1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * linux/arch/arm/mm/proc-xscale.S
5 * Author: Nicolas Pitre
6 * Created: November 2000
7 * Copyright: (C) 2000, 2001 MontaVista Software Inc.
9 * MMU functions for the Intel XScale CPUs
12 * some contributions by Brett Gaines <brett.w.gaines@intel.com>
13 * Copyright 2001 by Intel Corp.
16 * Completely revisited, many important fixes
17 * Nicolas Pitre <nico@fluxnic.net>
20 #include <linux/linkage.h>
21 #include <linux/init.h>
22 #include <linux/pgtable.h>
23 #include <asm/assembler.h>
24 #include <asm/hwcap.h>
25 #include <asm/pgtable-hwdef.h>
27 #include <asm/ptrace.h>
28 #include "proc-macros.S"
31 * This is the maximum size of an area which will be flushed. If the area
32 * is larger than this, then we flush the whole cache
34 #define MAX_AREA_SIZE 32768
37 * the cache line size of the I and D cache
39 #define CACHELINESIZE 32
42 * the size of the data cache
44 #define CACHESIZE 32768
47 * Virtual address used to allocate the cache when flushed
49 * This must be an address range which is _never_ used. It should
50 * apparently have a mapping in the corresponding page table for
51 * compatibility with future CPUs that _could_ require it. For instance we
54 * This must be aligned on a 2*CACHESIZE boundary. The code selects one of
55 * the 2 areas in alternance each time the clean_d_cache macro is used.
56 * Without this the XScale core exhibits cache eviction problems and no one
59 * Reminder: the vector table is located at 0xffff0000-0xffff0fff.
61 #define CLEAN_ADDR 0xfffe0000
64 * This macro is used to wait for a CP15 write and is needed
65 * when we have to ensure that the last operation to the co-pro
66 * was completed before continuing with operation.
69 mrc p15, 0, \rd, c2, c0, 0 @ arbitrary read of cp15
70 mov \rd, \rd @ wait for completion
71 sub pc, pc, #4 @ flush instruction pipeline
74 .macro cpwait_ret, lr, rd
75 mrc p15, 0, \rd, c2, c0, 0 @ arbitrary read of cp15
76 sub pc, \lr, \rd, LSR #32 @ wait for completion and
77 @ flush instruction pipeline
81 * This macro cleans the entire dcache using line allocate.
82 * The main loop has been unrolled to reduce loop overhead.
83 * rd and rs are two scratch registers.
85 .macro clean_d_cache, rd, rs
88 eor \rd, \rd, #CACHESIZE
90 add \rs, \rd, #CACHESIZE
91 1: mcr p15, 0, \rd, c7, c2, 5 @ allocate D cache line
92 add \rd, \rd, #CACHELINESIZE
93 mcr p15, 0, \rd, c7, c2, 5 @ allocate D cache line
94 add \rd, \rd, #CACHELINESIZE
95 mcr p15, 0, \rd, c7, c2, 5 @ allocate D cache line
96 add \rd, \rd, #CACHELINESIZE
97 mcr p15, 0, \rd, c7, c2, 5 @ allocate D cache line
98 add \rd, \rd, #CACHELINESIZE
105 clean_addr: .word CLEAN_ADDR
110 * cpu_xscale_proc_init()
112 * Nothing too exciting at the moment
114 ENTRY(cpu_xscale_proc_init)
115 @ enable write buffer coalescing. Some bootloader disable it
116 mrc p15, 0, r1, c1, c0, 1
118 mcr p15, 0, r1, c1, c0, 1
122 * cpu_xscale_proc_fin()
124 ENTRY(cpu_xscale_proc_fin)
125 mrc p15, 0, r0, c1, c0, 0 @ ctrl register
126 bic r0, r0, #0x1800 @ ...IZ...........
127 bic r0, r0, #0x0006 @ .............CA.
128 mcr p15, 0, r0, c1, c0, 0 @ disable caches
132 * cpu_xscale_reset(loc)
134 * Perform a soft reset of the system. Put the CPU into the
135 * same state as it would be if it had been reset, and branch
136 * to what would be the reset vector.
138 * loc: location to jump to for soft reset
140 * Beware PXA270 erratum E7.
143 .pushsection .idmap.text, "ax"
144 ENTRY(cpu_xscale_reset)
145 mov r1, #PSR_F_BIT|PSR_I_BIT|SVC_MODE
146 msr cpsr_c, r1 @ reset CPSR
147 mcr p15, 0, r1, c10, c4, 1 @ unlock I-TLB
148 mcr p15, 0, r1, c8, c5, 0 @ invalidate I-TLB
149 mrc p15, 0, r1, c1, c0, 0 @ ctrl register
150 bic r1, r1, #0x0086 @ ........B....CA.
151 bic r1, r1, #0x3900 @ ..VIZ..S........
152 sub pc, pc, #4 @ flush pipeline
153 @ *** cache line aligned ***
154 mcr p15, 0, r1, c1, c0, 0 @ ctrl register
155 bic r1, r1, #0x0001 @ ...............M
156 mcr p15, 0, ip, c7, c7, 0 @ invalidate I,D caches & BTB
157 mcr p15, 0, r1, c1, c0, 0 @ ctrl register
158 @ CAUTION: MMU turned off from this point. We count on the pipeline
159 @ already containing those two last instructions to survive.
160 mcr p15, 0, ip, c8, c7, 0 @ invalidate I & D TLBs
162 ENDPROC(cpu_xscale_reset)
166 * cpu_xscale_do_idle()
168 * Cause the processor to idle
170 * For now we do nothing but go to idle mode for every case
172 * XScale supports clock switching, but using idle mode support
173 * allows external hardware to react to system state changes.
177 ENTRY(cpu_xscale_do_idle)
179 mcr p14, 0, r0, c7, c0, 0 @ Go to IDLE
182 /* ================================= CACHE ================================ */
187 * Unconditionally clean and invalidate the entire icache.
189 ENTRY(xscale_flush_icache_all)
191 mcr p15, 0, r0, c7, c5, 0 @ invalidate I cache
193 ENDPROC(xscale_flush_icache_all)
196 * flush_user_cache_all()
198 * Invalidate all cache entries in a particular address
201 ENTRY(xscale_flush_user_cache_all)
205 * flush_kern_cache_all()
207 * Clean and invalidate the entire cache.
209 ENTRY(xscale_flush_kern_cache_all)
215 mcrne p15, 0, ip, c7, c5, 0 @ Invalidate I cache & BTB
216 mcrne p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer
220 * flush_user_cache_range(start, end, vm_flags)
222 * Invalidate a range of cache entries in the specified
225 * - start - start address (may not be aligned)
226 * - end - end address (exclusive, may not be aligned)
227 * - vma - vma_area_struct describing address space
230 ENTRY(xscale_flush_user_cache_range)
232 sub r3, r1, r0 @ calculate total size
233 cmp r3, #MAX_AREA_SIZE
234 bhs __flush_whole_cache
237 mcrne p15, 0, r0, c7, c5, 1 @ Invalidate I cache line
238 mcr p15, 0, r0, c7, c10, 1 @ Clean D cache line
239 mcr p15, 0, r0, c7, c6, 1 @ Invalidate D cache line
240 add r0, r0, #CACHELINESIZE
244 mcrne p15, 0, ip, c7, c5, 6 @ Invalidate BTB
245 mcrne p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer
249 * coherent_kern_range(start, end)
251 * Ensure coherency between the Icache and the Dcache in the
252 * region described by start. If you have non-snooping
253 * Harvard caches, you need to implement this function.
255 * - start - virtual start address
256 * - end - virtual end address
258 * Note: single I-cache line invalidation isn't used here since
259 * it also trashes the mini I-cache used by JTAG debuggers.
261 ENTRY(xscale_coherent_kern_range)
262 bic r0, r0, #CACHELINESIZE - 1
263 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
264 add r0, r0, #CACHELINESIZE
268 mcr p15, 0, r0, c7, c5, 0 @ Invalidate I cache & BTB
269 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
273 * coherent_user_range(start, end)
275 * Ensure coherency between the Icache and the Dcache in the
276 * region described by start. If you have non-snooping
277 * Harvard caches, you need to implement this function.
279 * - start - virtual start address
280 * - end - virtual end address
282 ENTRY(xscale_coherent_user_range)
283 bic r0, r0, #CACHELINESIZE - 1
284 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
285 mcr p15, 0, r0, c7, c5, 1 @ Invalidate I cache entry
286 add r0, r0, #CACHELINESIZE
290 mcr p15, 0, r0, c7, c5, 6 @ Invalidate BTB
291 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
295 * flush_kern_dcache_area(void *addr, size_t size)
297 * Ensure no D cache aliasing occurs, either with itself or
300 * - addr - kernel address
301 * - size - region size
303 ENTRY(xscale_flush_kern_dcache_area)
305 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
306 mcr p15, 0, r0, c7, c6, 1 @ invalidate D entry
307 add r0, r0, #CACHELINESIZE
311 mcr p15, 0, r0, c7, c5, 0 @ Invalidate I cache & BTB
312 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
316 * dma_inv_range(start, end)
318 * Invalidate (discard) the specified virtual address range.
319 * May not write back any entries. If 'start' or 'end'
320 * are not cache line aligned, those lines must be written
323 * - start - virtual start address
324 * - end - virtual end address
326 xscale_dma_inv_range:
327 tst r0, #CACHELINESIZE - 1
328 bic r0, r0, #CACHELINESIZE - 1
329 mcrne p15, 0, r0, c7, c10, 1 @ clean D entry
330 tst r1, #CACHELINESIZE - 1
331 mcrne p15, 0, r1, c7, c10, 1 @ clean D entry
332 1: mcr p15, 0, r0, c7, c6, 1 @ invalidate D entry
333 add r0, r0, #CACHELINESIZE
336 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
340 * dma_clean_range(start, end)
342 * Clean the specified virtual address range.
344 * - start - virtual start address
345 * - end - virtual end address
347 xscale_dma_clean_range:
348 bic r0, r0, #CACHELINESIZE - 1
349 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
350 add r0, r0, #CACHELINESIZE
353 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
357 * dma_flush_range(start, end)
359 * Clean and invalidate the specified virtual address range.
361 * - start - virtual start address
362 * - end - virtual end address
364 ENTRY(xscale_dma_flush_range)
365 bic r0, r0, #CACHELINESIZE - 1
366 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
367 mcr p15, 0, r0, c7, c6, 1 @ invalidate D entry
368 add r0, r0, #CACHELINESIZE
371 mcr p15, 0, r0, c7, c10, 4 @ Drain Write (& Fill) Buffer
375 * dma_map_area(start, size, dir)
376 * - start - kernel virtual start address
377 * - size - size of region
378 * - dir - DMA direction
380 ENTRY(xscale_dma_map_area)
382 cmp r2, #DMA_TO_DEVICE
383 beq xscale_dma_clean_range
384 bcs xscale_dma_inv_range
385 b xscale_dma_flush_range
386 ENDPROC(xscale_dma_map_area)
389 * dma_map_area(start, size, dir)
390 * - start - kernel virtual start address
391 * - size - size of region
392 * - dir - DMA direction
394 ENTRY(xscale_80200_A0_A1_dma_map_area)
396 teq r2, #DMA_TO_DEVICE
397 beq xscale_dma_clean_range
398 b xscale_dma_flush_range
399 ENDPROC(xscale_80200_A0_A1_dma_map_area)
402 * dma_unmap_area(start, size, dir)
403 * - start - kernel virtual start address
404 * - size - size of region
405 * - dir - DMA direction
407 ENTRY(xscale_dma_unmap_area)
409 ENDPROC(xscale_dma_unmap_area)
411 .globl xscale_flush_kern_cache_louis
412 .equ xscale_flush_kern_cache_louis, xscale_flush_kern_cache_all
414 @ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
415 define_cache_functions xscale
418 * On stepping A0/A1 of the 80200, invalidating D-cache by line doesn't
419 * clear the dirty bits, which means that if we invalidate a dirty line,
420 * the dirty data can still be written back to external memory later on.
422 * The recommended workaround is to always do a clean D-cache line before
423 * doing an invalidate D-cache line, so on the affected processors,
424 * dma_inv_range() is implemented as dma_flush_range().
426 * See erratum #25 of "Intel 80200 Processor Specification Update",
427 * revision January 22, 2003, available at:
428 * http://www.intel.com/design/iio/specupdt/273415.htm
430 .macro a0_alias basename
431 .globl xscale_80200_A0_A1_\basename
432 .type xscale_80200_A0_A1_\basename , %function
433 .equ xscale_80200_A0_A1_\basename , xscale_\basename
437 * Most of the cache functions are unchanged for these processor revisions.
438 * Export suitable alias symbols for the unchanged functions:
440 a0_alias flush_icache_all
441 a0_alias flush_user_cache_all
442 a0_alias flush_kern_cache_all
443 a0_alias flush_kern_cache_louis
444 a0_alias flush_user_cache_range
445 a0_alias coherent_kern_range
446 a0_alias coherent_user_range
447 a0_alias flush_kern_dcache_area
448 a0_alias dma_flush_range
449 a0_alias dma_unmap_area
451 @ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S)
452 define_cache_functions xscale_80200_A0_A1
454 ENTRY(cpu_xscale_dcache_clean_area)
455 1: mcr p15, 0, r0, c7, c10, 1 @ clean D entry
456 add r0, r0, #CACHELINESIZE
457 subs r1, r1, #CACHELINESIZE
461 /* =============================== PageTable ============================== */
464 * cpu_xscale_switch_mm(pgd)
466 * Set the translation base pointer to be as described by pgd.
468 * pgd: new page tables
471 ENTRY(cpu_xscale_switch_mm)
473 mcr p15, 0, ip, c7, c5, 0 @ Invalidate I cache & BTB
474 mcr p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer
475 mcr p15, 0, r0, c2, c0, 0 @ load page table pointer
476 mcr p15, 0, ip, c8, c7, 0 @ invalidate I & D TLBs
480 * cpu_xscale_set_pte_ext(ptep, pte, ext)
482 * Set a PTE and flush it out
484 * Errata 40: must set memory to write-through for user read-only pages.
487 .long 0x00 @ L_PTE_MT_UNCACHED
488 .long PTE_BUFFERABLE @ L_PTE_MT_BUFFERABLE
489 .long PTE_CACHEABLE @ L_PTE_MT_WRITETHROUGH
490 .long PTE_CACHEABLE | PTE_BUFFERABLE @ L_PTE_MT_WRITEBACK
491 .long PTE_EXT_TEX(1) | PTE_BUFFERABLE @ L_PTE_MT_DEV_SHARED
493 .long PTE_EXT_TEX(1) | PTE_CACHEABLE @ L_PTE_MT_MINICACHE
494 .long PTE_EXT_TEX(1) | PTE_CACHEABLE | PTE_BUFFERABLE @ L_PTE_MT_WRITEALLOC
496 .long PTE_BUFFERABLE @ L_PTE_MT_DEV_WC
498 .long PTE_CACHEABLE | PTE_BUFFERABLE @ L_PTE_MT_DEV_CACHED
499 .long 0x00 @ L_PTE_MT_DEV_NONSHARED
505 ENTRY(cpu_xscale_set_pte_ext)
506 xscale_set_pte_ext_prologue
509 @ Erratum 40: must set memory to write-through for user read-only pages
511 and ip, r1, #(L_PTE_MT_MASK | L_PTE_USER | L_PTE_RDONLY) & ~(4 << 2)
512 teq ip, #L_PTE_MT_WRITEBACK | L_PTE_USER | L_PTE_RDONLY
514 moveq r1, #L_PTE_MT_WRITETHROUGH
515 and r1, r1, #L_PTE_MT_MASK
516 adr ip, cpu_xscale_mt_table
521 xscale_set_pte_ext_epilogue
527 .globl cpu_xscale_suspend_size
528 .equ cpu_xscale_suspend_size, 4 * 6
529 #ifdef CONFIG_ARM_CPU_SUSPEND
530 ENTRY(cpu_xscale_do_suspend)
531 stmfd sp!, {r4 - r9, lr}
532 mrc p14, 0, r4, c6, c0, 0 @ clock configuration, for turbo mode
533 mrc p15, 0, r5, c15, c1, 0 @ CP access reg
534 mrc p15, 0, r6, c13, c0, 0 @ PID
535 mrc p15, 0, r7, c3, c0, 0 @ domain ID
536 mrc p15, 0, r8, c1, c0, 1 @ auxiliary control reg
537 mrc p15, 0, r9, c1, c0, 0 @ control reg
538 bic r4, r4, #2 @ clear frequency change bit
539 stmia r0, {r4 - r9} @ store cp regs
540 ldmfd sp!, {r4 - r9, pc}
541 ENDPROC(cpu_xscale_do_suspend)
543 ENTRY(cpu_xscale_do_resume)
544 ldmia r0, {r4 - r9} @ load cp regs
546 mcr p15, 0, ip, c8, c7, 0 @ invalidate I & D TLBs
547 mcr p15, 0, ip, c7, c7, 0 @ invalidate I & D caches, BTB
548 mcr p14, 0, r4, c6, c0, 0 @ clock configuration, turbo mode.
549 mcr p15, 0, r5, c15, c1, 0 @ CP access reg
550 mcr p15, 0, r6, c13, c0, 0 @ PID
551 mcr p15, 0, r7, c3, c0, 0 @ domain ID
552 mcr p15, 0, r1, c2, c0, 0 @ translation table base addr
553 mcr p15, 0, r8, c1, c0, 1 @ auxiliary control reg
554 mov r0, r9 @ control register
556 ENDPROC(cpu_xscale_do_resume)
559 .type __xscale_setup, #function
561 mcr p15, 0, ip, c7, c7, 0 @ invalidate I, D caches & BTB
562 mcr p15, 0, ip, c7, c10, 4 @ Drain Write (& Fill) Buffer
563 mcr p15, 0, ip, c8, c7, 0 @ invalidate I, D TLBs
564 mov r0, #1 << 6 @ cp6 for IOP3xx and Bulverde
565 orr r0, r0, #1 << 13 @ Its undefined whether this
566 mcr p15, 0, r0, c15, c1, 0 @ affects USR or SVC modes
570 mrc p15, 0, r0, c1, c0, 0 @ get control register
574 .size __xscale_setup, . - __xscale_setup
578 * .RVI ZFRS BLDP WCAM
579 * ..11 1.01 .... .101
582 .type xscale_crval, #object
584 crval clear=0x00003b07, mmuset=0x00003905, ucset=0x00001900
588 @ define struct processor (see <asm/proc-fns.h> and proc-macros.S)
589 define_processor_functions xscale, dabort=v5t_early_abort, pabort=legacy_pabort, suspend=1
593 string cpu_arch_name, "armv5te"
594 string cpu_elf_name, "v5"
596 string cpu_80200_A0_A1_name, "XScale-80200 A0/A1"
597 string cpu_80200_name, "XScale-80200"
598 string cpu_80219_name, "XScale-80219"
599 string cpu_8032x_name, "XScale-IOP8032x Family"
600 string cpu_8033x_name, "XScale-IOP8033x Family"
601 string cpu_pxa250_name, "XScale-PXA250"
602 string cpu_pxa210_name, "XScale-PXA210"
603 string cpu_ixp42x_name, "XScale-IXP42x Family"
604 string cpu_ixp43x_name, "XScale-IXP43x Family"
605 string cpu_ixp46x_name, "XScale-IXP46x Family"
606 string cpu_ixp2400_name, "XScale-IXP2400"
607 string cpu_ixp2800_name, "XScale-IXP2800"
608 string cpu_pxa255_name, "XScale-PXA255"
609 string cpu_pxa270_name, "XScale-PXA270"
613 .section ".proc.info.init", "a"
615 .macro xscale_proc_info name:req, cpu_val:req, cpu_mask:req, cpu_name:req, cache
616 .type __\name\()_proc_info,#object
617 __\name\()_proc_info:
620 .long PMD_TYPE_SECT | \
621 PMD_SECT_BUFFERABLE | \
622 PMD_SECT_CACHEABLE | \
623 PMD_SECT_AP_WRITE | \
625 .long PMD_TYPE_SECT | \
626 PMD_SECT_AP_WRITE | \
628 initfn __xscale_setup, __\name\()_proc_info
631 .long HWCAP_SWP|HWCAP_HALF|HWCAP_THUMB|HWCAP_FAST_MULT|HWCAP_EDSP
633 .long xscale_processor_functions
635 .long xscale_mc_user_fns
637 .long xscale_cache_fns
641 .size __\name\()_proc_info, . - __\name\()_proc_info
644 xscale_proc_info 80200_A0_A1, 0x69052000, 0xfffffffe, cpu_80200_name, \
645 cache=xscale_80200_A0_A1_cache_fns
646 xscale_proc_info 80200, 0x69052000, 0xfffffff0, cpu_80200_name
647 xscale_proc_info 80219, 0x69052e20, 0xffffffe0, cpu_80219_name
648 xscale_proc_info 8032x, 0x69052420, 0xfffff7e0, cpu_8032x_name
649 xscale_proc_info 8033x, 0x69054010, 0xfffffd30, cpu_8033x_name
650 xscale_proc_info pxa250, 0x69052100, 0xfffff7f0, cpu_pxa250_name
651 xscale_proc_info pxa210, 0x69052120, 0xfffff3f0, cpu_pxa210_name
652 xscale_proc_info ixp2400, 0x69054190, 0xfffffff0, cpu_ixp2400_name
653 xscale_proc_info ixp2800, 0x690541a0, 0xfffffff0, cpu_ixp2800_name
654 xscale_proc_info ixp42x, 0x690541c0, 0xffffffc0, cpu_ixp42x_name
655 xscale_proc_info ixp43x, 0x69054040, 0xfffffff0, cpu_ixp43x_name
656 xscale_proc_info ixp46x, 0x69054200, 0xffffff00, cpu_ixp46x_name
657 xscale_proc_info pxa255, 0x69052d00, 0xfffffff0, cpu_pxa255_name
658 xscale_proc_info pxa270, 0x69054110, 0xfffffff0, cpu_pxa270_name