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>
15 #include <asm/pgtable.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>
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 * - one bit for each possible page, but only in low memory, which means
44 * 2^32/4096/8 = 128K worst case (4G/4G split.)
45 * - enough space to map all low memory, which means
46 * (2^32/4096) / 1024 pages (worst case, non PAE)
47 * (2^32/4096) / 512 + 4 pages (worst case for PAE)
48 * - a few pages for allocator use before the kernel pagetable has
51 * Modulo rounding, each megabyte assigned here requires a kilobyte of
52 * memory, which is currently unreclaimed.
54 * This should be a multiple of a page.
56 LOW_PAGES = 1<<(32-PAGE_SHIFT_asm)
59 * To preserve the DMA pool in PAGEALLOC kernels, we'll allocate
60 * pagetables from above the 16MB DMA limit, so we'll have to set
61 * up pagetables 16MB more (worst-case):
63 #ifdef CONFIG_DEBUG_PAGEALLOC
64 LOW_PAGES = LOW_PAGES + 0x1000000
68 PAGE_TABLE_SIZE = (LOW_PAGES / PTRS_PER_PMD) + PTRS_PER_PGD
70 PAGE_TABLE_SIZE = (LOW_PAGES / PTRS_PER_PGD)
72 BOOTBITMAP_SIZE = LOW_PAGES / 8
75 INIT_MAP_BEYOND_END = BOOTBITMAP_SIZE + (PAGE_TABLE_SIZE + ALLOCATOR_SLOP)*PAGE_SIZE_asm
78 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
79 * %esi points to the real-mode code as a 32-bit pointer.
80 * CS and DS must be 4 GB flat segments, but we don't depend on
81 * any particular GDT layout, because we load our own as soon as we
84 .section .text.head,"ax",@progbits
86 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
87 us to not reload segments */
88 testb $(1<<6), BP_loadflags(%esi)
92 * Set segments to known values.
94 lgdt pa(boot_gdt_descr)
95 movl $(__BOOT_DS),%eax
103 * Clear BSS first so that there are no surprises...
107 movl $pa(__bss_start),%edi
108 movl $pa(__bss_stop),%ecx
113 * Copy bootup parameters out of the way.
114 * Note: %esi still has the pointer to the real-mode data.
115 * With the kexec as boot loader, parameter segment might be loaded beyond
116 * kernel image and might not even be addressable by early boot page tables.
117 * (kexec on panic case). Hence copy out the parameters before initializing
120 movl $pa(boot_params),%edi
121 movl $(PARAM_SIZE/4),%ecx
125 movl pa(boot_params) + NEW_CL_POINTER,%esi
127 jz 1f # No comand line
128 movl $pa(boot_command_line),%edi
129 movl $(COMMAND_LINE_SIZE/4),%ecx
134 #ifdef CONFIG_PARAVIRT
135 /* This is can only trip for a broken bootloader... */
136 cmpw $0x207, pa(boot_params + BP_version)
139 /* Paravirt-compatible boot parameters. Look to see what architecture
140 we're booting under. */
141 movl pa(boot_params + BP_hardware_subarch), %eax
142 cmpl $num_subarch_entries, %eax
145 movl pa(subarch_entries)(,%eax,4), %eax
146 subl $__PAGE_OFFSET, %eax
152 /* Unknown implementation; there's really
153 nothing we can do at this point. */
159 .long default_entry /* normal x86/PC */
160 .long lguest_entry /* lguest hypervisor */
161 .long xen_entry /* Xen hypervisor */
162 num_subarch_entries = (. - subarch_entries) / 4
164 #endif /* CONFIG_PARAVIRT */
167 * Initialize page tables. This creates a PDE and a set of page
168 * tables, which are located immediately beyond _end. The variable
169 * init_pg_tables_end is set up to point to the first "safe" location.
170 * Mappings are created both at virtual address 0 (identity mapping)
171 * and PAGE_OFFSET for up to _end+sizeof(page tables)+INIT_MAP_BEYOND_END.
173 * Note that the stack is not yet set up!
176 #ifdef CONFIG_X86_PAE
179 * In PAE mode swapper_pg_dir is statically defined to contain enough
180 * entries to cover the VMSPLIT option (that is the top 1, 2 or 3
181 * entries). The identity mapping is handled by pointing two PGD
182 * entries to the first kernel PMD.
184 * Note the upper half of each PMD or PTE are always zero at
188 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
190 xorl %ebx,%ebx /* %ebx is kept at zero */
193 movl %edi, pa(init_pg_tables_start)
194 movl $pa(swapper_pg_pmd), %edx
195 movl $PTE_IDENT_ATTR, %eax
197 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
198 movl %ecx,(%edx) /* Store PMD entry */
199 /* Upper half already zero */
211 * End condition: we must map up to and including INIT_MAP_BEYOND_END
212 * bytes beyond the end of our own page tables.
214 leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
218 movl %edi,pa(init_pg_tables_end)
220 movl %eax, pa(max_pfn_mapped)
222 /* Do early initialization of the fixmap area */
223 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
224 movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8)
227 page_pde_offset = (__PAGE_OFFSET >> 20);
230 movl %edi, pa(init_pg_tables_start)
231 movl $pa(swapper_pg_dir), %edx
232 movl $PTE_IDENT_ATTR, %eax
234 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
235 movl %ecx,(%edx) /* Store identity PDE entry */
236 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
244 * End condition: we must map up to and including INIT_MAP_BEYOND_END
245 * bytes beyond the end of our own page tables; the +0x007 is
248 leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
251 movl %edi,pa(init_pg_tables_end)
253 movl %eax, pa(max_pfn_mapped)
255 /* Do early initialization of the fixmap area */
256 movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
257 movl %eax,pa(swapper_pg_dir+0xffc)
261 * Non-boot CPU entry point; entered from trampoline.S
262 * We can't lgdt here, because lgdt itself uses a data segment, but
263 * we know the trampoline has already loaded the boot_gdt for us.
265 * If cpu hotplug is not supported then this code can go in init section
266 * which will be freed later
269 #ifndef CONFIG_HOTPLUG_CPU
270 .section .init.text,"ax",@progbits
274 ENTRY(startup_32_smp)
276 movl $(__BOOT_DS),%eax
281 #endif /* CONFIG_SMP */
285 * New page tables may be in 4Mbyte page mode and may
286 * be using the global pages.
288 * NOTE! If we are on a 486 we may have no cr4 at all!
289 * So we do not try to touch it unless we really have
290 * some bits in it to set. This won't work if the BSP
291 * implements cr4 but this AP does not -- very unlikely
292 * but be warned! The same applies to the pse feature
293 * if not equally supported. --macro
295 * NOTE! We have to correct for the fact that we're
296 * not yet offset PAGE_OFFSET..
298 #define cr4_bits pa(mmu_cr4_features)
302 movl %cr4,%eax # Turn on paging options (PSE,PAE,..)
306 btl $5, %eax # check if PAE is enabled
309 /* Check if extended functions are implemented */
310 movl $0x80000000, %eax
312 cmpl $0x80000000, %eax
314 mov $0x80000001, %eax
316 /* Execute Disable bit supported? */
320 /* Setup EFER (Extended Feature Enable Register) */
321 movl $0xc0000080, %ecx
325 /* Make changes effective */
333 movl $pa(swapper_pg_dir),%eax
334 movl %eax,%cr3 /* set the page table pointer.. */
337 movl %eax,%cr0 /* ..and set paging (PG) bit */
338 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
340 /* Set up the stack pointer */
344 * Initialize eflags. Some BIOS's leave bits like NT set. This would
345 * confuse the debugger if this code is traced.
346 * XXX - best to initialize before switching to protected mode.
353 jz 1f /* Initial CPU cleans BSS */
356 #endif /* CONFIG_SMP */
359 * start system 32-bit setup. We need to re-do some of the things done
360 * in 16-bit mode for the "real" operations.
366 movl $-1,X86_CPUID # -1 for no CPUID initially
368 /* check if it is 486 or 386. */
370 * XXX - this does a lot of unnecessary setup. Alignment checks don't
371 * apply at our cpl of 0 and the stack ought to be aligned already, and
372 * we don't need to preserve eflags.
375 movb $3,X86 # at least 386
377 popl %eax # get EFLAGS
378 movl %eax,%ecx # save original EFLAGS
379 xorl $0x240000,%eax # flip AC and ID bits in EFLAGS
380 pushl %eax # copy to EFLAGS
382 pushfl # get new EFLAGS
383 popl %eax # put it in eax
384 xorl %ecx,%eax # change in flags
385 pushl %ecx # restore original EFLAGS
387 testl $0x40000,%eax # check if AC bit changed
390 movb $4,X86 # at least 486
391 testl $0x200000,%eax # check if ID bit changed
394 /* get vendor info */
395 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
397 movl %eax,X86_CPUID # save CPUID level
398 movl %ebx,X86_VENDOR_ID # lo 4 chars
399 movl %edx,X86_VENDOR_ID+4 # next 4 chars
400 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
402 orl %eax,%eax # do we have processor info as well?
405 movl $1,%eax # Use the CPUID instruction to get CPU type
407 movb %al,%cl # save reg for future use
408 andb $0x0f,%ah # mask processor family
410 andb $0xf0,%al # mask model
413 andb $0x0f,%cl # mask mask revision
415 movl %edx,X86_CAPABILITY
417 is486: movl $0x50022,%ecx # set AM, WP, NE and MP
420 is386: movl $2,%ecx # set MP
422 andl $0x80000011,%eax # Save PG,PE,ET
429 ljmp $(__KERNEL_CS),$1f
430 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
431 movl %eax,%ss # after changing gdt.
432 movl %eax,%fs # gets reset once there's real percpu
434 movl $(__USER_DS),%eax # DS/ES contains default USER segment
438 xorl %eax,%eax # Clear GS and LDT
442 cld # gcc2 wants the direction flag cleared at all times
443 pushl $0 # fake return address for unwinder
447 cmpb $0,%cl # the first CPU calls start_kernel
449 movl $(__KERNEL_PERCPU), %eax
450 movl %eax,%fs # set this cpu's percpu
451 movl (stack_start), %esp
453 #endif /* CONFIG_SMP */
457 * We depend on ET to be correct. This checks for 287/387.
460 movb $0,X86_HARD_MATH
466 movl %cr0,%eax /* no coprocessor: have to set bits */
467 xorl $4,%eax /* set EM */
471 1: movb $1,X86_HARD_MATH
472 .byte 0xDB,0xE4 /* fsetpm for 287, ignored by 387 */
478 * sets up a idt with 256 entries pointing to
479 * ignore_int, interrupt gates. It doesn't actually load
480 * idt - that can be done only after paging has been enabled
481 * and the kernel moved to PAGE_OFFSET. Interrupts
482 * are enabled elsewhere, when we can be relatively
483 * sure everything is ok.
485 * Warning: %esi is live across this function.
489 movl $(__KERNEL_CS << 16),%eax
490 movw %dx,%ax /* selector = 0x0010 = cs */
491 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
502 .macro set_early_handler handler,trapno
504 movl $(__KERNEL_CS << 16),%eax
506 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
508 movl %eax,8*\trapno(%edi)
509 movl %edx,8*\trapno+4(%edi)
512 set_early_handler handler=early_divide_err,trapno=0
513 set_early_handler handler=early_illegal_opcode,trapno=6
514 set_early_handler handler=early_protection_fault,trapno=13
515 set_early_handler handler=early_page_fault,trapno=14
521 pushl $0 /* fake errcode */
524 early_illegal_opcode:
526 pushl $0 /* fake errcode */
529 early_protection_fault:
541 movl $(__KERNEL_DS),%eax
544 cmpl $2,early_recursion_flag
546 incl early_recursion_flag
549 pushl %edx /* trapno */
551 #ifdef CONFIG_EARLY_PRINTK
562 /* This is the default interrupt "handler" :-) */
572 movl $(__KERNEL_DS),%eax
575 cmpl $2,early_recursion_flag
577 incl early_recursion_flag
583 #ifdef CONFIG_EARLY_PRINTK
597 .section .cpuinit.data,"wa"
600 .long i386_start_kernel
604 * Real beginning of normal "text" segment
612 .section ".bss.page_aligned","wa"
614 #ifdef CONFIG_X86_PAE
618 ENTRY(swapper_pg_dir)
623 ENTRY(empty_zero_page)
626 * This starts the data section.
628 #ifdef CONFIG_X86_PAE
629 .section ".data.page_aligned","wa"
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 EIP %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 */