xref: /linux/arch/x86/kvm/mmu/spte.c (revision 63eb28bb1402891b1ad2be02a530f29a9dd7f1cd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * Macros and functions to access KVM PTEs (also known as SPTEs)
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2020 Red Hat, Inc. and/or its affiliates.
9  */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/kvm_host.h>
13 #include "mmu.h"
14 #include "mmu_internal.h"
15 #include "x86.h"
16 #include "spte.h"
17 
18 #include <asm/e820/api.h>
19 #include <asm/memtype.h>
20 #include <asm/vmx.h>
21 
22 bool __read_mostly enable_mmio_caching = true;
23 static bool __ro_after_init allow_mmio_caching;
24 module_param_named(mmio_caching, enable_mmio_caching, bool, 0444);
25 EXPORT_SYMBOL_GPL(enable_mmio_caching);
26 
27 bool __read_mostly kvm_ad_enabled;
28 
29 u64 __read_mostly shadow_host_writable_mask;
30 u64 __read_mostly shadow_mmu_writable_mask;
31 u64 __read_mostly shadow_nx_mask;
32 u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
33 u64 __read_mostly shadow_user_mask;
34 u64 __read_mostly shadow_accessed_mask;
35 u64 __read_mostly shadow_dirty_mask;
36 u64 __read_mostly shadow_mmio_value;
37 u64 __read_mostly shadow_mmio_mask;
38 u64 __read_mostly shadow_mmio_access_mask;
39 u64 __read_mostly shadow_present_mask;
40 u64 __read_mostly shadow_me_value;
41 u64 __read_mostly shadow_me_mask;
42 u64 __read_mostly shadow_acc_track_mask;
43 
44 u64 __read_mostly shadow_nonpresent_or_rsvd_mask;
45 u64 __read_mostly shadow_nonpresent_or_rsvd_lower_gfn_mask;
46 
kvm_get_host_maxphyaddr(void)47 static u8 __init kvm_get_host_maxphyaddr(void)
48 {
49 	/*
50 	 * boot_cpu_data.x86_phys_bits is reduced when MKTME or SME are detected
51 	 * in CPU detection code, but the processor treats those reduced bits as
52 	 * 'keyID' thus they are not reserved bits. Therefore KVM needs to look at
53 	 * the physical address bits reported by CPUID, i.e. the raw MAXPHYADDR,
54 	 * when reasoning about CPU behavior with respect to MAXPHYADDR.
55 	 */
56 	if (likely(boot_cpu_data.extended_cpuid_level >= 0x80000008))
57 		return cpuid_eax(0x80000008) & 0xff;
58 
59 	/*
60 	 * Quite weird to have VMX or SVM but not MAXPHYADDR; probably a VM with
61 	 * custom CPUID.  Proceed with whatever the kernel found since these features
62 	 * aren't virtualizable (SME/SEV also require CPUIDs higher than 0x80000008).
63 	 */
64 	return boot_cpu_data.x86_phys_bits;
65 }
66 
kvm_mmu_spte_module_init(void)67 void __init kvm_mmu_spte_module_init(void)
68 {
69 	/*
70 	 * Snapshot userspace's desire to allow MMIO caching.  Whether or not
71 	 * KVM can actually enable MMIO caching depends on vendor-specific
72 	 * hardware capabilities and other module params that can't be resolved
73 	 * until the vendor module is loaded, i.e. enable_mmio_caching can and
74 	 * will change when the vendor module is (re)loaded.
75 	 */
76 	allow_mmio_caching = enable_mmio_caching;
77 
78 	kvm_host.maxphyaddr = kvm_get_host_maxphyaddr();
79 }
80 
generation_mmio_spte_mask(u64 gen)81 static u64 generation_mmio_spte_mask(u64 gen)
82 {
83 	u64 mask;
84 
85 	WARN_ON_ONCE(gen & ~MMIO_SPTE_GEN_MASK);
86 
87 	mask = (gen << MMIO_SPTE_GEN_LOW_SHIFT) & MMIO_SPTE_GEN_LOW_MASK;
88 	mask |= (gen << MMIO_SPTE_GEN_HIGH_SHIFT) & MMIO_SPTE_GEN_HIGH_MASK;
89 	return mask;
90 }
91 
make_mmio_spte(struct kvm_vcpu * vcpu,u64 gfn,unsigned int access)92 u64 make_mmio_spte(struct kvm_vcpu *vcpu, u64 gfn, unsigned int access)
93 {
94 	u64 gen = kvm_vcpu_memslots(vcpu)->generation & MMIO_SPTE_GEN_MASK;
95 	u64 spte = generation_mmio_spte_mask(gen);
96 	u64 gpa = gfn << PAGE_SHIFT;
97 
98 	access &= shadow_mmio_access_mask;
99 	spte |= vcpu->kvm->arch.shadow_mmio_value | access;
100 	spte |= gpa | shadow_nonpresent_or_rsvd_mask;
101 	spte |= (gpa & shadow_nonpresent_or_rsvd_mask)
102 		<< SHADOW_NONPRESENT_OR_RSVD_MASK_LEN;
103 
104 	return spte;
105 }
106 
__kvm_is_mmio_pfn(kvm_pfn_t pfn)107 static bool __kvm_is_mmio_pfn(kvm_pfn_t pfn)
108 {
109 	if (pfn_valid(pfn))
110 		return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn)) &&
111 			/*
112 			 * Some reserved pages, such as those from NVDIMM
113 			 * DAX devices, are not for MMIO, and can be mapped
114 			 * with cached memory type for better performance.
115 			 * However, the above check misconceives those pages
116 			 * as MMIO, and results in KVM mapping them with UC
117 			 * memory type, which would hurt the performance.
118 			 * Therefore, we check the host memory type in addition
119 			 * and only treat UC/UC-/WC pages as MMIO.
120 			 */
121 			(!pat_enabled() || pat_pfn_immune_to_uc_mtrr(pfn));
122 
123 	return !e820__mapped_raw_any(pfn_to_hpa(pfn),
124 				     pfn_to_hpa(pfn + 1) - 1,
125 				     E820_TYPE_RAM);
126 }
127 
kvm_is_mmio_pfn(kvm_pfn_t pfn,int * is_host_mmio)128 static bool kvm_is_mmio_pfn(kvm_pfn_t pfn, int *is_host_mmio)
129 {
130 	/*
131 	 * Determining if a PFN is host MMIO is relative expensive.  Cache the
132 	 * result locally (in the sole caller) to avoid doing the full query
133 	 * multiple times when creating a single SPTE.
134 	 */
135 	if (*is_host_mmio < 0)
136 		*is_host_mmio = __kvm_is_mmio_pfn(pfn);
137 
138 	return *is_host_mmio;
139 }
140 
kvm_track_host_mmio_mapping(struct kvm_vcpu * vcpu)141 static void kvm_track_host_mmio_mapping(struct kvm_vcpu *vcpu)
142 {
143 	struct kvm_mmu_page *root = root_to_sp(vcpu->arch.mmu->root.hpa);
144 
145 	if (root)
146 		WRITE_ONCE(root->has_mapped_host_mmio, true);
147 	else
148 		WRITE_ONCE(vcpu->kvm->arch.has_mapped_host_mmio, true);
149 
150 	/*
151 	 * Force vCPUs to exit and flush CPU buffers if the vCPU is using the
152 	 * affected root(s).
153 	 */
154 	kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_OUTSIDE_GUEST_MODE);
155 }
156 
157 /*
158  * Returns true if the SPTE needs to be updated atomically due to having bits
159  * that may be changed without holding mmu_lock, and for which KVM must not
160  * lose information.  E.g. KVM must not drop Dirty bit information.  The caller
161  * is responsible for checking if the SPTE is shadow-present, and for
162  * determining whether or not the caller cares about non-leaf SPTEs.
163  */
spte_needs_atomic_update(u64 spte)164 bool spte_needs_atomic_update(u64 spte)
165 {
166 	/* SPTEs can be made Writable bit by KVM's fast page fault handler. */
167 	if (!is_writable_pte(spte) && is_mmu_writable_spte(spte))
168 		return true;
169 
170 	/*
171 	 * A/D-disabled SPTEs can be access-tracked by aging, and access-tracked
172 	 * SPTEs can be restored by KVM's fast page fault handler.
173 	 */
174 	if (!spte_ad_enabled(spte))
175 		return true;
176 
177 	/*
178 	 * Dirty and Accessed bits can be set by the CPU.  Ignore the Accessed
179 	 * bit, as KVM tolerates false negatives/positives, e.g. KVM doesn't
180 	 * invalidate TLBs when aging SPTEs, and so it's safe to clobber the
181 	 * Accessed bit (and rare in practice).
182 	 */
183 	return is_writable_pte(spte) && !(spte & shadow_dirty_mask);
184 }
185 
make_spte(struct kvm_vcpu * vcpu,struct kvm_mmu_page * sp,const struct kvm_memory_slot * slot,unsigned int pte_access,gfn_t gfn,kvm_pfn_t pfn,u64 old_spte,bool prefetch,bool synchronizing,bool host_writable,u64 * new_spte)186 bool make_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
187 	       const struct kvm_memory_slot *slot,
188 	       unsigned int pte_access, gfn_t gfn, kvm_pfn_t pfn,
189 	       u64 old_spte, bool prefetch, bool synchronizing,
190 	       bool host_writable, u64 *new_spte)
191 {
192 	int level = sp->role.level;
193 	u64 spte = SPTE_MMU_PRESENT_MASK;
194 	int is_host_mmio = -1;
195 	bool wrprot = false;
196 
197 	/*
198 	 * For the EPT case, shadow_present_mask has no RWX bits set if
199 	 * exec-only page table entries are supported.  In that case,
200 	 * ACC_USER_MASK and shadow_user_mask are used to represent
201 	 * read access.  See FNAME(gpte_access) in paging_tmpl.h.
202 	 */
203 	WARN_ON_ONCE((pte_access | shadow_present_mask) == SHADOW_NONPRESENT_VALUE);
204 
205 	if (sp->role.ad_disabled)
206 		spte |= SPTE_TDP_AD_DISABLED;
207 	else if (kvm_mmu_page_ad_need_write_protect(vcpu->kvm, sp))
208 		spte |= SPTE_TDP_AD_WRPROT_ONLY;
209 
210 	spte |= shadow_present_mask;
211 	if (!prefetch || synchronizing)
212 		spte |= shadow_accessed_mask;
213 
214 	/*
215 	 * For simplicity, enforce the NX huge page mitigation even if not
216 	 * strictly necessary.  KVM could ignore the mitigation if paging is
217 	 * disabled in the guest, as the guest doesn't have any page tables to
218 	 * abuse.  But to safely ignore the mitigation, KVM would have to
219 	 * ensure a new MMU is loaded (or all shadow pages zapped) when CR0.PG
220 	 * is toggled on, and that's a net negative for performance when TDP is
221 	 * enabled.  When TDP is disabled, KVM will always switch to a new MMU
222 	 * when CR0.PG is toggled, but leveraging that to ignore the mitigation
223 	 * would tie make_spte() further to vCPU/MMU state, and add complexity
224 	 * just to optimize a mode that is anything but performance critical.
225 	 */
226 	if (level > PG_LEVEL_4K && (pte_access & ACC_EXEC_MASK) &&
227 	    is_nx_huge_page_enabled(vcpu->kvm)) {
228 		pte_access &= ~ACC_EXEC_MASK;
229 	}
230 
231 	if (pte_access & ACC_EXEC_MASK)
232 		spte |= shadow_x_mask;
233 	else
234 		spte |= shadow_nx_mask;
235 
236 	if (pte_access & ACC_USER_MASK)
237 		spte |= shadow_user_mask;
238 
239 	if (level > PG_LEVEL_4K)
240 		spte |= PT_PAGE_SIZE_MASK;
241 
242 	if (kvm_x86_ops.get_mt_mask)
243 		spte |= kvm_x86_call(get_mt_mask)(vcpu, gfn,
244 						  kvm_is_mmio_pfn(pfn, &is_host_mmio));
245 	if (host_writable)
246 		spte |= shadow_host_writable_mask;
247 	else
248 		pte_access &= ~ACC_WRITE_MASK;
249 
250 	if (shadow_me_value && !kvm_is_mmio_pfn(pfn, &is_host_mmio))
251 		spte |= shadow_me_value;
252 
253 	spte |= (u64)pfn << PAGE_SHIFT;
254 
255 	if (pte_access & ACC_WRITE_MASK) {
256 		/*
257 		 * Unsync shadow pages that are reachable by the new, writable
258 		 * SPTE.  Write-protect the SPTE if the page can't be unsync'd,
259 		 * e.g. it's write-tracked (upper-level SPs) or has one or more
260 		 * shadow pages and unsync'ing pages is not allowed.
261 		 *
262 		 * When overwriting an existing leaf SPTE, and the old SPTE was
263 		 * writable, skip trying to unsync shadow pages as any relevant
264 		 * shadow pages must already be unsync, i.e. the hash lookup is
265 		 * unnecessary (and expensive).  Note, this relies on KVM not
266 		 * changing PFNs without first zapping the old SPTE, which is
267 		 * guaranteed by both the shadow MMU and the TDP MMU.
268 		 */
269 		if ((!is_last_spte(old_spte, level) || !is_writable_pte(old_spte)) &&
270 		    mmu_try_to_unsync_pages(vcpu->kvm, slot, gfn, synchronizing, prefetch))
271 			wrprot = true;
272 		else
273 			spte |= PT_WRITABLE_MASK | shadow_mmu_writable_mask |
274 				shadow_dirty_mask;
275 	}
276 
277 	if (prefetch && !synchronizing)
278 		spte = mark_spte_for_access_track(spte);
279 
280 	WARN_ONCE(is_rsvd_spte(&vcpu->arch.mmu->shadow_zero_check, spte, level),
281 		  "spte = 0x%llx, level = %d, rsvd bits = 0x%llx", spte, level,
282 		  get_rsvd_bits(&vcpu->arch.mmu->shadow_zero_check, spte, level));
283 
284 	/*
285 	 * Mark the memslot dirty *after* modifying it for access tracking.
286 	 * Unlike folios, memslots can be safely marked dirty out of mmu_lock,
287 	 * i.e. in the fast page fault handler.
288 	 */
289 	if ((spte & PT_WRITABLE_MASK) && kvm_slot_dirty_track_enabled(slot)) {
290 		/* Enforced by kvm_mmu_hugepage_adjust. */
291 		WARN_ON_ONCE(level > PG_LEVEL_4K);
292 		mark_page_dirty_in_slot(vcpu->kvm, slot, gfn);
293 	}
294 
295 	if (static_branch_unlikely(&cpu_buf_vm_clear) &&
296 	    !kvm_vcpu_can_access_host_mmio(vcpu) &&
297 	    kvm_is_mmio_pfn(pfn, &is_host_mmio))
298 		kvm_track_host_mmio_mapping(vcpu);
299 
300 	*new_spte = spte;
301 	return wrprot;
302 }
303 
modify_spte_protections(u64 spte,u64 set,u64 clear)304 static u64 modify_spte_protections(u64 spte, u64 set, u64 clear)
305 {
306 	bool is_access_track = is_access_track_spte(spte);
307 
308 	if (is_access_track)
309 		spte = restore_acc_track_spte(spte);
310 
311 	KVM_MMU_WARN_ON(set & clear);
312 	spte = (spte | set) & ~clear;
313 
314 	if (is_access_track)
315 		spte = mark_spte_for_access_track(spte);
316 
317 	return spte;
318 }
319 
make_spte_executable(u64 spte)320 static u64 make_spte_executable(u64 spte)
321 {
322 	return modify_spte_protections(spte, shadow_x_mask, shadow_nx_mask);
323 }
324 
make_spte_nonexecutable(u64 spte)325 static u64 make_spte_nonexecutable(u64 spte)
326 {
327 	return modify_spte_protections(spte, shadow_nx_mask, shadow_x_mask);
328 }
329 
330 /*
331  * Construct an SPTE that maps a sub-page of the given huge page SPTE where
332  * `index` identifies which sub-page.
333  *
334  * This is used during huge page splitting to build the SPTEs that make up the
335  * new page table.
336  */
make_small_spte(struct kvm * kvm,u64 huge_spte,union kvm_mmu_page_role role,int index)337 u64 make_small_spte(struct kvm *kvm, u64 huge_spte,
338 		    union kvm_mmu_page_role role, int index)
339 {
340 	u64 child_spte = huge_spte;
341 
342 	KVM_BUG_ON(!is_shadow_present_pte(huge_spte) || !is_large_pte(huge_spte), kvm);
343 
344 	/*
345 	 * The child_spte already has the base address of the huge page being
346 	 * split. So we just have to OR in the offset to the page at the next
347 	 * lower level for the given index.
348 	 */
349 	child_spte |= (index * KVM_PAGES_PER_HPAGE(role.level)) << PAGE_SHIFT;
350 
351 	if (role.level == PG_LEVEL_4K) {
352 		child_spte &= ~PT_PAGE_SIZE_MASK;
353 
354 		/*
355 		 * When splitting to a 4K page where execution is allowed, mark
356 		 * the page executable as the NX hugepage mitigation no longer
357 		 * applies.
358 		 */
359 		if ((role.access & ACC_EXEC_MASK) && is_nx_huge_page_enabled(kvm))
360 			child_spte = make_spte_executable(child_spte);
361 	}
362 
363 	return child_spte;
364 }
365 
make_huge_spte(struct kvm * kvm,u64 small_spte,int level)366 u64 make_huge_spte(struct kvm *kvm, u64 small_spte, int level)
367 {
368 	u64 huge_spte;
369 
370 	KVM_BUG_ON(!is_shadow_present_pte(small_spte) || level == PG_LEVEL_4K, kvm);
371 
372 	huge_spte = small_spte | PT_PAGE_SIZE_MASK;
373 
374 	/*
375 	 * huge_spte already has the address of the sub-page being collapsed
376 	 * from small_spte, so just clear the lower address bits to create the
377 	 * huge page address.
378 	 */
379 	huge_spte &= KVM_HPAGE_MASK(level) | ~PAGE_MASK;
380 
381 	if (is_nx_huge_page_enabled(kvm))
382 		huge_spte = make_spte_nonexecutable(huge_spte);
383 
384 	return huge_spte;
385 }
386 
make_nonleaf_spte(u64 * child_pt,bool ad_disabled)387 u64 make_nonleaf_spte(u64 *child_pt, bool ad_disabled)
388 {
389 	u64 spte = SPTE_MMU_PRESENT_MASK;
390 
391 	spte |= __pa(child_pt) | shadow_present_mask | PT_WRITABLE_MASK |
392 		shadow_user_mask | shadow_x_mask | shadow_me_value;
393 
394 	if (ad_disabled)
395 		spte |= SPTE_TDP_AD_DISABLED;
396 	else
397 		spte |= shadow_accessed_mask;
398 
399 	return spte;
400 }
401 
mark_spte_for_access_track(u64 spte)402 u64 mark_spte_for_access_track(u64 spte)
403 {
404 	if (spte_ad_enabled(spte))
405 		return spte & ~shadow_accessed_mask;
406 
407 	if (is_access_track_spte(spte))
408 		return spte;
409 
410 	check_spte_writable_invariants(spte);
411 
412 	WARN_ONCE(spte & (SHADOW_ACC_TRACK_SAVED_BITS_MASK <<
413 			  SHADOW_ACC_TRACK_SAVED_BITS_SHIFT),
414 		  "Access Tracking saved bit locations are not zero\n");
415 
416 	spte |= (spte & SHADOW_ACC_TRACK_SAVED_BITS_MASK) <<
417 		SHADOW_ACC_TRACK_SAVED_BITS_SHIFT;
418 	spte &= ~(shadow_acc_track_mask | shadow_accessed_mask);
419 
420 	return spte;
421 }
422 
kvm_mmu_set_mmio_spte_mask(u64 mmio_value,u64 mmio_mask,u64 access_mask)423 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask)
424 {
425 	BUG_ON((u64)(unsigned)access_mask != access_mask);
426 	WARN_ON(mmio_value & shadow_nonpresent_or_rsvd_lower_gfn_mask);
427 
428 	/*
429 	 * Reset to the original module param value to honor userspace's desire
430 	 * to (dis)allow MMIO caching.  Update the param itself so that
431 	 * userspace can see whether or not KVM is actually using MMIO caching.
432 	 */
433 	enable_mmio_caching = allow_mmio_caching;
434 	if (!enable_mmio_caching)
435 		mmio_value = 0;
436 
437 	/*
438 	 * The mask must contain only bits that are carved out specifically for
439 	 * the MMIO SPTE mask, e.g. to ensure there's no overlap with the MMIO
440 	 * generation.
441 	 */
442 	if (WARN_ON(mmio_mask & ~SPTE_MMIO_ALLOWED_MASK))
443 		mmio_value = 0;
444 
445 	/*
446 	 * Disable MMIO caching if the MMIO value collides with the bits that
447 	 * are used to hold the relocated GFN when the L1TF mitigation is
448 	 * enabled.  This should never fire as there is no known hardware that
449 	 * can trigger this condition, e.g. SME/SEV CPUs that require a custom
450 	 * MMIO value are not susceptible to L1TF.
451 	 */
452 	if (WARN_ON(mmio_value & (shadow_nonpresent_or_rsvd_mask <<
453 				  SHADOW_NONPRESENT_OR_RSVD_MASK_LEN)))
454 		mmio_value = 0;
455 
456 	/*
457 	 * The masked MMIO value must obviously match itself and a frozen SPTE
458 	 * must not get a false positive.  Frozen SPTEs and MMIO SPTEs should
459 	 * never collide as MMIO must set some RWX bits, and frozen SPTEs must
460 	 * not set any RWX bits.
461 	 */
462 	if (WARN_ON((mmio_value & mmio_mask) != mmio_value) ||
463 	    WARN_ON(mmio_value && (FROZEN_SPTE & mmio_mask) == mmio_value))
464 		mmio_value = 0;
465 
466 	if (!mmio_value)
467 		enable_mmio_caching = false;
468 
469 	shadow_mmio_value = mmio_value;
470 	shadow_mmio_mask  = mmio_mask;
471 	shadow_mmio_access_mask = access_mask;
472 }
473 EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_mask);
474 
kvm_mmu_set_mmio_spte_value(struct kvm * kvm,u64 mmio_value)475 void kvm_mmu_set_mmio_spte_value(struct kvm *kvm, u64 mmio_value)
476 {
477 	kvm->arch.shadow_mmio_value = mmio_value;
478 }
479 EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_value);
480 
kvm_mmu_set_me_spte_mask(u64 me_value,u64 me_mask)481 void kvm_mmu_set_me_spte_mask(u64 me_value, u64 me_mask)
482 {
483 	/* shadow_me_value must be a subset of shadow_me_mask */
484 	if (WARN_ON(me_value & ~me_mask))
485 		me_value = me_mask = 0;
486 
487 	shadow_me_value = me_value;
488 	shadow_me_mask = me_mask;
489 }
490 EXPORT_SYMBOL_GPL(kvm_mmu_set_me_spte_mask);
491 
kvm_mmu_set_ept_masks(bool has_ad_bits,bool has_exec_only)492 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only)
493 {
494 	kvm_ad_enabled		= has_ad_bits;
495 
496 	shadow_user_mask	= VMX_EPT_READABLE_MASK;
497 	shadow_accessed_mask	= VMX_EPT_ACCESS_BIT;
498 	shadow_dirty_mask	= VMX_EPT_DIRTY_BIT;
499 	shadow_nx_mask		= 0ull;
500 	shadow_x_mask		= VMX_EPT_EXECUTABLE_MASK;
501 	/* VMX_EPT_SUPPRESS_VE_BIT is needed for W or X violation. */
502 	shadow_present_mask	=
503 		(has_exec_only ? 0ull : VMX_EPT_READABLE_MASK) | VMX_EPT_SUPPRESS_VE_BIT;
504 
505 	shadow_acc_track_mask	= VMX_EPT_RWX_MASK;
506 	shadow_host_writable_mask = EPT_SPTE_HOST_WRITABLE;
507 	shadow_mmu_writable_mask  = EPT_SPTE_MMU_WRITABLE;
508 
509 	/*
510 	 * EPT Misconfigurations are generated if the value of bits 2:0
511 	 * of an EPT paging-structure entry is 110b (write/execute).
512 	 */
513 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE,
514 				   VMX_EPT_RWX_MASK | VMX_EPT_SUPPRESS_VE_BIT, 0);
515 }
516 EXPORT_SYMBOL_GPL(kvm_mmu_set_ept_masks);
517 
kvm_mmu_reset_all_pte_masks(void)518 void kvm_mmu_reset_all_pte_masks(void)
519 {
520 	u8 low_phys_bits;
521 	u64 mask;
522 
523 	kvm_ad_enabled = true;
524 
525 	/*
526 	 * If the CPU has 46 or less physical address bits, then set an
527 	 * appropriate mask to guard against L1TF attacks. Otherwise, it is
528 	 * assumed that the CPU is not vulnerable to L1TF.
529 	 *
530 	 * Some Intel CPUs address the L1 cache using more PA bits than are
531 	 * reported by CPUID. Use the PA width of the L1 cache when possible
532 	 * to achieve more effective mitigation, e.g. if system RAM overlaps
533 	 * the most significant bits of legal physical address space.
534 	 */
535 	shadow_nonpresent_or_rsvd_mask = 0;
536 	low_phys_bits = boot_cpu_data.x86_phys_bits;
537 	if (boot_cpu_has_bug(X86_BUG_L1TF) &&
538 	    !WARN_ON_ONCE(boot_cpu_data.x86_cache_bits >=
539 			  52 - SHADOW_NONPRESENT_OR_RSVD_MASK_LEN)) {
540 		low_phys_bits = boot_cpu_data.x86_cache_bits
541 			- SHADOW_NONPRESENT_OR_RSVD_MASK_LEN;
542 		shadow_nonpresent_or_rsvd_mask =
543 			rsvd_bits(low_phys_bits, boot_cpu_data.x86_cache_bits - 1);
544 	}
545 
546 	shadow_nonpresent_or_rsvd_lower_gfn_mask =
547 		GENMASK_ULL(low_phys_bits - 1, PAGE_SHIFT);
548 
549 	shadow_user_mask	= PT_USER_MASK;
550 	shadow_accessed_mask	= PT_ACCESSED_MASK;
551 	shadow_dirty_mask	= PT_DIRTY_MASK;
552 	shadow_nx_mask		= PT64_NX_MASK;
553 	shadow_x_mask		= 0;
554 	shadow_present_mask	= PT_PRESENT_MASK;
555 
556 	shadow_acc_track_mask	= 0;
557 	shadow_me_mask		= 0;
558 	shadow_me_value		= 0;
559 
560 	shadow_host_writable_mask = DEFAULT_SPTE_HOST_WRITABLE;
561 	shadow_mmu_writable_mask  = DEFAULT_SPTE_MMU_WRITABLE;
562 
563 	/*
564 	 * Set a reserved PA bit in MMIO SPTEs to generate page faults with
565 	 * PFEC.RSVD=1 on MMIO accesses.  64-bit PTEs (PAE, x86-64, and EPT
566 	 * paging) support a maximum of 52 bits of PA, i.e. if the CPU supports
567 	 * 52-bit physical addresses then there are no reserved PA bits in the
568 	 * PTEs and so the reserved PA approach must be disabled.
569 	 */
570 	if (kvm_host.maxphyaddr < 52)
571 		mask = BIT_ULL(51) | PT_PRESENT_MASK;
572 	else
573 		mask = 0;
574 
575 	kvm_mmu_set_mmio_spte_mask(mask, mask, ACC_WRITE_MASK | ACC_USER_MASK);
576 }
577