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>
26 /* Physical address */
27 #define pa(X) ((X) - __PAGE_OFFSET)
30 * References to members of the new_cpu_data structure.
33 #define X86 new_cpu_data+CPUINFO_x86
34 #define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor
35 #define X86_MODEL new_cpu_data+CPUINFO_x86_model
36 #define X86_MASK new_cpu_data+CPUINFO_x86_mask
37 #define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math
38 #define X86_CPUID new_cpu_data+CPUINFO_cpuid_level
39 #define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability
40 #define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id
43 * This is how much memory in addition to the memory covered up to
44 * and including _end we need mapped initially.
46 * (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
47 * (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
49 * Modulo rounding, each megabyte assigned here requires a kilobyte of
50 * memory, which is currently unreclaimed.
52 * This should be a multiple of a page.
54 * KERNEL_IMAGE_SIZE should be greater than pa(_end)
55 * and small than max_low_pfn, otherwise will waste some page table entries
59 #define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
61 #define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
64 /* Number of possible pages in the lowmem region */
65 LOWMEM_PAGES = (((1<<32) - __PAGE_OFFSET) >> PAGE_SHIFT)
67 /* Enough space to fit pagetables for the low memory linear map */
68 MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
71 * Worst-case size of the kernel mapping we need to make:
72 * a relocatable kernel can live anywhere in lowmem, so we need to be able
73 * to map all of lowmem.
75 KERNEL_PAGES = LOWMEM_PAGES
77 INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE
78 RESERVE_BRK(pagetables, INIT_MAP_SIZE)
81 * 32-bit kernel entrypoint; only used by the boot CPU. On entry,
82 * %esi points to the real-mode code as a 32-bit pointer.
83 * CS and DS must be 4 GB flat segments, but we don't depend on
84 * any particular GDT layout, because we load our own as soon as we
89 movl pa(stack_start),%ecx
91 /* test KEEP_SEGMENTS flag to see if the bootloader is asking
92 us to not reload segments */
93 testb $(1<<6), BP_loadflags(%esi)
97 * Set segments to known values.
99 lgdt pa(boot_gdt_descr)
100 movl $(__BOOT_DS),%eax
107 leal -__PAGE_OFFSET(%ecx),%esp
110 * Clear BSS first so that there are no surprises...
114 movl $pa(__bss_start),%edi
115 movl $pa(__bss_stop),%ecx
120 * Copy bootup parameters out of the way.
121 * Note: %esi still has the pointer to the real-mode data.
122 * With the kexec as boot loader, parameter segment might be loaded beyond
123 * kernel image and might not even be addressable by early boot page tables.
124 * (kexec on panic case). Hence copy out the parameters before initializing
127 movl $pa(boot_params),%edi
128 movl $(PARAM_SIZE/4),%ecx
132 movl pa(boot_params) + NEW_CL_POINTER,%esi
134 jz 1f # No command line
135 movl $pa(boot_command_line),%edi
136 movl $(COMMAND_LINE_SIZE/4),%ecx
142 /* save OFW's pgdir table for later use when calling into OFW */
144 movl %eax, pa(olpc_ofw_pgd)
147 #ifdef CONFIG_MICROCODE_EARLY
148 /* Early load ucode on BSP. */
153 * Initialize page tables. This creates a PDE and a set of page
154 * tables, which are located immediately beyond __brk_base. The variable
155 * _brk_end is set up to point to the first "safe" location.
156 * Mappings are created both at virtual address 0 (identity mapping)
157 * and PAGE_OFFSET for up to _end.
159 #ifdef CONFIG_X86_PAE
162 * In PAE mode initial_page_table is statically defined to contain
163 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
164 * entries). The identity mapping is handled by pointing two PGD entries
165 * to the first kernel PMD.
167 * Note the upper half of each PMD or PTE are always zero at this stage.
170 #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
172 xorl %ebx,%ebx /* %ebx is kept at zero */
174 movl $pa(__brk_base), %edi
175 movl $pa(initial_pg_pmd), %edx
176 movl $PTE_IDENT_ATTR, %eax
178 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PMD entry */
179 movl %ecx,(%edx) /* Store PMD entry */
180 /* Upper half already zero */
192 * End condition: we must map up to the end + MAPPING_BEYOND_END.
194 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
198 addl $__PAGE_OFFSET, %edi
199 movl %edi, pa(_brk_end)
201 movl %eax, pa(max_pfn_mapped)
203 /* Do early initialization of the fixmap area */
204 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
205 movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
208 page_pde_offset = (__PAGE_OFFSET >> 20);
210 movl $pa(__brk_base), %edi
211 movl $pa(initial_page_table), %edx
212 movl $PTE_IDENT_ATTR, %eax
214 leal PDE_IDENT_ATTR(%edi),%ecx /* Create PDE entry */
215 movl %ecx,(%edx) /* Store identity PDE entry */
216 movl %ecx,page_pde_offset(%edx) /* Store kernel PDE entry */
224 * End condition: we must map up to the end + MAPPING_BEYOND_END.
226 movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
229 addl $__PAGE_OFFSET, %edi
230 movl %edi, pa(_brk_end)
232 movl %eax, pa(max_pfn_mapped)
234 /* Do early initialization of the fixmap area */
235 movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
236 movl %eax,pa(initial_page_table+0xffc)
239 #ifdef CONFIG_PARAVIRT
240 /* This is can only trip for a broken bootloader... */
241 cmpw $0x207, pa(boot_params + BP_version)
244 /* Paravirt-compatible boot parameters. Look to see what architecture
245 we're booting under. */
246 movl pa(boot_params + BP_hardware_subarch), %eax
247 cmpl $num_subarch_entries, %eax
250 movl pa(subarch_entries)(,%eax,4), %eax
251 subl $__PAGE_OFFSET, %eax
257 /* Unknown implementation; there's really
258 nothing we can do at this point. */
264 .long default_entry /* normal x86/PC */
265 .long lguest_entry /* lguest hypervisor */
266 .long xen_entry /* Xen hypervisor */
267 .long default_entry /* Moorestown MID */
268 num_subarch_entries = (. - subarch_entries) / 4
272 #endif /* CONFIG_PARAVIRT */
274 #ifdef CONFIG_HOTPLUG_CPU
276 * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
277 * up already except stack. We just set up stack here. Then call
281 movl stack_start, %ecx
288 * Non-boot CPU entry point; entered from trampoline.S
289 * We can't lgdt here, because lgdt itself uses a data segment, but
290 * we know the trampoline has already loaded the boot_gdt for us.
292 * If cpu hotplug is not supported then this code can go in init section
293 * which will be freed later
295 ENTRY(startup_32_smp)
297 movl $(__BOOT_DS),%eax
302 movl pa(stack_start),%ecx
304 leal -__PAGE_OFFSET(%ecx),%esp
306 #ifdef CONFIG_MICROCODE_EARLY
307 /* Early load ucode on AP. */
313 #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
314 X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
316 movl $(CR0_STATE & ~X86_CR0_PG),%eax
320 * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave
321 * bits like NT set. This would confuse the debugger if this code is traced. So
322 * initialize them properly now before switching to protected mode. That means
323 * DF in particular (even though we have cleared it earlier after copying the
324 * command line) because GCC expects it.
330 * New page tables may be in 4Mbyte page mode and may be using the global pages.
332 * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists
333 * if and only if CPUID exists and has flags other than the FPU flag set.
335 movl $-1,pa(X86_CPUID) # preset CPUID level
336 movl $X86_EFLAGS_ID,%ecx
338 popfl # set EFLAGS=ID
340 popl %eax # get EFLAGS
341 testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set?
342 jz enable_paging # hw disallowed setting of ID bit
343 # which means no CPUID and no CR4
347 movl %eax,pa(X86_CPUID) # save largest std CPUID function
351 andl $~1,%edx # Ignore CPUID.FPU
352 jz enable_paging # No flags or only CPUID.FPU = no CR4
354 movl pa(mmu_cr4_features),%eax
357 testb $X86_CR4_PAE, %al # check if PAE is enabled
360 /* Check if extended functions are implemented */
361 movl $0x80000000, %eax
363 /* Value must be in the range 0x80000001 to 0x8000ffff */
364 subl $0x80000001, %eax
365 cmpl $(0x8000ffff-0x80000001), %eax
368 /* Clear bogus XD_DISABLE bits */
371 mov $0x80000001, %eax
373 /* Execute Disable bit supported? */
374 btl $(X86_FEATURE_NX & 31), %edx
377 /* Setup EFER (Extended Feature Enable Register) */
382 /* Make changes effective */
390 movl $pa(initial_page_table), %eax
391 movl %eax,%cr3 /* set the page table pointer.. */
393 movl %eax,%cr0 /* ..and set paging (PG) bit */
394 ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */
396 /* Shift the stack pointer to a virtual address */
397 addl $__PAGE_OFFSET, %esp
400 * start system 32-bit setup. We need to re-do some of the things done
401 * in 16-bit mode for the "real" operations.
403 movl setup_once_ref,%eax
405 jz 1f # Did we do this already?
412 movb $4,X86 # at least 486
416 /* get vendor info */
417 xorl %eax,%eax # call CPUID with 0 -> return vendor ID
419 movl %eax,X86_CPUID # save CPUID level
420 movl %ebx,X86_VENDOR_ID # lo 4 chars
421 movl %edx,X86_VENDOR_ID+4 # next 4 chars
422 movl %ecx,X86_VENDOR_ID+8 # last 4 chars
424 orl %eax,%eax # do we have processor info as well?
427 movl $1,%eax # Use the CPUID instruction to get CPU type
429 movb %al,%cl # save reg for future use
430 andb $0x0f,%ah # mask processor family
432 andb $0xf0,%al # mask model
435 andb $0x0f,%cl # mask mask revision
437 movl %edx,X86_CAPABILITY
440 movl $0x50022,%ecx # set AM, WP, NE and MP
442 andl $0x80000011,%eax # Save PG,PE,ET
448 ljmp $(__KERNEL_CS),$1f
449 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
450 movl %eax,%ss # after changing gdt.
452 movl $(__USER_DS),%eax # DS/ES contains default USER segment
456 movl $(__KERNEL_PERCPU), %eax
457 movl %eax,%fs # set this cpu's percpu
459 movl $(__KERNEL_STACK_CANARY),%eax
462 xorl %eax,%eax # Clear LDT
465 pushl $0 # fake return address for unwinder
468 #include "verify_cpu.S"
473 * The setup work we only want to run on the BSP.
475 * Warning: %esi is live across this function.
480 * Set up a idt with 256 entries pointing to ignore_int,
481 * interrupt gates. It doesn't actually load idt - that needs
482 * to be done on each CPU. Interrupts are enabled elsewhere,
483 * when we can be relatively sure everything is ok.
487 movl $early_idt_handlers,%eax
488 movl $NUM_EXCEPTION_VECTORS,%ecx
492 /* interrupt gate, dpl=0, present */
493 movl $(0x8E000000 + __KERNEL_CS),2(%edi)
498 movl $256 - NUM_EXCEPTION_VECTORS,%ecx
499 movl $ignore_int,%edx
500 movl $(__KERNEL_CS << 16),%eax
501 movw %dx,%ax /* selector = 0x0010 = cs */
502 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
509 #ifdef CONFIG_CC_STACKPROTECTOR
511 * Configure the stack canary. The linker can't handle this by
512 * relocation. Manually set base address in stack canary
513 * segment descriptor.
516 movl $stack_canary,%ecx
517 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
519 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
520 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
523 andl $0,setup_once_ref /* Once is enough, thanks */
526 ENTRY(early_idt_handlers)
530 # 24(%rsp) error code
532 .rept NUM_EXCEPTION_VECTORS
533 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
536 pushl $0 # Dummy error code, to make stack frame uniform
538 pushl $i # 20(%esp) Vector number
539 jmp early_idt_handler
542 ENDPROC(early_idt_handlers)
544 /* This is global to keep gas from relaxing the jumps */
545 ENTRY(early_idt_handler)
548 cmpl $2,(%esp) # X86_TRAP_NMI
549 je is_nmi # Ignore NMI
551 cmpl $2,%ss:early_recursion_flag
553 incl %ss:early_recursion_flag
560 movl $(__KERNEL_DS),%eax
564 cmpl $(__KERNEL_CS),32(%esp)
567 leal 28(%esp),%eax # Pointer to %eip
568 call early_fixup_exception
570 jnz ex_entry /* found an exception entry */
575 movw %ax,2(%esp) /* clean up the segment values on some cpus */
579 pushl %eax /* %esp before the exception */
586 pushl (20+6*4)(%esp) /* trapno */
601 decl %ss:early_recursion_flag
603 addl $8,%esp /* drop vector number and error code */
605 ENDPROC(early_idt_handler)
607 /* This is the default interrupt "handler" :-) */
617 movl $(__KERNEL_DS),%eax
620 cmpl $2,early_recursion_flag
622 incl early_recursion_flag
643 early_recursion_flag:
649 .long i386_start_kernel
650 ENTRY(setup_once_ref)
658 #ifdef CONFIG_X86_PAE
662 ENTRY(initial_page_table)
667 ENTRY(empty_zero_page)
669 ENTRY(swapper_pg_dir)
673 * This starts the data section.
675 #ifdef CONFIG_X86_PAE
677 /* Page-aligned for the benefit of paravirt? */
679 ENTRY(initial_page_table)
680 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
682 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
683 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
684 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
687 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
688 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
692 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
694 # error "Kernel PMDs should be 1, 2 or 3"
696 .align PAGE_SIZE /* needs to be page-sized too */
702 .long init_thread_union+THREAD_SIZE
706 .asciz "Unknown interrupt or fault at: %p %p %p\n"
710 .ascii "BUG: Int %d: CR2 %p\n"
711 /* regs pushed in early_idt_handler: */
712 .ascii " EDI %p ESI %p EBP %p EBX %p\n"
713 .ascii " ESP %p ES %p DS %p\n"
714 .ascii " EDX %p ECX %p EAX %p\n"
716 .ascii " vec %p err %p EIP %p CS %p flg %p\n"
717 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
718 .ascii " %p %p %p %p %p %p %p %p\n"
719 .asciz " %p %p %p %p %p %p %p %p\n"
721 #include "../../x86/xen/xen-head.S"
724 * The IDT and GDT 'descriptors' are a strange 48-bit object
725 * only used by the lidt and lgdt instructions. They are not
726 * like usual segment descriptors - they consist of a 16-bit
727 * segment size, and 32-bit linear address value:
731 .globl boot_gdt_descr
735 # early boot GDT descriptor (must use 1:1 address mapping)
736 .word 0 # 32 bit align gdt_desc.address
739 .long boot_gdt - __PAGE_OFFSET
741 .word 0 # 32-bit align idt_desc.address
743 .word IDT_ENTRIES*8-1 # idt contains 256 entries
746 # boot GDT descriptor (later on used by CPU#0):
747 .word 0 # 32 bit align gdt_desc.address
748 ENTRY(early_gdt_descr)
749 .word GDT_ENTRIES*8-1
750 .long gdt_page /* Overwritten for secondary CPUs */
753 * The boot_gdt must mirror the equivalent in setup.S and is
754 * used only for booting.
756 .align L1_CACHE_BYTES
758 .fill GDT_ENTRY_BOOT_CS,8,0
759 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
760 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */