Linux Audio

Check our new training course

Loading...
v6.13.7
  1/* SPDX-License-Identifier: GPL-2.0-only */
  2/*
  3 *  linux/arch/arm/mm/cache-v7.S
  4 *
  5 *  Copyright (C) 2001 Deep Blue Solutions Ltd.
  6 *  Copyright (C) 2005 ARM Ltd.
  7 *
  8 *  This is the "shell" of the ARMv7 processor support.
  9 */
 10#include <linux/linkage.h>
 11#include <linux/init.h>
 12#include <linux/cfi_types.h>
 13#include <asm/assembler.h>
 14#include <asm/errno.h>
 15#include <asm/unwind.h>
 16#include <asm/hardware/cache-b15-rac.h>
 17
 18#include "proc-macros.S"
 19
 20.arch armv7-a
 21
 22#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
 23.globl icache_size
 24	.data
 25	.align	2
 26icache_size:
 27	.long	64
 28	.text
 29#endif
 30/*
 31 * The secondary kernel init calls v7_flush_dcache_all before it enables
 32 * the L1; however, the L1 comes out of reset in an undefined state, so
 33 * the clean + invalidate performed by v7_flush_dcache_all causes a bunch
 34 * of cache lines with uninitialized data and uninitialized tags to get
 35 * written out to memory, which does really unpleasant things to the main
 36 * processor.  We fix this by performing an invalidate, rather than a
 37 * clean + invalidate, before jumping into the kernel.
 38 *
 39 * This function needs to be called for both secondary cores startup and
 40 * primary core resume procedures.
 41 */
 42ENTRY(v7_invalidate_l1)
 43	mov	r0, #0
 44	mcr	p15, 2, r0, c0, c0, 0	@ select L1 data cache in CSSELR
 45	isb
 46	mrc	p15, 1, r0, c0, c0, 0	@ read cache geometry from CCSIDR
 47
 48	movw	r3, #0x3ff
 49	and	r3, r3, r0, lsr #3	@ 'Associativity' in CCSIDR[12:3]
 50	clz	r1, r3			@ WayShift
 51	mov	r2, #1
 52	mov	r3, r3, lsl r1		@ NumWays-1 shifted into bits [31:...]
 53	movs	r1, r2, lsl r1		@ #1 shifted left by same amount
 54	moveq	r1, #1			@ r1 needs value > 0 even if only 1 way
 55
 56	and	r2, r0, #0x7
 57	add	r2, r2, #4		@ SetShift
 58
 591:	movw	ip, #0x7fff
 60	and	r0, ip, r0, lsr #13	@ 'NumSets' in CCSIDR[27:13]
 61
 622:	mov	ip, r0, lsl r2		@ NumSet << SetShift
 63	orr	ip, ip, r3		@ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
 64	mcr	p15, 0, ip, c7, c6, 2
 65	subs	r0, r0, #1		@ Set--
 66	bpl	2b
 67	subs	r3, r3, r1		@ Way--
 68	bcc	3f
 69	mrc	p15, 1, r0, c0, c0, 0	@ re-read cache geometry from CCSIDR
 70	b	1b
 713:	dsb	st
 72	isb
 73	ret	lr
 74ENDPROC(v7_invalidate_l1)
 75
 76/*
 77 *	v7_flush_icache_all()
 78 *
 79 *	Flush the whole I-cache.
 80 *
 81 *	Registers:
 82 *	r0 - set to 0
 83 */
 84SYM_TYPED_FUNC_START(v7_flush_icache_all)
 85	mov	r0, #0
 86	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)		@ invalidate I-cache inner shareable
 87	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)		@ I+BTB cache invalidate
 88	ret	lr
 89SYM_FUNC_END(v7_flush_icache_all)
 90
 91 /*
 92 *     v7_flush_dcache_louis()
 93 *
 94 *     Flush the D-cache up to the Level of Unification Inner Shareable
 95 *
 96 *     Corrupted registers: r0-r6, r9-r10
 97 */
 98
 99ENTRY(v7_flush_dcache_louis)
100	dmb					@ ensure ordering with previous memory accesses
101	mrc	p15, 1, r0, c0, c0, 1		@ read clidr, r0 = clidr
102ALT_SMP(mov	r3, r0, lsr #20)		@ move LoUIS into position
103ALT_UP(	mov	r3, r0, lsr #26)		@ move LoUU into position
104	ands	r3, r3, #7 << 1 		@ extract LoU*2 field from clidr
105	bne	start_flush_levels		@ LoU != 0, start flushing
106#ifdef CONFIG_ARM_ERRATA_643719
107ALT_SMP(mrc	p15, 0, r2, c0, c0, 0)		@ read main ID register
108ALT_UP(	ret	lr)				@ LoUU is zero, so nothing to do
109	movw	r1, #:lower16:(0x410fc090 >> 4)	@ ID of ARM Cortex A9 r0p?
110	movt	r1, #:upper16:(0x410fc090 >> 4)
111	teq	r1, r2, lsr #4			@ test for errata affected core and if so...
112	moveq	r3, #1 << 1			@   fix LoUIS value
113	beq	start_flush_levels		@   start flushing cache levels
114#endif
115	ret	lr
116ENDPROC(v7_flush_dcache_louis)
117
118/*
119 *	v7_flush_dcache_all()
120 *
121 *	Flush the whole D-cache.
122 *
123 *	Corrupted registers: r0-r6, r9-r10
124 *
125 *	- mm    - mm_struct describing address space
126 */
127ENTRY(v7_flush_dcache_all)
128	dmb					@ ensure ordering with previous memory accesses
129	mrc	p15, 1, r0, c0, c0, 1		@ read clidr
130	mov	r3, r0, lsr #23			@ move LoC into position
131	ands	r3, r3, #7 << 1			@ extract LoC*2 from clidr
132	beq	finished			@ if loc is 0, then no need to clean
133start_flush_levels:
134	mov	r10, #0				@ start clean at cache level 0
135flush_levels:
136	add	r2, r10, r10, lsr #1		@ work out 3x current cache level
137	mov	r1, r0, lsr r2			@ extract cache type bits from clidr
138	and	r1, r1, #7			@ mask of the bits for current cache only
139	cmp	r1, #2				@ see what cache we have at this level
140	blt	skip				@ skip if no cache, or just i-cache
141#ifdef CONFIG_PREEMPTION
142	save_and_disable_irqs_notrace r9	@ make cssr&csidr read atomic
143#endif
144	mcr	p15, 2, r10, c0, c0, 0		@ select current cache level in cssr
145	isb					@ isb to sych the new cssr&csidr
146	mrc	p15, 1, r1, c0, c0, 0		@ read the new csidr
147#ifdef CONFIG_PREEMPTION
148	restore_irqs_notrace r9
149#endif
150	and	r2, r1, #7			@ extract the length of the cache lines
151	add	r2, r2, #4			@ add 4 (line length offset)
152	movw	r4, #0x3ff
153	ands	r4, r4, r1, lsr #3		@ find maximum number on the way size
154	clz	r5, r4				@ find bit position of way size increment
155	movw	r6, #0x7fff
156	and	r1, r6, r1, lsr #13		@ extract max number of the index size
157	mov	r6, #1
158	movne	r4, r4, lsl r5			@ # of ways shifted into bits [31:...]
159	movne	r6, r6, lsl r5			@ 1 shifted left by same amount
160loop1:
161	mov	r9, r1				@ create working copy of max index
162loop2:
163	mov	r5, r9, lsl r2			@ factor set number into r5
164	orr	r5, r5, r4			@ factor way number into r5
165	orr	r5, r5, r10			@ factor cache level into r5
166	mcr	p15, 0, r5, c7, c14, 2		@ clean & invalidate by set/way
167	subs	r9, r9, #1			@ decrement the index
168	bge	loop2
169	subs	r4, r4, r6			@ decrement the way
170	bcs	loop1
171skip:
172	add	r10, r10, #2			@ increment cache number
173	cmp	r3, r10
174#ifdef CONFIG_ARM_ERRATA_814220
175	dsb
176#endif
177	bgt	flush_levels
178finished:
179	mov	r10, #0				@ switch back to cache level 0
180	mcr	p15, 2, r10, c0, c0, 0		@ select current cache level in cssr
181	dsb	st
182	isb
183	ret	lr
184ENDPROC(v7_flush_dcache_all)
185
186/*
187 *	v7_flush_cache_all()
188 *
189 *	Flush the entire cache system.
190 *  The data cache flush is now achieved using atomic clean / invalidates
191 *  working outwards from L1 cache. This is done using Set/Way based cache
192 *  maintenance instructions.
193 *  The instruction cache can still be invalidated back to the point of
194 *  unification in a single instruction.
195 *
196 */
197SYM_TYPED_FUNC_START(v7_flush_kern_cache_all)
198	stmfd	sp!, {r4-r6, r9-r10, lr}
199	bl	v7_flush_dcache_all
200	mov	r0, #0
201	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)	@ invalidate I-cache inner shareable
202	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)	@ I+BTB cache invalidate
203	ldmfd	sp!, {r4-r6, r9-r10, lr}
204	ret	lr
205SYM_FUNC_END(v7_flush_kern_cache_all)
206
207 /*
208 *     v7_flush_kern_cache_louis(void)
209 *
210 *     Flush the data cache up to Level of Unification Inner Shareable.
211 *     Invalidate the I-cache to the point of unification.
212 */
213SYM_TYPED_FUNC_START(v7_flush_kern_cache_louis)
214	stmfd	sp!, {r4-r6, r9-r10, lr}
215	bl	v7_flush_dcache_louis
216	mov	r0, #0
217	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)	@ invalidate I-cache inner shareable
218	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)	@ I+BTB cache invalidate
219	ldmfd	sp!, {r4-r6, r9-r10, lr}
220	ret	lr
221SYM_FUNC_END(v7_flush_kern_cache_louis)
222
223/*
224 *	v7_flush_cache_all()
225 *
226 *	Flush all TLB entries in a particular address space
227 *
228 *	- mm    - mm_struct describing address space
229 */
230SYM_TYPED_FUNC_START(v7_flush_user_cache_all)
231	ret	lr
232SYM_FUNC_END(v7_flush_user_cache_all)
233
234/*
235 *	v7_flush_cache_range(start, end, flags)
236 *
237 *	Flush a range of TLB entries in the specified address space.
238 *
239 *	- start - start address (may not be aligned)
240 *	- end   - end address (exclusive, may not be aligned)
241 *	- flags	- vm_area_struct flags describing address space
242 *
243 *	It is assumed that:
244 *	- we have a VIPT cache.
245 */
246SYM_TYPED_FUNC_START(v7_flush_user_cache_range)
247	ret	lr
248SYM_FUNC_END(v7_flush_user_cache_range)
 
249
250/*
251 *	v7_coherent_kern_range(start,end)
252 *
253 *	Ensure that the I and D caches are coherent within specified
254 *	region.  This is typically used when code has been written to
255 *	a memory region, and will be executed.
256 *
257 *	- start   - virtual start address of region
258 *	- end     - virtual end address of region
259 *
260 *	It is assumed that:
261 *	- the Icache does not read data from the write buffer
262 */
263SYM_TYPED_FUNC_START(v7_coherent_kern_range)
264#ifdef CONFIG_CFI_CLANG /* Fallthrough if !CFI */
265	b	v7_coherent_user_range
266#endif
267SYM_FUNC_END(v7_coherent_kern_range)
268
269/*
270 *	v7_coherent_user_range(start,end)
271 *
272 *	Ensure that the I and D caches are coherent within specified
273 *	region.  This is typically used when code has been written to
274 *	a memory region, and will be executed.
275 *
276 *	- start   - virtual start address of region
277 *	- end     - virtual end address of region
278 *
279 *	It is assumed that:
280 *	- the Icache does not read data from the write buffer
281 */
282SYM_TYPED_FUNC_START(v7_coherent_user_range)
283 UNWIND(.fnstart		)
284	dcache_line_size r2, r3
285	sub	r3, r2, #1
286	bic	r12, r0, r3
287#ifdef CONFIG_ARM_ERRATA_764369
288	ALT_SMP(W(dsb))
289	ALT_UP(W(nop))
290#endif
2911:
292 USER(	mcr	p15, 0, r12, c7, c11, 1	)	@ clean D line to the point of unification
293	add	r12, r12, r2
294	cmp	r12, r1
295	blo	1b
296	dsb	ishst
297#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
298	ldr	r3, =icache_size
299	ldr	r2, [r3, #0]
300#else
301	icache_line_size r2, r3
302#endif
303	sub	r3, r2, #1
304	bic	r12, r0, r3
3052:
306 USER(	mcr	p15, 0, r12, c7, c5, 1	)	@ invalidate I line
307	add	r12, r12, r2
308	cmp	r12, r1
309	blo	2b
310	mov	r0, #0
311	ALT_SMP(mcr	p15, 0, r0, c7, c1, 6)	@ invalidate BTB Inner Shareable
312	ALT_UP(mcr	p15, 0, r0, c7, c5, 6)	@ invalidate BTB
313	dsb	ishst
314	isb
315	ret	lr
316
317/*
318 * Fault handling for the cache operation above. If the virtual address in r0
319 * isn't mapped, fail with -EFAULT.
320 */
3219001:
322#ifdef CONFIG_ARM_ERRATA_775420
323	dsb
324#endif
325	mov	r0, #-EFAULT
326	ret	lr
327 UNWIND(.fnend		)
328SYM_FUNC_END(v7_coherent_user_range)
 
329
330/*
331 *	v7_flush_kern_dcache_area(void *addr, size_t size)
332 *
333 *	Ensure that the data held in the page kaddr is written back
334 *	to the page in question.
335 *
336 *	- addr	- kernel address
337 *	- size	- region size
338 */
339SYM_TYPED_FUNC_START(v7_flush_kern_dcache_area)
340	dcache_line_size r2, r3
341	add	r1, r0, r1
342	sub	r3, r2, #1
343	bic	r0, r0, r3
344#ifdef CONFIG_ARM_ERRATA_764369
345	ALT_SMP(W(dsb))
346	ALT_UP(W(nop))
347#endif
3481:
349	mcr	p15, 0, r0, c7, c14, 1		@ clean & invalidate D line / unified line
350	add	r0, r0, r2
351	cmp	r0, r1
352	blo	1b
353	dsb	st
354	ret	lr
355SYM_FUNC_END(v7_flush_kern_dcache_area)
356
357/*
358 *	v7_dma_inv_range(start,end)
359 *
360 *	Invalidate the data cache within the specified region; we will
361 *	be performing a DMA operation in this region and we want to
362 *	purge old data in the cache.
363 *
364 *	- start   - virtual start address of region
365 *	- end     - virtual end address of region
366 */
367v7_dma_inv_range:
368	dcache_line_size r2, r3
369	sub	r3, r2, #1
370	tst	r0, r3
371	bic	r0, r0, r3
372#ifdef CONFIG_ARM_ERRATA_764369
373	ALT_SMP(W(dsb))
374	ALT_UP(W(nop))
375#endif
376	mcrne	p15, 0, r0, c7, c14, 1		@ clean & invalidate D / U line
377	addne	r0, r0, r2
378
379	tst	r1, r3
380	bic	r1, r1, r3
381	mcrne	p15, 0, r1, c7, c14, 1		@ clean & invalidate D / U line
382	cmp	r0, r1
3831:
384	mcrlo	p15, 0, r0, c7, c6, 1		@ invalidate D / U line
385	addlo	r0, r0, r2
386	cmplo	r0, r1
387	blo	1b
388	dsb	st
389	ret	lr
390ENDPROC(v7_dma_inv_range)
391
392/*
393 *	v7_dma_clean_range(start,end)
394 *	- start   - virtual start address of region
395 *	- end     - virtual end address of region
396 */
397v7_dma_clean_range:
398	dcache_line_size r2, r3
399	sub	r3, r2, #1
400	bic	r0, r0, r3
401#ifdef CONFIG_ARM_ERRATA_764369
402	ALT_SMP(W(dsb))
403	ALT_UP(W(nop))
404#endif
4051:
406	mcr	p15, 0, r0, c7, c10, 1		@ clean D / U line
407	add	r0, r0, r2
408	cmp	r0, r1
409	blo	1b
410	dsb	st
411	ret	lr
412ENDPROC(v7_dma_clean_range)
413
414/*
415 *	v7_dma_flush_range(start,end)
416 *	- start   - virtual start address of region
417 *	- end     - virtual end address of region
418 */
419SYM_TYPED_FUNC_START(v7_dma_flush_range)
420	dcache_line_size r2, r3
421	sub	r3, r2, #1
422	bic	r0, r0, r3
423#ifdef CONFIG_ARM_ERRATA_764369
424	ALT_SMP(W(dsb))
425	ALT_UP(W(nop))
426#endif
4271:
428	mcr	p15, 0, r0, c7, c14, 1		@ clean & invalidate D / U line
429	add	r0, r0, r2
430	cmp	r0, r1
431	blo	1b
432	dsb	st
433	ret	lr
434SYM_FUNC_END(v7_dma_flush_range)
435
436/*
437 *	dma_map_area(start, size, dir)
438 *	- start	- kernel virtual start address
439 *	- size	- size of region
440 *	- dir	- DMA direction
441 */
442SYM_TYPED_FUNC_START(v7_dma_map_area)
443	add	r1, r1, r0
444	teq	r2, #DMA_FROM_DEVICE
445	beq	v7_dma_inv_range
446	b	v7_dma_clean_range
447SYM_FUNC_END(v7_dma_map_area)
448
449/*
450 *	dma_unmap_area(start, size, dir)
451 *	- start	- kernel virtual start address
452 *	- size	- size of region
453 *	- dir	- DMA direction
454 */
455SYM_TYPED_FUNC_START(v7_dma_unmap_area)
456	add	r1, r1, r0
457	teq	r2, #DMA_TO_DEVICE
458	bne	v7_dma_inv_range
459	ret	lr
460SYM_FUNC_END(v7_dma_unmap_area)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v6.8
  1/* SPDX-License-Identifier: GPL-2.0-only */
  2/*
  3 *  linux/arch/arm/mm/cache-v7.S
  4 *
  5 *  Copyright (C) 2001 Deep Blue Solutions Ltd.
  6 *  Copyright (C) 2005 ARM Ltd.
  7 *
  8 *  This is the "shell" of the ARMv7 processor support.
  9 */
 10#include <linux/linkage.h>
 11#include <linux/init.h>
 
 12#include <asm/assembler.h>
 13#include <asm/errno.h>
 14#include <asm/unwind.h>
 15#include <asm/hardware/cache-b15-rac.h>
 16
 17#include "proc-macros.S"
 18
 19.arch armv7-a
 20
 21#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
 22.globl icache_size
 23	.data
 24	.align	2
 25icache_size:
 26	.long	64
 27	.text
 28#endif
 29/*
 30 * The secondary kernel init calls v7_flush_dcache_all before it enables
 31 * the L1; however, the L1 comes out of reset in an undefined state, so
 32 * the clean + invalidate performed by v7_flush_dcache_all causes a bunch
 33 * of cache lines with uninitialized data and uninitialized tags to get
 34 * written out to memory, which does really unpleasant things to the main
 35 * processor.  We fix this by performing an invalidate, rather than a
 36 * clean + invalidate, before jumping into the kernel.
 37 *
 38 * This function needs to be called for both secondary cores startup and
 39 * primary core resume procedures.
 40 */
 41ENTRY(v7_invalidate_l1)
 42	mov	r0, #0
 43	mcr	p15, 2, r0, c0, c0, 0	@ select L1 data cache in CSSELR
 44	isb
 45	mrc	p15, 1, r0, c0, c0, 0	@ read cache geometry from CCSIDR
 46
 47	movw	r3, #0x3ff
 48	and	r3, r3, r0, lsr #3	@ 'Associativity' in CCSIDR[12:3]
 49	clz	r1, r3			@ WayShift
 50	mov	r2, #1
 51	mov	r3, r3, lsl r1		@ NumWays-1 shifted into bits [31:...]
 52	movs	r1, r2, lsl r1		@ #1 shifted left by same amount
 53	moveq	r1, #1			@ r1 needs value > 0 even if only 1 way
 54
 55	and	r2, r0, #0x7
 56	add	r2, r2, #4		@ SetShift
 57
 581:	movw	ip, #0x7fff
 59	and	r0, ip, r0, lsr #13	@ 'NumSets' in CCSIDR[27:13]
 60
 612:	mov	ip, r0, lsl r2		@ NumSet << SetShift
 62	orr	ip, ip, r3		@ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
 63	mcr	p15, 0, ip, c7, c6, 2
 64	subs	r0, r0, #1		@ Set--
 65	bpl	2b
 66	subs	r3, r3, r1		@ Way--
 67	bcc	3f
 68	mrc	p15, 1, r0, c0, c0, 0	@ re-read cache geometry from CCSIDR
 69	b	1b
 703:	dsb	st
 71	isb
 72	ret	lr
 73ENDPROC(v7_invalidate_l1)
 74
 75/*
 76 *	v7_flush_icache_all()
 77 *
 78 *	Flush the whole I-cache.
 79 *
 80 *	Registers:
 81 *	r0 - set to 0
 82 */
 83ENTRY(v7_flush_icache_all)
 84	mov	r0, #0
 85	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)		@ invalidate I-cache inner shareable
 86	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)		@ I+BTB cache invalidate
 87	ret	lr
 88ENDPROC(v7_flush_icache_all)
 89
 90 /*
 91 *     v7_flush_dcache_louis()
 92 *
 93 *     Flush the D-cache up to the Level of Unification Inner Shareable
 94 *
 95 *     Corrupted registers: r0-r6, r9-r10
 96 */
 97
 98ENTRY(v7_flush_dcache_louis)
 99	dmb					@ ensure ordering with previous memory accesses
100	mrc	p15, 1, r0, c0, c0, 1		@ read clidr, r0 = clidr
101ALT_SMP(mov	r3, r0, lsr #20)		@ move LoUIS into position
102ALT_UP(	mov	r3, r0, lsr #26)		@ move LoUU into position
103	ands	r3, r3, #7 << 1 		@ extract LoU*2 field from clidr
104	bne	start_flush_levels		@ LoU != 0, start flushing
105#ifdef CONFIG_ARM_ERRATA_643719
106ALT_SMP(mrc	p15, 0, r2, c0, c0, 0)		@ read main ID register
107ALT_UP(	ret	lr)				@ LoUU is zero, so nothing to do
108	movw	r1, #:lower16:(0x410fc090 >> 4)	@ ID of ARM Cortex A9 r0p?
109	movt	r1, #:upper16:(0x410fc090 >> 4)
110	teq	r1, r2, lsr #4			@ test for errata affected core and if so...
111	moveq	r3, #1 << 1			@   fix LoUIS value
112	beq	start_flush_levels		@   start flushing cache levels
113#endif
114	ret	lr
115ENDPROC(v7_flush_dcache_louis)
116
117/*
118 *	v7_flush_dcache_all()
119 *
120 *	Flush the whole D-cache.
121 *
122 *	Corrupted registers: r0-r6, r9-r10
123 *
124 *	- mm    - mm_struct describing address space
125 */
126ENTRY(v7_flush_dcache_all)
127	dmb					@ ensure ordering with previous memory accesses
128	mrc	p15, 1, r0, c0, c0, 1		@ read clidr
129	mov	r3, r0, lsr #23			@ move LoC into position
130	ands	r3, r3, #7 << 1			@ extract LoC*2 from clidr
131	beq	finished			@ if loc is 0, then no need to clean
132start_flush_levels:
133	mov	r10, #0				@ start clean at cache level 0
134flush_levels:
135	add	r2, r10, r10, lsr #1		@ work out 3x current cache level
136	mov	r1, r0, lsr r2			@ extract cache type bits from clidr
137	and	r1, r1, #7			@ mask of the bits for current cache only
138	cmp	r1, #2				@ see what cache we have at this level
139	blt	skip				@ skip if no cache, or just i-cache
140#ifdef CONFIG_PREEMPTION
141	save_and_disable_irqs_notrace r9	@ make cssr&csidr read atomic
142#endif
143	mcr	p15, 2, r10, c0, c0, 0		@ select current cache level in cssr
144	isb					@ isb to sych the new cssr&csidr
145	mrc	p15, 1, r1, c0, c0, 0		@ read the new csidr
146#ifdef CONFIG_PREEMPTION
147	restore_irqs_notrace r9
148#endif
149	and	r2, r1, #7			@ extract the length of the cache lines
150	add	r2, r2, #4			@ add 4 (line length offset)
151	movw	r4, #0x3ff
152	ands	r4, r4, r1, lsr #3		@ find maximum number on the way size
153	clz	r5, r4				@ find bit position of way size increment
154	movw	r6, #0x7fff
155	and	r1, r6, r1, lsr #13		@ extract max number of the index size
156	mov	r6, #1
157	movne	r4, r4, lsl r5			@ # of ways shifted into bits [31:...]
158	movne	r6, r6, lsl r5			@ 1 shifted left by same amount
159loop1:
160	mov	r9, r1				@ create working copy of max index
161loop2:
162	mov	r5, r9, lsl r2			@ factor set number into r5
163	orr	r5, r5, r4			@ factor way number into r5
164	orr	r5, r5, r10			@ factor cache level into r5
165	mcr	p15, 0, r5, c7, c14, 2		@ clean & invalidate by set/way
166	subs	r9, r9, #1			@ decrement the index
167	bge	loop2
168	subs	r4, r4, r6			@ decrement the way
169	bcs	loop1
170skip:
171	add	r10, r10, #2			@ increment cache number
172	cmp	r3, r10
173#ifdef CONFIG_ARM_ERRATA_814220
174	dsb
175#endif
176	bgt	flush_levels
177finished:
178	mov	r10, #0				@ switch back to cache level 0
179	mcr	p15, 2, r10, c0, c0, 0		@ select current cache level in cssr
180	dsb	st
181	isb
182	ret	lr
183ENDPROC(v7_flush_dcache_all)
184
185/*
186 *	v7_flush_cache_all()
187 *
188 *	Flush the entire cache system.
189 *  The data cache flush is now achieved using atomic clean / invalidates
190 *  working outwards from L1 cache. This is done using Set/Way based cache
191 *  maintenance instructions.
192 *  The instruction cache can still be invalidated back to the point of
193 *  unification in a single instruction.
194 *
195 */
196ENTRY(v7_flush_kern_cache_all)
197	stmfd	sp!, {r4-r6, r9-r10, lr}
198	bl	v7_flush_dcache_all
199	mov	r0, #0
200	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)	@ invalidate I-cache inner shareable
201	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)	@ I+BTB cache invalidate
202	ldmfd	sp!, {r4-r6, r9-r10, lr}
203	ret	lr
204ENDPROC(v7_flush_kern_cache_all)
205
206 /*
207 *     v7_flush_kern_cache_louis(void)
208 *
209 *     Flush the data cache up to Level of Unification Inner Shareable.
210 *     Invalidate the I-cache to the point of unification.
211 */
212ENTRY(v7_flush_kern_cache_louis)
213	stmfd	sp!, {r4-r6, r9-r10, lr}
214	bl	v7_flush_dcache_louis
215	mov	r0, #0
216	ALT_SMP(mcr	p15, 0, r0, c7, c1, 0)	@ invalidate I-cache inner shareable
217	ALT_UP(mcr	p15, 0, r0, c7, c5, 0)	@ I+BTB cache invalidate
218	ldmfd	sp!, {r4-r6, r9-r10, lr}
219	ret	lr
220ENDPROC(v7_flush_kern_cache_louis)
221
222/*
223 *	v7_flush_cache_all()
224 *
225 *	Flush all TLB entries in a particular address space
226 *
227 *	- mm    - mm_struct describing address space
228 */
229ENTRY(v7_flush_user_cache_all)
230	/*FALLTHROUGH*/
 
231
232/*
233 *	v7_flush_cache_range(start, end, flags)
234 *
235 *	Flush a range of TLB entries in the specified address space.
236 *
237 *	- start - start address (may not be aligned)
238 *	- end   - end address (exclusive, may not be aligned)
239 *	- flags	- vm_area_struct flags describing address space
240 *
241 *	It is assumed that:
242 *	- we have a VIPT cache.
243 */
244ENTRY(v7_flush_user_cache_range)
245	ret	lr
246ENDPROC(v7_flush_user_cache_all)
247ENDPROC(v7_flush_user_cache_range)
248
249/*
250 *	v7_coherent_kern_range(start,end)
251 *
252 *	Ensure that the I and D caches are coherent within specified
253 *	region.  This is typically used when code has been written to
254 *	a memory region, and will be executed.
255 *
256 *	- start   - virtual start address of region
257 *	- end     - virtual end address of region
258 *
259 *	It is assumed that:
260 *	- the Icache does not read data from the write buffer
261 */
262ENTRY(v7_coherent_kern_range)
263	/* FALLTHROUGH */
 
 
 
264
265/*
266 *	v7_coherent_user_range(start,end)
267 *
268 *	Ensure that the I and D caches are coherent within specified
269 *	region.  This is typically used when code has been written to
270 *	a memory region, and will be executed.
271 *
272 *	- start   - virtual start address of region
273 *	- end     - virtual end address of region
274 *
275 *	It is assumed that:
276 *	- the Icache does not read data from the write buffer
277 */
278ENTRY(v7_coherent_user_range)
279 UNWIND(.fnstart		)
280	dcache_line_size r2, r3
281	sub	r3, r2, #1
282	bic	r12, r0, r3
283#ifdef CONFIG_ARM_ERRATA_764369
284	ALT_SMP(W(dsb))
285	ALT_UP(W(nop))
286#endif
2871:
288 USER(	mcr	p15, 0, r12, c7, c11, 1	)	@ clean D line to the point of unification
289	add	r12, r12, r2
290	cmp	r12, r1
291	blo	1b
292	dsb	ishst
293#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
294	ldr	r3, =icache_size
295	ldr	r2, [r3, #0]
296#else
297	icache_line_size r2, r3
298#endif
299	sub	r3, r2, #1
300	bic	r12, r0, r3
3012:
302 USER(	mcr	p15, 0, r12, c7, c5, 1	)	@ invalidate I line
303	add	r12, r12, r2
304	cmp	r12, r1
305	blo	2b
306	mov	r0, #0
307	ALT_SMP(mcr	p15, 0, r0, c7, c1, 6)	@ invalidate BTB Inner Shareable
308	ALT_UP(mcr	p15, 0, r0, c7, c5, 6)	@ invalidate BTB
309	dsb	ishst
310	isb
311	ret	lr
312
313/*
314 * Fault handling for the cache operation above. If the virtual address in r0
315 * isn't mapped, fail with -EFAULT.
316 */
3179001:
318#ifdef CONFIG_ARM_ERRATA_775420
319	dsb
320#endif
321	mov	r0, #-EFAULT
322	ret	lr
323 UNWIND(.fnend		)
324ENDPROC(v7_coherent_kern_range)
325ENDPROC(v7_coherent_user_range)
326
327/*
328 *	v7_flush_kern_dcache_area(void *addr, size_t size)
329 *
330 *	Ensure that the data held in the page kaddr is written back
331 *	to the page in question.
332 *
333 *	- addr	- kernel address
334 *	- size	- region size
335 */
336ENTRY(v7_flush_kern_dcache_area)
337	dcache_line_size r2, r3
338	add	r1, r0, r1
339	sub	r3, r2, #1
340	bic	r0, r0, r3
341#ifdef CONFIG_ARM_ERRATA_764369
342	ALT_SMP(W(dsb))
343	ALT_UP(W(nop))
344#endif
3451:
346	mcr	p15, 0, r0, c7, c14, 1		@ clean & invalidate D line / unified line
347	add	r0, r0, r2
348	cmp	r0, r1
349	blo	1b
350	dsb	st
351	ret	lr
352ENDPROC(v7_flush_kern_dcache_area)
353
354/*
355 *	v7_dma_inv_range(start,end)
356 *
357 *	Invalidate the data cache within the specified region; we will
358 *	be performing a DMA operation in this region and we want to
359 *	purge old data in the cache.
360 *
361 *	- start   - virtual start address of region
362 *	- end     - virtual end address of region
363 */
364v7_dma_inv_range:
365	dcache_line_size r2, r3
366	sub	r3, r2, #1
367	tst	r0, r3
368	bic	r0, r0, r3
369#ifdef CONFIG_ARM_ERRATA_764369
370	ALT_SMP(W(dsb))
371	ALT_UP(W(nop))
372#endif
373	mcrne	p15, 0, r0, c7, c14, 1		@ clean & invalidate D / U line
374	addne	r0, r0, r2
375
376	tst	r1, r3
377	bic	r1, r1, r3
378	mcrne	p15, 0, r1, c7, c14, 1		@ clean & invalidate D / U line
379	cmp	r0, r1
3801:
381	mcrlo	p15, 0, r0, c7, c6, 1		@ invalidate D / U line
382	addlo	r0, r0, r2
383	cmplo	r0, r1
384	blo	1b
385	dsb	st
386	ret	lr
387ENDPROC(v7_dma_inv_range)
388
389/*
390 *	v7_dma_clean_range(start,end)
391 *	- start   - virtual start address of region
392 *	- end     - virtual end address of region
393 */
394v7_dma_clean_range:
395	dcache_line_size r2, r3
396	sub	r3, r2, #1
397	bic	r0, r0, r3
398#ifdef CONFIG_ARM_ERRATA_764369
399	ALT_SMP(W(dsb))
400	ALT_UP(W(nop))
401#endif
4021:
403	mcr	p15, 0, r0, c7, c10, 1		@ clean D / U line
404	add	r0, r0, r2
405	cmp	r0, r1
406	blo	1b
407	dsb	st
408	ret	lr
409ENDPROC(v7_dma_clean_range)
410
411/*
412 *	v7_dma_flush_range(start,end)
413 *	- start   - virtual start address of region
414 *	- end     - virtual end address of region
415 */
416ENTRY(v7_dma_flush_range)
417	dcache_line_size r2, r3
418	sub	r3, r2, #1
419	bic	r0, r0, r3
420#ifdef CONFIG_ARM_ERRATA_764369
421	ALT_SMP(W(dsb))
422	ALT_UP(W(nop))
423#endif
4241:
425	mcr	p15, 0, r0, c7, c14, 1		@ clean & invalidate D / U line
426	add	r0, r0, r2
427	cmp	r0, r1
428	blo	1b
429	dsb	st
430	ret	lr
431ENDPROC(v7_dma_flush_range)
432
433/*
434 *	dma_map_area(start, size, dir)
435 *	- start	- kernel virtual start address
436 *	- size	- size of region
437 *	- dir	- DMA direction
438 */
439ENTRY(v7_dma_map_area)
440	add	r1, r1, r0
441	teq	r2, #DMA_FROM_DEVICE
442	beq	v7_dma_inv_range
443	b	v7_dma_clean_range
444ENDPROC(v7_dma_map_area)
445
446/*
447 *	dma_unmap_area(start, size, dir)
448 *	- start	- kernel virtual start address
449 *	- size	- size of region
450 *	- dir	- DMA direction
451 */
452ENTRY(v7_dma_unmap_area)
453	add	r1, r1, r0
454	teq	r2, #DMA_TO_DEVICE
455	bne	v7_dma_inv_range
456	ret	lr
457ENDPROC(v7_dma_unmap_area)
458
459	__INITDATA
460
461	@ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
462	define_cache_functions v7
463
464	/* The Broadcom Brahma-B15 read-ahead cache requires some modifications
465	 * to the v7_cache_fns, we only override the ones we need
466	 */
467#ifndef CONFIG_CACHE_B15_RAC
468	globl_equ	b15_flush_kern_cache_all,	v7_flush_kern_cache_all
469#endif
470	globl_equ	b15_flush_icache_all,		v7_flush_icache_all
471	globl_equ	b15_flush_kern_cache_louis,	v7_flush_kern_cache_louis
472	globl_equ	b15_flush_user_cache_all,	v7_flush_user_cache_all
473	globl_equ	b15_flush_user_cache_range,	v7_flush_user_cache_range
474	globl_equ	b15_coherent_kern_range,	v7_coherent_kern_range
475	globl_equ	b15_coherent_user_range,	v7_coherent_user_range
476	globl_equ	b15_flush_kern_dcache_area,	v7_flush_kern_dcache_area
477
478	globl_equ	b15_dma_map_area,		v7_dma_map_area
479	globl_equ	b15_dma_unmap_area,		v7_dma_unmap_area
480	globl_equ	b15_dma_flush_range,		v7_dma_flush_range
481
482	define_cache_functions b15