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 <asm/segment.h> 17#include <asm/pgtable.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 28#ifdef CONFIG_PARAVIRT 29#include <asm/asm-offsets.h> 30#include <asm/paravirt.h> 31#define GET_CR2_INTO(reg) GET_CR2_INTO_RAX ; movq %rax, reg 32#else 33#define GET_CR2_INTO(reg) movq %cr2, reg 34#define INTERRUPT_RETURN iretq 35#endif 36 37/* we are not able to switch in one step to the final KERNEL ADDRESS SPACE 38 * because we need identity-mapped pages. 39 * 40 */ 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 53 .globl startup_64 54startup_64: 55 UNWIND_HINT_EMPTY 56 /* 57 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0, 58 * and someone has loaded an identity mapped page table 59 * for us. These identity mapped page tables map all of the 60 * kernel pages and possibly all of memory. 61 * 62 * %rsi holds a physical pointer to real_mode_data. 63 * 64 * We come here either directly from a 64bit bootloader, or from 65 * arch/x86/boot/compressed/head_64.S. 66 * 67 * We only come here initially at boot nothing else comes here. 68 * 69 * Since we may be loaded at an address different from what we were 70 * compiled to run at we first fixup the physical addresses in our page 71 * tables and then reload them. 72 */ 73 74 /* Set up the stack for verify_cpu(), similar to initial_stack below */ 75 leaq (__end_init_task - SIZEOF_PTREGS)(%rip), %rsp 76 77 /* Sanitize CPU configuration */ 78 call verify_cpu 79 80 /* 81 * Perform pagetable fixups. Additionally, if SME is active, encrypt 82 * the kernel and retrieve the modifier (SME encryption mask if SME 83 * is active) to be added to the initial pgdir entry that will be 84 * programmed into CR3. 85 */ 86 leaq _text(%rip), %rdi 87 pushq %rsi 88 call __startup_64 89 popq %rsi 90 91 /* Form the CR3 value being sure to include the CR3 modifier */ 92 addq $(early_top_pgt - __START_KERNEL_map), %rax 93 jmp 1f 94ENTRY(secondary_startup_64) 95 UNWIND_HINT_EMPTY 96 /* 97 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0, 98 * and someone has loaded a mapped page table. 99 * 100 * %rsi holds a physical pointer to real_mode_data. 101 * 102 * We come here either from startup_64 (using physical addresses) 103 * or from trampoline.S (using virtual addresses). 104 * 105 * Using virtual addresses from trampoline.S removes the need 106 * to have any identity mapped pages in the kernel page table 107 * after the boot processor executes this code. 108 */ 109 110 /* Sanitize CPU configuration */ 111 call verify_cpu 112 113 /* 114 * Retrieve the modifier (SME encryption mask if SME is active) to be 115 * added to the initial pgdir entry that will be programmed into CR3. 116 */ 117 pushq %rsi 118 call __startup_secondary_64 119 popq %rsi 120 121 /* Form the CR3 value being sure to include the CR3 modifier */ 122 addq $(init_top_pgt - __START_KERNEL_map), %rax 1231: 124 125 /* Enable PAE mode, PGE and LA57 */ 126 movl $(X86_CR4_PAE | X86_CR4_PGE), %ecx 127#ifdef CONFIG_X86_5LEVEL 128 testl $1, __pgtable_l5_enabled(%rip) 129 jz 1f 130 orl $X86_CR4_LA57, %ecx 1311: 132#endif 133 movq %rcx, %cr4 134 135 /* Setup early boot stage 4-/5-level pagetables. */ 136 addq phys_base(%rip), %rax 137 movq %rax, %cr3 138 139 /* Ensure I am executing from virtual addresses */ 140 movq $1f, %rax 141 ANNOTATE_RETPOLINE_SAFE 142 jmp *%rax 1431: 144 UNWIND_HINT_EMPTY 145 146 /* Check if nx is implemented */ 147 movl $0x80000001, %eax 148 cpuid 149 movl %edx,%edi 150 151 /* Setup EFER (Extended Feature Enable Register) */ 152 movl $MSR_EFER, %ecx 153 rdmsr 154 btsl $_EFER_SCE, %eax /* Enable System Call */ 155 btl $20,%edi /* No Execute supported? */ 156 jnc 1f 157 btsl $_EFER_NX, %eax 158 btsq $_PAGE_BIT_NX,early_pmd_flags(%rip) 1591: wrmsr /* Make changes effective */ 160 161 /* Setup cr0 */ 162 movl $CR0_STATE, %eax 163 /* Make changes effective */ 164 movq %rax, %cr0 165 166 /* Setup a boot time stack */ 167 movq initial_stack(%rip), %rsp 168 169 /* zero EFLAGS after setting rsp */ 170 pushq $0 171 popfq 172 173 /* 174 * We must switch to a new descriptor in kernel space for the GDT 175 * because soon the kernel won't have access anymore to the userspace 176 * addresses where we're currently running on. We have to do that here 177 * because in 32bit we couldn't load a 64bit linear address. 178 */ 179 lgdt early_gdt_descr(%rip) 180 181 /* set up data segments */ 182 xorl %eax,%eax 183 movl %eax,%ds 184 movl %eax,%ss 185 movl %eax,%es 186 187 /* 188 * We don't really need to load %fs or %gs, but load them anyway 189 * to kill any stale realmode selectors. This allows execution 190 * under VT hardware. 191 */ 192 movl %eax,%fs 193 movl %eax,%gs 194 195 /* Set up %gs. 196 * 197 * The base of %gs always points to the bottom of the irqstack 198 * union. If the stack protector canary is enabled, it is 199 * located at %gs:40. Note that, on SMP, the boot cpu uses 200 * init data section till per cpu areas are set up. 201 */ 202 movl $MSR_GS_BASE,%ecx 203 movl initial_gs(%rip),%eax 204 movl initial_gs+4(%rip),%edx 205 wrmsr 206 207 /* rsi is pointer to real mode structure with interesting info. 208 pass it to C */ 209 movq %rsi, %rdi 210 211.Ljump_to_C_code: 212 /* 213 * Jump to run C code and to be on a real kernel address. 214 * Since we are running on identity-mapped space we have to jump 215 * to the full 64bit address, this is only possible as indirect 216 * jump. In addition we need to ensure %cs is set so we make this 217 * a far return. 218 * 219 * Note: do not change to far jump indirect with 64bit offset. 220 * 221 * AMD does not support far jump indirect with 64bit offset. 222 * AMD64 Architecture Programmer's Manual, Volume 3: states only 223 * JMP FAR mem16:16 FF /5 Far jump indirect, 224 * with the target specified by a far pointer in memory. 225 * JMP FAR mem16:32 FF /5 Far jump indirect, 226 * with the target specified by a far pointer in memory. 227 * 228 * Intel64 does support 64bit offset. 229 * Software Developer Manual Vol 2: states: 230 * FF /5 JMP m16:16 Jump far, absolute indirect, 231 * address given in m16:16 232 * FF /5 JMP m16:32 Jump far, absolute indirect, 233 * address given in m16:32. 234 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect, 235 * address given in m16:64. 236 */ 237 pushq $.Lafter_lret # put return address on stack for unwinder 238 xorl %ebp, %ebp # clear frame pointer 239 movq initial_code(%rip), %rax 240 pushq $__KERNEL_CS # set correct cs 241 pushq %rax # target address in negative space 242 lretq 243.Lafter_lret: 244END(secondary_startup_64) 245 246#include "verify_cpu.S" 247 248#ifdef CONFIG_HOTPLUG_CPU 249/* 250 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set 251 * up already except stack. We just set up stack here. Then call 252 * start_secondary() via .Ljump_to_C_code. 253 */ 254ENTRY(start_cpu0) 255 movq initial_stack(%rip), %rsp 256 UNWIND_HINT_EMPTY 257 jmp .Ljump_to_C_code 258ENDPROC(start_cpu0) 259#endif 260 261 /* Both SMP bootup and ACPI suspend change these variables */ 262 __REFDATA 263 .balign 8 264 GLOBAL(initial_code) 265 .quad x86_64_start_kernel 266 GLOBAL(initial_gs) 267 .quad INIT_PER_CPU_VAR(irq_stack_union) 268 GLOBAL(initial_stack) 269 /* 270 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel 271 * unwinder reliably detect the end of the stack. 272 */ 273 .quad init_thread_union + THREAD_SIZE - SIZEOF_PTREGS 274 __FINITDATA 275 276 __INIT 277ENTRY(early_idt_handler_array) 278 i = 0 279 .rept NUM_EXCEPTION_VECTORS 280 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0 281 UNWIND_HINT_IRET_REGS 282 pushq $0 # Dummy error code, to make stack frame uniform 283 .else 284 UNWIND_HINT_IRET_REGS offset=8 285 .endif 286 pushq $i # 72(%rsp) Vector number 287 jmp early_idt_handler_common 288 UNWIND_HINT_IRET_REGS 289 i = i + 1 290 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc 291 .endr 292 UNWIND_HINT_IRET_REGS offset=16 293END(early_idt_handler_array) 294 295early_idt_handler_common: 296 /* 297 * The stack is the hardware frame, an error code or zero, and the 298 * vector number. 299 */ 300 cld 301 302 incl early_recursion_flag(%rip) 303 304 /* The vector number is currently in the pt_regs->di slot. */ 305 pushq %rsi /* pt_regs->si */ 306 movq 8(%rsp), %rsi /* RSI = vector number */ 307 movq %rdi, 8(%rsp) /* pt_regs->di = RDI */ 308 pushq %rdx /* pt_regs->dx */ 309 pushq %rcx /* pt_regs->cx */ 310 pushq %rax /* pt_regs->ax */ 311 pushq %r8 /* pt_regs->r8 */ 312 pushq %r9 /* pt_regs->r9 */ 313 pushq %r10 /* pt_regs->r10 */ 314 pushq %r11 /* pt_regs->r11 */ 315 pushq %rbx /* pt_regs->bx */ 316 pushq %rbp /* pt_regs->bp */ 317 pushq %r12 /* pt_regs->r12 */ 318 pushq %r13 /* pt_regs->r13 */ 319 pushq %r14 /* pt_regs->r14 */ 320 pushq %r15 /* pt_regs->r15 */ 321 UNWIND_HINT_REGS 322 323 cmpq $14,%rsi /* Page fault? */ 324 jnz 10f 325 GET_CR2_INTO(%rdi) /* Can clobber any volatile register if pv */ 326 call early_make_pgtable 327 andl %eax,%eax 328 jz 20f /* All good */ 329 33010: 331 movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */ 332 call early_fixup_exception 333 33420: 335 decl early_recursion_flag(%rip) 336 jmp restore_regs_and_return_to_kernel 337END(early_idt_handler_common) 338 339 __INITDATA 340 341 .balign 4 342GLOBAL(early_recursion_flag) 343 .long 0 344 345#define NEXT_PAGE(name) \ 346 .balign PAGE_SIZE; \ 347GLOBAL(name) 348 349#ifdef CONFIG_PAGE_TABLE_ISOLATION 350/* 351 * Each PGD needs to be 8k long and 8k aligned. We do not 352 * ever go out to userspace with these, so we do not 353 * strictly *need* the second page, but this allows us to 354 * have a single set_pgd() implementation that does not 355 * need to worry about whether it has 4k or 8k to work 356 * with. 357 * 358 * This ensures PGDs are 8k long: 359 */ 360#define PTI_USER_PGD_FILL 512 361/* This ensures they are 8k-aligned: */ 362#define NEXT_PGD_PAGE(name) \ 363 .balign 2 * PAGE_SIZE; \ 364GLOBAL(name) 365#else 366#define NEXT_PGD_PAGE(name) NEXT_PAGE(name) 367#define PTI_USER_PGD_FILL 0 368#endif 369 370/* Automate the creation of 1 to 1 mapping pmd entries */ 371#define PMDS(START, PERM, COUNT) \ 372 i = 0 ; \ 373 .rept (COUNT) ; \ 374 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \ 375 i = i + 1 ; \ 376 .endr 377 378 __INITDATA 379NEXT_PGD_PAGE(early_top_pgt) 380 .fill 512,8,0 381 .fill PTI_USER_PGD_FILL,8,0 382 383NEXT_PAGE(early_dynamic_pgts) 384 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0 385 386 .data 387 388#if defined(CONFIG_XEN_PV) || defined(CONFIG_XEN_PVH) 389NEXT_PGD_PAGE(init_top_pgt) 390 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 391 .org init_top_pgt + L4_PAGE_OFFSET*8, 0 392 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 393 .org init_top_pgt + L4_START_KERNEL*8, 0 394 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */ 395 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 396 .fill PTI_USER_PGD_FILL,8,0 397 398NEXT_PAGE(level3_ident_pgt) 399 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 400 .fill 511, 8, 0 401NEXT_PAGE(level2_ident_pgt) 402 /* 403 * Since I easily can, map the first 1G. 404 * Don't set NX because code runs from these pages. 405 * 406 * Note: This sets _PAGE_GLOBAL despite whether 407 * the CPU supports it or it is enabled. But, 408 * the CPU should ignore the bit. 409 */ 410 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD) 411#else 412NEXT_PGD_PAGE(init_top_pgt) 413 .fill 512,8,0 414 .fill PTI_USER_PGD_FILL,8,0 415#endif 416 417#ifdef CONFIG_X86_5LEVEL 418NEXT_PAGE(level4_kernel_pgt) 419 .fill 511,8,0 420 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 421#endif 422 423NEXT_PAGE(level3_kernel_pgt) 424 .fill L3_START_KERNEL,8,0 425 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */ 426 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC 427 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 428 429NEXT_PAGE(level2_kernel_pgt) 430 /* 431 * 512 MB kernel mapping. We spend a full page on this pagetable 432 * anyway. 433 * 434 * The kernel code+data+bss must not be bigger than that. 435 * 436 * (NOTE: at +512MB starts the module area, see MODULES_VADDR. 437 * If you want to increase this then increase MODULES_VADDR 438 * too.) 439 * 440 * This table is eventually used by the kernel during normal 441 * runtime. Care must be taken to clear out undesired bits 442 * later, like _PAGE_RW or _PAGE_GLOBAL in some cases. 443 */ 444 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, 445 KERNEL_IMAGE_SIZE/PMD_SIZE) 446 447NEXT_PAGE(level2_fixmap_pgt) 448 .fill 506,8,0 449 .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC 450 /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */ 451 .fill 5,8,0 452 453NEXT_PAGE(level1_fixmap_pgt) 454 .fill 512,8,0 455 456#undef PMDS 457 458 .data 459 .align 16 460 .globl early_gdt_descr 461early_gdt_descr: 462 .word GDT_ENTRIES*8-1 463early_gdt_descr_base: 464 .quad INIT_PER_CPU_VAR(gdt_page) 465 466ENTRY(phys_base) 467 /* This must match the first entry in level2_kernel_pgt */ 468 .quad 0x0000000000000000 469EXPORT_SYMBOL(phys_base) 470 471#include "../../x86/xen/xen-head.S" 472 473 __PAGE_ALIGNED_BSS 474NEXT_PAGE(empty_zero_page) 475 .skip PAGE_SIZE 476EXPORT_SYMBOL(empty_zero_page) 477 478