arch/x86/kernel/head_64.S

   1 /*
   2  *  linux/arch/x86/kernel/head_64.S -- start in 32bit and switch to 64bit
   3  *
   4  *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
   5  *  Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
   6  *  Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
   7  *  Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
   8  *  Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
   9  */
  10
  11
  12 #include <linux/linkage.h>
  13 #include <linux/threads.h>
  14 #include <linux/init.h>
  15 #include <asm/segment.h>
  16 #include <asm/pgtable.h>
  17 #include <asm/page.h>
  18 #include <asm/msr.h>
  19 #include <asm/cache.h>
  20 #include <asm/processor-flags.h>
  21 #include <asm/percpu.h>
  22 #include <asm/nops.h>
  23 #include "../entry/calling.h"
  24 #include <asm/export.h>
  25
  26 #ifdef CONFIG_PARAVIRT
  27 #include <asm/asm-offsets.h>
  28 #include <asm/paravirt.h>
  29 #define GET_CR2_INTO(reg) GET_CR2_INTO_RAX ; movq %rax, reg
  30 #else
  31 #define GET_CR2_INTO(reg) movq %cr2, reg
  32 #define INTERRUPT_RETURN iretq
  33 #endif
  34
  35 /* we are not able to switch in one step to the final KERNEL ADDRESS SPACE
  36  * because we need identity-mapped pages.
  37  *
  38  */
  39
  40 #define p4d_index(x)    (((x) >> P4D_SHIFT) & (PTRS_PER_P4D-1))
  41 #define pud_index(x)    (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
  42
  43 PGD_PAGE_OFFSET = pgd_index(__PAGE_OFFSET_BASE)
  44 PGD_START_KERNEL = pgd_index(__START_KERNEL_map)
  45 L3_START_KERNEL = pud_index(__START_KERNEL_map)
  46
  47         .text
  48         __HEAD
  49         .code64
  50         .globl startup_64
  51 startup_64:
  52         /*
  53          * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
  54          * and someone has loaded an identity mapped page table
  55          * for us.  These identity mapped page tables map all of the
  56          * kernel pages and possibly all of memory.
  57          *
  58          * %rsi holds a physical pointer to real_mode_data.
  59          *
  60          * We come here either directly from a 64bit bootloader, or from
  61          * arch/x86/boot/compressed/head_64.S.
  62          *
  63          * We only come here initially at boot nothing else comes here.
  64          *
  65          * Since we may be loaded at an address different from what we were
  66          * compiled to run at we first fixup the physical addresses in our page
  67          * tables and then reload them.
  68          */
  69
  70         /* Set up the stack for verify_cpu(), similar to initial_stack below */
  71         leaq    (__end_init_task - SIZEOF_PTREGS)(%rip), %rsp
  72
  73         /* Sanitize CPU configuration */
  74         call verify_cpu
  75
  76         /*
  77          * Perform pagetable fixups. Additionally, if SME is active, encrypt
  78          * the kernel and retrieve the modifier (SME encryption mask if SME
  79          * is active) to be added to the initial pgdir entry that will be
  80          * programmed into CR3.
  81          */
  82         leaq    _text(%rip), %rdi
  83         pushq   %rsi
  84         call    __startup_64
  85         popq    %rsi
  86
  87         /* Form the CR3 value being sure to include the CR3 modifier */
  88         addq    $(early_top_pgt - __START_KERNEL_map), %rax
  89         jmp 1f
  90 ENTRY(secondary_startup_64)
  91         /*
  92          * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
  93          * and someone has loaded a mapped page table.
  94          *
  95          * %rsi holds a physical pointer to real_mode_data.
  96          *
  97          * We come here either from startup_64 (using physical addresses)
  98          * or from trampoline.S (using virtual addresses).
  99          *
 100          * Using virtual addresses from trampoline.S removes the need
 101          * to have any identity mapped pages in the kernel page table
 102          * after the boot processor executes this code.
 103          */
 104
 105         /* Sanitize CPU configuration */
 106         call verify_cpu
 107
 108         /*
 109          * Retrieve the modifier (SME encryption mask if SME is active) to be
 110          * added to the initial pgdir entry that will be programmed into CR3.
 111          */
 112         pushq   %rsi
 113         call    __startup_secondary_64
 114         popq    %rsi
 115
 116         /* Form the CR3 value being sure to include the CR3 modifier */
 117         addq    $(init_top_pgt - __START_KERNEL_map), %rax
 118 1:
 119
 120         /* Enable PAE mode, PGE and LA57 */
 121         movl    $(X86_CR4_PAE | X86_CR4_PGE), %ecx
 122 #ifdef CONFIG_X86_5LEVEL
 123         orl     $X86_CR4_LA57, %ecx
 124 #endif
 125         movq    %rcx, %cr4
 126
 127         /* Setup early boot stage 4-/5-level pagetables. */
 128         addq    phys_base(%rip), %rax
 129         movq    %rax, %cr3
 130
 131         /* Ensure I am executing from virtual addresses */
 132         movq    $1f, %rax
 133         jmp     *%rax
 134 1:
 135
 136         /* Check if nx is implemented */
 137         movl    $0x80000001, %eax
 138         cpuid
 139         movl    %edx,%edi
 140
 141         /* Setup EFER (Extended Feature Enable Register) */
 142         movl    $MSR_EFER, %ecx
 143         rdmsr
 144         btsl    $_EFER_SCE, %eax        /* Enable System Call */
 145         btl     $20,%edi                /* No Execute supported? */
 146         jnc     1f
 147         btsl    $_EFER_NX, %eax
 148         btsq    $_PAGE_BIT_NX,early_pmd_flags(%rip)
 149 1:      wrmsr                           /* Make changes effective */
 150
 151         /* Setup cr0 */
 152 #define CR0_STATE       (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
 153                          X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
 154                          X86_CR0_PG)
 155         movl    $CR0_STATE, %eax
 156         /* Make changes effective */
 157         movq    %rax, %cr0
 158
 159         /* Setup a boot time stack */
 160         movq initial_stack(%rip), %rsp
 161
 162         /* zero EFLAGS after setting rsp */
 163         pushq $0
 164         popfq
 165
 166         /*
 167          * We must switch to a new descriptor in kernel space for the GDT
 168          * because soon the kernel won't have access anymore to the userspace
 169          * addresses where we're currently running on. We have to do that here
 170          * because in 32bit we couldn't load a 64bit linear address.
 171          */
 172         lgdt    early_gdt_descr(%rip)
 173
 174         /* set up data segments */
 175         xorl %eax,%eax
 176         movl %eax,%ds
 177         movl %eax,%ss
 178         movl %eax,%es
 179
 180         /*
 181          * We don't really need to load %fs or %gs, but load them anyway
 182          * to kill any stale realmode selectors.  This allows execution
 183          * under VT hardware.
 184          */
 185         movl %eax,%fs
 186         movl %eax,%gs
 187
 188         /* Set up %gs.
 189          *
 190          * The base of %gs always points to the bottom of the irqstack
 191          * union.  If the stack protector canary is enabled, it is
 192          * located at %gs:40.  Note that, on SMP, the boot cpu uses
 193          * init data section till per cpu areas are set up.
 194          */
 195         movl    $MSR_GS_BASE,%ecx
 196         movl    initial_gs(%rip),%eax
 197         movl    initial_gs+4(%rip),%edx
 198         wrmsr
 199
 200         /* rsi is pointer to real mode structure with interesting info.
 201            pass it to C */
 202         movq    %rsi, %rdi
 203
 204 .Ljump_to_C_code:
 205         /*
 206          * Jump to run C code and to be on a real kernel address.
 207          * Since we are running on identity-mapped space we have to jump
 208          * to the full 64bit address, this is only possible as indirect
 209          * jump.  In addition we need to ensure %cs is set so we make this
 210          * a far return.
 211          *
 212          * Note: do not change to far jump indirect with 64bit offset.
 213          *
 214          * AMD does not support far jump indirect with 64bit offset.
 215          * AMD64 Architecture Programmer's Manual, Volume 3: states only
 216          *      JMP FAR mem16:16 FF /5 Far jump indirect,
 217          *              with the target specified by a far pointer in memory.
 218          *      JMP FAR mem16:32 FF /5 Far jump indirect,
 219          *              with the target specified by a far pointer in memory.
 220          *
 221          * Intel64 does support 64bit offset.
 222          * Software Developer Manual Vol 2: states:
 223          *      FF /5 JMP m16:16 Jump far, absolute indirect,
 224          *              address given in m16:16
 225          *      FF /5 JMP m16:32 Jump far, absolute indirect,
 226          *              address given in m16:32.
 227          *      REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
 228          *              address given in m16:64.
 229          */
 230         pushq   $.Lafter_lret   # put return address on stack for unwinder
 231         xorq    %rbp, %rbp      # clear frame pointer
 232         movq    initial_code(%rip), %rax
 233         pushq   $__KERNEL_CS    # set correct cs
 234         pushq   %rax            # target address in negative space
 235         lretq
 236 .Lafter_lret:
 237 ENDPROC(secondary_startup_64)
 238
 239 #include "verify_cpu.S"
 240
 241 #ifdef CONFIG_HOTPLUG_CPU
 242 /*
 243  * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
 244  * up already except stack. We just set up stack here. Then call
 245  * start_secondary() via .Ljump_to_C_code.
 246  */
 247 ENTRY(start_cpu0)
 248         movq    initial_stack(%rip), %rsp
 249         jmp     .Ljump_to_C_code
 250 ENDPROC(start_cpu0)
 251 #endif
 252
 253         /* Both SMP bootup and ACPI suspend change these variables */
 254         __REFDATA
 255         .balign 8
 256         GLOBAL(initial_code)
 257         .quad   x86_64_start_kernel
 258         GLOBAL(initial_gs)
 259         .quad   INIT_PER_CPU_VAR(irq_stack_union)
 260         GLOBAL(initial_stack)
 261         /*
 262          * The SIZEOF_PTREGS gap is a convention which helps the in-kernel
 263          * unwinder reliably detect the end of the stack.
 264          */
 265         .quad  init_thread_union + THREAD_SIZE - SIZEOF_PTREGS
 266         __FINITDATA
 267
 268 bad_address:
 269         jmp bad_address
 270
 271         __INIT
 272 ENTRY(early_idt_handler_array)
 273         # 104(%rsp) %rflags
 274         #  96(%rsp) %cs
 275         #  88(%rsp) %rip
 276         #  80(%rsp) error code
 277         i = 0
 278         .rept NUM_EXCEPTION_VECTORS
 279         .ifeq (EXCEPTION_ERRCODE_MASK >> i) & 1
 280         pushq $0                # Dummy error code, to make stack frame uniform
 281         .endif
 282         pushq $i                # 72(%rsp) Vector number
 283         jmp early_idt_handler_common
 284         i = i + 1
 285         .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
 286         .endr
 287 ENDPROC(early_idt_handler_array)
 288
 289 early_idt_handler_common:
 290         /*
 291          * The stack is the hardware frame, an error code or zero, and the
 292          * vector number.
 293          */
 294         cld
 295
 296         incl early_recursion_flag(%rip)
 297
 298         /* The vector number is currently in the pt_regs->di slot. */
 299         pushq %rsi                              /* pt_regs->si */
 300         movq 8(%rsp), %rsi                      /* RSI = vector number */
 301         movq %rdi, 8(%rsp)                      /* pt_regs->di = RDI */
 302         pushq %rdx                              /* pt_regs->dx */
 303         pushq %rcx                              /* pt_regs->cx */
 304         pushq %rax                              /* pt_regs->ax */
 305         pushq %r8                               /* pt_regs->r8 */
 306         pushq %r9                               /* pt_regs->r9 */
 307         pushq %r10                              /* pt_regs->r10 */
 308         pushq %r11                              /* pt_regs->r11 */
 309         pushq %rbx                              /* pt_regs->bx */
 310         pushq %rbp                              /* pt_regs->bp */
 311         pushq %r12                              /* pt_regs->r12 */
 312         pushq %r13                              /* pt_regs->r13 */
 313         pushq %r14                              /* pt_regs->r14 */
 314         pushq %r15                              /* pt_regs->r15 */
 315
 316         cmpq $14,%rsi           /* Page fault? */
 317         jnz 10f
 318         GET_CR2_INTO(%rdi)      /* Can clobber any volatile register if pv */
 319         call early_make_pgtable
 320         andl %eax,%eax
 321         jz 20f                  /* All good */
 322
 323 10:
 324         movq %rsp,%rdi          /* RDI = pt_regs; RSI is already trapnr */
 325         call early_fixup_exception
 326
 327 20:
 328         decl early_recursion_flag(%rip)
 329         jmp restore_regs_and_iret
 330 ENDPROC(early_idt_handler_common)
 331
 332         __INITDATA
 333
 334         .balign 4
 335 GLOBAL(early_recursion_flag)
 336         .long 0
 337
 338 #define NEXT_PAGE(name) \
 339         .balign PAGE_SIZE; \
 340 GLOBAL(name)
 341
 342 /* Automate the creation of 1 to 1 mapping pmd entries */
 343 #define PMDS(START, PERM, COUNT)                        \
 344         i = 0 ;                                         \
 345         .rept (COUNT) ;                                 \
 346         .quad   (START) + (i << PMD_SHIFT) + (PERM) ;   \
 347         i = i + 1 ;                                     \
 348         .endr
 349
 350         __INITDATA
 351 NEXT_PAGE(early_top_pgt)
 352         .fill   511,8,0
 353 #ifdef CONFIG_X86_5LEVEL
 354         .quad   level4_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 355 #else
 356         .quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 357 #endif
 358
 359 NEXT_PAGE(early_dynamic_pgts)
 360         .fill   512*EARLY_DYNAMIC_PAGE_TABLES,8,0
 361
 362         .data
 363
 364 #ifndef CONFIG_XEN
 365 NEXT_PAGE(init_top_pgt)
 366         .fill   512,8,0
 367 #else
 368 NEXT_PAGE(init_top_pgt)
 369         .quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
 370         .org    init_top_pgt + PGD_PAGE_OFFSET*8, 0
 371         .quad   level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
 372         .org    init_top_pgt + PGD_START_KERNEL*8, 0
 373         /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
 374         .quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 375
 376 NEXT_PAGE(level3_ident_pgt)
 377         .quad   level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
 378         .fill   511, 8, 0
 379 NEXT_PAGE(level2_ident_pgt)
 380         /* Since I easily can, map the first 1G.
 381          * Don't set NX because code runs from these pages.
 382          */
 383         PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
 384 #endif
 385
 386 #ifdef CONFIG_X86_5LEVEL
 387 NEXT_PAGE(level4_kernel_pgt)
 388         .fill   511,8,0
 389         .quad   level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 390 #endif
 391
 392 NEXT_PAGE(level3_kernel_pgt)
 393         .fill   L3_START_KERNEL,8,0
 394         /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
 395         .quad   level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
 396         .quad   level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 397
 398 NEXT_PAGE(level2_kernel_pgt)
 399         /*
 400          * 512 MB kernel mapping. We spend a full page on this pagetable
 401          * anyway.
 402          *
 403          * The kernel code+data+bss must not be bigger than that.
 404          *
 405          * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
 406          *  If you want to increase this then increase MODULES_VADDR
 407          *  too.)
 408          */
 409         PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
 410                 KERNEL_IMAGE_SIZE/PMD_SIZE)
 411
 412 NEXT_PAGE(level2_fixmap_pgt)
 413         .fill   506,8,0
 414         .quad   level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
 415         /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
 416         .fill   5,8,0
 417
 418 NEXT_PAGE(level1_fixmap_pgt)
 419         .fill   512,8,0
 420
 421 #undef PMDS
 422
 423         .data
 424         .align 16
 425         .globl early_gdt_descr
 426 early_gdt_descr:
 427         .word   GDT_ENTRIES*8-1
 428 early_gdt_descr_base:
 429         .quad   INIT_PER_CPU_VAR(gdt_page)
 430
 431 ENTRY(phys_base)
 432         /* This must match the first entry in level2_kernel_pgt */
 433         .quad   0x0000000000000000
 434 EXPORT_SYMBOL(phys_base)
 435
 436 #include "../../x86/xen/xen-head.S"
 437
 438         __PAGE_ALIGNED_BSS
 439 NEXT_PAGE(empty_zero_page)
 440         .skip PAGE_SIZE
 441 EXPORT_SYMBOL(empty_zero_page)
 442