xref: /openbmc/linux/arch/x86/include/asm/tlbflush.h (revision db66795f)
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 DECLARE_PER_CPU(u64, tlbstate_untag_mask);
18 
19 void __flush_tlb_all(void);
20 
21 #define TLB_FLUSH_ALL	-1UL
22 #define TLB_GENERATION_INVALID	0
23 
24 void cr4_update_irqsoff(unsigned long set, unsigned long clear);
25 unsigned long cr4_read_shadow(void);
26 
27 /* Set in this cpu's CR4. */
28 static inline void cr4_set_bits_irqsoff(unsigned long mask)
29 {
30 	cr4_update_irqsoff(mask, 0);
31 }
32 
33 /* Clear in this cpu's CR4. */
34 static inline void cr4_clear_bits_irqsoff(unsigned long mask)
35 {
36 	cr4_update_irqsoff(0, mask);
37 }
38 
39 /* Set in this cpu's CR4. */
40 static inline void cr4_set_bits(unsigned long mask)
41 {
42 	unsigned long flags;
43 
44 	local_irq_save(flags);
45 	cr4_set_bits_irqsoff(mask);
46 	local_irq_restore(flags);
47 }
48 
49 /* Clear in this cpu's CR4. */
50 static inline void cr4_clear_bits(unsigned long mask)
51 {
52 	unsigned long flags;
53 
54 	local_irq_save(flags);
55 	cr4_clear_bits_irqsoff(mask);
56 	local_irq_restore(flags);
57 }
58 
59 #ifndef MODULE
60 /*
61  * 6 because 6 should be plenty and struct tlb_state will fit in two cache
62  * lines.
63  */
64 #define TLB_NR_DYN_ASIDS	6
65 
66 struct tlb_context {
67 	u64 ctx_id;
68 	u64 tlb_gen;
69 };
70 
71 struct tlb_state {
72 	/*
73 	 * cpu_tlbstate.loaded_mm should match CR3 whenever interrupts
74 	 * are on.  This means that it may not match current->active_mm,
75 	 * which will contain the previous user mm when we're in lazy TLB
76 	 * mode even if we've already switched back to swapper_pg_dir.
77 	 *
78 	 * During switch_mm_irqs_off(), loaded_mm will be set to
79 	 * LOADED_MM_SWITCHING during the brief interrupts-off window
80 	 * when CR3 and loaded_mm would otherwise be inconsistent.  This
81 	 * is for nmi_uaccess_okay()'s benefit.
82 	 */
83 	struct mm_struct *loaded_mm;
84 
85 #define LOADED_MM_SWITCHING ((struct mm_struct *)1UL)
86 
87 	/* Last user mm for optimizing IBPB */
88 	union {
89 		struct mm_struct	*last_user_mm;
90 		unsigned long		last_user_mm_spec;
91 	};
92 
93 	u16 loaded_mm_asid;
94 	u16 next_asid;
95 
96 	/*
97 	 * If set we changed the page tables in such a way that we
98 	 * needed an invalidation of all contexts (aka. PCIDs / ASIDs).
99 	 * This tells us to go invalidate all the non-loaded ctxs[]
100 	 * on the next context switch.
101 	 *
102 	 * The current ctx was kept up-to-date as it ran and does not
103 	 * need to be invalidated.
104 	 */
105 	bool invalidate_other;
106 
107 #ifdef CONFIG_ADDRESS_MASKING
108 	/*
109 	 * Active LAM mode.
110 	 *
111 	 * X86_CR3_LAM_U57/U48 shifted right by X86_CR3_LAM_U57_BIT or 0 if LAM
112 	 * disabled.
113 	 */
114 	u8 lam;
115 #endif
116 
117 	/*
118 	 * Mask that contains TLB_NR_DYN_ASIDS+1 bits to indicate
119 	 * the corresponding user PCID needs a flush next time we
120 	 * switch to it; see SWITCH_TO_USER_CR3.
121 	 */
122 	unsigned short user_pcid_flush_mask;
123 
124 	/*
125 	 * Access to this CR4 shadow and to H/W CR4 is protected by
126 	 * disabling interrupts when modifying either one.
127 	 */
128 	unsigned long cr4;
129 
130 	/*
131 	 * This is a list of all contexts that might exist in the TLB.
132 	 * There is one per ASID that we use, and the ASID (what the
133 	 * CPU calls PCID) is the index into ctxts.
134 	 *
135 	 * For each context, ctx_id indicates which mm the TLB's user
136 	 * entries came from.  As an invariant, the TLB will never
137 	 * contain entries that are out-of-date as when that mm reached
138 	 * the tlb_gen in the list.
139 	 *
140 	 * To be clear, this means that it's legal for the TLB code to
141 	 * flush the TLB without updating tlb_gen.  This can happen
142 	 * (for now, at least) due to paravirt remote flushes.
143 	 *
144 	 * NB: context 0 is a bit special, since it's also used by
145 	 * various bits of init code.  This is fine -- code that
146 	 * isn't aware of PCID will end up harmlessly flushing
147 	 * context 0.
148 	 */
149 	struct tlb_context ctxs[TLB_NR_DYN_ASIDS];
150 };
151 DECLARE_PER_CPU_ALIGNED(struct tlb_state, cpu_tlbstate);
152 
153 struct tlb_state_shared {
154 	/*
155 	 * We can be in one of several states:
156 	 *
157 	 *  - Actively using an mm.  Our CPU's bit will be set in
158 	 *    mm_cpumask(loaded_mm) and is_lazy == false;
159 	 *
160 	 *  - Not using a real mm.  loaded_mm == &init_mm.  Our CPU's bit
161 	 *    will not be set in mm_cpumask(&init_mm) and is_lazy == false.
162 	 *
163 	 *  - Lazily using a real mm.  loaded_mm != &init_mm, our bit
164 	 *    is set in mm_cpumask(loaded_mm), but is_lazy == true.
165 	 *    We're heuristically guessing that the CR3 load we
166 	 *    skipped more than makes up for the overhead added by
167 	 *    lazy mode.
168 	 */
169 	bool is_lazy;
170 };
171 DECLARE_PER_CPU_SHARED_ALIGNED(struct tlb_state_shared, cpu_tlbstate_shared);
172 
173 bool nmi_uaccess_okay(void);
174 #define nmi_uaccess_okay nmi_uaccess_okay
175 
176 /* Initialize cr4 shadow for this CPU. */
177 static inline void cr4_init_shadow(void)
178 {
179 	this_cpu_write(cpu_tlbstate.cr4, __read_cr4());
180 }
181 
182 extern unsigned long mmu_cr4_features;
183 extern u32 *trampoline_cr4_features;
184 
185 extern void initialize_tlbstate_and_flush(void);
186 
187 /*
188  * TLB flushing:
189  *
190  *  - flush_tlb_all() flushes all processes TLBs
191  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
192  *  - flush_tlb_page(vma, vmaddr) flushes one page
193  *  - flush_tlb_range(vma, start, end) flushes a range of pages
194  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
195  *  - flush_tlb_multi(cpumask, info) flushes TLBs on multiple cpus
196  *
197  * ..but the i386 has somewhat limited tlb flushing capabilities,
198  * and page-granular flushes are available only on i486 and up.
199  */
200 struct flush_tlb_info {
201 	/*
202 	 * We support several kinds of flushes.
203 	 *
204 	 * - Fully flush a single mm.  .mm will be set, .end will be
205 	 *   TLB_FLUSH_ALL, and .new_tlb_gen will be the tlb_gen to
206 	 *   which the IPI sender is trying to catch us up.
207 	 *
208 	 * - Partially flush a single mm.  .mm will be set, .start and
209 	 *   .end will indicate the range, and .new_tlb_gen will be set
210 	 *   such that the changes between generation .new_tlb_gen-1 and
211 	 *   .new_tlb_gen are entirely contained in the indicated range.
212 	 *
213 	 * - Fully flush all mms whose tlb_gens have been updated.  .mm
214 	 *   will be NULL, .end will be TLB_FLUSH_ALL, and .new_tlb_gen
215 	 *   will be zero.
216 	 */
217 	struct mm_struct	*mm;
218 	unsigned long		start;
219 	unsigned long		end;
220 	u64			new_tlb_gen;
221 	unsigned int		initiating_cpu;
222 	u8			stride_shift;
223 	u8			freed_tables;
224 };
225 
226 void flush_tlb_local(void);
227 void flush_tlb_one_user(unsigned long addr);
228 void flush_tlb_one_kernel(unsigned long addr);
229 void flush_tlb_multi(const struct cpumask *cpumask,
230 		      const struct flush_tlb_info *info);
231 
232 #ifdef CONFIG_PARAVIRT
233 #include <asm/paravirt.h>
234 #endif
235 
236 #define flush_tlb_mm(mm)						\
237 		flush_tlb_mm_range(mm, 0UL, TLB_FLUSH_ALL, 0UL, true)
238 
239 #define flush_tlb_range(vma, start, end)				\
240 	flush_tlb_mm_range((vma)->vm_mm, start, end,			\
241 			   ((vma)->vm_flags & VM_HUGETLB)		\
242 				? huge_page_shift(hstate_vma(vma))	\
243 				: PAGE_SHIFT, false)
244 
245 extern void flush_tlb_all(void);
246 extern void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
247 				unsigned long end, unsigned int stride_shift,
248 				bool freed_tables);
249 extern void flush_tlb_kernel_range(unsigned long start, unsigned long end);
250 
251 static inline void flush_tlb_page(struct vm_area_struct *vma, unsigned long a)
252 {
253 	flush_tlb_mm_range(vma->vm_mm, a, a + PAGE_SIZE, PAGE_SHIFT, false);
254 }
255 
256 static inline u64 inc_mm_tlb_gen(struct mm_struct *mm)
257 {
258 	/*
259 	 * Bump the generation count.  This also serves as a full barrier
260 	 * that synchronizes with switch_mm(): callers are required to order
261 	 * their read of mm_cpumask after their writes to the paging
262 	 * structures.
263 	 */
264 	return atomic64_inc_return(&mm->context.tlb_gen);
265 }
266 
267 static inline void arch_tlbbatch_add_mm(struct arch_tlbflush_unmap_batch *batch,
268 					struct mm_struct *mm)
269 {
270 	inc_mm_tlb_gen(mm);
271 	cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
272 }
273 
274 extern void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch);
275 
276 static inline bool pte_flags_need_flush(unsigned long oldflags,
277 					unsigned long newflags,
278 					bool ignore_access)
279 {
280 	/*
281 	 * Flags that require a flush when cleared but not when they are set.
282 	 * Only include flags that would not trigger spurious page-faults.
283 	 * Non-present entries are not cached. Hardware would set the
284 	 * dirty/access bit if needed without a fault.
285 	 */
286 	const pteval_t flush_on_clear = _PAGE_DIRTY | _PAGE_PRESENT |
287 					_PAGE_ACCESSED;
288 	const pteval_t software_flags = _PAGE_SOFTW1 | _PAGE_SOFTW2 |
289 					_PAGE_SOFTW3 | _PAGE_SOFTW4;
290 	const pteval_t flush_on_change = _PAGE_RW | _PAGE_USER | _PAGE_PWT |
291 			  _PAGE_PCD | _PAGE_PSE | _PAGE_GLOBAL | _PAGE_PAT |
292 			  _PAGE_PAT_LARGE | _PAGE_PKEY_BIT0 | _PAGE_PKEY_BIT1 |
293 			  _PAGE_PKEY_BIT2 | _PAGE_PKEY_BIT3 | _PAGE_NX;
294 	unsigned long diff = oldflags ^ newflags;
295 
296 	BUILD_BUG_ON(flush_on_clear & software_flags);
297 	BUILD_BUG_ON(flush_on_clear & flush_on_change);
298 	BUILD_BUG_ON(flush_on_change & software_flags);
299 
300 	/* Ignore software flags */
301 	diff &= ~software_flags;
302 
303 	if (ignore_access)
304 		diff &= ~_PAGE_ACCESSED;
305 
306 	/*
307 	 * Did any of the 'flush_on_clear' flags was clleared set from between
308 	 * 'oldflags' and 'newflags'?
309 	 */
310 	if (diff & oldflags & flush_on_clear)
311 		return true;
312 
313 	/* Flush on modified flags. */
314 	if (diff & flush_on_change)
315 		return true;
316 
317 	/* Ensure there are no flags that were left behind */
318 	if (IS_ENABLED(CONFIG_DEBUG_VM) &&
319 	    (diff & ~(flush_on_clear | software_flags | flush_on_change))) {
320 		VM_WARN_ON_ONCE(1);
321 		return true;
322 	}
323 
324 	return false;
325 }
326 
327 /*
328  * pte_needs_flush() checks whether permissions were demoted and require a
329  * flush. It should only be used for userspace PTEs.
330  */
331 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
332 {
333 	/* !PRESENT -> * ; no need for flush */
334 	if (!(pte_flags(oldpte) & _PAGE_PRESENT))
335 		return false;
336 
337 	/* PFN changed ; needs flush */
338 	if (pte_pfn(oldpte) != pte_pfn(newpte))
339 		return true;
340 
341 	/*
342 	 * check PTE flags; ignore access-bit; see comment in
343 	 * ptep_clear_flush_young().
344 	 */
345 	return pte_flags_need_flush(pte_flags(oldpte), pte_flags(newpte),
346 				    true);
347 }
348 #define pte_needs_flush pte_needs_flush
349 
350 /*
351  * huge_pmd_needs_flush() checks whether permissions were demoted and require a
352  * flush. It should only be used for userspace huge PMDs.
353  */
354 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
355 {
356 	/* !PRESENT -> * ; no need for flush */
357 	if (!(pmd_flags(oldpmd) & _PAGE_PRESENT))
358 		return false;
359 
360 	/* PFN changed ; needs flush */
361 	if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
362 		return true;
363 
364 	/*
365 	 * check PMD flags; do not ignore access-bit; see
366 	 * pmdp_clear_flush_young().
367 	 */
368 	return pte_flags_need_flush(pmd_flags(oldpmd), pmd_flags(newpmd),
369 				    false);
370 }
371 #define huge_pmd_needs_flush huge_pmd_needs_flush
372 
373 #ifdef CONFIG_ADDRESS_MASKING
374 static inline  u64 tlbstate_lam_cr3_mask(void)
375 {
376 	u64 lam = this_cpu_read(cpu_tlbstate.lam);
377 
378 	return lam << X86_CR3_LAM_U57_BIT;
379 }
380 
381 static inline void set_tlbstate_lam_mode(struct mm_struct *mm)
382 {
383 	this_cpu_write(cpu_tlbstate.lam,
384 		       mm->context.lam_cr3_mask >> X86_CR3_LAM_U57_BIT);
385 	this_cpu_write(tlbstate_untag_mask, mm->context.untag_mask);
386 }
387 
388 #else
389 
390 static inline u64 tlbstate_lam_cr3_mask(void)
391 {
392 	return 0;
393 }
394 
395 static inline void set_tlbstate_lam_mode(struct mm_struct *mm)
396 {
397 }
398 #endif
399 #endif /* !MODULE */
400 
401 static inline void __native_tlb_flush_global(unsigned long cr4)
402 {
403 	native_write_cr4(cr4 ^ X86_CR4_PGE);
404 	native_write_cr4(cr4);
405 }
406 #endif /* _ASM_X86_TLBFLUSH_H */
407