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/cpufeatures.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)
66 * Number of possible pages in the lowmem region.
68 * We shift 2 by 31 instead of 1 by 32 to the left in order to avoid a
69 * gas warning about overflowing shift count when gas has been compiled
70 * with only a host target support using a 32-bit type for internal
73 LOWMEM_PAGES = (((2<<31) - __PAGE_OFFSET) >> PAGE_SHIFT)
75 /* Enough space to fit pagetables for the low memory linear map */
76 MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
79 * Worst-case size of the kernel mapping we need to make:
80 * a relocatable kernel can live anywhere in lowmem, so we need to be able
81 * to map all of lowmem.
83 KERNEL_PAGES = LOWMEM_PAGES
85 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
86 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
89 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
90 * %esi points to the real-mode code as a 32-bit pointer.
91 * CS and DS must be 4 GB flat segments, but we don't depend on
92 * any particular GDT layout, because we load our own as soon as we
97 movl pa(stack_start),%ecx
99 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
100 us to not reload segments */
101 testb $KEEP_SEGMENTS, BP_loadflags(%esi)
105 * Set segments to known values.
107 lgdt pa(boot_gdt_descr)
108 movl $(__BOOT_DS),%eax
115 leal -__PAGE_OFFSET(%ecx),%esp
118 * Clear BSS first so that there are no surprises...
122 movl $pa(__bss_start),%edi
123 movl $pa(__bss_stop),%ecx
128 * Copy bootup parameters out of the way.
129 * Note: %esi still has the pointer to the real-mode data.
130 * With the kexec as boot loader, parameter segment might be loaded beyond
131 * kernel image and might not even be addressable by early boot page tables.
132 * (kexec on panic case). Hence copy out the parameters before initializing
135 movl $pa(boot_params),%edi
136 movl $(PARAM_SIZE/4),%ecx
140 movl pa(boot_params) + NEW_CL_POINTER,%esi
142 jz 1f # No command line
143 movl $pa(boot_command_line),%edi
144 movl $(COMMAND_LINE_SIZE/4),%ecx
150 /* save OFW's pgdir table for later use when calling into OFW */
152 movl %eax, pa(olpc_ofw_pgd)
155 #ifdef CONFIG_MICROCODE
156 /* Early load ucode on BSP. */
161 * Initialize page tables. This creates a PDE and a set of page
162 * tables, which are located immediately beyond __brk_base. The variable
163 * _brk_end is set up to point to the first "safe" location.
164 * Mappings are created both at virtual address 0 (identity mapping)
165 * and PAGE_OFFSET for up to _end.
167 #ifdef CONFIG_X86_PAE
170 * In PAE mode initial_page_table is statically defined to contain
171 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
172 * entries). The identity mapping is handled by pointing two PGD entries
173 * to the first kernel PMD.
175 * Note the upper half of each PMD or PTE are always zero at this stage.
178 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
180 xorl %ebx,%ebx /* %ebx is kept at zero */
182 movl $pa(__brk_base), %edi
183 movl $pa(initial_pg_pmd), %edx
184 movl $PTE_IDENT_ATTR, %eax
186 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
187 movl %ecx,(%edx) /* Store PMD entry */
188 /* Upper half already zero */
200 * End condition: we must map up to the end + MAPPING_BEYOND_END.
202 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
206 addl $__PAGE_OFFSET, %edi
207 movl %edi, pa(_brk_end)
209 movl %eax, pa(max_pfn_mapped)
211 /* Do early initialization of the fixmap area */
212 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
213 movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
216 page_pde_offset = (__PAGE_OFFSET >> 20);
218 movl $pa(__brk_base), %edi
219 movl $pa(initial_page_table), %edx
220 movl $PTE_IDENT_ATTR, %eax
222 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
223 movl %ecx,(%edx) /* Store identity PDE entry */
224 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
232 * End condition: we must map up to the end + MAPPING_BEYOND_END.
234 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
237 addl $__PAGE_OFFSET, %edi
238 movl %edi, pa(_brk_end)
240 movl %eax, pa(max_pfn_mapped)
242 /* Do early initialization of the fixmap area */
243 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
244 movl %eax,pa(initial_page_table+0xffc)
247 #ifdef CONFIG_PARAVIRT
248 /* This is can only trip for a broken bootloader... */
249 cmpw $0x207, pa(boot_params + BP_version)
252 /* Paravirt-compatible boot parameters. Look to see what architecture
253 we're booting under. */
254 movl pa(boot_params + BP_hardware_subarch), %eax
255 cmpl $num_subarch_entries, %eax
258 movl pa(subarch_entries)(,%eax,4), %eax
259 subl $__PAGE_OFFSET, %eax
265 /* Unknown implementation; there's really
266 nothing we can do at this point. */
272 .long default_entry /* normal x86/PC */
273 .long lguest_entry /* lguest hypervisor */
274 .long xen_entry /* Xen hypervisor */
275 .long default_entry /* Moorestown MID */
276 num_subarch_entries = (. - subarch_entries) / 4
280 #endif /* CONFIG_PARAVIRT */
282 #ifdef CONFIG_HOTPLUG_CPU
284 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
285 * up already except stack. We just set up stack here. Then call
289 movl stack_start, %ecx
296 * Non-boot CPU entry point; entered from trampoline.S
297 * We can't lgdt here, because lgdt itself uses a data segment, but
298 * we know the trampoline has already loaded the boot_gdt for us.
300 * If cpu hotplug is not supported then this code can go in init section
301 * which will be freed later
303 ENTRY(startup_32_smp)
305 movl $(__BOOT_DS),%eax
310 movl pa(stack_start),%ecx
312 leal -__PAGE_OFFSET(%ecx),%esp
314 #ifdef CONFIG_MICROCODE
315 /* Early load ucode on AP. */
320 #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
321 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
323 movl $(CR0_STATE & ~X86_CR0_PG),%eax
327 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
328 * bits like NT set. This would confuse the debugger if this code is traced. So
329 * initialize them properly now before switching to protected mode. That means
330 * DF in particular (even though we have cleared it earlier after copying the
331 * command line) because GCC expects it.
337 * New page tables may be in 4Mbyte page mode and may be using the global pages.
339 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
340 * if and only if CPUID exists and has flags other than the FPU flag set.
342 movl $-1,pa(X86_CPUID) # preset CPUID level
343 movl $X86_EFLAGS_ID,%ecx
345 popfl # set EFLAGS=ID
347 popl %eax # get EFLAGS
348 testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set?
349 jz enable_paging # hw disallowed setting of ID bit
350 # which means no CPUID and no CR4
354 movl %eax,pa(X86_CPUID) # save largest std CPUID function
358 andl $~1,%edx # Ignore CPUID.FPU
359 jz enable_paging # No flags or only CPUID.FPU = no CR4
361 movl pa(mmu_cr4_features),%eax
364 testb $X86_CR4_PAE, %al # check if PAE is enabled
367 /* Check if extended functions are implemented */
368 movl $0x80000000, %eax
370 /* Value must be in the range 0x80000001 to 0x8000ffff */
371 subl $0x80000001, %eax
372 cmpl $(0x8000ffff-0x80000001), %eax
375 /* Clear bogus XD_DISABLE bits */
378 mov $0x80000001, %eax
380 /* Execute Disable bit supported? */
381 btl $(X86_FEATURE_NX & 31), %edx
384 /* Setup EFER (Extended Feature Enable Register) */
389 /* Make changes effective */
397 movl $pa(initial_page_table), %eax
398 movl %eax,%cr3 /* set the page table pointer.. */
400 movl %eax,%cr0 /* ..and set paging (PG) bit */
401 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
403 /* Shift the stack pointer to a virtual address */
404 addl $__PAGE_OFFSET, %esp
407 * start system 32-bit setup. We need to re-do some of the things done
408 * in 16-bit mode for the "real" operations.
410 movl setup_once_ref,%eax
412 jz 1f # Did we do this already?
419 movb $4,X86 # at least 486
423 /* get vendor info */
424 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
426 movl %eax,X86_CPUID # save CPUID level
427 movl %ebx,X86_VENDOR_ID # lo 4 chars
428 movl %edx,X86_VENDOR_ID+4 # next 4 chars
429 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
431 orl %eax,%eax # do we have processor info as well?
434 movl $1,%eax # Use the CPUID instruction to get CPU type
436 movb %al,%cl # save reg for future use
437 andb $0x0f,%ah # mask processor family
439 andb $0xf0,%al # mask model
442 andb $0x0f,%cl # mask mask revision
444 movl %edx,X86_CAPABILITY
447 movl $0x50022,%ecx # set AM, WP, NE and MP
449 andl $0x80000011,%eax # Save PG,PE,ET
455 ljmp $(__KERNEL_CS),$1f
456 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
457 movl %eax,%ss # after changing gdt.
459 movl $(__USER_DS),%eax # DS/ES contains default USER segment
463 movl $(__KERNEL_PERCPU), %eax
464 movl %eax,%fs # set this cpu's percpu
466 movl $(__KERNEL_STACK_CANARY),%eax
469 xorl %eax,%eax # Clear LDT
472 pushl $0 # fake return address for unwinder
475 #include "verify_cpu.S"
480 * The setup work we only want to run on the BSP.
482 * Warning: %esi is live across this function.
487 * Set up a idt with 256 interrupt gates that push zero if there
488 * is no error code and then jump to early_idt_handler_common.
489 * It doesn't actually load the idt - that needs to be done on
490 * each CPU. Interrupts are enabled elsewhere, when we can be
491 * relatively sure everything is ok.
495 movl $early_idt_handler_array,%eax
496 movl $NUM_EXCEPTION_VECTORS,%ecx
500 /* interrupt gate, dpl=0, present */
501 movl $(0x8E000000 + __KERNEL_CS),2(%edi)
502 addl $EARLY_IDT_HANDLER_SIZE,%eax
506 movl $256 - NUM_EXCEPTION_VECTORS,%ecx
507 movl $ignore_int,%edx
508 movl $(__KERNEL_CS << 16),%eax
509 movw %dx,%ax /* selector = 0x0010 = cs */
510 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
517 #ifdef CONFIG_CC_STACKPROTECTOR
519 * Configure the stack canary. The linker can't handle this by
520 * relocation. Manually set base address in stack canary
521 * segment descriptor.
524 movl $stack_canary,%ecx
525 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
527 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
528 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
531 andl $0,setup_once_ref /* Once is enough, thanks */
534 ENTRY(early_idt_handler_array)
538 # 24(%rsp) error code
540 .rept NUM_EXCEPTION_VECTORS
541 .ifeq (EXCEPTION_ERRCODE_MASK >> i) & 1
542 pushl $0 # Dummy error code, to make stack frame uniform
544 pushl $i # 20(%esp) Vector number
545 jmp early_idt_handler_common
547 .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc
549 ENDPROC(early_idt_handler_array)
551 early_idt_handler_common:
553 * The stack is the hardware frame, an error code or zero, and the
558 incl %ss:early_recursion_flag
560 /* The vector number is in pt_regs->gs */
563 pushl %fs /* pt_regs->fs */
564 movw $0, 2(%esp) /* clear high bits (some CPUs leave garbage) */
565 pushl %es /* pt_regs->es */
566 movw $0, 2(%esp) /* clear high bits (some CPUs leave garbage) */
567 pushl %ds /* pt_regs->ds */
568 movw $0, 2(%esp) /* clear high bits (some CPUs leave garbage) */
569 pushl %eax /* pt_regs->ax */
570 pushl %ebp /* pt_regs->bp */
571 pushl %edi /* pt_regs->di */
572 pushl %esi /* pt_regs->si */
573 pushl %edx /* pt_regs->dx */
574 pushl %ecx /* pt_regs->cx */
575 pushl %ebx /* pt_regs->bx */
577 /* Fix up DS and ES */
578 movl $(__KERNEL_DS), %ecx
582 /* Load the vector number into EDX */
583 movl PT_GS(%esp), %edx
585 /* Load GS into pt_regs->gs and clear high bits */
586 movw %gs, PT_GS(%esp)
587 movw $0, PT_GS+2(%esp)
589 movl %esp, %eax /* args are pt_regs (EAX), trapnr (EDX) */
590 call early_fixup_exception
592 popl %ebx /* pt_regs->bx */
593 popl %ecx /* pt_regs->cx */
594 popl %edx /* pt_regs->dx */
595 popl %esi /* pt_regs->si */
596 popl %edi /* pt_regs->di */
597 popl %ebp /* pt_regs->bp */
598 popl %eax /* pt_regs->ax */
599 popl %ds /* pt_regs->ds */
600 popl %es /* pt_regs->es */
601 popl %fs /* pt_regs->fs */
602 popl %gs /* pt_regs->gs */
603 decl %ss:early_recursion_flag
604 addl $4, %esp /* pop pt_regs->orig_ax */
606 ENDPROC(early_idt_handler_common)
608 /* This is the default interrupt "handler" :-) */
618 movl $(__KERNEL_DS),%eax
621 cmpl $2,early_recursion_flag
623 incl early_recursion_flag
648 GLOBAL(early_recursion_flag)
654 .long i386_start_kernel
655 ENTRY(setup_once_ref)
663 #ifdef CONFIG_X86_PAE
667 ENTRY(initial_page_table)
672 ENTRY(empty_zero_page)
674 ENTRY(swapper_pg_dir)
678 * This starts the data section.
680 #ifdef CONFIG_X86_PAE
682 /* Page-aligned for the benefit of paravirt? */
684 ENTRY(initial_page_table)
685 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
687 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
688 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
689 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
692 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
693 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
697 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
699 # error "Kernel PMDs should be 1, 2 or 3"
701 .align PAGE_SIZE /* needs to be page-sized too */
707 .long init_thread_union+THREAD_SIZE
711 .asciz "Unknown interrupt or fault at: %p %p %p\n"
713 #include "../../x86/xen/xen-head.S"
716 * The IDT and GDT 'descriptors' are a strange 48-bit object
717 * only used by the lidt and lgdt instructions. They are not
718 * like usual segment descriptors - they consist of a 16-bit
719 * segment size, and 32-bit linear address value:
723 .globl boot_gdt_descr
727 # early boot GDT descriptor (must use 1:1 address mapping)
728 .word 0 # 32 bit align gdt_desc.address
731 .long boot_gdt - __PAGE_OFFSET
733 .word 0 # 32-bit align idt_desc.address
735 .word IDT_ENTRIES*8-1 # idt contains 256 entries
738 # boot GDT descriptor (later on used by CPU#0):
739 .word 0 # 32 bit align gdt_desc.address
740 ENTRY(early_gdt_descr)
741 .word GDT_ENTRIES*8-1
742 .long gdt_page /* Overwritten for secondary CPUs */
745 * The boot_gdt must mirror the equivalent in setup.S and is
746 * used only for booting.
748 .align L1_CACHE_BYTES
750 .fill GDT_ENTRY_BOOT_CS,8,0
751 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
752 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */