492 		return ERR_PTR(-EINVAL);
2493 	return dst;
2494 }
2495