1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * Based on arch/arm/include/asm/tlbflush.h
4 *
5 * Copyright (C) 1999-2003 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 */
8 #ifndef __ASM_TLBFLUSH_H
9 #define __ASM_TLBFLUSH_H
10
11 #ifndef __ASSEMBLER__
12
13 #include <linux/bitfield.h>
14 #include <linux/mm_types.h>
15 #include <linux/sched.h>
16 #include <linux/mmu_notifier.h>
17 #include <asm/cputype.h>
18 #include <asm/mmu.h>
19
20 /*
21 * Raw TLBI operations.
22 *
23 * Where necessary, use the __tlbi() macro to avoid asm()
24 * boilerplate. Drivers and most kernel code should use the TLB
25 * management routines in preference to the macro below.
26 *
27 * The macro can be used as __tlbi(op) or __tlbi(op, arg), depending
28 * on whether a particular TLBI operation takes an argument or
29 * not. The macros handles invoking the asm with or without the
30 * register argument as appropriate.
31 */
32 #define __TLBI_0(op, arg) asm (ARM64_ASM_PREAMBLE \
33 "tlbi " #op "\n" \
34 : : )
35
36 #define __TLBI_1(op, arg) asm (ARM64_ASM_PREAMBLE \
37 "tlbi " #op ", %x0\n" \
38 : : "rZ" (arg))
39
40 #define __TLBI_N(op, arg, n, ...) __TLBI_##n(op, arg)
41
42 #define __tlbi(op, ...) __TLBI_N(op, ##__VA_ARGS__, 1, 0)
43
44 #define __tlbi_user(op, arg) do { \
45 if (arm64_kernel_unmapped_at_el0()) \
46 __tlbi(op, (arg) | USER_ASID_FLAG); \
47 } while (0)
48
49 /* This macro creates a properly formatted VA operand for the TLBI */
50 #define __TLBI_VADDR(addr, asid) \
51 ({ \
52 unsigned long __ta = (addr) >> 12; \
53 __ta &= GENMASK_ULL(43, 0); \
54 __ta |= (unsigned long)(asid) << 48; \
55 __ta; \
56 })
57
58 /*
59 * Get translation granule of the system, which is decided by
60 * PAGE_SIZE. Used by TTL.
61 * - 4KB : 1
62 * - 16KB : 2
63 * - 64KB : 3
64 */
65 #define TLBI_TTL_TG_4K 1
66 #define TLBI_TTL_TG_16K 2
67 #define TLBI_TTL_TG_64K 3
68
get_trans_granule(void)69 static inline unsigned long get_trans_granule(void)
70 {
71 switch (PAGE_SIZE) {
72 case SZ_4K:
73 return TLBI_TTL_TG_4K;
74 case SZ_16K:
75 return TLBI_TTL_TG_16K;
76 case SZ_64K:
77 return TLBI_TTL_TG_64K;
78 default:
79 return 0;
80 }
81 }
82
83 #ifdef CONFIG_ARM64_ERRATUM_4193714
84
85 extern cpumask_t sme_active_cpus;
86
87 void sme_do_dvmsync(const struct cpumask *mask);
88
sme_dvmsync(struct mm_struct * mm)89 static inline void sme_dvmsync(struct mm_struct *mm)
90 {
91 if (!alternative_has_cap_unlikely(ARM64_WORKAROUND_4193714))
92 return;
93
94 sme_do_dvmsync(mm_cpumask(mm));
95 }
96
sme_dvmsync_batch(void)97 static inline void sme_dvmsync_batch(void)
98 {
99 if (!alternative_has_cap_unlikely(ARM64_WORKAROUND_4193714))
100 return;
101
102 sme_do_dvmsync(&sme_active_cpus);
103 }
104
105 #else
106
sme_dvmsync(struct mm_struct * mm)107 static inline void sme_dvmsync(struct mm_struct *mm)
108 {
109 }
sme_dvmsync_batch(void)110 static inline void sme_dvmsync_batch(void)
111 {
112 }
113
114 #endif /* CONFIG_ARM64_ERRATUM_4193714 */
115
116 /*
117 * Level-based TLBI operations.
118 *
119 * When ARMv8.4-TTL exists, TLBI operations take an additional hint for
120 * the level at which the invalidation must take place. If the level is
121 * wrong, no invalidation may take place. In the case where the level
122 * cannot be easily determined, the value TLBI_TTL_UNKNOWN will perform
123 * a non-hinted invalidation. Any provided level outside the hint range
124 * will also cause fall-back to non-hinted invalidation.
125 *
126 * For Stage-2 invalidation, use the level values provided to that effect
127 * in asm/stage2_pgtable.h.
128 */
129 #define TLBI_TTL_MASK GENMASK_ULL(47, 44)
130
131 #define TLBI_TTL_UNKNOWN INT_MAX
132
133 typedef void (*tlbi_op)(u64 arg);
134
vae1is(u64 arg)135 static __always_inline void vae1is(u64 arg)
136 {
137 __tlbi(vae1is, arg);
138 __tlbi_user(vae1is, arg);
139 }
140
vae2is(u64 arg)141 static __always_inline void vae2is(u64 arg)
142 {
143 __tlbi(vae2is, arg);
144 }
145
vale1(u64 arg)146 static __always_inline void vale1(u64 arg)
147 {
148 __tlbi(vale1, arg);
149 __tlbi_user(vale1, arg);
150 }
151
vale1is(u64 arg)152 static __always_inline void vale1is(u64 arg)
153 {
154 __tlbi(vale1is, arg);
155 __tlbi_user(vale1is, arg);
156 }
157
vale2is(u64 arg)158 static __always_inline void vale2is(u64 arg)
159 {
160 __tlbi(vale2is, arg);
161 }
162
vaale1is(u64 arg)163 static __always_inline void vaale1is(u64 arg)
164 {
165 __tlbi(vaale1is, arg);
166 }
167
ipas2e1(u64 arg)168 static __always_inline void ipas2e1(u64 arg)
169 {
170 __tlbi(ipas2e1, arg);
171 }
172
ipas2e1is(u64 arg)173 static __always_inline void ipas2e1is(u64 arg)
174 {
175 __tlbi(ipas2e1is, arg);
176 }
177
__tlbi_level_asid(tlbi_op op,u64 addr,u32 level,u16 asid)178 static __always_inline void __tlbi_level_asid(tlbi_op op, u64 addr, u32 level,
179 u16 asid)
180 {
181 u64 arg = __TLBI_VADDR(addr, asid);
182
183 if (alternative_has_cap_unlikely(ARM64_HAS_ARMv8_4_TTL) && level <= 3) {
184 u64 ttl = level | (get_trans_granule() << 2);
185
186 FIELD_MODIFY(TLBI_TTL_MASK, &arg, ttl);
187 }
188
189 op(arg);
190 }
191
__tlbi_level(tlbi_op op,u64 addr,u32 level)192 static inline void __tlbi_level(tlbi_op op, u64 addr, u32 level)
193 {
194 __tlbi_level_asid(op, addr, level, 0);
195 }
196
197 /*
198 * This macro creates a properly formatted VA operand for the TLB RANGE. The
199 * value bit assignments are:
200 *
201 * +----------+------+-------+-------+-------+----------------------+
202 * | ASID | TG | SCALE | NUM | TTL | BADDR |
203 * +-----------------+-------+-------+-------+----------------------+
204 * |63 48|47 46|45 44|43 39|38 37|36 0|
205 *
206 * The address range is determined by below formula: [BADDR, BADDR + (NUM + 1) *
207 * 2^(5*SCALE + 1) * PAGESIZE)
208 *
209 * Note that the first argument, baddr, is pre-shifted; If LPA2 is in use, BADDR
210 * holds addr[52:16]. Else BADDR holds page number. See for example ARM DDI
211 * 0487J.a section C5.5.60 "TLBI VAE1IS, TLBI VAE1ISNXS, TLB Invalidate by VA,
212 * EL1, Inner Shareable".
213 *
214 */
215 #define TLBIR_ASID_MASK GENMASK_ULL(63, 48)
216 #define TLBIR_TG_MASK GENMASK_ULL(47, 46)
217 #define TLBIR_SCALE_MASK GENMASK_ULL(45, 44)
218 #define TLBIR_NUM_MASK GENMASK_ULL(43, 39)
219 #define TLBIR_TTL_MASK GENMASK_ULL(38, 37)
220 #define TLBIR_BADDR_MASK GENMASK_ULL(36, 0)
221
222 /* These macros are used by the TLBI RANGE feature. */
223 #define __TLBI_RANGE_PAGES(num, scale) \
224 ((unsigned long)((num) + 1) << (5 * (scale) + 1))
225 #define MAX_TLBI_RANGE_PAGES __TLBI_RANGE_PAGES(31, 3)
226
227 /*
228 * Generate 'num' values from -1 to 31 with -1 rejected by the
229 * __flush_tlb_range() loop below. Its return value is only
230 * significant for a maximum of MAX_TLBI_RANGE_PAGES pages. If
231 * 'pages' is more than that, you must iterate over the overall
232 * range.
233 */
234 #define __TLBI_RANGE_NUM(pages, scale) \
235 (((pages) >> (5 * (scale) + 1)) - 1)
236
237 #define __repeat_tlbi_sync(op, arg...) \
238 do { \
239 if (!alternative_has_cap_unlikely(ARM64_WORKAROUND_REPEAT_TLBI_SYNC)) \
240 break; \
241 __tlbi(op, ##arg); \
242 dsb(ish); \
243 } while (0)
244
245 /*
246 * Complete broadcast TLB maintenance issued by the host which invalidates
247 * stage 1 information in the host's own translation regime.
248 */
__tlbi_sync_s1ish(struct mm_struct * mm)249 static inline void __tlbi_sync_s1ish(struct mm_struct *mm)
250 {
251 dsb(ish);
252 __repeat_tlbi_sync(vale1is, 0);
253 sme_dvmsync(mm);
254 }
255
__tlbi_sync_s1ish_batch(void)256 static inline void __tlbi_sync_s1ish_batch(void)
257 {
258 dsb(ish);
259 __repeat_tlbi_sync(vale1is, 0);
260 sme_dvmsync_batch();
261 }
262
__tlbi_sync_s1ish_kernel(void)263 static inline void __tlbi_sync_s1ish_kernel(void)
264 {
265 dsb(ish);
266 __repeat_tlbi_sync(vale1is, 0);
267 }
268
269 /*
270 * Complete broadcast TLB maintenance issued by hyp code which invalidates
271 * stage 1 translation information in any translation regime.
272 */
__tlbi_sync_s1ish_hyp(void)273 static inline void __tlbi_sync_s1ish_hyp(void)
274 {
275 dsb(ish);
276 __repeat_tlbi_sync(vale2is, 0);
277 }
278
279 /*
280 * TLB Invalidation
281 * ================
282 *
283 * This header file implements the low-level TLB invalidation routines
284 * (sometimes referred to as "flushing" in the kernel) for arm64.
285 *
286 * Every invalidation operation uses the following template:
287 *
288 * DSB ISHST // Ensure prior page-table updates have completed
289 * TLBI ... // Invalidate the TLB
290 * DSB ISH // Ensure the TLB invalidation has completed
291 * if (invalidated kernel mappings)
292 * ISB // Discard any instructions fetched from the old mapping
293 *
294 *
295 * The following functions form part of the "core" TLB invalidation API,
296 * as documented in Documentation/core-api/cachetlb.rst:
297 *
298 * flush_tlb_all()
299 * Invalidate the entire TLB (kernel + user) on all CPUs
300 *
301 * flush_tlb_mm(mm)
302 * Invalidate an entire user address space on all CPUs.
303 * The 'mm' argument identifies the ASID to invalidate.
304 *
305 * flush_tlb_range(vma, start, end)
306 * Invalidate the virtual-address range '[start, end)' on all
307 * CPUs for the user address space corresponding to 'vma->mm'.
308 * Note that this operation also invalidates any walk-cache
309 * entries associated with translations for the specified address
310 * range.
311 *
312 * flush_tlb_kernel_range(start, end)
313 * Same as flush_tlb_range(..., start, end), but applies to
314 * kernel mappings rather than a particular user address space.
315 * Whilst not explicitly documented, this function is used when
316 * unmapping pages from vmalloc/io space.
317 *
318 * flush_tlb_page(vma, addr)
319 * Equivalent to __flush_tlb_page(..., flags=TLBF_NONE)
320 *
321 *
322 * Next, we have some undocumented invalidation routines that you probably
323 * don't want to call unless you know what you're doing:
324 *
325 * local_flush_tlb_all()
326 * Same as flush_tlb_all(), but only applies to the calling CPU.
327 *
328 * __flush_tlb_kernel_pgtable(addr)
329 * Invalidate a single kernel mapping for address 'addr' on all
330 * CPUs, ensuring that any walk-cache entries associated with the
331 * translation are also invalidated.
332 *
333 * __flush_tlb_range(vma, start, end, stride, tlb_level, flags)
334 * Invalidate the virtual-address range '[start, end)' on all
335 * CPUs for the user address space corresponding to 'vma->mm'.
336 * The invalidation operations are issued at a granularity
337 * determined by 'stride'. tlb_level is the level at
338 * which the invalidation must take place. If the level is wrong,
339 * no invalidation may take place. In the case where the level
340 * cannot be easily determined, the value TLBI_TTL_UNKNOWN will
341 * perform a non-hinted invalidation. flags may be TLBF_NONE (0) or
342 * any combination of TLBF_NOWALKCACHE (elide eviction of walk
343 * cache entries), TLBF_NONOTIFY (don't call mmu notifiers),
344 * TLBF_NOSYNC (don't issue trailing dsb) and TLBF_NOBROADCAST
345 * (only perform the invalidation for the local cpu).
346 *
347 * __flush_tlb_page(vma, addr, flags)
348 * Invalidate a single user mapping for address 'addr' in the
349 * address space corresponding to 'vma->mm'. Note that this
350 * operation only invalidates a single level 3 page-table entry
351 * and therefore does not affect any walk-caches. flags may contain
352 * any combination of TLBF_NONOTIFY (don't call mmu notifiers),
353 * TLBF_NOSYNC (don't issue trailing dsb) and TLBF_NOBROADCAST
354 * (only perform the invalidation for the local cpu).
355 *
356 * Finally, take a look at asm/tlb.h to see how tlb_flush() is implemented
357 * on top of these routines, since that is our interface to the mmu_gather
358 * API as used by munmap() and friends.
359 */
local_flush_tlb_all(void)360 static inline void local_flush_tlb_all(void)
361 {
362 dsb(nshst);
363 __tlbi(vmalle1);
364 dsb(nsh);
365 isb();
366 }
367
flush_tlb_all(void)368 static inline void flush_tlb_all(void)
369 {
370 dsb(ishst);
371 __tlbi(vmalle1is);
372 __tlbi_sync_s1ish_kernel();
373 isb();
374 }
375
flush_tlb_mm(struct mm_struct * mm)376 static inline void flush_tlb_mm(struct mm_struct *mm)
377 {
378 unsigned long asid;
379
380 dsb(ishst);
381 asid = __TLBI_VADDR(0, ASID(mm));
382 __tlbi(aside1is, asid);
383 __tlbi_user(aside1is, asid);
384 __tlbi_sync_s1ish(mm);
385 mmu_notifier_arch_invalidate_secondary_tlbs(mm, 0, -1UL);
386 }
387
arch_tlbbatch_should_defer(struct mm_struct * mm)388 static inline bool arch_tlbbatch_should_defer(struct mm_struct *mm)
389 {
390 return true;
391 }
392
393 /*
394 * To support TLB batched flush for multiple pages unmapping, we only send
395 * the TLBI for each page in arch_tlbbatch_add_pending() and wait for the
396 * completion at the end in arch_tlbbatch_flush(). Since we've already issued
397 * TLBI for each page so only a DSB is needed to synchronise its effect on the
398 * other CPUs.
399 *
400 * This will save the time waiting on DSB comparing issuing a TLBI;DSB sequence
401 * for each page.
402 */
arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch * batch)403 static inline void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)
404 {
405 __tlbi_sync_s1ish_batch();
406 }
407
408 /*
409 * This is meant to avoid soft lock-ups on large TLB flushing ranges and not
410 * necessarily a performance improvement.
411 */
412 #define MAX_DVM_OPS PTRS_PER_PTE
413
414 /*
415 * __flush_tlb_range_op - Perform TLBI operation upon a range
416 *
417 * @lop: TLBI level operation to perform
418 * @rop: TLBI range operation to perform
419 * @start: The start address of the range
420 * @pages: Range as the number of pages from 'start'
421 * @stride: Flush granularity
422 * @asid: The ASID of the task (0 for IPA instructions)
423 * @level: Translation Table level hint, if known
424 * @lpa2: If 'true', the lpa2 scheme is used as set out below
425 *
426 * When the CPU does not support TLB range operations, flush the TLB
427 * entries one by one at the granularity of 'stride'. If the TLB
428 * range ops are supported, then:
429 *
430 * 1. If FEAT_LPA2 is in use, the start address of a range operation must be
431 * 64KB aligned, so flush pages one by one until the alignment is reached
432 * using the non-range operations. This step is skipped if LPA2 is not in
433 * use.
434 *
435 * 2. The minimum range granularity is decided by 'scale', so multiple range
436 * TLBI operations may be required. Start from scale = 3, flush the largest
437 * possible number of pages ((num+1)*2^(5*scale+1)) that fit into the
438 * requested range, then decrement scale and continue until one or zero pages
439 * are left. We must start from highest scale to ensure 64KB start alignment
440 * is maintained in the LPA2 case.
441 *
442 * 3. If there is 1 page remaining, flush it through non-range operations. Range
443 * operations can only span an even number of pages. We save this for last to
444 * ensure 64KB start alignment is maintained for the LPA2 case.
445 */
rvae1is(u64 arg)446 static __always_inline void rvae1is(u64 arg)
447 {
448 __tlbi(rvae1is, arg);
449 __tlbi_user(rvae1is, arg);
450 }
451
rvale1(u64 arg)452 static __always_inline void rvale1(u64 arg)
453 {
454 __tlbi(rvale1, arg);
455 __tlbi_user(rvale1, arg);
456 }
457
rvale1is(u64 arg)458 static __always_inline void rvale1is(u64 arg)
459 {
460 __tlbi(rvale1is, arg);
461 __tlbi_user(rvale1is, arg);
462 }
463
rvaale1is(u64 arg)464 static __always_inline void rvaale1is(u64 arg)
465 {
466 __tlbi(rvaale1is, arg);
467 }
468
ripas2e1is(u64 arg)469 static __always_inline void ripas2e1is(u64 arg)
470 {
471 __tlbi(ripas2e1is, arg);
472 }
473
__tlbi_range(tlbi_op op,u64 addr,u16 asid,int scale,int num,u32 level,bool lpa2)474 static __always_inline void __tlbi_range(tlbi_op op, u64 addr,
475 u16 asid, int scale, int num,
476 u32 level, bool lpa2)
477 {
478 u64 arg = 0;
479
480 arg |= FIELD_PREP(TLBIR_BADDR_MASK, addr >> (lpa2 ? 16 : PAGE_SHIFT));
481 arg |= FIELD_PREP(TLBIR_TTL_MASK, level > 3 ? 0 : level);
482 arg |= FIELD_PREP(TLBIR_NUM_MASK, num);
483 arg |= FIELD_PREP(TLBIR_SCALE_MASK, scale);
484 arg |= FIELD_PREP(TLBIR_TG_MASK, get_trans_granule());
485 arg |= FIELD_PREP(TLBIR_ASID_MASK, asid);
486
487 op(arg);
488 }
489
__flush_tlb_range_op(tlbi_op lop,tlbi_op rop,u64 start,size_t pages,u64 stride,u16 asid,u32 level,bool lpa2)490 static __always_inline void __flush_tlb_range_op(tlbi_op lop, tlbi_op rop,
491 u64 start, size_t pages,
492 u64 stride, u16 asid,
493 u32 level, bool lpa2)
494 {
495 u64 addr = start, end = start + pages * PAGE_SIZE;
496 int scale = 3;
497
498 while (addr != end) {
499 int num;
500
501 pages = (end - addr) >> PAGE_SHIFT;
502
503 if (!system_supports_tlb_range() || pages == 1)
504 goto invalidate_one;
505
506 if (lpa2 && !IS_ALIGNED(addr, SZ_64K))
507 goto invalidate_one;
508
509 num = __TLBI_RANGE_NUM(pages, scale);
510 if (num >= 0) {
511 __tlbi_range(rop, addr, asid, scale, num, level, lpa2);
512 addr += __TLBI_RANGE_PAGES(num, scale) << PAGE_SHIFT;
513 }
514
515 scale--;
516 continue;
517 invalidate_one:
518 __tlbi_level_asid(lop, addr, level, asid);
519 addr += stride;
520 }
521 }
522
523 #define __flush_s1_tlb_range_op(op, start, pages, stride, asid, tlb_level) \
524 __flush_tlb_range_op(op, r##op, start, pages, stride, asid, tlb_level, lpa2_is_enabled())
525
526 #define __flush_s2_tlb_range_op(op, start, pages, stride, tlb_level) \
527 __flush_tlb_range_op(op, r##op, start, pages, stride, 0, tlb_level, kvm_lpa2_is_enabled())
528
__flush_tlb_range_limit_excess(unsigned long pages,unsigned long stride)529 static inline bool __flush_tlb_range_limit_excess(unsigned long pages,
530 unsigned long stride)
531 {
532 /*
533 * Assume that the worst case number of DVM ops required to flush a
534 * given range on a system that supports tlb-range is 20 (4 scales, 1
535 * final page, 15 for alignment on LPA2 systems), which is much smaller
536 * than MAX_DVM_OPS.
537 */
538 if (system_supports_tlb_range())
539 return pages > MAX_TLBI_RANGE_PAGES;
540
541 return pages >= (MAX_DVM_OPS * stride) >> PAGE_SHIFT;
542 }
543
544 typedef unsigned __bitwise tlbf_t;
545
546 /* No special behaviour. */
547 #define TLBF_NONE ((__force tlbf_t)0)
548
549 /* Invalidate tlb entries only, leaving the page table walk cache intact. */
550 #define TLBF_NOWALKCACHE ((__force tlbf_t)BIT(0))
551
552 /* Skip the trailing dsb after issuing tlbi. */
553 #define TLBF_NOSYNC ((__force tlbf_t)BIT(1))
554
555 /* Suppress tlb notifier callbacks for this flush operation. */
556 #define TLBF_NONOTIFY ((__force tlbf_t)BIT(2))
557
558 /* Perform the tlbi locally without broadcasting to other CPUs. */
559 #define TLBF_NOBROADCAST ((__force tlbf_t)BIT(3))
560
__do_flush_tlb_range(struct vm_area_struct * vma,unsigned long start,unsigned long end,unsigned long stride,int tlb_level,tlbf_t flags)561 static __always_inline void __do_flush_tlb_range(struct vm_area_struct *vma,
562 unsigned long start, unsigned long end,
563 unsigned long stride, int tlb_level,
564 tlbf_t flags)
565 {
566 struct mm_struct *mm = vma->vm_mm;
567 unsigned long asid, pages;
568
569 pages = (end - start) >> PAGE_SHIFT;
570
571 if (__flush_tlb_range_limit_excess(pages, stride)) {
572 flush_tlb_mm(mm);
573 return;
574 }
575
576 if (!(flags & TLBF_NOBROADCAST))
577 dsb(ishst);
578 else
579 dsb(nshst);
580
581 asid = ASID(mm);
582
583 switch (flags & (TLBF_NOWALKCACHE | TLBF_NOBROADCAST)) {
584 case TLBF_NONE:
585 __flush_s1_tlb_range_op(vae1is, start, pages, stride,
586 asid, tlb_level);
587 break;
588 case TLBF_NOWALKCACHE:
589 __flush_s1_tlb_range_op(vale1is, start, pages, stride,
590 asid, tlb_level);
591 break;
592 case TLBF_NOBROADCAST:
593 /* Combination unused */
594 BUG();
595 break;
596 case TLBF_NOWALKCACHE | TLBF_NOBROADCAST:
597 __flush_s1_tlb_range_op(vale1, start, pages, stride,
598 asid, tlb_level);
599 break;
600 }
601
602 if (!(flags & TLBF_NONOTIFY))
603 mmu_notifier_arch_invalidate_secondary_tlbs(mm, start, end);
604
605 if (!(flags & TLBF_NOSYNC)) {
606 if (!(flags & TLBF_NOBROADCAST))
607 __tlbi_sync_s1ish(mm);
608 else
609 dsb(nsh);
610 }
611 }
612
__flush_tlb_range(struct vm_area_struct * vma,unsigned long start,unsigned long end,unsigned long stride,int tlb_level,tlbf_t flags)613 static inline void __flush_tlb_range(struct vm_area_struct *vma,
614 unsigned long start, unsigned long end,
615 unsigned long stride, int tlb_level,
616 tlbf_t flags)
617 {
618 start = round_down(start, stride);
619 end = round_up(end, stride);
620 __do_flush_tlb_range(vma, start, end, stride, tlb_level, flags);
621 }
622
flush_tlb_range(struct vm_area_struct * vma,unsigned long start,unsigned long end)623 static inline void flush_tlb_range(struct vm_area_struct *vma,
624 unsigned long start, unsigned long end)
625 {
626 /*
627 * We cannot use leaf-only invalidation here, since we may be invalidating
628 * table entries as part of collapsing hugepages or moving page tables.
629 * Set the tlb_level to TLBI_TTL_UNKNOWN because we can not get enough
630 * information here.
631 */
632 __flush_tlb_range(vma, start, end, PAGE_SIZE, TLBI_TTL_UNKNOWN, TLBF_NONE);
633 }
634
__flush_tlb_page(struct vm_area_struct * vma,unsigned long uaddr,tlbf_t flags)635 static inline void __flush_tlb_page(struct vm_area_struct *vma,
636 unsigned long uaddr, tlbf_t flags)
637 {
638 unsigned long start = round_down(uaddr, PAGE_SIZE);
639 unsigned long end = start + PAGE_SIZE;
640
641 __do_flush_tlb_range(vma, start, end, PAGE_SIZE, 3,
642 TLBF_NOWALKCACHE | flags);
643 }
644
flush_tlb_page(struct vm_area_struct * vma,unsigned long uaddr)645 static inline void flush_tlb_page(struct vm_area_struct *vma,
646 unsigned long uaddr)
647 {
648 __flush_tlb_page(vma, uaddr, TLBF_NONE);
649 }
650
flush_tlb_kernel_range(unsigned long start,unsigned long end)651 static inline void flush_tlb_kernel_range(unsigned long start, unsigned long end)
652 {
653 const unsigned long stride = PAGE_SIZE;
654 unsigned long pages;
655
656 start = round_down(start, stride);
657 end = round_up(end, stride);
658 pages = (end - start) >> PAGE_SHIFT;
659
660 if (__flush_tlb_range_limit_excess(pages, stride)) {
661 flush_tlb_all();
662 return;
663 }
664
665 dsb(ishst);
666 __flush_s1_tlb_range_op(vaale1is, start, pages, stride, 0,
667 TLBI_TTL_UNKNOWN);
668 __tlbi_sync_s1ish_kernel();
669 isb();
670 }
671
672 /*
673 * Used to invalidate the TLB (walk caches) corresponding to intermediate page
674 * table levels (pgd/pud/pmd).
675 */
__flush_tlb_kernel_pgtable(unsigned long kaddr)676 static inline void __flush_tlb_kernel_pgtable(unsigned long kaddr)
677 {
678 unsigned long addr = __TLBI_VADDR(kaddr, 0);
679
680 dsb(ishst);
681 __tlbi(vaae1is, addr);
682 __tlbi_sync_s1ish_kernel();
683 isb();
684 }
685
arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch * batch,struct mm_struct * mm,unsigned long start,unsigned long end)686 static inline void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch,
687 struct mm_struct *mm, unsigned long start, unsigned long end)
688 {
689 struct vm_area_struct vma = { .vm_mm = mm, .vm_flags = 0 };
690
691 __flush_tlb_range(&vma, start, end, PAGE_SIZE, 3,
692 TLBF_NOWALKCACHE | TLBF_NOSYNC);
693 }
694
__pte_flags_need_flush(ptval_t oldval,ptval_t newval)695 static inline bool __pte_flags_need_flush(ptval_t oldval, ptval_t newval)
696 {
697 ptval_t diff = oldval ^ newval;
698
699 /* invalid to valid transition requires no flush */
700 if (!(oldval & PTE_VALID))
701 return false;
702
703 /* Transition in the SW bits requires no flush */
704 diff &= ~PTE_SWBITS_MASK;
705
706 return diff;
707 }
708
pte_needs_flush(pte_t oldpte,pte_t newpte)709 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
710 {
711 return __pte_flags_need_flush(pte_val(oldpte), pte_val(newpte));
712 }
713 #define pte_needs_flush pte_needs_flush
714
huge_pmd_needs_flush(pmd_t oldpmd,pmd_t newpmd)715 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
716 {
717 return __pte_flags_need_flush(pmd_val(oldpmd), pmd_val(newpmd));
718 }
719 #define huge_pmd_needs_flush huge_pmd_needs_flush
720
721 #undef __tlbi_user
722 #undef __TLBI_VADDR
723 #endif
724
725 #endif
726