1/*
2 * vMTRR implementation
3 *
4 * Copyright (C) 2006 Qumranet, Inc.
5 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
6 * Copyright(C) 2015 Intel Corporation.
7 *
8 * Authors:
9 *   Yaniv Kamay  <yaniv@qumranet.com>
10 *   Avi Kivity   <avi@qumranet.com>
11 *   Marcelo Tosatti <mtosatti@redhat.com>
12 *   Paolo Bonzini <pbonzini@redhat.com>
13 *   Xiao Guangrong <guangrong.xiao@linux.intel.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2.  See
16 * the COPYING file in the top-level directory.
17 */
18
19#include <linux/kvm_host.h>
20#include <asm/mtrr.h>
21
22#include "cpuid.h"
23#include "mmu.h"
24
25#define IA32_MTRR_DEF_TYPE_E		(1ULL << 11)
26#define IA32_MTRR_DEF_TYPE_FE		(1ULL << 10)
27#define IA32_MTRR_DEF_TYPE_TYPE_MASK	(0xff)
28
29static bool msr_mtrr_valid(unsigned msr)
30{
31	switch (msr) {
32	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
33	case MSR_MTRRfix64K_00000:
34	case MSR_MTRRfix16K_80000:
35	case MSR_MTRRfix16K_A0000:
36	case MSR_MTRRfix4K_C0000:
37	case MSR_MTRRfix4K_C8000:
38	case MSR_MTRRfix4K_D0000:
39	case MSR_MTRRfix4K_D8000:
40	case MSR_MTRRfix4K_E0000:
41	case MSR_MTRRfix4K_E8000:
42	case MSR_MTRRfix4K_F0000:
43	case MSR_MTRRfix4K_F8000:
44	case MSR_MTRRdefType:
45	case MSR_IA32_CR_PAT:
46		return true;
47	}
48	return false;
49}
50
51static bool valid_pat_type(unsigned t)
52{
53	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
54}
55
56static bool valid_mtrr_type(unsigned t)
57{
58	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
59}
60
61bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
62{
63	int i;
64	u64 mask;
65
66	if (!msr_mtrr_valid(msr))
67		return false;
68
69	if (msr == MSR_IA32_CR_PAT) {
70		for (i = 0; i < 8; i++)
71			if (!valid_pat_type((data >> (i * 8)) & 0xff))
72				return false;
73		return true;
74	} else if (msr == MSR_MTRRdefType) {
75		if (data & ~0xcff)
76			return false;
77		return valid_mtrr_type(data & 0xff);
78	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
79		for (i = 0; i < 8 ; i++)
80			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
81				return false;
82		return true;
83	}
84
85	/* variable MTRRs */
86	WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
87
88	mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
89	if ((msr & 1) == 0) {
90		/* MTRR base */
91		if (!valid_mtrr_type(data & 0xff))
92			return false;
93		mask |= 0xf00;
94	} else
95		/* MTRR mask */
96		mask |= 0x7ff;
97	if (data & mask) {
98		kvm_inject_gp(vcpu, 0);
99		return false;
100	}
101
102	return true;
103}
104EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
105
106static bool mtrr_is_enabled(struct kvm_mtrr *mtrr_state)
107{
108	return !!(mtrr_state->deftype & IA32_MTRR_DEF_TYPE_E);
109}
110
111static bool fixed_mtrr_is_enabled(struct kvm_mtrr *mtrr_state)
112{
113	return !!(mtrr_state->deftype & IA32_MTRR_DEF_TYPE_FE);
114}
115
116static u8 mtrr_default_type(struct kvm_mtrr *mtrr_state)
117{
118	return mtrr_state->deftype & IA32_MTRR_DEF_TYPE_TYPE_MASK;
119}
120
121static u8 mtrr_disabled_type(struct kvm_vcpu *vcpu)
122{
123	/*
124	 * Intel SDM 11.11.2.2: all MTRRs are disabled when
125	 * IA32_MTRR_DEF_TYPE.E bit is cleared, and the UC
126	 * memory type is applied to all of physical memory.
127	 *
128	 * However, virtual machines can be run with CPUID such that
129	 * there are no MTRRs.  In that case, the firmware will never
130	 * enable MTRRs and it is obviously undesirable to run the
131	 * guest entirely with UC memory and we use WB.
132	 */
133	if (guest_cpuid_has_mtrr(vcpu))
134		return MTRR_TYPE_UNCACHABLE;
135	else
136		return MTRR_TYPE_WRBACK;
137}
138
139/*
140* Three terms are used in the following code:
141* - segment, it indicates the address segments covered by fixed MTRRs.
142* - unit, it corresponds to the MSR entry in the segment.
143* - range, a range is covered in one memory cache type.
144*/
145struct fixed_mtrr_segment {
146	u64 start;
147	u64 end;
148
149	int range_shift;
150
151	/* the start position in kvm_mtrr.fixed_ranges[]. */
152	int range_start;
153};
154
155static struct fixed_mtrr_segment fixed_seg_table[] = {
156	/* MSR_MTRRfix64K_00000, 1 unit. 64K fixed mtrr. */
157	{
158		.start = 0x0,
159		.end = 0x80000,
160		.range_shift = 16, /* 64K */
161		.range_start = 0,
162	},
163
164	/*
165	 * MSR_MTRRfix16K_80000 ... MSR_MTRRfix16K_A0000, 2 units,
166	 * 16K fixed mtrr.
167	 */
168	{
169		.start = 0x80000,
170		.end = 0xc0000,
171		.range_shift = 14, /* 16K */
172		.range_start = 8,
173	},
174
175	/*
176	 * MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000, 8 units,
177	 * 4K fixed mtrr.
178	 */
179	{
180		.start = 0xc0000,
181		.end = 0x100000,
182		.range_shift = 12, /* 12K */
183		.range_start = 24,
184	}
185};
186
187/*
188 * The size of unit is covered in one MSR, one MSR entry contains
189 * 8 ranges so that unit size is always 8 * 2^range_shift.
190 */
191static u64 fixed_mtrr_seg_unit_size(int seg)
192{
193	return 8 << fixed_seg_table[seg].range_shift;
194}
195
196static bool fixed_msr_to_seg_unit(u32 msr, int *seg, int *unit)
197{
198	switch (msr) {
199	case MSR_MTRRfix64K_00000:
200		*seg = 0;
201		*unit = 0;
202		break;
203	case MSR_MTRRfix16K_80000 ... MSR_MTRRfix16K_A0000:
204		*seg = 1;
205		*unit = msr - MSR_MTRRfix16K_80000;
206		break;
207	case MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000:
208		*seg = 2;
209		*unit = msr - MSR_MTRRfix4K_C0000;
210		break;
211	default:
212		return false;
213	}
214
215	return true;
216}
217
218static void fixed_mtrr_seg_unit_range(int seg, int unit, u64 *start, u64 *end)
219{
220	struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
221	u64 unit_size = fixed_mtrr_seg_unit_size(seg);
222
223	*start = mtrr_seg->start + unit * unit_size;
224	*end = *start + unit_size;
225	WARN_ON(*end > mtrr_seg->end);
226}
227
228static int fixed_mtrr_seg_unit_range_index(int seg, int unit)
229{
230	struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
231
232	WARN_ON(mtrr_seg->start + unit * fixed_mtrr_seg_unit_size(seg)
233		> mtrr_seg->end);
234
235	/* each unit has 8 ranges. */
236	return mtrr_seg->range_start + 8 * unit;
237}
238
239static int fixed_mtrr_seg_end_range_index(int seg)
240{
241	struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
242	int n;
243
244	n = (mtrr_seg->end - mtrr_seg->start) >> mtrr_seg->range_shift;
245	return mtrr_seg->range_start + n - 1;
246}
247
248static bool fixed_msr_to_range(u32 msr, u64 *start, u64 *end)
249{
250	int seg, unit;
251
252	if (!fixed_msr_to_seg_unit(msr, &seg, &unit))
253		return false;
254
255	fixed_mtrr_seg_unit_range(seg, unit, start, end);
256	return true;
257}
258
259static int fixed_msr_to_range_index(u32 msr)
260{
261	int seg, unit;
262
263	if (!fixed_msr_to_seg_unit(msr, &seg, &unit))
264		return -1;
265
266	return fixed_mtrr_seg_unit_range_index(seg, unit);
267}
268
269static int fixed_mtrr_addr_to_seg(u64 addr)
270{
271	struct fixed_mtrr_segment *mtrr_seg;
272	int seg, seg_num = ARRAY_SIZE(fixed_seg_table);
273
274	for (seg = 0; seg < seg_num; seg++) {
275		mtrr_seg = &fixed_seg_table[seg];
276		if (mtrr_seg->start <= addr && addr < mtrr_seg->end)
277			return seg;
278	}
279
280	return -1;
281}
282
283static int fixed_mtrr_addr_seg_to_range_index(u64 addr, int seg)
284{
285	struct fixed_mtrr_segment *mtrr_seg;
286	int index;
287
288	mtrr_seg = &fixed_seg_table[seg];
289	index = mtrr_seg->range_start;
290	index += (addr - mtrr_seg->start) >> mtrr_seg->range_shift;
291	return index;
292}
293
294static u64 fixed_mtrr_range_end_addr(int seg, int index)
295{
296	struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
297	int pos = index - mtrr_seg->range_start;
298
299	return mtrr_seg->start + ((pos + 1) << mtrr_seg->range_shift);
300}
301
302static void var_mtrr_range(struct kvm_mtrr_range *range, u64 *start, u64 *end)
303{
304	u64 mask;
305
306	*start = range->base & PAGE_MASK;
307
308	mask = range->mask & PAGE_MASK;
309
310	/* This cannot overflow because writing to the reserved bits of
311	 * variable MTRRs causes a #GP.
312	 */
313	*end = (*start | ~mask) + 1;
314}
315
316static void update_mtrr(struct kvm_vcpu *vcpu, u32 msr)
317{
318	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
319	gfn_t start, end;
320	int index;
321
322	if (msr == MSR_IA32_CR_PAT || !tdp_enabled ||
323	      !kvm_arch_has_noncoherent_dma(vcpu->kvm))
324		return;
325
326	if (!mtrr_is_enabled(mtrr_state) && msr != MSR_MTRRdefType)
327		return;
328
329	/* fixed MTRRs. */
330	if (fixed_msr_to_range(msr, &start, &end)) {
331		if (!fixed_mtrr_is_enabled(mtrr_state))
332			return;
333	} else if (msr == MSR_MTRRdefType) {
334		start = 0x0;
335		end = ~0ULL;
336	} else {
337		/* variable range MTRRs. */
338		index = (msr - 0x200) / 2;
339		var_mtrr_range(&mtrr_state->var_ranges[index], &start, &end);
340	}
341
342	kvm_zap_gfn_range(vcpu->kvm, gpa_to_gfn(start), gpa_to_gfn(end));
343}
344
345static bool var_mtrr_range_is_valid(struct kvm_mtrr_range *range)
346{
347	return (range->mask & (1 << 11)) != 0;
348}
349
350static void set_var_mtrr_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
351{
352	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
353	struct kvm_mtrr_range *tmp, *cur;
354	int index, is_mtrr_mask;
355
356	index = (msr - 0x200) / 2;
357	is_mtrr_mask = msr - 0x200 - 2 * index;
358	cur = &mtrr_state->var_ranges[index];
359
360	/* remove the entry if it's in the list. */
361	if (var_mtrr_range_is_valid(cur))
362		list_del(&mtrr_state->var_ranges[index].node);
363
364	/* Extend the mask with all 1 bits to the left, since those
365	 * bits must implicitly be 0.  The bits are then cleared
366	 * when reading them.
367	 */
368	if (!is_mtrr_mask)
369		cur->base = data;
370	else
371		cur->mask = data | (-1LL << cpuid_maxphyaddr(vcpu));
372
373	/* add it to the list if it's enabled. */
374	if (var_mtrr_range_is_valid(cur)) {
375		list_for_each_entry(tmp, &mtrr_state->head, node)
376			if (cur->base >= tmp->base)
377				break;
378		list_add_tail(&cur->node, &tmp->node);
379	}
380}
381
382int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
383{
384	int index;
385
386	if (!kvm_mtrr_valid(vcpu, msr, data))
387		return 1;
388
389	index = fixed_msr_to_range_index(msr);
390	if (index >= 0)
391		*(u64 *)&vcpu->arch.mtrr_state.fixed_ranges[index] = data;
392	else if (msr == MSR_MTRRdefType)
393		vcpu->arch.mtrr_state.deftype = data;
394	else if (msr == MSR_IA32_CR_PAT)
395		vcpu->arch.pat = data;
396	else
397		set_var_mtrr_msr(vcpu, msr, data);
398
399	update_mtrr(vcpu, msr);
400	return 0;
401}
402
403int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
404{
405	int index;
406
407	/* MSR_MTRRcap is a readonly MSR. */
408	if (msr == MSR_MTRRcap) {
409		/*
410		 * SMRR = 0
411		 * WC = 1
412		 * FIX = 1
413		 * VCNT = KVM_NR_VAR_MTRR
414		 */
415		*pdata = 0x500 | KVM_NR_VAR_MTRR;
416		return 0;
417	}
418
419	if (!msr_mtrr_valid(msr))
420		return 1;
421
422	index = fixed_msr_to_range_index(msr);
423	if (index >= 0)
424		*pdata = *(u64 *)&vcpu->arch.mtrr_state.fixed_ranges[index];
425	else if (msr == MSR_MTRRdefType)
426		*pdata = vcpu->arch.mtrr_state.deftype;
427	else if (msr == MSR_IA32_CR_PAT)
428		*pdata = vcpu->arch.pat;
429	else {	/* Variable MTRRs */
430		int is_mtrr_mask;
431
432		index = (msr - 0x200) / 2;
433		is_mtrr_mask = msr - 0x200 - 2 * index;
434		if (!is_mtrr_mask)
435			*pdata = vcpu->arch.mtrr_state.var_ranges[index].base;
436		else
437			*pdata = vcpu->arch.mtrr_state.var_ranges[index].mask;
438
439		*pdata &= (1ULL << cpuid_maxphyaddr(vcpu)) - 1;
440	}
441
442	return 0;
443}
444
445void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu)
446{
447	INIT_LIST_HEAD(&vcpu->arch.mtrr_state.head);
448}
449
450struct mtrr_iter {
451	/* input fields. */
452	struct kvm_mtrr *mtrr_state;
453	u64 start;
454	u64 end;
455
456	/* output fields. */
457	int mem_type;
458	/* mtrr is completely disabled? */
459	bool mtrr_disabled;
460	/* [start, end) is not fully covered in MTRRs? */
461	bool partial_map;
462
463	/* private fields. */
464	union {
465		/* used for fixed MTRRs. */
466		struct {
467			int index;
468			int seg;
469		};
470
471		/* used for var MTRRs. */
472		struct {
473			struct kvm_mtrr_range *range;
474			/* max address has been covered in var MTRRs. */
475			u64 start_max;
476		};
477	};
478
479	bool fixed;
480};
481
482static bool mtrr_lookup_fixed_start(struct mtrr_iter *iter)
483{
484	int seg, index;
485
486	if (!fixed_mtrr_is_enabled(iter->mtrr_state))
487		return false;
488
489	seg = fixed_mtrr_addr_to_seg(iter->start);
490	if (seg < 0)
491		return false;
492
493	iter->fixed = true;
494	index = fixed_mtrr_addr_seg_to_range_index(iter->start, seg);
495	iter->index = index;
496	iter->seg = seg;
497	return true;
498}
499
500static bool match_var_range(struct mtrr_iter *iter,
501			    struct kvm_mtrr_range *range)
502{
503	u64 start, end;
504
505	var_mtrr_range(range, &start, &end);
506	if (!(start >= iter->end || end <= iter->start)) {
507		iter->range = range;
508
509		/*
510		 * the function is called when we do kvm_mtrr.head walking.
511		 * Range has the minimum base address which interleaves
512		 * [looker->start_max, looker->end).
513		 */
514		iter->partial_map |= iter->start_max < start;
515
516		/* update the max address has been covered. */
517		iter->start_max = max(iter->start_max, end);
518		return true;
519	}
520
521	return false;
522}
523
524static void __mtrr_lookup_var_next(struct mtrr_iter *iter)
525{
526	struct kvm_mtrr *mtrr_state = iter->mtrr_state;
527
528	list_for_each_entry_continue(iter->range, &mtrr_state->head, node)
529		if (match_var_range(iter, iter->range))
530			return;
531
532	iter->range = NULL;
533	iter->partial_map |= iter->start_max < iter->end;
534}
535
536static void mtrr_lookup_var_start(struct mtrr_iter *iter)
537{
538	struct kvm_mtrr *mtrr_state = iter->mtrr_state;
539
540	iter->fixed = false;
541	iter->start_max = iter->start;
542	iter->range = list_prepare_entry(iter->range, &mtrr_state->head, node);
543
544	__mtrr_lookup_var_next(iter);
545}
546
547static void mtrr_lookup_fixed_next(struct mtrr_iter *iter)
548{
549	/* terminate the lookup. */
550	if (fixed_mtrr_range_end_addr(iter->seg, iter->index) >= iter->end) {
551		iter->fixed = false;
552		iter->range = NULL;
553		return;
554	}
555
556	iter->index++;
557
558	/* have looked up for all fixed MTRRs. */
559	if (iter->index >= ARRAY_SIZE(iter->mtrr_state->fixed_ranges))
560		return mtrr_lookup_var_start(iter);
561
562	/* switch to next segment. */
563	if (iter->index > fixed_mtrr_seg_end_range_index(iter->seg))
564		iter->seg++;
565}
566
567static void mtrr_lookup_var_next(struct mtrr_iter *iter)
568{
569	__mtrr_lookup_var_next(iter);
570}
571
572static void mtrr_lookup_start(struct mtrr_iter *iter)
573{
574	if (!mtrr_is_enabled(iter->mtrr_state)) {
575		iter->mtrr_disabled = true;
576		return;
577	}
578
579	if (!mtrr_lookup_fixed_start(iter))
580		mtrr_lookup_var_start(iter);
581}
582
583static void mtrr_lookup_init(struct mtrr_iter *iter,
584			     struct kvm_mtrr *mtrr_state, u64 start, u64 end)
585{
586	iter->mtrr_state = mtrr_state;
587	iter->start = start;
588	iter->end = end;
589	iter->mtrr_disabled = false;
590	iter->partial_map = false;
591	iter->fixed = false;
592	iter->range = NULL;
593
594	mtrr_lookup_start(iter);
595}
596
597static bool mtrr_lookup_okay(struct mtrr_iter *iter)
598{
599	if (iter->fixed) {
600		iter->mem_type = iter->mtrr_state->fixed_ranges[iter->index];
601		return true;
602	}
603
604	if (iter->range) {
605		iter->mem_type = iter->range->base & 0xff;
606		return true;
607	}
608
609	return false;
610}
611
612static void mtrr_lookup_next(struct mtrr_iter *iter)
613{
614	if (iter->fixed)
615		mtrr_lookup_fixed_next(iter);
616	else
617		mtrr_lookup_var_next(iter);
618}
619
620#define mtrr_for_each_mem_type(_iter_, _mtrr_, _gpa_start_, _gpa_end_) \
621	for (mtrr_lookup_init(_iter_, _mtrr_, _gpa_start_, _gpa_end_); \
622	     mtrr_lookup_okay(_iter_); mtrr_lookup_next(_iter_))
623
624u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
625{
626	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
627	struct mtrr_iter iter;
628	u64 start, end;
629	int type = -1;
630	const int wt_wb_mask = (1 << MTRR_TYPE_WRBACK)
631			       | (1 << MTRR_TYPE_WRTHROUGH);
632
633	start = gfn_to_gpa(gfn);
634	end = start + PAGE_SIZE;
635
636	mtrr_for_each_mem_type(&iter, mtrr_state, start, end) {
637		int curr_type = iter.mem_type;
638
639		/*
640		 * Please refer to Intel SDM Volume 3: 11.11.4.1 MTRR
641		 * Precedences.
642		 */
643
644		if (type == -1) {
645			type = curr_type;
646			continue;
647		}
648
649		/*
650		 * If two or more variable memory ranges match and the
651		 * memory types are identical, then that memory type is
652		 * used.
653		 */
654		if (type == curr_type)
655			continue;
656
657		/*
658		 * If two or more variable memory ranges match and one of
659		 * the memory types is UC, the UC memory type used.
660		 */
661		if (curr_type == MTRR_TYPE_UNCACHABLE)
662			return MTRR_TYPE_UNCACHABLE;
663
664		/*
665		 * If two or more variable memory ranges match and the
666		 * memory types are WT and WB, the WT memory type is used.
667		 */
668		if (((1 << type) & wt_wb_mask) &&
669		      ((1 << curr_type) & wt_wb_mask)) {
670			type = MTRR_TYPE_WRTHROUGH;
671			continue;
672		}
673
674		/*
675		 * For overlaps not defined by the above rules, processor
676		 * behavior is undefined.
677		 */
678
679		/* We use WB for this undefined behavior. :( */
680		return MTRR_TYPE_WRBACK;
681	}
682
683	if (iter.mtrr_disabled)
684		return mtrr_disabled_type(vcpu);
685
686	/* not contained in any MTRRs. */
687	if (type == -1)
688		return mtrr_default_type(mtrr_state);
689
690	/*
691	 * We just check one page, partially covered by MTRRs is
692	 * impossible.
693	 */
694	WARN_ON(iter.partial_map);
695
696	return type;
697}
698EXPORT_SYMBOL_GPL(kvm_mtrr_get_guest_memory_type);
699
700bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
701					  int page_num)
702{
703	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
704	struct mtrr_iter iter;
705	u64 start, end;
706	int type = -1;
707
708	start = gfn_to_gpa(gfn);
709	end = gfn_to_gpa(gfn + page_num);
710	mtrr_for_each_mem_type(&iter, mtrr_state, start, end) {
711		if (type == -1) {
712			type = iter.mem_type;
713			continue;
714		}
715
716		if (type != iter.mem_type)
717			return false;
718	}
719
720	if (iter.mtrr_disabled)
721		return true;
722
723	if (!iter.partial_map)
724		return true;
725
726	if (type == -1)
727		return true;
728
729	return type == mtrr_default_type(mtrr_state);
730}
731