xref: /linux/arch/x86/kernel/head_32.S (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1/*
2 *
3 *  Copyright (C) 1991, 1992  Linus Torvalds
4 *
5 *  Enhanced CPU detection and feature setting code by Mike Jagdis
6 *  and Martin Mares, November 1997.
7 */
8
9.text
10#include <linux/threads.h>
11#include <linux/init.h>
12#include <linux/linkage.h>
13#include <asm/segment.h>
14#include <asm/page_types.h>
15#include <asm/pgtable_types.h>
16#include <asm/cache.h>
17#include <asm/thread_info.h>
18#include <asm/asm-offsets.h>
19#include <asm/setup.h>
20#include <asm/processor-flags.h>
21#include <asm/msr-index.h>
22#include <asm/cpufeatures.h>
23#include <asm/percpu.h>
24#include <asm/nops.h>
25#include <asm/bootparam.h>
26#include <asm/export.h>
27#include <asm/pgtable_32.h>
28
29/* Physical address */
30#define pa(X) ((X) - __PAGE_OFFSET)
31
32/*
33 * References to members of the new_cpu_data structure.
34 */
35
36#define X86		new_cpu_data+CPUINFO_x86
37#define X86_VENDOR	new_cpu_data+CPUINFO_x86_vendor
38#define X86_MODEL	new_cpu_data+CPUINFO_x86_model
39#define X86_MASK	new_cpu_data+CPUINFO_x86_mask
40#define X86_HARD_MATH	new_cpu_data+CPUINFO_hard_math
41#define X86_CPUID	new_cpu_data+CPUINFO_cpuid_level
42#define X86_CAPABILITY	new_cpu_data+CPUINFO_x86_capability
43#define X86_VENDOR_ID	new_cpu_data+CPUINFO_x86_vendor_id
44
45
46#define SIZEOF_PTREGS 17*4
47
48/*
49 * Worst-case size of the kernel mapping we need to make:
50 * a relocatable kernel can live anywhere in lowmem, so we need to be able
51 * to map all of lowmem.
52 */
53KERNEL_PAGES = LOWMEM_PAGES
54
55INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
56RESERVE_BRK(pagetables, INIT_MAP_SIZE)
57
58/*
59 * 32-bit kernel entrypoint; only used by the boot CPU.  On entry,
60 * %esi points to the real-mode code as a 32-bit pointer.
61 * CS and DS must be 4 GB flat segments, but we don't depend on
62 * any particular GDT layout, because we load our own as soon as we
63 * can.
64 */
65__HEAD
66ENTRY(startup_32)
67	movl pa(initial_stack),%ecx
68
69	/* test KEEP_SEGMENTS flag to see if the bootloader is asking
70		us to not reload segments */
71	testb $KEEP_SEGMENTS, BP_loadflags(%esi)
72	jnz 2f
73
74/*
75 * Set segments to known values.
76 */
77	lgdt pa(boot_gdt_descr)
78	movl $(__BOOT_DS),%eax
79	movl %eax,%ds
80	movl %eax,%es
81	movl %eax,%fs
82	movl %eax,%gs
83	movl %eax,%ss
842:
85	leal -__PAGE_OFFSET(%ecx),%esp
86
87/*
88 * Clear BSS first so that there are no surprises...
89 */
90	cld
91	xorl %eax,%eax
92	movl $pa(__bss_start),%edi
93	movl $pa(__bss_stop),%ecx
94	subl %edi,%ecx
95	shrl $2,%ecx
96	rep ; stosl
97/*
98 * Copy bootup parameters out of the way.
99 * Note: %esi still has the pointer to the real-mode data.
100 * With the kexec as boot loader, parameter segment might be loaded beyond
101 * kernel image and might not even be addressable by early boot page tables.
102 * (kexec on panic case). Hence copy out the parameters before initializing
103 * page tables.
104 */
105	movl $pa(boot_params),%edi
106	movl $(PARAM_SIZE/4),%ecx
107	cld
108	rep
109	movsl
110	movl pa(boot_params) + NEW_CL_POINTER,%esi
111	andl %esi,%esi
112	jz 1f			# No command line
113	movl $pa(boot_command_line),%edi
114	movl $(COMMAND_LINE_SIZE/4),%ecx
115	rep
116	movsl
1171:
118
119#ifdef CONFIG_OLPC
120	/* save OFW's pgdir table for later use when calling into OFW */
121	movl %cr3, %eax
122	movl %eax, pa(olpc_ofw_pgd)
123#endif
124
125#ifdef CONFIG_MICROCODE
126	/* Early load ucode on BSP. */
127	call load_ucode_bsp
128#endif
129
130	/* Create early pagetables. */
131	call  mk_early_pgtbl_32
132
133	/* Do early initialization of the fixmap area */
134	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
135#ifdef  CONFIG_X86_PAE
136#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
137	movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
138#else
139	movl %eax,pa(initial_page_table+0xffc)
140#endif
141
142#ifdef CONFIG_PARAVIRT
143	/* This is can only trip for a broken bootloader... */
144	cmpw $0x207, pa(boot_params + BP_version)
145	jb .Ldefault_entry
146
147	/* Paravirt-compatible boot parameters.  Look to see what architecture
148		we're booting under. */
149	movl pa(boot_params + BP_hardware_subarch), %eax
150	cmpl $num_subarch_entries, %eax
151	jae .Lbad_subarch
152
153	movl pa(subarch_entries)(,%eax,4), %eax
154	subl $__PAGE_OFFSET, %eax
155	jmp *%eax
156
157.Lbad_subarch:
158WEAK(lguest_entry)
159WEAK(xen_entry)
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 lguest_entry		/* lguest hypervisor */
169	.long xen_entry			/* Xen hypervisor */
170	.long .Ldefault_entry		/* Moorestown MID */
171num_subarch_entries = (. - subarch_entries) / 4
172.previous
173#else
174	jmp .Ldefault_entry
175#endif /* CONFIG_PARAVIRT */
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 */
183ENTRY(start_cpu0)
184	movl initial_stack, %ecx
185	movl %ecx, %esp
186	call *(initial_code)
1871:	jmp 1b
188ENDPROC(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 */
199ENTRY(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#define CR0_STATE	(X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
217			 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
218			 X86_CR0_PG)
219	movl $(CR0_STATE & ~X86_CR0_PG),%eax
220	movl %eax,%cr0
221
222/*
223 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
224 * bits like NT set. This would confuse the debugger if this code is traced. So
225 * initialize them properly now before switching to protected mode. That means
226 * DF in particular (even though we have cleared it earlier after copying the
227 * command line) because GCC expects it.
228 */
229	pushl $0
230	popfl
231
232/*
233 * New page tables may be in 4Mbyte page mode and may be using the global pages.
234 *
235 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
236 * if and only if CPUID exists and has flags other than the FPU flag set.
237 */
238	movl $-1,pa(X86_CPUID)		# preset CPUID level
239	movl $X86_EFLAGS_ID,%ecx
240	pushl %ecx
241	popfl				# set EFLAGS=ID
242	pushfl
243	popl %eax			# get EFLAGS
244	testl $X86_EFLAGS_ID,%eax	# did EFLAGS.ID remained set?
245	jz .Lenable_paging		# hw disallowed setting of ID bit
246					# which means no CPUID and no CR4
247
248	xorl %eax,%eax
249	cpuid
250	movl %eax,pa(X86_CPUID)		# save largest std CPUID function
251
252	movl $1,%eax
253	cpuid
254	andl $~1,%edx			# Ignore CPUID.FPU
255	jz .Lenable_paging		# No flags or only CPUID.FPU = no CR4
256
257	movl pa(mmu_cr4_features),%eax
258	movl %eax,%cr4
259
260	testb $X86_CR4_PAE, %al		# check if PAE is enabled
261	jz .Lenable_paging
262
263	/* Check if extended functions are implemented */
264	movl $0x80000000, %eax
265	cpuid
266	/* Value must be in the range 0x80000001 to 0x8000ffff */
267	subl $0x80000001, %eax
268	cmpl $(0x8000ffff-0x80000001), %eax
269	ja .Lenable_paging
270
271	/* Clear bogus XD_DISABLE bits */
272	call verify_cpu
273
274	mov $0x80000001, %eax
275	cpuid
276	/* Execute Disable bit supported? */
277	btl $(X86_FEATURE_NX & 31), %edx
278	jnc .Lenable_paging
279
280	/* Setup EFER (Extended Feature Enable Register) */
281	movl $MSR_EFER, %ecx
282	rdmsr
283
284	btsl $_EFER_NX, %eax
285	/* Make changes effective */
286	wrmsr
287
288.Lenable_paging:
289
290/*
291 * Enable paging
292 */
293	movl $pa(initial_page_table), %eax
294	movl %eax,%cr3		/* set the page table pointer.. */
295	movl $CR0_STATE,%eax
296	movl %eax,%cr0		/* ..and set paging (PG) bit */
297	ljmp $__BOOT_CS,$1f	/* Clear prefetch and normalize %eip */
2981:
299	/* Shift the stack pointer to a virtual address */
300	addl $__PAGE_OFFSET, %esp
301
302/*
303 * start system 32-bit setup. We need to re-do some of the things done
304 * in 16-bit mode for the "real" operations.
305 */
306	movl setup_once_ref,%eax
307	andl %eax,%eax
308	jz 1f				# Did we do this already?
309	call *%eax
3101:
311
312/*
313 * Check if it is 486
314 */
315	movb $4,X86			# at least 486
316	cmpl $-1,X86_CPUID
317	je .Lis486
318
319	/* get vendor info */
320	xorl %eax,%eax			# call CPUID with 0 -> return vendor ID
321	cpuid
322	movl %eax,X86_CPUID		# save CPUID level
323	movl %ebx,X86_VENDOR_ID		# lo 4 chars
324	movl %edx,X86_VENDOR_ID+4	# next 4 chars
325	movl %ecx,X86_VENDOR_ID+8	# last 4 chars
326
327	orl %eax,%eax			# do we have processor info as well?
328	je .Lis486
329
330	movl $1,%eax		# Use the CPUID instruction to get CPU type
331	cpuid
332	movb %al,%cl		# save reg for future use
333	andb $0x0f,%ah		# mask processor family
334	movb %ah,X86
335	andb $0xf0,%al		# mask model
336	shrb $4,%al
337	movb %al,X86_MODEL
338	andb $0x0f,%cl		# mask mask revision
339	movb %cl,X86_MASK
340	movl %edx,X86_CAPABILITY
341
342.Lis486:
343	movl $0x50022,%ecx	# set AM, WP, NE and MP
344	movl %cr0,%eax
345	andl $0x80000011,%eax	# Save PG,PE,ET
346	orl %ecx,%eax
347	movl %eax,%cr0
348
349	lgdt early_gdt_descr
350	lidt idt_descr
351	ljmp $(__KERNEL_CS),$1f
3521:	movl $(__KERNEL_DS),%eax	# reload all the segment registers
353	movl %eax,%ss			# after changing gdt.
354
355	movl $(__USER_DS),%eax		# DS/ES contains default USER segment
356	movl %eax,%ds
357	movl %eax,%es
358
359	movl $(__KERNEL_PERCPU), %eax
360	movl %eax,%fs			# set this cpu's percpu
361
362	movl $(__KERNEL_STACK_CANARY),%eax
363	movl %eax,%gs
364
365	xorl %eax,%eax			# Clear LDT
366	lldt %ax
367
368	call *(initial_code)
3691:	jmp 1b
370ENDPROC(startup_32_smp)
371
372#include "verify_cpu.S"
373
374/*
375 *  setup_once
376 *
377 *  The setup work we only want to run on the BSP.
378 *
379 *  Warning: %esi is live across this function.
380 */
381__INIT
382setup_once:
383	/*
384	 * Set up a idt with 256 interrupt gates that push zero if there
385	 * is no error code and then jump to early_idt_handler_common.
386	 * It doesn't actually load the idt - that needs to be done on
387	 * each CPU. Interrupts are enabled elsewhere, when we can be
388	 * relatively sure everything is ok.
389	 */
390
391	movl $idt_table,%edi
392	movl $early_idt_handler_array,%eax
393	movl $NUM_EXCEPTION_VECTORS,%ecx
3941:
395	movl %eax,(%edi)
396	movl %eax,4(%edi)
397	/* interrupt gate, dpl=0, present */
398	movl $(0x8E000000 + __KERNEL_CS),2(%edi)
399	addl $EARLY_IDT_HANDLER_SIZE,%eax
400	addl $8,%edi
401	loop 1b
402
403	movl $256 - NUM_EXCEPTION_VECTORS,%ecx
404	movl $ignore_int,%edx
405	movl $(__KERNEL_CS << 16),%eax
406	movw %dx,%ax		/* selector = 0x0010 = cs */
407	movw $0x8E00,%dx	/* interrupt gate - dpl=0, present */
4082:
409	movl %eax,(%edi)
410	movl %edx,4(%edi)
411	addl $8,%edi
412	loop 2b
413
414#ifdef CONFIG_CC_STACKPROTECTOR
415	/*
416	 * Configure the stack canary. The linker can't handle this by
417	 * relocation.  Manually set base address in stack canary
418	 * segment descriptor.
419	 */
420	movl $gdt_page,%eax
421	movl $stack_canary,%ecx
422	movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
423	shrl $16, %ecx
424	movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
425	movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
426#endif
427
428	andl $0,setup_once_ref	/* Once is enough, thanks */
429	ret
430
431ENTRY(early_idt_handler_array)
432	# 36(%esp) %eflags
433	# 32(%esp) %cs
434	# 28(%esp) %eip
435	# 24(%rsp) error code
436	i = 0
437	.rept NUM_EXCEPTION_VECTORS
438	.ifeq (EXCEPTION_ERRCODE_MASK >> i) & 1
439	pushl $0		# Dummy error code, to make stack frame uniform
440	.endif
441	pushl $i		# 20(%esp) Vector number
442	jmp early_idt_handler_common
443	i = i + 1
444	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
445	.endr
446ENDPROC(early_idt_handler_array)
447
448early_idt_handler_common:
449	/*
450	 * The stack is the hardware frame, an error code or zero, and the
451	 * vector number.
452	 */
453	cld
454
455	incl %ss:early_recursion_flag
456
457	/* The vector number is in pt_regs->gs */
458
459	cld
460	pushl	%fs		/* pt_regs->fs */
461	movw	$0, 2(%esp)	/* clear high bits (some CPUs leave garbage) */
462	pushl	%es		/* pt_regs->es */
463	movw	$0, 2(%esp)	/* clear high bits (some CPUs leave garbage) */
464	pushl	%ds		/* pt_regs->ds */
465	movw	$0, 2(%esp)	/* clear high bits (some CPUs leave garbage) */
466	pushl	%eax		/* pt_regs->ax */
467	pushl	%ebp		/* pt_regs->bp */
468	pushl	%edi		/* pt_regs->di */
469	pushl	%esi		/* pt_regs->si */
470	pushl	%edx		/* pt_regs->dx */
471	pushl	%ecx		/* pt_regs->cx */
472	pushl	%ebx		/* pt_regs->bx */
473
474	/* Fix up DS and ES */
475	movl	$(__KERNEL_DS), %ecx
476	movl	%ecx, %ds
477	movl	%ecx, %es
478
479	/* Load the vector number into EDX */
480	movl	PT_GS(%esp), %edx
481
482	/* Load GS into pt_regs->gs and clear high bits */
483	movw	%gs, PT_GS(%esp)
484	movw	$0, PT_GS+2(%esp)
485
486	movl	%esp, %eax	/* args are pt_regs (EAX), trapnr (EDX) */
487	call	early_fixup_exception
488
489	popl	%ebx		/* pt_regs->bx */
490	popl	%ecx		/* pt_regs->cx */
491	popl	%edx		/* pt_regs->dx */
492	popl	%esi		/* pt_regs->si */
493	popl	%edi		/* pt_regs->di */
494	popl	%ebp		/* pt_regs->bp */
495	popl	%eax		/* pt_regs->ax */
496	popl	%ds		/* pt_regs->ds */
497	popl	%es		/* pt_regs->es */
498	popl	%fs		/* pt_regs->fs */
499	popl	%gs		/* pt_regs->gs */
500	decl	%ss:early_recursion_flag
501	addl	$4, %esp	/* pop pt_regs->orig_ax */
502	iret
503ENDPROC(early_idt_handler_common)
504
505/* This is the default interrupt "handler" :-) */
506	ALIGN
507ignore_int:
508	cld
509#ifdef CONFIG_PRINTK
510	pushl %eax
511	pushl %ecx
512	pushl %edx
513	pushl %es
514	pushl %ds
515	movl $(__KERNEL_DS),%eax
516	movl %eax,%ds
517	movl %eax,%es
518	cmpl $2,early_recursion_flag
519	je hlt_loop
520	incl early_recursion_flag
521	pushl 16(%esp)
522	pushl 24(%esp)
523	pushl 32(%esp)
524	pushl 40(%esp)
525	pushl $int_msg
526	call printk
527
528	call dump_stack
529
530	addl $(5*4),%esp
531	popl %ds
532	popl %es
533	popl %edx
534	popl %ecx
535	popl %eax
536#endif
537	iret
538
539hlt_loop:
540	hlt
541	jmp hlt_loop
542ENDPROC(ignore_int)
543__INITDATA
544	.align 4
545GLOBAL(early_recursion_flag)
546	.long 0
547
548__REFDATA
549	.align 4
550ENTRY(initial_code)
551	.long i386_start_kernel
552ENTRY(setup_once_ref)
553	.long setup_once
554
555/*
556 * BSS section
557 */
558__PAGE_ALIGNED_BSS
559	.align PAGE_SIZE
560#ifdef CONFIG_X86_PAE
561.globl initial_pg_pmd
562initial_pg_pmd:
563	.fill 1024*KPMDS,4,0
564#else
565.globl initial_page_table
566initial_page_table:
567	.fill 1024,4,0
568#endif
569initial_pg_fixmap:
570	.fill 1024,4,0
571.globl empty_zero_page
572empty_zero_page:
573	.fill 4096,1,0
574.globl swapper_pg_dir
575swapper_pg_dir:
576	.fill 1024,4,0
577EXPORT_SYMBOL(empty_zero_page)
578
579/*
580 * This starts the data section.
581 */
582#ifdef CONFIG_X86_PAE
583__PAGE_ALIGNED_DATA
584	/* Page-aligned for the benefit of paravirt? */
585	.align PAGE_SIZE
586ENTRY(initial_page_table)
587	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
588# if KPMDS == 3
589	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
590	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
591	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
592# elif KPMDS == 2
593	.long	0,0
594	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
595	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
596# elif KPMDS == 1
597	.long	0,0
598	.long	0,0
599	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
600# else
601#  error "Kernel PMDs should be 1, 2 or 3"
602# endif
603	.align PAGE_SIZE		/* needs to be page-sized too */
604#endif
605
606.data
607.balign 4
608ENTRY(initial_stack)
609	/*
610	 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel
611	 * unwinder reliably detect the end of the stack.
612	 */
613	.long init_thread_union + THREAD_SIZE - SIZEOF_PTREGS - \
614	      TOP_OF_KERNEL_STACK_PADDING;
615
616__INITRODATA
617int_msg:
618	.asciz "Unknown interrupt or fault at: %p %p %p\n"
619
620#include "../../x86/xen/xen-head.S"
621
622/*
623 * The IDT and GDT 'descriptors' are a strange 48-bit object
624 * only used by the lidt and lgdt instructions. They are not
625 * like usual segment descriptors - they consist of a 16-bit
626 * segment size, and 32-bit linear address value:
627 */
628
629	.data
630.globl boot_gdt_descr
631.globl idt_descr
632
633	ALIGN
634# early boot GDT descriptor (must use 1:1 address mapping)
635	.word 0				# 32 bit align gdt_desc.address
636boot_gdt_descr:
637	.word __BOOT_DS+7
638	.long boot_gdt - __PAGE_OFFSET
639
640	.word 0				# 32-bit align idt_desc.address
641idt_descr:
642	.word IDT_ENTRIES*8-1		# idt contains 256 entries
643	.long idt_table
644
645# boot GDT descriptor (later on used by CPU#0):
646	.word 0				# 32 bit align gdt_desc.address
647ENTRY(early_gdt_descr)
648	.word GDT_ENTRIES*8-1
649	.long gdt_page			/* Overwritten for secondary CPUs */
650
651/*
652 * The boot_gdt must mirror the equivalent in setup.S and is
653 * used only for booting.
654 */
655	.align L1_CACHE_BYTES
656ENTRY(boot_gdt)
657	.fill GDT_ENTRY_BOOT_CS,8,0
658	.quad 0x00cf9a000000ffff	/* kernel 4GB code at 0x00000000 */
659	.quad 0x00cf92000000ffff	/* kernel 4GB data at 0x00000000 */
660