2 * linux/arch/arm/mm/cache-v7.S
4 * Copyright (C) 2001 Deep Blue Solutions Ltd.
5 * Copyright (C) 2005 ARM Ltd.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * This is the "shell" of the ARMv7 processor support.
13 #include <linux/linkage.h>
14 #include <linux/init.h>
15 #include <asm/assembler.h>
16 #include <asm/errno.h>
17 #include <asm/unwind.h>
18 #include <asm/hardware/cache-b15-rac.h>
20 #include "proc-macros.S"
23 * The secondary kernel init calls v7_flush_dcache_all before it enables
24 * the L1; however, the L1 comes out of reset in an undefined state, so
25 * the clean + invalidate performed by v7_flush_dcache_all causes a bunch
26 * of cache lines with uninitialized data and uninitialized tags to get
27 * written out to memory, which does really unpleasant things to the main
28 * processor. We fix this by performing an invalidate, rather than a
29 * clean + invalidate, before jumping into the kernel.
31 * This function is cloned from arch/arm/mach-tegra/headsmp.S, and needs
32 * to be called for both secondary cores startup and primary core resume
35 ENTRY(v7_invalidate_l1)
37 mcr p15, 2, r0, c0, c0, 0
38 mrc p15, 1, r0, c0, c0, 0
41 and r2, r1, r0, lsr #13
45 and r3, r1, r0, lsr #3 @ NumWays - 1
46 add r2, r2, #1 @ NumSets
49 add r0, r0, #4 @ SetShift
52 add r4, r3, #1 @ NumWays
53 1: sub r2, r2, #1 @ NumSets--
54 mov r3, r4 @ Temp = NumWays
55 2: subs r3, r3, #1 @ Temp--
58 orr r5, r5, r6 @ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
59 mcr p15, 0, r5, c7, c6, 2
66 ENDPROC(v7_invalidate_l1)
69 * v7_flush_icache_all()
71 * Flush the whole I-cache.
76 ENTRY(v7_flush_icache_all)
78 ALT_SMP(mcr p15, 0, r0, c7, c1, 0) @ invalidate I-cache inner shareable
79 ALT_UP(mcr p15, 0, r0, c7, c5, 0) @ I+BTB cache invalidate
81 ENDPROC(v7_flush_icache_all)
84 * v7_flush_dcache_louis()
86 * Flush the D-cache up to the Level of Unification Inner Shareable
88 * Corrupted registers: r0-r7, r9-r11 (r6 only in Thumb mode)
91 ENTRY(v7_flush_dcache_louis)
92 dmb @ ensure ordering with previous memory accesses
93 mrc p15, 1, r0, c0, c0, 1 @ read clidr, r0 = clidr
94 ALT_SMP(mov r3, r0, lsr #20) @ move LoUIS into position
95 ALT_UP( mov r3, r0, lsr #26) @ move LoUU into position
96 ands r3, r3, #7 << 1 @ extract LoU*2 field from clidr
97 bne start_flush_levels @ LoU != 0, start flushing
98 #ifdef CONFIG_ARM_ERRATA_643719
99 ALT_SMP(mrc p15, 0, r2, c0, c0, 0) @ read main ID register
100 ALT_UP( ret lr) @ LoUU is zero, so nothing to do
101 movw r1, #:lower16:(0x410fc090 >> 4) @ ID of ARM Cortex A9 r0p?
102 movt r1, #:upper16:(0x410fc090 >> 4)
103 teq r1, r2, lsr #4 @ test for errata affected core and if so...
104 moveq r3, #1 << 1 @ fix LoUIS value
105 beq start_flush_levels @ start flushing cache levels
108 ENDPROC(v7_flush_dcache_louis)
111 * v7_flush_dcache_all()
113 * Flush the whole D-cache.
115 * Corrupted registers: r0-r7, r9-r11 (r6 only in Thumb mode)
117 * - mm - mm_struct describing address space
119 ENTRY(v7_flush_dcache_all)
120 dmb @ ensure ordering with previous memory accesses
121 mrc p15, 1, r0, c0, c0, 1 @ read clidr
122 mov r3, r0, lsr #23 @ move LoC into position
123 ands r3, r3, #7 << 1 @ extract LoC*2 from clidr
124 beq finished @ if loc is 0, then no need to clean
126 mov r10, #0 @ start clean at cache level 0
128 add r2, r10, r10, lsr #1 @ work out 3x current cache level
129 mov r1, r0, lsr r2 @ extract cache type bits from clidr
130 and r1, r1, #7 @ mask of the bits for current cache only
131 cmp r1, #2 @ see what cache we have at this level
132 blt skip @ skip if no cache, or just i-cache
133 #ifdef CONFIG_PREEMPT
134 save_and_disable_irqs_notrace r9 @ make cssr&csidr read atomic
136 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
137 isb @ isb to sych the new cssr&csidr
138 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
139 #ifdef CONFIG_PREEMPT
140 restore_irqs_notrace r9
142 and r2, r1, #7 @ extract the length of the cache lines
143 add r2, r2, #4 @ add 4 (line length offset)
145 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
146 clz r5, r4 @ find bit position of way size increment
148 ands r7, r7, r1, lsr #13 @ extract max number of the index size
150 mov r9, r7 @ create working copy of max index
152 ARM( orr r11, r10, r4, lsl r5 ) @ factor way and cache number into r11
153 THUMB( lsl r6, r4, r5 )
154 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
155 ARM( orr r11, r11, r9, lsl r2 ) @ factor index number into r11
156 THUMB( lsl r6, r9, r2 )
157 THUMB( orr r11, r11, r6 ) @ factor index number into r11
158 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
159 subs r9, r9, #1 @ decrement the index
161 subs r4, r4, #1 @ decrement the way
164 add r10, r10, #2 @ increment cache number
168 mov r10, #0 @ switch back to cache level 0
169 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
173 ENDPROC(v7_flush_dcache_all)
176 * v7_flush_cache_all()
178 * Flush the entire cache system.
179 * The data cache flush is now achieved using atomic clean / invalidates
180 * working outwards from L1 cache. This is done using Set/Way based cache
181 * maintenance instructions.
182 * The instruction cache can still be invalidated back to the point of
183 * unification in a single instruction.
186 ENTRY(v7_flush_kern_cache_all)
187 ARM( stmfd sp!, {r4-r5, r7, r9-r11, lr} )
188 THUMB( stmfd sp!, {r4-r7, r9-r11, lr} )
189 bl v7_flush_dcache_all
191 ALT_SMP(mcr p15, 0, r0, c7, c1, 0) @ invalidate I-cache inner shareable
192 ALT_UP(mcr p15, 0, r0, c7, c5, 0) @ I+BTB cache invalidate
193 ARM( ldmfd sp!, {r4-r5, r7, r9-r11, lr} )
194 THUMB( ldmfd sp!, {r4-r7, r9-r11, lr} )
196 ENDPROC(v7_flush_kern_cache_all)
199 * v7_flush_kern_cache_louis(void)
201 * Flush the data cache up to Level of Unification Inner Shareable.
202 * Invalidate the I-cache to the point of unification.
204 ENTRY(v7_flush_kern_cache_louis)
205 ARM( stmfd sp!, {r4-r5, r7, r9-r11, lr} )
206 THUMB( stmfd sp!, {r4-r7, r9-r11, lr} )
207 bl v7_flush_dcache_louis
209 ALT_SMP(mcr p15, 0, r0, c7, c1, 0) @ invalidate I-cache inner shareable
210 ALT_UP(mcr p15, 0, r0, c7, c5, 0) @ I+BTB cache invalidate
211 ARM( ldmfd sp!, {r4-r5, r7, r9-r11, lr} )
212 THUMB( ldmfd sp!, {r4-r7, r9-r11, lr} )
214 ENDPROC(v7_flush_kern_cache_louis)
217 * v7_flush_cache_all()
219 * Flush all TLB entries in a particular address space
221 * - mm - mm_struct describing address space
223 ENTRY(v7_flush_user_cache_all)
227 * v7_flush_cache_range(start, end, flags)
229 * Flush a range of TLB entries in the specified address space.
231 * - start - start address (may not be aligned)
232 * - end - end address (exclusive, may not be aligned)
233 * - flags - vm_area_struct flags describing address space
235 * It is assumed that:
236 * - we have a VIPT cache.
238 ENTRY(v7_flush_user_cache_range)
240 ENDPROC(v7_flush_user_cache_all)
241 ENDPROC(v7_flush_user_cache_range)
244 * v7_coherent_kern_range(start,end)
246 * Ensure that the I and D caches are coherent within specified
247 * region. This is typically used when code has been written to
248 * a memory region, and will be executed.
250 * - start - virtual start address of region
251 * - end - virtual end address of region
253 * It is assumed that:
254 * - the Icache does not read data from the write buffer
256 ENTRY(v7_coherent_kern_range)
260 * v7_coherent_user_range(start,end)
262 * Ensure that the I and D caches are coherent within specified
263 * region. This is typically used when code has been written to
264 * a memory region, and will be executed.
266 * - start - virtual start address of region
267 * - end - virtual end address of region
269 * It is assumed that:
270 * - the Icache does not read data from the write buffer
272 ENTRY(v7_coherent_user_range)
274 dcache_line_size r2, r3
277 #ifdef CONFIG_ARM_ERRATA_764369
282 USER( mcr p15, 0, r12, c7, c11, 1 ) @ clean D line to the point of unification
287 icache_line_size r2, r3
291 USER( mcr p15, 0, r12, c7, c5, 1 ) @ invalidate I line
296 ALT_SMP(mcr p15, 0, r0, c7, c1, 6) @ invalidate BTB Inner Shareable
297 ALT_UP(mcr p15, 0, r0, c7, c5, 6) @ invalidate BTB
303 * Fault handling for the cache operation above. If the virtual address in r0
304 * isn't mapped, fail with -EFAULT.
307 #ifdef CONFIG_ARM_ERRATA_775420
313 ENDPROC(v7_coherent_kern_range)
314 ENDPROC(v7_coherent_user_range)
317 * v7_flush_kern_dcache_area(void *addr, size_t size)
319 * Ensure that the data held in the page kaddr is written back
320 * to the page in question.
322 * - addr - kernel address
323 * - size - region size
325 ENTRY(v7_flush_kern_dcache_area)
326 dcache_line_size r2, r3
330 #ifdef CONFIG_ARM_ERRATA_764369
335 mcr p15, 0, r0, c7, c14, 1 @ clean & invalidate D line / unified line
341 ENDPROC(v7_flush_kern_dcache_area)
344 * v7_dma_inv_range(start,end)
346 * Invalidate the data cache within the specified region; we will
347 * be performing a DMA operation in this region and we want to
348 * purge old data in the cache.
350 * - start - virtual start address of region
351 * - end - virtual end address of region
354 dcache_line_size r2, r3
358 #ifdef CONFIG_ARM_ERRATA_764369
362 mcrne p15, 0, r0, c7, c14, 1 @ clean & invalidate D / U line
367 mcrne p15, 0, r1, c7, c14, 1 @ clean & invalidate D / U line
370 mcrlo p15, 0, r0, c7, c6, 1 @ invalidate D / U line
376 ENDPROC(v7_dma_inv_range)
379 * v7_dma_clean_range(start,end)
380 * - start - virtual start address of region
381 * - end - virtual end address of region
384 dcache_line_size r2, r3
387 #ifdef CONFIG_ARM_ERRATA_764369
392 mcr p15, 0, r0, c7, c10, 1 @ clean D / U line
398 ENDPROC(v7_dma_clean_range)
401 * v7_dma_flush_range(start,end)
402 * - start - virtual start address of region
403 * - end - virtual end address of region
405 ENTRY(v7_dma_flush_range)
406 dcache_line_size r2, r3
409 #ifdef CONFIG_ARM_ERRATA_764369
414 mcr p15, 0, r0, c7, c14, 1 @ clean & invalidate D / U line
420 ENDPROC(v7_dma_flush_range)
423 * dma_map_area(start, size, dir)
424 * - start - kernel virtual start address
425 * - size - size of region
426 * - dir - DMA direction
428 ENTRY(v7_dma_map_area)
430 teq r2, #DMA_FROM_DEVICE
433 ENDPROC(v7_dma_map_area)
436 * dma_unmap_area(start, size, dir)
437 * - start - kernel virtual start address
438 * - size - size of region
439 * - dir - DMA direction
441 ENTRY(v7_dma_unmap_area)
443 teq r2, #DMA_TO_DEVICE
446 ENDPROC(v7_dma_unmap_area)
450 @ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
451 define_cache_functions v7
453 /* The Broadcom Brahma-B15 read-ahead cache requires some modifications
454 * to the v7_cache_fns, we only override the ones we need
456 #ifndef CONFIG_CACHE_B15_RAC
457 globl_equ b15_flush_kern_cache_all, v7_flush_kern_cache_all
459 globl_equ b15_flush_icache_all, v7_flush_icache_all
460 globl_equ b15_flush_kern_cache_louis, v7_flush_kern_cache_louis
461 globl_equ b15_flush_user_cache_all, v7_flush_user_cache_all
462 globl_equ b15_flush_user_cache_range, v7_flush_user_cache_range
463 globl_equ b15_coherent_kern_range, v7_coherent_kern_range
464 globl_equ b15_coherent_user_range, v7_coherent_user_range
465 globl_equ b15_flush_kern_dcache_area, v7_flush_kern_dcache_area
467 globl_equ b15_dma_map_area, v7_dma_map_area
468 globl_equ b15_dma_unmap_area, v7_dma_unmap_area
469 globl_equ b15_dma_flush_range, v7_dma_flush_range
471 define_cache_functions b15