1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * arch/arm/include/asm/cacheflush.h
4 *
5 * Copyright (C) 1999-2002 Russell King
6 */
7 #ifndef _ASMARM_CACHEFLUSH_H
8 #define _ASMARM_CACHEFLUSH_H
9
10 #include <linux/mm.h>
11
12 #include <asm/glue-cache.h>
13 #include <asm/shmparam.h>
14 #include <asm/cachetype.h>
15 #include <asm/outercache.h>
16
17 #define CACHE_COLOUR(vaddr) ((vaddr & (SHMLBA - 1)) >> PAGE_SHIFT)
18
19 /*
20 * This flag is used to indicate that the page pointed to by a pte is clean
21 * and does not require cleaning before returning it to the user.
22 */
23 #define PG_dcache_clean PG_arch_1
24
25 /*
26 * MM Cache Management
27 * ===================
28 *
29 * The arch/arm/mm/cache-*.S and arch/arm/mm/proc-*.S files
30 * implement these methods.
31 *
32 * Start addresses are inclusive and end addresses are exclusive;
33 * start addresses should be rounded down, end addresses up.
34 *
35 * See Documentation/core-api/cachetlb.rst for more information.
36 * Please note that the implementation of these, and the required
37 * effects are cache-type (VIVT/VIPT/PIPT) specific.
38 *
39 * flush_icache_all()
40 *
41 * Unconditionally clean and invalidate the entire icache.
42 * Currently only needed for cache-v6.S and cache-v7.S, see
43 * __flush_icache_all for the generic implementation.
44 *
45 * flush_kern_all()
46 *
47 * Unconditionally clean and invalidate the entire cache.
48 *
49 * flush_kern_louis()
50 *
51 * Flush data cache levels up to the level of unification
52 * inner shareable and invalidate the I-cache.
53 * Only needed from v7 onwards, falls back to flush_cache_all()
54 * for all other processor versions.
55 *
56 * flush_user_all()
57 *
58 * Clean and invalidate all user space cache entries
59 * before a change of page tables.
60 *
61 * flush_user_range(start, end, flags)
62 *
63 * Clean and invalidate a range of cache entries in the
64 * specified address space before a change of page tables.
65 * - start - user start address (inclusive, page aligned)
66 * - end - user end address (exclusive, page aligned)
67 * - flags - vma->vm_flags field
68 *
69 * coherent_kern_range(start, end)
70 *
71 * Ensure coherency between the Icache and the Dcache in the
72 * region described by start, end. If you have non-snooping
73 * Harvard caches, you need to implement this function.
74 * - start - virtual start address
75 * - end - virtual end address
76 *
77 * coherent_user_range(start, end)
78 *
79 * Ensure coherency between the Icache and the Dcache in the
80 * region described by start, end. If you have non-snooping
81 * Harvard caches, you need to implement this function.
82 * - start - virtual start address
83 * - end - virtual end address
84 *
85 * flush_kern_dcache_area(kaddr, size)
86 *
87 * Ensure that the data held in page is written back.
88 * - kaddr - page address
89 * - size - region size
90 *
91 * DMA Cache Coherency
92 * ===================
93 *
94 * dma_flush_range(start, end)
95 *
96 * Clean and invalidate the specified virtual address range.
97 * - start - virtual start address
98 * - end - virtual end address
99 */
100
101 struct cpu_cache_fns {
102 void (*flush_icache_all)(void);
103 void (*flush_kern_all)(void);
104 void (*flush_kern_louis)(void);
105 void (*flush_user_all)(void);
106 void (*flush_user_range)(unsigned long, unsigned long, unsigned int);
107
108 void (*coherent_kern_range)(unsigned long, unsigned long);
109 int (*coherent_user_range)(unsigned long, unsigned long);
110 void (*flush_kern_dcache_area)(void *, size_t);
111
112 void (*dma_map_area)(const void *, size_t, int);
113 void (*dma_unmap_area)(const void *, size_t, int);
114
115 void (*dma_flush_range)(const void *, const void *);
116 } __no_randomize_layout;
117
118 /*
119 * Select the calling method
120 */
121 #ifdef MULTI_CACHE
122
123 extern struct cpu_cache_fns cpu_cache;
124
125 #define __cpuc_flush_icache_all cpu_cache.flush_icache_all
126 #define __cpuc_flush_kern_all cpu_cache.flush_kern_all
127 #define __cpuc_flush_kern_louis cpu_cache.flush_kern_louis
128 #define __cpuc_flush_user_all cpu_cache.flush_user_all
129 #define __cpuc_flush_user_range cpu_cache.flush_user_range
130 #define __cpuc_coherent_kern_range cpu_cache.coherent_kern_range
131 #define __cpuc_coherent_user_range cpu_cache.coherent_user_range
132 #define __cpuc_flush_dcache_area cpu_cache.flush_kern_dcache_area
133
134 /*
135 * These are private to the dma-mapping API. Do not use directly.
136 * Their sole purpose is to ensure that data held in the cache
137 * is visible to DMA, or data written by DMA to system memory is
138 * visible to the CPU.
139 */
140 #define dmac_flush_range cpu_cache.dma_flush_range
141
142 #else
143
144 extern void __cpuc_flush_icache_all(void);
145 extern void __cpuc_flush_kern_all(void);
146 extern void __cpuc_flush_kern_louis(void);
147 extern void __cpuc_flush_user_all(void);
148 extern void __cpuc_flush_user_range(unsigned long, unsigned long, unsigned int);
149 extern void __cpuc_coherent_kern_range(unsigned long, unsigned long);
150 extern int __cpuc_coherent_user_range(unsigned long, unsigned long);
151 extern void __cpuc_flush_dcache_area(void *, size_t);
152
153 /*
154 * These are private to the dma-mapping API. Do not use directly.
155 * Their sole purpose is to ensure that data held in the cache
156 * is visible to DMA, or data written by DMA to system memory is
157 * visible to the CPU.
158 */
159 extern void dmac_flush_range(const void *, const void *);
160
161 #endif
162
163 /*
164 * Copy user data from/to a page which is mapped into a different
165 * processes address space. Really, we want to allow our "user
166 * space" model to handle this.
167 */
168 extern void copy_to_user_page(struct vm_area_struct *, struct page *,
169 unsigned long, void *, const void *, unsigned long);
170 #define copy_from_user_page(vma, page, vaddr, dst, src, len) \
171 do { \
172 memcpy(dst, src, len); \
173 } while (0)
174
175 /*
176 * Convert calls to our calling convention.
177 */
178
179 /* Invalidate I-cache */
180 #define __flush_icache_all_generic() \
181 asm("mcr p15, 0, %0, c7, c5, 0" \
182 : : "r" (0));
183
184 /* Invalidate I-cache inner shareable */
185 #define __flush_icache_all_v7_smp() \
186 asm("mcr p15, 0, %0, c7, c1, 0" \
187 : : "r" (0));
188
189 /*
190 * Optimized __flush_icache_all for the common cases. Note that UP ARMv7
191 * will fall through to use __flush_icache_all_generic.
192 */
193 #if (defined(CONFIG_CPU_V7) && \
194 (defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K))) || \
195 defined(CONFIG_SMP_ON_UP)
196 #define __flush_icache_preferred __cpuc_flush_icache_all
197 #elif __LINUX_ARM_ARCH__ >= 7 && defined(CONFIG_SMP)
198 #define __flush_icache_preferred __flush_icache_all_v7_smp
199 #elif __LINUX_ARM_ARCH__ == 6 && defined(CONFIG_ARM_ERRATA_411920)
200 #define __flush_icache_preferred __cpuc_flush_icache_all
201 #else
202 #define __flush_icache_preferred __flush_icache_all_generic
203 #endif
204
__flush_icache_all(void)205 static inline void __flush_icache_all(void)
206 {
207 __flush_icache_preferred();
208 dsb(ishst);
209 }
210
211 /*
212 * Flush caches up to Level of Unification Inner Shareable
213 */
214 #define flush_cache_louis() __cpuc_flush_kern_louis()
215
216 #define flush_cache_all() __cpuc_flush_kern_all()
217
vivt_flush_cache_mm(struct mm_struct * mm)218 static inline void vivt_flush_cache_mm(struct mm_struct *mm)
219 {
220 if (cpumask_test_cpu(smp_processor_id(), mm_cpumask(mm)))
221 __cpuc_flush_user_all();
222 }
223
224 static inline void
vivt_flush_cache_range(struct vm_area_struct * vma,unsigned long start,unsigned long end)225 vivt_flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
226 {
227 struct mm_struct *mm = vma->vm_mm;
228
229 if (!mm || cpumask_test_cpu(smp_processor_id(), mm_cpumask(mm)))
230 __cpuc_flush_user_range(start & PAGE_MASK, PAGE_ALIGN(end),
231 vma->vm_flags);
232 }
233
234 static inline void
vivt_flush_cache_page(struct vm_area_struct * vma,unsigned long user_addr,unsigned long pfn)235 vivt_flush_cache_page(struct vm_area_struct *vma, unsigned long user_addr, unsigned long pfn)
236 {
237 struct mm_struct *mm = vma->vm_mm;
238
239 if (!mm || cpumask_test_cpu(smp_processor_id(), mm_cpumask(mm))) {
240 unsigned long addr = user_addr & PAGE_MASK;
241 __cpuc_flush_user_range(addr, addr + PAGE_SIZE, vma->vm_flags);
242 }
243 }
244
245 #ifndef CONFIG_CPU_CACHE_VIPT
246 #define flush_cache_mm(mm) \
247 vivt_flush_cache_mm(mm)
248 #define flush_cache_range(vma,start,end) \
249 vivt_flush_cache_range(vma,start,end)
250 #define flush_cache_page(vma,addr,pfn) \
251 vivt_flush_cache_page(vma,addr,pfn)
252 #else
253 extern void flush_cache_mm(struct mm_struct *mm);
254 extern void flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end);
255 extern void flush_cache_page(struct vm_area_struct *vma, unsigned long user_addr, unsigned long pfn);
256 #endif
257
258 #define flush_cache_dup_mm(mm) flush_cache_mm(mm)
259
260 /*
261 * flush_icache_user_range is used when we want to ensure that the
262 * Harvard caches are synchronised for the user space address range.
263 * This is used for the ARM private sys_cacheflush system call.
264 */
265 #define flush_icache_user_range(s,e) __cpuc_coherent_user_range(s,e)
266
267 /*
268 * Perform necessary cache operations to ensure that data previously
269 * stored within this range of addresses can be executed by the CPU.
270 */
271 #define flush_icache_range(s,e) __cpuc_coherent_kern_range(s,e)
272
273 /*
274 * Perform necessary cache operations to ensure that the TLB will
275 * see data written in the specified area.
276 */
277 #define clean_dcache_area(start,size) cpu_dcache_clean_area(start, size)
278
279 /*
280 * flush_dcache_page is used when the kernel has written to the page
281 * cache page at virtual address page->virtual.
282 *
283 * If this page isn't mapped (ie, page_mapping == NULL), or it might
284 * have userspace mappings, then we _must_ always clean + invalidate
285 * the dcache entries associated with the kernel mapping.
286 *
287 * Otherwise we can defer the operation, and clean the cache when we are
288 * about to change to user space. This is the same method as used on SPARC64.
289 * See update_mmu_cache for the user space part.
290 */
291 #define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1
292 extern void flush_dcache_page(struct page *);
293
flush_kernel_vmap_range(void * addr,int size)294 static inline void flush_kernel_vmap_range(void *addr, int size)
295 {
296 if ((cache_is_vivt() || cache_is_vipt_aliasing()))
297 __cpuc_flush_dcache_area(addr, (size_t)size);
298 }
invalidate_kernel_vmap_range(void * addr,int size)299 static inline void invalidate_kernel_vmap_range(void *addr, int size)
300 {
301 if ((cache_is_vivt() || cache_is_vipt_aliasing()))
302 __cpuc_flush_dcache_area(addr, (size_t)size);
303 }
304
305 #define ARCH_HAS_FLUSH_ANON_PAGE
flush_anon_page(struct vm_area_struct * vma,struct page * page,unsigned long vmaddr)306 static inline void flush_anon_page(struct vm_area_struct *vma,
307 struct page *page, unsigned long vmaddr)
308 {
309 extern void __flush_anon_page(struct vm_area_struct *vma,
310 struct page *, unsigned long);
311 if (PageAnon(page))
312 __flush_anon_page(vma, page, vmaddr);
313 }
314
315 #define ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
316 extern void flush_kernel_dcache_page(struct page *);
317
318 #define flush_dcache_mmap_lock(mapping) xa_lock_irq(&mapping->i_pages)
319 #define flush_dcache_mmap_unlock(mapping) xa_unlock_irq(&mapping->i_pages)
320
321 /*
322 * We don't appear to need to do anything here. In fact, if we did, we'd
323 * duplicate cache flushing elsewhere performed by flush_dcache_page().
324 */
325 #define flush_icache_page(vma,page) do { } while (0)
326
327 /*
328 * flush_cache_vmap() is used when creating mappings (eg, via vmap,
329 * vmalloc, ioremap etc) in kernel space for pages. On non-VIPT
330 * caches, since the direct-mappings of these pages may contain cached
331 * data, we need to do a full cache flush to ensure that writebacks
332 * don't corrupt data placed into these pages via the new mappings.
333 */
flush_cache_vmap(unsigned long start,unsigned long end)334 static inline void flush_cache_vmap(unsigned long start, unsigned long end)
335 {
336 if (!cache_is_vipt_nonaliasing())
337 flush_cache_all();
338 else
339 /*
340 * set_pte_at() called from vmap_pte_range() does not
341 * have a DSB after cleaning the cache line.
342 */
343 dsb(ishst);
344 }
345
flush_cache_vunmap(unsigned long start,unsigned long end)346 static inline void flush_cache_vunmap(unsigned long start, unsigned long end)
347 {
348 if (!cache_is_vipt_nonaliasing())
349 flush_cache_all();
350 }
351
352 /*
353 * Memory synchronization helpers for mixed cached vs non cached accesses.
354 *
355 * Some synchronization algorithms have to set states in memory with the
356 * cache enabled or disabled depending on the code path. It is crucial
357 * to always ensure proper cache maintenance to update main memory right
358 * away in that case.
359 *
360 * Any cached write must be followed by a cache clean operation.
361 * Any cached read must be preceded by a cache invalidate operation.
362 * Yet, in the read case, a cache flush i.e. atomic clean+invalidate
363 * operation is needed to avoid discarding possible concurrent writes to the
364 * accessed memory.
365 *
366 * Also, in order to prevent a cached writer from interfering with an
367 * adjacent non-cached writer, each state variable must be located to
368 * a separate cache line.
369 */
370
371 /*
372 * This needs to be >= the max cache writeback size of all
373 * supported platforms included in the current kernel configuration.
374 * This is used to align state variables to their own cache lines.
375 */
376 #define __CACHE_WRITEBACK_ORDER 6 /* guessed from existing platforms */
377 #define __CACHE_WRITEBACK_GRANULE (1 << __CACHE_WRITEBACK_ORDER)
378
379 /*
380 * There is no __cpuc_clean_dcache_area but we use it anyway for
381 * code intent clarity, and alias it to __cpuc_flush_dcache_area.
382 */
383 #define __cpuc_clean_dcache_area __cpuc_flush_dcache_area
384
385 /*
386 * Ensure preceding writes to *p by this CPU are visible to
387 * subsequent reads by other CPUs:
388 */
__sync_cache_range_w(volatile void * p,size_t size)389 static inline void __sync_cache_range_w(volatile void *p, size_t size)
390 {
391 char *_p = (char *)p;
392
393 __cpuc_clean_dcache_area(_p, size);
394 outer_clean_range(__pa(_p), __pa(_p + size));
395 }
396
397 /*
398 * Ensure preceding writes to *p by other CPUs are visible to
399 * subsequent reads by this CPU. We must be careful not to
400 * discard data simultaneously written by another CPU, hence the
401 * usage of flush rather than invalidate operations.
402 */
__sync_cache_range_r(volatile void * p,size_t size)403 static inline void __sync_cache_range_r(volatile void *p, size_t size)
404 {
405 char *_p = (char *)p;
406
407 #ifdef CONFIG_OUTER_CACHE
408 if (outer_cache.flush_range) {
409 /*
410 * Ensure dirty data migrated from other CPUs into our cache
411 * are cleaned out safely before the outer cache is cleaned:
412 */
413 __cpuc_clean_dcache_area(_p, size);
414
415 /* Clean and invalidate stale data for *p from outer ... */
416 outer_flush_range(__pa(_p), __pa(_p + size));
417 }
418 #endif
419
420 /* ... and inner cache: */
421 __cpuc_flush_dcache_area(_p, size);
422 }
423
424 #define sync_cache_w(ptr) __sync_cache_range_w(ptr, sizeof *(ptr))
425 #define sync_cache_r(ptr) __sync_cache_range_r(ptr, sizeof *(ptr))
426
427 /*
428 * Disabling cache access for one CPU in an ARMv7 SMP system is tricky.
429 * To do so we must:
430 *
431 * - Clear the SCTLR.C bit to prevent further cache allocations
432 * - Flush the desired level of cache
433 * - Clear the ACTLR "SMP" bit to disable local coherency
434 *
435 * ... and so without any intervening memory access in between those steps,
436 * not even to the stack.
437 *
438 * WARNING -- After this has been called:
439 *
440 * - No ldrex/strex (and similar) instructions must be used.
441 * - The CPU is obviously no longer coherent with the other CPUs.
442 * - This is unlikely to work as expected if Linux is running non-secure.
443 *
444 * Note:
445 *
446 * - This is known to apply to several ARMv7 processor implementations,
447 * however some exceptions may exist. Caveat emptor.
448 *
449 * - The clobber list is dictated by the call to v7_flush_dcache_*.
450 * fp is preserved to the stack explicitly prior disabling the cache
451 * since adding it to the clobber list is incompatible with having
452 * CONFIG_FRAME_POINTER=y. ip is saved as well if ever r12-clobbering
453 * trampoline are inserted by the linker and to keep sp 64-bit aligned.
454 */
455 #define v7_exit_coherency_flush(level) \
456 asm volatile( \
457 ".arch armv7-a \n\t" \
458 "stmfd sp!, {fp, ip} \n\t" \
459 "mrc p15, 0, r0, c1, c0, 0 @ get SCTLR \n\t" \
460 "bic r0, r0, #"__stringify(CR_C)" \n\t" \
461 "mcr p15, 0, r0, c1, c0, 0 @ set SCTLR \n\t" \
462 "isb \n\t" \
463 "bl v7_flush_dcache_"__stringify(level)" \n\t" \
464 "mrc p15, 0, r0, c1, c0, 1 @ get ACTLR \n\t" \
465 "bic r0, r0, #(1 << 6) @ disable local coherency \n\t" \
466 "mcr p15, 0, r0, c1, c0, 1 @ set ACTLR \n\t" \
467 "isb \n\t" \
468 "dsb \n\t" \
469 "ldmfd sp!, {fp, ip}" \
470 : : : "r0","r1","r2","r3","r4","r5","r6","r7", \
471 "r9","r10","lr","memory" )
472
473 void flush_uprobe_xol_access(struct page *page, unsigned long uaddr,
474 void *kaddr, unsigned long len);
475
476
477 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
478 void check_cpu_icache_size(int cpuid);
479 #else
check_cpu_icache_size(int cpuid)480 static inline void check_cpu_icache_size(int cpuid) { }
481 #endif
482
483 #endif
484
