arch/x86/kernel/head_32.S

   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_MASK        new_cpu_data+CPUINFO_x86_mask
  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  */
  54 KERNEL_PAGES = LOWMEM_PAGES
  55
  56 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
  57 RESERVE_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
  67 ENTRY(startup_32)
  68         movl pa(initial_stack),%ecx
  69
  70         /* test KEEP_SEGMENTS flag to see if the bootloader is asking
  71                 us to not reload segments */
  72         testb $KEEP_SEGMENTS, BP_loadflags(%esi)
  73         jnz 2f
  74
  75 /*
  76  * Set segments to known values.
  77  */
  78         lgdt pa(boot_gdt_descr)
  79         movl $(__BOOT_DS),%eax
  80         movl %eax,%ds
  81         movl %eax,%es
  82         movl %eax,%fs
  83         movl %eax,%gs
  84         movl %eax,%ss
  85 2:
  86         leal -__PAGE_OFFSET(%ecx),%esp
  87
  88 /*
  89  * Clear BSS first so that there are no surprises...
  90  */
  91         cld
  92         xorl %eax,%eax
  93         movl $pa(__bss_start),%edi
  94         movl $pa(__bss_stop),%ecx
  95         subl %edi,%ecx
  96         shrl $2,%ecx
  97         rep ; stosl
  98 /*
  99  * Copy bootup parameters out of the way.
 100  * Note: %esi still has the pointer to the real-mode data.
 101  * With the kexec as boot loader, parameter segment might be loaded beyond
 102  * kernel image and might not even be addressable by early boot page tables.
 103  * (kexec on panic case). Hence copy out the parameters before initializing
 104  * page tables.
 105  */
 106         movl $pa(boot_params),%edi
 107         movl $(PARAM_SIZE/4),%ecx
 108         cld
 109         rep
 110         movsl
 111         movl pa(boot_params) + NEW_CL_POINTER,%esi
 112         andl %esi,%esi
 113         jz 1f                   # No command line
 114         movl $pa(boot_command_line),%edi
 115         movl $(COMMAND_LINE_SIZE/4),%ecx
 116         rep
 117         movsl
 118 1:
 119
 120 #ifdef CONFIG_OLPC
 121         /* save OFW's pgdir table for later use when calling into OFW */
 122         movl %cr3, %eax
 123         movl %eax, pa(olpc_ofw_pgd)
 124 #endif
 125
 126 #ifdef CONFIG_MICROCODE
 127         /* Early load ucode on BSP. */
 128         call load_ucode_bsp
 129 #endif
 130
 131         /* Create early pagetables. */
 132         call  mk_early_pgtbl_32
 133
 134         /* Do early initialization of the fixmap area */
 135         movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
 136 #ifdef  CONFIG_X86_PAE
 137 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
 138         movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
 139 #else
 140         movl %eax,pa(initial_page_table+0xffc)
 141 #endif
 142
 143 #ifdef CONFIG_PARAVIRT
 144         /* This is can only trip for a broken bootloader... */
 145         cmpw $0x207, pa(boot_params + BP_version)
 146         jb .Ldefault_entry
 147
 148         /* Paravirt-compatible boot parameters.  Look to see what architecture
 149                 we're booting under. */
 150         movl pa(boot_params + BP_hardware_subarch), %eax
 151         cmpl $num_subarch_entries, %eax
 152         jae .Lbad_subarch
 153
 154         movl pa(subarch_entries)(,%eax,4), %eax
 155         subl $__PAGE_OFFSET, %eax
 156         jmp *%eax
 157
 158 .Lbad_subarch:
 159 WEAK(xen_entry)
 160         /* Unknown implementation; there's really
 161            nothing we can do at this point. */
 162         ud2a
 163
 164         __INITDATA
 165
 166 subarch_entries:
 167         .long .Ldefault_entry           /* normal x86/PC */
 168         .long xen_entry                 /* Xen hypervisor */
 169         .long .Ldefault_entry           /* Moorestown MID */
 170 num_subarch_entries = (. - subarch_entries) / 4
 171 .previous
 172 #else
 173         jmp .Ldefault_entry
 174 #endif /* CONFIG_PARAVIRT */
 175
 176 #ifdef CONFIG_HOTPLUG_CPU
 177 /*
 178  * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
 179  * up already except stack. We just set up stack here. Then call
 180  * start_secondary().
 181  */
 182 ENTRY(start_cpu0)
 183         movl initial_stack, %ecx
 184         movl %ecx, %esp
 185         call *(initial_code)
 186 1:      jmp 1b
 187 ENDPROC(start_cpu0)
 188 #endif
 189
 190 /*
 191  * Non-boot CPU entry point; entered from trampoline.S
 192  * We can't lgdt here, because lgdt itself uses a data segment, but
 193  * we know the trampoline has already loaded the boot_gdt for us.
 194  *
 195  * If cpu hotplug is not supported then this code can go in init section
 196  * which will be freed later
 197  */
 198 ENTRY(startup_32_smp)
 199         cld
 200         movl $(__BOOT_DS),%eax
 201         movl %eax,%ds
 202         movl %eax,%es
 203         movl %eax,%fs
 204         movl %eax,%gs
 205         movl pa(initial_stack),%ecx
 206         movl %eax,%ss
 207         leal -__PAGE_OFFSET(%ecx),%esp
 208
 209 #ifdef CONFIG_MICROCODE
 210         /* Early load ucode on AP. */
 211         call load_ucode_ap
 212 #endif
 213
 214 .Ldefault_entry:
 215 #define CR0_STATE       (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
 216                          X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
 217                          X86_CR0_PG)
 218         movl $(CR0_STATE & ~X86_CR0_PG),%eax
 219         movl %eax,%cr0
 220
 221 /*
 222  * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
 223  * bits like NT set. This would confuse the debugger if this code is traced. So
 224  * initialize them properly now before switching to protected mode. That means
 225  * DF in particular (even though we have cleared it earlier after copying the
 226  * command line) because GCC expects it.
 227  */
 228         pushl $0
 229         popfl
 230
 231 /*
 232  * New page tables may be in 4Mbyte page mode and may be using the global pages.
 233  *
 234  * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
 235  * if and only if CPUID exists and has flags other than the FPU flag set.
 236  */
 237         movl $-1,pa(X86_CPUID)          # preset CPUID level
 238         movl $X86_EFLAGS_ID,%ecx
 239         pushl %ecx
 240         popfl                           # set EFLAGS=ID
 241         pushfl
 242         popl %eax                       # get EFLAGS
 243         testl $X86_EFLAGS_ID,%eax       # did EFLAGS.ID remained set?
 244         jz .Lenable_paging              # hw disallowed setting of ID bit
 245                                         # which means no CPUID and no CR4
 246
 247         xorl %eax,%eax
 248         cpuid
 249         movl %eax,pa(X86_CPUID)         # save largest std CPUID function
 250
 251         movl $1,%eax
 252         cpuid
 253         andl $~1,%edx                   # Ignore CPUID.FPU
 254         jz .Lenable_paging              # No flags or only CPUID.FPU = no CR4
 255
 256         movl pa(mmu_cr4_features),%eax
 257         movl %eax,%cr4
 258
 259         testb $X86_CR4_PAE, %al         # check if PAE is enabled
 260         jz .Lenable_paging
 261
 262         /* Check if extended functions are implemented */
 263         movl $0x80000000, %eax
 264         cpuid
 265         /* Value must be in the range 0x80000001 to 0x8000ffff */
 266         subl $0x80000001, %eax
 267         cmpl $(0x8000ffff-0x80000001), %eax
 268         ja .Lenable_paging
 269
 270         /* Clear bogus XD_DISABLE bits */
 271         call verify_cpu
 272
 273         mov $0x80000001, %eax
 274         cpuid
 275         /* Execute Disable bit supported? */
 276         btl $(X86_FEATURE_NX & 31), %edx
 277         jnc .Lenable_paging
 278
 279         /* Setup EFER (Extended Feature Enable Register) */
 280         movl $MSR_EFER, %ecx
 281         rdmsr
 282
 283         btsl $_EFER_NX, %eax
 284         /* Make changes effective */
 285         wrmsr
 286
 287 .Lenable_paging:
 288
 289 /*
 290  * Enable paging
 291  */
 292         movl $pa(initial_page_table), %eax
 293         movl %eax,%cr3          /* set the page table pointer.. */
 294         movl $CR0_STATE,%eax
 295         movl %eax,%cr0          /* ..and set paging (PG) bit */
 296         ljmp $__BOOT_CS,$1f     /* Clear prefetch and normalize %eip */
 297 1:
 298         /* Shift the stack pointer to a virtual address */
 299         addl $__PAGE_OFFSET, %esp
 300
 301 /*
 302  * start system 32-bit setup. We need to re-do some of the things done
 303  * in 16-bit mode for the "real" operations.
 304  */
 305         movl setup_once_ref,%eax
 306         andl %eax,%eax
 307         jz 1f                           # Did we do this already?
 308         call *%eax
 309 1:
 310
 311 /*
 312  * Check if it is 486
 313  */
 314         movb $4,X86                     # at least 486
 315         cmpl $-1,X86_CPUID
 316         je .Lis486
 317
 318         /* get vendor info */
 319         xorl %eax,%eax                  # call CPUID with 0 -> return vendor ID
 320         cpuid
 321         movl %eax,X86_CPUID             # save CPUID level
 322         movl %ebx,X86_VENDOR_ID         # lo 4 chars
 323         movl %edx,X86_VENDOR_ID+4       # next 4 chars
 324         movl %ecx,X86_VENDOR_ID+8       # last 4 chars
 325
 326         orl %eax,%eax                   # do we have processor info as well?
 327         je .Lis486
 328
 329         movl $1,%eax            # Use the CPUID instruction to get CPU type
 330         cpuid
 331         movb %al,%cl            # save reg for future use
 332         andb $0x0f,%ah          # mask processor family
 333         movb %ah,X86
 334         andb $0xf0,%al          # mask model
 335         shrb $4,%al
 336         movb %al,X86_MODEL
 337         andb $0x0f,%cl          # mask mask revision
 338         movb %cl,X86_MASK
 339         movl %edx,X86_CAPABILITY
 340
 341 .Lis486:
 342         movl $0x50022,%ecx      # set AM, WP, NE and MP
 343         movl %cr0,%eax
 344         andl $0x80000011,%eax   # Save PG,PE,ET
 345         orl %ecx,%eax
 346         movl %eax,%cr0
 347
 348         lgdt early_gdt_descr
 349         ljmp $(__KERNEL_CS),$1f
 350 1:      movl $(__KERNEL_DS),%eax        # reload all the segment registers
 351         movl %eax,%ss                   # after changing gdt.
 352
 353         movl $(__USER_DS),%eax          # DS/ES contains default USER segment
 354         movl %eax,%ds
 355         movl %eax,%es
 356
 357         movl $(__KERNEL_PERCPU), %eax
 358         movl %eax,%fs                   # set this cpu's percpu
 359
 360         movl $(__KERNEL_STACK_CANARY),%eax
 361         movl %eax,%gs
 362
 363         xorl %eax,%eax                  # Clear LDT
 364         lldt %ax
 365
 366         call *(initial_code)
 367 1:      jmp 1b
 368 ENDPROC(startup_32_smp)
 369
 370 #include "verify_cpu.S"
 371
 372 /*
 373  *  setup_once
 374  *
 375  *  The setup work we only want to run on the BSP.
 376  *
 377  *  Warning: %esi is live across this function.
 378  */
 379 __INIT
 380 setup_once:
 381 #ifdef CONFIG_CC_STACKPROTECTOR
 382         /*
 383          * Configure the stack canary. The linker can't handle this by
 384          * relocation.  Manually set base address in stack canary
 385          * segment descriptor.
 386          */
 387         movl $gdt_page,%eax
 388         movl $stack_canary,%ecx
 389         movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
 390         shrl $16, %ecx
 391         movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
 392         movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
 393 #endif
 394
 395         andl $0,setup_once_ref  /* Once is enough, thanks */
 396         ret
 397
 398 ENTRY(early_idt_handler_array)
 399         # 36(%esp) %eflags
 400         # 32(%esp) %cs
 401         # 28(%esp) %eip
 402         # 24(%rsp) error code
 403         i = 0
 404         .rept NUM_EXCEPTION_VECTORS
 405         .ifeq (EXCEPTION_ERRCODE_MASK >> i) & 1
 406         pushl $0                # Dummy error code, to make stack frame uniform
 407         .endif
 408         pushl $i                # 20(%esp) Vector number
 409         jmp early_idt_handler_common
 410         i = i + 1
 411         .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
 412         .endr
 413 ENDPROC(early_idt_handler_array)
 414
 415 early_idt_handler_common:
 416         /*
 417          * The stack is the hardware frame, an error code or zero, and the
 418          * vector number.
 419          */
 420         cld
 421
 422         incl %ss:early_recursion_flag
 423
 424         /* The vector number is in pt_regs->gs */
 425
 426         cld
 427         pushl   %fs             /* pt_regs->fs (__fsh varies by model) */
 428         pushl   %es             /* pt_regs->es (__esh varies by model) */
 429         pushl   %ds             /* pt_regs->ds (__dsh varies by model) */
 430         pushl   %eax            /* pt_regs->ax */
 431         pushl   %ebp            /* pt_regs->bp */
 432         pushl   %edi            /* pt_regs->di */
 433         pushl   %esi            /* pt_regs->si */
 434         pushl   %edx            /* pt_regs->dx */
 435         pushl   %ecx            /* pt_regs->cx */
 436         pushl   %ebx            /* pt_regs->bx */
 437
 438         /* Fix up DS and ES */
 439         movl    $(__KERNEL_DS), %ecx
 440         movl    %ecx, %ds
 441         movl    %ecx, %es
 442
 443         /* Load the vector number into EDX */
 444         movl    PT_GS(%esp), %edx
 445
 446         /* Load GS into pt_regs->gs (and maybe clobber __gsh) */
 447         movw    %gs, PT_GS(%esp)
 448
 449         movl    %esp, %eax      /* args are pt_regs (EAX), trapnr (EDX) */
 450         call    early_fixup_exception
 451
 452         popl    %ebx            /* pt_regs->bx */
 453         popl    %ecx            /* pt_regs->cx */
 454         popl    %edx            /* pt_regs->dx */
 455         popl    %esi            /* pt_regs->si */
 456         popl    %edi            /* pt_regs->di */
 457         popl    %ebp            /* pt_regs->bp */
 458         popl    %eax            /* pt_regs->ax */
 459         popl    %ds             /* pt_regs->ds (always ignores __dsh) */
 460         popl    %es             /* pt_regs->es (always ignores __esh) */
 461         popl    %fs             /* pt_regs->fs (always ignores __fsh) */
 462         popl    %gs             /* pt_regs->gs (always ignores __gsh) */
 463         decl    %ss:early_recursion_flag
 464         addl    $4, %esp        /* pop pt_regs->orig_ax */
 465         iret
 466 ENDPROC(early_idt_handler_common)
 467
 468 /* This is the default interrupt "handler" :-) */
 469 ENTRY(early_ignore_irq)
 470         cld
 471 #ifdef CONFIG_PRINTK
 472         pushl %eax
 473         pushl %ecx
 474         pushl %edx
 475         pushl %es
 476         pushl %ds
 477         movl $(__KERNEL_DS),%eax
 478         movl %eax,%ds
 479         movl %eax,%es
 480         cmpl $2,early_recursion_flag
 481         je hlt_loop
 482         incl early_recursion_flag
 483         pushl 16(%esp)
 484         pushl 24(%esp)
 485         pushl 32(%esp)
 486         pushl 40(%esp)
 487         pushl $int_msg
 488         call printk
 489
 490         call dump_stack
 491
 492         addl $(5*4),%esp
 493         popl %ds
 494         popl %es
 495         popl %edx
 496         popl %ecx
 497         popl %eax
 498 #endif
 499         iret
 500
 501 hlt_loop:
 502         hlt
 503         jmp hlt_loop
 504 ENDPROC(early_ignore_irq)
 505
 506 __INITDATA
 507         .align 4
 508 GLOBAL(early_recursion_flag)
 509         .long 0
 510
 511 __REFDATA
 512         .align 4
 513 ENTRY(initial_code)
 514         .long i386_start_kernel
 515 ENTRY(setup_once_ref)
 516         .long setup_once
 517
 518 /*
 519  * BSS section
 520  */
 521 __PAGE_ALIGNED_BSS
 522         .align PAGE_SIZE
 523 #ifdef CONFIG_X86_PAE
 524 .globl initial_pg_pmd
 525 initial_pg_pmd:
 526         .fill 1024*KPMDS,4,0
 527 #else
 528 .globl initial_page_table
 529 initial_page_table:
 530         .fill 1024,4,0
 531 #endif
 532 initial_pg_fixmap:
 533         .fill 1024,4,0
 534 .globl empty_zero_page
 535 empty_zero_page:
 536         .fill 4096,1,0
 537 .globl swapper_pg_dir
 538 swapper_pg_dir:
 539         .fill 1024,4,0
 540 EXPORT_SYMBOL(empty_zero_page)
 541
 542 /*
 543  * This starts the data section.
 544  */
 545 #ifdef CONFIG_X86_PAE
 546 __PAGE_ALIGNED_DATA
 547         /* Page-aligned for the benefit of paravirt? */
 548         .align PAGE_SIZE
 549 ENTRY(initial_page_table)
 550         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0     /* low identity map */
 551 # if KPMDS == 3
 552         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 553         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
 554         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
 555 # elif KPMDS == 2
 556         .long   0,0
 557         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 558         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
 559 # elif KPMDS == 1
 560         .long   0,0
 561         .long   0,0
 562         .long   pa(initial_pg_pmd+PGD_IDENT_ATTR),0
 563 # else
 564 #  error "Kernel PMDs should be 1, 2 or 3"
 565 # endif
 566         .align PAGE_SIZE                /* needs to be page-sized too */
 567 #endif
 568
 569 .data
 570 .balign 4
 571 ENTRY(initial_stack)
 572         /*
 573          * The SIZEOF_PTREGS gap is a convention which helps the in-kernel
 574          * unwinder reliably detect the end of the stack.
 575          */
 576         .long init_thread_union + THREAD_SIZE - SIZEOF_PTREGS - \
 577               TOP_OF_KERNEL_STACK_PADDING;
 578
 579 __INITRODATA
 580 int_msg:
 581         .asciz "Unknown interrupt or fault at: %p %p %p\n"
 582
 583 #include "../../x86/xen/xen-head.S"
 584
 585 /*
 586  * The IDT and GDT 'descriptors' are a strange 48-bit object
 587  * only used by the lidt and lgdt instructions. They are not
 588  * like usual segment descriptors - they consist of a 16-bit
 589  * segment size, and 32-bit linear address value:
 590  */
 591
 592         .data
 593 .globl boot_gdt_descr
 594
 595         ALIGN
 596 # early boot GDT descriptor (must use 1:1 address mapping)
 597         .word 0                         # 32 bit align gdt_desc.address
 598 boot_gdt_descr:
 599         .word __BOOT_DS+7
 600         .long boot_gdt - __PAGE_OFFSET
 601
 602 # boot GDT descriptor (later on used by CPU#0):
 603         .word 0                         # 32 bit align gdt_desc.address
 604 ENTRY(early_gdt_descr)
 605         .word GDT_ENTRIES*8-1
 606         .long gdt_page                  /* Overwritten for secondary CPUs */
 607
 608 /*
 609  * The boot_gdt must mirror the equivalent in setup.S and is
 610  * used only for booting.
 611  */
 612         .align L1_CACHE_BYTES
 613 ENTRY(boot_gdt)
 614         .fill GDT_ENTRY_BOOT_CS,8,0
 615         .quad 0x00cf9a000000ffff        /* kernel 4GB code at 0x00000000 */
 616         .quad 0x00cf92000000ffff        /* kernel 4GB data at 0x00000000 */