xref: /linux/arch/x86/include/asm/tlbflush.h (revision 52990390f91c1c39ca742fc8f390b29891d95127)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_TLBFLUSH_H
3 #define _ASM_X86_TLBFLUSH_H
4 
5 #include <linux/mm_types.h>
6 #include <linux/sched.h>
7 
8 #include <asm/processor.h>
9 #include <asm/cpufeature.h>
10 #include <asm/special_insns.h>
11 #include <asm/smp.h>
12 #include <asm/invpcid.h>
13 #include <asm/pti.h>
14 #include <asm/processor-flags.h>
15 #include <asm/pgtable.h>
16 
17 void __flush_tlb_all(void);
18 
19 #define TLB_FLUSH_ALL	-1UL
20 #define TLB_GENERATION_INVALID	0
21 
22 void cr4_update_irqsoff(unsigned long set, unsigned long clear);
23 unsigned long cr4_read_shadow(void);
24 
25 /* Set in this cpu's CR4. */
26 static inline void cr4_set_bits_irqsoff(unsigned long mask)
27 {
28 	cr4_update_irqsoff(mask, 0);
29 }
30 
31 /* Clear in this cpu's CR4. */
32 static inline void cr4_clear_bits_irqsoff(unsigned long mask)
33 {
34 	cr4_update_irqsoff(0, mask);
35 }
36 
37 /* Set in this cpu's CR4. */
38 static inline void cr4_set_bits(unsigned long mask)
39 {
40 	unsigned long flags;
41 
42 	local_irq_save(flags);
43 	cr4_set_bits_irqsoff(mask);
44 	local_irq_restore(flags);
45 }
46 
47 /* Clear in this cpu's CR4. */
48 static inline void cr4_clear_bits(unsigned long mask)
49 {
50 	unsigned long flags;
51 
52 	local_irq_save(flags);
53 	cr4_clear_bits_irqsoff(mask);
54 	local_irq_restore(flags);
55 }
56 
57 #ifdef CONFIG_ADDRESS_MASKING
58 DECLARE_PER_CPU(u64, tlbstate_untag_mask);
59 
60 static inline u64 current_untag_mask(void)
61 {
62 	return this_cpu_read(tlbstate_untag_mask);
63 }
64 #endif
65 
66 #ifndef MODULE
67 /*
68  * 6 because 6 should be plenty and struct tlb_state will fit in two cache
69  * lines.
70  */
71 #define TLB_NR_DYN_ASIDS	6
72 
73 struct tlb_context {
74 	u64 ctx_id;
75 	u64 tlb_gen;
76 };
77 
78 struct tlb_state {
79 	/*
80 	 * cpu_tlbstate.loaded_mm should match CR3 whenever interrupts
81 	 * are on.  This means that it may not match current->active_mm,
82 	 * which will contain the previous user mm when we're in lazy TLB
83 	 * mode even if we've already switched back to swapper_pg_dir.
84 	 *
85 	 * During switch_mm_irqs_off(), loaded_mm will be set to
86 	 * LOADED_MM_SWITCHING during the brief interrupts-off window
87 	 * when CR3 and loaded_mm would otherwise be inconsistent.  This
88 	 * is for nmi_uaccess_okay()'s benefit.
89 	 */
90 	struct mm_struct *loaded_mm;
91 
92 #define LOADED_MM_SWITCHING ((struct mm_struct *)1UL)
93 
94 	/* Last user mm for optimizing IBPB */
95 	union {
96 		struct mm_struct	*last_user_mm;
97 		unsigned long		last_user_mm_spec;
98 	};
99 
100 	u16 loaded_mm_asid;
101 	u16 next_asid;
102 
103 	/*
104 	 * If set we changed the page tables in such a way that we
105 	 * needed an invalidation of all contexts (aka. PCIDs / ASIDs).
106 	 * This tells us to go invalidate all the non-loaded ctxs[]
107 	 * on the next context switch.
108 	 *
109 	 * The current ctx was kept up-to-date as it ran and does not
110 	 * need to be invalidated.
111 	 */
112 	bool invalidate_other;
113 
114 #ifdef CONFIG_ADDRESS_MASKING
115 	/*
116 	 * Active LAM mode.
117 	 *
118 	 * X86_CR3_LAM_U57/U48 shifted right by X86_CR3_LAM_U57_BIT or 0 if LAM
119 	 * disabled.
120 	 */
121 	u8 lam;
122 #endif
123 
124 	/*
125 	 * Mask that contains TLB_NR_DYN_ASIDS+1 bits to indicate
126 	 * the corresponding user PCID needs a flush next time we
127 	 * switch to it; see SWITCH_TO_USER_CR3.
128 	 */
129 	unsigned short user_pcid_flush_mask;
130 
131 	/*
132 	 * Access to this CR4 shadow and to H/W CR4 is protected by
133 	 * disabling interrupts when modifying either one.
134 	 */
135 	unsigned long cr4;
136 
137 	/*
138 	 * This is a list of all contexts that might exist in the TLB.
139 	 * There is one per ASID that we use, and the ASID (what the
140 	 * CPU calls PCID) is the index into ctxts.
141 	 *
142 	 * For each context, ctx_id indicates which mm the TLB's user
143 	 * entries came from.  As an invariant, the TLB will never
144 	 * contain entries that are out-of-date as when that mm reached
145 	 * the tlb_gen in the list.
146 	 *
147 	 * To be clear, this means that it's legal for the TLB code to
148 	 * flush the TLB without updating tlb_gen.  This can happen
149 	 * (for now, at least) due to paravirt remote flushes.
150 	 *
151 	 * NB: context 0 is a bit special, since it's also used by
152 	 * various bits of init code.  This is fine -- code that
153 	 * isn't aware of PCID will end up harmlessly flushing
154 	 * context 0.
155 	 */
156 	struct tlb_context ctxs[TLB_NR_DYN_ASIDS];
157 };
158 DECLARE_PER_CPU_ALIGNED(struct tlb_state, cpu_tlbstate);
159 
160 struct tlb_state_shared {
161 	/*
162 	 * We can be in one of several states:
163 	 *
164 	 *  - Actively using an mm.  Our CPU's bit will be set in
165 	 *    mm_cpumask(loaded_mm) and is_lazy == false;
166 	 *
167 	 *  - Not using a real mm.  loaded_mm == &init_mm.  Our CPU's bit
168 	 *    will not be set in mm_cpumask(&init_mm) and is_lazy == false.
169 	 *
170 	 *  - Lazily using a real mm.  loaded_mm != &init_mm, our bit
171 	 *    is set in mm_cpumask(loaded_mm), but is_lazy == true.
172 	 *    We're heuristically guessing that the CR3 load we
173 	 *    skipped more than makes up for the overhead added by
174 	 *    lazy mode.
175 	 */
176 	bool is_lazy;
177 };
178 DECLARE_PER_CPU_SHARED_ALIGNED(struct tlb_state_shared, cpu_tlbstate_shared);
179 
180 bool nmi_uaccess_okay(void);
181 #define nmi_uaccess_okay nmi_uaccess_okay
182 
183 /* Initialize cr4 shadow for this CPU. */
184 static inline void cr4_init_shadow(void)
185 {
186 	this_cpu_write(cpu_tlbstate.cr4, __read_cr4());
187 }
188 
189 extern unsigned long mmu_cr4_features;
190 extern u32 *trampoline_cr4_features;
191 
192 extern void initialize_tlbstate_and_flush(void);
193 
194 /*
195  * TLB flushing:
196  *
197  *  - flush_tlb_all() flushes all processes TLBs
198  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
199  *  - flush_tlb_page(vma, vmaddr) flushes one page
200  *  - flush_tlb_range(vma, start, end) flushes a range of pages
201  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
202  *  - flush_tlb_multi(cpumask, info) flushes TLBs on multiple cpus
203  *
204  * ..but the i386 has somewhat limited tlb flushing capabilities,
205  * and page-granular flushes are available only on i486 and up.
206  */
207 struct flush_tlb_info {
208 	/*
209 	 * We support several kinds of flushes.
210 	 *
211 	 * - Fully flush a single mm.  .mm will be set, .end will be
212 	 *   TLB_FLUSH_ALL, and .new_tlb_gen will be the tlb_gen to
213 	 *   which the IPI sender is trying to catch us up.
214 	 *
215 	 * - Partially flush a single mm.  .mm will be set, .start and
216 	 *   .end will indicate the range, and .new_tlb_gen will be set
217 	 *   such that the changes between generation .new_tlb_gen-1 and
218 	 *   .new_tlb_gen are entirely contained in the indicated range.
219 	 *
220 	 * - Fully flush all mms whose tlb_gens have been updated.  .mm
221 	 *   will be NULL, .end will be TLB_FLUSH_ALL, and .new_tlb_gen
222 	 *   will be zero.
223 	 */
224 	struct mm_struct	*mm;
225 	unsigned long		start;
226 	unsigned long		end;
227 	u64			new_tlb_gen;
228 	unsigned int		initiating_cpu;
229 	u8			stride_shift;
230 	u8			freed_tables;
231 };
232 
233 void flush_tlb_local(void);
234 void flush_tlb_one_user(unsigned long addr);
235 void flush_tlb_one_kernel(unsigned long addr);
236 void flush_tlb_multi(const struct cpumask *cpumask,
237 		      const struct flush_tlb_info *info);
238 
239 #ifdef CONFIG_PARAVIRT
240 #include <asm/paravirt.h>
241 #endif
242 
243 #define flush_tlb_mm(mm)						\
244 		flush_tlb_mm_range(mm, 0UL, TLB_FLUSH_ALL, 0UL, true)
245 
246 #define flush_tlb_range(vma, start, end)				\
247 	flush_tlb_mm_range((vma)->vm_mm, start, end,			\
248 			   ((vma)->vm_flags & VM_HUGETLB)		\
249 				? huge_page_shift(hstate_vma(vma))	\
250 				: PAGE_SHIFT, false)
251 
252 extern void flush_tlb_all(void);
253 extern void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
254 				unsigned long end, unsigned int stride_shift,
255 				bool freed_tables);
256 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
257 
258 static inline void flush_tlb_page(struct vm_area_struct *vma, unsigned long a)
259 {
260 	flush_tlb_mm_range(vma->vm_mm, a, a + PAGE_SIZE, PAGE_SHIFT, false);
261 }
262 
263 static inline u64 inc_mm_tlb_gen(struct mm_struct *mm)
264 {
265 	/*
266 	 * Bump the generation count.  This also serves as a full barrier
267 	 * that synchronizes with switch_mm(): callers are required to order
268 	 * their read of mm_cpumask after their writes to the paging
269 	 * structures.
270 	 */
271 	return atomic64_inc_return(&mm->context.tlb_gen);
272 }
273 
274 static inline void arch_tlbbatch_add_mm(struct arch_tlbflush_unmap_batch *batch,
275 					struct mm_struct *mm)
276 {
277 	inc_mm_tlb_gen(mm);
278 	cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
279 }
280 
281 extern void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch);
282 
283 static inline bool pte_flags_need_flush(unsigned long oldflags,
284 					unsigned long newflags,
285 					bool ignore_access)
286 {
287 	/*
288 	 * Flags that require a flush when cleared but not when they are set.
289 	 * Only include flags that would not trigger spurious page-faults.
290 	 * Non-present entries are not cached. Hardware would set the
291 	 * dirty/access bit if needed without a fault.
292 	 */
293 	const pteval_t flush_on_clear = _PAGE_DIRTY | _PAGE_PRESENT |
294 					_PAGE_ACCESSED;
295 	const pteval_t software_flags = _PAGE_SOFTW1 | _PAGE_SOFTW2 |
296 					_PAGE_SOFTW3 | _PAGE_SOFTW4;
297 	const pteval_t flush_on_change = _PAGE_RW | _PAGE_USER | _PAGE_PWT |
298 			  _PAGE_PCD | _PAGE_PSE | _PAGE_GLOBAL | _PAGE_PAT |
299 			  _PAGE_PAT_LARGE | _PAGE_PKEY_BIT0 | _PAGE_PKEY_BIT1 |
300 			  _PAGE_PKEY_BIT2 | _PAGE_PKEY_BIT3 | _PAGE_NX;
301 	unsigned long diff = oldflags ^ newflags;
302 
303 	BUILD_BUG_ON(flush_on_clear & software_flags);
304 	BUILD_BUG_ON(flush_on_clear & flush_on_change);
305 	BUILD_BUG_ON(flush_on_change & software_flags);
306 
307 	/* Ignore software flags */
308 	diff &= ~software_flags;
309 
310 	if (ignore_access)
311 		diff &= ~_PAGE_ACCESSED;
312 
313 	/*
314 	 * Did any of the 'flush_on_clear' flags was clleared set from between
315 	 * 'oldflags' and 'newflags'?
316 	 */
317 	if (diff & oldflags & flush_on_clear)
318 		return true;
319 
320 	/* Flush on modified flags. */
321 	if (diff & flush_on_change)
322 		return true;
323 
324 	/* Ensure there are no flags that were left behind */
325 	if (IS_ENABLED(CONFIG_DEBUG_VM) &&
326 	    (diff & ~(flush_on_clear | software_flags | flush_on_change))) {
327 		VM_WARN_ON_ONCE(1);
328 		return true;
329 	}
330 
331 	return false;
332 }
333 
334 /*
335  * pte_needs_flush() checks whether permissions were demoted and require a
336  * flush. It should only be used for userspace PTEs.
337  */
338 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
339 {
340 	/* !PRESENT -> * ; no need for flush */
341 	if (!(pte_flags(oldpte) & _PAGE_PRESENT))
342 		return false;
343 
344 	/* PFN changed ; needs flush */
345 	if (pte_pfn(oldpte) != pte_pfn(newpte))
346 		return true;
347 
348 	/*
349 	 * check PTE flags; ignore access-bit; see comment in
350 	 * ptep_clear_flush_young().
351 	 */
352 	return pte_flags_need_flush(pte_flags(oldpte), pte_flags(newpte),
353 				    true);
354 }
355 #define pte_needs_flush pte_needs_flush
356 
357 /*
358  * huge_pmd_needs_flush() checks whether permissions were demoted and require a
359  * flush. It should only be used for userspace huge PMDs.
360  */
361 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
362 {
363 	/* !PRESENT -> * ; no need for flush */
364 	if (!(pmd_flags(oldpmd) & _PAGE_PRESENT))
365 		return false;
366 
367 	/* PFN changed ; needs flush */
368 	if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
369 		return true;
370 
371 	/*
372 	 * check PMD flags; do not ignore access-bit; see
373 	 * pmdp_clear_flush_young().
374 	 */
375 	return pte_flags_need_flush(pmd_flags(oldpmd), pmd_flags(newpmd),
376 				    false);
377 }
378 #define huge_pmd_needs_flush huge_pmd_needs_flush
379 
380 #ifdef CONFIG_ADDRESS_MASKING
381 static inline  u64 tlbstate_lam_cr3_mask(void)
382 {
383 	u64 lam = this_cpu_read(cpu_tlbstate.lam);
384 
385 	return lam << X86_CR3_LAM_U57_BIT;
386 }
387 
388 static inline void set_tlbstate_lam_mode(struct mm_struct *mm)
389 {
390 	this_cpu_write(cpu_tlbstate.lam,
391 		       mm->context.lam_cr3_mask >> X86_CR3_LAM_U57_BIT);
392 	this_cpu_write(tlbstate_untag_mask, mm->context.untag_mask);
393 }
394 
395 #else
396 
397 static inline u64 tlbstate_lam_cr3_mask(void)
398 {
399 	return 0;
400 }
401 
402 static inline void set_tlbstate_lam_mode(struct mm_struct *mm)
403 {
404 }
405 #endif
406 #endif /* !MODULE */
407 
408 static inline void __native_tlb_flush_global(unsigned long cr4)
409 {
410 	native_write_cr4(cr4 ^ X86_CR4_PGE);
411 	native_write_cr4(cr4);
412 }
413 #endif /* _ASM_X86_TLBFLUSH_H */
414