3 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Enhanced CPU detection and feature setting code by Mike Jagdis
6 * and Martin Mares, November 1997.
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>
25 /* Physical address */
26 #define pa(X) ((X) - __PAGE_OFFSET)
29 * References to members of the new_cpu_data structure.
32 #define X86 new_cpu_data+CPUINFO_x86
33 #define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor
34 #define X86_MODEL new_cpu_data+CPUINFO_x86_model
35 #define X86_MASK new_cpu_data+CPUINFO_x86_mask
36 #define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math
37 #define X86_CPUID new_cpu_data+CPUINFO_cpuid_level
38 #define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability
39 #define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id
42 * This is how much memory in addition to the memory covered up to
43 * and including _end we need mapped initially.
45 * (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
46 * (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
48 * Modulo rounding, each megabyte assigned here requires a kilobyte of
49 * memory, which is currently unreclaimed.
51 * This should be a multiple of a page.
53 * KERNEL_IMAGE_SIZE should be greater than pa(_end)
54 * and small than max_low_pfn, otherwise will waste some page table entries
58 #define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
60 #define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
63 /* Number of possible pages in the lowmem region */
64 LOWMEM_PAGES = (((1<<32) - __PAGE_OFFSET) >> PAGE_SHIFT)
66 /* Enough space to fit pagetables for the low memory linear map */
67 MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
70 * Worst-case size of the kernel mapping we need to make:
71 * a relocatable kernel can live anywhere in lowmem, so we need to be able
72 * to map all of lowmem.
74 KERNEL_PAGES = LOWMEM_PAGES
76 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE_asm
77 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
80 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
81 * %esi points to the real-mode code as a 32-bit pointer.
82 * CS and DS must be 4 GB flat segments, but we don't depend on
83 * any particular GDT layout, because we load our own as soon as we
88 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
89 us to not reload segments */
90 testb $(1<<6), BP_loadflags(%esi)
94 * Set segments to known values.
96 lgdt pa(boot_gdt_descr)
97 movl $(__BOOT_DS),%eax
105 * Clear BSS first so that there are no surprises...
109 movl $pa(__bss_start),%edi
110 movl $pa(__bss_stop),%ecx
115 * Copy bootup parameters out of the way.
116 * Note: %esi still has the pointer to the real-mode data.
117 * With the kexec as boot loader, parameter segment might be loaded beyond
118 * kernel image and might not even be addressable by early boot page tables.
119 * (kexec on panic case). Hence copy out the parameters before initializing
122 movl $pa(boot_params),%edi
123 movl $(PARAM_SIZE/4),%ecx
127 movl pa(boot_params) + NEW_CL_POINTER,%esi
129 jz 1f # No comand line
130 movl $pa(boot_command_line),%edi
131 movl $(COMMAND_LINE_SIZE/4),%ecx
136 #ifdef CONFIG_OLPC_OPENFIRMWARE
137 /* save OFW's pgdir table for later use when calling into OFW */
139 movl %eax, pa(olpc_ofw_pgd)
142 #ifdef CONFIG_PARAVIRT
143 /* This is can only trip for a broken bootloader... */
144 cmpw $0x207, pa(boot_params + BP_version)
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
153 movl pa(subarch_entries)(,%eax,4), %eax
154 subl $__PAGE_OFFSET, %eax
160 /* Unknown implementation; there's really
161 nothing we can do at this point. */
167 .long default_entry /* normal x86/PC */
168 .long lguest_entry /* lguest hypervisor */
169 .long xen_entry /* Xen hypervisor */
170 .long default_entry /* Moorestown MID */
171 num_subarch_entries = (. - subarch_entries) / 4
173 #endif /* CONFIG_PARAVIRT */
176 * Initialize page tables. This creates a PDE and a set of page
177 * tables, which are located immediately beyond __brk_base. The variable
178 * _brk_end is set up to point to the first "safe" location.
179 * Mappings are created both at virtual address 0 (identity mapping)
180 * and PAGE_OFFSET for up to _end.
182 * Note that the stack is not yet set up!
185 #ifdef CONFIG_X86_PAE
188 * In PAE mode initial_page_table is statically defined to contain
189 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
190 * entries). The identity mapping is handled by pointing two PGD entries
191 * to the first kernel PMD.
193 * Note the upper half of each PMD or PTE are always zero at this stage.
196 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
198 xorl %ebx,%ebx /* %ebx is kept at zero */
200 movl $pa(__brk_base), %edi
201 movl $pa(initial_pg_pmd), %edx
202 movl $PTE_IDENT_ATTR, %eax
204 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
205 movl %ecx,(%edx) /* Store PMD entry */
206 /* Upper half already zero */
218 * End condition: we must map up to the end + MAPPING_BEYOND_END.
220 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
224 addl $__PAGE_OFFSET, %edi
225 movl %edi, pa(_brk_end)
227 movl %eax, pa(max_pfn_mapped)
229 /* Do early initialization of the fixmap area */
230 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
231 movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
234 page_pde_offset = (__PAGE_OFFSET >> 20);
236 movl $pa(__brk_base), %edi
237 movl $pa(initial_page_table), %edx
238 movl $PTE_IDENT_ATTR, %eax
240 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
241 movl %ecx,(%edx) /* Store identity PDE entry */
242 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
250 * End condition: we must map up to the end + MAPPING_BEYOND_END.
252 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
255 addl $__PAGE_OFFSET, %edi
256 movl %edi, pa(_brk_end)
258 movl %eax, pa(max_pfn_mapped)
260 /* Do early initialization of the fixmap area */
261 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
262 movl %eax,pa(initial_page_table+0xffc)
266 * Non-boot CPU entry point; entered from trampoline.S
267 * We can't lgdt here, because lgdt itself uses a data segment, but
268 * we know the trampoline has already loaded the boot_gdt for us.
270 * If cpu hotplug is not supported then this code can go in init section
271 * which will be freed later
277 ENTRY(startup_32_smp)
279 movl $(__BOOT_DS),%eax
284 #endif /* CONFIG_SMP */
288 * New page tables may be in 4Mbyte page mode and may
289 * be using the global pages.
291 * NOTE! If we are on a 486 we may have no cr4 at all!
292 * So we do not try to touch it unless we really have
293 * some bits in it to set. This won't work if the BSP
294 * implements cr4 but this AP does not -- very unlikely
295 * but be warned! The same applies to the pse feature
296 * if not equally supported. --macro
298 * NOTE! We have to correct for the fact that we're
299 * not yet offset PAGE_OFFSET..
301 #define cr4_bits pa(mmu_cr4_features)
305 movl %cr4,%eax # Turn on paging options (PSE,PAE,..)
309 testb $X86_CR4_PAE, %al # check if PAE is enabled
312 /* Check if extended functions are implemented */
313 movl $0x80000000, %eax
315 /* Value must be in the range 0x80000001 to 0x8000ffff */
316 subl $0x80000001, %eax
317 cmpl $(0x8000ffff-0x80000001), %eax
319 mov $0x80000001, %eax
321 /* Execute Disable bit supported? */
322 btl $(X86_FEATURE_NX & 31), %edx
325 /* Setup EFER (Extended Feature Enable Register) */
330 /* Make changes effective */
338 movl $pa(initial_page_table), %eax
339 movl %eax,%cr3 /* set the page table pointer.. */
342 movl %eax,%cr0 /* ..and set paging (PG) bit */
343 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
345 /* Set up the stack pointer */
349 * Initialize eflags. Some BIOS's leave bits like NT set. This would
350 * confuse the debugger if this code is traced.
351 * XXX - best to initialize before switching to protected mode.
358 jz 1f /* Initial CPU cleans BSS */
361 #endif /* CONFIG_SMP */
364 * start system 32-bit setup. We need to re-do some of the things done
365 * in 16-bit mode for the "real" operations.
371 movl $-1,X86_CPUID # -1 for no CPUID initially
373 /* check if it is 486 or 386. */
375 * XXX - this does a lot of unnecessary setup. Alignment checks don't
376 * apply at our cpl of 0 and the stack ought to be aligned already, and
377 * we don't need to preserve eflags.
380 movb $3,X86 # at least 386
382 popl %eax # get EFLAGS
383 movl %eax,%ecx # save original EFLAGS
384 xorl $0x240000,%eax # flip AC and ID bits in EFLAGS
385 pushl %eax # copy to EFLAGS
387 pushfl # get new EFLAGS
388 popl %eax # put it in eax
389 xorl %ecx,%eax # change in flags
390 pushl %ecx # restore original EFLAGS
392 testl $0x40000,%eax # check if AC bit changed
395 movb $4,X86 # at least 486
396 testl $0x200000,%eax # check if ID bit changed
399 /* get vendor info */
400 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
402 movl %eax,X86_CPUID # save CPUID level
403 movl %ebx,X86_VENDOR_ID # lo 4 chars
404 movl %edx,X86_VENDOR_ID+4 # next 4 chars
405 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
407 orl %eax,%eax # do we have processor info as well?
410 movl $1,%eax # Use the CPUID instruction to get CPU type
412 movb %al,%cl # save reg for future use
413 andb $0x0f,%ah # mask processor family
415 andb $0xf0,%al # mask model
418 andb $0x0f,%cl # mask mask revision
420 movl %edx,X86_CAPABILITY
422 is486: movl $0x50022,%ecx # set AM, WP, NE and MP
425 is386: movl $2,%ecx # set MP
427 andl $0x80000011,%eax # Save PG,PE,ET
434 ljmp $(__KERNEL_CS),$1f
435 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
436 movl %eax,%ss # after changing gdt.
438 movl $(__USER_DS),%eax # DS/ES contains default USER segment
442 movl $(__KERNEL_PERCPU), %eax
443 movl %eax,%fs # set this cpu's percpu
445 #ifdef CONFIG_CC_STACKPROTECTOR
447 * The linker can't handle this by relocation. Manually set
448 * base address in stack canary segment descriptor.
453 movl $stack_canary,%ecx
454 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
456 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
457 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
460 movl $(__KERNEL_STACK_CANARY),%eax
463 xorl %eax,%eax # Clear LDT
466 cld # gcc2 wants the direction flag cleared at all times
467 pushl $0 # fake return address for unwinder
471 cmpb $0,%cl # the first CPU calls start_kernel
473 movl (stack_start), %esp
475 #endif /* CONFIG_SMP */
479 * We depend on ET to be correct. This checks for 287/387.
482 movb $0,X86_HARD_MATH
488 movl %cr0,%eax /* no coprocessor: have to set bits */
489 xorl $4,%eax /* set EM */
493 1: movb $1,X86_HARD_MATH
494 .byte 0xDB,0xE4 /* fsetpm for 287, ignored by 387 */
500 * sets up a idt with 256 entries pointing to
501 * ignore_int, interrupt gates. It doesn't actually load
502 * idt - that can be done only after paging has been enabled
503 * and the kernel moved to PAGE_OFFSET. Interrupts
504 * are enabled elsewhere, when we can be relatively
505 * sure everything is ok.
507 * Warning: %esi is live across this function.
511 movl $(__KERNEL_CS << 16),%eax
512 movw %dx,%ax /* selector = 0x0010 = cs */
513 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
524 .macro set_early_handler handler,trapno
526 movl $(__KERNEL_CS << 16),%eax
528 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
530 movl %eax,8*\trapno(%edi)
531 movl %edx,8*\trapno+4(%edi)
534 set_early_handler handler=early_divide_err,trapno=0
535 set_early_handler handler=early_illegal_opcode,trapno=6
536 set_early_handler handler=early_protection_fault,trapno=13
537 set_early_handler handler=early_page_fault,trapno=14
543 pushl $0 /* fake errcode */
546 early_illegal_opcode:
548 pushl $0 /* fake errcode */
551 early_protection_fault:
563 movl $(__KERNEL_DS),%eax
566 cmpl $2,early_recursion_flag
568 incl early_recursion_flag
571 pushl %edx /* trapno */
580 /* This is the default interrupt "handler" :-) */
590 movl $(__KERNEL_DS),%eax
593 cmpl $2,early_recursion_flag
595 incl early_recursion_flag
617 .long i386_start_kernel
624 #ifdef CONFIG_X86_PAE
625 ENTRY(initial_pg_pmd)
628 ENTRY(initial_page_table)
631 ENTRY(initial_pg_fixmap)
633 ENTRY(empty_zero_page)
635 ENTRY(swapper_pg_dir)
639 * This starts the data section.
641 #ifdef CONFIG_X86_PAE
643 /* Page-aligned for the benefit of paravirt? */
645 ENTRY(initial_page_table)
646 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
648 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
649 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
650 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
653 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
654 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
658 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
660 # error "Kernel PMDs should be 1, 2 or 3"
662 .align PAGE_SIZE_asm /* needs to be page-sized too */
667 .long init_thread_union+THREAD_SIZE
672 early_recursion_flag:
676 .asciz "Unknown interrupt or fault at: %p %p %p\n"
680 .ascii "BUG: Int %d: CR2 %p\n"
682 .ascii " EDI %p ESI %p EBP %p ESP %p\n"
683 .ascii " EBX %p EDX %p ECX %p EAX %p\n"
685 .ascii " err %p EIP %p CS %p flg %p\n"
686 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
687 .ascii " %p %p %p %p %p %p %p %p\n"
688 .asciz " %p %p %p %p %p %p %p %p\n"
690 #include "../../x86/xen/xen-head.S"
693 * The IDT and GDT 'descriptors' are a strange 48-bit object
694 * only used by the lidt and lgdt instructions. They are not
695 * like usual segment descriptors - they consist of a 16-bit
696 * segment size, and 32-bit linear address value:
699 .globl boot_gdt_descr
703 # early boot GDT descriptor (must use 1:1 address mapping)
704 .word 0 # 32 bit align gdt_desc.address
707 .long boot_gdt - __PAGE_OFFSET
709 .word 0 # 32-bit align idt_desc.address
711 .word IDT_ENTRIES*8-1 # idt contains 256 entries
714 # boot GDT descriptor (later on used by CPU#0):
715 .word 0 # 32 bit align gdt_desc.address
716 ENTRY(early_gdt_descr)
717 .word GDT_ENTRIES*8-1
718 .long gdt_page /* Overwritten for secondary CPUs */
721 * The boot_gdt must mirror the equivalent in setup.S and is
722 * used only for booting.
724 .align L1_CACHE_BYTES
726 .fill GDT_ENTRY_BOOT_CS,8,0
727 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
728 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */