xref: /openbmc/linux/arch/x86/include/asm/tlbflush.h (revision da1d9caf)
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 
20 void cr4_update_irqsoff(unsigned long set, unsigned long clear);
21 unsigned long cr4_read_shadow(void);
22 
23 /* Set in this cpu's CR4. */
24 static inline void cr4_set_bits_irqsoff(unsigned long mask)
25 {
26 	cr4_update_irqsoff(mask, 0);
27 }
28 
29 /* Clear in this cpu's CR4. */
30 static inline void cr4_clear_bits_irqsoff(unsigned long mask)
31 {
32 	cr4_update_irqsoff(0, mask);
33 }
34 
35 /* Set in this cpu's CR4. */
36 static inline void cr4_set_bits(unsigned long mask)
37 {
38 	unsigned long flags;
39 
40 	local_irq_save(flags);
41 	cr4_set_bits_irqsoff(mask);
42 	local_irq_restore(flags);
43 }
44 
45 /* Clear in this cpu's CR4. */
46 static inline void cr4_clear_bits(unsigned long mask)
47 {
48 	unsigned long flags;
49 
50 	local_irq_save(flags);
51 	cr4_clear_bits_irqsoff(mask);
52 	local_irq_restore(flags);
53 }
54 
55 #ifndef MODULE
56 /*
57  * 6 because 6 should be plenty and struct tlb_state will fit in two cache
58  * lines.
59  */
60 #define TLB_NR_DYN_ASIDS	6
61 
62 struct tlb_context {
63 	u64 ctx_id;
64 	u64 tlb_gen;
65 };
66 
67 struct tlb_state {
68 	/*
69 	 * cpu_tlbstate.loaded_mm should match CR3 whenever interrupts
70 	 * are on.  This means that it may not match current->active_mm,
71 	 * which will contain the previous user mm when we're in lazy TLB
72 	 * mode even if we've already switched back to swapper_pg_dir.
73 	 *
74 	 * During switch_mm_irqs_off(), loaded_mm will be set to
75 	 * LOADED_MM_SWITCHING during the brief interrupts-off window
76 	 * when CR3 and loaded_mm would otherwise be inconsistent.  This
77 	 * is for nmi_uaccess_okay()'s benefit.
78 	 */
79 	struct mm_struct *loaded_mm;
80 
81 #define LOADED_MM_SWITCHING ((struct mm_struct *)1UL)
82 
83 	/* Last user mm for optimizing IBPB */
84 	union {
85 		struct mm_struct	*last_user_mm;
86 		unsigned long		last_user_mm_spec;
87 	};
88 
89 	u16 loaded_mm_asid;
90 	u16 next_asid;
91 
92 	/*
93 	 * If set we changed the page tables in such a way that we
94 	 * needed an invalidation of all contexts (aka. PCIDs / ASIDs).
95 	 * This tells us to go invalidate all the non-loaded ctxs[]
96 	 * on the next context switch.
97 	 *
98 	 * The current ctx was kept up-to-date as it ran and does not
99 	 * need to be invalidated.
100 	 */
101 	bool invalidate_other;
102 
103 	/*
104 	 * Mask that contains TLB_NR_DYN_ASIDS+1 bits to indicate
105 	 * the corresponding user PCID needs a flush next time we
106 	 * switch to it; see SWITCH_TO_USER_CR3.
107 	 */
108 	unsigned short user_pcid_flush_mask;
109 
110 	/*
111 	 * Access to this CR4 shadow and to H/W CR4 is protected by
112 	 * disabling interrupts when modifying either one.
113 	 */
114 	unsigned long cr4;
115 
116 	/*
117 	 * This is a list of all contexts that might exist in the TLB.
118 	 * There is one per ASID that we use, and the ASID (what the
119 	 * CPU calls PCID) is the index into ctxts.
120 	 *
121 	 * For each context, ctx_id indicates which mm the TLB's user
122 	 * entries came from.  As an invariant, the TLB will never
123 	 * contain entries that are out-of-date as when that mm reached
124 	 * the tlb_gen in the list.
125 	 *
126 	 * To be clear, this means that it's legal for the TLB code to
127 	 * flush the TLB without updating tlb_gen.  This can happen
128 	 * (for now, at least) due to paravirt remote flushes.
129 	 *
130 	 * NB: context 0 is a bit special, since it's also used by
131 	 * various bits of init code.  This is fine -- code that
132 	 * isn't aware of PCID will end up harmlessly flushing
133 	 * context 0.
134 	 */
135 	struct tlb_context ctxs[TLB_NR_DYN_ASIDS];
136 };
137 DECLARE_PER_CPU_ALIGNED(struct tlb_state, cpu_tlbstate);
138 
139 struct tlb_state_shared {
140 	/*
141 	 * We can be in one of several states:
142 	 *
143 	 *  - Actively using an mm.  Our CPU's bit will be set in
144 	 *    mm_cpumask(loaded_mm) and is_lazy == false;
145 	 *
146 	 *  - Not using a real mm.  loaded_mm == &init_mm.  Our CPU's bit
147 	 *    will not be set in mm_cpumask(&init_mm) and is_lazy == false.
148 	 *
149 	 *  - Lazily using a real mm.  loaded_mm != &init_mm, our bit
150 	 *    is set in mm_cpumask(loaded_mm), but is_lazy == true.
151 	 *    We're heuristically guessing that the CR3 load we
152 	 *    skipped more than makes up for the overhead added by
153 	 *    lazy mode.
154 	 */
155 	bool is_lazy;
156 };
157 DECLARE_PER_CPU_SHARED_ALIGNED(struct tlb_state_shared, cpu_tlbstate_shared);
158 
159 bool nmi_uaccess_okay(void);
160 #define nmi_uaccess_okay nmi_uaccess_okay
161 
162 /* Initialize cr4 shadow for this CPU. */
163 static inline void cr4_init_shadow(void)
164 {
165 	this_cpu_write(cpu_tlbstate.cr4, __read_cr4());
166 }
167 
168 extern unsigned long mmu_cr4_features;
169 extern u32 *trampoline_cr4_features;
170 
171 extern void initialize_tlbstate_and_flush(void);
172 
173 /*
174  * TLB flushing:
175  *
176  *  - flush_tlb_all() flushes all processes TLBs
177  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
178  *  - flush_tlb_page(vma, vmaddr) flushes one page
179  *  - flush_tlb_range(vma, start, end) flushes a range of pages
180  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
181  *  - flush_tlb_multi(cpumask, info) flushes TLBs on multiple cpus
182  *
183  * ..but the i386 has somewhat limited tlb flushing capabilities,
184  * and page-granular flushes are available only on i486 and up.
185  */
186 struct flush_tlb_info {
187 	/*
188 	 * We support several kinds of flushes.
189 	 *
190 	 * - Fully flush a single mm.  .mm will be set, .end will be
191 	 *   TLB_FLUSH_ALL, and .new_tlb_gen will be the tlb_gen to
192 	 *   which the IPI sender is trying to catch us up.
193 	 *
194 	 * - Partially flush a single mm.  .mm will be set, .start and
195 	 *   .end will indicate the range, and .new_tlb_gen will be set
196 	 *   such that the changes between generation .new_tlb_gen-1 and
197 	 *   .new_tlb_gen are entirely contained in the indicated range.
198 	 *
199 	 * - Fully flush all mms whose tlb_gens have been updated.  .mm
200 	 *   will be NULL, .end will be TLB_FLUSH_ALL, and .new_tlb_gen
201 	 *   will be zero.
202 	 */
203 	struct mm_struct	*mm;
204 	unsigned long		start;
205 	unsigned long		end;
206 	u64			new_tlb_gen;
207 	unsigned int		initiating_cpu;
208 	u8			stride_shift;
209 	u8			freed_tables;
210 };
211 
212 void flush_tlb_local(void);
213 void flush_tlb_one_user(unsigned long addr);
214 void flush_tlb_one_kernel(unsigned long addr);
215 void flush_tlb_multi(const struct cpumask *cpumask,
216 		      const struct flush_tlb_info *info);
217 
218 #ifdef CONFIG_PARAVIRT
219 #include <asm/paravirt.h>
220 #endif
221 
222 #define flush_tlb_mm(mm)						\
223 		flush_tlb_mm_range(mm, 0UL, TLB_FLUSH_ALL, 0UL, true)
224 
225 #define flush_tlb_range(vma, start, end)				\
226 	flush_tlb_mm_range((vma)->vm_mm, start, end,			\
227 			   ((vma)->vm_flags & VM_HUGETLB)		\
228 				? huge_page_shift(hstate_vma(vma))	\
229 				: PAGE_SHIFT, false)
230 
231 extern void flush_tlb_all(void);
232 extern void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
233 				unsigned long end, unsigned int stride_shift,
234 				bool freed_tables);
235 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
236 
237 static inline void flush_tlb_page(struct vm_area_struct *vma, unsigned long a)
238 {
239 	flush_tlb_mm_range(vma->vm_mm, a, a + PAGE_SIZE, PAGE_SHIFT, false);
240 }
241 
242 static inline u64 inc_mm_tlb_gen(struct mm_struct *mm)
243 {
244 	/*
245 	 * Bump the generation count.  This also serves as a full barrier
246 	 * that synchronizes with switch_mm(): callers are required to order
247 	 * their read of mm_cpumask after their writes to the paging
248 	 * structures.
249 	 */
250 	return atomic64_inc_return(&mm->context.tlb_gen);
251 }
252 
253 static inline void arch_tlbbatch_add_mm(struct arch_tlbflush_unmap_batch *batch,
254 					struct mm_struct *mm)
255 {
256 	inc_mm_tlb_gen(mm);
257 	cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
258 }
259 
260 extern void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch);
261 
262 static inline bool pte_flags_need_flush(unsigned long oldflags,
263 					unsigned long newflags,
264 					bool ignore_access)
265 {
266 	/*
267 	 * Flags that require a flush when cleared but not when they are set.
268 	 * Only include flags that would not trigger spurious page-faults.
269 	 * Non-present entries are not cached. Hardware would set the
270 	 * dirty/access bit if needed without a fault.
271 	 */
272 	const pteval_t flush_on_clear = _PAGE_DIRTY | _PAGE_PRESENT |
273 					_PAGE_ACCESSED;
274 	const pteval_t software_flags = _PAGE_SOFTW1 | _PAGE_SOFTW2 |
275 					_PAGE_SOFTW3 | _PAGE_SOFTW4;
276 	const pteval_t flush_on_change = _PAGE_RW | _PAGE_USER | _PAGE_PWT |
277 			  _PAGE_PCD | _PAGE_PSE | _PAGE_GLOBAL | _PAGE_PAT |
278 			  _PAGE_PAT_LARGE | _PAGE_PKEY_BIT0 | _PAGE_PKEY_BIT1 |
279 			  _PAGE_PKEY_BIT2 | _PAGE_PKEY_BIT3 | _PAGE_NX;
280 	unsigned long diff = oldflags ^ newflags;
281 
282 	BUILD_BUG_ON(flush_on_clear & software_flags);
283 	BUILD_BUG_ON(flush_on_clear & flush_on_change);
284 	BUILD_BUG_ON(flush_on_change & software_flags);
285 
286 	/* Ignore software flags */
287 	diff &= ~software_flags;
288 
289 	if (ignore_access)
290 		diff &= ~_PAGE_ACCESSED;
291 
292 	/*
293 	 * Did any of the 'flush_on_clear' flags was clleared set from between
294 	 * 'oldflags' and 'newflags'?
295 	 */
296 	if (diff & oldflags & flush_on_clear)
297 		return true;
298 
299 	/* Flush on modified flags. */
300 	if (diff & flush_on_change)
301 		return true;
302 
303 	/* Ensure there are no flags that were left behind */
304 	if (IS_ENABLED(CONFIG_DEBUG_VM) &&
305 	    (diff & ~(flush_on_clear | software_flags | flush_on_change))) {
306 		VM_WARN_ON_ONCE(1);
307 		return true;
308 	}
309 
310 	return false;
311 }
312 
313 /*
314  * pte_needs_flush() checks whether permissions were demoted and require a
315  * flush. It should only be used for userspace PTEs.
316  */
317 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
318 {
319 	/* !PRESENT -> * ; no need for flush */
320 	if (!(pte_flags(oldpte) & _PAGE_PRESENT))
321 		return false;
322 
323 	/* PFN changed ; needs flush */
324 	if (pte_pfn(oldpte) != pte_pfn(newpte))
325 		return true;
326 
327 	/*
328 	 * check PTE flags; ignore access-bit; see comment in
329 	 * ptep_clear_flush_young().
330 	 */
331 	return pte_flags_need_flush(pte_flags(oldpte), pte_flags(newpte),
332 				    true);
333 }
334 #define pte_needs_flush pte_needs_flush
335 
336 /*
337  * huge_pmd_needs_flush() checks whether permissions were demoted and require a
338  * flush. It should only be used for userspace huge PMDs.
339  */
340 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
341 {
342 	/* !PRESENT -> * ; no need for flush */
343 	if (!(pmd_flags(oldpmd) & _PAGE_PRESENT))
344 		return false;
345 
346 	/* PFN changed ; needs flush */
347 	if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
348 		return true;
349 
350 	/*
351 	 * check PMD flags; do not ignore access-bit; see
352 	 * pmdp_clear_flush_young().
353 	 */
354 	return pte_flags_need_flush(pmd_flags(oldpmd), pmd_flags(newpmd),
355 				    false);
356 }
357 #define huge_pmd_needs_flush huge_pmd_needs_flush
358 
359 #endif /* !MODULE */
360 
361 static inline void __native_tlb_flush_global(unsigned long cr4)
362 {
363 	native_write_cr4(cr4 ^ X86_CR4_PGE);
364 	native_write_cr4(cr4);
365 }
366 #endif /* _ASM_X86_TLBFLUSH_H */
367