1 /* SPDX-License-Identifier: GPL-2.0 */
3 * linux/arch/x86/kernel/head_64.S -- start in 32bit and switch to 64bit
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>
20 #include <asm/cache.h>
21 #include <asm/processor-flags.h>
22 #include <asm/percpu.h>
24 #include "../entry/calling.h"
25 #include <asm/export.h>
26 #include <asm/nospec-branch.h>
27 #include <asm/apicdef.h>
28 #include <asm/fixmap.h>
32 * We are not able to switch in one step to the final KERNEL ADDRESS SPACE
33 * because we need identity-mapped pages.
35 #define l4_index(x) (((x) >> 39) & 511)
36 #define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
38 L4_PAGE_OFFSET = l4_index(__PAGE_OFFSET_BASE_L4)
39 L4_START_KERNEL = l4_index(__START_KERNEL_map)
41 L3_START_KERNEL = pud_index(__START_KERNEL_map)
46 SYM_CODE_START_NOALIGN(startup_64)
47 UNWIND_HINT_END_OF_STACK
49 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
50 * and someone has loaded an identity mapped page table
51 * for us. These identity mapped page tables map all of the
52 * kernel pages and possibly all of memory.
54 * %RSI holds the physical address of the boot_params structure
55 * provided by the bootloader. Preserve it in %R15 so C function calls
56 * will not clobber it.
58 * We come here either directly from a 64bit bootloader, or from
59 * arch/x86/boot/compressed/head_64.S.
61 * We only come here initially at boot nothing else comes here.
63 * Since we may be loaded at an address different from what we were
64 * compiled to run at we first fixup the physical addresses in our page
65 * tables and then reload them.
69 /* Set up the stack for verify_cpu() */
70 leaq (__end_init_task - PTREGS_SIZE)(%rip), %rsp
72 leaq _text(%rip), %rdi
74 /* Setup GSBASE to allow stack canary access for C code */
75 movl $MSR_GS_BASE, %ecx
76 leaq INIT_PER_CPU_VAR(fixed_percpu_data)(%rip), %rdx
81 call startup_64_setup_env
83 /* Now switch to __KERNEL_CS so IRET works reliably */
85 leaq .Lon_kernel_cs(%rip), %rax
90 UNWIND_HINT_END_OF_STACK
92 #ifdef CONFIG_AMD_MEM_ENCRYPT
94 * Activate SEV/SME memory encryption if supported/enabled. This needs to
95 * be done now, since this also includes setup of the SEV-SNP CPUID table,
96 * which needs to be done before any CPUID instructions are executed in
97 * subsequent code. Pass the boot_params pointer as the first argument.
103 /* Sanitize CPU configuration */
107 * Perform pagetable fixups. Additionally, if SME is active, encrypt
108 * the kernel and retrieve the modifier (SME encryption mask if SME
109 * is active) to be added to the initial pgdir entry that will be
110 * programmed into CR3.
112 leaq _text(%rip), %rdi
116 /* Form the CR3 value being sure to include the CR3 modifier */
117 addq $(early_top_pgt - __START_KERNEL_map), %rax
119 SYM_CODE_END(startup_64)
121 SYM_CODE_START(secondary_startup_64)
122 UNWIND_HINT_END_OF_STACK
125 * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 0,
126 * and someone has loaded a mapped page table.
128 * We come here either from startup_64 (using physical addresses)
129 * or from trampoline.S (using virtual addresses).
131 * Using virtual addresses from trampoline.S removes the need
132 * to have any identity mapped pages in the kernel page table
133 * after the boot processor executes this code.
136 /* Sanitize CPU configuration */
140 * The secondary_startup_64_no_verify entry point is only used by
141 * SEV-ES guests. In those guests the call to verify_cpu() would cause
142 * #VC exceptions which can not be handled at this stage of secondary
145 * All non SEV-ES systems, especially Intel systems, need to execute
146 * verify_cpu() above to make sure NX is enabled.
148 SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL)
149 UNWIND_HINT_END_OF_STACK
152 /* Clear %R15 which holds the boot_params pointer on the boot CPU */
156 * Retrieve the modifier (SME encryption mask if SME is active) to be
157 * added to the initial pgdir entry that will be programmed into CR3.
159 #ifdef CONFIG_AMD_MEM_ENCRYPT
160 movq sme_me_mask, %rax
165 /* Form the CR3 value being sure to include the CR3 modifier */
166 addq $(init_top_pgt - __START_KERNEL_map), %rax
169 #ifdef CONFIG_X86_MCE
171 * Preserve CR4.MCE if the kernel will enable #MC support.
172 * Clearing MCE may fault in some environments (that also force #MC
173 * support). Any machine check that occurs before #MC support is fully
174 * configured will crash the system regardless of the CR4.MCE value set
178 andl $X86_CR4_MCE, %ecx
183 /* Enable PAE mode, PGE and LA57 */
184 orl $(X86_CR4_PAE | X86_CR4_PGE), %ecx
185 #ifdef CONFIG_X86_5LEVEL
186 testl $1, __pgtable_l5_enabled(%rip)
188 orl $X86_CR4_LA57, %ecx
193 /* Setup early boot stage 4-/5-level pagetables. */
194 addq phys_base(%rip), %rax
197 * For SEV guests: Verify that the C-bit is correct. A malicious
198 * hypervisor could lie about the C-bit position to perform a ROP
199 * attack on the guest by writing to the unencrypted stack and wait for
200 * the next RET instruction.
206 * Switch to new page-table
208 * For the boot CPU this switches to early_top_pgt which still has the
209 * indentity mappings present. The secondary CPUs will switch to the
210 * init_top_pgt here, away from the trampoline_pgd and unmap the
211 * indentity mapped ranges.
216 * Do a global TLB flush after the CR3 switch to make sure the TLB
217 * entries from the identity mapping are flushed.
221 xorq $X86_CR4_PGE, %rcx
225 /* Ensure I am executing from virtual addresses */
227 ANNOTATE_RETPOLINE_SAFE
230 UNWIND_HINT_END_OF_STACK
231 ANNOTATE_NOENDBR // above
235 * For parallel boot, the APIC ID is read from the APIC, and then
236 * used to look up the CPU number. For booting a single CPU, the
237 * CPU number is encoded in smpboot_control.
239 * Bit 31 STARTUP_READ_APICID (Read APICID from APIC)
240 * Bit 0-23 CPU# if STARTUP_xx flags are not set
242 movl smpboot_control(%rip), %ecx
243 testl $STARTUP_READ_APICID, %ecx
246 * No control bit set, single CPU bringup. CPU number is provided
247 * in bit 0-23. This is also the boot CPU case (CPU number 0).
249 andl $(~STARTUP_PARALLEL_MASK), %ecx
253 /* Check whether X2APIC mode is already enabled */
254 mov $MSR_IA32_APICBASE, %ecx
256 testl $X2APIC_ENABLE, %eax
257 jnz .Lread_apicid_msr
259 /* Read the APIC ID from the fix-mapped MMIO space. */
260 movq apic_mmio_base(%rip), %rcx
267 mov $APIC_X2APIC_ID_MSR, %ecx
271 /* EAX contains the APIC ID of the current CPU */
273 leaq cpuid_to_apicid(%rip), %rbx
276 cmpl (%rbx,%rcx,4), %eax
279 #ifdef CONFIG_FORCE_NR_CPUS
282 cmpl nr_cpu_ids(%rip), %ecx
286 /* APIC ID not found in the table. Drop the trampoline lock and bail. */
287 movq trampoline_lock(%rip), %rax
295 /* Get the per cpu offset for the given CPU# which is in ECX */
296 movq __per_cpu_offset(,%rcx,8), %rdx
298 xorl %edx, %edx /* zero-extended to clear all of RDX */
299 #endif /* CONFIG_SMP */
302 * Setup a boot time stack - Any secondary CPU will have lost its stack
303 * by now because the cr3-switch above unmaps the real-mode stack.
305 * RDX contains the per-cpu offset
307 movq pcpu_hot + X86_current_task(%rdx), %rax
308 movq TASK_threadsp(%rax), %rsp
311 * Now that this CPU is running on its own stack, drop the realmode
312 * protection. For the boot CPU the pointer is NULL!
314 movq trampoline_lock(%rip), %rax
321 * We must switch to a new descriptor in kernel space for the GDT
322 * because soon the kernel won't have access anymore to the userspace
323 * addresses where we're currently running on. We have to do that here
324 * because in 32bit we couldn't load a 64bit linear address.
327 movw $(GDT_SIZE-1), (%rsp)
328 leaq gdt_page(%rdx), %rax
333 /* set up data segments */
340 * We don't really need to load %fs or %gs, but load them anyway
341 * to kill any stale realmode selectors. This allows execution
349 * The base of %gs always points to fixed_percpu_data. If the
350 * stack protector canary is enabled, it is located at %gs:40.
351 * Note that, on SMP, the boot cpu uses init data section until
352 * the per cpu areas are set up.
354 movl $MSR_GS_BASE,%ecx
356 leaq INIT_PER_CPU_VAR(fixed_percpu_data)(%rip), %rdx
362 /* Setup and Load IDT */
365 /* Check if nx is implemented */
366 movl $0x80000001, %eax
370 /* Setup EFER (Extended Feature Enable Register) */
374 * Preserve current value of EFER for comparison and to skip
375 * EFER writes if no change was made (for TDX guest)
378 btsl $_EFER_SCE, %eax /* Enable System Call */
379 btl $20,%edi /* No Execute supported? */
382 btsq $_PAGE_BIT_NX,early_pmd_flags(%rip)
384 /* Avoid writing EFER if no change was made (for TDX guest) */
388 wrmsr /* Make changes effective */
391 movl $CR0_STATE, %eax
392 /* Make changes effective */
395 /* zero EFLAGS after setting rsp */
399 /* Pass the boot_params pointer as first argument */
404 * Jump to run C code and to be on a real kernel address.
405 * Since we are running on identity-mapped space we have to jump
406 * to the full 64bit address, this is only possible as indirect
407 * jump. In addition we need to ensure %cs is set so we make this
410 * Note: do not change to far jump indirect with 64bit offset.
412 * AMD does not support far jump indirect with 64bit offset.
413 * AMD64 Architecture Programmer's Manual, Volume 3: states only
414 * JMP FAR mem16:16 FF /5 Far jump indirect,
415 * with the target specified by a far pointer in memory.
416 * JMP FAR mem16:32 FF /5 Far jump indirect,
417 * with the target specified by a far pointer in memory.
419 * Intel64 does support 64bit offset.
420 * Software Developer Manual Vol 2: states:
421 * FF /5 JMP m16:16 Jump far, absolute indirect,
422 * address given in m16:16
423 * FF /5 JMP m16:32 Jump far, absolute indirect,
424 * address given in m16:32.
425 * REX.W + FF /5 JMP m16:64 Jump far, absolute indirect,
426 * address given in m16:64.
428 pushq $.Lafter_lret # put return address on stack for unwinder
429 xorl %ebp, %ebp # clear frame pointer
430 movq initial_code(%rip), %rax
431 pushq $__KERNEL_CS # set correct cs
432 pushq %rax # target address in negative space
436 SYM_CODE_END(secondary_startup_64)
438 #include "verify_cpu.S"
439 #include "sev_verify_cbit.S"
441 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_AMD_MEM_ENCRYPT)
443 * Entry point for soft restart of a CPU. Invoked from xxx_play_dead() for
444 * restarting the boot CPU or for restarting SEV guest CPUs after CPU hot
445 * unplug. Everything is set up already except the stack.
447 SYM_CODE_START(soft_restart_cpu)
449 UNWIND_HINT_END_OF_STACK
451 /* Find the idle task stack */
452 movq PER_CPU_VAR(pcpu_hot) + X86_current_task, %rcx
453 movq TASK_threadsp(%rcx), %rsp
456 SYM_CODE_END(soft_restart_cpu)
459 #ifdef CONFIG_AMD_MEM_ENCRYPT
461 * VC Exception handler used during early boot when running on kernel
462 * addresses, but before the switch to the idt_table can be made.
463 * The early_idt_handler_array can't be used here because it calls into a lot
464 * of __init code and this handler is also used during CPU offlining/onlining.
465 * Therefore this handler ends up in the .text section so that it stays around
466 * when .init.text is freed.
468 SYM_CODE_START_NOALIGN(vc_boot_ghcb)
469 UNWIND_HINT_IRET_REGS offset=8
477 movq ORIG_RAX(%rsp), %rsi
478 movq initial_vc_handler(%rip), %rax
479 ANNOTATE_RETPOLINE_SAFE
485 /* Remove Error Code */
489 SYM_CODE_END(vc_boot_ghcb)
492 /* Both SMP bootup and ACPI suspend change these variables */
495 SYM_DATA(initial_code, .quad x86_64_start_kernel)
496 #ifdef CONFIG_AMD_MEM_ENCRYPT
497 SYM_DATA(initial_vc_handler, .quad handle_vc_boot_ghcb)
500 SYM_DATA(trampoline_lock, .quad 0);
504 SYM_CODE_START(early_idt_handler_array)
506 .rept NUM_EXCEPTION_VECTORS
507 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
508 UNWIND_HINT_IRET_REGS
510 pushq $0 # Dummy error code, to make stack frame uniform
512 UNWIND_HINT_IRET_REGS offset=8
515 pushq $i # 72(%rsp) Vector number
516 jmp early_idt_handler_common
517 UNWIND_HINT_IRET_REGS
519 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
521 SYM_CODE_END(early_idt_handler_array)
522 ANNOTATE_NOENDBR // early_idt_handler_array[NUM_EXCEPTION_VECTORS]
524 SYM_CODE_START_LOCAL(early_idt_handler_common)
525 UNWIND_HINT_IRET_REGS offset=16
527 * The stack is the hardware frame, an error code or zero, and the
532 incl early_recursion_flag(%rip)
534 /* The vector number is currently in the pt_regs->di slot. */
535 pushq %rsi /* pt_regs->si */
536 movq 8(%rsp), %rsi /* RSI = vector number */
537 movq %rdi, 8(%rsp) /* pt_regs->di = RDI */
538 pushq %rdx /* pt_regs->dx */
539 pushq %rcx /* pt_regs->cx */
540 pushq %rax /* pt_regs->ax */
541 pushq %r8 /* pt_regs->r8 */
542 pushq %r9 /* pt_regs->r9 */
543 pushq %r10 /* pt_regs->r10 */
544 pushq %r11 /* pt_regs->r11 */
545 pushq %rbx /* pt_regs->bx */
546 pushq %rbp /* pt_regs->bp */
547 pushq %r12 /* pt_regs->r12 */
548 pushq %r13 /* pt_regs->r13 */
549 pushq %r14 /* pt_regs->r14 */
550 pushq %r15 /* pt_regs->r15 */
553 movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */
554 call do_early_exception
556 decl early_recursion_flag(%rip)
557 jmp restore_regs_and_return_to_kernel
558 SYM_CODE_END(early_idt_handler_common)
560 #ifdef CONFIG_AMD_MEM_ENCRYPT
562 * VC Exception handler used during very early boot. The
563 * early_idt_handler_array can't be used because it returns via the
564 * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early.
566 * XXX it does, fix this.
568 * This handler will end up in the .init.text section and not be
569 * available to boot secondary CPUs.
571 SYM_CODE_START_NOALIGN(vc_no_ghcb)
572 UNWIND_HINT_IRET_REGS offset=8
580 movq ORIG_RAX(%rsp), %rsi
586 /* Remove Error Code */
589 /* Pure iret required here - don't use INTERRUPT_RETURN */
591 SYM_CODE_END(vc_no_ghcb)
594 #define SYM_DATA_START_PAGE_ALIGNED(name) \
595 SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
597 #ifdef CONFIG_PAGE_TABLE_ISOLATION
599 * Each PGD needs to be 8k long and 8k aligned. We do not
600 * ever go out to userspace with these, so we do not
601 * strictly *need* the second page, but this allows us to
602 * have a single set_pgd() implementation that does not
603 * need to worry about whether it has 4k or 8k to work
606 * This ensures PGDs are 8k long:
608 #define PTI_USER_PGD_FILL 512
609 /* This ensures they are 8k-aligned: */
610 #define SYM_DATA_START_PTI_ALIGNED(name) \
611 SYM_START(name, SYM_L_GLOBAL, .balign 2 * PAGE_SIZE)
613 #define SYM_DATA_START_PTI_ALIGNED(name) \
614 SYM_DATA_START_PAGE_ALIGNED(name)
615 #define PTI_USER_PGD_FILL 0
618 /* Automate the creation of 1 to 1 mapping pmd entries */
619 #define PMDS(START, PERM, COUNT) \
622 .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
629 SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
631 .fill PTI_USER_PGD_FILL,8,0
632 SYM_DATA_END(early_top_pgt)
634 SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts)
635 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,8,0
636 SYM_DATA_END(early_dynamic_pgts)
638 SYM_DATA(early_recursion_flag, .long 0)
642 #if defined(CONFIG_XEN_PV) || defined(CONFIG_PVH)
643 SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
644 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
645 .org init_top_pgt + L4_PAGE_OFFSET*8, 0
646 .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
647 .org init_top_pgt + L4_START_KERNEL*8, 0
648 /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
649 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
650 .fill PTI_USER_PGD_FILL,8,0
651 SYM_DATA_END(init_top_pgt)
653 SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
654 .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
656 SYM_DATA_END(level3_ident_pgt)
657 SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
659 * Since I easily can, map the first 1G.
660 * Don't set NX because code runs from these pages.
662 * Note: This sets _PAGE_GLOBAL despite whether
663 * the CPU supports it or it is enabled. But,
664 * the CPU should ignore the bit.
666 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
667 SYM_DATA_END(level2_ident_pgt)
669 SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
671 .fill PTI_USER_PGD_FILL,8,0
672 SYM_DATA_END(init_top_pgt)
675 #ifdef CONFIG_X86_5LEVEL
676 SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
678 .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
679 SYM_DATA_END(level4_kernel_pgt)
682 SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
683 .fill L3_START_KERNEL,8,0
684 /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
685 .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
686 .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
687 SYM_DATA_END(level3_kernel_pgt)
689 SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
691 * Kernel high mapping.
693 * The kernel code+data+bss must be located below KERNEL_IMAGE_SIZE in
694 * virtual address space, which is 1 GiB if RANDOMIZE_BASE is enabled,
697 * (NOTE: after that starts the module area, see MODULES_VADDR.)
699 * This table is eventually used by the kernel during normal runtime.
700 * Care must be taken to clear out undesired bits later, like _PAGE_RW
701 * or _PAGE_GLOBAL in some cases.
703 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERNEL_IMAGE_SIZE/PMD_SIZE)
704 SYM_DATA_END(level2_kernel_pgt)
706 SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
707 .fill (512 - 4 - FIXMAP_PMD_NUM),8,0
709 .rept (FIXMAP_PMD_NUM)
710 .quad level1_fixmap_pgt + (pgtno << PAGE_SHIFT) - __START_KERNEL_map \
714 /* 6 MB reserved space + a 2MB hole */
716 SYM_DATA_END(level2_fixmap_pgt)
718 SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
719 .rept (FIXMAP_PMD_NUM)
722 SYM_DATA_END(level1_fixmap_pgt)
729 SYM_DATA(smpboot_control, .long 0)
732 /* This must match the first entry in level2_kernel_pgt */
733 SYM_DATA(phys_base, .quad 0x0)
734 EXPORT_SYMBOL(phys_base)
736 #include "../../x86/xen/xen-head.S"
739 SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
741 SYM_DATA_END(empty_zero_page)
742 EXPORT_SYMBOL(empty_zero_page)