xref: /linux/arch/x86/kernel/head_32.S (revision 621cde16e49b3ecf7d59a8106a20aaebfb4a59a9) 
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/export.h>
12 #include <linux/threads.h>
13 #include <linux/init.h>
14 #include <linux/linkage.h>
15 #include <asm/segment.h>
16 #include <asm/page_types.h>
17 #include <asm/pgtable_types.h>
18 #include <asm/cache.h>
19 #include <asm/thread_info.h>
20 #include <asm/asm-offsets.h>
21 #include <asm/setup.h>
22 #include <asm/processor-flags.h>
23 #include <asm/msr-index.h>
24 #include <asm/cpufeatures.h>
25 #include <asm/percpu.h>
26 #include <asm/nops.h>
27 #include <asm/nospec-branch.h>
28 #include <asm/bootparam.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  * Worst-case size of the kernel mapping we need to make:
49  * a relocatable kernel can live anywhere in lowmem, so we need to be able
50  * to map all of lowmem.
51  */
52 KERNEL_PAGES = LOWMEM_PAGES
53 
54 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
55 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
56 
57 /*
58  * 32-bit kernel entrypoint; only used by the boot CPU.  On entry,
59  * %esi points to the real-mode code as a 32-bit pointer.
60  * CS and DS must be 4 GB flat segments, but we don't depend on
61  * any particular GDT layout, because we load our own as soon as we
62  * can.
63  */
64 __HEAD
65 SYM_CODE_START(startup_32)
66 	movl pa(initial_stack),%ecx
67 
68 /*
69  * Set segments to known values.
70  */
71 	lgdt pa(boot_gdt_descr)
72 	movl $(__BOOT_DS),%eax
73 	movl %eax,%ds
74 	movl %eax,%es
75 	movl %eax,%fs
76 	movl %eax,%gs
77 	movl %eax,%ss
78 	leal -__PAGE_OFFSET(%ecx),%esp
79 
80 /*
81  * Clear BSS first so that there are no surprises...
82  */
83 	cld
84 	xorl %eax,%eax
85 	movl $pa(__bss_start),%edi
86 	movl $pa(__bss_stop),%ecx
87 	subl %edi,%ecx
88 	shrl $2,%ecx
89 	rep ; stosl
90 /*
91  * Copy bootup parameters out of the way.
92  * Note: %esi still has the pointer to the real-mode data.
93  * With the kexec as boot loader, parameter segment might be loaded beyond
94  * kernel image and might not even be addressable by early boot page tables.
95  * (kexec on panic case). Hence copy out the parameters before initializing
96  * page tables.
97  */
98 	movl $pa(boot_params),%edi
99 	movl $(PARAM_SIZE/4),%ecx
100 	cld
101 	rep
102 	movsl
103 	movl pa(boot_params) + NEW_CL_POINTER,%esi
104 	andl %esi,%esi
105 	jz 1f			# No command line
106 	movl $pa(boot_command_line),%edi
107 	movl $(COMMAND_LINE_SIZE/4),%ecx
108 	rep
109 	movsl
110 1:
111 
112 #ifdef CONFIG_OLPC
113 	/* save OFW's pgdir table for later use when calling into OFW */
114 	movl %cr3, %eax
115 	movl %eax, pa(olpc_ofw_pgd)
116 #endif
117 
118 	/* Create early pagetables. */
119 	call  mk_early_pgtbl_32
120 
121 	/* Do early initialization of the fixmap area */
122 	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
123 #ifdef  CONFIG_X86_PAE
124 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
125 	movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
126 #else
127 	movl %eax,pa(initial_page_table+0xffc)
128 #endif
129 
130 	jmp .Ldefault_entry
131 SYM_CODE_END(startup_32)
132 
133 /*
134  * Non-boot CPU entry point; entered from trampoline.S
135  * We can't lgdt here, because lgdt itself uses a data segment, but
136  * we know the trampoline has already loaded the boot_gdt for us.
137  *
138  * If cpu hotplug is not supported then this code can go in init section
139  * which will be freed later
140  */
141 SYM_FUNC_START(startup_32_smp)
142 	cld
143 	movl $(__BOOT_DS),%eax
144 	movl %eax,%ds
145 	movl %eax,%es
146 	movl %eax,%fs
147 	movl %eax,%gs
148 	movl pa(initial_stack),%ecx
149 	movl %eax,%ss
150 	leal -__PAGE_OFFSET(%ecx),%esp
151 
152 .Ldefault_entry:
153 	movl $(CR0_STATE & ~X86_CR0_PG),%eax
154 	movl %eax,%cr0
155 
156 /*
157  * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
158  * bits like NT set. This would confuse the debugger if this code is traced. So
159  * initialize them properly now before switching to protected mode. That means
160  * DF in particular (even though we have cleared it earlier after copying the
161  * command line) because GCC expects it.
162  */
163 	pushl $0
164 	popfl
165 
166 /*
167  * New page tables may be in 4Mbyte page mode and may be using the global pages.
168  *
169  * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
170  * if and only if CPUID exists and has flags other than the FPU flag set.
171  */
172 	movl $-1,pa(X86_CPUID)		# preset CPUID level
173 	movl $X86_EFLAGS_ID,%ecx
174 	pushl %ecx
175 	popfl				# set EFLAGS=ID
176 	pushfl
177 	popl %eax			# get EFLAGS
178 	testl $X86_EFLAGS_ID,%eax	# did EFLAGS.ID remained set?
179 	jz .Lenable_paging		# hw disallowed setting of ID bit
180 					# which means no CPUID and no CR4
181 
182 	xorl %eax,%eax
183 	cpuid
184 	movl %eax,pa(X86_CPUID)		# save largest std CPUID function
185 
186 	movl $1,%eax
187 	cpuid
188 	andl $~1,%edx			# Ignore CPUID.FPU
189 	jz .Lenable_paging		# No flags or only CPUID.FPU = no CR4
190 
191 	movl pa(mmu_cr4_features),%eax
192 	movl %eax,%cr4
193 
194 	testb $X86_CR4_PAE, %al		# check if PAE is enabled
195 	jz .Lenable_paging
196 
197 	/* Check if extended functions are implemented */
198 	movl $0x80000000, %eax
199 	cpuid
200 	/* Value must be in the range 0x80000001 to 0x8000ffff */
201 	subl $0x80000001, %eax
202 	cmpl $(0x8000ffff-0x80000001), %eax
203 	ja .Lenable_paging
204 
205 	/* Clear bogus XD_DISABLE bits */
206 	call verify_cpu
207 
208 	mov $0x80000001, %eax
209 	cpuid
210 	/* Execute Disable bit supported? */
211 	btl $(X86_FEATURE_NX & 31), %edx
212 	jnc .Lenable_paging
213 
214 	/* Setup EFER (Extended Feature Enable Register) */
215 	movl $MSR_EFER, %ecx
216 	rdmsr
217 
218 	btsl $_EFER_NX, %eax
219 	/* Make changes effective */
220 	wrmsr
221 
222 .Lenable_paging:
223 
224 /*
225  * Enable paging
226  */
227 	movl $pa(initial_page_table), %eax
228 	movl %eax,%cr3		/* set the page table pointer.. */
229 	movl $CR0_STATE,%eax
230 	movl %eax,%cr0		/* ..and set paging (PG) bit */
231 	ljmp $__BOOT_CS,$1f	/* Clear prefetch and normalize %eip */
232 1:
233 	/* Shift the stack pointer to a virtual address */
234 	addl $__PAGE_OFFSET, %esp
235 
236 /*
237  * Check if it is 486
238  */
239 	movb $4,X86			# at least 486
240 	cmpl $-1,X86_CPUID
241 	je .Lis486
242 
243 	/* get vendor info */
244 	xorl %eax,%eax			# call CPUID with 0 -> return vendor ID
245 	cpuid
246 	movl %eax,X86_CPUID		# save CPUID level
247 	movl %ebx,X86_VENDOR_ID		# lo 4 chars
248 	movl %edx,X86_VENDOR_ID+4	# next 4 chars
249 	movl %ecx,X86_VENDOR_ID+8	# last 4 chars
250 
251 	orl %eax,%eax			# do we have processor info as well?
252 	je .Lis486
253 
254 	movl $1,%eax		# Use the CPUID instruction to get CPU type
255 	cpuid
256 	movb %al,%cl		# save reg for future use
257 	andb $0x0f,%ah		# mask processor family
258 	movb %ah,X86
259 	andb $0xf0,%al		# mask model
260 	shrb $4,%al
261 	movb %al,X86_MODEL
262 	andb $0x0f,%cl		# mask mask revision
263 	movb %cl,X86_STEPPING
264 	movl %edx,X86_CAPABILITY
265 
266 .Lis486:
267 	movl $0x50022,%ecx	# set AM, WP, NE and MP
268 	movl %cr0,%eax
269 	andl $0x80000011,%eax	# Save PG,PE,ET
270 	orl %ecx,%eax
271 	movl %eax,%cr0
272 
273 	lgdt early_gdt_descr
274 	ljmp $(__KERNEL_CS),$1f
275 1:	movl $(__KERNEL_DS),%eax	# reload all the segment registers
276 	movl %eax,%ss			# after changing gdt.
277 
278 	movl $(__USER_DS),%eax		# DS/ES contains default USER segment
279 	movl %eax,%ds
280 	movl %eax,%es
281 
282 	movl $(__KERNEL_PERCPU), %eax
283 	movl %eax,%fs			# set this cpu's percpu
284 
285 	xorl %eax,%eax
286 	movl %eax,%gs			# clear possible garbage in %gs
287 
288 	xorl %eax,%eax			# Clear LDT
289 	lldt %ax
290 
291 	call *(initial_code)
292 1:	jmp 1b
293 SYM_FUNC_END(startup_32_smp)
294 
295 #include "verify_cpu.S"
296 
297 __INIT
298 SYM_FUNC_START(early_idt_handler_array)
299 	# 36(%esp) %eflags
300 	# 32(%esp) %cs
301 	# 28(%esp) %eip
302 	# 24(%rsp) error code
303 	i = 0
304 	.rept NUM_EXCEPTION_VECTORS
305 	.if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
306 	pushl $0		# Dummy error code, to make stack frame uniform
307 	.endif
308 	pushl $i		# 20(%esp) Vector number
309 	jmp early_idt_handler_common
310 	i = i + 1
311 	.fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
312 	.endr
313 SYM_FUNC_END(early_idt_handler_array)
314 
315 SYM_CODE_START_LOCAL(early_idt_handler_common)
316 	/*
317 	 * The stack is the hardware frame, an error code or zero, and the
318 	 * vector number.
319 	 */
320 	cld
321 
322 	incl %ss:early_recursion_flag
323 
324 	/* The vector number is in pt_regs->gs */
325 
326 	cld
327 	pushl	%fs		/* pt_regs->fs (__fsh varies by model) */
328 	pushl	%es		/* pt_regs->es (__esh varies by model) */
329 	pushl	%ds		/* pt_regs->ds (__dsh varies by model) */
330 	pushl	%eax		/* pt_regs->ax */
331 	pushl	%ebp		/* pt_regs->bp */
332 	pushl	%edi		/* pt_regs->di */
333 	pushl	%esi		/* pt_regs->si */
334 	pushl	%edx		/* pt_regs->dx */
335 	pushl	%ecx		/* pt_regs->cx */
336 	pushl	%ebx		/* pt_regs->bx */
337 
338 	/* Fix up DS and ES */
339 	movl	$(__KERNEL_DS), %ecx
340 	movl	%ecx, %ds
341 	movl	%ecx, %es
342 
343 	/* Load the vector number into EDX */
344 	movl	PT_GS(%esp), %edx
345 
346 	/* Load GS into pt_regs->gs (and maybe clobber __gsh) */
347 	movw	%gs, PT_GS(%esp)
348 
349 	movl	%esp, %eax	/* args are pt_regs (EAX), trapnr (EDX) */
350 	call	early_fixup_exception
351 
352 	popl	%ebx		/* pt_regs->bx */
353 	popl	%ecx		/* pt_regs->cx */
354 	popl	%edx		/* pt_regs->dx */
355 	popl	%esi		/* pt_regs->si */
356 	popl	%edi		/* pt_regs->di */
357 	popl	%ebp		/* pt_regs->bp */
358 	popl	%eax		/* pt_regs->ax */
359 	popl	%ds		/* pt_regs->ds (always ignores __dsh) */
360 	popl	%es		/* pt_regs->es (always ignores __esh) */
361 	popl	%fs		/* pt_regs->fs (always ignores __fsh) */
362 	popl	%gs		/* pt_regs->gs (always ignores __gsh) */
363 	decl	%ss:early_recursion_flag
364 	addl	$4, %esp	/* pop pt_regs->orig_ax */
365 	iret
366 SYM_CODE_END(early_idt_handler_common)
367 
368 /* This is the default interrupt "handler" :-) */
369 SYM_FUNC_START(early_ignore_irq)
370 	cld
371 #ifdef CONFIG_PRINTK
372 	pushl %eax
373 	pushl %ecx
374 	pushl %edx
375 	pushl %es
376 	pushl %ds
377 	movl $(__KERNEL_DS),%eax
378 	movl %eax,%ds
379 	movl %eax,%es
380 	cmpl $2,early_recursion_flag
381 	je hlt_loop
382 	incl early_recursion_flag
383 	pushl 16(%esp)
384 	pushl 24(%esp)
385 	pushl 32(%esp)
386 	pushl 40(%esp)
387 	pushl $int_msg
388 	call _printk
389 
390 	call dump_stack
391 
392 	addl $(5*4),%esp
393 	popl %ds
394 	popl %es
395 	popl %edx
396 	popl %ecx
397 	popl %eax
398 #endif
399 	iret
400 
401 hlt_loop:
402 	hlt
403 	jmp hlt_loop
404 SYM_FUNC_END(early_ignore_irq)
405 
406 __INITDATA
407 	.align 4
408 SYM_DATA(early_recursion_flag, .long 0)
409 
410 __REFDATA
411 	.align 4
412 SYM_DATA(initial_code,		.long i386_start_kernel)
413 
414 #ifdef CONFIG_MITIGATION_PAGE_TABLE_ISOLATION
415 #define	PGD_ALIGN	(2 * PAGE_SIZE)
416 #define PTI_USER_PGD_FILL	1024
417 #else
418 #define	PGD_ALIGN	(PAGE_SIZE)
419 #define PTI_USER_PGD_FILL	0
420 #endif
421 /*
422  * BSS section
423  */
424 __PAGE_ALIGNED_BSS
425 	.align PGD_ALIGN
426 #ifdef CONFIG_X86_PAE
427 .globl initial_pg_pmd
428 initial_pg_pmd:
429 	.fill 1024*KPMDS,4,0
430 #else
431 .globl initial_page_table
432 initial_page_table:
433 	.fill 1024,4,0
434 #endif
435 	.align PGD_ALIGN
436 initial_pg_fixmap:
437 	.fill 1024,4,0
438 .globl swapper_pg_dir
439 	.align PGD_ALIGN
440 swapper_pg_dir:
441 	.fill 1024,4,0
442 	.fill PTI_USER_PGD_FILL,4,0
443 .globl empty_zero_page
444 empty_zero_page:
445 	.fill 4096,1,0
446 EXPORT_SYMBOL(empty_zero_page)
447 
448 /*
449  * This starts the data section.
450  */
451 #ifdef CONFIG_X86_PAE
452 __PAGE_ALIGNED_DATA
453 	/* Page-aligned for the benefit of paravirt? */
454 	.align PGD_ALIGN
455 SYM_DATA_START(initial_page_table)
456 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
457 # if KPMDS == 3
458 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
459 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
460 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
461 # elif KPMDS == 2
462 	.long	0,0
463 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
464 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
465 # elif KPMDS == 1
466 	.long	0,0
467 	.long	0,0
468 	.long	pa(initial_pg_pmd+PGD_IDENT_ATTR),0
469 # else
470 #  error "Kernel PMDs should be 1, 2 or 3"
471 # endif
472 	.align PAGE_SIZE		/* needs to be page-sized too */
473 
474 #ifdef CONFIG_MITIGATION_PAGE_TABLE_ISOLATION
475 	/*
476 	 * PTI needs another page so sync_initial_pagetable() works correctly
477 	 * and does not scribble over the data which is placed behind the
478 	 * actual initial_page_table. See clone_pgd_range().
479 	 */
480 	.fill 1024, 4, 0
481 #endif
482 
483 SYM_DATA_END(initial_page_table)
484 #endif
485 
486 .data
487 .balign 4
488 SYM_DATA(initial_stack, .long __top_init_kernel_stack)
489 
490 __INITRODATA
491 int_msg:
492 	.asciz "Unknown interrupt or fault at: %p %p %p\n"
493 
494 #include "../xen/xen-head.S"
495 
496 /*
497  * The IDT and GDT 'descriptors' are a strange 48-bit object
498  * only used by the lidt and lgdt instructions. They are not
499  * like usual segment descriptors - they consist of a 16-bit
500  * segment size, and 32-bit linear address value:
501  */
502 
503 	.data
504 	ALIGN
505 # early boot GDT descriptor (must use 1:1 address mapping)
506 	.word 0				# 32 bit align gdt_desc.address
507 SYM_DATA_START_LOCAL(boot_gdt_descr)
508 	.word __BOOT_DS+7
509 	.long boot_gdt - __PAGE_OFFSET
510 SYM_DATA_END(boot_gdt_descr)
511 
512 # boot GDT descriptor (later on used by CPU#0):
513 	.word 0				# 32 bit align gdt_desc.address
514 SYM_DATA_START(early_gdt_descr)
515 	.word GDT_ENTRIES*8-1
516 	.long gdt_page			/* Overwritten for secondary CPUs */
517 SYM_DATA_END(early_gdt_descr)
518 
519 /*
520  * The boot_gdt must mirror the equivalent in setup.S and is
521  * used only for booting.
522  */
523 	.align L1_CACHE_BYTES
524 SYM_DATA_START(boot_gdt)
525 	.fill GDT_ENTRY_BOOT_CS,8,0
526 	.quad 0x00cf9a000000ffff	/* kernel 4GB code at 0x00000000 */
527 	.quad 0x00cf92000000ffff	/* kernel 4GB data at 0x00000000 */
528 SYM_DATA_END(boot_gdt)
529