xref: /openbmc/linux/arch/arm64/net/bpf_jit_comp.c (revision 43ffe817)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * BPF JIT compiler for ARM64
4  *
5  * Copyright (C) 2014-2016 Zi Shen Lim <zlim.lnx@gmail.com>
6  */
7 
8 #define pr_fmt(fmt) "bpf_jit: " fmt
9 
10 #include <linux/bitfield.h>
11 #include <linux/bpf.h>
12 #include <linux/filter.h>
13 #include <linux/printk.h>
14 #include <linux/slab.h>
15 
16 #include <asm/byteorder.h>
17 #include <asm/cacheflush.h>
18 #include <asm/debug-monitors.h>
19 #include <asm/set_memory.h>
20 
21 #include "bpf_jit.h"
22 
23 #define TMP_REG_1 (MAX_BPF_JIT_REG + 0)
24 #define TMP_REG_2 (MAX_BPF_JIT_REG + 1)
25 #define TCALL_CNT (MAX_BPF_JIT_REG + 2)
26 #define TMP_REG_3 (MAX_BPF_JIT_REG + 3)
27 
28 /* Map BPF registers to A64 registers */
29 static const int bpf2a64[] = {
30 	/* return value from in-kernel function, and exit value from eBPF */
31 	[BPF_REG_0] = A64_R(7),
32 	/* arguments from eBPF program to in-kernel function */
33 	[BPF_REG_1] = A64_R(0),
34 	[BPF_REG_2] = A64_R(1),
35 	[BPF_REG_3] = A64_R(2),
36 	[BPF_REG_4] = A64_R(3),
37 	[BPF_REG_5] = A64_R(4),
38 	/* callee saved registers that in-kernel function will preserve */
39 	[BPF_REG_6] = A64_R(19),
40 	[BPF_REG_7] = A64_R(20),
41 	[BPF_REG_8] = A64_R(21),
42 	[BPF_REG_9] = A64_R(22),
43 	/* read-only frame pointer to access stack */
44 	[BPF_REG_FP] = A64_R(25),
45 	/* temporary registers for internal BPF JIT */
46 	[TMP_REG_1] = A64_R(10),
47 	[TMP_REG_2] = A64_R(11),
48 	[TMP_REG_3] = A64_R(12),
49 	/* tail_call_cnt */
50 	[TCALL_CNT] = A64_R(26),
51 	/* temporary register for blinding constants */
52 	[BPF_REG_AX] = A64_R(9),
53 };
54 
55 struct jit_ctx {
56 	const struct bpf_prog *prog;
57 	int idx;
58 	int epilogue_offset;
59 	int *offset;
60 	int exentry_idx;
61 	__le32 *image;
62 	u32 stack_size;
63 };
64 
65 static inline void emit(const u32 insn, struct jit_ctx *ctx)
66 {
67 	if (ctx->image != NULL)
68 		ctx->image[ctx->idx] = cpu_to_le32(insn);
69 
70 	ctx->idx++;
71 }
72 
73 static inline void emit_a64_mov_i(const int is64, const int reg,
74 				  const s32 val, struct jit_ctx *ctx)
75 {
76 	u16 hi = val >> 16;
77 	u16 lo = val & 0xffff;
78 
79 	if (hi & 0x8000) {
80 		if (hi == 0xffff) {
81 			emit(A64_MOVN(is64, reg, (u16)~lo, 0), ctx);
82 		} else {
83 			emit(A64_MOVN(is64, reg, (u16)~hi, 16), ctx);
84 			if (lo != 0xffff)
85 				emit(A64_MOVK(is64, reg, lo, 0), ctx);
86 		}
87 	} else {
88 		emit(A64_MOVZ(is64, reg, lo, 0), ctx);
89 		if (hi)
90 			emit(A64_MOVK(is64, reg, hi, 16), ctx);
91 	}
92 }
93 
94 static int i64_i16_blocks(const u64 val, bool inverse)
95 {
96 	return (((val >>  0) & 0xffff) != (inverse ? 0xffff : 0x0000)) +
97 	       (((val >> 16) & 0xffff) != (inverse ? 0xffff : 0x0000)) +
98 	       (((val >> 32) & 0xffff) != (inverse ? 0xffff : 0x0000)) +
99 	       (((val >> 48) & 0xffff) != (inverse ? 0xffff : 0x0000));
100 }
101 
102 static inline void emit_a64_mov_i64(const int reg, const u64 val,
103 				    struct jit_ctx *ctx)
104 {
105 	u64 nrm_tmp = val, rev_tmp = ~val;
106 	bool inverse;
107 	int shift;
108 
109 	if (!(nrm_tmp >> 32))
110 		return emit_a64_mov_i(0, reg, (u32)val, ctx);
111 
112 	inverse = i64_i16_blocks(nrm_tmp, true) < i64_i16_blocks(nrm_tmp, false);
113 	shift = max(round_down((inverse ? (fls64(rev_tmp) - 1) :
114 					  (fls64(nrm_tmp) - 1)), 16), 0);
115 	if (inverse)
116 		emit(A64_MOVN(1, reg, (rev_tmp >> shift) & 0xffff, shift), ctx);
117 	else
118 		emit(A64_MOVZ(1, reg, (nrm_tmp >> shift) & 0xffff, shift), ctx);
119 	shift -= 16;
120 	while (shift >= 0) {
121 		if (((nrm_tmp >> shift) & 0xffff) != (inverse ? 0xffff : 0x0000))
122 			emit(A64_MOVK(1, reg, (nrm_tmp >> shift) & 0xffff, shift), ctx);
123 		shift -= 16;
124 	}
125 }
126 
127 /*
128  * Kernel addresses in the vmalloc space use at most 48 bits, and the
129  * remaining bits are guaranteed to be 0x1. So we can compose the address
130  * with a fixed length movn/movk/movk sequence.
131  */
132 static inline void emit_addr_mov_i64(const int reg, const u64 val,
133 				     struct jit_ctx *ctx)
134 {
135 	u64 tmp = val;
136 	int shift = 0;
137 
138 	emit(A64_MOVN(1, reg, ~tmp & 0xffff, shift), ctx);
139 	while (shift < 32) {
140 		tmp >>= 16;
141 		shift += 16;
142 		emit(A64_MOVK(1, reg, tmp & 0xffff, shift), ctx);
143 	}
144 }
145 
146 static inline int bpf2a64_offset(int bpf_insn, int off,
147 				 const struct jit_ctx *ctx)
148 {
149 	/* BPF JMP offset is relative to the next instruction */
150 	bpf_insn++;
151 	/*
152 	 * Whereas arm64 branch instructions encode the offset
153 	 * from the branch itself, so we must subtract 1 from the
154 	 * instruction offset.
155 	 */
156 	return ctx->offset[bpf_insn + off] - (ctx->offset[bpf_insn] - 1);
157 }
158 
159 static void jit_fill_hole(void *area, unsigned int size)
160 {
161 	__le32 *ptr;
162 	/* We are guaranteed to have aligned memory. */
163 	for (ptr = area; size >= sizeof(u32); size -= sizeof(u32))
164 		*ptr++ = cpu_to_le32(AARCH64_BREAK_FAULT);
165 }
166 
167 static inline int epilogue_offset(const struct jit_ctx *ctx)
168 {
169 	int to = ctx->epilogue_offset;
170 	int from = ctx->idx;
171 
172 	return to - from;
173 }
174 
175 static bool is_addsub_imm(u32 imm)
176 {
177 	/* Either imm12 or shifted imm12. */
178 	return !(imm & ~0xfff) || !(imm & ~0xfff000);
179 }
180 
181 /* Stack must be multiples of 16B */
182 #define STACK_ALIGN(sz) (((sz) + 15) & ~15)
183 
184 /* Tail call offset to jump into */
185 #if IS_ENABLED(CONFIG_ARM64_BTI_KERNEL)
186 #define PROLOGUE_OFFSET 8
187 #else
188 #define PROLOGUE_OFFSET 7
189 #endif
190 
191 static int build_prologue(struct jit_ctx *ctx, bool ebpf_from_cbpf)
192 {
193 	const struct bpf_prog *prog = ctx->prog;
194 	const u8 r6 = bpf2a64[BPF_REG_6];
195 	const u8 r7 = bpf2a64[BPF_REG_7];
196 	const u8 r8 = bpf2a64[BPF_REG_8];
197 	const u8 r9 = bpf2a64[BPF_REG_9];
198 	const u8 fp = bpf2a64[BPF_REG_FP];
199 	const u8 tcc = bpf2a64[TCALL_CNT];
200 	const int idx0 = ctx->idx;
201 	int cur_offset;
202 
203 	/*
204 	 * BPF prog stack layout
205 	 *
206 	 *                         high
207 	 * original A64_SP =>   0:+-----+ BPF prologue
208 	 *                        |FP/LR|
209 	 * current A64_FP =>  -16:+-----+
210 	 *                        | ... | callee saved registers
211 	 * BPF fp register => -64:+-----+ <= (BPF_FP)
212 	 *                        |     |
213 	 *                        | ... | BPF prog stack
214 	 *                        |     |
215 	 *                        +-----+ <= (BPF_FP - prog->aux->stack_depth)
216 	 *                        |RSVD | padding
217 	 * current A64_SP =>      +-----+ <= (BPF_FP - ctx->stack_size)
218 	 *                        |     |
219 	 *                        | ... | Function call stack
220 	 *                        |     |
221 	 *                        +-----+
222 	 *                          low
223 	 *
224 	 */
225 
226 	/* BTI landing pad */
227 	if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
228 		emit(A64_BTI_C, ctx);
229 
230 	/* Save FP and LR registers to stay align with ARM64 AAPCS */
231 	emit(A64_PUSH(A64_FP, A64_LR, A64_SP), ctx);
232 	emit(A64_MOV(1, A64_FP, A64_SP), ctx);
233 
234 	/* Save callee-saved registers */
235 	emit(A64_PUSH(r6, r7, A64_SP), ctx);
236 	emit(A64_PUSH(r8, r9, A64_SP), ctx);
237 	emit(A64_PUSH(fp, tcc, A64_SP), ctx);
238 
239 	/* Set up BPF prog stack base register */
240 	emit(A64_MOV(1, fp, A64_SP), ctx);
241 
242 	if (!ebpf_from_cbpf) {
243 		/* Initialize tail_call_cnt */
244 		emit(A64_MOVZ(1, tcc, 0, 0), ctx);
245 
246 		cur_offset = ctx->idx - idx0;
247 		if (cur_offset != PROLOGUE_OFFSET) {
248 			pr_err_once("PROLOGUE_OFFSET = %d, expected %d!\n",
249 				    cur_offset, PROLOGUE_OFFSET);
250 			return -1;
251 		}
252 
253 		/* BTI landing pad for the tail call, done with a BR */
254 		if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
255 			emit(A64_BTI_J, ctx);
256 	}
257 
258 	ctx->stack_size = STACK_ALIGN(prog->aux->stack_depth);
259 
260 	/* Set up function call stack */
261 	emit(A64_SUB_I(1, A64_SP, A64_SP, ctx->stack_size), ctx);
262 	return 0;
263 }
264 
265 static int out_offset = -1; /* initialized on the first pass of build_body() */
266 static int emit_bpf_tail_call(struct jit_ctx *ctx)
267 {
268 	/* bpf_tail_call(void *prog_ctx, struct bpf_array *array, u64 index) */
269 	const u8 r2 = bpf2a64[BPF_REG_2];
270 	const u8 r3 = bpf2a64[BPF_REG_3];
271 
272 	const u8 tmp = bpf2a64[TMP_REG_1];
273 	const u8 prg = bpf2a64[TMP_REG_2];
274 	const u8 tcc = bpf2a64[TCALL_CNT];
275 	const int idx0 = ctx->idx;
276 #define cur_offset (ctx->idx - idx0)
277 #define jmp_offset (out_offset - (cur_offset))
278 	size_t off;
279 
280 	/* if (index >= array->map.max_entries)
281 	 *     goto out;
282 	 */
283 	off = offsetof(struct bpf_array, map.max_entries);
284 	emit_a64_mov_i64(tmp, off, ctx);
285 	emit(A64_LDR32(tmp, r2, tmp), ctx);
286 	emit(A64_MOV(0, r3, r3), ctx);
287 	emit(A64_CMP(0, r3, tmp), ctx);
288 	emit(A64_B_(A64_COND_CS, jmp_offset), ctx);
289 
290 	/* if (tail_call_cnt > MAX_TAIL_CALL_CNT)
291 	 *     goto out;
292 	 * tail_call_cnt++;
293 	 */
294 	emit_a64_mov_i64(tmp, MAX_TAIL_CALL_CNT, ctx);
295 	emit(A64_CMP(1, tcc, tmp), ctx);
296 	emit(A64_B_(A64_COND_HI, jmp_offset), ctx);
297 	emit(A64_ADD_I(1, tcc, tcc, 1), ctx);
298 
299 	/* prog = array->ptrs[index];
300 	 * if (prog == NULL)
301 	 *     goto out;
302 	 */
303 	off = offsetof(struct bpf_array, ptrs);
304 	emit_a64_mov_i64(tmp, off, ctx);
305 	emit(A64_ADD(1, tmp, r2, tmp), ctx);
306 	emit(A64_LSL(1, prg, r3, 3), ctx);
307 	emit(A64_LDR64(prg, tmp, prg), ctx);
308 	emit(A64_CBZ(1, prg, jmp_offset), ctx);
309 
310 	/* goto *(prog->bpf_func + prologue_offset); */
311 	off = offsetof(struct bpf_prog, bpf_func);
312 	emit_a64_mov_i64(tmp, off, ctx);
313 	emit(A64_LDR64(tmp, prg, tmp), ctx);
314 	emit(A64_ADD_I(1, tmp, tmp, sizeof(u32) * PROLOGUE_OFFSET), ctx);
315 	emit(A64_ADD_I(1, A64_SP, A64_SP, ctx->stack_size), ctx);
316 	emit(A64_BR(tmp), ctx);
317 
318 	/* out: */
319 	if (out_offset == -1)
320 		out_offset = cur_offset;
321 	if (cur_offset != out_offset) {
322 		pr_err_once("tail_call out_offset = %d, expected %d!\n",
323 			    cur_offset, out_offset);
324 		return -1;
325 	}
326 	return 0;
327 #undef cur_offset
328 #undef jmp_offset
329 }
330 
331 static void build_epilogue(struct jit_ctx *ctx)
332 {
333 	const u8 r0 = bpf2a64[BPF_REG_0];
334 	const u8 r6 = bpf2a64[BPF_REG_6];
335 	const u8 r7 = bpf2a64[BPF_REG_7];
336 	const u8 r8 = bpf2a64[BPF_REG_8];
337 	const u8 r9 = bpf2a64[BPF_REG_9];
338 	const u8 fp = bpf2a64[BPF_REG_FP];
339 
340 	/* We're done with BPF stack */
341 	emit(A64_ADD_I(1, A64_SP, A64_SP, ctx->stack_size), ctx);
342 
343 	/* Restore fs (x25) and x26 */
344 	emit(A64_POP(fp, A64_R(26), A64_SP), ctx);
345 
346 	/* Restore callee-saved register */
347 	emit(A64_POP(r8, r9, A64_SP), ctx);
348 	emit(A64_POP(r6, r7, A64_SP), ctx);
349 
350 	/* Restore FP/LR registers */
351 	emit(A64_POP(A64_FP, A64_LR, A64_SP), ctx);
352 
353 	/* Set return value */
354 	emit(A64_MOV(1, A64_R(0), r0), ctx);
355 
356 	emit(A64_RET(A64_LR), ctx);
357 }
358 
359 #define BPF_FIXUP_OFFSET_MASK	GENMASK(26, 0)
360 #define BPF_FIXUP_REG_MASK	GENMASK(31, 27)
361 
362 int arm64_bpf_fixup_exception(const struct exception_table_entry *ex,
363 			      struct pt_regs *regs)
364 {
365 	off_t offset = FIELD_GET(BPF_FIXUP_OFFSET_MASK, ex->fixup);
366 	int dst_reg = FIELD_GET(BPF_FIXUP_REG_MASK, ex->fixup);
367 
368 	regs->regs[dst_reg] = 0;
369 	regs->pc = (unsigned long)&ex->fixup - offset;
370 	return 1;
371 }
372 
373 /* For accesses to BTF pointers, add an entry to the exception table */
374 static int add_exception_handler(const struct bpf_insn *insn,
375 				 struct jit_ctx *ctx,
376 				 int dst_reg)
377 {
378 	off_t offset;
379 	unsigned long pc;
380 	struct exception_table_entry *ex;
381 
382 	if (!ctx->image)
383 		/* First pass */
384 		return 0;
385 
386 	if (BPF_MODE(insn->code) != BPF_PROBE_MEM)
387 		return 0;
388 
389 	if (!ctx->prog->aux->extable ||
390 	    WARN_ON_ONCE(ctx->exentry_idx >= ctx->prog->aux->num_exentries))
391 		return -EINVAL;
392 
393 	ex = &ctx->prog->aux->extable[ctx->exentry_idx];
394 	pc = (unsigned long)&ctx->image[ctx->idx - 1];
395 
396 	offset = pc - (long)&ex->insn;
397 	if (WARN_ON_ONCE(offset >= 0 || offset < INT_MIN))
398 		return -ERANGE;
399 	ex->insn = offset;
400 
401 	/*
402 	 * Since the extable follows the program, the fixup offset is always
403 	 * negative and limited to BPF_JIT_REGION_SIZE. Store a positive value
404 	 * to keep things simple, and put the destination register in the upper
405 	 * bits. We don't need to worry about buildtime or runtime sort
406 	 * modifying the upper bits because the table is already sorted, and
407 	 * isn't part of the main exception table.
408 	 */
409 	offset = (long)&ex->fixup - (pc + AARCH64_INSN_SIZE);
410 	if (!FIELD_FIT(BPF_FIXUP_OFFSET_MASK, offset))
411 		return -ERANGE;
412 
413 	ex->fixup = FIELD_PREP(BPF_FIXUP_OFFSET_MASK, offset) |
414 		    FIELD_PREP(BPF_FIXUP_REG_MASK, dst_reg);
415 
416 	ctx->exentry_idx++;
417 	return 0;
418 }
419 
420 /* JITs an eBPF instruction.
421  * Returns:
422  * 0  - successfully JITed an 8-byte eBPF instruction.
423  * >0 - successfully JITed a 16-byte eBPF instruction.
424  * <0 - failed to JIT.
425  */
426 static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
427 		      bool extra_pass)
428 {
429 	const u8 code = insn->code;
430 	const u8 dst = bpf2a64[insn->dst_reg];
431 	const u8 src = bpf2a64[insn->src_reg];
432 	const u8 tmp = bpf2a64[TMP_REG_1];
433 	const u8 tmp2 = bpf2a64[TMP_REG_2];
434 	const u8 tmp3 = bpf2a64[TMP_REG_3];
435 	const s16 off = insn->off;
436 	const s32 imm = insn->imm;
437 	const int i = insn - ctx->prog->insnsi;
438 	const bool is64 = BPF_CLASS(code) == BPF_ALU64 ||
439 			  BPF_CLASS(code) == BPF_JMP;
440 	const bool isdw = BPF_SIZE(code) == BPF_DW;
441 	u8 jmp_cond, reg;
442 	s32 jmp_offset;
443 	u32 a64_insn;
444 	int ret;
445 
446 #define check_imm(bits, imm) do {				\
447 	if ((((imm) > 0) && ((imm) >> (bits))) ||		\
448 	    (((imm) < 0) && (~(imm) >> (bits)))) {		\
449 		pr_info("[%2d] imm=%d(0x%x) out of range\n",	\
450 			i, imm, imm);				\
451 		return -EINVAL;					\
452 	}							\
453 } while (0)
454 #define check_imm19(imm) check_imm(19, imm)
455 #define check_imm26(imm) check_imm(26, imm)
456 
457 	switch (code) {
458 	/* dst = src */
459 	case BPF_ALU | BPF_MOV | BPF_X:
460 	case BPF_ALU64 | BPF_MOV | BPF_X:
461 		emit(A64_MOV(is64, dst, src), ctx);
462 		break;
463 	/* dst = dst OP src */
464 	case BPF_ALU | BPF_ADD | BPF_X:
465 	case BPF_ALU64 | BPF_ADD | BPF_X:
466 		emit(A64_ADD(is64, dst, dst, src), ctx);
467 		break;
468 	case BPF_ALU | BPF_SUB | BPF_X:
469 	case BPF_ALU64 | BPF_SUB | BPF_X:
470 		emit(A64_SUB(is64, dst, dst, src), ctx);
471 		break;
472 	case BPF_ALU | BPF_AND | BPF_X:
473 	case BPF_ALU64 | BPF_AND | BPF_X:
474 		emit(A64_AND(is64, dst, dst, src), ctx);
475 		break;
476 	case BPF_ALU | BPF_OR | BPF_X:
477 	case BPF_ALU64 | BPF_OR | BPF_X:
478 		emit(A64_ORR(is64, dst, dst, src), ctx);
479 		break;
480 	case BPF_ALU | BPF_XOR | BPF_X:
481 	case BPF_ALU64 | BPF_XOR | BPF_X:
482 		emit(A64_EOR(is64, dst, dst, src), ctx);
483 		break;
484 	case BPF_ALU | BPF_MUL | BPF_X:
485 	case BPF_ALU64 | BPF_MUL | BPF_X:
486 		emit(A64_MUL(is64, dst, dst, src), ctx);
487 		break;
488 	case BPF_ALU | BPF_DIV | BPF_X:
489 	case BPF_ALU64 | BPF_DIV | BPF_X:
490 	case BPF_ALU | BPF_MOD | BPF_X:
491 	case BPF_ALU64 | BPF_MOD | BPF_X:
492 		switch (BPF_OP(code)) {
493 		case BPF_DIV:
494 			emit(A64_UDIV(is64, dst, dst, src), ctx);
495 			break;
496 		case BPF_MOD:
497 			emit(A64_UDIV(is64, tmp, dst, src), ctx);
498 			emit(A64_MSUB(is64, dst, dst, tmp, src), ctx);
499 			break;
500 		}
501 		break;
502 	case BPF_ALU | BPF_LSH | BPF_X:
503 	case BPF_ALU64 | BPF_LSH | BPF_X:
504 		emit(A64_LSLV(is64, dst, dst, src), ctx);
505 		break;
506 	case BPF_ALU | BPF_RSH | BPF_X:
507 	case BPF_ALU64 | BPF_RSH | BPF_X:
508 		emit(A64_LSRV(is64, dst, dst, src), ctx);
509 		break;
510 	case BPF_ALU | BPF_ARSH | BPF_X:
511 	case BPF_ALU64 | BPF_ARSH | BPF_X:
512 		emit(A64_ASRV(is64, dst, dst, src), ctx);
513 		break;
514 	/* dst = -dst */
515 	case BPF_ALU | BPF_NEG:
516 	case BPF_ALU64 | BPF_NEG:
517 		emit(A64_NEG(is64, dst, dst), ctx);
518 		break;
519 	/* dst = BSWAP##imm(dst) */
520 	case BPF_ALU | BPF_END | BPF_FROM_LE:
521 	case BPF_ALU | BPF_END | BPF_FROM_BE:
522 #ifdef CONFIG_CPU_BIG_ENDIAN
523 		if (BPF_SRC(code) == BPF_FROM_BE)
524 			goto emit_bswap_uxt;
525 #else /* !CONFIG_CPU_BIG_ENDIAN */
526 		if (BPF_SRC(code) == BPF_FROM_LE)
527 			goto emit_bswap_uxt;
528 #endif
529 		switch (imm) {
530 		case 16:
531 			emit(A64_REV16(is64, dst, dst), ctx);
532 			/* zero-extend 16 bits into 64 bits */
533 			emit(A64_UXTH(is64, dst, dst), ctx);
534 			break;
535 		case 32:
536 			emit(A64_REV32(is64, dst, dst), ctx);
537 			/* upper 32 bits already cleared */
538 			break;
539 		case 64:
540 			emit(A64_REV64(dst, dst), ctx);
541 			break;
542 		}
543 		break;
544 emit_bswap_uxt:
545 		switch (imm) {
546 		case 16:
547 			/* zero-extend 16 bits into 64 bits */
548 			emit(A64_UXTH(is64, dst, dst), ctx);
549 			break;
550 		case 32:
551 			/* zero-extend 32 bits into 64 bits */
552 			emit(A64_UXTW(is64, dst, dst), ctx);
553 			break;
554 		case 64:
555 			/* nop */
556 			break;
557 		}
558 		break;
559 	/* dst = imm */
560 	case BPF_ALU | BPF_MOV | BPF_K:
561 	case BPF_ALU64 | BPF_MOV | BPF_K:
562 		emit_a64_mov_i(is64, dst, imm, ctx);
563 		break;
564 	/* dst = dst OP imm */
565 	case BPF_ALU | BPF_ADD | BPF_K:
566 	case BPF_ALU64 | BPF_ADD | BPF_K:
567 		if (is_addsub_imm(imm)) {
568 			emit(A64_ADD_I(is64, dst, dst, imm), ctx);
569 		} else if (is_addsub_imm(-imm)) {
570 			emit(A64_SUB_I(is64, dst, dst, -imm), ctx);
571 		} else {
572 			emit_a64_mov_i(is64, tmp, imm, ctx);
573 			emit(A64_ADD(is64, dst, dst, tmp), ctx);
574 		}
575 		break;
576 	case BPF_ALU | BPF_SUB | BPF_K:
577 	case BPF_ALU64 | BPF_SUB | BPF_K:
578 		if (is_addsub_imm(imm)) {
579 			emit(A64_SUB_I(is64, dst, dst, imm), ctx);
580 		} else if (is_addsub_imm(-imm)) {
581 			emit(A64_ADD_I(is64, dst, dst, -imm), ctx);
582 		} else {
583 			emit_a64_mov_i(is64, tmp, imm, ctx);
584 			emit(A64_SUB(is64, dst, dst, tmp), ctx);
585 		}
586 		break;
587 	case BPF_ALU | BPF_AND | BPF_K:
588 	case BPF_ALU64 | BPF_AND | BPF_K:
589 		a64_insn = A64_AND_I(is64, dst, dst, imm);
590 		if (a64_insn != AARCH64_BREAK_FAULT) {
591 			emit(a64_insn, ctx);
592 		} else {
593 			emit_a64_mov_i(is64, tmp, imm, ctx);
594 			emit(A64_AND(is64, dst, dst, tmp), ctx);
595 		}
596 		break;
597 	case BPF_ALU | BPF_OR | BPF_K:
598 	case BPF_ALU64 | BPF_OR | BPF_K:
599 		a64_insn = A64_ORR_I(is64, dst, dst, imm);
600 		if (a64_insn != AARCH64_BREAK_FAULT) {
601 			emit(a64_insn, ctx);
602 		} else {
603 			emit_a64_mov_i(is64, tmp, imm, ctx);
604 			emit(A64_ORR(is64, dst, dst, tmp), ctx);
605 		}
606 		break;
607 	case BPF_ALU | BPF_XOR | BPF_K:
608 	case BPF_ALU64 | BPF_XOR | BPF_K:
609 		a64_insn = A64_EOR_I(is64, dst, dst, imm);
610 		if (a64_insn != AARCH64_BREAK_FAULT) {
611 			emit(a64_insn, ctx);
612 		} else {
613 			emit_a64_mov_i(is64, tmp, imm, ctx);
614 			emit(A64_EOR(is64, dst, dst, tmp), ctx);
615 		}
616 		break;
617 	case BPF_ALU | BPF_MUL | BPF_K:
618 	case BPF_ALU64 | BPF_MUL | BPF_K:
619 		emit_a64_mov_i(is64, tmp, imm, ctx);
620 		emit(A64_MUL(is64, dst, dst, tmp), ctx);
621 		break;
622 	case BPF_ALU | BPF_DIV | BPF_K:
623 	case BPF_ALU64 | BPF_DIV | BPF_K:
624 		emit_a64_mov_i(is64, tmp, imm, ctx);
625 		emit(A64_UDIV(is64, dst, dst, tmp), ctx);
626 		break;
627 	case BPF_ALU | BPF_MOD | BPF_K:
628 	case BPF_ALU64 | BPF_MOD | BPF_K:
629 		emit_a64_mov_i(is64, tmp2, imm, ctx);
630 		emit(A64_UDIV(is64, tmp, dst, tmp2), ctx);
631 		emit(A64_MSUB(is64, dst, dst, tmp, tmp2), ctx);
632 		break;
633 	case BPF_ALU | BPF_LSH | BPF_K:
634 	case BPF_ALU64 | BPF_LSH | BPF_K:
635 		emit(A64_LSL(is64, dst, dst, imm), ctx);
636 		break;
637 	case BPF_ALU | BPF_RSH | BPF_K:
638 	case BPF_ALU64 | BPF_RSH | BPF_K:
639 		emit(A64_LSR(is64, dst, dst, imm), ctx);
640 		break;
641 	case BPF_ALU | BPF_ARSH | BPF_K:
642 	case BPF_ALU64 | BPF_ARSH | BPF_K:
643 		emit(A64_ASR(is64, dst, dst, imm), ctx);
644 		break;
645 
646 	/* JUMP off */
647 	case BPF_JMP | BPF_JA:
648 		jmp_offset = bpf2a64_offset(i, off, ctx);
649 		check_imm26(jmp_offset);
650 		emit(A64_B(jmp_offset), ctx);
651 		break;
652 	/* IF (dst COND src) JUMP off */
653 	case BPF_JMP | BPF_JEQ | BPF_X:
654 	case BPF_JMP | BPF_JGT | BPF_X:
655 	case BPF_JMP | BPF_JLT | BPF_X:
656 	case BPF_JMP | BPF_JGE | BPF_X:
657 	case BPF_JMP | BPF_JLE | BPF_X:
658 	case BPF_JMP | BPF_JNE | BPF_X:
659 	case BPF_JMP | BPF_JSGT | BPF_X:
660 	case BPF_JMP | BPF_JSLT | BPF_X:
661 	case BPF_JMP | BPF_JSGE | BPF_X:
662 	case BPF_JMP | BPF_JSLE | BPF_X:
663 	case BPF_JMP32 | BPF_JEQ | BPF_X:
664 	case BPF_JMP32 | BPF_JGT | BPF_X:
665 	case BPF_JMP32 | BPF_JLT | BPF_X:
666 	case BPF_JMP32 | BPF_JGE | BPF_X:
667 	case BPF_JMP32 | BPF_JLE | BPF_X:
668 	case BPF_JMP32 | BPF_JNE | BPF_X:
669 	case BPF_JMP32 | BPF_JSGT | BPF_X:
670 	case BPF_JMP32 | BPF_JSLT | BPF_X:
671 	case BPF_JMP32 | BPF_JSGE | BPF_X:
672 	case BPF_JMP32 | BPF_JSLE | BPF_X:
673 		emit(A64_CMP(is64, dst, src), ctx);
674 emit_cond_jmp:
675 		jmp_offset = bpf2a64_offset(i, off, ctx);
676 		check_imm19(jmp_offset);
677 		switch (BPF_OP(code)) {
678 		case BPF_JEQ:
679 			jmp_cond = A64_COND_EQ;
680 			break;
681 		case BPF_JGT:
682 			jmp_cond = A64_COND_HI;
683 			break;
684 		case BPF_JLT:
685 			jmp_cond = A64_COND_CC;
686 			break;
687 		case BPF_JGE:
688 			jmp_cond = A64_COND_CS;
689 			break;
690 		case BPF_JLE:
691 			jmp_cond = A64_COND_LS;
692 			break;
693 		case BPF_JSET:
694 		case BPF_JNE:
695 			jmp_cond = A64_COND_NE;
696 			break;
697 		case BPF_JSGT:
698 			jmp_cond = A64_COND_GT;
699 			break;
700 		case BPF_JSLT:
701 			jmp_cond = A64_COND_LT;
702 			break;
703 		case BPF_JSGE:
704 			jmp_cond = A64_COND_GE;
705 			break;
706 		case BPF_JSLE:
707 			jmp_cond = A64_COND_LE;
708 			break;
709 		default:
710 			return -EFAULT;
711 		}
712 		emit(A64_B_(jmp_cond, jmp_offset), ctx);
713 		break;
714 	case BPF_JMP | BPF_JSET | BPF_X:
715 	case BPF_JMP32 | BPF_JSET | BPF_X:
716 		emit(A64_TST(is64, dst, src), ctx);
717 		goto emit_cond_jmp;
718 	/* IF (dst COND imm) JUMP off */
719 	case BPF_JMP | BPF_JEQ | BPF_K:
720 	case BPF_JMP | BPF_JGT | BPF_K:
721 	case BPF_JMP | BPF_JLT | BPF_K:
722 	case BPF_JMP | BPF_JGE | BPF_K:
723 	case BPF_JMP | BPF_JLE | BPF_K:
724 	case BPF_JMP | BPF_JNE | BPF_K:
725 	case BPF_JMP | BPF_JSGT | BPF_K:
726 	case BPF_JMP | BPF_JSLT | BPF_K:
727 	case BPF_JMP | BPF_JSGE | BPF_K:
728 	case BPF_JMP | BPF_JSLE | BPF_K:
729 	case BPF_JMP32 | BPF_JEQ | BPF_K:
730 	case BPF_JMP32 | BPF_JGT | BPF_K:
731 	case BPF_JMP32 | BPF_JLT | BPF_K:
732 	case BPF_JMP32 | BPF_JGE | BPF_K:
733 	case BPF_JMP32 | BPF_JLE | BPF_K:
734 	case BPF_JMP32 | BPF_JNE | BPF_K:
735 	case BPF_JMP32 | BPF_JSGT | BPF_K:
736 	case BPF_JMP32 | BPF_JSLT | BPF_K:
737 	case BPF_JMP32 | BPF_JSGE | BPF_K:
738 	case BPF_JMP32 | BPF_JSLE | BPF_K:
739 		if (is_addsub_imm(imm)) {
740 			emit(A64_CMP_I(is64, dst, imm), ctx);
741 		} else if (is_addsub_imm(-imm)) {
742 			emit(A64_CMN_I(is64, dst, -imm), ctx);
743 		} else {
744 			emit_a64_mov_i(is64, tmp, imm, ctx);
745 			emit(A64_CMP(is64, dst, tmp), ctx);
746 		}
747 		goto emit_cond_jmp;
748 	case BPF_JMP | BPF_JSET | BPF_K:
749 	case BPF_JMP32 | BPF_JSET | BPF_K:
750 		a64_insn = A64_TST_I(is64, dst, imm);
751 		if (a64_insn != AARCH64_BREAK_FAULT) {
752 			emit(a64_insn, ctx);
753 		} else {
754 			emit_a64_mov_i(is64, tmp, imm, ctx);
755 			emit(A64_TST(is64, dst, tmp), ctx);
756 		}
757 		goto emit_cond_jmp;
758 	/* function call */
759 	case BPF_JMP | BPF_CALL:
760 	{
761 		const u8 r0 = bpf2a64[BPF_REG_0];
762 		bool func_addr_fixed;
763 		u64 func_addr;
764 
765 		ret = bpf_jit_get_func_addr(ctx->prog, insn, extra_pass,
766 					    &func_addr, &func_addr_fixed);
767 		if (ret < 0)
768 			return ret;
769 		emit_addr_mov_i64(tmp, func_addr, ctx);
770 		emit(A64_BLR(tmp), ctx);
771 		emit(A64_MOV(1, r0, A64_R(0)), ctx);
772 		break;
773 	}
774 	/* tail call */
775 	case BPF_JMP | BPF_TAIL_CALL:
776 		if (emit_bpf_tail_call(ctx))
777 			return -EFAULT;
778 		break;
779 	/* function return */
780 	case BPF_JMP | BPF_EXIT:
781 		/* Optimization: when last instruction is EXIT,
782 		   simply fallthrough to epilogue. */
783 		if (i == ctx->prog->len - 1)
784 			break;
785 		jmp_offset = epilogue_offset(ctx);
786 		check_imm26(jmp_offset);
787 		emit(A64_B(jmp_offset), ctx);
788 		break;
789 
790 	/* dst = imm64 */
791 	case BPF_LD | BPF_IMM | BPF_DW:
792 	{
793 		const struct bpf_insn insn1 = insn[1];
794 		u64 imm64;
795 
796 		imm64 = (u64)insn1.imm << 32 | (u32)imm;
797 		emit_a64_mov_i64(dst, imm64, ctx);
798 
799 		return 1;
800 	}
801 
802 	/* LDX: dst = *(size *)(src + off) */
803 	case BPF_LDX | BPF_MEM | BPF_W:
804 	case BPF_LDX | BPF_MEM | BPF_H:
805 	case BPF_LDX | BPF_MEM | BPF_B:
806 	case BPF_LDX | BPF_MEM | BPF_DW:
807 	case BPF_LDX | BPF_PROBE_MEM | BPF_DW:
808 	case BPF_LDX | BPF_PROBE_MEM | BPF_W:
809 	case BPF_LDX | BPF_PROBE_MEM | BPF_H:
810 	case BPF_LDX | BPF_PROBE_MEM | BPF_B:
811 		emit_a64_mov_i(1, tmp, off, ctx);
812 		switch (BPF_SIZE(code)) {
813 		case BPF_W:
814 			emit(A64_LDR32(dst, src, tmp), ctx);
815 			break;
816 		case BPF_H:
817 			emit(A64_LDRH(dst, src, tmp), ctx);
818 			break;
819 		case BPF_B:
820 			emit(A64_LDRB(dst, src, tmp), ctx);
821 			break;
822 		case BPF_DW:
823 			emit(A64_LDR64(dst, src, tmp), ctx);
824 			break;
825 		}
826 
827 		ret = add_exception_handler(insn, ctx, dst);
828 		if (ret)
829 			return ret;
830 		break;
831 
832 	/* ST: *(size *)(dst + off) = imm */
833 	case BPF_ST | BPF_MEM | BPF_W:
834 	case BPF_ST | BPF_MEM | BPF_H:
835 	case BPF_ST | BPF_MEM | BPF_B:
836 	case BPF_ST | BPF_MEM | BPF_DW:
837 		/* Load imm to a register then store it */
838 		emit_a64_mov_i(1, tmp2, off, ctx);
839 		emit_a64_mov_i(1, tmp, imm, ctx);
840 		switch (BPF_SIZE(code)) {
841 		case BPF_W:
842 			emit(A64_STR32(tmp, dst, tmp2), ctx);
843 			break;
844 		case BPF_H:
845 			emit(A64_STRH(tmp, dst, tmp2), ctx);
846 			break;
847 		case BPF_B:
848 			emit(A64_STRB(tmp, dst, tmp2), ctx);
849 			break;
850 		case BPF_DW:
851 			emit(A64_STR64(tmp, dst, tmp2), ctx);
852 			break;
853 		}
854 		break;
855 
856 	/* STX: *(size *)(dst + off) = src */
857 	case BPF_STX | BPF_MEM | BPF_W:
858 	case BPF_STX | BPF_MEM | BPF_H:
859 	case BPF_STX | BPF_MEM | BPF_B:
860 	case BPF_STX | BPF_MEM | BPF_DW:
861 		emit_a64_mov_i(1, tmp, off, ctx);
862 		switch (BPF_SIZE(code)) {
863 		case BPF_W:
864 			emit(A64_STR32(src, dst, tmp), ctx);
865 			break;
866 		case BPF_H:
867 			emit(A64_STRH(src, dst, tmp), ctx);
868 			break;
869 		case BPF_B:
870 			emit(A64_STRB(src, dst, tmp), ctx);
871 			break;
872 		case BPF_DW:
873 			emit(A64_STR64(src, dst, tmp), ctx);
874 			break;
875 		}
876 		break;
877 
878 	/* STX XADD: lock *(u32 *)(dst + off) += src */
879 	case BPF_STX | BPF_XADD | BPF_W:
880 	/* STX XADD: lock *(u64 *)(dst + off) += src */
881 	case BPF_STX | BPF_XADD | BPF_DW:
882 		if (!off) {
883 			reg = dst;
884 		} else {
885 			emit_a64_mov_i(1, tmp, off, ctx);
886 			emit(A64_ADD(1, tmp, tmp, dst), ctx);
887 			reg = tmp;
888 		}
889 		if (cpus_have_cap(ARM64_HAS_LSE_ATOMICS)) {
890 			emit(A64_STADD(isdw, reg, src), ctx);
891 		} else {
892 			emit(A64_LDXR(isdw, tmp2, reg), ctx);
893 			emit(A64_ADD(isdw, tmp2, tmp2, src), ctx);
894 			emit(A64_STXR(isdw, tmp2, reg, tmp3), ctx);
895 			jmp_offset = -3;
896 			check_imm19(jmp_offset);
897 			emit(A64_CBNZ(0, tmp3, jmp_offset), ctx);
898 		}
899 		break;
900 
901 	default:
902 		pr_err_once("unknown opcode %02x\n", code);
903 		return -EINVAL;
904 	}
905 
906 	return 0;
907 }
908 
909 static int build_body(struct jit_ctx *ctx, bool extra_pass)
910 {
911 	const struct bpf_prog *prog = ctx->prog;
912 	int i;
913 
914 	/*
915 	 * - offset[0] offset of the end of prologue,
916 	 *   start of the 1st instruction.
917 	 * - offset[1] - offset of the end of 1st instruction,
918 	 *   start of the 2nd instruction
919 	 * [....]
920 	 * - offset[3] - offset of the end of 3rd instruction,
921 	 *   start of 4th instruction
922 	 */
923 	for (i = 0; i < prog->len; i++) {
924 		const struct bpf_insn *insn = &prog->insnsi[i];
925 		int ret;
926 
927 		if (ctx->image == NULL)
928 			ctx->offset[i] = ctx->idx;
929 		ret = build_insn(insn, ctx, extra_pass);
930 		if (ret > 0) {
931 			i++;
932 			if (ctx->image == NULL)
933 				ctx->offset[i] = ctx->idx;
934 			continue;
935 		}
936 		if (ret)
937 			return ret;
938 	}
939 	/*
940 	 * offset is allocated with prog->len + 1 so fill in
941 	 * the last element with the offset after the last
942 	 * instruction (end of program)
943 	 */
944 	if (ctx->image == NULL)
945 		ctx->offset[i] = ctx->idx;
946 
947 	return 0;
948 }
949 
950 static int validate_code(struct jit_ctx *ctx)
951 {
952 	int i;
953 
954 	for (i = 0; i < ctx->idx; i++) {
955 		u32 a64_insn = le32_to_cpu(ctx->image[i]);
956 
957 		if (a64_insn == AARCH64_BREAK_FAULT)
958 			return -1;
959 	}
960 
961 	if (WARN_ON_ONCE(ctx->exentry_idx != ctx->prog->aux->num_exentries))
962 		return -1;
963 
964 	return 0;
965 }
966 
967 static inline void bpf_flush_icache(void *start, void *end)
968 {
969 	flush_icache_range((unsigned long)start, (unsigned long)end);
970 }
971 
972 struct arm64_jit_data {
973 	struct bpf_binary_header *header;
974 	u8 *image;
975 	struct jit_ctx ctx;
976 };
977 
978 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
979 {
980 	int image_size, prog_size, extable_size;
981 	struct bpf_prog *tmp, *orig_prog = prog;
982 	struct bpf_binary_header *header;
983 	struct arm64_jit_data *jit_data;
984 	bool was_classic = bpf_prog_was_classic(prog);
985 	bool tmp_blinded = false;
986 	bool extra_pass = false;
987 	struct jit_ctx ctx;
988 	u8 *image_ptr;
989 
990 	if (!prog->jit_requested)
991 		return orig_prog;
992 
993 	tmp = bpf_jit_blind_constants(prog);
994 	/* If blinding was requested and we failed during blinding,
995 	 * we must fall back to the interpreter.
996 	 */
997 	if (IS_ERR(tmp))
998 		return orig_prog;
999 	if (tmp != prog) {
1000 		tmp_blinded = true;
1001 		prog = tmp;
1002 	}
1003 
1004 	jit_data = prog->aux->jit_data;
1005 	if (!jit_data) {
1006 		jit_data = kzalloc(sizeof(*jit_data), GFP_KERNEL);
1007 		if (!jit_data) {
1008 			prog = orig_prog;
1009 			goto out;
1010 		}
1011 		prog->aux->jit_data = jit_data;
1012 	}
1013 	if (jit_data->ctx.offset) {
1014 		ctx = jit_data->ctx;
1015 		image_ptr = jit_data->image;
1016 		header = jit_data->header;
1017 		extra_pass = true;
1018 		prog_size = sizeof(u32) * ctx.idx;
1019 		goto skip_init_ctx;
1020 	}
1021 	memset(&ctx, 0, sizeof(ctx));
1022 	ctx.prog = prog;
1023 
1024 	ctx.offset = kcalloc(prog->len + 1, sizeof(int), GFP_KERNEL);
1025 	if (ctx.offset == NULL) {
1026 		prog = orig_prog;
1027 		goto out_off;
1028 	}
1029 
1030 	/* 1. Initial fake pass to compute ctx->idx. */
1031 
1032 	/* Fake pass to fill in ctx->offset. */
1033 	if (build_body(&ctx, extra_pass)) {
1034 		prog = orig_prog;
1035 		goto out_off;
1036 	}
1037 
1038 	if (build_prologue(&ctx, was_classic)) {
1039 		prog = orig_prog;
1040 		goto out_off;
1041 	}
1042 
1043 	ctx.epilogue_offset = ctx.idx;
1044 	build_epilogue(&ctx);
1045 
1046 	extable_size = prog->aux->num_exentries *
1047 		sizeof(struct exception_table_entry);
1048 
1049 	/* Now we know the actual image size. */
1050 	prog_size = sizeof(u32) * ctx.idx;
1051 	image_size = prog_size + extable_size;
1052 	header = bpf_jit_binary_alloc(image_size, &image_ptr,
1053 				      sizeof(u32), jit_fill_hole);
1054 	if (header == NULL) {
1055 		prog = orig_prog;
1056 		goto out_off;
1057 	}
1058 
1059 	/* 2. Now, the actual pass. */
1060 
1061 	ctx.image = (__le32 *)image_ptr;
1062 	if (extable_size)
1063 		prog->aux->extable = (void *)image_ptr + prog_size;
1064 skip_init_ctx:
1065 	ctx.idx = 0;
1066 	ctx.exentry_idx = 0;
1067 
1068 	build_prologue(&ctx, was_classic);
1069 
1070 	if (build_body(&ctx, extra_pass)) {
1071 		bpf_jit_binary_free(header);
1072 		prog = orig_prog;
1073 		goto out_off;
1074 	}
1075 
1076 	build_epilogue(&ctx);
1077 
1078 	/* 3. Extra pass to validate JITed code. */
1079 	if (validate_code(&ctx)) {
1080 		bpf_jit_binary_free(header);
1081 		prog = orig_prog;
1082 		goto out_off;
1083 	}
1084 
1085 	/* And we're done. */
1086 	if (bpf_jit_enable > 1)
1087 		bpf_jit_dump(prog->len, prog_size, 2, ctx.image);
1088 
1089 	bpf_flush_icache(header, ctx.image + ctx.idx);
1090 
1091 	if (!prog->is_func || extra_pass) {
1092 		if (extra_pass && ctx.idx != jit_data->ctx.idx) {
1093 			pr_err_once("multi-func JIT bug %d != %d\n",
1094 				    ctx.idx, jit_data->ctx.idx);
1095 			bpf_jit_binary_free(header);
1096 			prog->bpf_func = NULL;
1097 			prog->jited = 0;
1098 			goto out_off;
1099 		}
1100 		bpf_jit_binary_lock_ro(header);
1101 	} else {
1102 		jit_data->ctx = ctx;
1103 		jit_data->image = image_ptr;
1104 		jit_data->header = header;
1105 	}
1106 	prog->bpf_func = (void *)ctx.image;
1107 	prog->jited = 1;
1108 	prog->jited_len = prog_size;
1109 
1110 	if (!prog->is_func || extra_pass) {
1111 		bpf_prog_fill_jited_linfo(prog, ctx.offset + 1);
1112 out_off:
1113 		kfree(ctx.offset);
1114 		kfree(jit_data);
1115 		prog->aux->jit_data = NULL;
1116 	}
1117 out:
1118 	if (tmp_blinded)
1119 		bpf_jit_prog_release_other(prog, prog == orig_prog ?
1120 					   tmp : orig_prog);
1121 	return prog;
1122 }
1123 
1124 void *bpf_jit_alloc_exec(unsigned long size)
1125 {
1126 	return __vmalloc_node_range(size, PAGE_SIZE, BPF_JIT_REGION_START,
1127 				    BPF_JIT_REGION_END, GFP_KERNEL,
1128 				    PAGE_KERNEL, 0, NUMA_NO_NODE,
1129 				    __builtin_return_address(0));
1130 }
1131 
1132 void bpf_jit_free_exec(void *addr)
1133 {
1134 	return vfree(addr);
1135 }
1136