1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * linux/arch/x86/kernel/head_64.S -- start in 32bit and switch to 64bit 4 * 5 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 6 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> 7 * Copyright (C) 2000 Karsten Keil <kkeil@suse.de> 8 * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de> 9 * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com> 10 */ 11 12 13#include <linux/linkage.h> 14#include <linux/threads.h> 15#include <linux/init.h> 16#include <linux/pgtable.h> 17#include <asm/segment.h> 18#include <asm/page.h> 19#include <asm/msr.h> 20#include <asm/cache.h> 21#include <asm/processor-flags.h> 22#include <asm/percpu.h> 23#include <asm/nops.h> 24#include "../entry/calling.h" 25#include <asm/export.h> 26#include <asm/nospec-branch.h> 27#include <asm/fixmap.h> 28 29#ifdef CONFIG_PARAVIRT_XXL 30#include <asm/asm-offsets.h> 31#include <asm/paravirt.h> 32#define GET_CR2_INTO(reg) GET_CR2_INTO_AX ; _ASM_MOV %_ASM_AX, reg 33#else 34#define INTERRUPT_RETURN iretq 35#define GET_CR2_INTO(reg) _ASM_MOV %cr2, reg 36#endif 37 38/* 39 * We are not able to switch in one step to the final KERNEL ADDRESS SPACE 40 * because we need identity-mapped pages. 41 */ 42#define l4_index(x) (((x) >> 39) & 511) 43#define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1)) 44 45L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4) 46L4_START_KERNEL = l4_index(__START_KERNEL_map) 47 48L3_START_KERNEL = pud_index(__START_KERNEL_map) 49 50 .text 51 __HEAD 52 .code64 53SYM_CODE_START_NOALIGN(startup_64) 54 UNWIND_HINT_EMPTY 55 /* 56 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0, 57 * and someone has loaded an identity mapped page table 58 * for us. These identity mapped page tables map all of the 59 * kernel pages and possibly all of memory. 60 * 61 * %rsi holds a physical pointer to real_mode_data. 62 * 63 * We come here either directly from a 64bit bootloader, or from 64 * arch/x86/boot/compressed/head_64.S. 65 * 66 * We only come here initially at boot nothing else comes here. 67 * 68 * Since we may be loaded at an address different from what we were 69 * compiled to run at we first fixup the physical addresses in our page 70 * tables and then reload them. 71 */ 72 73 /* Set up the stack for verify_cpu(), similar to initial_stack below */ 74 leaq (__end_init_task - SIZEOF_PTREGS)(%rip), %rsp 75 76 leaq _text(%rip), %rdi 77 pushq %rsi 78 call startup_64_setup_env 79 popq %rsi 80 81 /* Now switch to __KERNEL_CS so IRET works reliably */ 82 pushq $__KERNEL_CS 83 leaq .Lon_kernel_cs(%rip), %rax 84 pushq %rax 85 lretq 86 87.Lon_kernel_cs: 88 UNWIND_HINT_EMPTY 89 90 /* Sanitize CPU configuration */ 91 call verify_cpu 92 93 /* 94 * Perform pagetable fixups. Additionally, if SME is active, encrypt 95 * the kernel and retrieve the modifier (SME encryption mask if SME 96 * is active) to be added to the initial pgdir entry that will be 97 * programmed into CR3. 98 */ 99 leaq _text(%rip), %rdi 100 pushq %rsi 101 call __startup_64 102 popq %rsi 103 104 /* Form the CR3 value being sure to include the CR3 modifier */ 105 addq $(early_top_pgt - __START_KERNEL_map), %rax 106 jmp 1f 107SYM_CODE_END(startup_64) 108 109SYM_CODE_START(secondary_startup_64) 110 UNWIND_HINT_EMPTY 111 /* 112 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0, 113 * and someone has loaded a mapped page table. 114 * 115 * %rsi holds a physical pointer to real_mode_data. 116 * 117 * We come here either from startup_64 (using physical addresses) 118 * or from trampoline.S (using virtual addresses). 119 * 120 * Using virtual addresses from trampoline.S removes the need 121 * to have any identity mapped pages in the kernel page table 122 * after the boot processor executes this code. 123 */ 124 125 /* Sanitize CPU configuration */ 126 call verify_cpu 127 128 /* 129 * Retrieve the modifier (SME encryption mask if SME is active) to be 130 * added to the initial pgdir entry that will be programmed into CR3. 131 */ 132 pushq %rsi 133 call __startup_secondary_64 134 popq %rsi 135 136 /* Form the CR3 value being sure to include the CR3 modifier */ 137 addq $(init_top_pgt - __START_KERNEL_map), %rax 1381: 139 140 /* Enable PAE mode, PGE and LA57 */ 141 movl $(X86_CR4_PAE | X86_CR4_PGE), %ecx 142#ifdef CONFIG_X86_5LEVEL 143 testl $1, __pgtable_l5_enabled(%rip) 144 jz 1f 145 orl $X86_CR4_LA57, %ecx 1461: 147#endif 148 movq %rcx, %cr4 149 150 /* Setup early boot stage 4-/5-level pagetables. */ 151 addq phys_base(%rip), %rax 152 movq %rax, %cr3 153 154 /* Ensure I am executing from virtual addresses */ 155 movq $1f, %rax 156 ANNOTATE_RETPOLINE_SAFE 157 jmp *%rax 1581: 159 UNWIND_HINT_EMPTY 160 161 /* Check if nx is implemented */ 162 movl $0x80000001, %eax 163 cpuid 164 movl %edx,%edi 165 166 /* Setup EFER (Extended Feature Enable Register) */ 167 movl $MSR_EFER, %ecx 168 rdmsr 169 btsl $_EFER_SCE, %eax /* Enable System Call */ 170 btl $20,%edi /* No Execute supported? */ 171 jnc 1f 172 btsl $_EFER_NX, %eax 173 btsq $_PAGE_BIT_NX,early_pmd_flags(%rip) 1741: wrmsr /* Make changes effective */ 175 176 /* Setup cr0 */ 177 movl $CR0_STATE, %eax 178 /* Make changes effective */ 179 movq %rax, %cr0 180 181 /* Setup a boot time stack */ 182 movq initial_stack(%rip), %rsp 183 184 /* zero EFLAGS after setting rsp */ 185 pushq $0 186 popfq 187 188 /* 189 * We must switch to a new descriptor in kernel space for the GDT 190 * because soon the kernel won't have access anymore to the userspace 191 * addresses where we're currently running on. We have to do that here 192 * because in 32bit we couldn't load a 64bit linear address. 193 */ 194 lgdt early_gdt_descr(%rip) 195 196 /* set up data segments */ 197 xorl %eax,%eax 198 movl %eax,%ds 199 movl %eax,%ss 200 movl %eax,%es 201 202 /* 203 * We don't really need to load %fs or %gs, but load them anyway 204 * to kill any stale realmode selectors. This allows execution 205 * under VT hardware. 206 */ 207 movl %eax,%fs 208 movl %eax,%gs 209 210 /* Set up %gs. 211 * 212 * The base of %gs always points to fixed_percpu_data. If the 213 * stack protector canary is enabled, it is located at %gs:40. 214 * Note that, on SMP, the boot cpu uses init data section until 215 * the per cpu areas are set up. 216 */ 217 movl $MSR_GS_BASE,%ecx 218 movl initial_gs(%rip),%eax 219 movl initial_gs+4(%rip),%edx 220 wrmsr 221 222 /* rsi is pointer to real mode structure with interesting info. 223 pass it to C */ 224 movq %rsi, %rdi 225 226.Ljump_to_C_code: 227 /* 228 * Jump to run C code and to be on a real kernel address. 229 * Since we are running on identity-mapped space we have to jump 230 * to the full 64bit address, this is only possible as indirect 231 * jump. In addition we need to ensure %cs is set so we make this 232 * a far return. 233 * 234 * Note: do not change to far jump indirect with 64bit offset. 235 * 236 * AMD does not support far jump indirect with 64bit offset. 237 * AMD64 Architecture Programmer's Manual, Volume 3: states only 238 * JMP FAR mem16:16 FF /5 Far jump indirect, 239 * with the target specified by a far pointer in memory. 240 * JMP FAR mem16:32 FF /5 Far jump indirect, 241 * with the target specified by a far pointer in memory. 242 * 243 * Intel64 does support 64bit offset. 244 * Software Developer Manual Vol 2: states: 245 * FF /5 JMP m16:16 Jump far, absolute indirect, 246 * address given in m16:16 247 * FF /5 JMP m16:32 Jump far, absolute indirect, 248 * address given in m16:32. 249 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect, 250 * address given in m16:64. 251 */ 252 pushq $.Lafter_lret # put return address on stack for unwinder 253 xorl %ebp, %ebp # clear frame pointer 254 movq initial_code(%rip), %rax 255 pushq $__KERNEL_CS # set correct cs 256 pushq %rax # target address in negative space 257 lretq 258.Lafter_lret: 259SYM_CODE_END(secondary_startup_64) 260 261#include "verify_cpu.S" 262 263#ifdef CONFIG_HOTPLUG_CPU 264/* 265 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set 266 * up already except stack. We just set up stack here. Then call 267 * start_secondary() via .Ljump_to_C_code. 268 */ 269SYM_CODE_START(start_cpu0) 270 UNWIND_HINT_EMPTY 271 movq initial_stack(%rip), %rsp 272 jmp .Ljump_to_C_code 273SYM_CODE_END(start_cpu0) 274#endif 275 276 /* Both SMP bootup and ACPI suspend change these variables */ 277 __REFDATA 278 .balign 8 279SYM_DATA(initial_code, .quad x86_64_start_kernel) 280SYM_DATA(initial_gs, .quad INIT_PER_CPU_VAR(fixed_percpu_data)) 281 282/* 283 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder 284 * reliably detect the end of the stack. 285 */ 286SYM_DATA(initial_stack, .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS) 287 __FINITDATA 288 289 __INIT 290SYM_CODE_START(early_idt_handler_array) 291 i = 0 292 .rept NUM_EXCEPTION_VECTORS 293 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0 294 UNWIND_HINT_IRET_REGS 295 pushq $0 # Dummy error code, to make stack frame uniform 296 .else 297 UNWIND_HINT_IRET_REGS offset=8 298 .endif 299 pushq $i # 72(%rsp) Vector number 300 jmp early_idt_handler_common 301 UNWIND_HINT_IRET_REGS 302 i = i + 1 303 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc 304 .endr 305 UNWIND_HINT_IRET_REGS offset=16 306SYM_CODE_END(early_idt_handler_array) 307 308SYM_CODE_START_LOCAL(early_idt_handler_common) 309 /* 310 * The stack is the hardware frame, an error code or zero, and the 311 * vector number. 312 */ 313 cld 314 315 incl early_recursion_flag(%rip) 316 317 /* The vector number is currently in the pt_regs->di slot. */ 318 pushq %rsi /* pt_regs->si */ 319 movq 8(%rsp), %rsi /* RSI = vector number */ 320 movq %rdi, 8(%rsp) /* pt_regs->di = RDI */ 321 pushq %rdx /* pt_regs->dx */ 322 pushq %rcx /* pt_regs->cx */ 323 pushq %rax /* pt_regs->ax */ 324 pushq %r8 /* pt_regs->r8 */ 325 pushq %r9 /* pt_regs->r9 */ 326 pushq %r10 /* pt_regs->r10 */ 327 pushq %r11 /* pt_regs->r11 */ 328 pushq %rbx /* pt_regs->bx */ 329 pushq %rbp /* pt_regs->bp */ 330 pushq %r12 /* pt_regs->r12 */ 331 pushq %r13 /* pt_regs->r13 */ 332 pushq %r14 /* pt_regs->r14 */ 333 pushq %r15 /* pt_regs->r15 */ 334 UNWIND_HINT_REGS 335 336 cmpq $14,%rsi /* Page fault? */ 337 jnz 10f 338 GET_CR2_INTO(%rdi) /* can clobber %rax if pv */ 339 call early_make_pgtable 340 andl %eax,%eax 341 jz 20f /* All good */ 342 34310: 344 movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */ 345 call early_fixup_exception 346 34720: 348 decl early_recursion_flag(%rip) 349 jmp restore_regs_and_return_to_kernel 350SYM_CODE_END(early_idt_handler_common) 351 352 353#define SYM_DATA_START_PAGE_ALIGNED(name) \ 354 SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE) 355 356#ifdef CONFIG_PAGE_TABLE_ISOLATION 357/* 358 * Each PGD needs to be 8k long and 8k aligned. We do not 359 * ever go out to userspace with these, so we do not 360 * strictly *need* the second page, but this allows us to 361 * have a single set_pgd() implementation that does not 362 * need to worry about whether it has 4k or 8k to work 363 * with. 364 * 365 * This ensures PGDs are 8k long: 366 */ 367#define PTI_USER_PGD_FILL 512 368/* This ensures they are 8k-aligned: */ 369#define SYM_DATA_START_PTI_ALIGNED(name) \ 370 SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE) 371#else 372#define SYM_DATA_START_PTI_ALIGNED(name) \ 373 SYM_DATA_START_PAGE_ALIGNED(name) 374#define PTI_USER_PGD_FILL 0 375#endif 376 377/* Automate the creation of 1 to 1 mapping pmd entries */ 378#define PMDS(START, PERM, COUNT) \ 379 i = 0 ; \ 380 .rept (COUNT) ; \ 381 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \ 382 i = i + 1 ; \ 383 .endr 384 385 __INITDATA 386 .balign 4 387 388SYM_DATA_START_PTI_ALIGNED(early_top_pgt) 389 .fill 512,8,0 390 .fill PTI_USER_PGD_FILL,8,0 391SYM_DATA_END(early_top_pgt) 392 393SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts) 394 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0 395SYM_DATA_END(early_dynamic_pgts) 396 397SYM_DATA(early_recursion_flag, .long 0) 398 399 .data 400 401#if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH) 402SYM_DATA_START_PTI_ALIGNED(init_top_pgt) 403 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 404 .org init_top_pgt + L4_PAGE_OFFSET*8, 0 405 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 406 .org init_top_pgt + L4_START_KERNEL*8, 0 407 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ 408 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 409 .fill PTI_USER_PGD_FILL,8,0 410SYM_DATA_END(init_top_pgt) 411 412SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt) 413 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 414 .fill 511, 8, 0 415SYM_DATA_END(level3_ident_pgt) 416SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt) 417 /* 418 * Since I easily can, map the first 1G. 419 * Don't set NX because code runs from these pages. 420 * 421 * Note: This sets _PAGE_GLOBAL despite whether 422 * the CPU supports it or it is enabled. But, 423 * the CPU should ignore the bit. 424 */ 425 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD) 426SYM_DATA_END(level2_ident_pgt) 427#else 428SYM_DATA_START_PTI_ALIGNED(init_top_pgt) 429 .fill 512,8,0 430 .fill PTI_USER_PGD_FILL,8,0 431SYM_DATA_END(init_top_pgt) 432#endif 433 434#ifdef CONFIG_X86_5LEVEL 435SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt) 436 .fill 511,8,0 437 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 438SYM_DATA_END(level4_kernel_pgt) 439#endif 440 441SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt) 442 .fill L3_START_KERNEL,8,0 443 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */ 444 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 445 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 446SYM_DATA_END(level3_kernel_pgt) 447 448SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt) 449 /* 450 * 512 MB kernel mapping. We spend a full page on this pagetable 451 * anyway. 452 * 453 * The kernel code+data+bss must not be bigger than that. 454 * 455 * (NOTE: at +512MB starts the module area, see MODULES_VADDR. 456 * If you want to increase this then increase MODULES_VADDR 457 * too.) 458 * 459 * This table is eventually used by the kernel during normal 460 * runtime. Care must be taken to clear out undesired bits 461 * later, like _PAGE_RW or _PAGE_GLOBAL in some cases. 462 */ 463 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, 464 KERNEL_IMAGE_SIZE/PMD_SIZE) 465SYM_DATA_END(level2_kernel_pgt) 466 467SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt) 468 .fill (512 - 4 - FIXMAP_PMD_NUM),8,0 469 pgtno = 0 470 .rept (FIXMAP_PMD_NUM) 471 .quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \ 472 + _PAGE_TABLE_NOENC; 473 pgtno = pgtno + 1 474 .endr 475 /* 6 MB reserved space + a 2MB hole */ 476 .fill 4,8,0 477SYM_DATA_END(level2_fixmap_pgt) 478 479SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt) 480 .rept (FIXMAP_PMD_NUM) 481 .fill 512,8,0 482 .endr 483SYM_DATA_END(level1_fixmap_pgt) 484 485#undef PMDS 486 487 .data 488 .align 16 489 490SYM_DATA(early_gdt_descr, .word GDT_ENTRIES*8-1) 491SYM_DATA_LOCAL(early_gdt_descr_base, .quad INIT_PER_CPU_VAR(gdt_page)) 492 493 .align 16 494/* This must match the first entry in level2_kernel_pgt */ 495SYM_DATA(phys_base, .quad 0x0) 496EXPORT_SYMBOL(phys_base) 497 498#include "../../x86/xen/xen-head.S" 499 500 __PAGE_ALIGNED_BSS 501SYM_DATA_START_PAGE_ALIGNED(empty_zero_page) 502 .skip PAGE_SIZE 503SYM_DATA_END(empty_zero_page) 504EXPORT_SYMBOL(empty_zero_page) 505 506