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 #include <asm/bootparam.h>
27 /* Physical address */
28 #define pa(X) ((X) - __PAGE_OFFSET)
31 * References to members of the new_cpu_data structure.
34 #define X86 new_cpu_data+CPUINFO_x86
35 #define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor
36 #define X86_MODEL new_cpu_data+CPUINFO_x86_model
37 #define X86_MASK new_cpu_data+CPUINFO_x86_mask
38 #define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math
39 #define X86_CPUID new_cpu_data+CPUINFO_cpuid_level
40 #define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability
41 #define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id
44 * This is how much memory in addition to the memory covered up to
45 * and including _end we need mapped initially.
47 * (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
48 * (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
50 * Modulo rounding, each megabyte assigned here requires a kilobyte of
51 * memory, which is currently unreclaimed.
53 * This should be a multiple of a page.
55 * KERNEL_IMAGE_SIZE should be greater than pa(_end)
56 * and small than max_low_pfn, otherwise will waste some page table entries
60 #define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
62 #define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
65 /* Number of possible pages in the lowmem region */
66 LOWMEM_PAGES = (((1<<32) - __PAGE_OFFSET) >> PAGE_SHIFT)
68 /* Enough space to fit pagetables for the low memory linear map */
69 MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
72 * Worst-case size of the kernel mapping we need to make:
73 * a relocatable kernel can live anywhere in lowmem, so we need to be able
74 * to map all of lowmem.
76 KERNEL_PAGES = LOWMEM_PAGES
78 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
79 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
82 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
83 * %esi points to the real-mode code as a 32-bit pointer.
84 * CS and DS must be 4 GB flat segments, but we don't depend on
85 * any particular GDT layout, because we load our own as soon as we
90 movl pa(stack_start),%ecx
92 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
93 us to not reload segments */
94 testb $KEEP_SEGMENTS, BP_loadflags(%esi)
98 * Set segments to known values.
100 lgdt pa(boot_gdt_descr)
101 movl $(__BOOT_DS),%eax
108 leal -__PAGE_OFFSET(%ecx),%esp
111 * Clear BSS first so that there are no surprises...
115 movl $pa(__bss_start),%edi
116 movl $pa(__bss_stop),%ecx
121 * Copy bootup parameters out of the way.
122 * Note: %esi still has the pointer to the real-mode data.
123 * With the kexec as boot loader, parameter segment might be loaded beyond
124 * kernel image and might not even be addressable by early boot page tables.
125 * (kexec on panic case). Hence copy out the parameters before initializing
128 movl $pa(boot_params),%edi
129 movl $(PARAM_SIZE/4),%ecx
133 movl pa(boot_params) + NEW_CL_POINTER,%esi
135 jz 1f # No command line
136 movl $pa(boot_command_line),%edi
137 movl $(COMMAND_LINE_SIZE/4),%ecx
143 /* save OFW's pgdir table for later use when calling into OFW */
145 movl %eax, pa(olpc_ofw_pgd)
148 #ifdef CONFIG_MICROCODE_EARLY
149 /* Early load ucode on BSP. */
154 * Initialize page tables. This creates a PDE and a set of page
155 * tables, which are located immediately beyond __brk_base. The variable
156 * _brk_end is set up to point to the first "safe" location.
157 * Mappings are created both at virtual address 0 (identity mapping)
158 * and PAGE_OFFSET for up to _end.
160 #ifdef CONFIG_X86_PAE
163 * In PAE mode initial_page_table is statically defined to contain
164 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
165 * entries). The identity mapping is handled by pointing two PGD entries
166 * to the first kernel PMD.
168 * Note the upper half of each PMD or PTE are always zero at this stage.
171 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
173 xorl %ebx,%ebx /* %ebx is kept at zero */
175 movl $pa(__brk_base), %edi
176 movl $pa(initial_pg_pmd), %edx
177 movl $PTE_IDENT_ATTR, %eax
179 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
180 movl %ecx,(%edx) /* Store PMD entry */
181 /* Upper half already zero */
193 * End condition: we must map up to the end + MAPPING_BEYOND_END.
195 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
199 addl $__PAGE_OFFSET, %edi
200 movl %edi, pa(_brk_end)
202 movl %eax, pa(max_pfn_mapped)
204 /* Do early initialization of the fixmap area */
205 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
206 movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
209 page_pde_offset = (__PAGE_OFFSET >> 20);
211 movl $pa(__brk_base), %edi
212 movl $pa(initial_page_table), %edx
213 movl $PTE_IDENT_ATTR, %eax
215 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
216 movl %ecx,(%edx) /* Store identity PDE entry */
217 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
225 * End condition: we must map up to the end + MAPPING_BEYOND_END.
227 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
230 addl $__PAGE_OFFSET, %edi
231 movl %edi, pa(_brk_end)
233 movl %eax, pa(max_pfn_mapped)
235 /* Do early initialization of the fixmap area */
236 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
237 movl %eax,pa(initial_page_table+0xffc)
240 #ifdef CONFIG_PARAVIRT
241 /* This is can only trip for a broken bootloader... */
242 cmpw $0x207, pa(boot_params + BP_version)
245 /* Paravirt-compatible boot parameters. Look to see what architecture
246 we're booting under. */
247 movl pa(boot_params + BP_hardware_subarch), %eax
248 cmpl $num_subarch_entries, %eax
251 movl pa(subarch_entries)(,%eax,4), %eax
252 subl $__PAGE_OFFSET, %eax
258 /* Unknown implementation; there's really
259 nothing we can do at this point. */
265 .long default_entry /* normal x86/PC */
266 .long lguest_entry /* lguest hypervisor */
267 .long xen_entry /* Xen hypervisor */
268 .long default_entry /* Moorestown MID */
269 num_subarch_entries = (. - subarch_entries) / 4
273 #endif /* CONFIG_PARAVIRT */
275 #ifdef CONFIG_HOTPLUG_CPU
277 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
278 * up already except stack. We just set up stack here. Then call
282 movl stack_start, %ecx
289 * Non-boot CPU entry point; entered from trampoline.S
290 * We can't lgdt here, because lgdt itself uses a data segment, but
291 * we know the trampoline has already loaded the boot_gdt for us.
293 * If cpu hotplug is not supported then this code can go in init section
294 * which will be freed later
296 ENTRY(startup_32_smp)
298 movl $(__BOOT_DS),%eax
303 movl pa(stack_start),%ecx
305 leal -__PAGE_OFFSET(%ecx),%esp
307 #ifdef CONFIG_MICROCODE_EARLY
308 /* Early load ucode on AP. */
314 #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
315 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
317 movl $(CR0_STATE & ~X86_CR0_PG),%eax
321 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
322 * bits like NT set. This would confuse the debugger if this code is traced. So
323 * initialize them properly now before switching to protected mode. That means
324 * DF in particular (even though we have cleared it earlier after copying the
325 * command line) because GCC expects it.
331 * New page tables may be in 4Mbyte page mode and may be using the global pages.
333 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
334 * if and only if CPUID exists and has flags other than the FPU flag set.
336 movl $-1,pa(X86_CPUID) # preset CPUID level
337 movl $X86_EFLAGS_ID,%ecx
339 popfl # set EFLAGS=ID
341 popl %eax # get EFLAGS
342 testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set?
343 jz enable_paging # hw disallowed setting of ID bit
344 # which means no CPUID and no CR4
348 movl %eax,pa(X86_CPUID) # save largest std CPUID function
352 andl $~1,%edx # Ignore CPUID.FPU
353 jz enable_paging # No flags or only CPUID.FPU = no CR4
355 movl pa(mmu_cr4_features),%eax
358 testb $X86_CR4_PAE, %al # check if PAE is enabled
361 /* Check if extended functions are implemented */
362 movl $0x80000000, %eax
364 /* Value must be in the range 0x80000001 to 0x8000ffff */
365 subl $0x80000001, %eax
366 cmpl $(0x8000ffff-0x80000001), %eax
369 /* Clear bogus XD_DISABLE bits */
372 mov $0x80000001, %eax
374 /* Execute Disable bit supported? */
375 btl $(X86_FEATURE_NX & 31), %edx
378 /* Setup EFER (Extended Feature Enable Register) */
383 /* Make changes effective */
391 movl $pa(initial_page_table), %eax
392 movl %eax,%cr3 /* set the page table pointer.. */
394 movl %eax,%cr0 /* ..and set paging (PG) bit */
395 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
397 /* Shift the stack pointer to a virtual address */
398 addl $__PAGE_OFFSET, %esp
401 * start system 32-bit setup. We need to re-do some of the things done
402 * in 16-bit mode for the "real" operations.
404 movl setup_once_ref,%eax
406 jz 1f # Did we do this already?
413 movb $4,X86 # at least 486
417 /* get vendor info */
418 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
420 movl %eax,X86_CPUID # save CPUID level
421 movl %ebx,X86_VENDOR_ID # lo 4 chars
422 movl %edx,X86_VENDOR_ID+4 # next 4 chars
423 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
425 orl %eax,%eax # do we have processor info as well?
428 movl $1,%eax # Use the CPUID instruction to get CPU type
430 movb %al,%cl # save reg for future use
431 andb $0x0f,%ah # mask processor family
433 andb $0xf0,%al # mask model
436 andb $0x0f,%cl # mask mask revision
438 movl %edx,X86_CAPABILITY
441 movl $0x50022,%ecx # set AM, WP, NE and MP
443 andl $0x80000011,%eax # Save PG,PE,ET
449 ljmp $(__KERNEL_CS),$1f
450 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
451 movl %eax,%ss # after changing gdt.
453 movl $(__USER_DS),%eax # DS/ES contains default USER segment
457 movl $(__KERNEL_PERCPU), %eax
458 movl %eax,%fs # set this cpu's percpu
460 movl $(__KERNEL_STACK_CANARY),%eax
463 xorl %eax,%eax # Clear LDT
466 pushl $0 # fake return address for unwinder
469 #include "verify_cpu.S"
474 * The setup work we only want to run on the BSP.
476 * Warning: %esi is live across this function.
481 * Set up a idt with 256 entries pointing to ignore_int,
482 * interrupt gates. It doesn't actually load idt - that needs
483 * to be done on each CPU. Interrupts are enabled elsewhere,
484 * when we can be relatively sure everything is ok.
488 movl $early_idt_handlers,%eax
489 movl $NUM_EXCEPTION_VECTORS,%ecx
493 /* interrupt gate, dpl=0, present */
494 movl $(0x8E000000 + __KERNEL_CS),2(%edi)
499 movl $256 - NUM_EXCEPTION_VECTORS,%ecx
500 movl $ignore_int,%edx
501 movl $(__KERNEL_CS << 16),%eax
502 movw %dx,%ax /* selector = 0x0010 = cs */
503 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
510 #ifdef CONFIG_CC_STACKPROTECTOR
512 * Configure the stack canary. The linker can't handle this by
513 * relocation. Manually set base address in stack canary
514 * segment descriptor.
517 movl $stack_canary,%ecx
518 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
520 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
521 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
524 andl $0,setup_once_ref /* Once is enough, thanks */
527 ENTRY(early_idt_handlers)
531 # 24(%rsp) error code
533 .rept NUM_EXCEPTION_VECTORS
534 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
537 pushl $0 # Dummy error code, to make stack frame uniform
539 pushl $i # 20(%esp) Vector number
540 jmp early_idt_handler
543 ENDPROC(early_idt_handlers)
545 /* This is global to keep gas from relaxing the jumps */
546 ENTRY(early_idt_handler)
549 cmpl $2,(%esp) # X86_TRAP_NMI
550 je is_nmi # Ignore NMI
552 cmpl $2,%ss:early_recursion_flag
554 incl %ss:early_recursion_flag
561 movl $(__KERNEL_DS),%eax
565 cmpl $(__KERNEL_CS),32(%esp)
568 leal 28(%esp),%eax # Pointer to %eip
569 call early_fixup_exception
571 jnz ex_entry /* found an exception entry */
576 movw %ax,2(%esp) /* clean up the segment values on some cpus */
580 pushl %eax /* %esp before the exception */
587 pushl (20+6*4)(%esp) /* trapno */
602 decl %ss:early_recursion_flag
604 addl $8,%esp /* drop vector number and error code */
606 ENDPROC(early_idt_handler)
608 /* This is the default interrupt "handler" :-) */
618 movl $(__KERNEL_DS),%eax
621 cmpl $2,early_recursion_flag
623 incl early_recursion_flag
644 early_recursion_flag:
650 .long i386_start_kernel
651 ENTRY(setup_once_ref)
659 #ifdef CONFIG_X86_PAE
663 ENTRY(initial_page_table)
668 ENTRY(empty_zero_page)
670 ENTRY(swapper_pg_dir)
674 * This starts the data section.
676 #ifdef CONFIG_X86_PAE
678 /* Page-aligned for the benefit of paravirt? */
680 ENTRY(initial_page_table)
681 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
683 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
684 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
685 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
688 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
689 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
693 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
695 # error "Kernel PMDs should be 1, 2 or 3"
697 .align PAGE_SIZE /* needs to be page-sized too */
703 .long init_thread_union+THREAD_SIZE
707 .asciz "Unknown interrupt or fault at: %p %p %p\n"
711 .ascii "BUG: Int %d: CR2 %p\n"
712 /* regs pushed in early_idt_handler: */
713 .ascii " EDI %p ESI %p EBP %p EBX %p\n"
714 .ascii " ESP %p ES %p DS %p\n"
715 .ascii " EDX %p ECX %p EAX %p\n"
717 .ascii " vec %p err %p EIP %p CS %p flg %p\n"
718 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
719 .ascii " %p %p %p %p %p %p %p %p\n"
720 .asciz " %p %p %p %p %p %p %p %p\n"
722 #include "../../x86/xen/xen-head.S"
725 * The IDT and GDT 'descriptors' are a strange 48-bit object
726 * only used by the lidt and lgdt instructions. They are not
727 * like usual segment descriptors - they consist of a 16-bit
728 * segment size, and 32-bit linear address value:
732 .globl boot_gdt_descr
736 # early boot GDT descriptor (must use 1:1 address mapping)
737 .word 0 # 32 bit align gdt_desc.address
740 .long boot_gdt - __PAGE_OFFSET
742 .word 0 # 32-bit align idt_desc.address
744 .word IDT_ENTRIES*8-1 # idt contains 256 entries
747 # boot GDT descriptor (later on used by CPU#0):
748 .word 0 # 32 bit align gdt_desc.address
749 ENTRY(early_gdt_descr)
750 .word GDT_ENTRIES*8-1
751 .long gdt_page /* Overwritten for secondary CPUs */
754 * The boot_gdt must mirror the equivalent in setup.S and is
755 * used only for booting.
757 .align L1_CACHE_BYTES
759 .fill GDT_ENTRY_BOOT_CS,8,0
760 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
761 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */