Linux 5.7.6
[linux/fpc-iii.git] / arch / x86 / lib / insn-eval.c
blob31600d851fd8e3c327f524a87443debdee52be71
1 /*
2 * Utility functions for x86 operand and address decoding
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>
18 #undef pr_fmt
19 #define pr_fmt(fmt) "insn: " fmt
21 enum reg_type {
22 REG_TYPE_RM = 0,
23 REG_TYPE_INDEX,
24 REG_TYPE_BASE,
27 /**
28 * is_string_insn() - Determine if instruction is a string instruction
29 * @insn: Instruction containing the opcode to inspect
31 * Returns:
33 * true if the instruction, determined by the opcode, is any of the
34 * string instructions as defined in the Intel Software Development manual.
35 * False otherwise.
37 static bool is_string_insn(struct insn *insn)
39 insn_get_opcode(insn);
41 /* All string instructions have a 1-byte opcode. */
42 if (insn->opcode.nbytes != 1)
43 return false;
45 switch (insn->opcode.bytes[0]) {
46 case 0x6c ... 0x6f: /* INS, OUTS */
47 case 0xa4 ... 0xa7: /* MOVS, CMPS */
48 case 0xaa ... 0xaf: /* STOS, LODS, SCAS */
49 return true;
50 default:
51 return false;
55 /**
56 * get_seg_reg_override_idx() - obtain segment register override index
57 * @insn: Valid instruction with segment override prefixes
59 * Inspect the instruction prefixes in @insn and find segment overrides, if any.
61 * Returns:
63 * A constant identifying the segment register to use, among CS, SS, DS,
64 * ES, FS, or GS. INAT_SEG_REG_DEFAULT is returned if no segment override
65 * prefixes were found.
67 * -EINVAL in case of error.
69 static int get_seg_reg_override_idx(struct insn *insn)
71 int idx = INAT_SEG_REG_DEFAULT;
72 int num_overrides = 0, i;
74 insn_get_prefixes(insn);
76 /* Look for any segment override prefixes. */
77 for (i = 0; i < insn->prefixes.nbytes; i++) {
78 insn_attr_t attr;
80 attr = inat_get_opcode_attribute(insn->prefixes.bytes[i]);
81 switch (attr) {
82 case INAT_MAKE_PREFIX(INAT_PFX_CS):
83 idx = INAT_SEG_REG_CS;
84 num_overrides++;
85 break;
86 case INAT_MAKE_PREFIX(INAT_PFX_SS):
87 idx = INAT_SEG_REG_SS;
88 num_overrides++;
89 break;
90 case INAT_MAKE_PREFIX(INAT_PFX_DS):
91 idx = INAT_SEG_REG_DS;
92 num_overrides++;
93 break;
94 case INAT_MAKE_PREFIX(INAT_PFX_ES):
95 idx = INAT_SEG_REG_ES;
96 num_overrides++;
97 break;
98 case INAT_MAKE_PREFIX(INAT_PFX_FS):
99 idx = INAT_SEG_REG_FS;
100 num_overrides++;
101 break;
102 case INAT_MAKE_PREFIX(INAT_PFX_GS):
103 idx = INAT_SEG_REG_GS;
104 num_overrides++;
105 break;
106 /* No default action needed. */
110 /* More than one segment override prefix leads to undefined behavior. */
111 if (num_overrides > 1)
112 return -EINVAL;
114 return idx;
118 * check_seg_overrides() - check if segment override prefixes are allowed
119 * @insn: Valid instruction with segment override prefixes
120 * @regoff: Operand offset, in pt_regs, for which the check is performed
122 * For a particular register used in register-indirect addressing, determine if
123 * segment override prefixes can be used. Specifically, no overrides are allowed
124 * for rDI if used with a string instruction.
126 * Returns:
128 * True if segment override prefixes can be used with the register indicated
129 * in @regoff. False if otherwise.
131 static bool check_seg_overrides(struct insn *insn, int regoff)
133 if (regoff == offsetof(struct pt_regs, di) && is_string_insn(insn))
134 return false;
136 return true;
140 * resolve_default_seg() - resolve default segment register index for an operand
141 * @insn: Instruction with opcode and address size. Must be valid.
142 * @regs: Register values as seen when entering kernel mode
143 * @off: Operand offset, in pt_regs, for which resolution is needed
145 * Resolve the default segment register index associated with the instruction
146 * operand register indicated by @off. Such index is resolved based on defaults
147 * described in the Intel Software Development Manual.
149 * Returns:
151 * If in protected mode, a constant identifying the segment register to use,
152 * among CS, SS, ES or DS. If in long mode, INAT_SEG_REG_IGNORE.
154 * -EINVAL in case of error.
156 static int resolve_default_seg(struct insn *insn, struct pt_regs *regs, int off)
158 if (any_64bit_mode(regs))
159 return INAT_SEG_REG_IGNORE;
161 * Resolve the default segment register as described in Section 3.7.4
162 * of the Intel Software Development Manual Vol. 1:
164 * + DS for all references involving r[ABCD]X, and rSI.
165 * + If used in a string instruction, ES for rDI. Otherwise, DS.
166 * + AX, CX and DX are not valid register operands in 16-bit address
167 * encodings but are valid for 32-bit and 64-bit encodings.
168 * + -EDOM is reserved to identify for cases in which no register
169 * is used (i.e., displacement-only addressing). Use DS.
170 * + SS for rSP or rBP.
171 * + CS for rIP.
174 switch (off) {
175 case offsetof(struct pt_regs, ax):
176 case offsetof(struct pt_regs, cx):
177 case offsetof(struct pt_regs, dx):
178 /* Need insn to verify address size. */
179 if (insn->addr_bytes == 2)
180 return -EINVAL;
182 /* fall through */
184 case -EDOM:
185 case offsetof(struct pt_regs, bx):
186 case offsetof(struct pt_regs, si):
187 return INAT_SEG_REG_DS;
189 case offsetof(struct pt_regs, di):
190 if (is_string_insn(insn))
191 return INAT_SEG_REG_ES;
192 return INAT_SEG_REG_DS;
194 case offsetof(struct pt_regs, bp):
195 case offsetof(struct pt_regs, sp):
196 return INAT_SEG_REG_SS;
198 case offsetof(struct pt_regs, ip):
199 return INAT_SEG_REG_CS;
201 default:
202 return -EINVAL;
207 * resolve_seg_reg() - obtain segment register index
208 * @insn: Instruction with operands
209 * @regs: Register values as seen when entering kernel mode
210 * @regoff: Operand offset, in pt_regs, used to deterimine segment register
212 * Determine the segment register associated with the operands and, if
213 * applicable, prefixes and the instruction pointed by @insn.
215 * The segment register associated to an operand used in register-indirect
216 * addressing depends on:
218 * a) Whether running in long mode (in such a case segments are ignored, except
219 * if FS or GS are used).
221 * b) Whether segment override prefixes can be used. Certain instructions and
222 * registers do not allow override prefixes.
224 * c) Whether segment overrides prefixes are found in the instruction prefixes.
226 * d) If there are not segment override prefixes or they cannot be used, the
227 * default segment register associated with the operand register is used.
229 * The function checks first if segment override prefixes can be used with the
230 * operand indicated by @regoff. If allowed, obtain such overridden segment
231 * register index. Lastly, if not prefixes were found or cannot be used, resolve
232 * the segment register index to use based on the defaults described in the
233 * Intel documentation. In long mode, all segment register indexes will be
234 * ignored, except if overrides were found for FS or GS. All these operations
235 * are done using helper functions.
237 * The operand register, @regoff, is represented as the offset from the base of
238 * pt_regs.
240 * As stated, the main use of this function is to determine the segment register
241 * index based on the instruction, its operands and prefixes. Hence, @insn
242 * must be valid. However, if @regoff indicates rIP, we don't need to inspect
243 * @insn at all as in this case CS is used in all cases. This case is checked
244 * before proceeding further.
246 * Please note that this function does not return the value in the segment
247 * register (i.e., the segment selector) but our defined index. The segment
248 * selector needs to be obtained using get_segment_selector() and passing the
249 * segment register index resolved by this function.
251 * Returns:
253 * An index identifying the segment register to use, among CS, SS, DS,
254 * ES, FS, or GS. INAT_SEG_REG_IGNORE is returned if running in long mode.
256 * -EINVAL in case of error.
258 static int resolve_seg_reg(struct insn *insn, struct pt_regs *regs, int regoff)
260 int idx;
263 * In the unlikely event of having to resolve the segment register
264 * index for rIP, do it first. Segment override prefixes should not
265 * be used. Hence, it is not necessary to inspect the instruction,
266 * which may be invalid at this point.
268 if (regoff == offsetof(struct pt_regs, ip)) {
269 if (any_64bit_mode(regs))
270 return INAT_SEG_REG_IGNORE;
271 else
272 return INAT_SEG_REG_CS;
275 if (!insn)
276 return -EINVAL;
278 if (!check_seg_overrides(insn, regoff))
279 return resolve_default_seg(insn, regs, regoff);
281 idx = get_seg_reg_override_idx(insn);
282 if (idx < 0)
283 return idx;
285 if (idx == INAT_SEG_REG_DEFAULT)
286 return resolve_default_seg(insn, regs, regoff);
289 * In long mode, segment override prefixes are ignored, except for
290 * overrides for FS and GS.
292 if (any_64bit_mode(regs)) {
293 if (idx != INAT_SEG_REG_FS &&
294 idx != INAT_SEG_REG_GS)
295 idx = INAT_SEG_REG_IGNORE;
298 return idx;
302 * get_segment_selector() - obtain segment selector
303 * @regs: Register values as seen when entering kernel mode
304 * @seg_reg_idx: Segment register index to use
306 * Obtain the segment selector from any of the CS, SS, DS, ES, FS, GS segment
307 * registers. In CONFIG_X86_32, the segment is obtained from either pt_regs or
308 * kernel_vm86_regs as applicable. In CONFIG_X86_64, CS and SS are obtained
309 * from pt_regs. DS, ES, FS and GS are obtained by reading the actual CPU
310 * registers. This done for only for completeness as in CONFIG_X86_64 segment
311 * registers are ignored.
313 * Returns:
315 * Value of the segment selector, including null when running in
316 * long mode.
318 * -EINVAL on error.
320 static short get_segment_selector(struct pt_regs *regs, int seg_reg_idx)
322 #ifdef CONFIG_X86_64
323 unsigned short sel;
325 switch (seg_reg_idx) {
326 case INAT_SEG_REG_IGNORE:
327 return 0;
328 case INAT_SEG_REG_CS:
329 return (unsigned short)(regs->cs & 0xffff);
330 case INAT_SEG_REG_SS:
331 return (unsigned short)(regs->ss & 0xffff);
332 case INAT_SEG_REG_DS:
333 savesegment(ds, sel);
334 return sel;
335 case INAT_SEG_REG_ES:
336 savesegment(es, sel);
337 return sel;
338 case INAT_SEG_REG_FS:
339 savesegment(fs, sel);
340 return sel;
341 case INAT_SEG_REG_GS:
342 savesegment(gs, sel);
343 return sel;
344 default:
345 return -EINVAL;
347 #else /* CONFIG_X86_32 */
348 struct kernel_vm86_regs *vm86regs = (struct kernel_vm86_regs *)regs;
350 if (v8086_mode(regs)) {
351 switch (seg_reg_idx) {
352 case INAT_SEG_REG_CS:
353 return (unsigned short)(regs->cs & 0xffff);
354 case INAT_SEG_REG_SS:
355 return (unsigned short)(regs->ss & 0xffff);
356 case INAT_SEG_REG_DS:
357 return vm86regs->ds;
358 case INAT_SEG_REG_ES:
359 return vm86regs->es;
360 case INAT_SEG_REG_FS:
361 return vm86regs->fs;
362 case INAT_SEG_REG_GS:
363 return vm86regs->gs;
364 case INAT_SEG_REG_IGNORE:
365 /* fall through */
366 default:
367 return -EINVAL;
371 switch (seg_reg_idx) {
372 case INAT_SEG_REG_CS:
373 return (unsigned short)(regs->cs & 0xffff);
374 case INAT_SEG_REG_SS:
375 return (unsigned short)(regs->ss & 0xffff);
376 case INAT_SEG_REG_DS:
377 return (unsigned short)(regs->ds & 0xffff);
378 case INAT_SEG_REG_ES:
379 return (unsigned short)(regs->es & 0xffff);
380 case INAT_SEG_REG_FS:
381 return (unsigned short)(regs->fs & 0xffff);
382 case INAT_SEG_REG_GS:
384 * GS may or may not be in regs as per CONFIG_X86_32_LAZY_GS.
385 * The macro below takes care of both cases.
387 return get_user_gs(regs);
388 case INAT_SEG_REG_IGNORE:
389 /* fall through */
390 default:
391 return -EINVAL;
393 #endif /* CONFIG_X86_64 */
396 static int get_reg_offset(struct insn *insn, struct pt_regs *regs,
397 enum reg_type type)
399 int regno = 0;
401 static const int regoff[] = {
402 offsetof(struct pt_regs, ax),
403 offsetof(struct pt_regs, cx),
404 offsetof(struct pt_regs, dx),
405 offsetof(struct pt_regs, bx),
406 offsetof(struct pt_regs, sp),
407 offsetof(struct pt_regs, bp),
408 offsetof(struct pt_regs, si),
409 offsetof(struct pt_regs, di),
410 #ifdef CONFIG_X86_64
411 offsetof(struct pt_regs, r8),
412 offsetof(struct pt_regs, r9),
413 offsetof(struct pt_regs, r10),
414 offsetof(struct pt_regs, r11),
415 offsetof(struct pt_regs, r12),
416 offsetof(struct pt_regs, r13),
417 offsetof(struct pt_regs, r14),
418 offsetof(struct pt_regs, r15),
419 #endif
421 int nr_registers = ARRAY_SIZE(regoff);
423 * Don't possibly decode a 32-bit instructions as
424 * reading a 64-bit-only register.
426 if (IS_ENABLED(CONFIG_X86_64) && !insn->x86_64)
427 nr_registers -= 8;
429 switch (type) {
430 case REG_TYPE_RM:
431 regno = X86_MODRM_RM(insn->modrm.value);
434 * ModRM.mod == 0 and ModRM.rm == 5 means a 32-bit displacement
435 * follows the ModRM byte.
437 if (!X86_MODRM_MOD(insn->modrm.value) && regno == 5)
438 return -EDOM;
440 if (X86_REX_B(insn->rex_prefix.value))
441 regno += 8;
442 break;
444 case REG_TYPE_INDEX:
445 regno = X86_SIB_INDEX(insn->sib.value);
446 if (X86_REX_X(insn->rex_prefix.value))
447 regno += 8;
450 * If ModRM.mod != 3 and SIB.index = 4 the scale*index
451 * portion of the address computation is null. This is
452 * true only if REX.X is 0. In such a case, the SIB index
453 * is used in the address computation.
455 if (X86_MODRM_MOD(insn->modrm.value) != 3 && regno == 4)
456 return -EDOM;
457 break;
459 case REG_TYPE_BASE:
460 regno = X86_SIB_BASE(insn->sib.value);
462 * If ModRM.mod is 0 and SIB.base == 5, the base of the
463 * register-indirect addressing is 0. In this case, a
464 * 32-bit displacement follows the SIB byte.
466 if (!X86_MODRM_MOD(insn->modrm.value) && regno == 5)
467 return -EDOM;
469 if (X86_REX_B(insn->rex_prefix.value))
470 regno += 8;
471 break;
473 default:
474 pr_err_ratelimited("invalid register type: %d\n", type);
475 return -EINVAL;
478 if (regno >= nr_registers) {
479 WARN_ONCE(1, "decoded an instruction with an invalid register");
480 return -EINVAL;
482 return regoff[regno];
486 * get_reg_offset_16() - Obtain offset of register indicated by instruction
487 * @insn: Instruction containing ModRM byte
488 * @regs: Register values as seen when entering kernel mode
489 * @offs1: Offset of the first operand register
490 * @offs2: Offset of the second opeand register, if applicable
492 * Obtain the offset, in pt_regs, of the registers indicated by the ModRM byte
493 * in @insn. This function is to be used with 16-bit address encodings. The
494 * @offs1 and @offs2 will be written with the offset of the two registers
495 * indicated by the instruction. In cases where any of the registers is not
496 * referenced by the instruction, the value will be set to -EDOM.
498 * Returns:
500 * 0 on success, -EINVAL on error.
502 static int get_reg_offset_16(struct insn *insn, struct pt_regs *regs,
503 int *offs1, int *offs2)
506 * 16-bit addressing can use one or two registers. Specifics of
507 * encodings are given in Table 2-1. "16-Bit Addressing Forms with the
508 * ModR/M Byte" of the Intel Software Development Manual.
510 static const int regoff1[] = {
511 offsetof(struct pt_regs, bx),
512 offsetof(struct pt_regs, bx),
513 offsetof(struct pt_regs, bp),
514 offsetof(struct pt_regs, bp),
515 offsetof(struct pt_regs, si),
516 offsetof(struct pt_regs, di),
517 offsetof(struct pt_regs, bp),
518 offsetof(struct pt_regs, bx),
521 static const int regoff2[] = {
522 offsetof(struct pt_regs, si),
523 offsetof(struct pt_regs, di),
524 offsetof(struct pt_regs, si),
525 offsetof(struct pt_regs, di),
526 -EDOM,
527 -EDOM,
528 -EDOM,
529 -EDOM,
532 if (!offs1 || !offs2)
533 return -EINVAL;
535 /* Operand is a register, use the generic function. */
536 if (X86_MODRM_MOD(insn->modrm.value) == 3) {
537 *offs1 = insn_get_modrm_rm_off(insn, regs);
538 *offs2 = -EDOM;
539 return 0;
542 *offs1 = regoff1[X86_MODRM_RM(insn->modrm.value)];
543 *offs2 = regoff2[X86_MODRM_RM(insn->modrm.value)];
546 * If ModRM.mod is 0 and ModRM.rm is 110b, then we use displacement-
547 * only addressing. This means that no registers are involved in
548 * computing the effective address. Thus, ensure that the first
549 * register offset is invalild. The second register offset is already
550 * invalid under the aforementioned conditions.
552 if ((X86_MODRM_MOD(insn->modrm.value) == 0) &&
553 (X86_MODRM_RM(insn->modrm.value) == 6))
554 *offs1 = -EDOM;
556 return 0;
560 * get_desc() - Obtain contents of a segment descriptor
561 * @out: Segment descriptor contents on success
562 * @sel: Segment selector
564 * Given a segment selector, obtain a pointer to the segment descriptor.
565 * Both global and local descriptor tables are supported.
567 * Returns:
569 * True on success, false on failure.
571 * NULL on error.
573 static bool get_desc(struct desc_struct *out, unsigned short sel)
575 struct desc_ptr gdt_desc = {0, 0};
576 unsigned long desc_base;
578 #ifdef CONFIG_MODIFY_LDT_SYSCALL
579 if ((sel & SEGMENT_TI_MASK) == SEGMENT_LDT) {
580 bool success = false;
581 struct ldt_struct *ldt;
583 /* Bits [15:3] contain the index of the desired entry. */
584 sel >>= 3;
586 mutex_lock(&current->active_mm->context.lock);
587 ldt = current->active_mm->context.ldt;
588 if (ldt && sel < ldt->nr_entries) {
589 *out = ldt->entries[sel];
590 success = true;
593 mutex_unlock(&current->active_mm->context.lock);
595 return success;
597 #endif
598 native_store_gdt(&gdt_desc);
601 * Segment descriptors have a size of 8 bytes. Thus, the index is
602 * multiplied by 8 to obtain the memory offset of the desired descriptor
603 * from the base of the GDT. As bits [15:3] of the segment selector
604 * contain the index, it can be regarded as multiplied by 8 already.
605 * All that remains is to clear bits [2:0].
607 desc_base = sel & ~(SEGMENT_RPL_MASK | SEGMENT_TI_MASK);
609 if (desc_base > gdt_desc.size)
610 return false;
612 *out = *(struct desc_struct *)(gdt_desc.address + desc_base);
613 return true;
617 * insn_get_seg_base() - Obtain base address of segment descriptor.
618 * @regs: Register values as seen when entering kernel mode
619 * @seg_reg_idx: Index of the segment register pointing to seg descriptor
621 * Obtain the base address of the segment as indicated by the segment descriptor
622 * pointed by the segment selector. The segment selector is obtained from the
623 * input segment register index @seg_reg_idx.
625 * Returns:
627 * In protected mode, base address of the segment. Zero in long mode,
628 * except when FS or GS are used. In virtual-8086 mode, the segment
629 * selector shifted 4 bits to the right.
631 * -1L in case of error.
633 unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx)
635 struct desc_struct desc;
636 short sel;
638 sel = get_segment_selector(regs, seg_reg_idx);
639 if (sel < 0)
640 return -1L;
642 if (v8086_mode(regs))
644 * Base is simply the segment selector shifted 4
645 * bits to the right.
647 return (unsigned long)(sel << 4);
649 if (any_64bit_mode(regs)) {
651 * Only FS or GS will have a base address, the rest of
652 * the segments' bases are forced to 0.
654 unsigned long base;
656 if (seg_reg_idx == INAT_SEG_REG_FS) {
657 rdmsrl(MSR_FS_BASE, base);
658 } else if (seg_reg_idx == INAT_SEG_REG_GS) {
660 * swapgs was called at the kernel entry point. Thus,
661 * MSR_KERNEL_GS_BASE will have the user-space GS base.
663 if (user_mode(regs))
664 rdmsrl(MSR_KERNEL_GS_BASE, base);
665 else
666 rdmsrl(MSR_GS_BASE, base);
667 } else {
668 base = 0;
670 return base;
673 /* In protected mode the segment selector cannot be null. */
674 if (!sel)
675 return -1L;
677 if (!get_desc(&desc, sel))
678 return -1L;
680 return get_desc_base(&desc);
684 * get_seg_limit() - Obtain the limit of a segment descriptor
685 * @regs: Register values as seen when entering kernel mode
686 * @seg_reg_idx: Index of the segment register pointing to seg descriptor
688 * Obtain the limit of the segment as indicated by the segment descriptor
689 * pointed by the segment selector. The segment selector is obtained from the
690 * input segment register index @seg_reg_idx.
692 * Returns:
694 * In protected mode, the limit of the segment descriptor in bytes.
695 * In long mode and virtual-8086 mode, segment limits are not enforced. Thus,
696 * limit is returned as -1L to imply a limit-less segment.
698 * Zero is returned on error.
700 static unsigned long get_seg_limit(struct pt_regs *regs, int seg_reg_idx)
702 struct desc_struct desc;
703 unsigned long limit;
704 short sel;
706 sel = get_segment_selector(regs, seg_reg_idx);
707 if (sel < 0)
708 return 0;
710 if (any_64bit_mode(regs) || v8086_mode(regs))
711 return -1L;
713 if (!sel)
714 return 0;
716 if (!get_desc(&desc, sel))
717 return 0;
720 * If the granularity bit is set, the limit is given in multiples
721 * of 4096. This also means that the 12 least significant bits are
722 * not tested when checking the segment limits. In practice,
723 * this means that the segment ends in (limit << 12) + 0xfff.
725 limit = get_desc_limit(&desc);
726 if (desc.g)
727 limit = (limit << 12) + 0xfff;
729 return limit;
733 * insn_get_code_seg_params() - Obtain code segment parameters
734 * @regs: Structure with register values as seen when entering kernel mode
736 * Obtain address and operand sizes of the code segment. It is obtained from the
737 * selector contained in the CS register in regs. In protected mode, the default
738 * address is determined by inspecting the L and D bits of the segment
739 * descriptor. In virtual-8086 mode, the default is always two bytes for both
740 * address and operand sizes.
742 * Returns:
744 * An int containing ORed-in default parameters on success.
746 * -EINVAL on error.
748 int insn_get_code_seg_params(struct pt_regs *regs)
750 struct desc_struct desc;
751 short sel;
753 if (v8086_mode(regs))
754 /* Address and operand size are both 16-bit. */
755 return INSN_CODE_SEG_PARAMS(2, 2);
757 sel = get_segment_selector(regs, INAT_SEG_REG_CS);
758 if (sel < 0)
759 return sel;
761 if (!get_desc(&desc, sel))
762 return -EINVAL;
765 * The most significant byte of the Type field of the segment descriptor
766 * determines whether a segment contains data or code. If this is a data
767 * segment, return error.
769 if (!(desc.type & BIT(3)))
770 return -EINVAL;
772 switch ((desc.l << 1) | desc.d) {
773 case 0: /*
774 * Legacy mode. CS.L=0, CS.D=0. Address and operand size are
775 * both 16-bit.
777 return INSN_CODE_SEG_PARAMS(2, 2);
778 case 1: /*
779 * Legacy mode. CS.L=0, CS.D=1. Address and operand size are
780 * both 32-bit.
782 return INSN_CODE_SEG_PARAMS(4, 4);
783 case 2: /*
784 * IA-32e 64-bit mode. CS.L=1, CS.D=0. Address size is 64-bit;
785 * operand size is 32-bit.
787 return INSN_CODE_SEG_PARAMS(4, 8);
788 case 3: /* Invalid setting. CS.L=1, CS.D=1 */
789 /* fall through */
790 default:
791 return -EINVAL;
796 * insn_get_modrm_rm_off() - Obtain register in r/m part of the ModRM byte
797 * @insn: Instruction containing the ModRM byte
798 * @regs: Register values as seen when entering kernel mode
800 * Returns:
802 * The register indicated by the r/m part of the ModRM byte. The
803 * register is obtained as an offset from the base of pt_regs. In specific
804 * cases, the returned value can be -EDOM to indicate that the particular value
805 * of ModRM does not refer to a register and shall be ignored.
807 int insn_get_modrm_rm_off(struct insn *insn, struct pt_regs *regs)
809 return get_reg_offset(insn, regs, REG_TYPE_RM);
813 * get_seg_base_limit() - obtain base address and limit of a segment
814 * @insn: Instruction. Must be valid.
815 * @regs: Register values as seen when entering kernel mode
816 * @regoff: Operand offset, in pt_regs, used to resolve segment descriptor
817 * @base: Obtained segment base
818 * @limit: Obtained segment limit
820 * Obtain the base address and limit of the segment associated with the operand
821 * @regoff and, if any or allowed, override prefixes in @insn. This function is
822 * different from insn_get_seg_base() as the latter does not resolve the segment
823 * associated with the instruction operand. If a limit is not needed (e.g.,
824 * when running in long mode), @limit can be NULL.
826 * Returns:
828 * 0 on success. @base and @limit will contain the base address and of the
829 * resolved segment, respectively.
831 * -EINVAL on error.
833 static int get_seg_base_limit(struct insn *insn, struct pt_regs *regs,
834 int regoff, unsigned long *base,
835 unsigned long *limit)
837 int seg_reg_idx;
839 if (!base)
840 return -EINVAL;
842 seg_reg_idx = resolve_seg_reg(insn, regs, regoff);
843 if (seg_reg_idx < 0)
844 return seg_reg_idx;
846 *base = insn_get_seg_base(regs, seg_reg_idx);
847 if (*base == -1L)
848 return -EINVAL;
850 if (!limit)
851 return 0;
853 *limit = get_seg_limit(regs, seg_reg_idx);
854 if (!(*limit))
855 return -EINVAL;
857 return 0;
861 * get_eff_addr_reg() - Obtain effective address from register operand
862 * @insn: Instruction. Must be valid.
863 * @regs: Register values as seen when entering kernel mode
864 * @regoff: Obtained operand offset, in pt_regs, with the effective address
865 * @eff_addr: Obtained effective address
867 * Obtain the effective address stored in the register operand as indicated by
868 * the ModRM byte. This function is to be used only with register addressing
869 * (i.e., ModRM.mod is 3). The effective address is saved in @eff_addr. The
870 * register operand, as an offset from the base of pt_regs, is saved in @regoff;
871 * such offset can then be used to resolve the segment associated with the
872 * operand. This function can be used with any of the supported address sizes
873 * in x86.
875 * Returns:
877 * 0 on success. @eff_addr will have the effective address stored in the
878 * operand indicated by ModRM. @regoff will have such operand as an offset from
879 * the base of pt_regs.
881 * -EINVAL on error.
883 static int get_eff_addr_reg(struct insn *insn, struct pt_regs *regs,
884 int *regoff, long *eff_addr)
886 insn_get_modrm(insn);
888 if (!insn->modrm.nbytes)
889 return -EINVAL;
891 if (X86_MODRM_MOD(insn->modrm.value) != 3)
892 return -EINVAL;
894 *regoff = get_reg_offset(insn, regs, REG_TYPE_RM);
895 if (*regoff < 0)
896 return -EINVAL;
898 /* Ignore bytes that are outside the address size. */
899 if (insn->addr_bytes == 2)
900 *eff_addr = regs_get_register(regs, *regoff) & 0xffff;
901 else if (insn->addr_bytes == 4)
902 *eff_addr = regs_get_register(regs, *regoff) & 0xffffffff;
903 else /* 64-bit address */
904 *eff_addr = regs_get_register(regs, *regoff);
906 return 0;
910 * get_eff_addr_modrm() - Obtain referenced effective address via ModRM
911 * @insn: Instruction. Must be valid.
912 * @regs: Register values as seen when entering kernel mode
913 * @regoff: Obtained operand offset, in pt_regs, associated with segment
914 * @eff_addr: Obtained effective address
916 * Obtain the effective address referenced by the ModRM byte of @insn. After
917 * identifying the registers involved in the register-indirect memory reference,
918 * its value is obtained from the operands in @regs. The computed address is
919 * stored @eff_addr. Also, the register operand that indicates the associated
920 * segment is stored in @regoff, this parameter can later be used to determine
921 * such segment.
923 * Returns:
925 * 0 on success. @eff_addr will have the referenced effective address. @regoff
926 * will have a register, as an offset from the base of pt_regs, that can be used
927 * to resolve the associated segment.
929 * -EINVAL on error.
931 static int get_eff_addr_modrm(struct insn *insn, struct pt_regs *regs,
932 int *regoff, long *eff_addr)
934 long tmp;
936 if (insn->addr_bytes != 8 && insn->addr_bytes != 4)
937 return -EINVAL;
939 insn_get_modrm(insn);
941 if (!insn->modrm.nbytes)
942 return -EINVAL;
944 if (X86_MODRM_MOD(insn->modrm.value) > 2)
945 return -EINVAL;
947 *regoff = get_reg_offset(insn, regs, REG_TYPE_RM);
950 * -EDOM means that we must ignore the address_offset. In such a case,
951 * in 64-bit mode the effective address relative to the rIP of the
952 * following instruction.
954 if (*regoff == -EDOM) {
955 if (any_64bit_mode(regs))
956 tmp = regs->ip + insn->length;
957 else
958 tmp = 0;
959 } else if (*regoff < 0) {
960 return -EINVAL;
961 } else {
962 tmp = regs_get_register(regs, *regoff);
965 if (insn->addr_bytes == 4) {
966 int addr32 = (int)(tmp & 0xffffffff) + insn->displacement.value;
968 *eff_addr = addr32 & 0xffffffff;
969 } else {
970 *eff_addr = tmp + insn->displacement.value;
973 return 0;
977 * get_eff_addr_modrm_16() - Obtain referenced effective address via ModRM
978 * @insn: Instruction. Must be valid.
979 * @regs: Register values as seen when entering kernel mode
980 * @regoff: Obtained operand offset, in pt_regs, associated with segment
981 * @eff_addr: Obtained effective address
983 * Obtain the 16-bit effective address referenced by the ModRM byte of @insn.
984 * After identifying the registers involved in the register-indirect memory
985 * reference, its value is obtained from the operands in @regs. The computed
986 * address is stored @eff_addr. Also, the register operand that indicates
987 * the associated segment is stored in @regoff, this parameter can later be used
988 * to determine such segment.
990 * Returns:
992 * 0 on success. @eff_addr will have the referenced effective address. @regoff
993 * will have a register, as an offset from the base of pt_regs, that can be used
994 * to resolve the associated segment.
996 * -EINVAL on error.
998 static int get_eff_addr_modrm_16(struct insn *insn, struct pt_regs *regs,
999 int *regoff, short *eff_addr)
1001 int addr_offset1, addr_offset2, ret;
1002 short addr1 = 0, addr2 = 0, displacement;
1004 if (insn->addr_bytes != 2)
1005 return -EINVAL;
1007 insn_get_modrm(insn);
1009 if (!insn->modrm.nbytes)
1010 return -EINVAL;
1012 if (X86_MODRM_MOD(insn->modrm.value) > 2)
1013 return -EINVAL;
1015 ret = get_reg_offset_16(insn, regs, &addr_offset1, &addr_offset2);
1016 if (ret < 0)
1017 return -EINVAL;
1020 * Don't fail on invalid offset values. They might be invalid because
1021 * they cannot be used for this particular value of ModRM. Instead, use
1022 * them in the computation only if they contain a valid value.
1024 if (addr_offset1 != -EDOM)
1025 addr1 = regs_get_register(regs, addr_offset1) & 0xffff;
1027 if (addr_offset2 != -EDOM)
1028 addr2 = regs_get_register(regs, addr_offset2) & 0xffff;
1030 displacement = insn->displacement.value & 0xffff;
1031 *eff_addr = addr1 + addr2 + displacement;
1034 * The first operand register could indicate to use of either SS or DS
1035 * registers to obtain the segment selector. The second operand
1036 * register can only indicate the use of DS. Thus, the first operand
1037 * will be used to obtain the segment selector.
1039 *regoff = addr_offset1;
1041 return 0;
1045 * get_eff_addr_sib() - Obtain referenced effective address via SIB
1046 * @insn: Instruction. Must be valid.
1047 * @regs: Register values as seen when entering kernel mode
1048 * @regoff: Obtained operand offset, in pt_regs, associated with segment
1049 * @eff_addr: Obtained effective address
1051 * Obtain the effective address referenced by the SIB byte of @insn. After
1052 * identifying the registers involved in the indexed, register-indirect memory
1053 * reference, its value is obtained from the operands in @regs. The computed
1054 * address is stored @eff_addr. Also, the register operand that indicates the
1055 * associated segment is stored in @regoff, this parameter can later be used to
1056 * determine such segment.
1058 * Returns:
1060 * 0 on success. @eff_addr will have the referenced effective address.
1061 * @base_offset will have a register, as an offset from the base of pt_regs,
1062 * that can be used to resolve the associated segment.
1064 * -EINVAL on error.
1066 static int get_eff_addr_sib(struct insn *insn, struct pt_regs *regs,
1067 int *base_offset, long *eff_addr)
1069 long base, indx;
1070 int indx_offset;
1072 if (insn->addr_bytes != 8 && insn->addr_bytes != 4)
1073 return -EINVAL;
1075 insn_get_modrm(insn);
1077 if (!insn->modrm.nbytes)
1078 return -EINVAL;
1080 if (X86_MODRM_MOD(insn->modrm.value) > 2)
1081 return -EINVAL;
1083 insn_get_sib(insn);
1085 if (!insn->sib.nbytes)
1086 return -EINVAL;
1088 *base_offset = get_reg_offset(insn, regs, REG_TYPE_BASE);
1089 indx_offset = get_reg_offset(insn, regs, REG_TYPE_INDEX);
1092 * Negative values in the base and index offset means an error when
1093 * decoding the SIB byte. Except -EDOM, which means that the registers
1094 * should not be used in the address computation.
1096 if (*base_offset == -EDOM)
1097 base = 0;
1098 else if (*base_offset < 0)
1099 return -EINVAL;
1100 else
1101 base = regs_get_register(regs, *base_offset);
1103 if (indx_offset == -EDOM)
1104 indx = 0;
1105 else if (indx_offset < 0)
1106 return -EINVAL;
1107 else
1108 indx = regs_get_register(regs, indx_offset);
1110 if (insn->addr_bytes == 4) {
1111 int addr32, base32, idx32;
1113 base32 = base & 0xffffffff;
1114 idx32 = indx & 0xffffffff;
1116 addr32 = base32 + idx32 * (1 << X86_SIB_SCALE(insn->sib.value));
1117 addr32 += insn->displacement.value;
1119 *eff_addr = addr32 & 0xffffffff;
1120 } else {
1121 *eff_addr = base + indx * (1 << X86_SIB_SCALE(insn->sib.value));
1122 *eff_addr += insn->displacement.value;
1125 return 0;
1129 * get_addr_ref_16() - Obtain the 16-bit address referred by instruction
1130 * @insn: Instruction containing ModRM byte and displacement
1131 * @regs: Register values as seen when entering kernel mode
1133 * This function is to be used with 16-bit address encodings. Obtain the memory
1134 * address referred by the instruction's ModRM and displacement bytes. Also, the
1135 * segment used as base is determined by either any segment override prefixes in
1136 * @insn or the default segment of the registers involved in the address
1137 * computation. In protected mode, segment limits are enforced.
1139 * Returns:
1141 * Linear address referenced by the instruction operands on success.
1143 * -1L on error.
1145 static void __user *get_addr_ref_16(struct insn *insn, struct pt_regs *regs)
1147 unsigned long linear_addr = -1L, seg_base, seg_limit;
1148 int ret, regoff;
1149 short eff_addr;
1150 long tmp;
1152 insn_get_modrm(insn);
1153 insn_get_displacement(insn);
1155 if (insn->addr_bytes != 2)
1156 goto out;
1158 if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1159 ret = get_eff_addr_reg(insn, regs, &regoff, &tmp);
1160 if (ret)
1161 goto out;
1163 eff_addr = tmp;
1164 } else {
1165 ret = get_eff_addr_modrm_16(insn, regs, &regoff, &eff_addr);
1166 if (ret)
1167 goto out;
1170 ret = get_seg_base_limit(insn, regs, regoff, &seg_base, &seg_limit);
1171 if (ret)
1172 goto out;
1175 * Before computing the linear address, make sure the effective address
1176 * is within the limits of the segment. In virtual-8086 mode, segment
1177 * limits are not enforced. In such a case, the segment limit is -1L to
1178 * reflect this fact.
1180 if ((unsigned long)(eff_addr & 0xffff) > seg_limit)
1181 goto out;
1183 linear_addr = (unsigned long)(eff_addr & 0xffff) + seg_base;
1185 /* Limit linear address to 20 bits */
1186 if (v8086_mode(regs))
1187 linear_addr &= 0xfffff;
1189 out:
1190 return (void __user *)linear_addr;
1194 * get_addr_ref_32() - Obtain a 32-bit linear address
1195 * @insn: Instruction with ModRM, SIB bytes and displacement
1196 * @regs: Register values as seen when entering kernel mode
1198 * This function is to be used with 32-bit address encodings to obtain the
1199 * linear memory address referred by the instruction's ModRM, SIB,
1200 * displacement bytes and segment base address, as applicable. If in protected
1201 * mode, segment limits are enforced.
1203 * Returns:
1205 * Linear address referenced by instruction and registers on success.
1207 * -1L on error.
1209 static void __user *get_addr_ref_32(struct insn *insn, struct pt_regs *regs)
1211 unsigned long linear_addr = -1L, seg_base, seg_limit;
1212 int eff_addr, regoff;
1213 long tmp;
1214 int ret;
1216 if (insn->addr_bytes != 4)
1217 goto out;
1219 if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1220 ret = get_eff_addr_reg(insn, regs, &regoff, &tmp);
1221 if (ret)
1222 goto out;
1224 eff_addr = tmp;
1226 } else {
1227 if (insn->sib.nbytes) {
1228 ret = get_eff_addr_sib(insn, regs, &regoff, &tmp);
1229 if (ret)
1230 goto out;
1232 eff_addr = tmp;
1233 } else {
1234 ret = get_eff_addr_modrm(insn, regs, &regoff, &tmp);
1235 if (ret)
1236 goto out;
1238 eff_addr = tmp;
1242 ret = get_seg_base_limit(insn, regs, regoff, &seg_base, &seg_limit);
1243 if (ret)
1244 goto out;
1247 * In protected mode, before computing the linear address, make sure
1248 * the effective address is within the limits of the segment.
1249 * 32-bit addresses can be used in long and virtual-8086 modes if an
1250 * address override prefix is used. In such cases, segment limits are
1251 * not enforced. When in virtual-8086 mode, the segment limit is -1L
1252 * to reflect this situation.
1254 * After computed, the effective address is treated as an unsigned
1255 * quantity.
1257 if (!any_64bit_mode(regs) && ((unsigned int)eff_addr > seg_limit))
1258 goto out;
1261 * Even though 32-bit address encodings are allowed in virtual-8086
1262 * mode, the address range is still limited to [0x-0xffff].
1264 if (v8086_mode(regs) && (eff_addr & ~0xffff))
1265 goto out;
1268 * Data type long could be 64 bits in size. Ensure that our 32-bit
1269 * effective address is not sign-extended when computing the linear
1270 * address.
1272 linear_addr = (unsigned long)(eff_addr & 0xffffffff) + seg_base;
1274 /* Limit linear address to 20 bits */
1275 if (v8086_mode(regs))
1276 linear_addr &= 0xfffff;
1278 out:
1279 return (void __user *)linear_addr;
1283 * get_addr_ref_64() - Obtain a 64-bit linear address
1284 * @insn: Instruction struct with ModRM and SIB bytes and displacement
1285 * @regs: Structure with register values as seen when entering kernel mode
1287 * This function is to be used with 64-bit address encodings to obtain the
1288 * linear memory address referred by the instruction's ModRM, SIB,
1289 * displacement bytes and segment base address, as applicable.
1291 * Returns:
1293 * Linear address referenced by instruction and registers on success.
1295 * -1L on error.
1297 #ifndef CONFIG_X86_64
1298 static void __user *get_addr_ref_64(struct insn *insn, struct pt_regs *regs)
1300 return (void __user *)-1L;
1302 #else
1303 static void __user *get_addr_ref_64(struct insn *insn, struct pt_regs *regs)
1305 unsigned long linear_addr = -1L, seg_base;
1306 int regoff, ret;
1307 long eff_addr;
1309 if (insn->addr_bytes != 8)
1310 goto out;
1312 if (X86_MODRM_MOD(insn->modrm.value) == 3) {
1313 ret = get_eff_addr_reg(insn, regs, &regoff, &eff_addr);
1314 if (ret)
1315 goto out;
1317 } else {
1318 if (insn->sib.nbytes) {
1319 ret = get_eff_addr_sib(insn, regs, &regoff, &eff_addr);
1320 if (ret)
1321 goto out;
1322 } else {
1323 ret = get_eff_addr_modrm(insn, regs, &regoff, &eff_addr);
1324 if (ret)
1325 goto out;
1330 ret = get_seg_base_limit(insn, regs, regoff, &seg_base, NULL);
1331 if (ret)
1332 goto out;
1334 linear_addr = (unsigned long)eff_addr + seg_base;
1336 out:
1337 return (void __user *)linear_addr;
1339 #endif /* CONFIG_X86_64 */
1342 * insn_get_addr_ref() - Obtain the linear address referred by instruction
1343 * @insn: Instruction structure containing ModRM byte and displacement
1344 * @regs: Structure with register values as seen when entering kernel mode
1346 * Obtain the linear address referred by the instruction's ModRM, SIB and
1347 * displacement bytes, and segment base, as applicable. In protected mode,
1348 * segment limits are enforced.
1350 * Returns:
1352 * Linear address referenced by instruction and registers on success.
1354 * -1L on error.
1356 void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs)
1358 if (!insn || !regs)
1359 return (void __user *)-1L;
1361 switch (insn->addr_bytes) {
1362 case 2:
1363 return get_addr_ref_16(insn, regs);
1364 case 4:
1365 return get_addr_ref_32(insn, regs);
1366 case 8:
1367 return get_addr_ref_64(insn, regs);
1368 default:
1369 return (void __user *)-1L;