1/* 2 * linux/arch/arm/kernel/head.S 3 * 4 * Copyright (C) 1994-2002 Russell King 5 * Copyright (c) 2003 ARM Limited 6 * All Rights Reserved 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License version 2 as 10 * published by the Free Software Foundation. 11 * 12 * Kernel startup code for all 32-bit CPUs 13 */ 14#include <linux/linkage.h> 15#include <linux/init.h> 16 17#include <asm/assembler.h> 18#include <asm/cp15.h> 19#include <asm/domain.h> 20#include <asm/ptrace.h> 21#include <asm/asm-offsets.h> 22#include <asm/memory.h> 23#include <asm/thread_info.h> 24#include <asm/pgtable.h> 25 26#if defined(CONFIG_DEBUG_LL) && !defined(CONFIG_DEBUG_SEMIHOSTING) 27#include CONFIG_DEBUG_LL_INCLUDE 28#endif 29 30/* 31 * swapper_pg_dir is the virtual address of the initial page table. 32 * We place the page tables 16K below KERNEL_RAM_VADDR. Therefore, we must 33 * make sure that KERNEL_RAM_VADDR is correctly set. Currently, we expect 34 * the least significant 16 bits to be 0x8000, but we could probably 35 * relax this restriction to KERNEL_RAM_VADDR >= PAGE_OFFSET + 0x4000. 36 */ 37#define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET) 38#if (KERNEL_RAM_VADDR & 0xffff) != 0x8000 39#error KERNEL_RAM_VADDR must start at 0xXXXX8000 40#endif 41 42#ifdef CONFIG_ARM_LPAE 43 /* LPAE requires an additional page for the PGD */ 44#define PG_DIR_SIZE 0x5000 45#define PMD_ORDER 3 46#else 47#define PG_DIR_SIZE 0x4000 48#define PMD_ORDER 2 49#endif 50 51 .globl swapper_pg_dir 52 .equ swapper_pg_dir, KERNEL_RAM_VADDR - PG_DIR_SIZE 53 54 .macro pgtbl, rd, phys 55 add \rd, \phys, #TEXT_OFFSET - PG_DIR_SIZE 56 .endm 57 58/* 59 * Kernel startup entry point. 60 * --------------------------- 61 * 62 * This is normally called from the decompressor code. The requirements 63 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0, 64 * r1 = machine nr, r2 = atags or dtb pointer. 65 * 66 * This code is mostly position independent, so if you link the kernel at 67 * 0xc0008000, you call this at __pa(0xc0008000). 68 * 69 * See linux/arch/arm/tools/mach-types for the complete list of machine 70 * numbers for r1. 71 * 72 * We're trying to keep crap to a minimum; DO NOT add any machine specific 73 * crap here - that's what the boot loader (or in extreme, well justified 74 * circumstances, zImage) is for. 75 */ 76 .arm 77 78 __HEAD 79ENTRY(stext) 80 81 THUMB( adr r9, BSYM(1f) ) @ Kernel is always entered in ARM. 82 THUMB( bx r9 ) @ If this is a Thumb-2 kernel, 83 THUMB( .thumb ) @ switch to Thumb now. 84 THUMB(1: ) 85 86#ifdef CONFIG_ARM_VIRT_EXT 87 bl __hyp_stub_install 88#endif 89 @ ensure svc mode and all interrupts masked 90 safe_svcmode_maskall r9 91 92 mrc p15, 0, r9, c0, c0 @ get processor id 93 bl __lookup_processor_type @ r5=procinfo r9=cpuid 94 movs r10, r5 @ invalid processor (r5=0)? 95 THUMB( it eq ) @ force fixup-able long branch encoding 96 beq __error_p @ yes, error 'p' 97 98#ifdef CONFIG_ARM_LPAE 99 mrc p15, 0, r3, c0, c1, 4 @ read ID_MMFR0 100 and r3, r3, #0xf @ extract VMSA support 101 cmp r3, #5 @ long-descriptor translation table format? 102 THUMB( it lo ) @ force fixup-able long branch encoding 103 blo __error_p @ only classic page table format 104#endif 105 106#ifndef CONFIG_XIP_KERNEL 107 adr r3, 2f 108 ldmia r3, {r4, r8} 109 sub r4, r3, r4 @ (PHYS_OFFSET - PAGE_OFFSET) 110 add r8, r8, r4 @ PHYS_OFFSET 111#else 112 ldr r8, =PHYS_OFFSET @ always constant in this case 113#endif 114 115 /* 116 * r1 = machine no, r2 = atags or dtb, 117 * r8 = phys_offset, r9 = cpuid, r10 = procinfo 118 */ 119 bl __vet_atags 120#ifdef CONFIG_SMP_ON_UP 121 bl __fixup_smp 122#endif 123#ifdef CONFIG_ARM_PATCH_PHYS_VIRT 124 bl __fixup_pv_table 125#endif 126 bl __create_page_tables 127 128 /* 129 * The following calls CPU specific code in a position independent 130 * manner. See arch/arm/mm/proc-*.S for details. r10 = base of 131 * xxx_proc_info structure selected by __lookup_processor_type 132 * above. On return, the CPU will be ready for the MMU to be 133 * turned on, and r0 will hold the CPU control register value. 134 */ 135 ldr r13, =__mmap_switched @ address to jump to after 136 @ mmu has been enabled 137 adr lr, BSYM(1f) @ return (PIC) address 138 mov r8, r4 @ set TTBR1 to swapper_pg_dir 139 ARM( add pc, r10, #PROCINFO_INITFUNC ) 140 THUMB( add r12, r10, #PROCINFO_INITFUNC ) 141 THUMB( mov pc, r12 ) 1421: b __enable_mmu 143ENDPROC(stext) 144 .ltorg 145#ifndef CONFIG_XIP_KERNEL 1462: .long . 147 .long PAGE_OFFSET 148#endif 149 150/* 151 * Setup the initial page tables. We only setup the barest 152 * amount which are required to get the kernel running, which 153 * generally means mapping in the kernel code. 154 * 155 * r8 = phys_offset, r9 = cpuid, r10 = procinfo 156 * 157 * Returns: 158 * r0, r3, r5-r7 corrupted 159 * r4 = page table (see ARCH_PGD_SHIFT in asm/memory.h) 160 */ 161__create_page_tables: 162 pgtbl r4, r8 @ page table address 163 164 /* 165 * Clear the swapper page table 166 */ 167 mov r0, r4 168 mov r3, #0 169 add r6, r0, #PG_DIR_SIZE 1701: str r3, [r0], #4 171 str r3, [r0], #4 172 str r3, [r0], #4 173 str r3, [r0], #4 174 teq r0, r6 175 bne 1b 176 177#ifdef CONFIG_ARM_LPAE 178 /* 179 * Build the PGD table (first level) to point to the PMD table. A PGD 180 * entry is 64-bit wide. 181 */ 182 mov r0, r4 183 add r3, r4, #0x1000 @ first PMD table address 184 orr r3, r3, #3 @ PGD block type 185 mov r6, #4 @ PTRS_PER_PGD 186 mov r7, #1 << (55 - 32) @ L_PGD_SWAPPER 1871: 188#ifdef CONFIG_CPU_ENDIAN_BE8 189 str r7, [r0], #4 @ set top PGD entry bits 190 str r3, [r0], #4 @ set bottom PGD entry bits 191#else 192 str r3, [r0], #4 @ set bottom PGD entry bits 193 str r7, [r0], #4 @ set top PGD entry bits 194#endif 195 add r3, r3, #0x1000 @ next PMD table 196 subs r6, r6, #1 197 bne 1b 198 199 add r4, r4, #0x1000 @ point to the PMD tables 200#ifdef CONFIG_CPU_ENDIAN_BE8 201 add r4, r4, #4 @ we only write the bottom word 202#endif 203#endif 204 205 ldr r7, [r10, #PROCINFO_MM_MMUFLAGS] @ mm_mmuflags 206 207 /* 208 * Create identity mapping to cater for __enable_mmu. 209 * This identity mapping will be removed by paging_init(). 210 */ 211 adr r0, __turn_mmu_on_loc 212 ldmia r0, {r3, r5, r6} 213 sub r0, r0, r3 @ virt->phys offset 214 add r5, r5, r0 @ phys __turn_mmu_on 215 add r6, r6, r0 @ phys __turn_mmu_on_end 216 mov r5, r5, lsr #SECTION_SHIFT 217 mov r6, r6, lsr #SECTION_SHIFT 218 2191: orr r3, r7, r5, lsl #SECTION_SHIFT @ flags + kernel base 220 str r3, [r4, r5, lsl #PMD_ORDER] @ identity mapping 221 cmp r5, r6 222 addlo r5, r5, #1 @ next section 223 blo 1b 224 225 /* 226 * Map our RAM from the start to the end of the kernel .bss section. 227 */ 228 add r0, r4, #PAGE_OFFSET >> (SECTION_SHIFT - PMD_ORDER) 229 ldr r6, =(_end - 1) 230 orr r3, r8, r7 231 add r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER) 2321: str r3, [r0], #1 << PMD_ORDER 233 add r3, r3, #1 << SECTION_SHIFT 234 cmp r0, r6 235 bls 1b 236 237#ifdef CONFIG_XIP_KERNEL 238 /* 239 * Map the kernel image separately as it is not located in RAM. 240 */ 241#define XIP_START XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR) 242 mov r3, pc 243 mov r3, r3, lsr #SECTION_SHIFT 244 orr r3, r7, r3, lsl #SECTION_SHIFT 245 add r0, r4, #(XIP_START & 0xff000000) >> (SECTION_SHIFT - PMD_ORDER) 246 str r3, [r0, #((XIP_START & 0x00f00000) >> SECTION_SHIFT) << PMD_ORDER]! 247 ldr r6, =(_edata_loc - 1) 248 add r0, r0, #1 << PMD_ORDER 249 add r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER) 2501: cmp r0, r6 251 add r3, r3, #1 << SECTION_SHIFT 252 strls r3, [r0], #1 << PMD_ORDER 253 bls 1b 254#endif 255 256 /* 257 * Then map boot params address in r2 if specified. 258 * We map 2 sections in case the ATAGs/DTB crosses a section boundary. 259 */ 260 mov r0, r2, lsr #SECTION_SHIFT 261 movs r0, r0, lsl #SECTION_SHIFT 262 subne r3, r0, r8 263 addne r3, r3, #PAGE_OFFSET 264 addne r3, r4, r3, lsr #(SECTION_SHIFT - PMD_ORDER) 265 orrne r6, r7, r0 266 strne r6, [r3], #1 << PMD_ORDER 267 addne r6, r6, #1 << SECTION_SHIFT 268 strne r6, [r3] 269 270#if defined(CONFIG_ARM_LPAE) && defined(CONFIG_CPU_ENDIAN_BE8) 271 sub r4, r4, #4 @ Fixup page table pointer 272 @ for 64-bit descriptors 273#endif 274 275#ifdef CONFIG_DEBUG_LL 276#if !defined(CONFIG_DEBUG_ICEDCC) && !defined(CONFIG_DEBUG_SEMIHOSTING) 277 /* 278 * Map in IO space for serial debugging. 279 * This allows debug messages to be output 280 * via a serial console before paging_init. 281 */ 282 addruart r7, r3, r0 283 284 mov r3, r3, lsr #SECTION_SHIFT 285 mov r3, r3, lsl #PMD_ORDER 286 287 add r0, r4, r3 288 mov r3, r7, lsr #SECTION_SHIFT 289 ldr r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags 290 orr r3, r7, r3, lsl #SECTION_SHIFT 291#ifdef CONFIG_ARM_LPAE 292 mov r7, #1 << (54 - 32) @ XN 293#ifdef CONFIG_CPU_ENDIAN_BE8 294 str r7, [r0], #4 295 str r3, [r0], #4 296#else 297 str r3, [r0], #4 298 str r7, [r0], #4 299#endif 300#else 301 orr r3, r3, #PMD_SECT_XN 302 str r3, [r0], #4 303#endif 304 305#else /* CONFIG_DEBUG_ICEDCC || CONFIG_DEBUG_SEMIHOSTING */ 306 /* we don't need any serial debugging mappings */ 307 ldr r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags 308#endif 309 310#if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS) 311 /* 312 * If we're using the NetWinder or CATS, we also need to map 313 * in the 16550-type serial port for the debug messages 314 */ 315 add r0, r4, #0xff000000 >> (SECTION_SHIFT - PMD_ORDER) 316 orr r3, r7, #0x7c000000 317 str r3, [r0] 318#endif 319#ifdef CONFIG_ARCH_RPC 320 /* 321 * Map in screen at 0x02000000 & SCREEN2_BASE 322 * Similar reasons here - for debug. This is 323 * only for Acorn RiscPC architectures. 324 */ 325 add r0, r4, #0x02000000 >> (SECTION_SHIFT - PMD_ORDER) 326 orr r3, r7, #0x02000000 327 str r3, [r0] 328 add r0, r4, #0xd8000000 >> (SECTION_SHIFT - PMD_ORDER) 329 str r3, [r0] 330#endif 331#endif 332#ifdef CONFIG_ARM_LPAE 333 sub r4, r4, #0x1000 @ point to the PGD table 334 mov r4, r4, lsr #ARCH_PGD_SHIFT 335#endif 336 mov pc, lr 337ENDPROC(__create_page_tables) 338 .ltorg 339 .align 340__turn_mmu_on_loc: 341 .long . 342 .long __turn_mmu_on 343 .long __turn_mmu_on_end 344 345#if defined(CONFIG_SMP) 346ENTRY(secondary_startup) 347 /* 348 * Common entry point for secondary CPUs. 349 * 350 * Ensure that we're in SVC mode, and IRQs are disabled. Lookup 351 * the processor type - there is no need to check the machine type 352 * as it has already been validated by the primary processor. 353 */ 354#ifdef CONFIG_ARM_VIRT_EXT 355 bl __hyp_stub_install_secondary 356#endif 357 safe_svcmode_maskall r9 358 359 mrc p15, 0, r9, c0, c0 @ get processor id 360 bl __lookup_processor_type 361 movs r10, r5 @ invalid processor? 362 moveq r0, #'p' @ yes, error 'p' 363 THUMB( it eq ) @ force fixup-able long branch encoding 364 beq __error_p 365 366 /* 367 * Use the page tables supplied from __cpu_up. 368 */ 369 adr r4, __secondary_data 370 ldmia r4, {r5, r7, r12} @ address to jump to after 371 sub lr, r4, r5 @ mmu has been enabled 372 ldr r4, [r7, lr] @ get secondary_data.pgdir 373 add r7, r7, #4 374 ldr r8, [r7, lr] @ get secondary_data.swapper_pg_dir 375 adr lr, BSYM(__enable_mmu) @ return address 376 mov r13, r12 @ __secondary_switched address 377 ARM( add pc, r10, #PROCINFO_INITFUNC ) @ initialise processor 378 @ (return control reg) 379 THUMB( add r12, r10, #PROCINFO_INITFUNC ) 380 THUMB( mov pc, r12 ) 381ENDPROC(secondary_startup) 382 383 /* 384 * r6 = &secondary_data 385 */ 386ENTRY(__secondary_switched) 387 ldr sp, [r7, #4] @ get secondary_data.stack 388 mov fp, #0 389 b secondary_start_kernel 390ENDPROC(__secondary_switched) 391 392 .align 393 394 .type __secondary_data, %object 395__secondary_data: 396 .long . 397 .long secondary_data 398 .long __secondary_switched 399#endif /* defined(CONFIG_SMP) */ 400 401 402 403/* 404 * Setup common bits before finally enabling the MMU. Essentially 405 * this is just loading the page table pointer and domain access 406 * registers. 407 * 408 * r0 = cp#15 control register 409 * r1 = machine ID 410 * r2 = atags or dtb pointer 411 * r4 = page table (see ARCH_PGD_SHIFT in asm/memory.h) 412 * r9 = processor ID 413 * r13 = *virtual* address to jump to upon completion 414 */ 415__enable_mmu: 416#if defined(CONFIG_ALIGNMENT_TRAP) && __LINUX_ARM_ARCH__ < 6 417 orr r0, r0, #CR_A 418#else 419 bic r0, r0, #CR_A 420#endif 421#ifdef CONFIG_CPU_DCACHE_DISABLE 422 bic r0, r0, #CR_C 423#endif 424#ifdef CONFIG_CPU_BPREDICT_DISABLE 425 bic r0, r0, #CR_Z 426#endif 427#ifdef CONFIG_CPU_ICACHE_DISABLE 428 bic r0, r0, #CR_I 429#endif 430#ifndef CONFIG_ARM_LPAE 431 mov r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \ 432 domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \ 433 domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \ 434 domain_val(DOMAIN_IO, DOMAIN_CLIENT)) 435 mcr p15, 0, r5, c3, c0, 0 @ load domain access register 436 mcr p15, 0, r4, c2, c0, 0 @ load page table pointer 437#endif 438 b __turn_mmu_on 439ENDPROC(__enable_mmu) 440 441/* 442 * Enable the MMU. This completely changes the structure of the visible 443 * memory space. You will not be able to trace execution through this. 444 * If you have an enquiry about this, *please* check the linux-arm-kernel 445 * mailing list archives BEFORE sending another post to the list. 446 * 447 * r0 = cp#15 control register 448 * r1 = machine ID 449 * r2 = atags or dtb pointer 450 * r9 = processor ID 451 * r13 = *virtual* address to jump to upon completion 452 * 453 * other registers depend on the function called upon completion 454 */ 455 .align 5 456 .pushsection .idmap.text, "ax" 457ENTRY(__turn_mmu_on) 458 mov r0, r0 459 instr_sync 460 mcr p15, 0, r0, c1, c0, 0 @ write control reg 461 mrc p15, 0, r3, c0, c0, 0 @ read id reg 462 instr_sync 463 mov r3, r3 464 mov r3, r13 465 mov pc, r3 466__turn_mmu_on_end: 467ENDPROC(__turn_mmu_on) 468 .popsection 469 470 471#ifdef CONFIG_SMP_ON_UP 472 __INIT 473__fixup_smp: 474 and r3, r9, #0x000f0000 @ architecture version 475 teq r3, #0x000f0000 @ CPU ID supported? 476 bne __fixup_smp_on_up @ no, assume UP 477 478 bic r3, r9, #0x00ff0000 479 bic r3, r3, #0x0000000f @ mask 0xff00fff0 480 mov r4, #0x41000000 481 orr r4, r4, #0x0000b000 482 orr r4, r4, #0x00000020 @ val 0x4100b020 483 teq r3, r4 @ ARM 11MPCore? 484 moveq pc, lr @ yes, assume SMP 485 486 mrc p15, 0, r0, c0, c0, 5 @ read MPIDR 487 and r0, r0, #0xc0000000 @ multiprocessing extensions and 488 teq r0, #0x80000000 @ not part of a uniprocessor system? 489 moveq pc, lr @ yes, assume SMP 490 491__fixup_smp_on_up: 492 adr r0, 1f 493 ldmia r0, {r3 - r5} 494 sub r3, r0, r3 495 add r4, r4, r3 496 add r5, r5, r3 497 b __do_fixup_smp_on_up 498ENDPROC(__fixup_smp) 499 500 .align 5011: .word . 502 .word __smpalt_begin 503 .word __smpalt_end 504 505 .pushsection .data 506 .globl smp_on_up 507smp_on_up: 508 ALT_SMP(.long 1) 509 ALT_UP(.long 0) 510 .popsection 511#endif 512 513 .text 514__do_fixup_smp_on_up: 515 cmp r4, r5 516 movhs pc, lr 517 ldmia r4!, {r0, r6} 518 ARM( str r6, [r0, r3] ) 519 THUMB( add r0, r0, r3 ) 520#ifdef __ARMEB__ 521 THUMB( mov r6, r6, ror #16 ) @ Convert word order for big-endian. 522#endif 523 THUMB( strh r6, [r0], #2 ) @ For Thumb-2, store as two halfwords 524 THUMB( mov r6, r6, lsr #16 ) @ to be robust against misaligned r3. 525 THUMB( strh r6, [r0] ) 526 b __do_fixup_smp_on_up 527ENDPROC(__do_fixup_smp_on_up) 528 529ENTRY(fixup_smp) 530 stmfd sp!, {r4 - r6, lr} 531 mov r4, r0 532 add r5, r0, r1 533 mov r3, #0 534 bl __do_fixup_smp_on_up 535 ldmfd sp!, {r4 - r6, pc} 536ENDPROC(fixup_smp) 537 538#ifdef CONFIG_ARM_PATCH_PHYS_VIRT 539 540/* __fixup_pv_table - patch the stub instructions with the delta between 541 * PHYS_OFFSET and PAGE_OFFSET, which is assumed to be 16MiB aligned and 542 * can be expressed by an immediate shifter operand. The stub instruction 543 * has a form of '(add|sub) rd, rn, #imm'. 544 */ 545 __HEAD 546__fixup_pv_table: 547 adr r0, 1f 548 ldmia r0, {r3-r5, r7} 549 sub r3, r0, r3 @ PHYS_OFFSET - PAGE_OFFSET 550 add r4, r4, r3 @ adjust table start address 551 add r5, r5, r3 @ adjust table end address 552 add r7, r7, r3 @ adjust __pv_phys_offset address 553 str r8, [r7] @ save computed PHYS_OFFSET to __pv_phys_offset 554 mov r6, r3, lsr #24 @ constant for add/sub instructions 555 teq r3, r6, lsl #24 @ must be 16MiB aligned 556THUMB( it ne @ cross section branch ) 557 bne __error 558 str r6, [r7, #4] @ save to __pv_offset 559 b __fixup_a_pv_table 560ENDPROC(__fixup_pv_table) 561 562 .align 5631: .long . 564 .long __pv_table_begin 565 .long __pv_table_end 5662: .long __pv_phys_offset 567 568 .text 569__fixup_a_pv_table: 570#ifdef CONFIG_THUMB2_KERNEL 571 lsls r6, #24 572 beq 2f 573 clz r7, r6 574 lsr r6, #24 575 lsl r6, r7 576 bic r6, #0x0080 577 lsrs r7, #1 578 orrcs r6, #0x0080 579 orr r6, r6, r7, lsl #12 580 orr r6, #0x4000 581 b 2f 5821: add r7, r3 583 ldrh ip, [r7, #2] 584 and ip, 0x8f00 585 orr ip, r6 @ mask in offset bits 31-24 586 strh ip, [r7, #2] 5872: cmp r4, r5 588 ldrcc r7, [r4], #4 @ use branch for delay slot 589 bcc 1b 590 bx lr 591#else 592 b 2f 5931: ldr ip, [r7, r3] 594 bic ip, ip, #0x000000ff 595 orr ip, ip, r6 @ mask in offset bits 31-24 596 str ip, [r7, r3] 5972: cmp r4, r5 598 ldrcc r7, [r4], #4 @ use branch for delay slot 599 bcc 1b 600 mov pc, lr 601#endif 602ENDPROC(__fixup_a_pv_table) 603 604ENTRY(fixup_pv_table) 605 stmfd sp!, {r4 - r7, lr} 606 ldr r2, 2f @ get address of __pv_phys_offset 607 mov r3, #0 @ no offset 608 mov r4, r0 @ r0 = table start 609 add r5, r0, r1 @ r1 = table size 610 ldr r6, [r2, #4] @ get __pv_offset 611 bl __fixup_a_pv_table 612 ldmfd sp!, {r4 - r7, pc} 613ENDPROC(fixup_pv_table) 614 615 .align 6162: .long __pv_phys_offset 617 618 .data 619 .globl __pv_phys_offset 620 .type __pv_phys_offset, %object 621__pv_phys_offset: 622 .long 0 623 .size __pv_phys_offset, . - __pv_phys_offset 624__pv_offset: 625 .long 0 626#endif 627 628#include "head-common.S" 629