xref: /openbmc/linux/arch/x86/include/asm/tlbflush.h (revision 12cecbf9)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_TLBFLUSH_H
3 #define _ASM_X86_TLBFLUSH_H
4 
5 #include <linux/mm.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 
16 void __flush_tlb_all(void);
17 
18 #define TLB_FLUSH_ALL	-1UL
19 #define TLB_GENERATION_INVALID	0
20 
21 void cr4_update_irqsoff(unsigned long set, unsigned long clear);
22 unsigned long cr4_read_shadow(void);
23 
24 /* Set in this cpu's CR4. */
25 static inline void cr4_set_bits_irqsoff(unsigned long mask)
26 {
27 	cr4_update_irqsoff(mask, 0);
28 }
29 
30 /* Clear in this cpu's CR4. */
31 static inline void cr4_clear_bits_irqsoff(unsigned long mask)
32 {
33 	cr4_update_irqsoff(0, mask);
34 }
35 
36 /* Set in this cpu's CR4. */
37 static inline void cr4_set_bits(unsigned long mask)
38 {
39 	unsigned long flags;
40 
41 	local_irq_save(flags);
42 	cr4_set_bits_irqsoff(mask);
43 	local_irq_restore(flags);
44 }
45 
46 /* Clear in this cpu's CR4. */
47 static inline void cr4_clear_bits(unsigned long mask)
48 {
49 	unsigned long flags;
50 
51 	local_irq_save(flags);
52 	cr4_clear_bits_irqsoff(mask);
53 	local_irq_restore(flags);
54 }
55 
56 #ifndef MODULE
57 /*
58  * 6 because 6 should be plenty and struct tlb_state will fit in two cache
59  * lines.
60  */
61 #define TLB_NR_DYN_ASIDS	6
62 
63 struct tlb_context {
64 	u64 ctx_id;
65 	u64 tlb_gen;
66 };
67 
68 struct tlb_state {
69 	/*
70 	 * cpu_tlbstate.loaded_mm should match CR3 whenever interrupts
71 	 * are on.  This means that it may not match current->active_mm,
72 	 * which will contain the previous user mm when we're in lazy TLB
73 	 * mode even if we've already switched back to swapper_pg_dir.
74 	 *
75 	 * During switch_mm_irqs_off(), loaded_mm will be set to
76 	 * LOADED_MM_SWITCHING during the brief interrupts-off window
77 	 * when CR3 and loaded_mm would otherwise be inconsistent.  This
78 	 * is for nmi_uaccess_okay()'s benefit.
79 	 */
80 	struct mm_struct *loaded_mm;
81 
82 #define LOADED_MM_SWITCHING ((struct mm_struct *)1UL)
83 
84 	/* Last user mm for optimizing IBPB */
85 	union {
86 		struct mm_struct	*last_user_mm;
87 		unsigned long		last_user_mm_spec;
88 	};
89 
90 	u16 loaded_mm_asid;
91 	u16 next_asid;
92 
93 	/*
94 	 * If set we changed the page tables in such a way that we
95 	 * needed an invalidation of all contexts (aka. PCIDs / ASIDs).
96 	 * This tells us to go invalidate all the non-loaded ctxs[]
97 	 * on the next context switch.
98 	 *
99 	 * The current ctx was kept up-to-date as it ran and does not
100 	 * need to be invalidated.
101 	 */
102 	bool invalidate_other;
103 
104 	/*
105 	 * Mask that contains TLB_NR_DYN_ASIDS+1 bits to indicate
106 	 * the corresponding user PCID needs a flush next time we
107 	 * switch to it; see SWITCH_TO_USER_CR3.
108 	 */
109 	unsigned short user_pcid_flush_mask;
110 
111 	/*
112 	 * Access to this CR4 shadow and to H/W CR4 is protected by
113 	 * disabling interrupts when modifying either one.
114 	 */
115 	unsigned long cr4;
116 
117 	/*
118 	 * This is a list of all contexts that might exist in the TLB.
119 	 * There is one per ASID that we use, and the ASID (what the
120 	 * CPU calls PCID) is the index into ctxts.
121 	 *
122 	 * For each context, ctx_id indicates which mm the TLB's user
123 	 * entries came from.  As an invariant, the TLB will never
124 	 * contain entries that are out-of-date as when that mm reached
125 	 * the tlb_gen in the list.
126 	 *
127 	 * To be clear, this means that it's legal for the TLB code to
128 	 * flush the TLB without updating tlb_gen.  This can happen
129 	 * (for now, at least) due to paravirt remote flushes.
130 	 *
131 	 * NB: context 0 is a bit special, since it's also used by
132 	 * various bits of init code.  This is fine -- code that
133 	 * isn't aware of PCID will end up harmlessly flushing
134 	 * context 0.
135 	 */
136 	struct tlb_context ctxs[TLB_NR_DYN_ASIDS];
137 };
138 DECLARE_PER_CPU_ALIGNED(struct tlb_state, cpu_tlbstate);
139 
140 struct tlb_state_shared {
141 	/*
142 	 * We can be in one of several states:
143 	 *
144 	 *  - Actively using an mm.  Our CPU's bit will be set in
145 	 *    mm_cpumask(loaded_mm) and is_lazy == false;
146 	 *
147 	 *  - Not using a real mm.  loaded_mm == &init_mm.  Our CPU's bit
148 	 *    will not be set in mm_cpumask(&init_mm) and is_lazy == false.
149 	 *
150 	 *  - Lazily using a real mm.  loaded_mm != &init_mm, our bit
151 	 *    is set in mm_cpumask(loaded_mm), but is_lazy == true.
152 	 *    We're heuristically guessing that the CR3 load we
153 	 *    skipped more than makes up for the overhead added by
154 	 *    lazy mode.
155 	 */
156 	bool is_lazy;
157 };
158 DECLARE_PER_CPU_SHARED_ALIGNED(struct tlb_state_shared, cpu_tlbstate_shared);
159 
160 bool nmi_uaccess_okay(void);
161 #define nmi_uaccess_okay nmi_uaccess_okay
162 
163 /* Initialize cr4 shadow for this CPU. */
164 static inline void cr4_init_shadow(void)
165 {
166 	this_cpu_write(cpu_tlbstate.cr4, __read_cr4());
167 }
168 
169 extern unsigned long mmu_cr4_features;
170 extern u32 *trampoline_cr4_features;
171 
172 extern void initialize_tlbstate_and_flush(void);
173 
174 /*
175  * TLB flushing:
176  *
177  *  - flush_tlb_all() flushes all processes TLBs
178  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
179  *  - flush_tlb_page(vma, vmaddr) flushes one page
180  *  - flush_tlb_range(vma, start, end) flushes a range of pages
181  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
182  *  - flush_tlb_multi(cpumask, info) flushes TLBs on multiple cpus
183  *
184  * ..but the i386 has somewhat limited tlb flushing capabilities,
185  * and page-granular flushes are available only on i486 and up.
186  */
187 struct flush_tlb_info {
188 	/*
189 	 * We support several kinds of flushes.
190 	 *
191 	 * - Fully flush a single mm.  .mm will be set, .end will be
192 	 *   TLB_FLUSH_ALL, and .new_tlb_gen will be the tlb_gen to
193 	 *   which the IPI sender is trying to catch us up.
194 	 *
195 	 * - Partially flush a single mm.  .mm will be set, .start and
196 	 *   .end will indicate the range, and .new_tlb_gen will be set
197 	 *   such that the changes between generation .new_tlb_gen-1 and
198 	 *   .new_tlb_gen are entirely contained in the indicated range.
199 	 *
200 	 * - Fully flush all mms whose tlb_gens have been updated.  .mm
201 	 *   will be NULL, .end will be TLB_FLUSH_ALL, and .new_tlb_gen
202 	 *   will be zero.
203 	 */
204 	struct mm_struct	*mm;
205 	unsigned long		start;
206 	unsigned long		end;
207 	u64			new_tlb_gen;
208 	unsigned int		initiating_cpu;
209 	u8			stride_shift;
210 	u8			freed_tables;
211 };
212 
213 void flush_tlb_local(void);
214 void flush_tlb_one_user(unsigned long addr);
215 void flush_tlb_one_kernel(unsigned long addr);
216 void flush_tlb_multi(const struct cpumask *cpumask,
217 		      const struct flush_tlb_info *info);
218 
219 #ifdef CONFIG_PARAVIRT
220 #include <asm/paravirt.h>
221 #endif
222 
223 #define flush_tlb_mm(mm)						\
224 		flush_tlb_mm_range(mm, 0UL, TLB_FLUSH_ALL, 0UL, true)
225 
226 #define flush_tlb_range(vma, start, end)				\
227 	flush_tlb_mm_range((vma)->vm_mm, start, end,			\
228 			   ((vma)->vm_flags & VM_HUGETLB)		\
229 				? huge_page_shift(hstate_vma(vma))	\
230 				: PAGE_SHIFT, false)
231 
232 extern void flush_tlb_all(void);
233 extern void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
234 				unsigned long end, unsigned int stride_shift,
235 				bool freed_tables);
236 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
237 
238 static inline void flush_tlb_page(struct vm_area_struct *vma, unsigned long a)
239 {
240 	flush_tlb_mm_range(vma->vm_mm, a, a + PAGE_SIZE, PAGE_SHIFT, false);
241 }
242 
243 static inline u64 inc_mm_tlb_gen(struct mm_struct *mm)
244 {
245 	/*
246 	 * Bump the generation count.  This also serves as a full barrier
247 	 * that synchronizes with switch_mm(): callers are required to order
248 	 * their read of mm_cpumask after their writes to the paging
249 	 * structures.
250 	 */
251 	return atomic64_inc_return(&mm->context.tlb_gen);
252 }
253 
254 static inline void arch_tlbbatch_add_mm(struct arch_tlbflush_unmap_batch *batch,
255 					struct mm_struct *mm)
256 {
257 	inc_mm_tlb_gen(mm);
258 	cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
259 }
260 
261 extern void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch);
262 
263 static inline bool pte_flags_need_flush(unsigned long oldflags,
264 					unsigned long newflags,
265 					bool ignore_access)
266 {
267 	/*
268 	 * Flags that require a flush when cleared but not when they are set.
269 	 * Only include flags that would not trigger spurious page-faults.
270 	 * Non-present entries are not cached. Hardware would set the
271 	 * dirty/access bit if needed without a fault.
272 	 */
273 	const pteval_t flush_on_clear = _PAGE_DIRTY | _PAGE_PRESENT |
274 					_PAGE_ACCESSED;
275 	const pteval_t software_flags = _PAGE_SOFTW1 | _PAGE_SOFTW2 |
276 					_PAGE_SOFTW3 | _PAGE_SOFTW4;
277 	const pteval_t flush_on_change = _PAGE_RW | _PAGE_USER | _PAGE_PWT |
278 			  _PAGE_PCD | _PAGE_PSE | _PAGE_GLOBAL | _PAGE_PAT |
279 			  _PAGE_PAT_LARGE | _PAGE_PKEY_BIT0 | _PAGE_PKEY_BIT1 |
280 			  _PAGE_PKEY_BIT2 | _PAGE_PKEY_BIT3 | _PAGE_NX;
281 	unsigned long diff = oldflags ^ newflags;
282 
283 	BUILD_BUG_ON(flush_on_clear & software_flags);
284 	BUILD_BUG_ON(flush_on_clear & flush_on_change);
285 	BUILD_BUG_ON(flush_on_change & software_flags);
286 
287 	/* Ignore software flags */
288 	diff &= ~software_flags;
289 
290 	if (ignore_access)
291 		diff &= ~_PAGE_ACCESSED;
292 
293 	/*
294 	 * Did any of the 'flush_on_clear' flags was clleared set from between
295 	 * 'oldflags' and 'newflags'?
296 	 */
297 	if (diff & oldflags & flush_on_clear)
298 		return true;
299 
300 	/* Flush on modified flags. */
301 	if (diff & flush_on_change)
302 		return true;
303 
304 	/* Ensure there are no flags that were left behind */
305 	if (IS_ENABLED(CONFIG_DEBUG_VM) &&
306 	    (diff & ~(flush_on_clear | software_flags | flush_on_change))) {
307 		VM_WARN_ON_ONCE(1);
308 		return true;
309 	}
310 
311 	return false;
312 }
313 
314 /*
315  * pte_needs_flush() checks whether permissions were demoted and require a
316  * flush. It should only be used for userspace PTEs.
317  */
318 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
319 {
320 	/* !PRESENT -> * ; no need for flush */
321 	if (!(pte_flags(oldpte) & _PAGE_PRESENT))
322 		return false;
323 
324 	/* PFN changed ; needs flush */
325 	if (pte_pfn(oldpte) != pte_pfn(newpte))
326 		return true;
327 
328 	/*
329 	 * check PTE flags; ignore access-bit; see comment in
330 	 * ptep_clear_flush_young().
331 	 */
332 	return pte_flags_need_flush(pte_flags(oldpte), pte_flags(newpte),
333 				    true);
334 }
335 #define pte_needs_flush pte_needs_flush
336 
337 /*
338  * huge_pmd_needs_flush() checks whether permissions were demoted and require a
339  * flush. It should only be used for userspace huge PMDs.
340  */
341 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
342 {
343 	/* !PRESENT -> * ; no need for flush */
344 	if (!(pmd_flags(oldpmd) & _PAGE_PRESENT))
345 		return false;
346 
347 	/* PFN changed ; needs flush */
348 	if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
349 		return true;
350 
351 	/*
352 	 * check PMD flags; do not ignore access-bit; see
353 	 * pmdp_clear_flush_young().
354 	 */
355 	return pte_flags_need_flush(pmd_flags(oldpmd), pmd_flags(newpmd),
356 				    false);
357 }
358 #define huge_pmd_needs_flush huge_pmd_needs_flush
359 
360 #endif /* !MODULE */
361 
362 static inline void __native_tlb_flush_global(unsigned long cr4)
363 {
364 	native_write_cr4(cr4 ^ X86_CR4_PGE);
365 	native_write_cr4(cr4);
366 }
367 #endif /* _ASM_X86_TLBFLUSH_H */
368