xref: /openbmc/linux/arch/x86/kvm/mmu/spte.h (revision e40573a4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 #ifndef KVM_X86_MMU_SPTE_H
4 #define KVM_X86_MMU_SPTE_H
5 
6 #include "mmu_internal.h"
7 
8 /*
9  * A MMU present SPTE is backed by actual memory and may or may not be present
10  * in hardware.  E.g. MMIO SPTEs are not considered present.  Use bit 11, as it
11  * is ignored by all flavors of SPTEs and checking a low bit often generates
12  * better code than for a high bit, e.g. 56+.  MMU present checks are pervasive
13  * enough that the improved code generation is noticeable in KVM's footprint.
14  */
15 #define SPTE_MMU_PRESENT_MASK		BIT_ULL(11)
16 
17 /*
18  * TDP SPTES (more specifically, EPT SPTEs) may not have A/D bits, and may also
19  * be restricted to using write-protection (for L2 when CPU dirty logging, i.e.
20  * PML, is enabled).  Use bits 52 and 53 to hold the type of A/D tracking that
21  * is must be employed for a given TDP SPTE.
22  *
23  * Note, the "enabled" mask must be '0', as bits 62:52 are _reserved_ for PAE
24  * paging, including NPT PAE.  This scheme works because legacy shadow paging
25  * is guaranteed to have A/D bits and write-protection is forced only for
26  * TDP with CPU dirty logging (PML).  If NPT ever gains PML-like support, it
27  * must be restricted to 64-bit KVM.
28  */
29 #define SPTE_TDP_AD_SHIFT		52
30 #define SPTE_TDP_AD_MASK		(3ULL << SPTE_TDP_AD_SHIFT)
31 #define SPTE_TDP_AD_ENABLED_MASK	(0ULL << SPTE_TDP_AD_SHIFT)
32 #define SPTE_TDP_AD_DISABLED_MASK	(1ULL << SPTE_TDP_AD_SHIFT)
33 #define SPTE_TDP_AD_WRPROT_ONLY_MASK	(2ULL << SPTE_TDP_AD_SHIFT)
34 static_assert(SPTE_TDP_AD_ENABLED_MASK == 0);
35 
36 #ifdef CONFIG_DYNAMIC_PHYSICAL_MASK
37 #define SPTE_BASE_ADDR_MASK (physical_mask & ~(u64)(PAGE_SIZE-1))
38 #else
39 #define SPTE_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
40 #endif
41 
42 #define SPTE_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | shadow_user_mask \
43 			| shadow_x_mask | shadow_nx_mask | shadow_me_mask)
44 
45 #define ACC_EXEC_MASK    1
46 #define ACC_WRITE_MASK   PT_WRITABLE_MASK
47 #define ACC_USER_MASK    PT_USER_MASK
48 #define ACC_ALL          (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
49 
50 /* The mask for the R/X bits in EPT PTEs */
51 #define SPTE_EPT_READABLE_MASK			0x1ull
52 #define SPTE_EPT_EXECUTABLE_MASK		0x4ull
53 
54 #define SPTE_LEVEL_BITS			9
55 #define SPTE_LEVEL_SHIFT(level)		__PT_LEVEL_SHIFT(level, SPTE_LEVEL_BITS)
56 #define SPTE_INDEX(address, level)	__PT_INDEX(address, level, SPTE_LEVEL_BITS)
57 #define SPTE_ENT_PER_PAGE		__PT_ENT_PER_PAGE(SPTE_LEVEL_BITS)
58 
59 /*
60  * The mask/shift to use for saving the original R/X bits when marking the PTE
61  * as not-present for access tracking purposes. We do not save the W bit as the
62  * PTEs being access tracked also need to be dirty tracked, so the W bit will be
63  * restored only when a write is attempted to the page.  This mask obviously
64  * must not overlap the A/D type mask.
65  */
66 #define SHADOW_ACC_TRACK_SAVED_BITS_MASK (SPTE_EPT_READABLE_MASK | \
67 					  SPTE_EPT_EXECUTABLE_MASK)
68 #define SHADOW_ACC_TRACK_SAVED_BITS_SHIFT 54
69 #define SHADOW_ACC_TRACK_SAVED_MASK	(SHADOW_ACC_TRACK_SAVED_BITS_MASK << \
70 					 SHADOW_ACC_TRACK_SAVED_BITS_SHIFT)
71 static_assert(!(SPTE_TDP_AD_MASK & SHADOW_ACC_TRACK_SAVED_MASK));
72 
73 /*
74  * {DEFAULT,EPT}_SPTE_{HOST,MMU}_WRITABLE are used to keep track of why a given
75  * SPTE is write-protected. See is_writable_pte() for details.
76  */
77 
78 /* Bits 9 and 10 are ignored by all non-EPT PTEs. */
79 #define DEFAULT_SPTE_HOST_WRITABLE	BIT_ULL(9)
80 #define DEFAULT_SPTE_MMU_WRITABLE	BIT_ULL(10)
81 
82 /*
83  * Low ignored bits are at a premium for EPT, use high ignored bits, taking care
84  * to not overlap the A/D type mask or the saved access bits of access-tracked
85  * SPTEs when A/D bits are disabled.
86  */
87 #define EPT_SPTE_HOST_WRITABLE		BIT_ULL(57)
88 #define EPT_SPTE_MMU_WRITABLE		BIT_ULL(58)
89 
90 static_assert(!(EPT_SPTE_HOST_WRITABLE & SPTE_TDP_AD_MASK));
91 static_assert(!(EPT_SPTE_MMU_WRITABLE & SPTE_TDP_AD_MASK));
92 static_assert(!(EPT_SPTE_HOST_WRITABLE & SHADOW_ACC_TRACK_SAVED_MASK));
93 static_assert(!(EPT_SPTE_MMU_WRITABLE & SHADOW_ACC_TRACK_SAVED_MASK));
94 
95 /* Defined only to keep the above static asserts readable. */
96 #undef SHADOW_ACC_TRACK_SAVED_MASK
97 
98 /*
99  * Due to limited space in PTEs, the MMIO generation is a 19 bit subset of
100  * the memslots generation and is derived as follows:
101  *
102  * Bits 0-7 of the MMIO generation are propagated to spte bits 3-10
103  * Bits 8-18 of the MMIO generation are propagated to spte bits 52-62
104  *
105  * The KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS flag is intentionally not included in
106  * the MMIO generation number, as doing so would require stealing a bit from
107  * the "real" generation number and thus effectively halve the maximum number
108  * of MMIO generations that can be handled before encountering a wrap (which
109  * requires a full MMU zap).  The flag is instead explicitly queried when
110  * checking for MMIO spte cache hits.
111  */
112 
113 #define MMIO_SPTE_GEN_LOW_START		3
114 #define MMIO_SPTE_GEN_LOW_END		10
115 
116 #define MMIO_SPTE_GEN_HIGH_START	52
117 #define MMIO_SPTE_GEN_HIGH_END		62
118 
119 #define MMIO_SPTE_GEN_LOW_MASK		GENMASK_ULL(MMIO_SPTE_GEN_LOW_END, \
120 						    MMIO_SPTE_GEN_LOW_START)
121 #define MMIO_SPTE_GEN_HIGH_MASK		GENMASK_ULL(MMIO_SPTE_GEN_HIGH_END, \
122 						    MMIO_SPTE_GEN_HIGH_START)
123 static_assert(!(SPTE_MMU_PRESENT_MASK &
124 		(MMIO_SPTE_GEN_LOW_MASK | MMIO_SPTE_GEN_HIGH_MASK)));
125 
126 /*
127  * The SPTE MMIO mask must NOT overlap the MMIO generation bits or the
128  * MMU-present bit.  The generation obviously co-exists with the magic MMIO
129  * mask/value, and MMIO SPTEs are considered !MMU-present.
130  *
131  * The SPTE MMIO mask is allowed to use hardware "present" bits (i.e. all EPT
132  * RWX bits), all physical address bits (legal PA bits are used for "fast" MMIO
133  * and so they're off-limits for generation; additional checks ensure the mask
134  * doesn't overlap legal PA bits), and bit 63 (carved out for future usage).
135  */
136 #define SPTE_MMIO_ALLOWED_MASK (BIT_ULL(63) | GENMASK_ULL(51, 12) | GENMASK_ULL(2, 0))
137 static_assert(!(SPTE_MMIO_ALLOWED_MASK &
138 		(SPTE_MMU_PRESENT_MASK | MMIO_SPTE_GEN_LOW_MASK | MMIO_SPTE_GEN_HIGH_MASK)));
139 
140 #define MMIO_SPTE_GEN_LOW_BITS		(MMIO_SPTE_GEN_LOW_END - MMIO_SPTE_GEN_LOW_START + 1)
141 #define MMIO_SPTE_GEN_HIGH_BITS		(MMIO_SPTE_GEN_HIGH_END - MMIO_SPTE_GEN_HIGH_START + 1)
142 
143 /* remember to adjust the comment above as well if you change these */
144 static_assert(MMIO_SPTE_GEN_LOW_BITS == 8 && MMIO_SPTE_GEN_HIGH_BITS == 11);
145 
146 #define MMIO_SPTE_GEN_LOW_SHIFT		(MMIO_SPTE_GEN_LOW_START - 0)
147 #define MMIO_SPTE_GEN_HIGH_SHIFT	(MMIO_SPTE_GEN_HIGH_START - MMIO_SPTE_GEN_LOW_BITS)
148 
149 #define MMIO_SPTE_GEN_MASK		GENMASK_ULL(MMIO_SPTE_GEN_LOW_BITS + MMIO_SPTE_GEN_HIGH_BITS - 1, 0)
150 
151 extern u64 __read_mostly shadow_host_writable_mask;
152 extern u64 __read_mostly shadow_mmu_writable_mask;
153 extern u64 __read_mostly shadow_nx_mask;
154 extern u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
155 extern u64 __read_mostly shadow_user_mask;
156 extern u64 __read_mostly shadow_accessed_mask;
157 extern u64 __read_mostly shadow_dirty_mask;
158 extern u64 __read_mostly shadow_mmio_value;
159 extern u64 __read_mostly shadow_mmio_mask;
160 extern u64 __read_mostly shadow_mmio_access_mask;
161 extern u64 __read_mostly shadow_present_mask;
162 extern u64 __read_mostly shadow_memtype_mask;
163 extern u64 __read_mostly shadow_me_value;
164 extern u64 __read_mostly shadow_me_mask;
165 
166 /*
167  * SPTEs in MMUs without A/D bits are marked with SPTE_TDP_AD_DISABLED_MASK;
168  * shadow_acc_track_mask is the set of bits to be cleared in non-accessed
169  * pages.
170  */
171 extern u64 __read_mostly shadow_acc_track_mask;
172 
173 /*
174  * This mask must be set on all non-zero Non-Present or Reserved SPTEs in order
175  * to guard against L1TF attacks.
176  */
177 extern u64 __read_mostly shadow_nonpresent_or_rsvd_mask;
178 
179 /*
180  * The number of high-order 1 bits to use in the mask above.
181  */
182 #define SHADOW_NONPRESENT_OR_RSVD_MASK_LEN 5
183 
184 /*
185  * If a thread running without exclusive control of the MMU lock must perform a
186  * multi-part operation on an SPTE, it can set the SPTE to REMOVED_SPTE as a
187  * non-present intermediate value. Other threads which encounter this value
188  * should not modify the SPTE.
189  *
190  * Use a semi-arbitrary value that doesn't set RWX bits, i.e. is not-present on
191  * bot AMD and Intel CPUs, and doesn't set PFN bits, i.e. doesn't create a L1TF
192  * vulnerability.  Use only low bits to avoid 64-bit immediates.
193  *
194  * Only used by the TDP MMU.
195  */
196 #define REMOVED_SPTE	0x5a0ULL
197 
198 /* Removed SPTEs must not be misconstrued as shadow present PTEs. */
199 static_assert(!(REMOVED_SPTE & SPTE_MMU_PRESENT_MASK));
200 
201 static inline bool is_removed_spte(u64 spte)
202 {
203 	return spte == REMOVED_SPTE;
204 }
205 
206 /* Get an SPTE's index into its parent's page table (and the spt array). */
207 static inline int spte_index(u64 *sptep)
208 {
209 	return ((unsigned long)sptep / sizeof(*sptep)) & (SPTE_ENT_PER_PAGE - 1);
210 }
211 
212 /*
213  * In some cases, we need to preserve the GFN of a non-present or reserved
214  * SPTE when we usurp the upper five bits of the physical address space to
215  * defend against L1TF, e.g. for MMIO SPTEs.  To preserve the GFN, we'll
216  * shift bits of the GFN that overlap with shadow_nonpresent_or_rsvd_mask
217  * left into the reserved bits, i.e. the GFN in the SPTE will be split into
218  * high and low parts.  This mask covers the lower bits of the GFN.
219  */
220 extern u64 __read_mostly shadow_nonpresent_or_rsvd_lower_gfn_mask;
221 
222 static inline bool is_mmio_spte(u64 spte)
223 {
224 	return (spte & shadow_mmio_mask) == shadow_mmio_value &&
225 	       likely(enable_mmio_caching);
226 }
227 
228 static inline bool is_shadow_present_pte(u64 pte)
229 {
230 	return !!(pte & SPTE_MMU_PRESENT_MASK);
231 }
232 
233 /*
234  * Returns true if A/D bits are supported in hardware and are enabled by KVM.
235  * When enabled, KVM uses A/D bits for all non-nested MMUs.  Because L1 can
236  * disable A/D bits in EPTP12, SP and SPTE variants are needed to handle the
237  * scenario where KVM is using A/D bits for L1, but not L2.
238  */
239 static inline bool kvm_ad_enabled(void)
240 {
241 	return !!shadow_accessed_mask;
242 }
243 
244 static inline bool sp_ad_disabled(struct kvm_mmu_page *sp)
245 {
246 	return sp->role.ad_disabled;
247 }
248 
249 static inline bool spte_ad_enabled(u64 spte)
250 {
251 	MMU_WARN_ON(!is_shadow_present_pte(spte));
252 	return (spte & SPTE_TDP_AD_MASK) != SPTE_TDP_AD_DISABLED_MASK;
253 }
254 
255 static inline bool spte_ad_need_write_protect(u64 spte)
256 {
257 	MMU_WARN_ON(!is_shadow_present_pte(spte));
258 	/*
259 	 * This is benign for non-TDP SPTEs as SPTE_TDP_AD_ENABLED_MASK is '0',
260 	 * and non-TDP SPTEs will never set these bits.  Optimize for 64-bit
261 	 * TDP and do the A/D type check unconditionally.
262 	 */
263 	return (spte & SPTE_TDP_AD_MASK) != SPTE_TDP_AD_ENABLED_MASK;
264 }
265 
266 static inline u64 spte_shadow_accessed_mask(u64 spte)
267 {
268 	MMU_WARN_ON(!is_shadow_present_pte(spte));
269 	return spte_ad_enabled(spte) ? shadow_accessed_mask : 0;
270 }
271 
272 static inline u64 spte_shadow_dirty_mask(u64 spte)
273 {
274 	MMU_WARN_ON(!is_shadow_present_pte(spte));
275 	return spte_ad_enabled(spte) ? shadow_dirty_mask : 0;
276 }
277 
278 static inline bool is_access_track_spte(u64 spte)
279 {
280 	return !spte_ad_enabled(spte) && (spte & shadow_acc_track_mask) == 0;
281 }
282 
283 static inline bool is_large_pte(u64 pte)
284 {
285 	return pte & PT_PAGE_SIZE_MASK;
286 }
287 
288 static inline bool is_last_spte(u64 pte, int level)
289 {
290 	return (level == PG_LEVEL_4K) || is_large_pte(pte);
291 }
292 
293 static inline bool is_executable_pte(u64 spte)
294 {
295 	return (spte & (shadow_x_mask | shadow_nx_mask)) == shadow_x_mask;
296 }
297 
298 static inline kvm_pfn_t spte_to_pfn(u64 pte)
299 {
300 	return (pte & SPTE_BASE_ADDR_MASK) >> PAGE_SHIFT;
301 }
302 
303 static inline bool is_accessed_spte(u64 spte)
304 {
305 	u64 accessed_mask = spte_shadow_accessed_mask(spte);
306 
307 	return accessed_mask ? spte & accessed_mask
308 			     : !is_access_track_spte(spte);
309 }
310 
311 static inline bool is_dirty_spte(u64 spte)
312 {
313 	u64 dirty_mask = spte_shadow_dirty_mask(spte);
314 
315 	return dirty_mask ? spte & dirty_mask : spte & PT_WRITABLE_MASK;
316 }
317 
318 static inline u64 get_rsvd_bits(struct rsvd_bits_validate *rsvd_check, u64 pte,
319 				int level)
320 {
321 	int bit7 = (pte >> 7) & 1;
322 
323 	return rsvd_check->rsvd_bits_mask[bit7][level-1];
324 }
325 
326 static inline bool __is_rsvd_bits_set(struct rsvd_bits_validate *rsvd_check,
327 				      u64 pte, int level)
328 {
329 	return pte & get_rsvd_bits(rsvd_check, pte, level);
330 }
331 
332 static inline bool __is_bad_mt_xwr(struct rsvd_bits_validate *rsvd_check,
333 				   u64 pte)
334 {
335 	return rsvd_check->bad_mt_xwr & BIT_ULL(pte & 0x3f);
336 }
337 
338 static __always_inline bool is_rsvd_spte(struct rsvd_bits_validate *rsvd_check,
339 					 u64 spte, int level)
340 {
341 	return __is_bad_mt_xwr(rsvd_check, spte) ||
342 	       __is_rsvd_bits_set(rsvd_check, spte, level);
343 }
344 
345 /*
346  * An shadow-present leaf SPTE may be non-writable for 3 possible reasons:
347  *
348  *  1. To intercept writes for dirty logging. KVM write-protects huge pages
349  *     so that they can be split be split down into the dirty logging
350  *     granularity (4KiB) whenever the guest writes to them. KVM also
351  *     write-protects 4KiB pages so that writes can be recorded in the dirty log
352  *     (e.g. if not using PML). SPTEs are write-protected for dirty logging
353  *     during the VM-iotcls that enable dirty logging.
354  *
355  *  2. To intercept writes to guest page tables that KVM is shadowing. When a
356  *     guest writes to its page table the corresponding shadow page table will
357  *     be marked "unsync". That way KVM knows which shadow page tables need to
358  *     be updated on the next TLB flush, INVLPG, etc. and which do not.
359  *
360  *  3. To prevent guest writes to read-only memory, such as for memory in a
361  *     read-only memslot or guest memory backed by a read-only VMA. Writes to
362  *     such pages are disallowed entirely.
363  *
364  * To keep track of why a given SPTE is write-protected, KVM uses 2
365  * software-only bits in the SPTE:
366  *
367  *  shadow_mmu_writable_mask, aka MMU-writable -
368  *    Cleared on SPTEs that KVM is currently write-protecting for shadow paging
369  *    purposes (case 2 above).
370  *
371  *  shadow_host_writable_mask, aka Host-writable -
372  *    Cleared on SPTEs that are not host-writable (case 3 above)
373  *
374  * Note, not all possible combinations of PT_WRITABLE_MASK,
375  * shadow_mmu_writable_mask, and shadow_host_writable_mask are valid. A given
376  * SPTE can be in only one of the following states, which map to the
377  * aforementioned 3 cases:
378  *
379  *   shadow_host_writable_mask | shadow_mmu_writable_mask | PT_WRITABLE_MASK
380  *   ------------------------- | ------------------------ | ----------------
381  *   1                         | 1                        | 1       (writable)
382  *   1                         | 1                        | 0       (case 1)
383  *   1                         | 0                        | 0       (case 2)
384  *   0                         | 0                        | 0       (case 3)
385  *
386  * The valid combinations of these bits are checked by
387  * check_spte_writable_invariants() whenever an SPTE is modified.
388  *
389  * Clearing the MMU-writable bit is always done under the MMU lock and always
390  * accompanied by a TLB flush before dropping the lock to avoid corrupting the
391  * shadow page tables between vCPUs. Write-protecting an SPTE for dirty logging
392  * (which does not clear the MMU-writable bit), does not flush TLBs before
393  * dropping the lock, as it only needs to synchronize guest writes with the
394  * dirty bitmap.
395  *
396  * So, there is the problem: clearing the MMU-writable bit can encounter a
397  * write-protected SPTE while CPUs still have writable mappings for that SPTE
398  * cached in their TLB. To address this, KVM always flushes TLBs when
399  * write-protecting SPTEs if the MMU-writable bit is set on the old SPTE.
400  *
401  * The Host-writable bit is not modified on present SPTEs, it is only set or
402  * cleared when an SPTE is first faulted in from non-present and then remains
403  * immutable.
404  */
405 static inline bool is_writable_pte(unsigned long pte)
406 {
407 	return pte & PT_WRITABLE_MASK;
408 }
409 
410 /* Note: spte must be a shadow-present leaf SPTE. */
411 static inline void check_spte_writable_invariants(u64 spte)
412 {
413 	if (spte & shadow_mmu_writable_mask)
414 		WARN_ONCE(!(spte & shadow_host_writable_mask),
415 			  "kvm: MMU-writable SPTE is not Host-writable: %llx",
416 			  spte);
417 	else
418 		WARN_ONCE(is_writable_pte(spte),
419 			  "kvm: Writable SPTE is not MMU-writable: %llx", spte);
420 }
421 
422 static inline bool is_mmu_writable_spte(u64 spte)
423 {
424 	return spte & shadow_mmu_writable_mask;
425 }
426 
427 static inline u64 get_mmio_spte_generation(u64 spte)
428 {
429 	u64 gen;
430 
431 	gen = (spte & MMIO_SPTE_GEN_LOW_MASK) >> MMIO_SPTE_GEN_LOW_SHIFT;
432 	gen |= (spte & MMIO_SPTE_GEN_HIGH_MASK) >> MMIO_SPTE_GEN_HIGH_SHIFT;
433 	return gen;
434 }
435 
436 bool spte_has_volatile_bits(u64 spte);
437 
438 bool make_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
439 	       const struct kvm_memory_slot *slot,
440 	       unsigned int pte_access, gfn_t gfn, kvm_pfn_t pfn,
441 	       u64 old_spte, bool prefetch, bool can_unsync,
442 	       bool host_writable, u64 *new_spte);
443 u64 make_huge_page_split_spte(struct kvm *kvm, u64 huge_spte,
444 		      	      union kvm_mmu_page_role role, int index);
445 u64 make_nonleaf_spte(u64 *child_pt, bool ad_disabled);
446 u64 make_mmio_spte(struct kvm_vcpu *vcpu, u64 gfn, unsigned int access);
447 u64 mark_spte_for_access_track(u64 spte);
448 
449 /* Restore an acc-track PTE back to a regular PTE */
450 static inline u64 restore_acc_track_spte(u64 spte)
451 {
452 	u64 saved_bits = (spte >> SHADOW_ACC_TRACK_SAVED_BITS_SHIFT)
453 			 & SHADOW_ACC_TRACK_SAVED_BITS_MASK;
454 
455 	spte &= ~shadow_acc_track_mask;
456 	spte &= ~(SHADOW_ACC_TRACK_SAVED_BITS_MASK <<
457 		  SHADOW_ACC_TRACK_SAVED_BITS_SHIFT);
458 	spte |= saved_bits;
459 
460 	return spte;
461 }
462 
463 u64 kvm_mmu_changed_pte_notifier_make_spte(u64 old_spte, kvm_pfn_t new_pfn);
464 
465 void __init kvm_mmu_spte_module_init(void);
466 void kvm_mmu_reset_all_pte_masks(void);
467 
468 #endif
469