xref: /openbmc/linux/arch/x86/kernel/head_32.S (revision 979ac5ef)
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 * Check if it is 486
265 */
266	movb $4,X86			# at least 486
267	cmpl $-1,X86_CPUID
268	je .Lis486
269
270	/* get vendor info */
271	xorl %eax,%eax			# call CPUID with 0 -> return vendor ID
272	cpuid
273	movl %eax,X86_CPUID		# save CPUID level
274	movl %ebx,X86_VENDOR_ID		# lo 4 chars
275	movl %edx,X86_VENDOR_ID+4	# next 4 chars
276	movl %ecx,X86_VENDOR_ID+8	# last 4 chars
277
278	orl %eax,%eax			# do we have processor info as well?
279	je .Lis486
280
281	movl $1,%eax		# Use the CPUID instruction to get CPU type
282	cpuid
283	movb %al,%cl		# save reg for future use
284	andb $0x0f,%ah		# mask processor family
285	movb %ah,X86
286	andb $0xf0,%al		# mask model
287	shrb $4,%al
288	movb %al,X86_MODEL
289	andb $0x0f,%cl		# mask mask revision
290	movb %cl,X86_STEPPING
291	movl %edx,X86_CAPABILITY
292
293.Lis486:
294	movl $0x50022,%ecx	# set AM, WP, NE and MP
295	movl %cr0,%eax
296	andl $0x80000011,%eax	# Save PG,PE,ET
297	orl %ecx,%eax
298	movl %eax,%cr0
299
300	lgdt early_gdt_descr
301	ljmp $(__KERNEL_CS),$1f
3021:	movl $(__KERNEL_DS),%eax	# reload all the segment registers
303	movl %eax,%ss			# after changing gdt.
304
305	movl $(__USER_DS),%eax		# DS/ES contains default USER segment
306	movl %eax,%ds
307	movl %eax,%es
308
309	movl $(__KERNEL_PERCPU), %eax
310	movl %eax,%fs			# set this cpu's percpu
311
312	xorl %eax,%eax
313	movl %eax,%gs			# clear possible garbage in %gs
314
315	xorl %eax,%eax			# Clear LDT
316	lldt %ax
317
318	call *(initial_code)
3191:	jmp 1b
320SYM_FUNC_END(startup_32_smp)
321
322#include "verify_cpu.S"
323
324__INIT
325SYM_FUNC_START(early_idt_handler_array)
326	# 36(%esp) %eflags
327	# 32(%esp) %cs
328	# 28(%esp) %eip
329	# 24(%rsp) error code
330	i = 0
331	.rept NUM_EXCEPTION_VECTORS
332	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
333	pushl $0		# Dummy error code, to make stack frame uniform
334	.endif
335	pushl $i		# 20(%esp) Vector number
336	jmp early_idt_handler_common
337	i = i + 1
338	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
339	.endr
340SYM_FUNC_END(early_idt_handler_array)
341
342SYM_CODE_START_LOCAL(early_idt_handler_common)
343	/*
344	 * The stack is the hardware frame, an error code or zero, and the
345	 * vector number.
346	 */
347	cld
348
349	incl %ss:early_recursion_flag
350
351	/* The vector number is in pt_regs->gs */
352
353	cld
354	pushl	%fs		/* pt_regs->fs (__fsh varies by model) */
355	pushl	%es		/* pt_regs->es (__esh varies by model) */
356	pushl	%ds		/* pt_regs->ds (__dsh varies by model) */
357	pushl	%eax		/* pt_regs->ax */
358	pushl	%ebp		/* pt_regs->bp */
359	pushl	%edi		/* pt_regs->di */
360	pushl	%esi		/* pt_regs->si */
361	pushl	%edx		/* pt_regs->dx */
362	pushl	%ecx		/* pt_regs->cx */
363	pushl	%ebx		/* pt_regs->bx */
364
365	/* Fix up DS and ES */
366	movl	$(__KERNEL_DS), %ecx
367	movl	%ecx, %ds
368	movl	%ecx, %es
369
370	/* Load the vector number into EDX */
371	movl	PT_GS(%esp), %edx
372
373	/* Load GS into pt_regs->gs (and maybe clobber __gsh) */
374	movw	%gs, PT_GS(%esp)
375
376	movl	%esp, %eax	/* args are pt_regs (EAX), trapnr (EDX) */
377	call	early_fixup_exception
378
379	popl	%ebx		/* pt_regs->bx */
380	popl	%ecx		/* pt_regs->cx */
381	popl	%edx		/* pt_regs->dx */
382	popl	%esi		/* pt_regs->si */
383	popl	%edi		/* pt_regs->di */
384	popl	%ebp		/* pt_regs->bp */
385	popl	%eax		/* pt_regs->ax */
386	popl	%ds		/* pt_regs->ds (always ignores __dsh) */
387	popl	%es		/* pt_regs->es (always ignores __esh) */
388	popl	%fs		/* pt_regs->fs (always ignores __fsh) */
389	popl	%gs		/* pt_regs->gs (always ignores __gsh) */
390	decl	%ss:early_recursion_flag
391	addl	$4, %esp	/* pop pt_regs->orig_ax */
392	iret
393SYM_CODE_END(early_idt_handler_common)
394
395/* This is the default interrupt "handler" :-) */
396SYM_FUNC_START(early_ignore_irq)
397	cld
398#ifdef CONFIG_PRINTK
399	pushl %eax
400	pushl %ecx
401	pushl %edx
402	pushl %es
403	pushl %ds
404	movl $(__KERNEL_DS),%eax
405	movl %eax,%ds
406	movl %eax,%es
407	cmpl $2,early_recursion_flag
408	je hlt_loop
409	incl early_recursion_flag
410	pushl 16(%esp)
411	pushl 24(%esp)
412	pushl 32(%esp)
413	pushl 40(%esp)
414	pushl $int_msg
415	call _printk
416
417	call dump_stack
418
419	addl $(5*4),%esp
420	popl %ds
421	popl %es
422	popl %edx
423	popl %ecx
424	popl %eax
425#endif
426	iret
427
428hlt_loop:
429	hlt
430	jmp hlt_loop
431SYM_FUNC_END(early_ignore_irq)
432
433__INITDATA
434	.align 4
435SYM_DATA(early_recursion_flag, .long 0)
436
437__REFDATA
438	.align 4
439SYM_DATA(initial_code,		.long i386_start_kernel)
440
441#ifdef CONFIG_PAGE_TABLE_ISOLATION
442#define	PGD_ALIGN	(2 * PAGE_SIZE)
443#define PTI_USER_PGD_FILL	1024
444#else
445#define	PGD_ALIGN	(PAGE_SIZE)
446#define PTI_USER_PGD_FILL	0
447#endif
448/*
449 * BSS section
450 */
451__PAGE_ALIGNED_BSS
452	.align PGD_ALIGN
453#ifdef CONFIG_X86_PAE
454.globl initial_pg_pmd
455initial_pg_pmd:
456	.fill 1024*KPMDS,4,0
457#else
458.globl initial_page_table
459initial_page_table:
460	.fill 1024,4,0
461#endif
462	.align PGD_ALIGN
463initial_pg_fixmap:
464	.fill 1024,4,0
465.globl swapper_pg_dir
466	.align PGD_ALIGN
467swapper_pg_dir:
468	.fill 1024,4,0
469	.fill PTI_USER_PGD_FILL,4,0
470.globl empty_zero_page
471empty_zero_page:
472	.fill 4096,1,0
473EXPORT_SYMBOL(empty_zero_page)
474
475/*
476 * This starts the data section.
477 */
478#ifdef CONFIG_X86_PAE
479__PAGE_ALIGNED_DATA
480	/* Page-aligned for the benefit of paravirt? */
481	.align PGD_ALIGN
482SYM_DATA_START(initial_page_table)
483	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
484# if KPMDS == 3
485	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
486	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
487	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
488# elif KPMDS == 2
489	.long	0,0
490	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
491	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
492# elif KPMDS == 1
493	.long	0,0
494	.long	0,0
495	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
496# else
497#  error "Kernel PMDs should be 1, 2 or 3"
498# endif
499	.align PAGE_SIZE		/* needs to be page-sized too */
500
501#ifdef CONFIG_PAGE_TABLE_ISOLATION
502	/*
503	 * PTI needs another page so sync_initial_pagetable() works correctly
504	 * and does not scribble over the data which is placed behind the
505	 * actual initial_page_table. See clone_pgd_range().
506	 */
507	.fill 1024, 4, 0
508#endif
509
510SYM_DATA_END(initial_page_table)
511#endif
512
513.data
514.balign 4
515/*
516 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
517 * reliably detect the end of the stack.
518 */
519SYM_DATA(initial_stack,
520		.long init_thread_union + THREAD_SIZE -
521		SIZEOF_PTREGS - TOP_OF_KERNEL_STACK_PADDING)
522
523__INITRODATA
524int_msg:
525	.asciz "Unknown interrupt or fault at: %p %p %p\n"
526
527#include "../../x86/xen/xen-head.S"
528
529/*
530 * The IDT and GDT 'descriptors' are a strange 48-bit object
531 * only used by the lidt and lgdt instructions. They are not
532 * like usual segment descriptors - they consist of a 16-bit
533 * segment size, and 32-bit linear address value:
534 */
535
536	.data
537	ALIGN
538# early boot GDT descriptor (must use 1:1 address mapping)
539	.word 0				# 32 bit align gdt_desc.address
540SYM_DATA_START_LOCAL(boot_gdt_descr)
541	.word __BOOT_DS+7
542	.long boot_gdt - __PAGE_OFFSET
543SYM_DATA_END(boot_gdt_descr)
544
545# boot GDT descriptor (later on used by CPU#0):
546	.word 0				# 32 bit align gdt_desc.address
547SYM_DATA_START(early_gdt_descr)
548	.word GDT_ENTRIES*8-1
549	.long gdt_page			/* Overwritten for secondary CPUs */
550SYM_DATA_END(early_gdt_descr)
551
552/*
553 * The boot_gdt must mirror the equivalent in setup.S and is
554 * used only for booting.
555 */
556	.align L1_CACHE_BYTES
557SYM_DATA_START(boot_gdt)
558	.fill GDT_ENTRY_BOOT_CS,8,0
559	.quad 0x00cf9a000000ffff	/* kernel 4GB code at 0x00000000 */
560	.quad 0x00cf92000000ffff	/* kernel 4GB data at 0x00000000 */
561SYM_DATA_END(boot_gdt)
562