xref: /linux/arch/x86/kernel/head_32.S (revision b5bee6ced21ca98389000b7017dd41b0cc37fa50)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 *
4 *  Copyright (C) 1991, 1992  Linus Torvalds
5 *
6 *  Enhanced CPU detection and feature setting code by Mike Jagdis
7 *  and Martin Mares, November 1997.
8 */
9
10.text
11#include <linux/threads.h>
12#include <linux/init.h>
13#include <linux/linkage.h>
14#include <asm/segment.h>
15#include <asm/page_types.h>
16#include <asm/pgtable_types.h>
17#include <asm/cache.h>
18#include <asm/thread_info.h>
19#include <asm/asm-offsets.h>
20#include <asm/setup.h>
21#include <asm/processor-flags.h>
22#include <asm/msr-index.h>
23#include <asm/cpufeatures.h>
24#include <asm/percpu.h>
25#include <asm/nops.h>
26#include <asm/nospec-branch.h>
27#include <asm/bootparam.h>
28#include <asm/export.h>
29#include <asm/pgtable_32.h>
30
31/* Physical address */
32#define pa(X) ((X) - __PAGE_OFFSET)
33
34/*
35 * References to members of the new_cpu_data structure.
36 */
37
38#define X86		new_cpu_data+CPUINFO_x86
39#define X86_VENDOR	new_cpu_data+CPUINFO_x86_vendor
40#define X86_MODEL	new_cpu_data+CPUINFO_x86_model
41#define X86_STEPPING	new_cpu_data+CPUINFO_x86_stepping
42#define X86_HARD_MATH	new_cpu_data+CPUINFO_hard_math
43#define X86_CPUID	new_cpu_data+CPUINFO_cpuid_level
44#define X86_CAPABILITY	new_cpu_data+CPUINFO_x86_capability
45#define X86_VENDOR_ID	new_cpu_data+CPUINFO_x86_vendor_id
46
47
48#define SIZEOF_PTREGS 17*4
49
50/*
51 * Worst-case size of the kernel mapping we need to make:
52 * a relocatable kernel can live anywhere in lowmem, so we need to be able
53 * to map all of lowmem.
54 */
55KERNEL_PAGES = LOWMEM_PAGES
56
57INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
58RESERVE_BRK(pagetables, INIT_MAP_SIZE)
59
60/*
61 * 32-bit kernel entrypoint; only used by the boot CPU.  On entry,
62 * %esi points to the real-mode code as a 32-bit pointer.
63 * CS and DS must be 4 GB flat segments, but we don't depend on
64 * any particular GDT layout, because we load our own as soon as we
65 * can.
66 */
67__HEAD
68SYM_CODE_START(startup_32)
69	movl pa(initial_stack),%ecx
70
71/*
72 * Set segments to known values.
73 */
74	lgdt pa(boot_gdt_descr)
75	movl $(__BOOT_DS),%eax
76	movl %eax,%ds
77	movl %eax,%es
78	movl %eax,%fs
79	movl %eax,%gs
80	movl %eax,%ss
81	leal -__PAGE_OFFSET(%ecx),%esp
82
83/*
84 * Clear BSS first so that there are no surprises...
85 */
86	cld
87	xorl %eax,%eax
88	movl $pa(__bss_start),%edi
89	movl $pa(__bss_stop),%ecx
90	subl %edi,%ecx
91	shrl $2,%ecx
92	rep ; stosl
93/*
94 * Copy bootup parameters out of the way.
95 * Note: %esi still has the pointer to the real-mode data.
96 * With the kexec as boot loader, parameter segment might be loaded beyond
97 * kernel image and might not even be addressable by early boot page tables.
98 * (kexec on panic case). Hence copy out the parameters before initializing
99 * page tables.
100 */
101	movl $pa(boot_params),%edi
102	movl $(PARAM_SIZE/4),%ecx
103	cld
104	rep
105	movsl
106	movl pa(boot_params) + NEW_CL_POINTER,%esi
107	andl %esi,%esi
108	jz 1f			# No command line
109	movl $pa(boot_command_line),%edi
110	movl $(COMMAND_LINE_SIZE/4),%ecx
111	rep
112	movsl
1131:
114
115#ifdef CONFIG_OLPC
116	/* save OFW's pgdir table for later use when calling into OFW */
117	movl %cr3, %eax
118	movl %eax, pa(olpc_ofw_pgd)
119#endif
120
121#ifdef CONFIG_MICROCODE
122	/* Early load ucode on BSP. */
123	call load_ucode_bsp
124#endif
125
126	/* Create early pagetables. */
127	call  mk_early_pgtbl_32
128
129	/* Do early initialization of the fixmap area */
130	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
131#ifdef  CONFIG_X86_PAE
132#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
133	movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
134#else
135	movl %eax,pa(initial_page_table+0xffc)
136#endif
137
138	jmp .Ldefault_entry
139SYM_CODE_END(startup_32)
140
141#ifdef CONFIG_HOTPLUG_CPU
142/*
143 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
144 * up already except stack. We just set up stack here. Then call
145 * start_secondary().
146 */
147SYM_FUNC_START(start_cpu0)
148	movl initial_stack, %ecx
149	movl %ecx, %esp
150	call *(initial_code)
1511:	jmp 1b
152SYM_FUNC_END(start_cpu0)
153#endif
154
155/*
156 * Non-boot CPU entry point; entered from trampoline.S
157 * We can't lgdt here, because lgdt itself uses a data segment, but
158 * we know the trampoline has already loaded the boot_gdt for us.
159 *
160 * If cpu hotplug is not supported then this code can go in init section
161 * which will be freed later
162 */
163SYM_FUNC_START(startup_32_smp)
164	cld
165	movl $(__BOOT_DS),%eax
166	movl %eax,%ds
167	movl %eax,%es
168	movl %eax,%fs
169	movl %eax,%gs
170	movl pa(initial_stack),%ecx
171	movl %eax,%ss
172	leal -__PAGE_OFFSET(%ecx),%esp
173
174#ifdef CONFIG_MICROCODE
175	/* Early load ucode on AP. */
176	call load_ucode_ap
177#endif
178
179.Ldefault_entry:
180	movl $(CR0_STATE & ~X86_CR0_PG),%eax
181	movl %eax,%cr0
182
183/*
184 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
185 * bits like NT set. This would confuse the debugger if this code is traced. So
186 * initialize them properly now before switching to protected mode. That means
187 * DF in particular (even though we have cleared it earlier after copying the
188 * command line) because GCC expects it.
189 */
190	pushl $0
191	popfl
192
193/*
194 * New page tables may be in 4Mbyte page mode and may be using the global pages.
195 *
196 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
197 * if and only if CPUID exists and has flags other than the FPU flag set.
198 */
199	movl $-1,pa(X86_CPUID)		# preset CPUID level
200	movl $X86_EFLAGS_ID,%ecx
201	pushl %ecx
202	popfl				# set EFLAGS=ID
203	pushfl
204	popl %eax			# get EFLAGS
205	testl $X86_EFLAGS_ID,%eax	# did EFLAGS.ID remained set?
206	jz .Lenable_paging		# hw disallowed setting of ID bit
207					# which means no CPUID and no CR4
208
209	xorl %eax,%eax
210	cpuid
211	movl %eax,pa(X86_CPUID)		# save largest std CPUID function
212
213	movl $1,%eax
214	cpuid
215	andl $~1,%edx			# Ignore CPUID.FPU
216	jz .Lenable_paging		# No flags or only CPUID.FPU = no CR4
217
218	movl pa(mmu_cr4_features),%eax
219	movl %eax,%cr4
220
221	testb $X86_CR4_PAE, %al		# check if PAE is enabled
222	jz .Lenable_paging
223
224	/* Check if extended functions are implemented */
225	movl $0x80000000, %eax
226	cpuid
227	/* Value must be in the range 0x80000001 to 0x8000ffff */
228	subl $0x80000001, %eax
229	cmpl $(0x8000ffff-0x80000001), %eax
230	ja .Lenable_paging
231
232	/* Clear bogus XD_DISABLE bits */
233	call verify_cpu
234
235	mov $0x80000001, %eax
236	cpuid
237	/* Execute Disable bit supported? */
238	btl $(X86_FEATURE_NX & 31), %edx
239	jnc .Lenable_paging
240
241	/* Setup EFER (Extended Feature Enable Register) */
242	movl $MSR_EFER, %ecx
243	rdmsr
244
245	btsl $_EFER_NX, %eax
246	/* Make changes effective */
247	wrmsr
248
249.Lenable_paging:
250
251/*
252 * Enable paging
253 */
254	movl $pa(initial_page_table), %eax
255	movl %eax,%cr3		/* set the page table pointer.. */
256	movl $CR0_STATE,%eax
257	movl %eax,%cr0		/* ..and set paging (PG) bit */
258	ljmp $__BOOT_CS,$1f	/* Clear prefetch and normalize %eip */
2591:
260	/* Shift the stack pointer to a virtual address */
261	addl $__PAGE_OFFSET, %esp
262
263/*
264 * start system 32-bit setup. We need to re-do some of the things done
265 * in 16-bit mode for the "real" operations.
266 */
267	movl setup_once_ref,%eax
268	andl %eax,%eax
269	jz 1f				# Did we do this already?
270	call *%eax
2711:
272
273/*
274 * Check if it is 486
275 */
276	movb $4,X86			# at least 486
277	cmpl $-1,X86_CPUID
278	je .Lis486
279
280	/* get vendor info */
281	xorl %eax,%eax			# call CPUID with 0 -> return vendor ID
282	cpuid
283	movl %eax,X86_CPUID		# save CPUID level
284	movl %ebx,X86_VENDOR_ID		# lo 4 chars
285	movl %edx,X86_VENDOR_ID+4	# next 4 chars
286	movl %ecx,X86_VENDOR_ID+8	# last 4 chars
287
288	orl %eax,%eax			# do we have processor info as well?
289	je .Lis486
290
291	movl $1,%eax		# Use the CPUID instruction to get CPU type
292	cpuid
293	movb %al,%cl		# save reg for future use
294	andb $0x0f,%ah		# mask processor family
295	movb %ah,X86
296	andb $0xf0,%al		# mask model
297	shrb $4,%al
298	movb %al,X86_MODEL
299	andb $0x0f,%cl		# mask mask revision
300	movb %cl,X86_STEPPING
301	movl %edx,X86_CAPABILITY
302
303.Lis486:
304	movl $0x50022,%ecx	# set AM, WP, NE and MP
305	movl %cr0,%eax
306	andl $0x80000011,%eax	# Save PG,PE,ET
307	orl %ecx,%eax
308	movl %eax,%cr0
309
310	lgdt early_gdt_descr
311	ljmp $(__KERNEL_CS),$1f
3121:	movl $(__KERNEL_DS),%eax	# reload all the segment registers
313	movl %eax,%ss			# after changing gdt.
314
315	movl $(__USER_DS),%eax		# DS/ES contains default USER segment
316	movl %eax,%ds
317	movl %eax,%es
318
319	movl $(__KERNEL_PERCPU), %eax
320	movl %eax,%fs			# set this cpu's percpu
321
322	xorl %eax,%eax
323	movl %eax,%gs			# clear possible garbage in %gs
324
325	xorl %eax,%eax			# Clear LDT
326	lldt %ax
327
328	call *(initial_code)
3291:	jmp 1b
330SYM_FUNC_END(startup_32_smp)
331
332#include "verify_cpu.S"
333
334/*
335 *  setup_once
336 *
337 *  The setup work we only want to run on the BSP.
338 *
339 *  Warning: %esi is live across this function.
340 */
341__INIT
342setup_once:
343	andl $0,setup_once_ref	/* Once is enough, thanks */
344	RET
345
346SYM_FUNC_START(early_idt_handler_array)
347	# 36(%esp) %eflags
348	# 32(%esp) %cs
349	# 28(%esp) %eip
350	# 24(%rsp) error code
351	i = 0
352	.rept NUM_EXCEPTION_VECTORS
353	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
354	pushl $0		# Dummy error code, to make stack frame uniform
355	.endif
356	pushl $i		# 20(%esp) Vector number
357	jmp early_idt_handler_common
358	i = i + 1
359	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
360	.endr
361SYM_FUNC_END(early_idt_handler_array)
362
363SYM_CODE_START_LOCAL(early_idt_handler_common)
364	/*
365	 * The stack is the hardware frame, an error code or zero, and the
366	 * vector number.
367	 */
368	cld
369
370	incl %ss:early_recursion_flag
371
372	/* The vector number is in pt_regs->gs */
373
374	cld
375	pushl	%fs		/* pt_regs->fs (__fsh varies by model) */
376	pushl	%es		/* pt_regs->es (__esh varies by model) */
377	pushl	%ds		/* pt_regs->ds (__dsh varies by model) */
378	pushl	%eax		/* pt_regs->ax */
379	pushl	%ebp		/* pt_regs->bp */
380	pushl	%edi		/* pt_regs->di */
381	pushl	%esi		/* pt_regs->si */
382	pushl	%edx		/* pt_regs->dx */
383	pushl	%ecx		/* pt_regs->cx */
384	pushl	%ebx		/* pt_regs->bx */
385
386	/* Fix up DS and ES */
387	movl	$(__KERNEL_DS), %ecx
388	movl	%ecx, %ds
389	movl	%ecx, %es
390
391	/* Load the vector number into EDX */
392	movl	PT_GS(%esp), %edx
393
394	/* Load GS into pt_regs->gs (and maybe clobber __gsh) */
395	movw	%gs, PT_GS(%esp)
396
397	movl	%esp, %eax	/* args are pt_regs (EAX), trapnr (EDX) */
398	call	early_fixup_exception
399
400	popl	%ebx		/* pt_regs->bx */
401	popl	%ecx		/* pt_regs->cx */
402	popl	%edx		/* pt_regs->dx */
403	popl	%esi		/* pt_regs->si */
404	popl	%edi		/* pt_regs->di */
405	popl	%ebp		/* pt_regs->bp */
406	popl	%eax		/* pt_regs->ax */
407	popl	%ds		/* pt_regs->ds (always ignores __dsh) */
408	popl	%es		/* pt_regs->es (always ignores __esh) */
409	popl	%fs		/* pt_regs->fs (always ignores __fsh) */
410	popl	%gs		/* pt_regs->gs (always ignores __gsh) */
411	decl	%ss:early_recursion_flag
412	addl	$4, %esp	/* pop pt_regs->orig_ax */
413	iret
414SYM_CODE_END(early_idt_handler_common)
415
416/* This is the default interrupt "handler" :-) */
417SYM_FUNC_START(early_ignore_irq)
418	cld
419#ifdef CONFIG_PRINTK
420	pushl %eax
421	pushl %ecx
422	pushl %edx
423	pushl %es
424	pushl %ds
425	movl $(__KERNEL_DS),%eax
426	movl %eax,%ds
427	movl %eax,%es
428	cmpl $2,early_recursion_flag
429	je hlt_loop
430	incl early_recursion_flag
431	pushl 16(%esp)
432	pushl 24(%esp)
433	pushl 32(%esp)
434	pushl 40(%esp)
435	pushl $int_msg
436	call _printk
437
438	call dump_stack
439
440	addl $(5*4),%esp
441	popl %ds
442	popl %es
443	popl %edx
444	popl %ecx
445	popl %eax
446#endif
447	iret
448
449hlt_loop:
450	hlt
451	jmp hlt_loop
452SYM_FUNC_END(early_ignore_irq)
453
454__INITDATA
455	.align 4
456SYM_DATA(early_recursion_flag, .long 0)
457
458__REFDATA
459	.align 4
460SYM_DATA(initial_code,		.long i386_start_kernel)
461SYM_DATA(setup_once_ref,	.long setup_once)
462
463#ifdef CONFIG_PAGE_TABLE_ISOLATION
464#define	PGD_ALIGN	(2 * PAGE_SIZE)
465#define PTI_USER_PGD_FILL	1024
466#else
467#define	PGD_ALIGN	(PAGE_SIZE)
468#define PTI_USER_PGD_FILL	0
469#endif
470/*
471 * BSS section
472 */
473__PAGE_ALIGNED_BSS
474	.align PGD_ALIGN
475#ifdef CONFIG_X86_PAE
476.globl initial_pg_pmd
477initial_pg_pmd:
478	.fill 1024*KPMDS,4,0
479#else
480.globl initial_page_table
481initial_page_table:
482	.fill 1024,4,0
483#endif
484	.align PGD_ALIGN
485initial_pg_fixmap:
486	.fill 1024,4,0
487.globl swapper_pg_dir
488	.align PGD_ALIGN
489swapper_pg_dir:
490	.fill 1024,4,0
491	.fill PTI_USER_PGD_FILL,4,0
492.globl empty_zero_page
493empty_zero_page:
494	.fill 4096,1,0
495EXPORT_SYMBOL(empty_zero_page)
496
497/*
498 * This starts the data section.
499 */
500#ifdef CONFIG_X86_PAE
501__PAGE_ALIGNED_DATA
502	/* Page-aligned for the benefit of paravirt? */
503	.align PGD_ALIGN
504SYM_DATA_START(initial_page_table)
505	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
506# if KPMDS == 3
507	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
508	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
509	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
510# elif KPMDS == 2
511	.long	0,0
512	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
513	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
514# elif KPMDS == 1
515	.long	0,0
516	.long	0,0
517	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
518# else
519#  error "Kernel PMDs should be 1, 2 or 3"
520# endif
521	.align PAGE_SIZE		/* needs to be page-sized too */
522
523#ifdef CONFIG_PAGE_TABLE_ISOLATION
524	/*
525	 * PTI needs another page so sync_initial_pagetable() works correctly
526	 * and does not scribble over the data which is placed behind the
527	 * actual initial_page_table. See clone_pgd_range().
528	 */
529	.fill 1024, 4, 0
530#endif
531
532SYM_DATA_END(initial_page_table)
533#endif
534
535.data
536.balign 4
537/*
538 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
539 * reliably detect the end of the stack.
540 */
541SYM_DATA(initial_stack,
542		.long init_thread_union + THREAD_SIZE -
543		SIZEOF_PTREGS - TOP_OF_KERNEL_STACK_PADDING)
544
545__INITRODATA
546int_msg:
547	.asciz "Unknown interrupt or fault at: %p %p %p\n"
548
549#include "../../x86/xen/xen-head.S"
550
551/*
552 * The IDT and GDT 'descriptors' are a strange 48-bit object
553 * only used by the lidt and lgdt instructions. They are not
554 * like usual segment descriptors - they consist of a 16-bit
555 * segment size, and 32-bit linear address value:
556 */
557
558	.data
559	ALIGN
560# early boot GDT descriptor (must use 1:1 address mapping)
561	.word 0				# 32 bit align gdt_desc.address
562SYM_DATA_START_LOCAL(boot_gdt_descr)
563	.word __BOOT_DS+7
564	.long boot_gdt - __PAGE_OFFSET
565SYM_DATA_END(boot_gdt_descr)
566
567# boot GDT descriptor (later on used by CPU#0):
568	.word 0				# 32 bit align gdt_desc.address
569SYM_DATA_START(early_gdt_descr)
570	.word GDT_ENTRIES*8-1
571	.long gdt_page			/* Overwritten for secondary CPUs */
572SYM_DATA_END(early_gdt_descr)
573
574/*
575 * The boot_gdt must mirror the equivalent in setup.S and is
576 * used only for booting.
577 */
578	.align L1_CACHE_BYTES
579SYM_DATA_START(boot_gdt)
580	.fill GDT_ENTRY_BOOT_CS,8,0
581	.quad 0x00cf9a000000ffff	/* kernel 4GB code at 0x00000000 */
582	.quad 0x00cf92000000ffff	/* kernel 4GB data at 0x00000000 */
583SYM_DATA_END(boot_gdt)
584