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/percpu.h>
23 /* Physical address */
24 #define pa(X) ((X) - __PAGE_OFFSET)
27 * References to members of the new_cpu_data structure.
30 #define X86 new_cpu_data+CPUINFO_x86
31 #define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor
32 #define X86_MODEL new_cpu_data+CPUINFO_x86_model
33 #define X86_MASK new_cpu_data+CPUINFO_x86_mask
34 #define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math
35 #define X86_CPUID new_cpu_data+CPUINFO_cpuid_level
36 #define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability
37 #define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id
40 * This is how much memory in addition to the memory covered up to
41 * and including _end we need mapped initially.
43 * (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
44 * (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
46 * Modulo rounding, each megabyte assigned here requires a kilobyte of
47 * memory, which is currently unreclaimed.
49 * This should be a multiple of a page.
51 * KERNEL_IMAGE_SIZE should be greater than pa(_end)
52 * and small than max_low_pfn, otherwise will waste some page table entries
56 #define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
58 #define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
61 /* Enough space to fit pagetables for the low memory linear map */
62 MAPPING_BEYOND_END = \
63 PAGE_TABLE_SIZE(((1<<32) - __PAGE_OFFSET) >> PAGE_SHIFT) << PAGE_SHIFT
66 * Worst-case size of the kernel mapping we need to make:
67 * the worst-case size of the kernel itself, plus the extra we need
68 * to map for the linear map.
70 KERNEL_PAGES = (KERNEL_IMAGE_SIZE + MAPPING_BEYOND_END)>>PAGE_SHIFT
72 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE_asm
73 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
76 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
77 * %esi points to the real-mode code as a 32-bit pointer.
78 * CS and DS must be 4 GB flat segments, but we don't depend on
79 * any particular GDT layout, because we load our own as soon as we
84 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
85 us to not reload segments */
86 testb $(1<<6), BP_loadflags(%esi)
90 * Set segments to known values.
92 lgdt pa(boot_gdt_descr)
93 movl $(__BOOT_DS),%eax
101 * Clear BSS first so that there are no surprises...
105 movl $pa(__bss_start),%edi
106 movl $pa(__bss_stop),%ecx
111 * Copy bootup parameters out of the way.
112 * Note: %esi still has the pointer to the real-mode data.
113 * With the kexec as boot loader, parameter segment might be loaded beyond
114 * kernel image and might not even be addressable by early boot page tables.
115 * (kexec on panic case). Hence copy out the parameters before initializing
118 movl $pa(boot_params),%edi
119 movl $(PARAM_SIZE/4),%ecx
123 movl pa(boot_params) + NEW_CL_POINTER,%esi
125 jz 1f # No comand line
126 movl $pa(boot_command_line),%edi
127 movl $(COMMAND_LINE_SIZE/4),%ecx
132 #ifdef CONFIG_PARAVIRT
133 /* This is can only trip for a broken bootloader... */
134 cmpw $0x207, pa(boot_params + BP_version)
137 /* Paravirt-compatible boot parameters. Look to see what architecture
138 we're booting under. */
139 movl pa(boot_params + BP_hardware_subarch), %eax
140 cmpl $num_subarch_entries, %eax
143 movl pa(subarch_entries)(,%eax,4), %eax
144 subl $__PAGE_OFFSET, %eax
150 /* Unknown implementation; there's really
151 nothing we can do at this point. */
157 .long default_entry /* normal x86/PC */
158 .long lguest_entry /* lguest hypervisor */
159 .long xen_entry /* Xen hypervisor */
160 .long default_entry /* Moorestown MID */
161 num_subarch_entries = (. - subarch_entries) / 4
163 #endif /* CONFIG_PARAVIRT */
166 * Initialize page tables. This creates a PDE and a set of page
167 * tables, which are located immediately beyond __brk_base. The variable
168 * _brk_end is set up to point to the first "safe" location.
169 * Mappings are created both at virtual address 0 (identity mapping)
170 * and PAGE_OFFSET for up to _end.
172 * Note that the stack is not yet set up!
175 #ifdef CONFIG_X86_PAE
178 * In PAE mode swapper_pg_dir is statically defined to contain enough
179 * entries to cover the VMSPLIT option (that is the top 1, 2 or 3
180 * entries). The identity mapping is handled by pointing two PGD
181 * entries to the first kernel PMD.
183 * Note the upper half of each PMD or PTE are always zero at
187 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
189 xorl %ebx,%ebx /* %ebx is kept at zero */
191 movl $pa(__brk_base), %edi
192 movl $pa(swapper_pg_pmd), %edx
193 movl $PTE_IDENT_ATTR, %eax
195 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
196 movl %ecx,(%edx) /* Store PMD entry */
197 /* Upper half already zero */
209 * End condition: we must map up to the end + MAPPING_BEYOND_END.
211 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
215 addl $__PAGE_OFFSET, %edi
216 movl %edi, pa(_brk_end)
218 movl %eax, pa(max_pfn_mapped)
220 /* Do early initialization of the fixmap area */
221 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
222 movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8)
225 page_pde_offset = (__PAGE_OFFSET >> 20);
227 movl $pa(__brk_base), %edi
228 movl $pa(swapper_pg_dir), %edx
229 movl $PTE_IDENT_ATTR, %eax
231 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
232 movl %ecx,(%edx) /* Store identity PDE entry */
233 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
241 * End condition: we must map up to the end + MAPPING_BEYOND_END.
243 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
246 addl $__PAGE_OFFSET, %edi
247 movl %edi, pa(_brk_end)
249 movl %eax, pa(max_pfn_mapped)
251 /* Do early initialization of the fixmap area */
252 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
253 movl %eax,pa(swapper_pg_dir+0xffc)
257 * Non-boot CPU entry point; entered from trampoline.S
258 * We can't lgdt here, because lgdt itself uses a data segment, but
259 * we know the trampoline has already loaded the boot_gdt for us.
261 * If cpu hotplug is not supported then this code can go in init section
262 * which will be freed later
268 ENTRY(startup_32_smp)
270 movl $(__BOOT_DS),%eax
275 #endif /* CONFIG_SMP */
279 * New page tables may be in 4Mbyte page mode and may
280 * be using the global pages.
282 * NOTE! If we are on a 486 we may have no cr4 at all!
283 * So we do not try to touch it unless we really have
284 * some bits in it to set. This won't work if the BSP
285 * implements cr4 but this AP does not -- very unlikely
286 * but be warned! The same applies to the pse feature
287 * if not equally supported. --macro
289 * NOTE! We have to correct for the fact that we're
290 * not yet offset PAGE_OFFSET..
292 #define cr4_bits pa(mmu_cr4_features)
296 movl %cr4,%eax # Turn on paging options (PSE,PAE,..)
300 btl $5, %eax # check if PAE is enabled
303 /* Check if extended functions are implemented */
304 movl $0x80000000, %eax
306 cmpl $0x80000000, %eax
308 mov $0x80000001, %eax
310 /* Execute Disable bit supported? */
314 /* Setup EFER (Extended Feature Enable Register) */
315 movl $0xc0000080, %ecx
319 /* Make changes effective */
327 movl $pa(swapper_pg_dir),%eax
328 movl %eax,%cr3 /* set the page table pointer.. */
331 movl %eax,%cr0 /* ..and set paging (PG) bit */
332 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
334 /* Set up the stack pointer */
338 * Initialize eflags. Some BIOS's leave bits like NT set. This would
339 * confuse the debugger if this code is traced.
340 * XXX - best to initialize before switching to protected mode.
347 jz 1f /* Initial CPU cleans BSS */
350 #endif /* CONFIG_SMP */
353 * start system 32-bit setup. We need to re-do some of the things done
354 * in 16-bit mode for the "real" operations.
360 movl $-1,X86_CPUID # -1 for no CPUID initially
362 /* check if it is 486 or 386. */
364 * XXX - this does a lot of unnecessary setup. Alignment checks don't
365 * apply at our cpl of 0 and the stack ought to be aligned already, and
366 * we don't need to preserve eflags.
369 movb $3,X86 # at least 386
371 popl %eax # get EFLAGS
372 movl %eax,%ecx # save original EFLAGS
373 xorl $0x240000,%eax # flip AC and ID bits in EFLAGS
374 pushl %eax # copy to EFLAGS
376 pushfl # get new EFLAGS
377 popl %eax # put it in eax
378 xorl %ecx,%eax # change in flags
379 pushl %ecx # restore original EFLAGS
381 testl $0x40000,%eax # check if AC bit changed
384 movb $4,X86 # at least 486
385 testl $0x200000,%eax # check if ID bit changed
388 /* get vendor info */
389 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
391 movl %eax,X86_CPUID # save CPUID level
392 movl %ebx,X86_VENDOR_ID # lo 4 chars
393 movl %edx,X86_VENDOR_ID+4 # next 4 chars
394 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
396 orl %eax,%eax # do we have processor info as well?
399 movl $1,%eax # Use the CPUID instruction to get CPU type
401 movb %al,%cl # save reg for future use
402 andb $0x0f,%ah # mask processor family
404 andb $0xf0,%al # mask model
407 andb $0x0f,%cl # mask mask revision
409 movl %edx,X86_CAPABILITY
411 is486: movl $0x50022,%ecx # set AM, WP, NE and MP
414 is386: movl $2,%ecx # set MP
416 andl $0x80000011,%eax # Save PG,PE,ET
423 ljmp $(__KERNEL_CS),$1f
424 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
425 movl %eax,%ss # after changing gdt.
427 movl $(__USER_DS),%eax # DS/ES contains default USER segment
431 movl $(__KERNEL_PERCPU), %eax
432 movl %eax,%fs # set this cpu's percpu
434 #ifdef CONFIG_CC_STACKPROTECTOR
436 * The linker can't handle this by relocation. Manually set
437 * base address in stack canary segment descriptor.
441 movl $per_cpu__gdt_page,%eax
442 movl $per_cpu__stack_canary,%ecx
443 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
445 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
446 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
449 movl $(__KERNEL_STACK_CANARY),%eax
452 xorl %eax,%eax # Clear LDT
455 cld # gcc2 wants the direction flag cleared at all times
456 pushl $0 # fake return address for unwinder
460 cmpb $0,%cl # the first CPU calls start_kernel
462 movl (stack_start), %esp
464 #endif /* CONFIG_SMP */
468 * We depend on ET to be correct. This checks for 287/387.
471 movb $0,X86_HARD_MATH
477 movl %cr0,%eax /* no coprocessor: have to set bits */
478 xorl $4,%eax /* set EM */
482 1: movb $1,X86_HARD_MATH
483 .byte 0xDB,0xE4 /* fsetpm for 287, ignored by 387 */
489 * sets up a idt with 256 entries pointing to
490 * ignore_int, interrupt gates. It doesn't actually load
491 * idt - that can be done only after paging has been enabled
492 * and the kernel moved to PAGE_OFFSET. Interrupts
493 * are enabled elsewhere, when we can be relatively
494 * sure everything is ok.
496 * Warning: %esi is live across this function.
500 movl $(__KERNEL_CS << 16),%eax
501 movw %dx,%ax /* selector = 0x0010 = cs */
502 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
513 .macro set_early_handler handler,trapno
515 movl $(__KERNEL_CS << 16),%eax
517 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
519 movl %eax,8*\trapno(%edi)
520 movl %edx,8*\trapno+4(%edi)
523 set_early_handler handler=early_divide_err,trapno=0
524 set_early_handler handler=early_illegal_opcode,trapno=6
525 set_early_handler handler=early_protection_fault,trapno=13
526 set_early_handler handler=early_page_fault,trapno=14
532 pushl $0 /* fake errcode */
535 early_illegal_opcode:
537 pushl $0 /* fake errcode */
540 early_protection_fault:
552 movl $(__KERNEL_DS),%eax
555 cmpl $2,early_recursion_flag
557 incl early_recursion_flag
560 pushl %edx /* trapno */
569 /* This is the default interrupt "handler" :-) */
579 movl $(__KERNEL_DS),%eax
582 cmpl $2,early_recursion_flag
584 incl early_recursion_flag
606 .long i386_start_kernel
613 #ifdef CONFIG_X86_PAE
617 ENTRY(swapper_pg_dir)
622 ENTRY(empty_zero_page)
626 * This starts the data section.
628 #ifdef CONFIG_X86_PAE
630 /* Page-aligned for the benefit of paravirt? */
632 ENTRY(swapper_pg_dir)
633 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
635 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
636 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
637 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x2000),0
640 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
641 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
645 .long pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
647 # error "Kernel PMDs should be 1, 2 or 3"
649 .align PAGE_SIZE_asm /* needs to be page-sized too */
654 .long init_thread_union+THREAD_SIZE
659 early_recursion_flag:
663 .asciz "Unknown interrupt or fault at: %p %p %p\n"
667 .ascii "BUG: Int %d: CR2 %p\n"
669 .ascii " EDI %p ESI %p EBP %p ESP %p\n"
670 .ascii " EBX %p EDX %p ECX %p EAX %p\n"
672 .ascii " err %p EIP %p CS %p flg %p\n"
673 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
674 .ascii " %p %p %p %p %p %p %p %p\n"
675 .asciz " %p %p %p %p %p %p %p %p\n"
677 #include "../../x86/xen/xen-head.S"
680 * The IDT and GDT 'descriptors' are a strange 48-bit object
681 * only used by the lidt and lgdt instructions. They are not
682 * like usual segment descriptors - they consist of a 16-bit
683 * segment size, and 32-bit linear address value:
686 .globl boot_gdt_descr
690 # early boot GDT descriptor (must use 1:1 address mapping)
691 .word 0 # 32 bit align gdt_desc.address
694 .long boot_gdt - __PAGE_OFFSET
696 .word 0 # 32-bit align idt_desc.address
698 .word IDT_ENTRIES*8-1 # idt contains 256 entries
701 # boot GDT descriptor (later on used by CPU#0):
702 .word 0 # 32 bit align gdt_desc.address
703 ENTRY(early_gdt_descr)
704 .word GDT_ENTRIES*8-1
705 .long per_cpu__gdt_page /* Overwritten for secondary CPUs */
708 * The boot_gdt must mirror the equivalent in setup.S and is
709 * used only for booting.
711 .align L1_CACHE_BYTES
713 .fill GDT_ENTRY_BOOT_CS,8,0
714 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
715 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */