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#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND 20.globl icache_size 21 .data 22 .align 2 23icache_size: 24 .long 64 25 .text 26#endif 27/* 28 * The secondary kernel init calls v7_flush_dcache_all before it enables 29 * the L1; however, the L1 comes out of reset in an undefined state, so 30 * the clean + invalidate performed by v7_flush_dcache_all causes a bunch 31 * of cache lines with uninitialized data and uninitialized tags to get 32 * written out to memory, which does really unpleasant things to the main 33 * processor. We fix this by performing an invalidate, rather than a 34 * clean + invalidate, before jumping into the kernel. 35 * 36 * This function needs to be called for both secondary cores startup and 37 * primary core resume procedures. 38 */ 39ENTRY(v7_invalidate_l1) 40 mov r0, #0 41 mcr p15, 2, r0, c0, c0, 0 @ select L1 data cache in CSSELR 42 isb 43 mrc p15, 1, r0, c0, c0, 0 @ read cache geometry from CCSIDR 44 45 movw r3, #0x3ff 46 and r3, r3, r0, lsr #3 @ 'Associativity' in CCSIDR[12:3] 47 clz r1, r3 @ WayShift 48 mov r2, #1 49 mov r3, r3, lsl r1 @ NumWays-1 shifted into bits [31:...] 50 movs r1, r2, lsl r1 @ #1 shifted left by same amount 51 moveq r1, #1 @ r1 needs value > 0 even if only 1 way 52 53 and r2, r0, #0x7 54 add r2, r2, #4 @ SetShift 55 561: movw ip, #0x7fff 57 and r0, ip, r0, lsr #13 @ 'NumSets' in CCSIDR[27:13] 58 592: mov ip, r0, lsl r2 @ NumSet << SetShift 60 orr ip, ip, r3 @ Reg = (Temp<<WayShift)|(NumSets<<SetShift) 61 mcr p15, 0, ip, c7, c6, 2 62 subs r0, r0, #1 @ Set-- 63 bpl 2b 64 subs r3, r3, r1 @ Way-- 65 bcc 3f 66 mrc p15, 1, r0, c0, c0, 0 @ re-read cache geometry from CCSIDR 67 b 1b 683: dsb st 69 isb 70 ret lr 71ENDPROC(v7_invalidate_l1) 72 73/* 74 * v7_flush_icache_all() 75 * 76 * Flush the whole I-cache. 77 * 78 * Registers: 79 * r0 - set to 0 80 */ 81ENTRY(v7_flush_icache_all) 82 mov r0, #0 83 ALT_SMP(mcr p15, 0, r0, c7, c1, 0) @ invalidate I-cache inner shareable 84 ALT_UP(mcr p15, 0, r0, c7, c5, 0) @ I+BTB cache invalidate 85 ret lr 86ENDPROC(v7_flush_icache_all) 87 88 /* 89 * v7_flush_dcache_louis() 90 * 91 * Flush the D-cache up to the Level of Unification Inner Shareable 92 * 93 * Corrupted registers: r0-r7, r9-r11 (r6 only in Thumb mode) 94 */ 95 96ENTRY(v7_flush_dcache_louis) 97 dmb @ ensure ordering with previous memory accesses 98 mrc p15, 1, r0, c0, c0, 1 @ read clidr, r0 = clidr 99ALT_SMP(mov r3, r0, lsr #20) @ move LoUIS into position 100ALT_UP( mov r3, r0, lsr #26) @ move LoUU into position 101 ands r3, r3, #7 << 1 @ extract LoU*2 field from clidr 102 bne start_flush_levels @ LoU != 0, start flushing 103#ifdef CONFIG_ARM_ERRATA_643719 104ALT_SMP(mrc p15, 0, r2, c0, c0, 0) @ read main ID register 105ALT_UP( ret lr) @ LoUU is zero, so nothing to do 106 movw r1, #:lower16:(0x410fc090 >> 4) @ ID of ARM Cortex A9 r0p? 107 movt r1, #:upper16:(0x410fc090 >> 4) 108 teq r1, r2, lsr #4 @ test for errata affected core and if so... 109 moveq r3, #1 << 1 @ fix LoUIS value 110 beq start_flush_levels @ start flushing cache levels 111#endif 112 ret lr 113ENDPROC(v7_flush_dcache_louis) 114 115/* 116 * v7_flush_dcache_all() 117 * 118 * Flush the whole D-cache. 119 * 120 * Corrupted registers: r0-r7, r9-r11 (r6 only in Thumb mode) 121 * 122 * - mm - mm_struct describing address space 123 */ 124ENTRY(v7_flush_dcache_all) 125 dmb @ ensure ordering with previous memory accesses 126 mrc p15, 1, r0, c0, c0, 1 @ read clidr 127 mov r3, r0, lsr #23 @ move LoC into position 128 ands r3, r3, #7 << 1 @ extract LoC*2 from clidr 129 beq finished @ if loc is 0, then no need to clean 130start_flush_levels: 131 mov r10, #0 @ start clean at cache level 0 132flush_levels: 133 add r2, r10, r10, lsr #1 @ work out 3x current cache level 134 mov r1, r0, lsr r2 @ extract cache type bits from clidr 135 and r1, r1, #7 @ mask of the bits for current cache only 136 cmp r1, #2 @ see what cache we have at this level 137 blt skip @ skip if no cache, or just i-cache 138#ifdef CONFIG_PREEMPTION 139 save_and_disable_irqs_notrace r9 @ make cssr&csidr read atomic 140#endif 141 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr 142 isb @ isb to sych the new cssr&csidr 143 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr 144#ifdef CONFIG_PREEMPTION 145 restore_irqs_notrace r9 146#endif 147 and r2, r1, #7 @ extract the length of the cache lines 148 add r2, r2, #4 @ add 4 (line length offset) 149 movw r4, #0x3ff 150 ands r4, r4, r1, lsr #3 @ find maximum number on the way size 151 clz r5, r4 @ find bit position of way size increment 152 movw r7, #0x7fff 153 ands r7, r7, r1, lsr #13 @ extract max number of the index size 154loop1: 155 mov r9, r7 @ create working copy of max index 156loop2: 157 ARM( orr r11, r10, r4, lsl r5 ) @ factor way and cache number into r11 158 THUMB( lsl r6, r4, r5 ) 159 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11 160 ARM( orr r11, r11, r9, lsl r2 ) @ factor index number into r11 161 THUMB( lsl r6, r9, r2 ) 162 THUMB( orr r11, r11, r6 ) @ factor index number into r11 163 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way 164 subs r9, r9, #1 @ decrement the index 165 bge loop2 166 subs r4, r4, #1 @ decrement the way 167 bge loop1 168skip: 169 add r10, r10, #2 @ increment cache number 170 cmp r3, r10 171#ifdef CONFIG_ARM_ERRATA_814220 172 dsb 173#endif 174 bgt flush_levels 175finished: 176 mov r10, #0 @ switch back to cache level 0 177 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr 178 dsb st 179 isb 180 ret lr 181ENDPROC(v7_flush_dcache_all) 182 183/* 184 * v7_flush_cache_all() 185 * 186 * Flush the entire cache system. 187 * The data cache flush is now achieved using atomic clean / invalidates 188 * working outwards from L1 cache. This is done using Set/Way based cache 189 * maintenance instructions. 190 * The instruction cache can still be invalidated back to the point of 191 * unification in a single instruction. 192 * 193 */ 194ENTRY(v7_flush_kern_cache_all) 195 ARM( stmfd sp!, {r4-r5, r7, r9-r11, lr} ) 196 THUMB( stmfd sp!, {r4-r7, r9-r11, lr} ) 197 bl v7_flush_dcache_all 198 mov r0, #0 199 ALT_SMP(mcr p15, 0, r0, c7, c1, 0) @ invalidate I-cache inner shareable 200 ALT_UP(mcr p15, 0, r0, c7, c5, 0) @ I+BTB cache invalidate 201 ARM( ldmfd sp!, {r4-r5, r7, r9-r11, lr} ) 202 THUMB( ldmfd sp!, {r4-r7, r9-r11, 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 ARM( stmfd sp!, {r4-r5, r7, r9-r11, lr} ) 214 THUMB( stmfd sp!, {r4-r7, r9-r11, 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 ARM( ldmfd sp!, {r4-r5, r7, r9-r11, lr} ) 220 THUMB( ldmfd sp!, {r4-r7, r9-r11, lr} ) 221 ret lr 222ENDPROC(v7_flush_kern_cache_louis) 223 224/* 225 * v7_flush_cache_all() 226 * 227 * Flush all TLB entries in a particular address space 228 * 229 * - mm - mm_struct describing address space 230 */ 231ENTRY(v7_flush_user_cache_all) 232 /*FALLTHROUGH*/ 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 */ 246ENTRY(v7_flush_user_cache_range) 247 ret lr 248ENDPROC(v7_flush_user_cache_all) 249ENDPROC(v7_flush_user_cache_range) 250 251/* 252 * v7_coherent_kern_range(start,end) 253 * 254 * Ensure that the I and D caches are coherent within specified 255 * region. This is typically used when code has been written to 256 * a memory region, and will be executed. 257 * 258 * - start - virtual start address of region 259 * - end - virtual end address of region 260 * 261 * It is assumed that: 262 * - the Icache does not read data from the write buffer 263 */ 264ENTRY(v7_coherent_kern_range) 265 /* FALLTHROUGH */ 266 267/* 268 * v7_coherent_user_range(start,end) 269 * 270 * Ensure that the I and D caches are coherent within specified 271 * region. This is typically used when code has been written to 272 * a memory region, and will be executed. 273 * 274 * - start - virtual start address of region 275 * - end - virtual end address of region 276 * 277 * It is assumed that: 278 * - the Icache does not read data from the write buffer 279 */ 280ENTRY(v7_coherent_user_range) 281 UNWIND(.fnstart ) 282 dcache_line_size r2, r3 283 sub r3, r2, #1 284 bic r12, r0, r3 285#ifdef CONFIG_ARM_ERRATA_764369 286 ALT_SMP(W(dsb)) 287 ALT_UP(W(nop)) 288#endif 2891: 290 USER( mcr p15, 0, r12, c7, c11, 1 ) @ clean D line to the point of unification 291 add r12, r12, r2 292 cmp r12, r1 293 blo 1b 294 dsb ishst 295#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND 296 ldr r3, =icache_size 297 ldr r2, [r3, #0] 298#else 299 icache_line_size r2, r3 300#endif 301 sub r3, r2, #1 302 bic r12, r0, r3 3032: 304 USER( mcr p15, 0, r12, c7, c5, 1 ) @ invalidate I line 305 add r12, r12, r2 306 cmp r12, r1 307 blo 2b 308 mov r0, #0 309 ALT_SMP(mcr p15, 0, r0, c7, c1, 6) @ invalidate BTB Inner Shareable 310 ALT_UP(mcr p15, 0, r0, c7, c5, 6) @ invalidate BTB 311 dsb ishst 312 isb 313 ret lr 314 315/* 316 * Fault handling for the cache operation above. If the virtual address in r0 317 * isn't mapped, fail with -EFAULT. 318 */ 3199001: 320#ifdef CONFIG_ARM_ERRATA_775420 321 dsb 322#endif 323 mov r0, #-EFAULT 324 ret lr 325 UNWIND(.fnend ) 326ENDPROC(v7_coherent_kern_range) 327ENDPROC(v7_coherent_user_range) 328 329/* 330 * v7_flush_kern_dcache_area(void *addr, size_t size) 331 * 332 * Ensure that the data held in the page kaddr is written back 333 * to the page in question. 334 * 335 * - addr - kernel address 336 * - size - region size 337 */ 338ENTRY(v7_flush_kern_dcache_area) 339 dcache_line_size r2, r3 340 add r1, r0, r1 341 sub r3, r2, #1 342 bic r0, r0, r3 343#ifdef CONFIG_ARM_ERRATA_764369 344 ALT_SMP(W(dsb)) 345 ALT_UP(W(nop)) 346#endif 3471: 348 mcr p15, 0, r0, c7, c14, 1 @ 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(v7_flush_kern_dcache_area) 355 356/* 357 * v7_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 */ 366v7_dma_inv_range: 367 dcache_line_size r2, r3 368 sub r3, r2, #1 369 tst r0, r3 370 bic r0, r0, r3 371#ifdef CONFIG_ARM_ERRATA_764369 372 ALT_SMP(W(dsb)) 373 ALT_UP(W(nop)) 374#endif 375 mcrne p15, 0, r0, c7, c14, 1 @ clean & invalidate D / U line 376 addne r0, r0, r2 377 378 tst r1, r3 379 bic r1, r1, r3 380 mcrne p15, 0, r1, c7, c14, 1 @ clean & invalidate D / U line 381 cmp r0, r1 3821: 383 mcrlo p15, 0, r0, c7, c6, 1 @ invalidate D / U line 384 addlo r0, r0, r2 385 cmplo r0, r1 386 blo 1b 387 dsb st 388 ret lr 389ENDPROC(v7_dma_inv_range) 390 391/* 392 * v7_dma_clean_range(start,end) 393 * - start - virtual start address of region 394 * - end - virtual end address of region 395 */ 396v7_dma_clean_range: 397 dcache_line_size r2, r3 398 sub r3, r2, #1 399 bic r0, r0, r3 400#ifdef CONFIG_ARM_ERRATA_764369 401 ALT_SMP(W(dsb)) 402 ALT_UP(W(nop)) 403#endif 4041: 405 mcr p15, 0, r0, c7, c10, 1 @ clean D / U line 406 add r0, r0, r2 407 cmp r0, r1 408 blo 1b 409 dsb st 410 ret lr 411ENDPROC(v7_dma_clean_range) 412 413/* 414 * v7_dma_flush_range(start,end) 415 * - start - virtual start address of region 416 * - end - virtual end address of region 417 */ 418ENTRY(v7_dma_flush_range) 419 dcache_line_size r2, r3 420 sub r3, r2, #1 421 bic r0, r0, r3 422#ifdef CONFIG_ARM_ERRATA_764369 423 ALT_SMP(W(dsb)) 424 ALT_UP(W(nop)) 425#endif 4261: 427 mcr p15, 0, r0, c7, c14, 1 @ clean & invalidate D / U line 428 add r0, r0, r2 429 cmp r0, r1 430 blo 1b 431 dsb st 432 ret lr 433ENDPROC(v7_dma_flush_range) 434 435/* 436 * dma_map_area(start, size, dir) 437 * - start - kernel virtual start address 438 * - size - size of region 439 * - dir - DMA direction 440 */ 441ENTRY(v7_dma_map_area) 442 add r1, r1, r0 443 teq r2, #DMA_FROM_DEVICE 444 beq v7_dma_inv_range 445 b v7_dma_clean_range 446ENDPROC(v7_dma_map_area) 447 448/* 449 * dma_unmap_area(start, size, dir) 450 * - start - kernel virtual start address 451 * - size - size of region 452 * - dir - DMA direction 453 */ 454ENTRY(v7_dma_unmap_area) 455 add r1, r1, r0 456 teq r2, #DMA_TO_DEVICE 457 bne v7_dma_inv_range 458 ret lr 459ENDPROC(v7_dma_unmap_area) 460 461 __INITDATA 462 463 @ define struct cpu_cache_fns (see <asm/cacheflush.h> and proc-macros.S) 464 define_cache_functions v7 465 466 /* The Broadcom Brahma-B15 read-ahead cache requires some modifications 467 * to the v7_cache_fns, we only override the ones we need 468 */ 469#ifndef CONFIG_CACHE_B15_RAC 470 globl_equ b15_flush_kern_cache_all, v7_flush_kern_cache_all 471#endif 472 globl_equ b15_flush_icache_all, v7_flush_icache_all 473 globl_equ b15_flush_kern_cache_louis, v7_flush_kern_cache_louis 474 globl_equ b15_flush_user_cache_all, v7_flush_user_cache_all 475 globl_equ b15_flush_user_cache_range, v7_flush_user_cache_range 476 globl_equ b15_coherent_kern_range, v7_coherent_kern_range 477 globl_equ b15_coherent_user_range, v7_coherent_user_range 478 globl_equ b15_flush_kern_dcache_area, v7_flush_kern_dcache_area 479 480 globl_equ b15_dma_map_area, v7_dma_map_area 481 globl_equ b15_dma_unmap_area, v7_dma_unmap_area 482 globl_equ b15_dma_flush_range, v7_dma_flush_range 483 484 define_cache_functions b15 485