xref: /linux/arch/arm/mm/cache-v7m.S (revision 109d59b900e78834c66657dd4748fcedb9a1fe8d)
1/*
2 *  linux/arch/arm/mm/cache-v7m.S
3 *
4 *  Based on linux/arch/arm/mm/cache-v7.S
5 *
6 *  Copyright (C) 2001 Deep Blue Solutions Ltd.
7 *  Copyright (C) 2005 ARM Ltd.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 *  This is the "shell" of the ARMv7M processor support.
14 */
15#include <linux/linkage.h>
16#include <linux/init.h>
17#include <asm/assembler.h>
18#include <asm/errno.h>
19#include <asm/unwind.h>
20#include <asm/v7m.h>
21
22#include "proc-macros.S"
23
24/* Generic V7M read/write macros for memory mapped cache operations */
25.macro v7m_cache_read, rt, reg
26	movw	\rt, #:lower16:BASEADDR_V7M_SCB + \reg
27	movt	\rt, #:upper16:BASEADDR_V7M_SCB + \reg
28	ldr     \rt, [\rt]
29.endm
30
31.macro v7m_cacheop, rt, tmp, op, c = al
32	movw\c	\tmp, #:lower16:BASEADDR_V7M_SCB + \op
33	movt\c	\tmp, #:upper16:BASEADDR_V7M_SCB + \op
34	str\c	\rt, [\tmp]
35.endm
36
37
38.macro	read_ccsidr, rt
39	v7m_cache_read \rt, V7M_SCB_CCSIDR
40.endm
41
42.macro read_clidr, rt
43	v7m_cache_read \rt, V7M_SCB_CLIDR
44.endm
45
46.macro	write_csselr, rt, tmp
47	v7m_cacheop \rt, \tmp, V7M_SCB_CSSELR
48.endm
49
50/*
51 * dcisw: Invalidate data cache by set/way
52 */
53.macro dcisw, rt, tmp
54	v7m_cacheop \rt, \tmp, V7M_SCB_DCISW
55.endm
56
57/*
58 * dccisw: Clean and invalidate data cache by set/way
59 */
60.macro dccisw, rt, tmp
61	v7m_cacheop \rt, \tmp, V7M_SCB_DCCISW
62.endm
63
64/*
65 * dccimvac: Clean and invalidate data cache line by MVA to PoC.
66 */
67.irp    c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
68.macro dccimvac\c, rt, tmp
69	v7m_cacheop \rt, \tmp, V7M_SCB_DCCIMVAC, \c
70.endm
71.endr
72
73/*
74 * dcimvac: Invalidate data cache line by MVA to PoC
75 */
76.macro dcimvac, rt, tmp
77	v7m_cacheop \rt, \tmp, V7M_SCB_DCIMVAC
78.endm
79
80/*
81 * dccmvau: Clean data cache line by MVA to PoU
82 */
83.macro dccmvau, rt, tmp
84	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAU
85.endm
86
87/*
88 * dccmvac: Clean data cache line by MVA to PoC
89 */
90.macro dccmvac,  rt, tmp
91	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAC
92.endm
93
94/*
95 * icimvau: Invalidate instruction caches by MVA to PoU
96 */
97.macro icimvau, rt, tmp
98	v7m_cacheop \rt, \tmp, V7M_SCB_ICIMVAU
99.endm
100
101/*
102 * Invalidate the icache, inner shareable if SMP, invalidate BTB for UP.
103 * rt data ignored by ICIALLU(IS), so can be used for the address
104 */
105.macro invalidate_icache, rt
106	v7m_cacheop \rt, \rt, V7M_SCB_ICIALLU
107	mov \rt, #0
108.endm
109
110/*
111 * Invalidate the BTB, inner shareable if SMP.
112 * rt data ignored by BPIALL, so it can be used for the address
113 */
114.macro invalidate_bp, rt
115	v7m_cacheop \rt, \rt, V7M_SCB_BPIALL
116	mov \rt, #0
117.endm
118
119ENTRY(v7m_invalidate_l1)
120	mov	r0, #0
121
122	write_csselr r0, r1
123	read_ccsidr r0
124
125	movw	r1, #0x7fff
126	and	r2, r1, r0, lsr #13
127
128	movw	r1, #0x3ff
129
130	and	r3, r1, r0, lsr #3      @ NumWays - 1
131	add	r2, r2, #1              @ NumSets
132
133	and	r0, r0, #0x7
134	add	r0, r0, #4      @ SetShift
135
136	clz	r1, r3          @ WayShift
137	add	r4, r3, #1      @ NumWays
1381:	sub	r2, r2, #1      @ NumSets--
139	mov	r3, r4          @ Temp = NumWays
1402:	subs	r3, r3, #1      @ Temp--
141	mov	r5, r3, lsl r1
142	mov	r6, r2, lsl r0
143	orr	r5, r5, r6      @ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
144	dcisw	r5, r6
145	bgt	2b
146	cmp	r2, #0
147	bgt	1b
148	dsb	st
149	isb
150	ret	lr
151ENDPROC(v7m_invalidate_l1)
152
153/*
154 *	v7m_flush_icache_all()
155 *
156 *	Flush the whole I-cache.
157 *
158 *	Registers:
159 *	r0 - set to 0
160 */
161ENTRY(v7m_flush_icache_all)
162	invalidate_icache r0
163	ret	lr
164ENDPROC(v7m_flush_icache_all)
165
166/*
167 *	v7m_flush_dcache_all()
168 *
169 *	Flush the whole D-cache.
170 *
171 *	Corrupted registers: r0-r7, r9-r11
172 */
173ENTRY(v7m_flush_dcache_all)
174	dmb					@ ensure ordering with previous memory accesses
175	read_clidr r0
176	mov	r3, r0, lsr #23			@ move LoC into position
177	ands	r3, r3, #7 << 1			@ extract LoC*2 from clidr
178	beq	finished			@ if loc is 0, then no need to clean
179start_flush_levels:
180	mov	r10, #0				@ start clean at cache level 0
181flush_levels:
182	add	r2, r10, r10, lsr #1		@ work out 3x current cache level
183	mov	r1, r0, lsr r2			@ extract cache type bits from clidr
184	and	r1, r1, #7			@ mask of the bits for current cache only
185	cmp	r1, #2				@ see what cache we have at this level
186	blt	skip				@ skip if no cache, or just i-cache
187#ifdef CONFIG_PREEMPT
188	save_and_disable_irqs_notrace r9	@ make cssr&csidr read atomic
189#endif
190	write_csselr r10, r1			@ set current cache level
191	isb					@ isb to sych the new cssr&csidr
192	read_ccsidr r1				@ read the new csidr
193#ifdef CONFIG_PREEMPT
194	restore_irqs_notrace r9
195#endif
196	and	r2, r1, #7			@ extract the length of the cache lines
197	add	r2, r2, #4			@ add 4 (line length offset)
198	movw	r4, #0x3ff
199	ands	r4, r4, r1, lsr #3		@ find maximum number on the way size
200	clz	r5, r4				@ find bit position of way size increment
201	movw	r7, #0x7fff
202	ands	r7, r7, r1, lsr #13		@ extract max number of the index size
203loop1:
204	mov	r9, r7				@ create working copy of max index
205loop2:
206	lsl	r6, r4, r5
207	orr	r11, r10, r6			@ factor way and cache number into r11
208	lsl	r6, r9, r2
209	orr	r11, r11, r6			@ factor index number into r11
210	dccisw	r11, r6				@ clean/invalidate by set/way
211	subs	r9, r9, #1			@ decrement the index
212	bge	loop2
213	subs	r4, r4, #1			@ decrement the way
214	bge	loop1
215skip:
216	add	r10, r10, #2			@ increment cache number
217	cmp	r3, r10
218	bgt	flush_levels
219finished:
220	mov	r10, #0				@ switch back to cache level 0
221	write_csselr r10, r3			@ select current cache level in cssr
222	dsb	st
223	isb
224	ret	lr
225ENDPROC(v7m_flush_dcache_all)
226
227/*
228 *	v7m_flush_cache_all()
229 *
230 *	Flush the entire cache system.
231 *  The data cache flush is now achieved using atomic clean / invalidates
232 *  working outwards from L1 cache. This is done using Set/Way based cache
233 *  maintenance instructions.
234 *  The instruction cache can still be invalidated back to the point of
235 *  unification in a single instruction.
236 *
237 */
238ENTRY(v7m_flush_kern_cache_all)
239	stmfd	sp!, {r4-r7, r9-r11, lr}
240	bl	v7m_flush_dcache_all
241	invalidate_icache r0
242	ldmfd	sp!, {r4-r7, r9-r11, lr}
243	ret	lr
244ENDPROC(v7m_flush_kern_cache_all)
245
246/*
247 *	v7m_flush_cache_all()
248 *
249 *	Flush all TLB entries in a particular address space
250 *
251 *	- mm    - mm_struct describing address space
252 */
253ENTRY(v7m_flush_user_cache_all)
254	/*FALLTHROUGH*/
255
256/*
257 *	v7m_flush_cache_range(start, end, flags)
258 *
259 *	Flush a range of TLB entries in the specified address space.
260 *
261 *	- start - start address (may not be aligned)
262 *	- end   - end address (exclusive, may not be aligned)
263 *	- flags	- vm_area_struct flags describing address space
264 *
265 *	It is assumed that:
266 *	- we have a VIPT cache.
267 */
268ENTRY(v7m_flush_user_cache_range)
269	ret	lr
270ENDPROC(v7m_flush_user_cache_all)
271ENDPROC(v7m_flush_user_cache_range)
272
273/*
274 *	v7m_coherent_kern_range(start,end)
275 *
276 *	Ensure that the I and D caches are coherent within specified
277 *	region.  This is typically used when code has been written to
278 *	a memory region, and will be executed.
279 *
280 *	- start   - virtual start address of region
281 *	- end     - virtual end address of region
282 *
283 *	It is assumed that:
284 *	- the Icache does not read data from the write buffer
285 */
286ENTRY(v7m_coherent_kern_range)
287	/* FALLTHROUGH */
288
289/*
290 *	v7m_coherent_user_range(start,end)
291 *
292 *	Ensure that the I and D caches are coherent within specified
293 *	region.  This is typically used when code has been written to
294 *	a memory region, and will be executed.
295 *
296 *	- start   - virtual start address of region
297 *	- end     - virtual end address of region
298 *
299 *	It is assumed that:
300 *	- the Icache does not read data from the write buffer
301 */
302ENTRY(v7m_coherent_user_range)
303 UNWIND(.fnstart		)
304	dcache_line_size r2, r3
305	sub	r3, r2, #1
306	bic	r12, r0, r3
3071:
308/*
309 * We use open coded version of dccmvau otherwise USER() would
310 * point at movw instruction.
311 */
312	dccmvau	r12, r3
313	add	r12, r12, r2
314	cmp	r12, r1
315	blo	1b
316	dsb	ishst
317	icache_line_size r2, r3
318	sub	r3, r2, #1
319	bic	r12, r0, r3
3202:
321	icimvau r12, r3
322	add	r12, r12, r2
323	cmp	r12, r1
324	blo	2b
325	invalidate_bp r0
326	dsb	ishst
327	isb
328	ret	lr
329 UNWIND(.fnend		)
330ENDPROC(v7m_coherent_kern_range)
331ENDPROC(v7m_coherent_user_range)
332
333/*
334 *	v7m_flush_kern_dcache_area(void *addr, size_t size)
335 *
336 *	Ensure that the data held in the page kaddr is written back
337 *	to the page in question.
338 *
339 *	- addr	- kernel address
340 *	- size	- region size
341 */
342ENTRY(v7m_flush_kern_dcache_area)
343	dcache_line_size r2, r3
344	add	r1, r0, r1
345	sub	r3, r2, #1
346	bic	r0, r0, r3
3471:
348	dccimvac r0, r3		@ clean & invalidate D line / unified line
349	add	r0, r0, r2
350	cmp	r0, r1
351	blo	1b
352	dsb	st
353	ret	lr
354ENDPROC(v7m_flush_kern_dcache_area)
355
356/*
357 *	v7m_dma_inv_range(start,end)
358 *
359 *	Invalidate the data cache within the specified region; we will
360 *	be performing a DMA operation in this region and we want to
361 *	purge old data in the cache.
362 *
363 *	- start   - virtual start address of region
364 *	- end     - virtual end address of region
365 */
366v7m_dma_inv_range:
367	dcache_line_size r2, r3
368	sub	r3, r2, #1
369	tst	r0, r3
370	bic	r0, r0, r3
371	dccimvacne r0, r3
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
3761:
377	dcimvac r0, r3
378	add	r0, r0, r2
379	cmp	r0, r1
380	blo	1b
381	dsb	st
382	ret	lr
383ENDPROC(v7m_dma_inv_range)
384
385/*
386 *	v7m_dma_clean_range(start,end)
387 *	- start   - virtual start address of region
388 *	- end     - virtual end address of region
389 */
390v7m_dma_clean_range:
391	dcache_line_size r2, r3
392	sub	r3, r2, #1
393	bic	r0, r0, r3
3941:
395	dccmvac r0, r3			@ clean D / U line
396	add	r0, r0, r2
397	cmp	r0, r1
398	blo	1b
399	dsb	st
400	ret	lr
401ENDPROC(v7m_dma_clean_range)
402
403/*
404 *	v7m_dma_flush_range(start,end)
405 *	- start   - virtual start address of region
406 *	- end     - virtual end address of region
407 */
408ENTRY(v7m_dma_flush_range)
409	dcache_line_size r2, r3
410	sub	r3, r2, #1
411	bic	r0, r0, r3
4121:
413	dccimvac r0, r3			 @ clean & invalidate D / U line
414	add	r0, r0, r2
415	cmp	r0, r1
416	blo	1b
417	dsb	st
418	ret	lr
419ENDPROC(v7m_dma_flush_range)
420
421/*
422 *	dma_map_area(start, size, dir)
423 *	- start	- kernel virtual start address
424 *	- size	- size of region
425 *	- dir	- DMA direction
426 */
427ENTRY(v7m_dma_map_area)
428	add	r1, r1, r0
429	teq	r2, #DMA_FROM_DEVICE
430	beq	v7m_dma_inv_range
431	b	v7m_dma_clean_range
432ENDPROC(v7m_dma_map_area)
433
434/*
435 *	dma_unmap_area(start, size, dir)
436 *	- start	- kernel virtual start address
437 *	- size	- size of region
438 *	- dir	- DMA direction
439 */
440ENTRY(v7m_dma_unmap_area)
441	add	r1, r1, r0
442	teq	r2, #DMA_TO_DEVICE
443	bne	v7m_dma_inv_range
444	ret	lr
445ENDPROC(v7m_dma_unmap_area)
446
447	.globl	v7m_flush_kern_cache_louis
448	.equ	v7m_flush_kern_cache_louis, v7m_flush_kern_cache_all
449
450	__INITDATA
451
452	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
453	define_cache_functions v7m
454