xref: /openbmc/linux/arch/x86/lib/insn-eval.c (revision 2f0754f2)
1 /*
2  * Utility functions for x86 operand and address decoding
3  *
4  * Copyright (C) Intel Corporation 2017
5  */
6 #include <linux/kernel.h>
7 #include <linux/string.h>
8 #include <linux/ratelimit.h>
9 #include <linux/mmu_context.h>
10 #include <asm/desc_defs.h>
11 #include <asm/desc.h>
12 #include <asm/inat.h>
13 #include <asm/insn.h>
14 #include <asm/insn-eval.h>
15 #include <asm/ldt.h>
16 #include <asm/vm86.h>
17 
18 #undef pr_fmt
19 #define pr_fmt(fmt) "insn: " fmt
20 
21 enum reg_type {
22 	REG_TYPE_RM = 0,
23 	REG_TYPE_REG,
24 	REG_TYPE_INDEX,
25 	REG_TYPE_BASE,
26 };
27 
28 /**
29  * is_string_insn() - Determine if instruction is a string instruction
30  * @insn:	Instruction containing the opcode to inspect
31  *
32  * Returns:
33  *
34  * true if the instruction, determined by the opcode, is any of the
35  * string instructions as defined in the Intel Software Development manual.
36  * False otherwise.
37  */
38 static bool is_string_insn(struct insn *insn)
39 {
40 	/* All string instructions have a 1-byte opcode. */
41 	if (insn->opcode.nbytes != 1)
42 		return false;
43 
44 	switch (insn->opcode.bytes[0]) {
45 	case 0x6c ... 0x6f:	/* INS, OUTS */
46 	case 0xa4 ... 0xa7:	/* MOVS, CMPS */
47 	case 0xaa ... 0xaf:	/* STOS, LODS, SCAS */
48 		return true;
49 	default:
50 		return false;
51 	}
52 }
53 
54 /**
55  * insn_has_rep_prefix() - Determine if instruction has a REP prefix
56  * @insn:	Instruction containing the prefix to inspect
57  *
58  * Returns:
59  *
60  * true if the instruction has a REP prefix, false if not.
61  */
62 bool insn_has_rep_prefix(struct insn *insn)
63 {
64 	insn_byte_t p;
65 	int i;
66 
67 	insn_get_prefixes(insn);
68 
69 	for_each_insn_prefix(insn, i, p) {
70 		if (p == 0xf2 || p == 0xf3)
71 			return true;
72 	}
73 
74 	return false;
75 }
76 
77 /**
78  * get_seg_reg_override_idx() - obtain segment register override index
79  * @insn:	Valid instruction with segment override prefixes
80  *
81  * Inspect the instruction prefixes in @insn and find segment overrides, if any.
82  *
83  * Returns:
84  *
85  * A constant identifying the segment register to use, among CS, SS, DS,
86  * ES, FS, or GS. INAT_SEG_REG_DEFAULT is returned if no segment override
87  * prefixes were found.
88  *
89  * -EINVAL in case of error.
90  */
91 static int get_seg_reg_override_idx(struct insn *insn)
92 {
93 	int idx = INAT_SEG_REG_DEFAULT;
94 	int num_overrides = 0, i;
95 	insn_byte_t p;
96 
97 	insn_get_prefixes(insn);
98 
99 	/* Look for any segment override prefixes. */
100 	for_each_insn_prefix(insn, i, p) {
101 		insn_attr_t attr;
102 
103 		attr = inat_get_opcode_attribute(p);
104 		switch (attr) {
105 		case INAT_MAKE_PREFIX(INAT_PFX_CS):
106 			idx = INAT_SEG_REG_CS;
107 			num_overrides++;
108 			break;
109 		case INAT_MAKE_PREFIX(INAT_PFX_SS):
110 			idx = INAT_SEG_REG_SS;
111 			num_overrides++;
112 			break;
113 		case INAT_MAKE_PREFIX(INAT_PFX_DS):
114 			idx = INAT_SEG_REG_DS;
115 			num_overrides++;
116 			break;
117 		case INAT_MAKE_PREFIX(INAT_PFX_ES):
118 			idx = INAT_SEG_REG_ES;
119 			num_overrides++;
120 			break;
121 		case INAT_MAKE_PREFIX(INAT_PFX_FS):
122 			idx = INAT_SEG_REG_FS;
123 			num_overrides++;
124 			break;
125 		case INAT_MAKE_PREFIX(INAT_PFX_GS):
126 			idx = INAT_SEG_REG_GS;
127 			num_overrides++;
128 			break;
129 		/* No default action needed. */
130 		}
131 	}
132 
133 	/* More than one segment override prefix leads to undefined behavior. */
134 	if (num_overrides > 1)
135 		return -EINVAL;
136 
137 	return idx;
138 }
139 
140 /**
141  * check_seg_overrides() - check if segment override prefixes are allowed
142  * @insn:	Valid instruction with segment override prefixes
143  * @regoff:	Operand offset, in pt_regs, for which the check is performed
144  *
145  * For a particular register used in register-indirect addressing, determine if
146  * segment override prefixes can be used. Specifically, no overrides are allowed
147  * for rDI if used with a string instruction.
148  *
149  * Returns:
150  *
151  * True if segment override prefixes can be used with the register indicated
152  * in @regoff. False if otherwise.
153  */
154 static bool check_seg_overrides(struct insn *insn, int regoff)
155 {
156 	if (regoff == offsetof(struct pt_regs, di) && is_string_insn(insn))
157 		return false;
158 
159 	return true;
160 }
161 
162 /**
163  * resolve_default_seg() - resolve default segment register index for an operand
164  * @insn:	Instruction with opcode and address size. Must be valid.
165  * @regs:	Register values as seen when entering kernel mode
166  * @off:	Operand offset, in pt_regs, for which resolution is needed
167  *
168  * Resolve the default segment register index associated with the instruction
169  * operand register indicated by @off. Such index is resolved based on defaults
170  * described in the Intel Software Development Manual.
171  *
172  * Returns:
173  *
174  * If in protected mode, a constant identifying the segment register to use,
175  * among CS, SS, ES or DS. If in long mode, INAT_SEG_REG_IGNORE.
176  *
177  * -EINVAL in case of error.
178  */
179 static int resolve_default_seg(struct insn *insn, struct pt_regs *regs, int off)
180 {
181 	if (any_64bit_mode(regs))
182 		return INAT_SEG_REG_IGNORE;
183 	/*
184 	 * Resolve the default segment register as described in Section 3.7.4
185 	 * of the Intel Software Development Manual Vol. 1:
186 	 *
187 	 *  + DS for all references involving r[ABCD]X, and rSI.
188 	 *  + If used in a string instruction, ES for rDI. Otherwise, DS.
189 	 *  + AX, CX and DX are not valid register operands in 16-bit address
190 	 *    encodings but are valid for 32-bit and 64-bit encodings.
191 	 *  + -EDOM is reserved to identify for cases in which no register
192 	 *    is used (i.e., displacement-only addressing). Use DS.
193 	 *  + SS for rSP or rBP.
194 	 *  + CS for rIP.
195 	 */
196 
197 	switch (off) {
198 	case offsetof(struct pt_regs, ax):
199 	case offsetof(struct pt_regs, cx):
200 	case offsetof(struct pt_regs, dx):
201 		/* Need insn to verify address size. */
202 		if (insn->addr_bytes == 2)
203 			return -EINVAL;
204 
205 		fallthrough;
206 
207 	case -EDOM:
208 	case offsetof(struct pt_regs, bx):
209 	case offsetof(struct pt_regs, si):
210 		return INAT_SEG_REG_DS;
211 
212 	case offsetof(struct pt_regs, di):
213 		if (is_string_insn(insn))
214 			return INAT_SEG_REG_ES;
215 		return INAT_SEG_REG_DS;
216 
217 	case offsetof(struct pt_regs, bp):
218 	case offsetof(struct pt_regs, sp):
219 		return INAT_SEG_REG_SS;
220 
221 	case offsetof(struct pt_regs, ip):
222 		return INAT_SEG_REG_CS;
223 
224 	default:
225 		return -EINVAL;
226 	}
227 }
228 
229 /**
230  * resolve_seg_reg() - obtain segment register index
231  * @insn:	Instruction with operands
232  * @regs:	Register values as seen when entering kernel mode
233  * @regoff:	Operand offset, in pt_regs, used to determine segment register
234  *
235  * Determine the segment register associated with the operands and, if
236  * applicable, prefixes and the instruction pointed by @insn.
237  *
238  * The segment register associated to an operand used in register-indirect
239  * addressing depends on:
240  *
241  * a) Whether running in long mode (in such a case segments are ignored, except
242  * if FS or GS are used).
243  *
244  * b) Whether segment override prefixes can be used. Certain instructions and
245  *    registers do not allow override prefixes.
246  *
247  * c) Whether segment overrides prefixes are found in the instruction prefixes.
248  *
249  * d) If there are not segment override prefixes or they cannot be used, the
250  *    default segment register associated with the operand register is used.
251  *
252  * The function checks first if segment override prefixes can be used with the
253  * operand indicated by @regoff. If allowed, obtain such overridden segment
254  * register index. Lastly, if not prefixes were found or cannot be used, resolve
255  * the segment register index to use based on the defaults described in the
256  * Intel documentation. In long mode, all segment register indexes will be
257  * ignored, except if overrides were found for FS or GS. All these operations
258  * are done using helper functions.
259  *
260  * The operand register, @regoff, is represented as the offset from the base of
261  * pt_regs.
262  *
263  * As stated, the main use of this function is to determine the segment register
264  * index based on the instruction, its operands and prefixes. Hence, @insn
265  * must be valid. However, if @regoff indicates rIP, we don't need to inspect
266  * @insn at all as in this case CS is used in all cases. This case is checked
267  * before proceeding further.
268  *
269  * Please note that this function does not return the value in the segment
270  * register (i.e., the segment selector) but our defined index. The segment
271  * selector needs to be obtained using get_segment_selector() and passing the
272  * segment register index resolved by this function.
273  *
274  * Returns:
275  *
276  * An index identifying the segment register to use, among CS, SS, DS,
277  * ES, FS, or GS. INAT_SEG_REG_IGNORE is returned if running in long mode.
278  *
279  * -EINVAL in case of error.
280  */
281 static int resolve_seg_reg(struct insn *insn, struct pt_regs *regs, int regoff)
282 {
283 	int idx;
284 
285 	/*
286 	 * In the unlikely event of having to resolve the segment register
287 	 * index for rIP, do it first. Segment override prefixes should not
288 	 * be used. Hence, it is not necessary to inspect the instruction,
289 	 * which may be invalid at this point.
290 	 */
291 	if (regoff == offsetof(struct pt_regs, ip)) {
292 		if (any_64bit_mode(regs))
293 			return INAT_SEG_REG_IGNORE;
294 		else
295 			return INAT_SEG_REG_CS;
296 	}
297 
298 	if (!insn)
299 		return -EINVAL;
300 
301 	if (!check_seg_overrides(insn, regoff))
302 		return resolve_default_seg(insn, regs, regoff);
303 
304 	idx = get_seg_reg_override_idx(insn);
305 	if (idx < 0)
306 		return idx;
307 
308 	if (idx == INAT_SEG_REG_DEFAULT)
309 		return resolve_default_seg(insn, regs, regoff);
310 
311 	/*
312 	 * In long mode, segment override prefixes are ignored, except for
313 	 * overrides for FS and GS.
314 	 */
315 	if (any_64bit_mode(regs)) {
316 		if (idx != INAT_SEG_REG_FS &&
317 		    idx != INAT_SEG_REG_GS)
318 			idx = INAT_SEG_REG_IGNORE;
319 	}
320 
321 	return idx;
322 }
323 
324 /**
325  * get_segment_selector() - obtain segment selector
326  * @regs:		Register values as seen when entering kernel mode
327  * @seg_reg_idx:	Segment register index to use
328  *
329  * Obtain the segment selector from any of the CS, SS, DS, ES, FS, GS segment
330  * registers. In CONFIG_X86_32, the segment is obtained from either pt_regs or
331  * kernel_vm86_regs as applicable. In CONFIG_X86_64, CS and SS are obtained
332  * from pt_regs. DS, ES, FS and GS are obtained by reading the actual CPU
333  * registers. This done for only for completeness as in CONFIG_X86_64 segment
334  * registers are ignored.
335  *
336  * Returns:
337  *
338  * Value of the segment selector, including null when running in
339  * long mode.
340  *
341  * -EINVAL on error.
342  */
343 static short get_segment_selector(struct pt_regs *regs, int seg_reg_idx)
344 {
345 #ifdef CONFIG_X86_64
346 	unsigned short sel;
347 
348 	switch (seg_reg_idx) {
349 	case INAT_SEG_REG_IGNORE:
350 		return 0;
351 	case INAT_SEG_REG_CS:
352 		return (unsigned short)(regs->cs & 0xffff);
353 	case INAT_SEG_REG_SS:
354 		return (unsigned short)(regs->ss & 0xffff);
355 	case INAT_SEG_REG_DS:
356 		savesegment(ds, sel);
357 		return sel;
358 	case INAT_SEG_REG_ES:
359 		savesegment(es, sel);
360 		return sel;
361 	case INAT_SEG_REG_FS:
362 		savesegment(fs, sel);
363 		return sel;
364 	case INAT_SEG_REG_GS:
365 		savesegment(gs, sel);
366 		return sel;
367 	default:
368 		return -EINVAL;
369 	}
370 #else /* CONFIG_X86_32 */
371 	struct kernel_vm86_regs *vm86regs = (struct kernel_vm86_regs *)regs;
372 
373 	if (v8086_mode(regs)) {
374 		switch (seg_reg_idx) {
375 		case INAT_SEG_REG_CS:
376 			return (unsigned short)(regs->cs & 0xffff);
377 		case INAT_SEG_REG_SS:
378 			return (unsigned short)(regs->ss & 0xffff);
379 		case INAT_SEG_REG_DS:
380 			return vm86regs->ds;
381 		case INAT_SEG_REG_ES:
382 			return vm86regs->es;
383 		case INAT_SEG_REG_FS:
384 			return vm86regs->fs;
385 		case INAT_SEG_REG_GS:
386 			return vm86regs->gs;
387 		case INAT_SEG_REG_IGNORE:
388 		default:
389 			return -EINVAL;
390 		}
391 	}
392 
393 	switch (seg_reg_idx) {
394 	case INAT_SEG_REG_CS:
395 		return (unsigned short)(regs->cs & 0xffff);
396 	case INAT_SEG_REG_SS:
397 		return (unsigned short)(regs->ss & 0xffff);
398 	case INAT_SEG_REG_DS:
399 		return (unsigned short)(regs->ds & 0xffff);
400 	case INAT_SEG_REG_ES:
401 		return (unsigned short)(regs->es & 0xffff);
402 	case INAT_SEG_REG_FS:
403 		return (unsigned short)(regs->fs & 0xffff);
404 	case INAT_SEG_REG_GS:
405 		return get_user_gs(regs);
406 	case INAT_SEG_REG_IGNORE:
407 	default:
408 		return -EINVAL;
409 	}
410 #endif /* CONFIG_X86_64 */
411 }
412 
413 static const int pt_regoff[] = {
414 	offsetof(struct pt_regs, ax),
415 	offsetof(struct pt_regs, cx),
416 	offsetof(struct pt_regs, dx),
417 	offsetof(struct pt_regs, bx),
418 	offsetof(struct pt_regs, sp),
419 	offsetof(struct pt_regs, bp),
420 	offsetof(struct pt_regs, si),
421 	offsetof(struct pt_regs, di),
422 #ifdef CONFIG_X86_64
423 	offsetof(struct pt_regs, r8),
424 	offsetof(struct pt_regs, r9),
425 	offsetof(struct pt_regs, r10),
426 	offsetof(struct pt_regs, r11),
427 	offsetof(struct pt_regs, r12),
428 	offsetof(struct pt_regs, r13),
429 	offsetof(struct pt_regs, r14),
430 	offsetof(struct pt_regs, r15),
431 #else
432 	offsetof(struct pt_regs, ds),
433 	offsetof(struct pt_regs, es),
434 	offsetof(struct pt_regs, fs),
435 	offsetof(struct pt_regs, gs),
436 #endif
437 };
438 
439 int pt_regs_offset(struct pt_regs *regs, int regno)
440 {
441 	if ((unsigned)regno < ARRAY_SIZE(pt_regoff))
442 		return pt_regoff[regno];
443 	return -EDOM;
444 }
445 
446 static int get_regno(struct insn *insn, enum reg_type type)
447 {
448 	int nr_registers = ARRAY_SIZE(pt_regoff);
449 	int regno = 0;
450 
451 	/*
452 	 * Don't possibly decode a 32-bit instructions as
453 	 * reading a 64-bit-only register.
454 	 */
455 	if (IS_ENABLED(CONFIG_X86_64) && !insn->x86_64)
456 		nr_registers -= 8;
457 
458 	switch (type) {
459 	case REG_TYPE_RM:
460 		regno = X86_MODRM_RM(insn->modrm.value);
461 
462 		/*
463 		 * ModRM.mod == 0 and ModRM.rm == 5 means a 32-bit displacement
464 		 * follows the ModRM byte.
465 		 */
466 		if (!X86_MODRM_MOD(insn->modrm.value) && regno == 5)
467 			return -EDOM;
468 
469 		if (X86_REX_B(insn->rex_prefix.value))
470 			regno += 8;
471 		break;
472 
473 	case REG_TYPE_REG:
474 		regno = X86_MODRM_REG(insn->modrm.value);
475 
476 		if (X86_REX_R(insn->rex_prefix.value))
477 			regno += 8;
478 		break;
479 
480 	case REG_TYPE_INDEX:
481 		regno = X86_SIB_INDEX(insn->sib.value);
482 		if (X86_REX_X(insn->rex_prefix.value))
483 			regno += 8;
484 
485 		/*
486 		 * If ModRM.mod != 3 and SIB.index = 4 the scale*index
487 		 * portion of the address computation is null. This is
488 		 * true only if REX.X is 0. In such a case, the SIB index
489 		 * is used in the address computation.
490 		 */
491 		if (X86_MODRM_MOD(insn->modrm.value) != 3 && regno == 4)
492 			return -EDOM;
493 		break;
494 
495 	case REG_TYPE_BASE:
496 		regno = X86_SIB_BASE(insn->sib.value);
497 		/*
498 		 * If ModRM.mod is 0 and SIB.base == 5, the base of the
499 		 * register-indirect addressing is 0. In this case, a
500 		 * 32-bit displacement follows the SIB byte.
501 		 */
502 		if (!X86_MODRM_MOD(insn->modrm.value) && regno == 5)
503 			return -EDOM;
504 
505 		if (X86_REX_B(insn->rex_prefix.value))
506 			regno += 8;
507 		break;
508 
509 	default:
510 		pr_err_ratelimited("invalid register type: %d\n", type);
511 		return -EINVAL;
512 	}
513 
514 	if (regno >= nr_registers) {
515 		WARN_ONCE(1, "decoded an instruction with an invalid register");
516 		return -EINVAL;
517 	}
518 	return regno;
519 }
520 
521 static int get_reg_offset(struct insn *insn, struct pt_regs *regs,
522 			  enum reg_type type)
523 {
524 	int regno = get_regno(insn, type);
525 
526 	if (regno < 0)
527 		return regno;
528 
529 	return pt_regs_offset(regs, regno);
530 }
531 
532 /**
533  * get_reg_offset_16() - Obtain offset of register indicated by instruction
534  * @insn:	Instruction containing ModRM byte
535  * @regs:	Register values as seen when entering kernel mode
536  * @offs1:	Offset of the first operand register
537  * @offs2:	Offset of the second operand register, if applicable
538  *
539  * Obtain the offset, in pt_regs, of the registers indicated by the ModRM byte
540  * in @insn. This function is to be used with 16-bit address encodings. The
541  * @offs1 and @offs2 will be written with the offset of the two registers
542  * indicated by the instruction. In cases where any of the registers is not
543  * referenced by the instruction, the value will be set to -EDOM.
544  *
545  * Returns:
546  *
547  * 0 on success, -EINVAL on error.
548  */
549 static int get_reg_offset_16(struct insn *insn, struct pt_regs *regs,
550 			     int *offs1, int *offs2)
551 {
552 	/*
553 	 * 16-bit addressing can use one or two registers. Specifics of
554 	 * encodings are given in Table 2-1. "16-Bit Addressing Forms with the
555 	 * ModR/M Byte" of the Intel Software Development Manual.
556 	 */
557 	static const int regoff1[] = {
558 		offsetof(struct pt_regs, bx),
559 		offsetof(struct pt_regs, bx),
560 		offsetof(struct pt_regs, bp),
561 		offsetof(struct pt_regs, bp),
562 		offsetof(struct pt_regs, si),
563 		offsetof(struct pt_regs, di),
564 		offsetof(struct pt_regs, bp),
565 		offsetof(struct pt_regs, bx),
566 	};
567 
568 	static const int regoff2[] = {
569 		offsetof(struct pt_regs, si),
570 		offsetof(struct pt_regs, di),
571 		offsetof(struct pt_regs, si),
572 		offsetof(struct pt_regs, di),
573 		-EDOM,
574 		-EDOM,
575 		-EDOM,
576 		-EDOM,
577 	};
578 
579 	if (!offs1 || !offs2)
580 		return -EINVAL;
581 
582 	/* Operand is a register, use the generic function. */
583 	if (X86_MODRM_MOD(insn->modrm.value) == 3) {
584 		*offs1 = insn_get_modrm_rm_off(insn, regs);
585 		*offs2 = -EDOM;
586 		return 0;
587 	}
588 
589 	*offs1 = regoff1[X86_MODRM_RM(insn->modrm.value)];
590 	*offs2 = regoff2[X86_MODRM_RM(insn->modrm.value)];
591 
592 	/*
593 	 * If ModRM.mod is 0 and ModRM.rm is 110b, then we use displacement-
594 	 * only addressing. This means that no registers are involved in
595 	 * computing the effective address. Thus, ensure that the first
596 	 * register offset is invalid. The second register offset is already
597 	 * invalid under the aforementioned conditions.
598 	 */
599 	if ((X86_MODRM_MOD(insn->modrm.value) == 0) &&
600 	    (X86_MODRM_RM(insn->modrm.value) == 6))
601 		*offs1 = -EDOM;
602 
603 	return 0;
604 }
605 
606 /**
607  * get_desc() - Obtain contents of a segment descriptor
608  * @out:	Segment descriptor contents on success
609  * @sel:	Segment selector
610  *
611  * Given a segment selector, obtain a pointer to the segment descriptor.
612  * Both global and local descriptor tables are supported.
613  *
614  * Returns:
615  *
616  * True on success, false on failure.
617  *
618  * NULL on error.
619  */
620 static bool get_desc(struct desc_struct *out, unsigned short sel)
621 {
622 	struct desc_ptr gdt_desc = {0, 0};
623 	unsigned long desc_base;
624 
625 #ifdef CONFIG_MODIFY_LDT_SYSCALL
626 	if ((sel & SEGMENT_TI_MASK) == SEGMENT_LDT) {
627 		bool success = false;
628 		struct ldt_struct *ldt;
629 
630 		/* Bits [15:3] contain the index of the desired entry. */
631 		sel >>= 3;
632 
633 		mutex_lock(&current->active_mm->context.lock);
634 		ldt = current->active_mm->context.ldt;
635 		if (ldt && sel < ldt->nr_entries) {
636 			*out = ldt->entries[sel];
637 			success = true;
638 		}
639 
640 		mutex_unlock(&current->active_mm->context.lock);
641 
642 		return success;
643 	}
644 #endif
645 	native_store_gdt(&gdt_desc);
646 
647 	/*
648 	 * Segment descriptors have a size of 8 bytes. Thus, the index is
649 	 * multiplied by 8 to obtain the memory offset of the desired descriptor
650 	 * from the base of the GDT. As bits [15:3] of the segment selector
651 	 * contain the index, it can be regarded as multiplied by 8 already.
652 	 * All that remains is to clear bits [2:0].
653 	 */
654 	desc_base = sel & ~(SEGMENT_RPL_MASK | SEGMENT_TI_MASK);
655 
656 	if (desc_base > gdt_desc.size)
657 		return false;
658 
659 	*out = *(struct desc_struct *)(gdt_desc.address + desc_base);
660 	return true;
661 }
662 
663 /**
664  * insn_get_seg_base() - Obtain base address of segment descriptor.
665  * @regs:		Register values as seen when entering kernel mode
666  * @seg_reg_idx:	Index of the segment register pointing to seg descriptor
667  *
668  * Obtain the base address of the segment as indicated by the segment descriptor
669  * pointed by the segment selector. The segment selector is obtained from the
670  * input segment register index @seg_reg_idx.
671  *
672  * Returns:
673  *
674  * In protected mode, base address of the segment. Zero in long mode,
675  * except when FS or GS are used. In virtual-8086 mode, the segment
676  * selector shifted 4 bits to the right.
677  *
678  * -1L in case of error.
679  */
680 unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx)
681 {
682 	struct desc_struct desc;
683 	short sel;
684 
685 	sel = get_segment_selector(regs, seg_reg_idx);
686 	if (sel < 0)
687 		return -1L;
688 
689 	if (v8086_mode(regs))
690 		/*
691 		 * Base is simply the segment selector shifted 4
692 		 * bits to the right.
693 		 */
694 		return (unsigned long)(sel << 4);
695 
696 	if (any_64bit_mode(regs)) {
697 		/*
698 		 * Only FS or GS will have a base address, the rest of
699 		 * the segments' bases are forced to 0.
700 		 */
701 		unsigned long base;
702 
703 		if (seg_reg_idx == INAT_SEG_REG_FS) {
704 			rdmsrl(MSR_FS_BASE, base);
705 		} else if (seg_reg_idx == INAT_SEG_REG_GS) {
706 			/*
707 			 * swapgs was called at the kernel entry point. Thus,
708 			 * MSR_KERNEL_GS_BASE will have the user-space GS base.
709 			 */
710 			if (user_mode(regs))
711 				rdmsrl(MSR_KERNEL_GS_BASE, base);
712 			else
713 				rdmsrl(MSR_GS_BASE, base);
714 		} else {
715 			base = 0;
716 		}
717 		return base;
718 	}
719 
720 	/* In protected mode the segment selector cannot be null. */
721 	if (!sel)
722 		return -1L;
723 
724 	if (!get_desc(&desc, sel))
725 		return -1L;
726 
727 	return get_desc_base(&desc);
728 }
729 
730 /**
731  * get_seg_limit() - Obtain the limit of a segment descriptor
732  * @regs:		Register values as seen when entering kernel mode
733  * @seg_reg_idx:	Index of the segment register pointing to seg descriptor
734  *
735  * Obtain the limit of the segment as indicated by the segment descriptor
736  * pointed by the segment selector. The segment selector is obtained from the
737  * input segment register index @seg_reg_idx.
738  *
739  * Returns:
740  *
741  * In protected mode, the limit of the segment descriptor in bytes.
742  * In long mode and virtual-8086 mode, segment limits are not enforced. Thus,
743  * limit is returned as -1L to imply a limit-less segment.
744  *
745  * Zero is returned on error.
746  */
747 static unsigned long get_seg_limit(struct pt_regs *regs, int seg_reg_idx)
748 {
749 	struct desc_struct desc;
750 	unsigned long limit;
751 	short sel;
752 
753 	sel = get_segment_selector(regs, seg_reg_idx);
754 	if (sel < 0)
755 		return 0;
756 
757 	if (any_64bit_mode(regs) || v8086_mode(regs))
758 		return -1L;
759 
760 	if (!sel)
761 		return 0;
762 
763 	if (!get_desc(&desc, sel))
764 		return 0;
765 
766 	/*
767 	 * If the granularity bit is set, the limit is given in multiples
768 	 * of 4096. This also means that the 12 least significant bits are
769 	 * not tested when checking the segment limits. In practice,
770 	 * this means that the segment ends in (limit << 12) + 0xfff.
771 	 */
772 	limit = get_desc_limit(&desc);
773 	if (desc.g)
774 		limit = (limit << 12) + 0xfff;
775 
776 	return limit;
777 }
778 
779 /**
780  * insn_get_code_seg_params() - Obtain code segment parameters
781  * @regs:	Structure with register values as seen when entering kernel mode
782  *
783  * Obtain address and operand sizes of the code segment. It is obtained from the
784  * selector contained in the CS register in regs. In protected mode, the default
785  * address is determined by inspecting the L and D bits of the segment
786  * descriptor. In virtual-8086 mode, the default is always two bytes for both
787  * address and operand sizes.
788  *
789  * Returns:
790  *
791  * An int containing ORed-in default parameters on success.
792  *
793  * -EINVAL on error.
794  */
795 int insn_get_code_seg_params(struct pt_regs *regs)
796 {
797 	struct desc_struct desc;
798 	short sel;
799 
800 	if (v8086_mode(regs))
801 		/* Address and operand size are both 16-bit. */
802 		return INSN_CODE_SEG_PARAMS(2, 2);
803 
804 	sel = get_segment_selector(regs, INAT_SEG_REG_CS);
805 	if (sel < 0)
806 		return sel;
807 
808 	if (!get_desc(&desc, sel))
809 		return -EINVAL;
810 
811 	/*
812 	 * The most significant byte of the Type field of the segment descriptor
813 	 * determines whether a segment contains data or code. If this is a data
814 	 * segment, return error.
815 	 */
816 	if (!(desc.type & BIT(3)))
817 		return -EINVAL;
818 
819 	switch ((desc.l << 1) | desc.d) {
820 	case 0: /*
821 		 * Legacy mode. CS.L=0, CS.D=0. Address and operand size are
822 		 * both 16-bit.
823 		 */
824 		return INSN_CODE_SEG_PARAMS(2, 2);
825 	case 1: /*
826 		 * Legacy mode. CS.L=0, CS.D=1. Address and operand size are
827 		 * both 32-bit.
828 		 */
829 		return INSN_CODE_SEG_PARAMS(4, 4);
830 	case 2: /*
831 		 * IA-32e 64-bit mode. CS.L=1, CS.D=0. Address size is 64-bit;
832 		 * operand size is 32-bit.
833 		 */
834 		return INSN_CODE_SEG_PARAMS(4, 8);
835 	case 3: /* Invalid setting. CS.L=1, CS.D=1 */
836 		fallthrough;
837 	default:
838 		return -EINVAL;
839 	}
840 }
841 
842 /**
843  * insn_get_modrm_rm_off() - Obtain register in r/m part of the ModRM byte
844  * @insn:	Instruction containing the ModRM byte
845  * @regs:	Register values as seen when entering kernel mode
846  *
847  * Returns:
848  *
849  * The register indicated by the r/m part of the ModRM byte. The
850  * register is obtained as an offset from the base of pt_regs. In specific
851  * cases, the returned value can be -EDOM to indicate that the particular value
852  * of ModRM does not refer to a register and shall be ignored.
853  */
854 int insn_get_modrm_rm_off(struct insn *insn, struct pt_regs *regs)
855 {
856 	return get_reg_offset(insn, regs, REG_TYPE_RM);
857 }
858 
859 /**
860  * insn_get_modrm_reg_off() - Obtain register in reg part of the ModRM byte
861  * @insn:	Instruction containing the ModRM byte
862  * @regs:	Register values as seen when entering kernel mode
863  *
864  * Returns:
865  *
866  * The register indicated by the reg part of the ModRM byte. The
867  * register is obtained as an offset from the base of pt_regs.
868  */
869 int insn_get_modrm_reg_off(struct insn *insn, struct pt_regs *regs)
870 {
871 	return get_reg_offset(insn, regs, REG_TYPE_REG);
872 }
873 
874 /**
875  * insn_get_modrm_reg_ptr() - Obtain register pointer based on ModRM byte
876  * @insn:	Instruction containing the ModRM byte
877  * @regs:	Register values as seen when entering kernel mode
878  *
879  * Returns:
880  *
881  * The register indicated by the reg part of the ModRM byte.
882  * The register is obtained as a pointer within pt_regs.
883  */
884 unsigned long *insn_get_modrm_reg_ptr(struct insn *insn, struct pt_regs *regs)
885 {
886 	int offset;
887 
888 	offset = insn_get_modrm_reg_off(insn, regs);
889 	if (offset < 0)
890 		return NULL;
891 	return (void *)regs + offset;
892 }
893 
894 /**
895  * get_seg_base_limit() - obtain base address and limit of a segment
896  * @insn:	Instruction. Must be valid.
897  * @regs:	Register values as seen when entering kernel mode
898  * @regoff:	Operand offset, in pt_regs, used to resolve segment descriptor
899  * @base:	Obtained segment base
900  * @limit:	Obtained segment limit
901  *
902  * Obtain the base address and limit of the segment associated with the operand
903  * @regoff and, if any or allowed, override prefixes in @insn. This function is
904  * different from insn_get_seg_base() as the latter does not resolve the segment
905  * associated with the instruction operand. If a limit is not needed (e.g.,
906  * when running in long mode), @limit can be NULL.
907  *
908  * Returns:
909  *
910  * 0 on success. @base and @limit will contain the base address and of the
911  * resolved segment, respectively.
912  *
913  * -EINVAL on error.
914  */
915 static int get_seg_base_limit(struct insn *insn, struct pt_regs *regs,
916 			      int regoff, unsigned long *base,
917 			      unsigned long *limit)
918 {
919 	int seg_reg_idx;
920 
921 	if (!base)
922 		return -EINVAL;
923 
924 	seg_reg_idx = resolve_seg_reg(insn, regs, regoff);
925 	if (seg_reg_idx < 0)
926 		return seg_reg_idx;
927 
928 	*base = insn_get_seg_base(regs, seg_reg_idx);
929 	if (*base == -1L)
930 		return -EINVAL;
931 
932 	if (!limit)
933 		return 0;
934 
935 	*limit = get_seg_limit(regs, seg_reg_idx);
936 	if (!(*limit))
937 		return -EINVAL;
938 
939 	return 0;
940 }
941 
942 /**
943  * get_eff_addr_reg() - Obtain effective address from register operand
944  * @insn:	Instruction. Must be valid.
945  * @regs:	Register values as seen when entering kernel mode
946  * @regoff:	Obtained operand offset, in pt_regs, with the effective address
947  * @eff_addr:	Obtained effective address
948  *
949  * Obtain the effective address stored in the register operand as indicated by
950  * the ModRM byte. This function is to be used only with register addressing
951  * (i.e.,  ModRM.mod is 3). The effective address is saved in @eff_addr. The
952  * register operand, as an offset from the base of pt_regs, is saved in @regoff;
953  * such offset can then be used to resolve the segment associated with the
954  * operand. This function can be used with any of the supported address sizes
955  * in x86.
956  *
957  * Returns:
958  *
959  * 0 on success. @eff_addr will have the effective address stored in the
960  * operand indicated by ModRM. @regoff will have such operand as an offset from
961  * the base of pt_regs.
962  *
963  * -EINVAL on error.
964  */
965 static int get_eff_addr_reg(struct insn *insn, struct pt_regs *regs,
966 			    int *regoff, long *eff_addr)
967 {
968 	int ret;
969 
970 	ret = insn_get_modrm(insn);
971 	if (ret)
972 		return ret;
973 
974 	if (X86_MODRM_MOD(insn->modrm.value) != 3)
975 		return -EINVAL;
976 
977 	*regoff = get_reg_offset(insn, regs, REG_TYPE_RM);
978 	if (*regoff < 0)
979 		return -EINVAL;
980 
981 	/* Ignore bytes that are outside the address size. */
982 	if (insn->addr_bytes == 2)
983 		*eff_addr = regs_get_register(regs, *regoff) & 0xffff;
984 	else if (insn->addr_bytes == 4)
985 		*eff_addr = regs_get_register(regs, *regoff) & 0xffffffff;
986 	else /* 64-bit address */
987 		*eff_addr = regs_get_register(regs, *regoff);
988 
989 	return 0;
990 }
991 
992 /**
993  * get_eff_addr_modrm() - Obtain referenced effective address via ModRM
994  * @insn:	Instruction. Must be valid.
995  * @regs:	Register values as seen when entering kernel mode
996  * @regoff:	Obtained operand offset, in pt_regs, associated with segment
997  * @eff_addr:	Obtained effective address
998  *
999  * Obtain the effective address referenced by the ModRM byte of @insn. After
1000  * identifying the registers involved in the register-indirect memory reference,
1001  * its value is obtained from the operands in @regs. The computed address is
1002  * stored @eff_addr. Also, the register operand that indicates the associated
1003  * segment is stored in @regoff, this parameter can later be used to determine
1004  * such segment.
1005  *
1006  * Returns:
1007  *
1008  * 0 on success. @eff_addr will have the referenced effective address. @regoff
1009  * will have a register, as an offset from the base of pt_regs, that can be used
1010  * to resolve the associated segment.
1011  *
1012  * -EINVAL on error.
1013  */
1014 static int get_eff_addr_modrm(struct insn *insn, struct pt_regs *regs,
1015 			      int *regoff, long *eff_addr)
1016 {
1017 	long tmp;
1018 	int ret;
1019 
1020 	if (insn->addr_bytes != 8 && insn->addr_bytes != 4)
1021 		return -EINVAL;
1022 
1023 	ret = insn_get_modrm(insn);
1024 	if (ret)
1025 		return ret;
1026 
1027 	if (X86_MODRM_MOD(insn->modrm.value) > 2)
1028 		return -EINVAL;
1029 
1030 	*regoff = get_reg_offset(insn, regs, REG_TYPE_RM);
1031 
1032 	/*
1033 	 * -EDOM means that we must ignore the address_offset. In such a case,
1034 	 * in 64-bit mode the effective address relative to the rIP of the
1035 	 * following instruction.
1036 	 */
1037 	if (*regoff == -EDOM) {
1038 		if (any_64bit_mode(regs))
1039 			tmp = regs->ip + insn->length;
1040 		else
1041 			tmp = 0;
1042 	} else if (*regoff < 0) {
1043 		return -EINVAL;
1044 	} else {
1045 		tmp = regs_get_register(regs, *regoff);
1046 	}
1047 
1048 	if (insn->addr_bytes == 4) {
1049 		int addr32 = (int)(tmp & 0xffffffff) + insn->displacement.value;
1050 
1051 		*eff_addr = addr32 & 0xffffffff;
1052 	} else {
1053 		*eff_addr = tmp + insn->displacement.value;
1054 	}
1055 
1056 	return 0;
1057 }
1058 
1059 /**
1060  * get_eff_addr_modrm_16() - Obtain referenced effective address via ModRM
1061  * @insn:	Instruction. Must be valid.
1062  * @regs:	Register values as seen when entering kernel mode
1063  * @regoff:	Obtained operand offset, in pt_regs, associated with segment
1064  * @eff_addr:	Obtained effective address
1065  *
1066  * Obtain the 16-bit effective address referenced by the ModRM byte of @insn.
1067  * After identifying the registers involved in the register-indirect memory
1068  * reference, its value is obtained from the operands in @regs. The computed
1069  * address is stored @eff_addr. Also, the register operand that indicates
1070  * the associated segment is stored in @regoff, this parameter can later be used
1071  * to determine such segment.
1072  *
1073  * Returns:
1074  *
1075  * 0 on success. @eff_addr will have the referenced effective address. @regoff
1076  * will have a register, as an offset from the base of pt_regs, that can be used
1077  * to resolve the associated segment.
1078  *
1079  * -EINVAL on error.
1080  */
1081 static int get_eff_addr_modrm_16(struct insn *insn, struct pt_regs *regs,
1082 				 int *regoff, short *eff_addr)
1083 {
1084 	int addr_offset1, addr_offset2, ret;
1085 	short addr1 = 0, addr2 = 0, displacement;
1086 
1087 	if (insn->addr_bytes != 2)
1088 		return -EINVAL;
1089 
1090 	insn_get_modrm(insn);
1091 
1092 	if (!insn->modrm.nbytes)
1093 		return -EINVAL;
1094 
1095 	if (X86_MODRM_MOD(insn->modrm.value) > 2)
1096 		return -EINVAL;
1097 
1098 	ret = get_reg_offset_16(insn, regs, &addr_offset1, &addr_offset2);
1099 	if (ret < 0)
1100 		return -EINVAL;
1101 
1102 	/*
1103 	 * Don't fail on invalid offset values. They might be invalid because
1104 	 * they cannot be used for this particular value of ModRM. Instead, use
1105 	 * them in the computation only if they contain a valid value.
1106 	 */
1107 	if (addr_offset1 != -EDOM)
1108 		addr1 = regs_get_register(regs, addr_offset1) & 0xffff;
1109 
1110 	if (addr_offset2 != -EDOM)
1111 		addr2 = regs_get_register(regs, addr_offset2) & 0xffff;
1112 
1113 	displacement = insn->displacement.value & 0xffff;
1114 	*eff_addr = addr1 + addr2 + displacement;
1115 
1116 	/*
1117 	 * The first operand register could indicate to use of either SS or DS
1118 	 * registers to obtain the segment selector.  The second operand
1119 	 * register can only indicate the use of DS. Thus, the first operand
1120 	 * will be used to obtain the segment selector.
1121 	 */
1122 	*regoff = addr_offset1;
1123 
1124 	return 0;
1125 }
1126 
1127 /**
1128  * get_eff_addr_sib() - Obtain referenced effective address via SIB
1129  * @insn:	Instruction. Must be valid.
1130  * @regs:	Register values as seen when entering kernel mode
1131  * @regoff:	Obtained operand offset, in pt_regs, associated with segment
1132  * @eff_addr:	Obtained effective address
1133  *
1134  * Obtain the effective address referenced by the SIB byte of @insn. After
1135  * identifying the registers involved in the indexed, register-indirect memory
1136  * reference, its value is obtained from the operands in @regs. The computed
1137  * address is stored @eff_addr. Also, the register operand that indicates the
1138  * associated segment is stored in @regoff, this parameter can later be used to
1139  * determine such segment.
1140  *
1141  * Returns:
1142  *
1143  * 0 on success. @eff_addr will have the referenced effective address.
1144  * @base_offset will have a register, as an offset from the base of pt_regs,
1145  * that can be used to resolve the associated segment.
1146  *
1147  * Negative value on error.
1148  */
1149 static int get_eff_addr_sib(struct insn *insn, struct pt_regs *regs,
1150 			    int *base_offset, long *eff_addr)
1151 {
1152 	long base, indx;
1153 	int indx_offset;
1154 	int ret;
1155 
1156 	if (insn->addr_bytes != 8 && insn->addr_bytes != 4)
1157 		return -EINVAL;
1158 
1159 	ret = insn_get_modrm(insn);
1160 	if (ret)
1161 		return ret;
1162 
1163 	if (!insn->modrm.nbytes)
1164 		return -EINVAL;
1165 
1166 	if (X86_MODRM_MOD(insn->modrm.value) > 2)
1167 		return -EINVAL;
1168 
1169 	ret = insn_get_sib(insn);
1170 	if (ret)
1171 		return ret;
1172 
1173 	if (!insn->sib.nbytes)
1174 		return -EINVAL;
1175 
1176 	*base_offset = get_reg_offset(insn, regs, REG_TYPE_BASE);
1177 	indx_offset = get_reg_offset(insn, regs, REG_TYPE_INDEX);
1178 
1179 	/*
1180 	 * Negative values in the base and index offset means an error when
1181 	 * decoding the SIB byte. Except -EDOM, which means that the registers
1182 	 * should not be used in the address computation.
1183 	 */
1184 	if (*base_offset == -EDOM)
1185 		base = 0;
1186 	else if (*base_offset < 0)
1187 		return -EINVAL;
1188 	else
1189 		base = regs_get_register(regs, *base_offset);
1190 
1191 	if (indx_offset == -EDOM)
1192 		indx = 0;
1193 	else if (indx_offset < 0)
1194 		return -EINVAL;
1195 	else
1196 		indx = regs_get_register(regs, indx_offset);
1197 
1198 	if (insn->addr_bytes == 4) {
1199 		int addr32, base32, idx32;
1200 
1201 		base32 = base & 0xffffffff;
1202 		idx32 = indx & 0xffffffff;
1203 
1204 		addr32 = base32 + idx32 * (1 << X86_SIB_SCALE(insn->sib.value));
1205 		addr32 += insn->displacement.value;
1206 
1207 		*eff_addr = addr32 & 0xffffffff;
1208 	} else {
1209 		*eff_addr = base + indx * (1 << X86_SIB_SCALE(insn->sib.value));
1210 		*eff_addr += insn->displacement.value;
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 /**
1217  * get_addr_ref_16() - Obtain the 16-bit address referred by instruction
1218  * @insn:	Instruction containing ModRM byte and displacement
1219  * @regs:	Register values as seen when entering kernel mode
1220  *
1221  * This function is to be used with 16-bit address encodings. Obtain the memory
1222  * address referred by the instruction's ModRM and displacement bytes. Also, the
1223  * segment used as base is determined by either any segment override prefixes in
1224  * @insn or the default segment of the registers involved in the address
1225  * computation. In protected mode, segment limits are enforced.
1226  *
1227  * Returns:
1228  *
1229  * Linear address referenced by the instruction operands on success.
1230  *
1231  * -1L on error.
1232  */
1233 static void __user *get_addr_ref_16(struct insn *insn, struct pt_regs *regs)
1234 {
1235 	unsigned long linear_addr = -1L, seg_base, seg_limit;
1236 	int ret, regoff;
1237 	short eff_addr;
1238 	long tmp;
1239 
1240 	if (insn_get_displacement(insn))
1241 		goto out;
1242 
1243 	if (insn->addr_bytes != 2)
1244 		goto out;
1245 
1246 	if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1247 		ret = get_eff_addr_reg(insn, regs, &regoff, &tmp);
1248 		if (ret)
1249 			goto out;
1250 
1251 		eff_addr = tmp;
1252 	} else {
1253 		ret = get_eff_addr_modrm_16(insn, regs, &regoff, &eff_addr);
1254 		if (ret)
1255 			goto out;
1256 	}
1257 
1258 	ret = get_seg_base_limit(insn, regs, regoff, &seg_base, &seg_limit);
1259 	if (ret)
1260 		goto out;
1261 
1262 	/*
1263 	 * Before computing the linear address, make sure the effective address
1264 	 * is within the limits of the segment. In virtual-8086 mode, segment
1265 	 * limits are not enforced. In such a case, the segment limit is -1L to
1266 	 * reflect this fact.
1267 	 */
1268 	if ((unsigned long)(eff_addr & 0xffff) > seg_limit)
1269 		goto out;
1270 
1271 	linear_addr = (unsigned long)(eff_addr & 0xffff) + seg_base;
1272 
1273 	/* Limit linear address to 20 bits */
1274 	if (v8086_mode(regs))
1275 		linear_addr &= 0xfffff;
1276 
1277 out:
1278 	return (void __user *)linear_addr;
1279 }
1280 
1281 /**
1282  * get_addr_ref_32() - Obtain a 32-bit linear address
1283  * @insn:	Instruction with ModRM, SIB bytes and displacement
1284  * @regs:	Register values as seen when entering kernel mode
1285  *
1286  * This function is to be used with 32-bit address encodings to obtain the
1287  * linear memory address referred by the instruction's ModRM, SIB,
1288  * displacement bytes and segment base address, as applicable. If in protected
1289  * mode, segment limits are enforced.
1290  *
1291  * Returns:
1292  *
1293  * Linear address referenced by instruction and registers on success.
1294  *
1295  * -1L on error.
1296  */
1297 static void __user *get_addr_ref_32(struct insn *insn, struct pt_regs *regs)
1298 {
1299 	unsigned long linear_addr = -1L, seg_base, seg_limit;
1300 	int eff_addr, regoff;
1301 	long tmp;
1302 	int ret;
1303 
1304 	if (insn->addr_bytes != 4)
1305 		goto out;
1306 
1307 	if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1308 		ret = get_eff_addr_reg(insn, regs, &regoff, &tmp);
1309 		if (ret)
1310 			goto out;
1311 
1312 		eff_addr = tmp;
1313 
1314 	} else {
1315 		if (insn->sib.nbytes) {
1316 			ret = get_eff_addr_sib(insn, regs, &regoff, &tmp);
1317 			if (ret)
1318 				goto out;
1319 
1320 			eff_addr = tmp;
1321 		} else {
1322 			ret = get_eff_addr_modrm(insn, regs, &regoff, &tmp);
1323 			if (ret)
1324 				goto out;
1325 
1326 			eff_addr = tmp;
1327 		}
1328 	}
1329 
1330 	ret = get_seg_base_limit(insn, regs, regoff, &seg_base, &seg_limit);
1331 	if (ret)
1332 		goto out;
1333 
1334 	/*
1335 	 * In protected mode, before computing the linear address, make sure
1336 	 * the effective address is within the limits of the segment.
1337 	 * 32-bit addresses can be used in long and virtual-8086 modes if an
1338 	 * address override prefix is used. In such cases, segment limits are
1339 	 * not enforced. When in virtual-8086 mode, the segment limit is -1L
1340 	 * to reflect this situation.
1341 	 *
1342 	 * After computed, the effective address is treated as an unsigned
1343 	 * quantity.
1344 	 */
1345 	if (!any_64bit_mode(regs) && ((unsigned int)eff_addr > seg_limit))
1346 		goto out;
1347 
1348 	/*
1349 	 * Even though 32-bit address encodings are allowed in virtual-8086
1350 	 * mode, the address range is still limited to [0x-0xffff].
1351 	 */
1352 	if (v8086_mode(regs) && (eff_addr & ~0xffff))
1353 		goto out;
1354 
1355 	/*
1356 	 * Data type long could be 64 bits in size. Ensure that our 32-bit
1357 	 * effective address is not sign-extended when computing the linear
1358 	 * address.
1359 	 */
1360 	linear_addr = (unsigned long)(eff_addr & 0xffffffff) + seg_base;
1361 
1362 	/* Limit linear address to 20 bits */
1363 	if (v8086_mode(regs))
1364 		linear_addr &= 0xfffff;
1365 
1366 out:
1367 	return (void __user *)linear_addr;
1368 }
1369 
1370 /**
1371  * get_addr_ref_64() - Obtain a 64-bit linear address
1372  * @insn:	Instruction struct with ModRM and SIB bytes and displacement
1373  * @regs:	Structure with register values as seen when entering kernel mode
1374  *
1375  * This function is to be used with 64-bit address encodings to obtain the
1376  * linear memory address referred by the instruction's ModRM, SIB,
1377  * displacement bytes and segment base address, as applicable.
1378  *
1379  * Returns:
1380  *
1381  * Linear address referenced by instruction and registers on success.
1382  *
1383  * -1L on error.
1384  */
1385 #ifndef CONFIG_X86_64
1386 static void __user *get_addr_ref_64(struct insn *insn, struct pt_regs *regs)
1387 {
1388 	return (void __user *)-1L;
1389 }
1390 #else
1391 static void __user *get_addr_ref_64(struct insn *insn, struct pt_regs *regs)
1392 {
1393 	unsigned long linear_addr = -1L, seg_base;
1394 	int regoff, ret;
1395 	long eff_addr;
1396 
1397 	if (insn->addr_bytes != 8)
1398 		goto out;
1399 
1400 	if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1401 		ret = get_eff_addr_reg(insn, regs, &regoff, &eff_addr);
1402 		if (ret)
1403 			goto out;
1404 
1405 	} else {
1406 		if (insn->sib.nbytes) {
1407 			ret = get_eff_addr_sib(insn, regs, &regoff, &eff_addr);
1408 			if (ret)
1409 				goto out;
1410 		} else {
1411 			ret = get_eff_addr_modrm(insn, regs, &regoff, &eff_addr);
1412 			if (ret)
1413 				goto out;
1414 		}
1415 
1416 	}
1417 
1418 	ret = get_seg_base_limit(insn, regs, regoff, &seg_base, NULL);
1419 	if (ret)
1420 		goto out;
1421 
1422 	linear_addr = (unsigned long)eff_addr + seg_base;
1423 
1424 out:
1425 	return (void __user *)linear_addr;
1426 }
1427 #endif /* CONFIG_X86_64 */
1428 
1429 /**
1430  * insn_get_addr_ref() - Obtain the linear address referred by instruction
1431  * @insn:	Instruction structure containing ModRM byte and displacement
1432  * @regs:	Structure with register values as seen when entering kernel mode
1433  *
1434  * Obtain the linear address referred by the instruction's ModRM, SIB and
1435  * displacement bytes, and segment base, as applicable. In protected mode,
1436  * segment limits are enforced.
1437  *
1438  * Returns:
1439  *
1440  * Linear address referenced by instruction and registers on success.
1441  *
1442  * -1L on error.
1443  */
1444 void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs)
1445 {
1446 	if (!insn || !regs)
1447 		return (void __user *)-1L;
1448 
1449 	if (insn_get_opcode(insn))
1450 		return (void __user *)-1L;
1451 
1452 	switch (insn->addr_bytes) {
1453 	case 2:
1454 		return get_addr_ref_16(insn, regs);
1455 	case 4:
1456 		return get_addr_ref_32(insn, regs);
1457 	case 8:
1458 		return get_addr_ref_64(insn, regs);
1459 	default:
1460 		return (void __user *)-1L;
1461 	}
1462 }
1463 
1464 int insn_get_effective_ip(struct pt_regs *regs, unsigned long *ip)
1465 {
1466 	unsigned long seg_base = 0;
1467 
1468 	/*
1469 	 * If not in user-space long mode, a custom code segment could be in
1470 	 * use. This is true in protected mode (if the process defined a local
1471 	 * descriptor table), or virtual-8086 mode. In most of the cases
1472 	 * seg_base will be zero as in USER_CS.
1473 	 */
1474 	if (!user_64bit_mode(regs)) {
1475 		seg_base = insn_get_seg_base(regs, INAT_SEG_REG_CS);
1476 		if (seg_base == -1L)
1477 			return -EINVAL;
1478 	}
1479 
1480 	*ip = seg_base + regs->ip;
1481 
1482 	return 0;
1483 }
1484 
1485 /**
1486  * insn_fetch_from_user() - Copy instruction bytes from user-space memory
1487  * @regs:	Structure with register values as seen when entering kernel mode
1488  * @buf:	Array to store the fetched instruction
1489  *
1490  * Gets the linear address of the instruction and copies the instruction bytes
1491  * to the buf.
1492  *
1493  * Returns:
1494  *
1495  * - number of instruction bytes copied.
1496  * - 0 if nothing was copied.
1497  * - -EINVAL if the linear address of the instruction could not be calculated
1498  */
1499 int insn_fetch_from_user(struct pt_regs *regs, unsigned char buf[MAX_INSN_SIZE])
1500 {
1501 	unsigned long ip;
1502 	int not_copied;
1503 
1504 	if (insn_get_effective_ip(regs, &ip))
1505 		return -EINVAL;
1506 
1507 	not_copied = copy_from_user(buf, (void __user *)ip, MAX_INSN_SIZE);
1508 
1509 	return MAX_INSN_SIZE - not_copied;
1510 }
1511 
1512 /**
1513  * insn_fetch_from_user_inatomic() - Copy instruction bytes from user-space memory
1514  *                                   while in atomic code
1515  * @regs:	Structure with register values as seen when entering kernel mode
1516  * @buf:	Array to store the fetched instruction
1517  *
1518  * Gets the linear address of the instruction and copies the instruction bytes
1519  * to the buf. This function must be used in atomic context.
1520  *
1521  * Returns:
1522  *
1523  *  - number of instruction bytes copied.
1524  *  - 0 if nothing was copied.
1525  *  - -EINVAL if the linear address of the instruction could not be calculated.
1526  */
1527 int insn_fetch_from_user_inatomic(struct pt_regs *regs, unsigned char buf[MAX_INSN_SIZE])
1528 {
1529 	unsigned long ip;
1530 	int not_copied;
1531 
1532 	if (insn_get_effective_ip(regs, &ip))
1533 		return -EINVAL;
1534 
1535 	not_copied = __copy_from_user_inatomic(buf, (void __user *)ip, MAX_INSN_SIZE);
1536 
1537 	return MAX_INSN_SIZE - not_copied;
1538 }
1539 
1540 /**
1541  * insn_decode_from_regs() - Decode an instruction
1542  * @insn:	Structure to store decoded instruction
1543  * @regs:	Structure with register values as seen when entering kernel mode
1544  * @buf:	Buffer containing the instruction bytes
1545  * @buf_size:   Number of instruction bytes available in buf
1546  *
1547  * Decodes the instruction provided in buf and stores the decoding results in
1548  * insn. Also determines the correct address and operand sizes.
1549  *
1550  * Returns:
1551  *
1552  * True if instruction was decoded, False otherwise.
1553  */
1554 bool insn_decode_from_regs(struct insn *insn, struct pt_regs *regs,
1555 			   unsigned char buf[MAX_INSN_SIZE], int buf_size)
1556 {
1557 	int seg_defs;
1558 
1559 	insn_init(insn, buf, buf_size, user_64bit_mode(regs));
1560 
1561 	/*
1562 	 * Override the default operand and address sizes with what is specified
1563 	 * in the code segment descriptor. The instruction decoder only sets
1564 	 * the address size it to either 4 or 8 address bytes and does nothing
1565 	 * for the operand bytes. This OK for most of the cases, but we could
1566 	 * have special cases where, for instance, a 16-bit code segment
1567 	 * descriptor is used.
1568 	 * If there is an address override prefix, the instruction decoder
1569 	 * correctly updates these values, even for 16-bit defaults.
1570 	 */
1571 	seg_defs = insn_get_code_seg_params(regs);
1572 	if (seg_defs == -EINVAL)
1573 		return false;
1574 
1575 	insn->addr_bytes = INSN_CODE_SEG_ADDR_SZ(seg_defs);
1576 	insn->opnd_bytes = INSN_CODE_SEG_OPND_SZ(seg_defs);
1577 
1578 	if (insn_get_length(insn))
1579 		return false;
1580 
1581 	if (buf_size < insn->length)
1582 		return false;
1583 
1584 	return true;
1585 }
1586 
1587 /**
1588  * insn_decode_mmio() - Decode a MMIO instruction
1589  * @insn:	Structure to store decoded instruction
1590  * @bytes:	Returns size of memory operand
1591  *
1592  * Decodes instruction that used for Memory-mapped I/O.
1593  *
1594  * Returns:
1595  *
1596  * Type of the instruction. Size of the memory operand is stored in
1597  * @bytes. If decode failed, MMIO_DECODE_FAILED returned.
1598  */
1599 enum mmio_type insn_decode_mmio(struct insn *insn, int *bytes)
1600 {
1601 	enum mmio_type type = MMIO_DECODE_FAILED;
1602 
1603 	*bytes = 0;
1604 
1605 	if (insn_get_opcode(insn))
1606 		return MMIO_DECODE_FAILED;
1607 
1608 	switch (insn->opcode.bytes[0]) {
1609 	case 0x88: /* MOV m8,r8 */
1610 		*bytes = 1;
1611 		fallthrough;
1612 	case 0x89: /* MOV m16/m32/m64, r16/m32/m64 */
1613 		if (!*bytes)
1614 			*bytes = insn->opnd_bytes;
1615 		type = MMIO_WRITE;
1616 		break;
1617 
1618 	case 0xc6: /* MOV m8, imm8 */
1619 		*bytes = 1;
1620 		fallthrough;
1621 	case 0xc7: /* MOV m16/m32/m64, imm16/imm32/imm64 */
1622 		if (!*bytes)
1623 			*bytes = insn->opnd_bytes;
1624 		type = MMIO_WRITE_IMM;
1625 		break;
1626 
1627 	case 0x8a: /* MOV r8, m8 */
1628 		*bytes = 1;
1629 		fallthrough;
1630 	case 0x8b: /* MOV r16/r32/r64, m16/m32/m64 */
1631 		if (!*bytes)
1632 			*bytes = insn->opnd_bytes;
1633 		type = MMIO_READ;
1634 		break;
1635 
1636 	case 0xa4: /* MOVS m8, m8 */
1637 		*bytes = 1;
1638 		fallthrough;
1639 	case 0xa5: /* MOVS m16/m32/m64, m16/m32/m64 */
1640 		if (!*bytes)
1641 			*bytes = insn->opnd_bytes;
1642 		type = MMIO_MOVS;
1643 		break;
1644 
1645 	case 0x0f: /* Two-byte instruction */
1646 		switch (insn->opcode.bytes[1]) {
1647 		case 0xb6: /* MOVZX r16/r32/r64, m8 */
1648 			*bytes = 1;
1649 			fallthrough;
1650 		case 0xb7: /* MOVZX r32/r64, m16 */
1651 			if (!*bytes)
1652 				*bytes = 2;
1653 			type = MMIO_READ_ZERO_EXTEND;
1654 			break;
1655 
1656 		case 0xbe: /* MOVSX r16/r32/r64, m8 */
1657 			*bytes = 1;
1658 			fallthrough;
1659 		case 0xbf: /* MOVSX r32/r64, m16 */
1660 			if (!*bytes)
1661 				*bytes = 2;
1662 			type = MMIO_READ_SIGN_EXTEND;
1663 			break;
1664 		}
1665 		break;
1666 	}
1667 
1668 	return type;
1669 }
1670