1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 *  linux/arch/arm/mm/cache-v7m.S
4 *
5 *  Based on linux/arch/arm/mm/cache-v7.S
6 *
7 *  Copyright (C) 2001 Deep Blue Solutions Ltd.
8 *  Copyright (C) 2005 ARM Ltd.
9 *
10 *  This is the "shell" of the ARMv7M processor support.
11 */
12#include <linux/linkage.h>
13#include <linux/init.h>
14#include <asm/assembler.h>
15#include <asm/errno.h>
16#include <asm/unwind.h>
17#include <asm/v7m.h>
18
19#include "proc-macros.S"
20
21/* Generic V7M read/write macros for memory mapped cache operations */
22.macro v7m_cache_read, rt, reg
23	movw	\rt, #:lower16:BASEADDR_V7M_SCB + \reg
24	movt	\rt, #:upper16:BASEADDR_V7M_SCB + \reg
25	ldr     \rt, [\rt]
26.endm
27
28.macro v7m_cacheop, rt, tmp, op, c = al
29	movw\c	\tmp, #:lower16:BASEADDR_V7M_SCB + \op
30	movt\c	\tmp, #:upper16:BASEADDR_V7M_SCB + \op
31	str\c	\rt, [\tmp]
32.endm
33
34
35.macro	read_ccsidr, rt
36	v7m_cache_read \rt, V7M_SCB_CCSIDR
37.endm
38
39.macro read_clidr, rt
40	v7m_cache_read \rt, V7M_SCB_CLIDR
41.endm
42
43.macro	write_csselr, rt, tmp
44	v7m_cacheop \rt, \tmp, V7M_SCB_CSSELR
45.endm
46
47/*
48 * dcisw: Invalidate data cache by set/way
49 */
50.macro dcisw, rt, tmp
51	v7m_cacheop \rt, \tmp, V7M_SCB_DCISW
52.endm
53
54/*
55 * dccisw: Clean and invalidate data cache by set/way
56 */
57.macro dccisw, rt, tmp
58	v7m_cacheop \rt, \tmp, V7M_SCB_DCCISW
59.endm
60
61/*
62 * dccimvac: Clean and invalidate data cache line by MVA to PoC.
63 */
64.irp    c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
65.macro dccimvac\c, rt, tmp
66	v7m_cacheop \rt, \tmp, V7M_SCB_DCCIMVAC, \c
67.endm
68.endr
69
70/*
71 * dcimvac: Invalidate data cache line by MVA to PoC
72 */
73.irp    c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
74.macro dcimvac\c, rt, tmp
75	v7m_cacheop \rt, \tmp, V7M_SCB_DCIMVAC, \c
76.endm
77.endr
78
79/*
80 * dccmvau: Clean data cache line by MVA to PoU
81 */
82.macro dccmvau, rt, tmp
83	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAU
84.endm
85
86/*
87 * dccmvac: Clean data cache line by MVA to PoC
88 */
89.macro dccmvac,  rt, tmp
90	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAC
91.endm
92
93/*
94 * icimvau: Invalidate instruction caches by MVA to PoU
95 */
96.macro icimvau, rt, tmp
97	v7m_cacheop \rt, \tmp, V7M_SCB_ICIMVAU
98.endm
99
100/*
101 * Invalidate the icache, inner shareable if SMP, invalidate BTB for UP.
102 * rt data ignored by ICIALLU(IS), so can be used for the address
103 */
104.macro invalidate_icache, rt
105	v7m_cacheop \rt, \rt, V7M_SCB_ICIALLU
106	mov \rt, #0
107.endm
108
109/*
110 * Invalidate the BTB, inner shareable if SMP.
111 * rt data ignored by BPIALL, so it can be used for the address
112 */
113.macro invalidate_bp, rt
114	v7m_cacheop \rt, \rt, V7M_SCB_BPIALL
115	mov \rt, #0
116.endm
117
118ENTRY(v7m_invalidate_l1)
119	mov	r0, #0
120
121	write_csselr r0, r1
122	read_ccsidr r0
123
124	movw	r1, #0x7fff
125	and	r2, r1, r0, lsr #13
126
127	movw	r1, #0x3ff
128
129	and	r3, r1, r0, lsr #3      @ NumWays - 1
130	add	r2, r2, #1              @ NumSets
131
132	and	r0, r0, #0x7
133	add	r0, r0, #4      @ SetShift
134
135	clz	r1, r3          @ WayShift
136	add	r4, r3, #1      @ NumWays
1371:	sub	r2, r2, #1      @ NumSets--
138	mov	r3, r4          @ Temp = NumWays
1392:	subs	r3, r3, #1      @ Temp--
140	mov	r5, r3, lsl r1
141	mov	r6, r2, lsl r0
142	orr	r5, r5, r6      @ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
143	dcisw	r5, r6
144	bgt	2b
145	cmp	r2, #0
146	bgt	1b
147	dsb	st
148	isb
149	ret	lr
150ENDPROC(v7m_invalidate_l1)
151
152/*
153 *	v7m_flush_icache_all()
154 *
155 *	Flush the whole I-cache.
156 *
157 *	Registers:
158 *	r0 - set to 0
159 */
160ENTRY(v7m_flush_icache_all)
161	invalidate_icache r0
162	ret	lr
163ENDPROC(v7m_flush_icache_all)
164
165/*
166 *	v7m_flush_dcache_all()
167 *
168 *	Flush the whole D-cache.
169 *
170 *	Corrupted registers: r0-r7, r9-r11
171 */
172ENTRY(v7m_flush_dcache_all)
173	dmb					@ ensure ordering with previous memory accesses
174	read_clidr r0
175	mov	r3, r0, lsr #23			@ move LoC into position
176	ands	r3, r3, #7 << 1			@ extract LoC*2 from clidr
177	beq	finished			@ if loc is 0, then no need to clean
178start_flush_levels:
179	mov	r10, #0				@ start clean at cache level 0
180flush_levels:
181	add	r2, r10, r10, lsr #1		@ work out 3x current cache level
182	mov	r1, r0, lsr r2			@ extract cache type bits from clidr
183	and	r1, r1, #7			@ mask of the bits for current cache only
184	cmp	r1, #2				@ see what cache we have at this level
185	blt	skip				@ skip if no cache, or just i-cache
186#ifdef CONFIG_PREEMPTION
187	save_and_disable_irqs_notrace r9	@ make cssr&csidr read atomic
188#endif
189	write_csselr r10, r1			@ set current cache level
190	isb					@ isb to sych the new cssr&csidr
191	read_ccsidr r1				@ read the new csidr
192#ifdef CONFIG_PREEMPTION
193	restore_irqs_notrace r9
194#endif
195	and	r2, r1, #7			@ extract the length of the cache lines
196	add	r2, r2, #4			@ add 4 (line length offset)
197	movw	r4, #0x3ff
198	ands	r4, r4, r1, lsr #3		@ find maximum number on the way size
199	clz	r5, r4				@ find bit position of way size increment
200	movw	r7, #0x7fff
201	ands	r7, r7, r1, lsr #13		@ extract max number of the index size
202loop1:
203	mov	r9, r7				@ create working copy of max index
204loop2:
205	lsl	r6, r4, r5
206	orr	r11, r10, r6			@ factor way and cache number into r11
207	lsl	r6, r9, r2
208	orr	r11, r11, r6			@ factor index number into r11
209	dccisw	r11, r6				@ clean/invalidate by set/way
210	subs	r9, r9, #1			@ decrement the index
211	bge	loop2
212	subs	r4, r4, #1			@ decrement the way
213	bge	loop1
214skip:
215	add	r10, r10, #2			@ increment cache number
216	cmp	r3, r10
217	bgt	flush_levels
218finished:
219	mov	r10, #0				@ switch back to cache level 0
220	write_csselr r10, r3			@ select current cache level in cssr
221	dsb	st
222	isb
223	ret	lr
224ENDPROC(v7m_flush_dcache_all)
225
226/*
227 *	v7m_flush_cache_all()
228 *
229 *	Flush the entire cache system.
230 *  The data cache flush is now achieved using atomic clean / invalidates
231 *  working outwards from L1 cache. This is done using Set/Way based cache
232 *  maintenance instructions.
233 *  The instruction cache can still be invalidated back to the point of
234 *  unification in a single instruction.
235 *
236 */
237ENTRY(v7m_flush_kern_cache_all)
238	stmfd	sp!, {r4-r7, r9-r11, lr}
239	bl	v7m_flush_dcache_all
240	invalidate_icache r0
241	ldmfd	sp!, {r4-r7, r9-r11, lr}
242	ret	lr
243ENDPROC(v7m_flush_kern_cache_all)
244
245/*
246 *	v7m_flush_cache_all()
247 *
248 *	Flush all TLB entries in a particular address space
249 *
250 *	- mm    - mm_struct describing address space
251 */
252ENTRY(v7m_flush_user_cache_all)
253	/*FALLTHROUGH*/
254
255/*
256 *	v7m_flush_cache_range(start, end, flags)
257 *
258 *	Flush a range of TLB entries in the specified address space.
259 *
260 *	- start - start address (may not be aligned)
261 *	- end   - end address (exclusive, may not be aligned)
262 *	- flags	- vm_area_struct flags describing address space
263 *
264 *	It is assumed that:
265 *	- we have a VIPT cache.
266 */
267ENTRY(v7m_flush_user_cache_range)
268	ret	lr
269ENDPROC(v7m_flush_user_cache_all)
270ENDPROC(v7m_flush_user_cache_range)
271
272/*
273 *	v7m_coherent_kern_range(start,end)
274 *
275 *	Ensure that the I and D caches are coherent within specified
276 *	region.  This is typically used when code has been written to
277 *	a memory region, and will be executed.
278 *
279 *	- start   - virtual start address of region
280 *	- end     - virtual end address of region
281 *
282 *	It is assumed that:
283 *	- the Icache does not read data from the write buffer
284 */
285ENTRY(v7m_coherent_kern_range)
286	/* FALLTHROUGH */
287
288/*
289 *	v7m_coherent_user_range(start,end)
290 *
291 *	Ensure that the I and D caches are coherent within specified
292 *	region.  This is typically used when code has been written to
293 *	a memory region, and will be executed.
294 *
295 *	- start   - virtual start address of region
296 *	- end     - virtual end address of region
297 *
298 *	It is assumed that:
299 *	- the Icache does not read data from the write buffer
300 */
301ENTRY(v7m_coherent_user_range)
302 UNWIND(.fnstart		)
303	dcache_line_size r2, r3
304	sub	r3, r2, #1
305	bic	r12, r0, r3
3061:
307/*
308 * We use open coded version of dccmvau otherwise USER() would
309 * point at movw instruction.
310 */
311	dccmvau	r12, r3
312	add	r12, r12, r2
313	cmp	r12, r1
314	blo	1b
315	dsb	ishst
316	icache_line_size r2, r3
317	sub	r3, r2, #1
318	bic	r12, r0, r3
3192:
320	icimvau r12, r3
321	add	r12, r12, r2
322	cmp	r12, r1
323	blo	2b
324	invalidate_bp r0
325	dsb	ishst
326	isb
327	ret	lr
328 UNWIND(.fnend		)
329ENDPROC(v7m_coherent_kern_range)
330ENDPROC(v7m_coherent_user_range)
331
332/*
333 *	v7m_flush_kern_dcache_area(void *addr, size_t size)
334 *
335 *	Ensure that the data held in the page kaddr is written back
336 *	to the page in question.
337 *
338 *	- addr	- kernel address
339 *	- size	- region size
340 */
341ENTRY(v7m_flush_kern_dcache_area)
342	dcache_line_size r2, r3
343	add	r1, r0, r1
344	sub	r3, r2, #1
345	bic	r0, r0, r3
3461:
347	dccimvac r0, r3		@ clean & invalidate D line / unified line
348	add	r0, r0, r2
349	cmp	r0, r1
350	blo	1b
351	dsb	st
352	ret	lr
353ENDPROC(v7m_flush_kern_dcache_area)
354
355/*
356 *	v7m_dma_inv_range(start,end)
357 *
358 *	Invalidate the data cache within the specified region; we will
359 *	be performing a DMA operation in this region and we want to
360 *	purge old data in the cache.
361 *
362 *	- start   - virtual start address of region
363 *	- end     - virtual end address of region
364 */
365v7m_dma_inv_range:
366	dcache_line_size r2, r3
367	sub	r3, r2, #1
368	tst	r0, r3
369	bic	r0, r0, r3
370	dccimvacne r0, r3
371	addne	r0, r0, r2
372	subne	r3, r2, #1	@ restore r3, corrupted by v7m's dccimvac
373	tst	r1, r3
374	bic	r1, r1, r3
375	dccimvacne r1, r3
376	cmp	r0, r1
3771:
378	dcimvaclo r0, r3
379	addlo	r0, r0, r2
380	cmplo	r0, r1
381	blo	1b
382	dsb	st
383	ret	lr
384ENDPROC(v7m_dma_inv_range)
385
386/*
387 *	v7m_dma_clean_range(start,end)
388 *	- start   - virtual start address of region
389 *	- end     - virtual end address of region
390 */
391v7m_dma_clean_range:
392	dcache_line_size r2, r3
393	sub	r3, r2, #1
394	bic	r0, r0, r3
3951:
396	dccmvac r0, r3			@ clean D / U line
397	add	r0, r0, r2
398	cmp	r0, r1
399	blo	1b
400	dsb	st
401	ret	lr
402ENDPROC(v7m_dma_clean_range)
403
404/*
405 *	v7m_dma_flush_range(start,end)
406 *	- start   - virtual start address of region
407 *	- end     - virtual end address of region
408 */
409ENTRY(v7m_dma_flush_range)
410	dcache_line_size r2, r3
411	sub	r3, r2, #1
412	bic	r0, r0, r3
4131:
414	dccimvac r0, r3			 @ clean & invalidate D / U line
415	add	r0, r0, r2
416	cmp	r0, r1
417	blo	1b
418	dsb	st
419	ret	lr
420ENDPROC(v7m_dma_flush_range)
421
422/*
423 *	dma_map_area(start, size, dir)
424 *	- start	- kernel virtual start address
425 *	- size	- size of region
426 *	- dir	- DMA direction
427 */
428ENTRY(v7m_dma_map_area)
429	add	r1, r1, r0
430	teq	r2, #DMA_FROM_DEVICE
431	beq	v7m_dma_inv_range
432	b	v7m_dma_clean_range
433ENDPROC(v7m_dma_map_area)
434
435/*
436 *	dma_unmap_area(start, size, dir)
437 *	- start	- kernel virtual start address
438 *	- size	- size of region
439 *	- dir	- DMA direction
440 */
441ENTRY(v7m_dma_unmap_area)
442	add	r1, r1, r0
443	teq	r2, #DMA_TO_DEVICE
444	bne	v7m_dma_inv_range
445	ret	lr
446ENDPROC(v7m_dma_unmap_area)
447
448	.globl	v7m_flush_kern_cache_louis
449	.equ	v7m_flush_kern_cache_louis, v7m_flush_kern_cache_all
450
451	__INITDATA
452
453	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
454	define_cache_functions v7m
455