xref: /openbmc/linux/arch/arm/mm/cache-v7m.S (revision df687341)
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.irp    c,,eq,ne,cs,cc,mi,pl,vs,vc,hi,ls,ge,lt,gt,le,hs,lo
77.macro dcimvac\c, rt, tmp
78	v7m_cacheop \rt, \tmp, V7M_SCB_DCIMVAC, \c
79.endm
80.endr
81
82/*
83 * dccmvau: Clean data cache line by MVA to PoU
84 */
85.macro dccmvau, rt, tmp
86	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAU
87.endm
88
89/*
90 * dccmvac: Clean data cache line by MVA to PoC
91 */
92.macro dccmvac,  rt, tmp
93	v7m_cacheop \rt, \tmp, V7M_SCB_DCCMVAC
94.endm
95
96/*
97 * icimvau: Invalidate instruction caches by MVA to PoU
98 */
99.macro icimvau, rt, tmp
100	v7m_cacheop \rt, \tmp, V7M_SCB_ICIMVAU
101.endm
102
103/*
104 * Invalidate the icache, inner shareable if SMP, invalidate BTB for UP.
105 * rt data ignored by ICIALLU(IS), so can be used for the address
106 */
107.macro invalidate_icache, rt
108	v7m_cacheop \rt, \rt, V7M_SCB_ICIALLU
109	mov \rt, #0
110.endm
111
112/*
113 * Invalidate the BTB, inner shareable if SMP.
114 * rt data ignored by BPIALL, so it can be used for the address
115 */
116.macro invalidate_bp, rt
117	v7m_cacheop \rt, \rt, V7M_SCB_BPIALL
118	mov \rt, #0
119.endm
120
121ENTRY(v7m_invalidate_l1)
122	mov	r0, #0
123
124	write_csselr r0, r1
125	read_ccsidr r0
126
127	movw	r1, #0x7fff
128	and	r2, r1, r0, lsr #13
129
130	movw	r1, #0x3ff
131
132	and	r3, r1, r0, lsr #3      @ NumWays - 1
133	add	r2, r2, #1              @ NumSets
134
135	and	r0, r0, #0x7
136	add	r0, r0, #4      @ SetShift
137
138	clz	r1, r3          @ WayShift
139	add	r4, r3, #1      @ NumWays
1401:	sub	r2, r2, #1      @ NumSets--
141	mov	r3, r4          @ Temp = NumWays
1422:	subs	r3, r3, #1      @ Temp--
143	mov	r5, r3, lsl r1
144	mov	r6, r2, lsl r0
145	orr	r5, r5, r6      @ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
146	dcisw	r5, r6
147	bgt	2b
148	cmp	r2, #0
149	bgt	1b
150	dsb	st
151	isb
152	ret	lr
153ENDPROC(v7m_invalidate_l1)
154
155/*
156 *	v7m_flush_icache_all()
157 *
158 *	Flush the whole I-cache.
159 *
160 *	Registers:
161 *	r0 - set to 0
162 */
163ENTRY(v7m_flush_icache_all)
164	invalidate_icache r0
165	ret	lr
166ENDPROC(v7m_flush_icache_all)
167
168/*
169 *	v7m_flush_dcache_all()
170 *
171 *	Flush the whole D-cache.
172 *
173 *	Corrupted registers: r0-r7, r9-r11
174 */
175ENTRY(v7m_flush_dcache_all)
176	dmb					@ ensure ordering with previous memory accesses
177	read_clidr r0
178	mov	r3, r0, lsr #23			@ move LoC into position
179	ands	r3, r3, #7 << 1			@ extract LoC*2 from clidr
180	beq	finished			@ if loc is 0, then no need to clean
181start_flush_levels:
182	mov	r10, #0				@ start clean at cache level 0
183flush_levels:
184	add	r2, r10, r10, lsr #1		@ work out 3x current cache level
185	mov	r1, r0, lsr r2			@ extract cache type bits from clidr
186	and	r1, r1, #7			@ mask of the bits for current cache only
187	cmp	r1, #2				@ see what cache we have at this level
188	blt	skip				@ skip if no cache, or just i-cache
189#ifdef CONFIG_PREEMPT
190	save_and_disable_irqs_notrace r9	@ make cssr&csidr read atomic
191#endif
192	write_csselr r10, r1			@ set current cache level
193	isb					@ isb to sych the new cssr&csidr
194	read_ccsidr r1				@ read the new csidr
195#ifdef CONFIG_PREEMPT
196	restore_irqs_notrace r9
197#endif
198	and	r2, r1, #7			@ extract the length of the cache lines
199	add	r2, r2, #4			@ add 4 (line length offset)
200	movw	r4, #0x3ff
201	ands	r4, r4, r1, lsr #3		@ find maximum number on the way size
202	clz	r5, r4				@ find bit position of way size increment
203	movw	r7, #0x7fff
204	ands	r7, r7, r1, lsr #13		@ extract max number of the index size
205loop1:
206	mov	r9, r7				@ create working copy of max index
207loop2:
208	lsl	r6, r4, r5
209	orr	r11, r10, r6			@ factor way and cache number into r11
210	lsl	r6, r9, r2
211	orr	r11, r11, r6			@ factor index number into r11
212	dccisw	r11, r6				@ clean/invalidate by set/way
213	subs	r9, r9, #1			@ decrement the index
214	bge	loop2
215	subs	r4, r4, #1			@ decrement the way
216	bge	loop1
217skip:
218	add	r10, r10, #2			@ increment cache number
219	cmp	r3, r10
220	bgt	flush_levels
221finished:
222	mov	r10, #0				@ switch back to cache level 0
223	write_csselr r10, r3			@ select current cache level in cssr
224	dsb	st
225	isb
226	ret	lr
227ENDPROC(v7m_flush_dcache_all)
228
229/*
230 *	v7m_flush_cache_all()
231 *
232 *	Flush the entire cache system.
233 *  The data cache flush is now achieved using atomic clean / invalidates
234 *  working outwards from L1 cache. This is done using Set/Way based cache
235 *  maintenance instructions.
236 *  The instruction cache can still be invalidated back to the point of
237 *  unification in a single instruction.
238 *
239 */
240ENTRY(v7m_flush_kern_cache_all)
241	stmfd	sp!, {r4-r7, r9-r11, lr}
242	bl	v7m_flush_dcache_all
243	invalidate_icache r0
244	ldmfd	sp!, {r4-r7, r9-r11, lr}
245	ret	lr
246ENDPROC(v7m_flush_kern_cache_all)
247
248/*
249 *	v7m_flush_cache_all()
250 *
251 *	Flush all TLB entries in a particular address space
252 *
253 *	- mm    - mm_struct describing address space
254 */
255ENTRY(v7m_flush_user_cache_all)
256	/*FALLTHROUGH*/
257
258/*
259 *	v7m_flush_cache_range(start, end, flags)
260 *
261 *	Flush a range of TLB entries in the specified address space.
262 *
263 *	- start - start address (may not be aligned)
264 *	- end   - end address (exclusive, may not be aligned)
265 *	- flags	- vm_area_struct flags describing address space
266 *
267 *	It is assumed that:
268 *	- we have a VIPT cache.
269 */
270ENTRY(v7m_flush_user_cache_range)
271	ret	lr
272ENDPROC(v7m_flush_user_cache_all)
273ENDPROC(v7m_flush_user_cache_range)
274
275/*
276 *	v7m_coherent_kern_range(start,end)
277 *
278 *	Ensure that the I and D caches are coherent within specified
279 *	region.  This is typically used when code has been written to
280 *	a memory region, and will be executed.
281 *
282 *	- start   - virtual start address of region
283 *	- end     - virtual end address of region
284 *
285 *	It is assumed that:
286 *	- the Icache does not read data from the write buffer
287 */
288ENTRY(v7m_coherent_kern_range)
289	/* FALLTHROUGH */
290
291/*
292 *	v7m_coherent_user_range(start,end)
293 *
294 *	Ensure that the I and D caches are coherent within specified
295 *	region.  This is typically used when code has been written to
296 *	a memory region, and will be executed.
297 *
298 *	- start   - virtual start address of region
299 *	- end     - virtual end address of region
300 *
301 *	It is assumed that:
302 *	- the Icache does not read data from the write buffer
303 */
304ENTRY(v7m_coherent_user_range)
305 UNWIND(.fnstart		)
306	dcache_line_size r2, r3
307	sub	r3, r2, #1
308	bic	r12, r0, r3
3091:
310/*
311 * We use open coded version of dccmvau otherwise USER() would
312 * point at movw instruction.
313 */
314	dccmvau	r12, r3
315	add	r12, r12, r2
316	cmp	r12, r1
317	blo	1b
318	dsb	ishst
319	icache_line_size r2, r3
320	sub	r3, r2, #1
321	bic	r12, r0, r3
3222:
323	icimvau r12, r3
324	add	r12, r12, r2
325	cmp	r12, r1
326	blo	2b
327	invalidate_bp r0
328	dsb	ishst
329	isb
330	ret	lr
331 UNWIND(.fnend		)
332ENDPROC(v7m_coherent_kern_range)
333ENDPROC(v7m_coherent_user_range)
334
335/*
336 *	v7m_flush_kern_dcache_area(void *addr, size_t size)
337 *
338 *	Ensure that the data held in the page kaddr is written back
339 *	to the page in question.
340 *
341 *	- addr	- kernel address
342 *	- size	- region size
343 */
344ENTRY(v7m_flush_kern_dcache_area)
345	dcache_line_size r2, r3
346	add	r1, r0, r1
347	sub	r3, r2, #1
348	bic	r0, r0, r3
3491:
350	dccimvac r0, r3		@ clean & invalidate D line / unified line
351	add	r0, r0, r2
352	cmp	r0, r1
353	blo	1b
354	dsb	st
355	ret	lr
356ENDPROC(v7m_flush_kern_dcache_area)
357
358/*
359 *	v7m_dma_inv_range(start,end)
360 *
361 *	Invalidate the data cache within the specified region; we will
362 *	be performing a DMA operation in this region and we want to
363 *	purge old data in the cache.
364 *
365 *	- start   - virtual start address of region
366 *	- end     - virtual end address of region
367 */
368v7m_dma_inv_range:
369	dcache_line_size r2, r3
370	sub	r3, r2, #1
371	tst	r0, r3
372	bic	r0, r0, r3
373	dccimvacne r0, r3
374	addne	r0, r0, r2
375	subne	r3, r2, #1	@ restore r3, corrupted by v7m's dccimvac
376	tst	r1, r3
377	bic	r1, r1, r3
378	dccimvacne r1, r3
379	cmp	r0, r1
3801:
381	dcimvaclo r0, r3
382	addlo	r0, r0, r2
383	cmplo	r0, r1
384	blo	1b
385	dsb	st
386	ret	lr
387ENDPROC(v7m_dma_inv_range)
388
389/*
390 *	v7m_dma_clean_range(start,end)
391 *	- start   - virtual start address of region
392 *	- end     - virtual end address of region
393 */
394v7m_dma_clean_range:
395	dcache_line_size r2, r3
396	sub	r3, r2, #1
397	bic	r0, r0, r3
3981:
399	dccmvac r0, r3			@ clean D / U line
400	add	r0, r0, r2
401	cmp	r0, r1
402	blo	1b
403	dsb	st
404	ret	lr
405ENDPROC(v7m_dma_clean_range)
406
407/*
408 *	v7m_dma_flush_range(start,end)
409 *	- start   - virtual start address of region
410 *	- end     - virtual end address of region
411 */
412ENTRY(v7m_dma_flush_range)
413	dcache_line_size r2, r3
414	sub	r3, r2, #1
415	bic	r0, r0, r3
4161:
417	dccimvac r0, r3			 @ clean & invalidate D / U line
418	add	r0, r0, r2
419	cmp	r0, r1
420	blo	1b
421	dsb	st
422	ret	lr
423ENDPROC(v7m_dma_flush_range)
424
425/*
426 *	dma_map_area(start, size, dir)
427 *	- start	- kernel virtual start address
428 *	- size	- size of region
429 *	- dir	- DMA direction
430 */
431ENTRY(v7m_dma_map_area)
432	add	r1, r1, r0
433	teq	r2, #DMA_FROM_DEVICE
434	beq	v7m_dma_inv_range
435	b	v7m_dma_clean_range
436ENDPROC(v7m_dma_map_area)
437
438/*
439 *	dma_unmap_area(start, size, dir)
440 *	- start	- kernel virtual start address
441 *	- size	- size of region
442 *	- dir	- DMA direction
443 */
444ENTRY(v7m_dma_unmap_area)
445	add	r1, r1, r0
446	teq	r2, #DMA_TO_DEVICE
447	bne	v7m_dma_inv_range
448	ret	lr
449ENDPROC(v7m_dma_unmap_area)
450
451	.globl	v7m_flush_kern_cache_louis
452	.equ	v7m_flush_kern_cache_louis, v7m_flush_kern_cache_all
453
454	__INITDATA
455
456	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
457	define_cache_functions v7m
458