Loading...
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
67ENTRY(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:
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 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 */
175
176#ifdef CONFIG_HOTPLUG_CPU
177/*
178 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
179 * up already except stack. We just set up stack here. Then call
180 * start_secondary().
181 */
182ENTRY(start_cpu0)
183 movl initial_stack, %ecx
184 movl %ecx, %esp
185 call *(initial_code)
1861: jmp 1b
187ENDPROC(start_cpu0)
188#endif
189
190/*
191 * Non-boot CPU entry point; entered from trampoline.S
192 * We can't lgdt here, because lgdt itself uses a data segment, but
193 * we know the trampoline has already loaded the boot_gdt for us.
194 *
195 * If cpu hotplug is not supported then this code can go in init section
196 * which will be freed later
197 */
198ENTRY(startup_32_smp)
199 cld
200 movl $(__BOOT_DS),%eax
201 movl %eax,%ds
202 movl %eax,%es
203 movl %eax,%fs
204 movl %eax,%gs
205 movl pa(initial_stack),%ecx
206 movl %eax,%ss
207 leal -__PAGE_OFFSET(%ecx),%esp
208
209#ifdef CONFIG_MICROCODE
210 /* Early load ucode on AP. */
211 call load_ucode_ap
212#endif
213
214.Ldefault_entry:
215 movl $(CR0_STATE & ~X86_CR0_PG),%eax
216 movl %eax,%cr0
217
218/*
219 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
220 * bits like NT set. This would confuse the debugger if this code is traced. So
221 * initialize them properly now before switching to protected mode. That means
222 * DF in particular (even though we have cleared it earlier after copying the
223 * command line) because GCC expects it.
224 */
225 pushl $0
226 popfl
227
228/*
229 * New page tables may be in 4Mbyte page mode and may be using the global pages.
230 *
231 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
232 * if and only if CPUID exists and has flags other than the FPU flag set.
233 */
234 movl $-1,pa(X86_CPUID) # preset CPUID level
235 movl $X86_EFLAGS_ID,%ecx
236 pushl %ecx
237 popfl # set EFLAGS=ID
238 pushfl
239 popl %eax # get EFLAGS
240 testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set?
241 jz .Lenable_paging # hw disallowed setting of ID bit
242 # which means no CPUID and no CR4
243
244 xorl %eax,%eax
245 cpuid
246 movl %eax,pa(X86_CPUID) # save largest std CPUID function
247
248 movl $1,%eax
249 cpuid
250 andl $~1,%edx # Ignore CPUID.FPU
251 jz .Lenable_paging # No flags or only CPUID.FPU = no CR4
252
253 movl pa(mmu_cr4_features),%eax
254 movl %eax,%cr4
255
256 testb $X86_CR4_PAE, %al # check if PAE is enabled
257 jz .Lenable_paging
258
259 /* Check if extended functions are implemented */
260 movl $0x80000000, %eax
261 cpuid
262 /* Value must be in the range 0x80000001 to 0x8000ffff */
263 subl $0x80000001, %eax
264 cmpl $(0x8000ffff-0x80000001), %eax
265 ja .Lenable_paging
266
267 /* Clear bogus XD_DISABLE bits */
268 call verify_cpu
269
270 mov $0x80000001, %eax
271 cpuid
272 /* Execute Disable bit supported? */
273 btl $(X86_FEATURE_NX & 31), %edx
274 jnc .Lenable_paging
275
276 /* Setup EFER (Extended Feature Enable Register) */
277 movl $MSR_EFER, %ecx
278 rdmsr
279
280 btsl $_EFER_NX, %eax
281 /* Make changes effective */
282 wrmsr
283
284.Lenable_paging:
285
286/*
287 * Enable paging
288 */
289 movl $pa(initial_page_table), %eax
290 movl %eax,%cr3 /* set the page table pointer.. */
291 movl $CR0_STATE,%eax
292 movl %eax,%cr0 /* ..and set paging (PG) bit */
293 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
2941:
295 /* Shift the stack pointer to a virtual address */
296 addl $__PAGE_OFFSET, %esp
297
298/*
299 * start system 32-bit setup. We need to re-do some of the things done
300 * in 16-bit mode for the "real" operations.
301 */
302 movl setup_once_ref,%eax
303 andl %eax,%eax
304 jz 1f # Did we do this already?
305 call *%eax
3061:
307
308/*
309 * Check if it is 486
310 */
311 movb $4,X86 # at least 486
312 cmpl $-1,X86_CPUID
313 je .Lis486
314
315 /* get vendor info */
316 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
317 cpuid
318 movl %eax,X86_CPUID # save CPUID level
319 movl %ebx,X86_VENDOR_ID # lo 4 chars
320 movl %edx,X86_VENDOR_ID+4 # next 4 chars
321 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
322
323 orl %eax,%eax # do we have processor info as well?
324 je .Lis486
325
326 movl $1,%eax # Use the CPUID instruction to get CPU type
327 cpuid
328 movb %al,%cl # save reg for future use
329 andb $0x0f,%ah # mask processor family
330 movb %ah,X86
331 andb $0xf0,%al # mask model
332 shrb $4,%al
333 movb %al,X86_MODEL
334 andb $0x0f,%cl # mask mask revision
335 movb %cl,X86_STEPPING
336 movl %edx,X86_CAPABILITY
337
338.Lis486:
339 movl $0x50022,%ecx # set AM, WP, NE and MP
340 movl %cr0,%eax
341 andl $0x80000011,%eax # Save PG,PE,ET
342 orl %ecx,%eax
343 movl %eax,%cr0
344
345 lgdt early_gdt_descr
346 ljmp $(__KERNEL_CS),$1f
3471: movl $(__KERNEL_DS),%eax # reload all the segment registers
348 movl %eax,%ss # after changing gdt.
349
350 movl $(__USER_DS),%eax # DS/ES contains default USER segment
351 movl %eax,%ds
352 movl %eax,%es
353
354 movl $(__KERNEL_PERCPU), %eax
355 movl %eax,%fs # set this cpu's percpu
356
357 movl $(__KERNEL_STACK_CANARY),%eax
358 movl %eax,%gs
359
360 xorl %eax,%eax # Clear LDT
361 lldt %ax
362
363 call *(initial_code)
3641: jmp 1b
365ENDPROC(startup_32_smp)
366
367#include "verify_cpu.S"
368
369/*
370 * setup_once
371 *
372 * The setup work we only want to run on the BSP.
373 *
374 * Warning: %esi is live across this function.
375 */
376__INIT
377setup_once:
378#ifdef CONFIG_CC_STACKPROTECTOR
379 /*
380 * Configure the stack canary. The linker can't handle this by
381 * relocation. Manually set base address in stack canary
382 * segment descriptor.
383 */
384 movl $gdt_page,%eax
385 movl $stack_canary,%ecx
386 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
387 shrl $16, %ecx
388 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
389 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
390#endif
391
392 andl $0,setup_once_ref /* Once is enough, thanks */
393 ret
394
395ENTRY(early_idt_handler_array)
396 # 36(%esp) %eflags
397 # 32(%esp) %cs
398 # 28(%esp) %eip
399 # 24(%rsp) error code
400 i = 0
401 .rept NUM_EXCEPTION_VECTORS
402 .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0
403 pushl $0 # Dummy error code, to make stack frame uniform
404 .endif
405 pushl $i # 20(%esp) Vector number
406 jmp early_idt_handler_common
407 i = i + 1
408 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
409 .endr
410ENDPROC(early_idt_handler_array)
411
412early_idt_handler_common:
413 /*
414 * The stack is the hardware frame, an error code or zero, and the
415 * vector number.
416 */
417 cld
418
419 incl %ss:early_recursion_flag
420
421 /* The vector number is in pt_regs->gs */
422
423 cld
424 pushl %fs /* pt_regs->fs (__fsh varies by model) */
425 pushl %es /* pt_regs->es (__esh varies by model) */
426 pushl %ds /* pt_regs->ds (__dsh varies by model) */
427 pushl %eax /* pt_regs->ax */
428 pushl %ebp /* pt_regs->bp */
429 pushl %edi /* pt_regs->di */
430 pushl %esi /* pt_regs->si */
431 pushl %edx /* pt_regs->dx */
432 pushl %ecx /* pt_regs->cx */
433 pushl %ebx /* pt_regs->bx */
434
435 /* Fix up DS and ES */
436 movl $(__KERNEL_DS), %ecx
437 movl %ecx, %ds
438 movl %ecx, %es
439
440 /* Load the vector number into EDX */
441 movl PT_GS(%esp), %edx
442
443 /* Load GS into pt_regs->gs (and maybe clobber __gsh) */
444 movw %gs, PT_GS(%esp)
445
446 movl %esp, %eax /* args are pt_regs (EAX), trapnr (EDX) */
447 call early_fixup_exception
448
449 popl %ebx /* pt_regs->bx */
450 popl %ecx /* pt_regs->cx */
451 popl %edx /* pt_regs->dx */
452 popl %esi /* pt_regs->si */
453 popl %edi /* pt_regs->di */
454 popl %ebp /* pt_regs->bp */
455 popl %eax /* pt_regs->ax */
456 popl %ds /* pt_regs->ds (always ignores __dsh) */
457 popl %es /* pt_regs->es (always ignores __esh) */
458 popl %fs /* pt_regs->fs (always ignores __fsh) */
459 popl %gs /* pt_regs->gs (always ignores __gsh) */
460 decl %ss:early_recursion_flag
461 addl $4, %esp /* pop pt_regs->orig_ax */
462 iret
463ENDPROC(early_idt_handler_common)
464
465/* This is the default interrupt "handler" :-) */
466ENTRY(early_ignore_irq)
467 cld
468#ifdef CONFIG_PRINTK
469 pushl %eax
470 pushl %ecx
471 pushl %edx
472 pushl %es
473 pushl %ds
474 movl $(__KERNEL_DS),%eax
475 movl %eax,%ds
476 movl %eax,%es
477 cmpl $2,early_recursion_flag
478 je hlt_loop
479 incl early_recursion_flag
480 pushl 16(%esp)
481 pushl 24(%esp)
482 pushl 32(%esp)
483 pushl 40(%esp)
484 pushl $int_msg
485 call printk
486
487 call dump_stack
488
489 addl $(5*4),%esp
490 popl %ds
491 popl %es
492 popl %edx
493 popl %ecx
494 popl %eax
495#endif
496 iret
497
498hlt_loop:
499 hlt
500 jmp hlt_loop
501ENDPROC(early_ignore_irq)
502
503__INITDATA
504 .align 4
505GLOBAL(early_recursion_flag)
506 .long 0
507
508__REFDATA
509 .align 4
510ENTRY(initial_code)
511 .long i386_start_kernel
512ENTRY(setup_once_ref)
513 .long setup_once
514
515/*
516 * BSS section
517 */
518__PAGE_ALIGNED_BSS
519 .align PAGE_SIZE
520#ifdef CONFIG_X86_PAE
521.globl initial_pg_pmd
522initial_pg_pmd:
523 .fill 1024*KPMDS,4,0
524#else
525.globl initial_page_table
526initial_page_table:
527 .fill 1024,4,0
528#endif
529initial_pg_fixmap:
530 .fill 1024,4,0
531.globl empty_zero_page
532empty_zero_page:
533 .fill 4096,1,0
534.globl swapper_pg_dir
535swapper_pg_dir:
536 .fill 1024,4,0
537EXPORT_SYMBOL(empty_zero_page)
538
539/*
540 * This starts the data section.
541 */
542#ifdef CONFIG_X86_PAE
543__PAGE_ALIGNED_DATA
544 /* Page-aligned for the benefit of paravirt? */
545 .align PAGE_SIZE
546ENTRY(initial_page_table)
547 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
548# if KPMDS == 3
549 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
550 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
551 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
552# elif KPMDS == 2
553 .long 0,0
554 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
555 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
556# elif KPMDS == 1
557 .long 0,0
558 .long 0,0
559 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
560# else
561# error "Kernel PMDs should be 1, 2 or 3"
562# endif
563 .align PAGE_SIZE /* needs to be page-sized too */
564#endif
565
566.data
567.balign 4
568ENTRY(initial_stack)
569 /*
570 * The SIZEOF_PTREGS gap is a convention which helps the in-kernel
571 * unwinder reliably detect the end of the stack.
572 */
573 .long init_thread_union + THREAD_SIZE - SIZEOF_PTREGS - \
574 TOP_OF_KERNEL_STACK_PADDING;
575
576__INITRODATA
577int_msg:
578 .asciz "Unknown interrupt or fault at: %p %p %p\n"
579
580#include "../../x86/xen/xen-head.S"
581
582/*
583 * The IDT and GDT 'descriptors' are a strange 48-bit object
584 * only used by the lidt and lgdt instructions. They are not
585 * like usual segment descriptors - they consist of a 16-bit
586 * segment size, and 32-bit linear address value:
587 */
588
589 .data
590.globl boot_gdt_descr
591
592 ALIGN
593# early boot GDT descriptor (must use 1:1 address mapping)
594 .word 0 # 32 bit align gdt_desc.address
595boot_gdt_descr:
596 .word __BOOT_DS+7
597 .long boot_gdt - __PAGE_OFFSET
598
599# boot GDT descriptor (later on used by CPU#0):
600 .word 0 # 32 bit align gdt_desc.address
601ENTRY(early_gdt_descr)
602 .word GDT_ENTRIES*8-1
603 .long gdt_page /* Overwritten for secondary CPUs */
604
605/*
606 * The boot_gdt must mirror the equivalent in setup.S and is
607 * used only for booting.
608 */
609 .align L1_CACHE_BYTES
610ENTRY(boot_gdt)
611 .fill GDT_ENTRY_BOOT_CS,8,0
612 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
613 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */
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/cpufeature.h>
23#include <asm/percpu.h>
24#include <asm/nops.h>
25
26/* Physical address */
27#define pa(X) ((X) - __PAGE_OFFSET)
28
29/*
30 * References to members of the new_cpu_data structure.
31 */
32
33#define X86 new_cpu_data+CPUINFO_x86
34#define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor
35#define X86_MODEL new_cpu_data+CPUINFO_x86_model
36#define X86_MASK new_cpu_data+CPUINFO_x86_mask
37#define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math
38#define X86_CPUID new_cpu_data+CPUINFO_cpuid_level
39#define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability
40#define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id
41
42/*
43 * This is how much memory in addition to the memory covered up to
44 * and including _end we need mapped initially.
45 * We need:
46 * (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
47 * (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
48 *
49 * Modulo rounding, each megabyte assigned here requires a kilobyte of
50 * memory, which is currently unreclaimed.
51 *
52 * This should be a multiple of a page.
53 *
54 * KERNEL_IMAGE_SIZE should be greater than pa(_end)
55 * and small than max_low_pfn, otherwise will waste some page table entries
56 */
57
58#if PTRS_PER_PMD > 1
59#define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
60#else
61#define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
62#endif
63
64/* Number of possible pages in the lowmem region */
65LOWMEM_PAGES = (((1<<32) - __PAGE_OFFSET) >> PAGE_SHIFT)
66
67/* Enough space to fit pagetables for the low memory linear map */
68MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
69
70/*
71 * Worst-case size of the kernel mapping we need to make:
72 * a relocatable kernel can live anywhere in lowmem, so we need to be able
73 * to map all of lowmem.
74 */
75KERNEL_PAGES = LOWMEM_PAGES
76
77INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
78RESERVE_BRK(pagetables, INIT_MAP_SIZE)
79
80/*
81 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
82 * %esi points to the real-mode code as a 32-bit pointer.
83 * CS and DS must be 4 GB flat segments, but we don't depend on
84 * any particular GDT layout, because we load our own as soon as we
85 * can.
86 */
87__HEAD
88ENTRY(startup_32)
89 movl pa(stack_start),%ecx
90
91 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
92 us to not reload segments */
93 testb $(1<<6), BP_loadflags(%esi)
94 jnz 2f
95
96/*
97 * Set segments to known values.
98 */
99 lgdt pa(boot_gdt_descr)
100 movl $(__BOOT_DS),%eax
101 movl %eax,%ds
102 movl %eax,%es
103 movl %eax,%fs
104 movl %eax,%gs
105 movl %eax,%ss
1062:
107 leal -__PAGE_OFFSET(%ecx),%esp
108
109/*
110 * Clear BSS first so that there are no surprises...
111 */
112 cld
113 xorl %eax,%eax
114 movl $pa(__bss_start),%edi
115 movl $pa(__bss_stop),%ecx
116 subl %edi,%ecx
117 shrl $2,%ecx
118 rep ; stosl
119/*
120 * Copy bootup parameters out of the way.
121 * Note: %esi still has the pointer to the real-mode data.
122 * With the kexec as boot loader, parameter segment might be loaded beyond
123 * kernel image and might not even be addressable by early boot page tables.
124 * (kexec on panic case). Hence copy out the parameters before initializing
125 * page tables.
126 */
127 movl $pa(boot_params),%edi
128 movl $(PARAM_SIZE/4),%ecx
129 cld
130 rep
131 movsl
132 movl pa(boot_params) + NEW_CL_POINTER,%esi
133 andl %esi,%esi
134 jz 1f # No command line
135 movl $pa(boot_command_line),%edi
136 movl $(COMMAND_LINE_SIZE/4),%ecx
137 rep
138 movsl
1391:
140
141#ifdef CONFIG_OLPC
142 /* save OFW's pgdir table for later use when calling into OFW */
143 movl %cr3, %eax
144 movl %eax, pa(olpc_ofw_pgd)
145#endif
146
147#ifdef CONFIG_MICROCODE_EARLY
148 /* Early load ucode on BSP. */
149 call load_ucode_bsp
150#endif
151
152/*
153 * Initialize page tables. This creates a PDE and a set of page
154 * tables, which are located immediately beyond __brk_base. The variable
155 * _brk_end is set up to point to the first "safe" location.
156 * Mappings are created both at virtual address 0 (identity mapping)
157 * and PAGE_OFFSET for up to _end.
158 */
159#ifdef CONFIG_X86_PAE
160
161 /*
162 * In PAE mode initial_page_table is statically defined to contain
163 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
164 * entries). The identity mapping is handled by pointing two PGD entries
165 * to the first kernel PMD.
166 *
167 * Note the upper half of each PMD or PTE are always zero at this stage.
168 */
169
170#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
171
172 xorl %ebx,%ebx /* %ebx is kept at zero */
173
174 movl $pa(__brk_base), %edi
175 movl $pa(initial_pg_pmd), %edx
176 movl $PTE_IDENT_ATTR, %eax
17710:
178 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
179 movl %ecx,(%edx) /* Store PMD entry */
180 /* Upper half already zero */
181 addl $8,%edx
182 movl $512,%ecx
18311:
184 stosl
185 xchgl %eax,%ebx
186 stosl
187 xchgl %eax,%ebx
188 addl $0x1000,%eax
189 loop 11b
190
191 /*
192 * End condition: we must map up to the end + MAPPING_BEYOND_END.
193 */
194 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
195 cmpl %ebp,%eax
196 jb 10b
1971:
198 addl $__PAGE_OFFSET, %edi
199 movl %edi, pa(_brk_end)
200 shrl $12, %eax
201 movl %eax, pa(max_pfn_mapped)
202
203 /* Do early initialization of the fixmap area */
204 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
205 movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
206#else /* Not PAE */
207
208page_pde_offset = (__PAGE_OFFSET >> 20);
209
210 movl $pa(__brk_base), %edi
211 movl $pa(initial_page_table), %edx
212 movl $PTE_IDENT_ATTR, %eax
21310:
214 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
215 movl %ecx,(%edx) /* Store identity PDE entry */
216 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
217 addl $4,%edx
218 movl $1024, %ecx
21911:
220 stosl
221 addl $0x1000,%eax
222 loop 11b
223 /*
224 * End condition: we must map up to the end + MAPPING_BEYOND_END.
225 */
226 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
227 cmpl %ebp,%eax
228 jb 10b
229 addl $__PAGE_OFFSET, %edi
230 movl %edi, pa(_brk_end)
231 shrl $12, %eax
232 movl %eax, pa(max_pfn_mapped)
233
234 /* Do early initialization of the fixmap area */
235 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
236 movl %eax,pa(initial_page_table+0xffc)
237#endif
238
239#ifdef CONFIG_PARAVIRT
240 /* This is can only trip for a broken bootloader... */
241 cmpw $0x207, pa(boot_params + BP_version)
242 jb default_entry
243
244 /* Paravirt-compatible boot parameters. Look to see what architecture
245 we're booting under. */
246 movl pa(boot_params + BP_hardware_subarch), %eax
247 cmpl $num_subarch_entries, %eax
248 jae bad_subarch
249
250 movl pa(subarch_entries)(,%eax,4), %eax
251 subl $__PAGE_OFFSET, %eax
252 jmp *%eax
253
254bad_subarch:
255WEAK(lguest_entry)
256WEAK(xen_entry)
257 /* Unknown implementation; there's really
258 nothing we can do at this point. */
259 ud2a
260
261 __INITDATA
262
263subarch_entries:
264 .long default_entry /* normal x86/PC */
265 .long lguest_entry /* lguest hypervisor */
266 .long xen_entry /* Xen hypervisor */
267 .long default_entry /* Moorestown MID */
268num_subarch_entries = (. - subarch_entries) / 4
269.previous
270#else
271 jmp default_entry
272#endif /* CONFIG_PARAVIRT */
273
274#ifdef CONFIG_HOTPLUG_CPU
275/*
276 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
277 * up already except stack. We just set up stack here. Then call
278 * start_secondary().
279 */
280ENTRY(start_cpu0)
281 movl stack_start, %ecx
282 movl %ecx, %esp
283 jmp *(initial_code)
284ENDPROC(start_cpu0)
285#endif
286
287/*
288 * Non-boot CPU entry point; entered from trampoline.S
289 * We can't lgdt here, because lgdt itself uses a data segment, but
290 * we know the trampoline has already loaded the boot_gdt for us.
291 *
292 * If cpu hotplug is not supported then this code can go in init section
293 * which will be freed later
294 */
295ENTRY(startup_32_smp)
296 cld
297 movl $(__BOOT_DS),%eax
298 movl %eax,%ds
299 movl %eax,%es
300 movl %eax,%fs
301 movl %eax,%gs
302 movl pa(stack_start),%ecx
303 movl %eax,%ss
304 leal -__PAGE_OFFSET(%ecx),%esp
305
306#ifdef CONFIG_MICROCODE_EARLY
307 /* Early load ucode on AP. */
308 call load_ucode_ap
309#endif
310
311
312default_entry:
313#define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
314 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
315 X86_CR0_PG)
316 movl $(CR0_STATE & ~X86_CR0_PG),%eax
317 movl %eax,%cr0
318
319/*
320 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
321 * bits like NT set. This would confuse the debugger if this code is traced. So
322 * initialize them properly now before switching to protected mode. That means
323 * DF in particular (even though we have cleared it earlier after copying the
324 * command line) because GCC expects it.
325 */
326 pushl $0
327 popfl
328
329/*
330 * New page tables may be in 4Mbyte page mode and may be using the global pages.
331 *
332 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
333 * if and only if CPUID exists and has flags other than the FPU flag set.
334 */
335 movl $-1,pa(X86_CPUID) # preset CPUID level
336 movl $X86_EFLAGS_ID,%ecx
337 pushl %ecx
338 popfl # set EFLAGS=ID
339 pushfl
340 popl %eax # get EFLAGS
341 testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set?
342 jz enable_paging # hw disallowed setting of ID bit
343 # which means no CPUID and no CR4
344
345 xorl %eax,%eax
346 cpuid
347 movl %eax,pa(X86_CPUID) # save largest std CPUID function
348
349 movl $1,%eax
350 cpuid
351 andl $~1,%edx # Ignore CPUID.FPU
352 jz enable_paging # No flags or only CPUID.FPU = no CR4
353
354 movl pa(mmu_cr4_features),%eax
355 movl %eax,%cr4
356
357 testb $X86_CR4_PAE, %al # check if PAE is enabled
358 jz enable_paging
359
360 /* Check if extended functions are implemented */
361 movl $0x80000000, %eax
362 cpuid
363 /* Value must be in the range 0x80000001 to 0x8000ffff */
364 subl $0x80000001, %eax
365 cmpl $(0x8000ffff-0x80000001), %eax
366 ja enable_paging
367
368 /* Clear bogus XD_DISABLE bits */
369 call verify_cpu
370
371 mov $0x80000001, %eax
372 cpuid
373 /* Execute Disable bit supported? */
374 btl $(X86_FEATURE_NX & 31), %edx
375 jnc enable_paging
376
377 /* Setup EFER (Extended Feature Enable Register) */
378 movl $MSR_EFER, %ecx
379 rdmsr
380
381 btsl $_EFER_NX, %eax
382 /* Make changes effective */
383 wrmsr
384
385enable_paging:
386
387/*
388 * Enable paging
389 */
390 movl $pa(initial_page_table), %eax
391 movl %eax,%cr3 /* set the page table pointer.. */
392 movl $CR0_STATE,%eax
393 movl %eax,%cr0 /* ..and set paging (PG) bit */
394 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
3951:
396 /* Shift the stack pointer to a virtual address */
397 addl $__PAGE_OFFSET, %esp
398
399/*
400 * start system 32-bit setup. We need to re-do some of the things done
401 * in 16-bit mode for the "real" operations.
402 */
403 movl setup_once_ref,%eax
404 andl %eax,%eax
405 jz 1f # Did we do this already?
406 call *%eax
4071:
408
409/*
410 * Check if it is 486
411 */
412 movb $4,X86 # at least 486
413 cmpl $-1,X86_CPUID
414 je is486
415
416 /* get vendor info */
417 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
418 cpuid
419 movl %eax,X86_CPUID # save CPUID level
420 movl %ebx,X86_VENDOR_ID # lo 4 chars
421 movl %edx,X86_VENDOR_ID+4 # next 4 chars
422 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
423
424 orl %eax,%eax # do we have processor info as well?
425 je is486
426
427 movl $1,%eax # Use the CPUID instruction to get CPU type
428 cpuid
429 movb %al,%cl # save reg for future use
430 andb $0x0f,%ah # mask processor family
431 movb %ah,X86
432 andb $0xf0,%al # mask model
433 shrb $4,%al
434 movb %al,X86_MODEL
435 andb $0x0f,%cl # mask mask revision
436 movb %cl,X86_MASK
437 movl %edx,X86_CAPABILITY
438
439is486:
440 movl $0x50022,%ecx # set AM, WP, NE and MP
441 movl %cr0,%eax
442 andl $0x80000011,%eax # Save PG,PE,ET
443 orl %ecx,%eax
444 movl %eax,%cr0
445
446 lgdt early_gdt_descr
447 lidt idt_descr
448 ljmp $(__KERNEL_CS),$1f
4491: movl $(__KERNEL_DS),%eax # reload all the segment registers
450 movl %eax,%ss # after changing gdt.
451
452 movl $(__USER_DS),%eax # DS/ES contains default USER segment
453 movl %eax,%ds
454 movl %eax,%es
455
456 movl $(__KERNEL_PERCPU), %eax
457 movl %eax,%fs # set this cpu's percpu
458
459 movl $(__KERNEL_STACK_CANARY),%eax
460 movl %eax,%gs
461
462 xorl %eax,%eax # Clear LDT
463 lldt %ax
464
465 pushl $0 # fake return address for unwinder
466 jmp *(initial_code)
467
468#include "verify_cpu.S"
469
470/*
471 * setup_once
472 *
473 * The setup work we only want to run on the BSP.
474 *
475 * Warning: %esi is live across this function.
476 */
477__INIT
478setup_once:
479 /*
480 * Set up a idt with 256 entries pointing to ignore_int,
481 * interrupt gates. It doesn't actually load idt - that needs
482 * to be done on each CPU. Interrupts are enabled elsewhere,
483 * when we can be relatively sure everything is ok.
484 */
485
486 movl $idt_table,%edi
487 movl $early_idt_handlers,%eax
488 movl $NUM_EXCEPTION_VECTORS,%ecx
4891:
490 movl %eax,(%edi)
491 movl %eax,4(%edi)
492 /* interrupt gate, dpl=0, present */
493 movl $(0x8E000000 + __KERNEL_CS),2(%edi)
494 addl $9,%eax
495 addl $8,%edi
496 loop 1b
497
498 movl $256 - NUM_EXCEPTION_VECTORS,%ecx
499 movl $ignore_int,%edx
500 movl $(__KERNEL_CS << 16),%eax
501 movw %dx,%ax /* selector = 0x0010 = cs */
502 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
5032:
504 movl %eax,(%edi)
505 movl %edx,4(%edi)
506 addl $8,%edi
507 loop 2b
508
509#ifdef CONFIG_CC_STACKPROTECTOR
510 /*
511 * Configure the stack canary. The linker can't handle this by
512 * relocation. Manually set base address in stack canary
513 * segment descriptor.
514 */
515 movl $gdt_page,%eax
516 movl $stack_canary,%ecx
517 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
518 shrl $16, %ecx
519 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
520 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
521#endif
522
523 andl $0,setup_once_ref /* Once is enough, thanks */
524 ret
525
526ENTRY(early_idt_handlers)
527 # 36(%esp) %eflags
528 # 32(%esp) %cs
529 # 28(%esp) %eip
530 # 24(%rsp) error code
531 i = 0
532 .rept NUM_EXCEPTION_VECTORS
533 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
534 ASM_NOP2
535 .else
536 pushl $0 # Dummy error code, to make stack frame uniform
537 .endif
538 pushl $i # 20(%esp) Vector number
539 jmp early_idt_handler
540 i = i + 1
541 .endr
542ENDPROC(early_idt_handlers)
543
544 /* This is global to keep gas from relaxing the jumps */
545ENTRY(early_idt_handler)
546 cld
547
548 cmpl $2,(%esp) # X86_TRAP_NMI
549 je is_nmi # Ignore NMI
550
551 cmpl $2,%ss:early_recursion_flag
552 je hlt_loop
553 incl %ss:early_recursion_flag
554
555 push %eax # 16(%esp)
556 push %ecx # 12(%esp)
557 push %edx # 8(%esp)
558 push %ds # 4(%esp)
559 push %es # 0(%esp)
560 movl $(__KERNEL_DS),%eax
561 movl %eax,%ds
562 movl %eax,%es
563
564 cmpl $(__KERNEL_CS),32(%esp)
565 jne 10f
566
567 leal 28(%esp),%eax # Pointer to %eip
568 call early_fixup_exception
569 andl %eax,%eax
570 jnz ex_entry /* found an exception entry */
571
57210:
573#ifdef CONFIG_PRINTK
574 xorl %eax,%eax
575 movw %ax,2(%esp) /* clean up the segment values on some cpus */
576 movw %ax,6(%esp)
577 movw %ax,34(%esp)
578 leal 40(%esp),%eax
579 pushl %eax /* %esp before the exception */
580 pushl %ebx
581 pushl %ebp
582 pushl %esi
583 pushl %edi
584 movl %cr2,%eax
585 pushl %eax
586 pushl (20+6*4)(%esp) /* trapno */
587 pushl $fault_msg
588 call printk
589#endif
590 call dump_stack
591hlt_loop:
592 hlt
593 jmp hlt_loop
594
595ex_entry:
596 pop %es
597 pop %ds
598 pop %edx
599 pop %ecx
600 pop %eax
601 decl %ss:early_recursion_flag
602is_nmi:
603 addl $8,%esp /* drop vector number and error code */
604 iret
605ENDPROC(early_idt_handler)
606
607/* This is the default interrupt "handler" :-) */
608 ALIGN
609ignore_int:
610 cld
611#ifdef CONFIG_PRINTK
612 pushl %eax
613 pushl %ecx
614 pushl %edx
615 pushl %es
616 pushl %ds
617 movl $(__KERNEL_DS),%eax
618 movl %eax,%ds
619 movl %eax,%es
620 cmpl $2,early_recursion_flag
621 je hlt_loop
622 incl early_recursion_flag
623 pushl 16(%esp)
624 pushl 24(%esp)
625 pushl 32(%esp)
626 pushl 40(%esp)
627 pushl $int_msg
628 call printk
629
630 call dump_stack
631
632 addl $(5*4),%esp
633 popl %ds
634 popl %es
635 popl %edx
636 popl %ecx
637 popl %eax
638#endif
639 iret
640ENDPROC(ignore_int)
641__INITDATA
642 .align 4
643early_recursion_flag:
644 .long 0
645
646__REFDATA
647 .align 4
648ENTRY(initial_code)
649 .long i386_start_kernel
650ENTRY(setup_once_ref)
651 .long setup_once
652
653/*
654 * BSS section
655 */
656__PAGE_ALIGNED_BSS
657 .align PAGE_SIZE
658#ifdef CONFIG_X86_PAE
659initial_pg_pmd:
660 .fill 1024*KPMDS,4,0
661#else
662ENTRY(initial_page_table)
663 .fill 1024,4,0
664#endif
665initial_pg_fixmap:
666 .fill 1024,4,0
667ENTRY(empty_zero_page)
668 .fill 4096,1,0
669ENTRY(swapper_pg_dir)
670 .fill 1024,4,0
671
672/*
673 * This starts the data section.
674 */
675#ifdef CONFIG_X86_PAE
676__PAGE_ALIGNED_DATA
677 /* Page-aligned for the benefit of paravirt? */
678 .align PAGE_SIZE
679ENTRY(initial_page_table)
680 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
681# if KPMDS == 3
682 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
683 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
684 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
685# elif KPMDS == 2
686 .long 0,0
687 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
688 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
689# elif KPMDS == 1
690 .long 0,0
691 .long 0,0
692 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
693# else
694# error "Kernel PMDs should be 1, 2 or 3"
695# endif
696 .align PAGE_SIZE /* needs to be page-sized too */
697#endif
698
699.data
700.balign 4
701ENTRY(stack_start)
702 .long init_thread_union+THREAD_SIZE
703
704__INITRODATA
705int_msg:
706 .asciz "Unknown interrupt or fault at: %p %p %p\n"
707
708fault_msg:
709/* fault info: */
710 .ascii "BUG: Int %d: CR2 %p\n"
711/* regs pushed in early_idt_handler: */
712 .ascii " EDI %p ESI %p EBP %p EBX %p\n"
713 .ascii " ESP %p ES %p DS %p\n"
714 .ascii " EDX %p ECX %p EAX %p\n"
715/* fault frame: */
716 .ascii " vec %p err %p EIP %p CS %p flg %p\n"
717 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
718 .ascii " %p %p %p %p %p %p %p %p\n"
719 .asciz " %p %p %p %p %p %p %p %p\n"
720
721#include "../../x86/xen/xen-head.S"
722
723/*
724 * The IDT and GDT 'descriptors' are a strange 48-bit object
725 * only used by the lidt and lgdt instructions. They are not
726 * like usual segment descriptors - they consist of a 16-bit
727 * segment size, and 32-bit linear address value:
728 */
729
730 .data
731.globl boot_gdt_descr
732.globl idt_descr
733
734 ALIGN
735# early boot GDT descriptor (must use 1:1 address mapping)
736 .word 0 # 32 bit align gdt_desc.address
737boot_gdt_descr:
738 .word __BOOT_DS+7
739 .long boot_gdt - __PAGE_OFFSET
740
741 .word 0 # 32-bit align idt_desc.address
742idt_descr:
743 .word IDT_ENTRIES*8-1 # idt contains 256 entries
744 .long idt_table
745
746# boot GDT descriptor (later on used by CPU#0):
747 .word 0 # 32 bit align gdt_desc.address
748ENTRY(early_gdt_descr)
749 .word GDT_ENTRIES*8-1
750 .long gdt_page /* Overwritten for secondary CPUs */
751
752/*
753 * The boot_gdt must mirror the equivalent in setup.S and is
754 * used only for booting.
755 */
756 .align L1_CACHE_BYTES
757ENTRY(boot_gdt)
758 .fill GDT_ENTRY_BOOT_CS,8,0
759 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
760 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */