1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * linux/arch/x86/kernel/head_64.S -- start i
4 *
5 * Copyright (C) 2000 Andrea Arcangeli <andrea
6 * Copyright (C) 2000 Pavel Machek <pavel@suse
7 * Copyright (C) 2000 Karsten Keil <kkeil@suse
8 * Copyright (C) 2001,2002 Andi Kleen <ak@suse
9 * Copyright (C) 2005 Eric Biederman <ebiederm
10 */
11
12 #include <linux/export.h>
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/nospec-branch.h>
26 #include <asm/apicdef.h>
27 #include <asm/fixmap.h>
28 #include <asm/smp.h>
29 #include <asm/thread_info.h>
30
31 /*
32 * We are not able to switch in one step to th
33 * because we need identity-mapped pages.
34 */
35
36 __HEAD
37 .code64
38 SYM_CODE_START_NOALIGN(startup_64)
39 UNWIND_HINT_END_OF_STACK
40 /*
41 * At this point the CPU runs in 64bit
42 * and someone has loaded an identity
43 * for us. These identity mapped page
44 * kernel pages and possibly all of me
45 *
46 * %RSI holds the physical address of
47 * provided by the bootloader. Preserv
48 * will not clobber it.
49 *
50 * We come here either directly from a
51 * arch/x86/boot/compressed/head_64.S.
52 *
53 * We only come here initially at boot
54 *
55 * Since we may be loaded at an addres
56 * compiled to run at we first fixup t
57 * tables and then reload them.
58 */
59 mov %rsi, %r15
60
61 /* Set up the stack for verify_cpu() *
62 leaq __top_init_kernel_stack(%rip),
63
64 /* Setup GSBASE to allow stack canary
65 movl $MSR_GS_BASE, %ecx
66 leaq INIT_PER_CPU_VAR(fixed_percpu_
67 movl %edx, %eax
68 shrq $32, %rdx
69 wrmsr
70
71 call startup_64_setup_gdt_idt
72
73 /* Now switch to __KERNEL_CS so IRET w
74 pushq $__KERNEL_CS
75 leaq .Lon_kernel_cs(%rip), %rax
76 pushq %rax
77 lretq
78
79 .Lon_kernel_cs:
80 UNWIND_HINT_END_OF_STACK
81
82 #ifdef CONFIG_AMD_MEM_ENCRYPT
83 /*
84 * Activate SEV/SME memory encryption
85 * be done now, since this also includ
86 * which needs to be done before any C
87 * subsequent code. Pass the boot_para
88 */
89 movq %r15, %rdi
90 call sme_enable
91 #endif
92
93 /* Sanitize CPU configuration */
94 call verify_cpu
95
96 /*
97 * Perform pagetable fixups. Additiona
98 * the kernel and retrieve the modifie
99 * is active) to be added to the initi
100 * programmed into CR3.
101 */
102 leaq _text(%rip), %rdi
103 movq %r15, %rsi
104 call __startup_64
105
106 /* Form the CR3 value being sure to in
107 leaq early_top_pgt(%rip), %rcx
108 addq %rcx, %rax
109
110 #ifdef CONFIG_AMD_MEM_ENCRYPT
111 mov %rax, %rdi
112
113 /*
114 * For SEV guests: Verify that the C-b
115 * hypervisor could lie about the C-bi
116 * attack on the guest by writing to t
117 * the next RET instruction.
118 */
119 call sev_verify_cbit
120 #endif
121
122 /*
123 * Switch to early_top_pgt which still
124 * present.
125 */
126 movq %rax, %cr3
127
128 /* Branch to the common startup code a
129 ANNOTATE_RETPOLINE_SAFE
130 jmp *0f(%rip)
131 SYM_CODE_END(startup_64)
132
133 __INITRODATA
134 0: .quad common_startup_64
135
136 .text
137 SYM_CODE_START(secondary_startup_64)
138 UNWIND_HINT_END_OF_STACK
139 ANNOTATE_NOENDBR
140 /*
141 * At this point the CPU runs in 64bit
142 * and someone has loaded a mapped pag
143 *
144 * We come here either from startup_64
145 * or from trampoline.S (using virtual
146 *
147 * Using virtual addresses from trampo
148 * to have any identity mapped pages i
149 * after the boot processor executes t
150 */
151
152 /* Sanitize CPU configuration */
153 call verify_cpu
154
155 /*
156 * The secondary_startup_64_no_verify
157 * SEV-ES guests. In those guests the
158 * #VC exceptions which can not be han
159 * CPU bringup.
160 *
161 * All non SEV-ES systems, especially
162 * verify_cpu() above to make sure NX
163 */
164 SYM_INNER_LABEL(secondary_startup_64_no_verify
165 UNWIND_HINT_END_OF_STACK
166 ANNOTATE_NOENDBR
167
168 /* Clear %R15 which holds the boot_par
169 xorl %r15d, %r15d
170
171 /* Derive the runtime physical address
172 movq phys_base(%rip), %rax
173 addq $(init_top_pgt - __START_KERNE
174
175 /*
176 * Retrieve the modifier (SME encrypti
177 * added to the initial pgdir entry th
178 */
179 #ifdef CONFIG_AMD_MEM_ENCRYPT
180 addq sme_me_mask(%rip), %rax
181 #endif
182 /*
183 * Switch to the init_top_pgt here, aw
184 * unmap the identity mapped ranges.
185 */
186 movq %rax, %cr3
187
188 SYM_INNER_LABEL(common_startup_64, SYM_L_LOCAL
189 UNWIND_HINT_END_OF_STACK
190 ANNOTATE_NOENDBR
191
192 /*
193 * Create a mask of CR4 bits to preser
194 * global 1:1 translations from the TL
195 *
196 * From the SDM:
197 * "If CR4.PGE is changing from 0 to 1
198 * entries before the execution; if C
199 * there will be no global TLB entrie
200 */
201 movl $(X86_CR4_PAE | X86_CR4_LA57),
202 #ifdef CONFIG_X86_MCE
203 /*
204 * Preserve CR4.MCE if the kernel will
205 * Clearing MCE may fault in some envi
206 * support). Any machine check that oc
207 * configured will crash the system re
208 * here.
209 */
210 orl $X86_CR4_MCE, %edx
211 #endif
212 movq %cr4, %rcx
213 andl %edx, %ecx
214
215 /* Even if ignored in long mode, set P
216 btsl $X86_CR4_PSE_BIT, %ecx
217 movq %rcx, %cr4
218
219 /*
220 * Set CR4.PGE to re-enable global tra
221 */
222 btsl $X86_CR4_PGE_BIT, %ecx
223 movq %rcx, %cr4
224
225 #ifdef CONFIG_SMP
226 /*
227 * For parallel boot, the APIC ID is r
228 * used to look up the CPU number. Fo
229 * CPU number is encoded in smpboot_co
230 *
231 * Bit 31 STARTUP_READ_APICID (R
232 * Bit 0-23 CPU# if STARTUP_xx fla
233 */
234 movl smpboot_control(%rip), %ecx
235 testl $STARTUP_READ_APICID, %ecx
236 jnz .Lread_apicid
237 /*
238 * No control bit set, single CPU brin
239 * in bit 0-23. This is also the boot
240 */
241 andl $(~STARTUP_PARALLEL_MASK), %ec
242 jmp .Lsetup_cpu
243
244 .Lread_apicid:
245 /* Check whether X2APIC mode is alread
246 mov $MSR_IA32_APICBASE, %ecx
247 rdmsr
248 testl $X2APIC_ENABLE, %eax
249 jnz .Lread_apicid_msr
250
251 #ifdef CONFIG_X86_X2APIC
252 /*
253 * If system is in X2APIC mode then MM
254 * mapped causing the MMIO read below
255 * be handled at that point.
256 */
257 cmpl $0, x2apic_mode(%rip)
258 jz .Lread_apicid_mmio
259
260 /* Force the AP into X2APIC mode. */
261 orl $X2APIC_ENABLE, %eax
262 wrmsr
263 jmp .Lread_apicid_msr
264 #endif
265
266 .Lread_apicid_mmio:
267 /* Read the APIC ID from the fix-mappe
268 movq apic_mmio_base(%rip), %rcx
269 addq $APIC_ID, %rcx
270 movl (%rcx), %eax
271 shr $24, %eax
272 jmp .Llookup_AP
273
274 .Lread_apicid_msr:
275 mov $APIC_X2APIC_ID_MSR, %ecx
276 rdmsr
277
278 .Llookup_AP:
279 /* EAX contains the APIC ID of the cur
280 xorl %ecx, %ecx
281 leaq cpuid_to_apicid(%rip), %rbx
282
283 .Lfind_cpunr:
284 cmpl (%rbx,%rcx,4), %eax
285 jz .Lsetup_cpu
286 inc %ecx
287 #ifdef CONFIG_FORCE_NR_CPUS
288 cmpl $NR_CPUS, %ecx
289 #else
290 cmpl nr_cpu_ids(%rip), %ecx
291 #endif
292 jb .Lfind_cpunr
293
294 /* APIC ID not found in the table. Dr
295 movq trampoline_lock(%rip), %rax
296 movl $0, (%rax)
297
298 1: cli
299 hlt
300 jmp 1b
301
302 .Lsetup_cpu:
303 /* Get the per cpu offset for the give
304 movq __per_cpu_offset(,%rcx,8), %rd
305 #else
306 xorl %edx, %edx /* zero-extended to
307 #endif /* CONFIG_SMP */
308
309 /*
310 * Setup a boot time stack - Any secon
311 * by now because the cr3-switch above
312 *
313 * RDX contains the per-cpu offset
314 */
315 movq pcpu_hot + X86_current_task(%r
316 movq TASK_threadsp(%rax), %rsp
317
318 /*
319 * Now that this CPU is running on its
320 * protection. For the boot CPU the po
321 */
322 movq trampoline_lock(%rip), %rax
323 testq %rax, %rax
324 jz .Lsetup_gdt
325 movl $0, (%rax)
326
327 .Lsetup_gdt:
328 /*
329 * We must switch to a new descriptor
330 * because soon the kernel won't have
331 * addresses where we're currently run
332 * because in 32bit we couldn't load a
333 */
334 subq $16, %rsp
335 movw $(GDT_SIZE-1), (%rsp)
336 leaq gdt_page(%rdx), %rax
337 movq %rax, 2(%rsp)
338 lgdt (%rsp)
339 addq $16, %rsp
340
341 /* set up data segments */
342 xorl %eax,%eax
343 movl %eax,%ds
344 movl %eax,%ss
345 movl %eax,%es
346
347 /*
348 * We don't really need to load %fs or
349 * to kill any stale realmode selector
350 * under VT hardware.
351 */
352 movl %eax,%fs
353 movl %eax,%gs
354
355 /* Set up %gs.
356 *
357 * The base of %gs always points to fi
358 * stack protector canary is enabled,
359 * Note that, on SMP, the boot cpu use
360 * the per cpu areas are set up.
361 */
362 movl $MSR_GS_BASE,%ecx
363 #ifndef CONFIG_SMP
364 leaq INIT_PER_CPU_VAR(fixed_percpu_
365 #endif
366 movl %edx, %eax
367 shrq $32, %rdx
368 wrmsr
369
370 /* Setup and Load IDT */
371 call early_setup_idt
372
373 /* Check if nx is implemented */
374 movl $0x80000001, %eax
375 cpuid
376 movl %edx,%edi
377
378 /* Setup EFER (Extended Feature Enable
379 movl $MSR_EFER, %ecx
380 rdmsr
381 /*
382 * Preserve current value of EFER for
383 * EFER writes if no change was made (
384 */
385 movl %eax, %edx
386 btsl $_EFER_SCE, %eax /* Ena
387 btl $20,%edi /* No
388 jnc 1f
389 btsl $_EFER_NX, %eax
390 btsq $_PAGE_BIT_NX,early_pmd_flags(
391
392 /* Avoid writing EFER if no change was
393 1: cmpl %edx, %eax
394 je 1f
395 xor %edx, %edx
396 wrmsr /* Mak
397 1:
398 /* Setup cr0 */
399 movl $CR0_STATE, %eax
400 /* Make changes effective */
401 movq %rax, %cr0
402
403 /* zero EFLAGS after setting rsp */
404 pushq $0
405 popfq
406
407 /* Pass the boot_params pointer as fir
408 movq %r15, %rdi
409
410 .Ljump_to_C_code:
411 xorl %ebp, %ebp # clear frame
412 ANNOTATE_RETPOLINE_SAFE
413 callq *initial_code(%rip)
414 ud2
415 SYM_CODE_END(secondary_startup_64)
416
417 #include "verify_cpu.S"
418 #include "sev_verify_cbit.S"
419
420 #if defined(CONFIG_HOTPLUG_CPU) && defined(CON
421 /*
422 * Entry point for soft restart of a CPU. Invo
423 * restarting the boot CPU or for restarting S
424 * unplug. Everything is set up already except
425 */
426 SYM_CODE_START(soft_restart_cpu)
427 ANNOTATE_NOENDBR
428 UNWIND_HINT_END_OF_STACK
429
430 /* Find the idle task stack */
431 movq PER_CPU_VAR(pcpu_hot + X86_cur
432 movq TASK_threadsp(%rcx), %rsp
433
434 jmp .Ljump_to_C_code
435 SYM_CODE_END(soft_restart_cpu)
436 #endif
437
438 #ifdef CONFIG_AMD_MEM_ENCRYPT
439 /*
440 * VC Exception handler used during early boot
441 * addresses, but before the switch to the idt
442 * The early_idt_handler_array can't be used h
443 * of __init code and this handler is also use
444 * Therefore this handler ends up in the .text
445 * when .init.text is freed.
446 */
447 SYM_CODE_START_NOALIGN(vc_boot_ghcb)
448 UNWIND_HINT_IRET_REGS offset=8
449 ENDBR
450
451 /* Build pt_regs */
452 PUSH_AND_CLEAR_REGS
453
454 /* Call C handler */
455 movq %rsp, %rdi
456 movq ORIG_RAX(%rsp), %rsi
457 movq initial_vc_handler(%rip), %rax
458 ANNOTATE_RETPOLINE_SAFE
459 call *%rax
460
461 /* Unwind pt_regs */
462 POP_REGS
463
464 /* Remove Error Code */
465 addq $8, %rsp
466
467 iretq
468 SYM_CODE_END(vc_boot_ghcb)
469 #endif
470
471 /* Both SMP bootup and ACPI suspend ch
472 __REFDATA
473 .balign 8
474 SYM_DATA(initial_code, .quad x86_64_start_ker
475 #ifdef CONFIG_AMD_MEM_ENCRYPT
476 SYM_DATA(initial_vc_handler, .quad handle_v
477 #endif
478
479 SYM_DATA(trampoline_lock, .quad 0);
480 __FINITDATA
481
482 __INIT
483 SYM_CODE_START(early_idt_handler_array)
484 i = 0
485 .rept NUM_EXCEPTION_VECTORS
486 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1
487 UNWIND_HINT_IRET_REGS
488 ENDBR
489 pushq $0 # Dummy error
490 .else
491 UNWIND_HINT_IRET_REGS offset=8
492 ENDBR
493 .endif
494 pushq $i # 72(%rsp) Vec
495 jmp early_idt_handler_common
496 UNWIND_HINT_IRET_REGS
497 i = i + 1
498 .fill early_idt_handler_array + i*EARL
499 .endr
500 SYM_CODE_END(early_idt_handler_array)
501 ANNOTATE_NOENDBR // early_idt_handler_
502
503 SYM_CODE_START_LOCAL(early_idt_handler_common)
504 UNWIND_HINT_IRET_REGS offset=16
505 /*
506 * The stack is the hardware frame, an
507 * vector number.
508 */
509 cld
510
511 incl early_recursion_flag(%rip)
512
513 /* The vector number is currently in t
514 pushq %rsi
515 movq 8(%rsp), %rsi
516 movq %rdi, 8(%rsp)
517 pushq %rdx
518 pushq %rcx
519 pushq %rax
520 pushq %r8
521 pushq %r9
522 pushq %r10
523 pushq %r11
524 pushq %rbx
525 pushq %rbp
526 pushq %r12
527 pushq %r13
528 pushq %r14
529 pushq %r15
530 UNWIND_HINT_REGS
531
532 movq %rsp,%rdi /* RDI = pt_re
533 call do_early_exception
534
535 decl early_recursion_flag(%rip)
536 jmp restore_regs_and_return_to_kernel
537 SYM_CODE_END(early_idt_handler_common)
538
539 #ifdef CONFIG_AMD_MEM_ENCRYPT
540 /*
541 * VC Exception handler used during very early
542 * early_idt_handler_array can't be used becau
543 * paravirtualized INTERRUPT_RETURN and pv-ops
544 *
545 * XXX it does, fix this.
546 *
547 * This handler will end up in the .init.text
548 * available to boot secondary CPUs.
549 */
550 SYM_CODE_START_NOALIGN(vc_no_ghcb)
551 UNWIND_HINT_IRET_REGS offset=8
552 ENDBR
553
554 /* Build pt_regs */
555 PUSH_AND_CLEAR_REGS
556
557 /* Call C handler */
558 movq %rsp, %rdi
559 movq ORIG_RAX(%rsp), %rsi
560 call do_vc_no_ghcb
561
562 /* Unwind pt_regs */
563 POP_REGS
564
565 /* Remove Error Code */
566 addq $8, %rsp
567
568 /* Pure iret required here - don't use
569 iretq
570 SYM_CODE_END(vc_no_ghcb)
571 #endif
572
573 #ifdef CONFIG_MITIGATION_PAGE_TABLE_ISOLATION
574 /*
575 * Each PGD needs to be 8k long and 8k aligned
576 * ever go out to userspace with these, so we
577 * strictly *need* the second page, but this a
578 * have a single set_pgd() implementation that
579 * need to worry about whether it has 4k or 8k
580 * with.
581 *
582 * This ensures PGDs are 8k long:
583 */
584 #define PTI_USER_PGD_FILL 512
585 /* This ensures they are 8k-aligned: */
586 #define SYM_DATA_START_PTI_ALIGNED(name) \
587 SYM_START(name, SYM_L_GLOBAL, .balign
588 #else
589 #define SYM_DATA_START_PTI_ALIGNED(name) \
590 SYM_DATA_START_PAGE_ALIGNED(name)
591 #define PTI_USER_PGD_FILL 0
592 #endif
593
594 __INITDATA
595 .balign 4
596
597 SYM_DATA_START_PTI_ALIGNED(early_top_pgt)
598 .fill 511,8,0
599 .quad level3_kernel_pgt - __START_KE
600 .fill PTI_USER_PGD_FILL,8,0
601 SYM_DATA_END(early_top_pgt)
602
603 SYM_DATA_START_PAGE_ALIGNED(early_dynamic_pgts
604 .fill 512*EARLY_DYNAMIC_PAGE_TABLES,
605 SYM_DATA_END(early_dynamic_pgts)
606
607 SYM_DATA(early_recursion_flag, .long 0)
608
609 .data
610
611 #if defined(CONFIG_XEN_PV) || defined(CONFIG_P
612 SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
613 .quad level3_ident_pgt - __START_KER
614 .org init_top_pgt + L4_PAGE_OFFSET*
615 .quad level3_ident_pgt - __START_KER
616 .org init_top_pgt + L4_START_KERNEL
617 /* (2^48-(2*1024*1024*1024))/(2^39) =
618 .quad level3_kernel_pgt - __START_KE
619 .fill PTI_USER_PGD_FILL,8,0
620 SYM_DATA_END(init_top_pgt)
621
622 SYM_DATA_START_PAGE_ALIGNED(level3_ident_pgt)
623 .quad level2_ident_pgt - __START_KER
624 .fill 511, 8, 0
625 SYM_DATA_END(level3_ident_pgt)
626 SYM_DATA_START_PAGE_ALIGNED(level2_ident_pgt)
627 /*
628 * Since I easily can, map the first 1
629 * Don't set NX because code runs from
630 *
631 * Note: This sets _PAGE_GLOBAL despit
632 * the CPU supports it or it is enable
633 * the CPU should ignore the bit.
634 */
635 PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC
636 SYM_DATA_END(level2_ident_pgt)
637 #else
638 SYM_DATA_START_PTI_ALIGNED(init_top_pgt)
639 .fill 512,8,0
640 .fill PTI_USER_PGD_FILL,8,0
641 SYM_DATA_END(init_top_pgt)
642 #endif
643
644 #ifdef CONFIG_X86_5LEVEL
645 SYM_DATA_START_PAGE_ALIGNED(level4_kernel_pgt)
646 .fill 511,8,0
647 .quad level3_kernel_pgt - __START_KE
648 SYM_DATA_END(level4_kernel_pgt)
649 #endif
650
651 SYM_DATA_START_PAGE_ALIGNED(level3_kernel_pgt)
652 .fill L3_START_KERNEL,8,0
653 /* (2^48-(2*1024*1024*1024)-((2^39)*51
654 .quad level2_kernel_pgt - __START_KE
655 .quad level2_fixmap_pgt - __START_KE
656 SYM_DATA_END(level3_kernel_pgt)
657
658 SYM_DATA_START_PAGE_ALIGNED(level2_kernel_pgt)
659 /*
660 * Kernel high mapping.
661 *
662 * The kernel code+data+bss must be lo
663 * virtual address space, which is 1 G
664 * 512 MiB otherwise.
665 *
666 * (NOTE: after that starts the module
667 *
668 * This table is eventually used by th
669 * Care must be taken to clear out und
670 * or _PAGE_GLOBAL in some cases.
671 */
672 PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERN
673 SYM_DATA_END(level2_kernel_pgt)
674
675 SYM_DATA_START_PAGE_ALIGNED(level2_fixmap_pgt)
676 .fill (512 - 4 - FIXMAP_PMD_NUM),8,0
677 pgtno = 0
678 .rept (FIXMAP_PMD_NUM)
679 .quad level1_fixmap_pgt + (pgtno << PA
680 + _PAGE_TABLE_NOENC;
681 pgtno = pgtno + 1
682 .endr
683 /* 6 MB reserved space + a 2MB hole */
684 .fill 4,8,0
685 SYM_DATA_END(level2_fixmap_pgt)
686
687 SYM_DATA_START_PAGE_ALIGNED(level1_fixmap_pgt)
688 .rept (FIXMAP_PMD_NUM)
689 .fill 512,8,0
690 .endr
691 SYM_DATA_END(level1_fixmap_pgt)
692
693 .data
694 .align 16
695
696 SYM_DATA(smpboot_control, .long
697
698 .align 16
699 /* This must match the first entry in level2_k
700 SYM_DATA(phys_base, .quad 0x0)
701 EXPORT_SYMBOL(phys_base)
702
703 #include "../xen/xen-head.S"
704
705 __PAGE_ALIGNED_BSS
706 SYM_DATA_START_PAGE_ALIGNED(empty_zero_page)
707 .skip PAGE_SIZE
708 SYM_DATA_END(empty_zero_page)
709 EXPORT_SYMBOL(empty_zero_page)
710
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.