LXR linux/arch/x86/kernel/head

   1/* SPDX-License-Identifier: GPL-2.0 */
   2/*
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 *  Enhanced CPU detection and feature setting code by Mike Jagdis
   7 *  and Martin Mares, November 1997.
   8 */
   9
  10.text
  11#include <linux/threads.h>
  12#include <linux/init.h>
  13#include <linux/linkage.h>
  14#include <asm/segment.h>
  15#include <asm/page_types.h>
  16#include <asm/pgtable_types.h>
  17#include <asm/cache.h>
  18#include <asm/thread_info.h>
  19#include <asm/asm-offsets.h>
  20#include <asm/setup.h>
  21#include <asm/processor-flags.h>
  22#include <asm/msr-index.h>
  23#include <asm/cpufeatures.h>
  24#include <asm/percpu.h>
  25#include <asm/nops.h>
  26#include <asm/bootparam.h>
  27#include <asm/export.h>
  28#include <asm/pgtable_32.h>
  29
  30/* Physical address */
  31#define pa(X) ((X) - __PAGE_OFFSET)
  32
  33/*
  34 * References to members of the new_cpu_data structure.
  35 */
  36
  37#define X86             new_cpu_data+CPUINFO_x86
  38#define X86_VENDOR      new_cpu_data+CPUINFO_x86_vendor
  39#define X86_MODEL       new_cpu_data+CPUINFO_x86_model
  40#define X86_STEPPING    new_cpu_data+CPUINFO_x86_stepping
  41#define X86_HARD_MATH   new_cpu_data+CPUINFO_hard_math
  42#define X86_CPUID       new_cpu_data+CPUINFO_cpuid_level
  43#define X86_CAPABILITY  new_cpu_data+CPUINFO_x86_capability
  44#define X86_VENDOR_ID   new_cpu_data+CPUINFO_x86_vendor_id
  45
  46
  47#define SIZEOF_PTREGS 17*4
  48
  49/*
  50 * Worst-case size of the kernel mapping we need to make:
  51 * a relocatable kernel can live anywhere in lowmem, so we need to be able
  52 * to map all of lowmem.
  53 */
  54KERNEL_PAGES = LOWMEM_PAGES
  55
  56INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
  57RESERVE_BRK(pagetables, INIT_MAP_SIZE)
  58
  59/*
  60 * 32-bit kernel entrypoint; only used by the boot CPU.  On entry,
  61 * %esi points to the real-mode code as a 32-bit pointer.
  62 * CS and DS must be 4 GB flat segments, but we don't depend on
  63 * any particular GDT layout, because we load our own as soon as we
  64 * can.
  65 */
  66__HEAD
  67SYM_CODE_START(startup_32)
  68        movl pa(initial_stack),%ecx
  69        
  70/*
  71 * Set segments to known values.
  72 */
  73        lgdt pa(boot_gdt_descr)
  74        movl $(__BOOT_DS),%eax
  75        movl %eax,%ds
  76        movl %eax,%es
  77        movl %eax,%fs
  78        movl %eax,%gs
  79        movl %eax,%ss
  80        leal -__PAGE_OFFSET(%ecx),%esp
  81
  82/*
  83 * Clear BSS first so that there are no surprises...
  84 */
  85        cld
  86        xorl %eax,%eax
  87        movl $pa(__bss_start),%edi
  88        movl $pa(__bss_stop),%ecx
  89        subl %edi,%ecx
  90        shrl $2,%ecx
  91        rep ; stosl
  92/*
  93 * Copy bootup parameters out of the way.
  94 * Note: %esi still has the pointer to the real-mode data.
  95 * With the kexec as boot loader, parameter segment might be loaded beyond
  96 * kernel image and might not even be addressable by early boot page tables.
  97 * (kexec on panic case). Hence copy out the parameters before initializing
  98 * page tables.
  99 */
 100        movl $pa(boot_params),%edi
 101        movl $(PARAM_SIZE/4),%ecx
 102        cld
 103        rep
 104        movsl
 105        movl pa(boot_params) + NEW_CL_POINTER,%esi
 106        andl %esi,%esi
 107        jz 1f                   # No command line
 108        movl $pa(boot_command_line),%edi
 109        movl $(COMMAND_LINE_SIZE/4),%ecx
 110        rep
 111        movsl
 1121:
 113
 114#ifdef CONFIG_OLPC
 115        /* save OFW's pgdir table for later use when calling into OFW */
 116        movl %cr3, %eax
 117        movl %eax, pa(olpc_ofw_pgd)
 118#endif
 119
 120#ifdef CONFIG_MICROCODE
 121        /* Early load ucode on BSP. */
 122        call load_ucode_bsp
 123#endif
 124
 125        /* Create early pagetables. */
 126        call  mk_early_pgtbl_32
 127
 128        /* Do early initialization of the fixmap area */
 129        movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
 130#ifdef  CONFIG_X86_PAE
 131#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
 132        movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
 133#else
 134        movl %eax,pa(initial_page_table+0xffc)
 135#endif
 136
 137        jmp .Ldefault_entry
 138SYM_CODE_END(startup_32)
 139
 140#ifdef CONFIG_HOTPLUG_CPU
 141/*
 142 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
 143 * up already except stack. We just set up stack here. Then call
 144 * start_secondary().
 145 */
 146SYM_FUNC_START(start_cpu0)
 147        movl initial_stack, %ecx
 148        movl %ecx, %esp
 149        call *(initial_code)
 1501:      jmp 1b
 151SYM_FUNC_END(start_cpu0)
 152#endif
 153
 154/*
 155 * Non-boot CPU entry point; entered from trampoline.S
 156 * We can't lgdt here, because lgdt itself uses a data segment, but
 157 * we know the trampoline has already loaded the boot_gdt for us.
 158 *
 159 * If cpu hotplug is not supported then this code can go in init section
 160 * which will be freed later
 161 */
 162SYM_FUNC_START(startup_32_smp)
 163        cld
 164        movl $(__BOOT_DS),%eax
 165        movl %eax,%ds
 166        movl %eax,%es
 167        movl %eax,%fs
 168        movl %eax,%gs
 169        movl pa(initial_stack),%ecx
 170        movl %eax,%ss
 171        leal -__PAGE_OFFSET(%ecx),%esp
 172
 173#ifdef CONFIG_MICROCODE
 174        /* Early load ucode on AP. */
 175        call load_ucode_ap
 176#endif
 177
 178.Ldefault_entry:
 179        movl $(CR0_STATE & ~X86_CR0_PG),%eax
 180        movl %eax,%cr0
 181
 182/*
 183 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
 184 * bits like NT set. This would confuse the debugger if this code is traced. So
 185 * initialize them properly now before switching to protected mode. That means
 186 * DF in particular (even though we have cleared it earlier after copying the
 187 * command line) because GCC expects it.
 188 */
 189        pushl $0
 190        popfl
 191
 192/*
 193 * New page tables may be in 4Mbyte page mode and may be using the global pages.
 194 *
 195 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
 196 * if and only if CPUID exists and has flags other than the FPU flag set.
 197 */
 198        movl $-1,pa(X86_CPUID)          # preset CPUID level
 199        movl $X86_EFLAGS_ID,%ecx
 200        pushl %ecx
 201        popfl                           # set EFLAGS=ID
 202        pushfl
 203        popl %eax                       # get EFLAGS
 204        testl $X86_EFLAGS_ID,%eax       # did EFLAGS.ID remained set?
 205        jz .Lenable_paging              # hw disallowed setting of ID bit
 206                                        # which means no CPUID and no CR4
 207
 208        xorl %eax,%eax
 209        cpuid
 210        movl %eax,pa(X86_CPUID)         # save largest std CPUID function
 211
 212        movl $1,%eax
 213        cpuid
 214        andl $~1,%edx                   # Ignore CPUID.FPU
 215        jz .Lenable_paging              # No flags or only CPUID.FPU = no CR4
 216
 217        movl pa(mmu_cr4_features),%eax
 218        movl %eax,%cr4
 219
 220        testb $X86_CR4_PAE, %al         # check if PAE is enabled
 221        jz .Lenable_paging
 222
 223        /* Check if extended functions are implemented */
 224        movl $0x80000000, %eax
 225        cpuid
 226        /* Value must be in the range 0x80000001 to 0x8000ffff */
 227        subl $0x80000001, %eax
 228        cmpl $(0x8000ffff-0x80000001), %eax
 229        ja .Lenable_paging
 230
 231        /* Clear bogus XD_DISABLE bits */
 232        call verify_cpu
 233
 234        mov $0x80000001, %eax
 235        cpuid
 236        /* Execute Disable bit supported? */
 237        btl $(X86_FEATURE_NX & 31), %edx
 238        jnc .Lenable_paging
 239
 240        /* Setup EFER (Extended Feature Enable Register) */
 241        movl $MSR_EFER, %ecx
 242        rdmsr
 243
 244        btsl $_EFER_NX, %eax
 245        /* Make changes effective */
 246        wrmsr
 247
 248.Lenable_paging:
 249
 250/*
 251 * Enable paging
 252 */
 253        movl $pa(initial_page_table), %eax
 254        movl %eax,%cr3          /* set the page table pointer.. */
 255        movl $CR0_STATE,%eax
 256        movl %eax,%cr0          /* ..and set paging (PG) bit */
 257        ljmp $__BOOT_CS,$1f     /* Clear prefetch and normalize %eip */
 2581:
 259        /* Shift the stack pointer to a virtual address */
 260        addl $__PAGE_OFFSET, %esp
 261
 262/*
 263 * start system 32-bit setup. We need to re-do some of the things done
 264 * in 16-bit mode for the "real" operations.
 265 */
 266        movl setup_once_ref,%eax
 267        andl %eax,%eax
 268        jz 1f                           # Did we do this already?
 269        call *%eax
 2701:
 271
 272/*
 273 * Check if it is 486
 274 */
 275        movb $4,X86                     # at least 486
 276        cmpl $-1,X86_CPUID
 277        je .Lis486
 278
 279        /* get vendor info */
 280        xorl %eax,%eax                  # call CPUID with 0 -> return vendor ID
 281        cpuid
 282        movl %eax,X86_CPUID             # save CPUID level
 283        movl %ebx,X86_VENDOR_ID         # lo 4 chars
 284        movl %edx,X86_VENDOR_ID+4       # next 4 chars
 285        movl %ecx,X86_VENDOR_ID+8       # last 4 chars
 286
 287        orl %eax,%eax                   # do we have processor info as well?
 288        je .Lis486
 289
 290        movl $1,%eax            # Use the CPUID instruction to get CPU type
 291        cpuid
 292        movb %al,%cl            # save reg for future use
 293        andb $0x0f,%ah          # mask processor family
 294        movb %ah,X86
 295        andb $0xf0,%al          # mask model
 296        shrb $4,%al
 297        movb %al,X86_MODEL
 298        andb $0x0f,%cl          # mask mask revision
 299        movb %cl,X86_STEPPING
 300        movl %edx,X86_CAPABILITY
 301
 302.Lis486:
 303        movl $0x50022,%ecx      # set AM, WP, NE and MP
 304        movl %cr0,%eax
 305        andl $0x80000011,%eax   # Save PG,PE,ET
 306        orl %ecx,%eax
 307        movl %eax,%cr0
 308
 309        lgdt early_gdt_descr
 310        ljmp $(__KERNEL_CS),$1f
 3111:      movl $(__KERNEL_DS),%eax        # reload all the segment registers
 312        movl %eax,%ss                   # after changing gdt.
 313
 314        movl $(__USER_DS),%eax          # DS/ES contains default USER segment
 315        movl %eax,%ds
 316        movl %eax,%es
 317
 318        movl $(__KERNEL_PERCPU), %eax
 319        movl %eax,%fs                   # set this cpu's percpu
 320
 321        xorl %eax,%eax
 322        movl %eax,%gs                   # clear possible garbage in %gs
 323
 324        xorl %eax,%eax                  # Clear LDT
 325        lldt %ax
 326
 327        call *(initial_code)
 3281:      jmp 1b
 329SYM_FUNC_END(startup_32_smp)
 330
 331#include "verify_cpu.S"
 332
 333/*
 334 *  setup_once
 335 *
 336 *  The setup work we only want to run on the BSP.
 337 *
 338 *  Warning: %esi is live across this function.
 339 */
 340__INIT
 341setup_once:
 342        andl $0,setup_once_ref  /* Once is enough, thanks */
 343        ret
 344
 345SYM_FUNC_START(early_idt_handler_array)
 346        # 36(%esp) %eflags
 347        # 32(%esp) %cs
 348        # 28(%esp) %eip
 349        # 24(%rsp) error code
 350        i = 0
 351        .rept NUM_EXCEPTION_VECTORS
 352        .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
 353        pushl $0                # Dummy error code, to make stack frame uniform
 354        .endif
 355        pushl $i                # 20(%esp) Vector number
 356        jmp early_idt_handler_common
 357        i = i + 1
 358        .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
 359        .endr
 360SYM_FUNC_END(early_idt_handler_array)
 361        
 362SYM_CODE_START_LOCAL(early_idt_handler_common)
 363        /*
 364         * The stack is the hardware frame, an error code or zero, and the
 365         * vector number.
 366         */
 367        cld
 368
 369        incl %ss:early_recursion_flag
 370
 371        /* The vector number is in pt_regs->gs */
 372
 373        cld
 374        pushl   %fs             /* pt_regs->fs (__fsh varies by model) */
 375        pushl   %es             /* pt_regs->es (__esh varies by model) */
 376        pushl   %ds             /* pt_regs->ds (__dsh varies by model) */
 377        pushl   %eax            /* pt_regs->ax */
 378        pushl   %ebp            /* pt_regs->bp */
 379        pushl   %edi            /* pt_regs->di */
 380        pushl   %esi            /* pt_regs->si */
 381        pushl   %edx            /* pt_regs->dx */
 382        pushl   %ecx            /* pt_regs->cx */
 383        pushl   %ebx            /* pt_regs->bx */
 384
 385        /* Fix up DS and ES */
 386        movl    $(__KERNEL_DS), %ecx
 387        movl    %ecx, %ds
 388        movl    %ecx, %es
 389
 390        /* Load the vector number into EDX */
 391        movl    PT_GS(%esp), %edx
 392
 393        /* Load GS into pt_regs->gs (and maybe clobber __gsh) */
 394        movw    %gs, PT_GS(%esp)
 395
 396        movl    %esp, %eax      /* args are pt_regs (EAX), trapnr (EDX) */
 397        call    early_fixup_exception
 398
 399        popl    %ebx            /* pt_regs->bx */
 400        popl    %ecx            /* pt_regs->cx */
 401        popl    %edx            /* pt_regs->dx */
 402        popl    %esi            /* pt_regs->si */
 403        popl    %edi            /* pt_regs->di */
 404        popl    %ebp            /* pt_regs->bp */
 405        popl    %eax            /* pt_regs->ax */
 406        popl    %ds             /* pt_regs->ds (always ignores __dsh) */
 407        popl    %es             /* pt_regs->es (always ignores __esh) */
 408        popl    %fs             /* pt_regs->fs (always ignores __fsh) */
 409        popl    %gs             /* pt_regs->gs (always ignores __gsh) */
 410        decl    %ss:early_recursion_flag
 411        addl    $4, %esp        /* pop pt_regs->orig_ax */
 412        iret
 413SYM_CODE_END(early_idt_handler_common)
 414
 415/* This is the default interrupt "handler" :-) */
 416SYM_FUNC_START(early_ignore_irq)
 417        cld
 418#ifdef CONFIG_PRINTK
 419        pushl %eax
 420        pushl %ecx
 421        pushl %edx
 422        pushl %es
 423        pushl %ds
 424        movl $(__KERNEL_DS),%eax
 425        movl %eax,%ds
 426        movl %eax,%es
 427        cmpl $2,early_recursion_flag
 428        je hlt_loop
 429        incl early_recursion_flag
 430        pushl 16(%esp)
 431        pushl 24(%esp)
 432        pushl 32(%esp)
 433        pushl 40(%esp)
 434        pushl $int_msg
 435        call _printk
 436
 437        call dump_stack
 438
 439        addl $(5*4),%esp
 440        popl %ds
 441        popl %es
 442        popl %edx
 443        popl %ecx
 444        popl %eax
 445#endif
 446        iret
 447
 448hlt_loop:
 449        hlt
 450        jmp hlt_loop
 451SYM_FUNC_END(early_ignore_irq)
 452
 453__INITDATA
 454        .align 4
 455SYM_DATA(early_recursion_flag, .long 0)
 456
 457__REFDATA
 458        .align 4
 459SYM_DATA(initial_code,          .long i386_start_kernel)
 460SYM_DATA(setup_once_ref,        .long setup_once)
 461
 462#ifdef CONFIG_PAGE_TABLE_ISOLATION
 463#define PGD_ALIGN       (2 * PAGE_SIZE)
 464#define PTI_USER_PGD_FILL       1024
 465#else
 466#define PGD_ALIGN       (PAGE_SIZE)
 467#define PTI_USER_PGD_FILL       0
 468#endif
 469/*
 470 * BSS section
 471 */
 472__PAGE_ALIGNED_BSS
 473        .align PGD_ALIGN
 474#ifdef CONFIG_X86_PAE
 475.globl initial_pg_pmd
 476initial_pg_pmd:
 477        .fill 1024*KPMDS,4,0
 478#else
 479.globl initial_page_table
 480initial_page_table:
 481        .fill 1024,4,0
 482#endif
 483        .align PGD_ALIGN
 484initial_pg_fixmap:
 485        .fill 1024,4,0
 486.globl swapper_pg_dir
 487        .align PGD_ALIGN
 488swapper_pg_dir:
 489        .fill 1024,4,0
 490        .fill PTI_USER_PGD_FILL,4,0
 491.globl empty_zero_page
 492empty_zero_page:
 493        .fill 4096,1,0
 494EXPORT_SYMBOL(empty_zero_page)
 495
 496/*
 497 * This starts the data section.
 498 */
 499#ifdef CONFIG_X86_PAE
 500__PAGE_ALIGNED_DATA
 501        /* Page-aligned for the benefit of paravirt? */
 502        .align PGD_ALIGN
 503SYM_DATA_START(initial_page_table)
 504        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0     /* low identity map */
 505# if KPMDS == 3
 506        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 507        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
 508        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
 509# elif KPMDS == 2
 510        .long   0,0
 511        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 512        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
 513# elif KPMDS == 1
 514        .long   0,0
 515        .long   0,0
 516        .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 517# else
 518#  error "Kernel PMDs should be 1, 2 or 3"
 519# endif
 520        .align PAGE_SIZE                /* needs to be page-sized too */
 521
 522#ifdef CONFIG_PAGE_TABLE_ISOLATION
 523        /*
 524         * PTI needs another page so sync_initial_pagetable() works correctly
 525         * and does not scribble over the data which is placed behind the
 526         * actual initial_page_table. See clone_pgd_range().
 527         */
 528        .fill 1024, 4, 0
 529#endif
 530
 531SYM_DATA_END(initial_page_table)
 532#endif
 533
 534.data
 535.balign 4
 536/*
 537 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
 538 * reliably detect the end of the stack.
 539 */
 540SYM_DATA(initial_stack,
 541                .long init_thread_union + THREAD_SIZE -
 542                SIZEOF_PTREGS - TOP_OF_KERNEL_STACK_PADDING)
 543
 544__INITRODATA
 545int_msg:
 546        .asciz "Unknown interrupt or fault at: %p %p %p\n"
 547
 548#include "../../x86/xen/xen-head.S"
 549
 550/*
 551 * The IDT and GDT 'descriptors' are a strange 48-bit object
 552 * only used by the lidt and lgdt instructions. They are not
 553 * like usual segment descriptors - they consist of a 16-bit
 554 * segment size, and 32-bit linear address value:
 555 */
 556
 557        .data
 558        ALIGN
 559# early boot GDT descriptor (must use 1:1 address mapping)
 560        .word 0                         # 32 bit align gdt_desc.address
 561SYM_DATA_START_LOCAL(boot_gdt_descr)
 562        .word __BOOT_DS+7
 563        .long boot_gdt - __PAGE_OFFSET
 564SYM_DATA_END(boot_gdt_descr)
 565
 566# boot GDT descriptor (later on used by CPU#0):
 567        .word 0                         # 32 bit align gdt_desc.address
 568SYM_DATA_START(early_gdt_descr)
 569        .word GDT_ENTRIES*8-1
 570        .long gdt_page                  /* Overwritten for secondary CPUs */
 571SYM_DATA_END(early_gdt_descr)
 572
 573/*
 574 * The boot_gdt must mirror the equivalent in setup.S and is
 575 * used only for booting.
 576 */
 577        .align L1_CACHE_BYTES
 578SYM_DATA_START(boot_gdt)
 579        .fill GDT_ENTRY_BOOT_CS,8,0
 580        .quad 0x00cf9a000000ffff        /* kernel 4GB code at 0x00000000 */
 581        .quad 0x00cf92000000ffff        /* kernel 4GB data at 0x00000000 */
 582SYM_DATA_END(boot_gdt)
 583