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 43PGD_PAGE_OFFSET = pgd_index(__PAGE_OFFSET_BASE) 44PGD_START_KERNEL = pgd_index(__START_KERNEL_map) 45L3_START_KERNEL = pud_index(__START_KERNEL_map) 46 47 .text 48 __HEAD 49 .code64 50 .globl startup_64 51startup_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 90ENTRY(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 1181: 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 1341: 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) 1491: wrmsr /* Make changes effective */ 150 151 /* Setup cr0 */ 152 movl $CR0_STATE, %eax 153 /* Make changes effective */ 154 movq %rax, %cr0 155 156 /* Setup a boot time stack */ 157 movq initial_stack(%rip), %rsp 158 159 /* zero EFLAGS after setting rsp */ 160 pushq $0 161 popfq 162 163 /* 164 * We must switch to a new descriptor in kernel space for the GDT 165 * because soon the kernel won't have access anymore to the userspace 166 * addresses where we're currently running on. We have to do that here 167 * because in 32bit we couldn't load a 64bit linear address. 168 */ 169 lgdt early_gdt_descr(%rip) 170 171 /* set up data segments */ 172 xorl %eax,%eax 173 movl %eax,%ds 174 movl %eax,%ss 175 movl %eax,%es 176 177 /* 178 * We don't really need to load %fs or %gs, but load them anyway 179 * to kill any stale realmode selectors. This allows execution 180 * under VT hardware. 181 */ 182 movl %eax,%fs 183 movl %eax,%gs 184 185 /* Set up %gs. 186 * 187 * The base of %gs always points to the bottom of the irqstack 188 * union. If the stack protector canary is enabled, it is 189 * located at %gs:40. Note that, on SMP, the boot cpu uses 190 * init data section till per cpu areas are set up. 191 */ 192 movl $MSR_GS_BASE,%ecx 193 movl initial_gs(%rip),%eax 194 movl initial_gs+4(%rip),%edx 195 wrmsr 196 197 /* rsi is pointer to real mode structure with interesting info. 198 pass it to C */ 199 movq %rsi, %rdi 200 201.Ljump_to_C_code: 202 /* 203 * Jump to run C code and to be on a real kernel address. 204 * Since we are running on identity-mapped space we have to jump 205 * to the full 64bit address, this is only possible as indirect 206 * jump. In addition we need to ensure %cs is set so we make this 207 * a far return. 208 * 209 * Note: do not change to far jump indirect with 64bit offset. 210 * 211 * AMD does not support far jump indirect with 64bit offset. 212 * AMD64 Architecture Programmer's Manual, Volume 3: states only 213 * JMP FAR mem16:16 FF /5 Far jump indirect, 214 * with the target specified by a far pointer in memory. 215 * JMP FAR mem16:32 FF /5 Far jump indirect, 216 * with the target specified by a far pointer in memory. 217 * 218 * Intel64 does support 64bit offset. 219 * Software Developer Manual Vol 2: states: 220 * FF /5 JMP m16:16 Jump far, absolute indirect, 221 * address given in m16:16 222 * FF /5 JMP m16:32 Jump far, absolute indirect, 223 * address given in m16:32. 224 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect, 225 * address given in m16:64. 226 */ 227 pushq $.Lafter_lret # put return address on stack for unwinder 228 xorq %rbp, %rbp # clear frame pointer 229 movq initial_code(%rip), %rax 230 pushq $__KERNEL_CS # set correct cs 231 pushq %rax # target address in negative space 232 lretq 233.Lafter_lret: 234ENDPROC(secondary_startup_64) 235 236#include "verify_cpu.S" 237 238#ifdef CONFIG_HOTPLUG_CPU 239/* 240 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set 241 * up already except stack. We just set up stack here. Then call 242 * start_secondary() via .Ljump_to_C_code. 243 */ 244ENTRY(start_cpu0) 245 movq initial_stack(%rip), %rsp 246 jmp .Ljump_to_C_code 247ENDPROC(start_cpu0) 248#endif 249 250 /* Both SMP bootup and ACPI suspend change these variables */ 251 __REFDATA 252 .balign 8 253 GLOBAL(initial_code) 254 .quad x86_64_start_kernel 255 GLOBAL(initial_gs) 256 .quad INIT_PER_CPU_VAR(irq_stack_union) 257 GLOBAL(initial_stack) 258 /* 259 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel 260 * unwinder reliably detect the end of the stack. 261 */ 262 .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS 263 __FINITDATA 264 265bad_address: 266 jmp bad_address 267 268 __INIT 269ENTRY(early_idt_handler_array) 270 # 104(%rsp) %rflags 271 # 96(%rsp) %cs 272 # 88(%rsp) %rip 273 # 80(%rsp) error code 274 i = 0 275 .rept NUM_EXCEPTION_VECTORS 276 .ifeq (EXCEPTION_ERRCODE_MASK >> i) & 1 277 pushq $0 # Dummy error code, to make stack frame uniform 278 .endif 279 pushq $i # 72(%rsp) Vector number 280 jmp early_idt_handler_common 281 i = i + 1 282 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc 283 .endr 284ENDPROC(early_idt_handler_array) 285 286early_idt_handler_common: 287 /* 288 * The stack is the hardware frame, an error code or zero, and the 289 * vector number. 290 */ 291 cld 292 293 incl early_recursion_flag(%rip) 294 295 /* The vector number is currently in the pt_regs->di slot. */ 296 pushq %rsi /* pt_regs->si */ 297 movq 8(%rsp), %rsi /* RSI = vector number */ 298 movq %rdi, 8(%rsp) /* pt_regs->di = RDI */ 299 pushq %rdx /* pt_regs->dx */ 300 pushq %rcx /* pt_regs->cx */ 301 pushq %rax /* pt_regs->ax */ 302 pushq %r8 /* pt_regs->r8 */ 303 pushq %r9 /* pt_regs->r9 */ 304 pushq %r10 /* pt_regs->r10 */ 305 pushq %r11 /* pt_regs->r11 */ 306 pushq %rbx /* pt_regs->bx */ 307 pushq %rbp /* pt_regs->bp */ 308 pushq %r12 /* pt_regs->r12 */ 309 pushq %r13 /* pt_regs->r13 */ 310 pushq %r14 /* pt_regs->r14 */ 311 pushq %r15 /* pt_regs->r15 */ 312 313 cmpq $14,%rsi /* Page fault? */ 314 jnz 10f 315 GET_CR2_INTO(%rdi) /* Can clobber any volatile register if pv */ 316 call early_make_pgtable 317 andl %eax,%eax 318 jz 20f /* All good */ 319 32010: 321 movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */ 322 call early_fixup_exception 323 32420: 325 decl early_recursion_flag(%rip) 326 jmp restore_regs_and_iret 327ENDPROC(early_idt_handler_common) 328 329 __INITDATA 330 331 .balign 4 332GLOBAL(early_recursion_flag) 333 .long 0 334 335#define NEXT_PAGE(name) \ 336 .balign PAGE_SIZE; \ 337GLOBAL(name) 338 339/* Automate the creation of 1 to 1 mapping pmd entries */ 340#define PMDS(START, PERM, COUNT) \ 341 i = 0 ; \ 342 .rept (COUNT) ; \ 343 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \ 344 i = i + 1 ; \ 345 .endr 346 347 __INITDATA 348NEXT_PAGE(early_top_pgt) 349 .fill 511,8,0 350#ifdef CONFIG_X86_5LEVEL 351 .quad level4_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 352#else 353 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 354#endif 355 356NEXT_PAGE(early_dynamic_pgts) 357 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0 358 359 .data 360 361#ifndef CONFIG_XEN 362NEXT_PAGE(init_top_pgt) 363 .fill 512,8,0 364#else 365NEXT_PAGE(init_top_pgt) 366 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 367 .org init_top_pgt + PGD_PAGE_OFFSET*8, 0 368 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 369 .org init_top_pgt + PGD_START_KERNEL*8, 0 370 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ 371 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 372 373NEXT_PAGE(level3_ident_pgt) 374 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 375 .fill 511, 8, 0 376NEXT_PAGE(level2_ident_pgt) 377 /* Since I easily can, map the first 1G. 378 * Don't set NX because code runs from these pages. 379 */ 380 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD) 381#endif 382 383#ifdef CONFIG_X86_5LEVEL 384NEXT_PAGE(level4_kernel_pgt) 385 .fill 511,8,0 386 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 387#endif 388 389NEXT_PAGE(level3_kernel_pgt) 390 .fill L3_START_KERNEL,8,0 391 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */ 392 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 393 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 394 395NEXT_PAGE(level2_kernel_pgt) 396 /* 397 * 512 MB kernel mapping. We spend a full page on this pagetable 398 * anyway. 399 * 400 * The kernel code+data+bss must not be bigger than that. 401 * 402 * (NOTE: at +512MB starts the module area, see MODULES_VADDR. 403 * If you want to increase this then increase MODULES_VADDR 404 * too.) 405 */ 406 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, 407 KERNEL_IMAGE_SIZE/PMD_SIZE) 408 409NEXT_PAGE(level2_fixmap_pgt) 410 .fill 506,8,0 411 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 412 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */ 413 .fill 5,8,0 414 415NEXT_PAGE(level1_fixmap_pgt) 416 .fill 512,8,0 417 418#undef PMDS 419 420 .data 421 .align 16 422 .globl early_gdt_descr 423early_gdt_descr: 424 .word GDT_ENTRIES*8-1 425early_gdt_descr_base: 426 .quad INIT_PER_CPU_VAR(gdt_page) 427 428ENTRY(phys_base) 429 /* This must match the first entry in level2_kernel_pgt */ 430 .quad 0x0000000000000000 431EXPORT_SYMBOL(phys_base) 432 433#include "../../x86/xen/xen-head.S" 434 435 __PAGE_ALIGNED_BSS 436NEXT_PAGE(empty_zero_page) 437 .skip PAGE_SIZE 438EXPORT_SYMBOL(empty_zero_page) 439 440