xref: /openbmc/linux/arch/x86/kernel/head_32.S (revision 4bb1e4e7) 
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/bootparam.h>
27#include <asm/export.h>
28#include <asm/pgtable_32.h>
29
30/* Physical address */
31#define pa(X) ((X) - __PAGE_OFFSET)
32
33/*
34 * References to members of the new_cpu_data structure.
35 */
36
37#define X86		new_cpu_data+CPUINFO_x86
38#define X86_VENDOR	new_cpu_data+CPUINFO_x86_vendor
39#define X86_MODEL	new_cpu_data+CPUINFO_x86_model
40#define X86_STEPPING	new_cpu_data+CPUINFO_x86_stepping
41#define X86_HARD_MATH	new_cpu_data+CPUINFO_hard_math
42#define X86_CPUID	new_cpu_data+CPUINFO_cpuid_level
43#define X86_CAPABILITY	new_cpu_data+CPUINFO_x86_capability
44#define X86_VENDOR_ID	new_cpu_data+CPUINFO_x86_vendor_id
45
46
47#define SIZEOF_PTREGS 17*4
48
49/*
50 * Worst-case size of the kernel mapping we need to make:
51 * a relocatable kernel can live anywhere in lowmem, so we need to be able
52 * to map all of lowmem.
53 */
54KERNEL_PAGES = LOWMEM_PAGES
55
56INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
57RESERVE_BRK(pagetables, INIT_MAP_SIZE)
58
59/*
60 * 32-bit kernel entrypoint; only used by the boot CPU.  On entry,
61 * %esi points to the real-mode code as a 32-bit pointer.
62 * CS and DS must be 4 GB flat segments, but we don't depend on
63 * any particular GDT layout, because we load our own as soon as we
64 * can.
65 */
66__HEAD
67SYM_CODE_START(startup_32)
68	movl pa(initial_stack),%ecx
69
70	/* test KEEP_SEGMENTS flag to see if the bootloader is asking
71		us to not reload segments */
72	testb $KEEP_SEGMENTS, BP_loadflags(%esi)
73	jnz 2f
74
75/*
76 * Set segments to known values.
77 */
78	lgdt pa(boot_gdt_descr)
79	movl $(__BOOT_DS),%eax
80	movl %eax,%ds
81	movl %eax,%es
82	movl %eax,%fs
83	movl %eax,%gs
84	movl %eax,%ss
852:
86	leal -__PAGE_OFFSET(%ecx),%esp
87
88/*
89 * Clear BSS first so that there are no surprises...
90 */
91	cld
92	xorl %eax,%eax
93	movl $pa(__bss_start),%edi
94	movl $pa(__bss_stop),%ecx
95	subl %edi,%ecx
96	shrl $2,%ecx
97	rep ; stosl
98/*
99 * Copy bootup parameters out of the way.
100 * Note: %esi still has the pointer to the real-mode data.
101 * With the kexec as boot loader, parameter segment might be loaded beyond
102 * kernel image and might not even be addressable by early boot page tables.
103 * (kexec on panic case). Hence copy out the parameters before initializing
104 * page tables.
105 */
106	movl $pa(boot_params),%edi
107	movl $(PARAM_SIZE/4),%ecx
108	cld
109	rep
110	movsl
111	movl pa(boot_params) + NEW_CL_POINTER,%esi
112	andl %esi,%esi
113	jz 1f			# No command line
114	movl $pa(boot_command_line),%edi
115	movl $(COMMAND_LINE_SIZE/4),%ecx
116	rep
117	movsl
1181:
119
120#ifdef CONFIG_OLPC
121	/* save OFW's pgdir table for later use when calling into OFW */
122	movl %cr3, %eax
123	movl %eax, pa(olpc_ofw_pgd)
124#endif
125
126#ifdef CONFIG_MICROCODE
127	/* Early load ucode on BSP. */
128	call load_ucode_bsp
129#endif
130
131	/* Create early pagetables. */
132	call  mk_early_pgtbl_32
133
134	/* Do early initialization of the fixmap area */
135	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
136#ifdef  CONFIG_X86_PAE
137#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
138	movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
139#else
140	movl %eax,pa(initial_page_table+0xffc)
141#endif
142
143#ifdef CONFIG_PARAVIRT
144	/* This is can only trip for a broken bootloader... */
145	cmpw $0x207, pa(boot_params + BP_version)
146	jb .Ldefault_entry
147
148	/* Paravirt-compatible boot parameters.  Look to see what architecture
149		we're booting under. */
150	movl pa(boot_params + BP_hardware_subarch), %eax
151	cmpl $num_subarch_entries, %eax
152	jae .Lbad_subarch
153
154	movl pa(subarch_entries)(,%eax,4), %eax
155	subl $__PAGE_OFFSET, %eax
156	jmp *%eax
157
158.Lbad_subarch:
159SYM_INNER_LABEL_ALIGN(xen_entry, SYM_L_WEAK)
160	/* Unknown implementation; there's really
161	   nothing we can do at this point. */
162	ud2a
163
164	__INITDATA
165
166subarch_entries:
167	.long .Ldefault_entry		/* normal x86/PC */
168	.long xen_entry			/* Xen hypervisor */
169	.long .Ldefault_entry		/* Moorestown MID */
170num_subarch_entries = (. - subarch_entries) / 4
171.previous
172#else
173	jmp .Ldefault_entry
174#endif /* CONFIG_PARAVIRT */
175SYM_CODE_END(startup_32)
176
177#ifdef CONFIG_HOTPLUG_CPU
178/*
179 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
180 * up already except stack. We just set up stack here. Then call
181 * start_secondary().
182 */
183SYM_FUNC_START(start_cpu0)
184	movl initial_stack, %ecx
185	movl %ecx, %esp
186	call *(initial_code)
1871:	jmp 1b
188SYM_FUNC_END(start_cpu0)
189#endif
190
191/*
192 * Non-boot CPU entry point; entered from trampoline.S
193 * We can't lgdt here, because lgdt itself uses a data segment, but
194 * we know the trampoline has already loaded the boot_gdt for us.
195 *
196 * If cpu hotplug is not supported then this code can go in init section
197 * which will be freed later
198 */
199SYM_FUNC_START(startup_32_smp)
200	cld
201	movl $(__BOOT_DS),%eax
202	movl %eax,%ds
203	movl %eax,%es
204	movl %eax,%fs
205	movl %eax,%gs
206	movl pa(initial_stack),%ecx
207	movl %eax,%ss
208	leal -__PAGE_OFFSET(%ecx),%esp
209
210#ifdef CONFIG_MICROCODE
211	/* Early load ucode on AP. */
212	call load_ucode_ap
213#endif
214
215.Ldefault_entry:
216	movl $(CR0_STATE & ~X86_CR0_PG),%eax
217	movl %eax,%cr0
218
219/*
220 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
221 * bits like NT set. This would confuse the debugger if this code is traced. So
222 * initialize them properly now before switching to protected mode. That means
223 * DF in particular (even though we have cleared it earlier after copying the
224 * command line) because GCC expects it.
225 */
226	pushl $0
227	popfl
228
229/*
230 * New page tables may be in 4Mbyte page mode and may be using the global pages.
231 *
232 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
233 * if and only if CPUID exists and has flags other than the FPU flag set.
234 */
235	movl $-1,pa(X86_CPUID)		# preset CPUID level
236	movl $X86_EFLAGS_ID,%ecx
237	pushl %ecx
238	popfl				# set EFLAGS=ID
239	pushfl
240	popl %eax			# get EFLAGS
241	testl $X86_EFLAGS_ID,%eax	# did EFLAGS.ID remained set?
242	jz .Lenable_paging		# hw disallowed setting of ID bit
243					# which means no CPUID and no CR4
244
245	xorl %eax,%eax
246	cpuid
247	movl %eax,pa(X86_CPUID)		# save largest std CPUID function
248
249	movl $1,%eax
250	cpuid
251	andl $~1,%edx			# Ignore CPUID.FPU
252	jz .Lenable_paging		# No flags or only CPUID.FPU = no CR4
253
254	movl pa(mmu_cr4_features),%eax
255	movl %eax,%cr4
256
257	testb $X86_CR4_PAE, %al		# check if PAE is enabled
258	jz .Lenable_paging
259
260	/* Check if extended functions are implemented */
261	movl $0x80000000, %eax
262	cpuid
263	/* Value must be in the range 0x80000001 to 0x8000ffff */
264	subl $0x80000001, %eax
265	cmpl $(0x8000ffff-0x80000001), %eax
266	ja .Lenable_paging
267
268	/* Clear bogus XD_DISABLE bits */
269	call verify_cpu
270
271	mov $0x80000001, %eax
272	cpuid
273	/* Execute Disable bit supported? */
274	btl $(X86_FEATURE_NX & 31), %edx
275	jnc .Lenable_paging
276
277	/* Setup EFER (Extended Feature Enable Register) */
278	movl $MSR_EFER, %ecx
279	rdmsr
280
281	btsl $_EFER_NX, %eax
282	/* Make changes effective */
283	wrmsr
284
285.Lenable_paging:
286
287/*
288 * Enable paging
289 */
290	movl $pa(initial_page_table), %eax
291	movl %eax,%cr3		/* set the page table pointer.. */
292	movl $CR0_STATE,%eax
293	movl %eax,%cr0		/* ..and set paging (PG) bit */
294	ljmp $__BOOT_CS,$1f	/* Clear prefetch and normalize %eip */
2951:
296	/* Shift the stack pointer to a virtual address */
297	addl $__PAGE_OFFSET, %esp
298
299/*
300 * start system 32-bit setup. We need to re-do some of the things done
301 * in 16-bit mode for the "real" operations.
302 */
303	movl setup_once_ref,%eax
304	andl %eax,%eax
305	jz 1f				# Did we do this already?
306	call *%eax
3071:
308
309/*
310 * Check if it is 486
311 */
312	movb $4,X86			# at least 486
313	cmpl $-1,X86_CPUID
314	je .Lis486
315
316	/* get vendor info */
317	xorl %eax,%eax			# call CPUID with 0 -> return vendor ID
318	cpuid
319	movl %eax,X86_CPUID		# save CPUID level
320	movl %ebx,X86_VENDOR_ID		# lo 4 chars
321	movl %edx,X86_VENDOR_ID+4	# next 4 chars
322	movl %ecx,X86_VENDOR_ID+8	# last 4 chars
323
324	orl %eax,%eax			# do we have processor info as well?
325	je .Lis486
326
327	movl $1,%eax		# Use the CPUID instruction to get CPU type
328	cpuid
329	movb %al,%cl		# save reg for future use
330	andb $0x0f,%ah		# mask processor family
331	movb %ah,X86
332	andb $0xf0,%al		# mask model
333	shrb $4,%al
334	movb %al,X86_MODEL
335	andb $0x0f,%cl		# mask mask revision
336	movb %cl,X86_STEPPING
337	movl %edx,X86_CAPABILITY
338
339.Lis486:
340	movl $0x50022,%ecx	# set AM, WP, NE and MP
341	movl %cr0,%eax
342	andl $0x80000011,%eax	# Save PG,PE,ET
343	orl %ecx,%eax
344	movl %eax,%cr0
345
346	lgdt early_gdt_descr
347	ljmp $(__KERNEL_CS),$1f
3481:	movl $(__KERNEL_DS),%eax	# reload all the segment registers
349	movl %eax,%ss			# after changing gdt.
350
351	movl $(__USER_DS),%eax		# DS/ES contains default USER segment
352	movl %eax,%ds
353	movl %eax,%es
354
355	movl $(__KERNEL_PERCPU), %eax
356	movl %eax,%fs			# set this cpu's percpu
357
358	movl $(__KERNEL_STACK_CANARY),%eax
359	movl %eax,%gs
360
361	xorl %eax,%eax			# Clear LDT
362	lldt %ax
363
364	call *(initial_code)
3651:	jmp 1b
366SYM_FUNC_END(startup_32_smp)
367
368#include "verify_cpu.S"
369
370/*
371 *  setup_once
372 *
373 *  The setup work we only want to run on the BSP.
374 *
375 *  Warning: %esi is live across this function.
376 */
377__INIT
378setup_once:
379#ifdef CONFIG_STACKPROTECTOR
380	/*
381	 * Configure the stack canary. The linker can't handle this by
382	 * relocation.  Manually set base address in stack canary
383	 * segment descriptor.
384	 */
385	movl $gdt_page,%eax
386	movl $stack_canary,%ecx
387	movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
388	shrl $16, %ecx
389	movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
390	movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
391#endif
392
393	andl $0,setup_once_ref	/* Once is enough, thanks */
394	ret
395
396SYM_FUNC_START(early_idt_handler_array)
397	# 36(%esp) %eflags
398	# 32(%esp) %cs
399	# 28(%esp) %eip
400	# 24(%rsp) error code
401	i = 0
402	.rept NUM_EXCEPTION_VECTORS
403	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
404	pushl $0		# Dummy error code, to make stack frame uniform
405	.endif
406	pushl $i		# 20(%esp) Vector number
407	jmp early_idt_handler_common
408	i = i + 1
409	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
410	.endr
411SYM_FUNC_END(early_idt_handler_array)
412
413SYM_CODE_START_LOCAL(early_idt_handler_common)
414	/*
415	 * The stack is the hardware frame, an error code or zero, and the
416	 * vector number.
417	 */
418	cld
419
420	incl %ss:early_recursion_flag
421
422	/* The vector number is in pt_regs->gs */
423
424	cld
425	pushl	%fs		/* pt_regs->fs (__fsh varies by model) */
426	pushl	%es		/* pt_regs->es (__esh varies by model) */
427	pushl	%ds		/* pt_regs->ds (__dsh varies by model) */
428	pushl	%eax		/* pt_regs->ax */
429	pushl	%ebp		/* pt_regs->bp */
430	pushl	%edi		/* pt_regs->di */
431	pushl	%esi		/* pt_regs->si */
432	pushl	%edx		/* pt_regs->dx */
433	pushl	%ecx		/* pt_regs->cx */
434	pushl	%ebx		/* pt_regs->bx */
435
436	/* Fix up DS and ES */
437	movl	$(__KERNEL_DS), %ecx
438	movl	%ecx, %ds
439	movl	%ecx, %es
440
441	/* Load the vector number into EDX */
442	movl	PT_GS(%esp), %edx
443
444	/* Load GS into pt_regs->gs (and maybe clobber __gsh) */
445	movw	%gs, PT_GS(%esp)
446
447	movl	%esp, %eax	/* args are pt_regs (EAX), trapnr (EDX) */
448	call	early_fixup_exception
449
450	popl	%ebx		/* pt_regs->bx */
451	popl	%ecx		/* pt_regs->cx */
452	popl	%edx		/* pt_regs->dx */
453	popl	%esi		/* pt_regs->si */
454	popl	%edi		/* pt_regs->di */
455	popl	%ebp		/* pt_regs->bp */
456	popl	%eax		/* pt_regs->ax */
457	popl	%ds		/* pt_regs->ds (always ignores __dsh) */
458	popl	%es		/* pt_regs->es (always ignores __esh) */
459	popl	%fs		/* pt_regs->fs (always ignores __fsh) */
460	popl	%gs		/* pt_regs->gs (always ignores __gsh) */
461	decl	%ss:early_recursion_flag
462	addl	$4, %esp	/* pop pt_regs->orig_ax */
463	iret
464SYM_CODE_END(early_idt_handler_common)
465
466/* This is the default interrupt "handler" :-) */
467SYM_FUNC_START(early_ignore_irq)
468	cld
469#ifdef CONFIG_PRINTK
470	pushl %eax
471	pushl %ecx
472	pushl %edx
473	pushl %es
474	pushl %ds
475	movl $(__KERNEL_DS),%eax
476	movl %eax,%ds
477	movl %eax,%es
478	cmpl $2,early_recursion_flag
479	je hlt_loop
480	incl early_recursion_flag
481	pushl 16(%esp)
482	pushl 24(%esp)
483	pushl 32(%esp)
484	pushl 40(%esp)
485	pushl $int_msg
486	call printk
487
488	call dump_stack
489
490	addl $(5*4),%esp
491	popl %ds
492	popl %es
493	popl %edx
494	popl %ecx
495	popl %eax
496#endif
497	iret
498
499hlt_loop:
500	hlt
501	jmp hlt_loop
502SYM_FUNC_END(early_ignore_irq)
503
504__INITDATA
505	.align 4
506SYM_DATA(early_recursion_flag, .long 0)
507
508__REFDATA
509	.align 4
510SYM_DATA(initial_code,		.long i386_start_kernel)
511SYM_DATA(setup_once_ref,	.long setup_once)
512
513#ifdef CONFIG_PAGE_TABLE_ISOLATION
514#define	PGD_ALIGN	(2 * PAGE_SIZE)
515#define PTI_USER_PGD_FILL	1024
516#else
517#define	PGD_ALIGN	(PAGE_SIZE)
518#define PTI_USER_PGD_FILL	0
519#endif
520/*
521 * BSS section
522 */
523__PAGE_ALIGNED_BSS
524	.align PGD_ALIGN
525#ifdef CONFIG_X86_PAE
526.globl initial_pg_pmd
527initial_pg_pmd:
528	.fill 1024*KPMDS,4,0
529#else
530.globl initial_page_table
531initial_page_table:
532	.fill 1024,4,0
533#endif
534	.align PGD_ALIGN
535initial_pg_fixmap:
536	.fill 1024,4,0
537.globl swapper_pg_dir
538	.align PGD_ALIGN
539swapper_pg_dir:
540	.fill 1024,4,0
541	.fill PTI_USER_PGD_FILL,4,0
542.globl empty_zero_page
543empty_zero_page:
544	.fill 4096,1,0
545EXPORT_SYMBOL(empty_zero_page)
546
547/*
548 * This starts the data section.
549 */
550#ifdef CONFIG_X86_PAE
551__PAGE_ALIGNED_DATA
552	/* Page-aligned for the benefit of paravirt? */
553	.align PGD_ALIGN
554SYM_DATA_START(initial_page_table)
555	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
556# if KPMDS == 3
557	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
558	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
559	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
560# elif KPMDS == 2
561	.long	0,0
562	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
563	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
564# elif KPMDS == 1
565	.long	0,0
566	.long	0,0
567	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
568# else
569#  error "Kernel PMDs should be 1, 2 or 3"
570# endif
571	.align PAGE_SIZE		/* needs to be page-sized too */
572
573#ifdef CONFIG_PAGE_TABLE_ISOLATION
574	/*
575	 * PTI needs another page so sync_initial_pagetable() works correctly
576	 * and does not scribble over the data which is placed behind the
577	 * actual initial_page_table. See clone_pgd_range().
578	 */
579	.fill 1024, 4, 0
580#endif
581
582SYM_DATA_END(initial_page_table)
583#endif
584
585.data
586.balign 4
587/*
588 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder
589 * reliably detect the end of the stack.
590 */
591SYM_DATA(initial_stack,
592		.long init_thread_union + THREAD_SIZE -
593		SIZEOF_PTREGS - TOP_OF_KERNEL_STACK_PADDING)
594
595__INITRODATA
596int_msg:
597	.asciz "Unknown interrupt or fault at: %p %p %p\n"
598
599#include "../../x86/xen/xen-head.S"
600
601/*
602 * The IDT and GDT 'descriptors' are a strange 48-bit object
603 * only used by the lidt and lgdt instructions. They are not
604 * like usual segment descriptors - they consist of a 16-bit
605 * segment size, and 32-bit linear address value:
606 */
607
608	.data
609	ALIGN
610# early boot GDT descriptor (must use 1:1 address mapping)
611	.word 0				# 32 bit align gdt_desc.address
612SYM_DATA_START_LOCAL(boot_gdt_descr)
613	.word __BOOT_DS+7
614	.long boot_gdt - __PAGE_OFFSET
615SYM_DATA_END(boot_gdt_descr)
616
617# boot GDT descriptor (later on used by CPU#0):
618	.word 0				# 32 bit align gdt_desc.address
619SYM_DATA_START(early_gdt_descr)
620	.word GDT_ENTRIES*8-1
621	.long gdt_page			/* Overwritten for secondary CPUs */
622SYM_DATA_END(early_gdt_descr)
623
624/*
625 * The boot_gdt must mirror the equivalent in setup.S and is
626 * used only for booting.
627 */
628	.align L1_CACHE_BYTES
629SYM_DATA_START(boot_gdt)
630	.fill GDT_ENTRY_BOOT_CS,8,0
631	.quad 0x00cf9a000000ffff	/* kernel 4GB code at 0x00000000 */
632	.quad 0x00cf92000000ffff	/* kernel 4GB data at 0x00000000 */
633SYM_DATA_END(boot_gdt)
634