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
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?
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
441 movl $0x50022,%ecx # set AM, WP, NE and MP
443 andl $0x80000011,%eax # Save PG,PE,ET
450 ljmp $(__KERNEL_CS),$1f
451 1: movl $(__KERNEL_DS),%eax # reload all the segment registers
452 movl %eax,%ss # after changing gdt.
454 movl $(__USER_DS),%eax # DS/ES contains default USER segment
458 movl $(__KERNEL_PERCPU), %eax
459 movl %eax,%fs # set this cpu's percpu
461 movl $(__KERNEL_STACK_CANARY),%eax
464 xorl %eax,%eax # Clear LDT
467 pushl $0 # fake return address for unwinder
471 * We depend on ET to be correct. This checks for 287/387.
474 movb $0,X86_HARD_MATH
480 movl %cr0,%eax /* no coprocessor: have to set bits */
481 xorl $4,%eax /* set EM */
485 1: movb $1,X86_HARD_MATH
486 .byte 0xDB,0xE4 /* fsetpm for 287, ignored by 387 */
490 #include "verify_cpu.S"
495 * The setup work we only want to run on the BSP.
497 * Warning: %esi is live across this function.
502 * Set up a idt with 256 entries pointing to ignore_int,
503 * interrupt gates. It doesn't actually load idt - that needs
504 * to be done on each CPU. Interrupts are enabled elsewhere,
505 * when we can be relatively sure everything is ok.
509 movl $early_idt_handlers,%eax
510 movl $NUM_EXCEPTION_VECTORS,%ecx
514 /* interrupt gate, dpl=0, present */
515 movl $(0x8E000000 + __KERNEL_CS),2(%edi)
520 movl $256 - NUM_EXCEPTION_VECTORS,%ecx
521 movl $ignore_int,%edx
522 movl $(__KERNEL_CS << 16),%eax
523 movw %dx,%ax /* selector = 0x0010 = cs */
524 movw $0x8E00,%dx /* interrupt gate - dpl=0, present */
531 #ifdef CONFIG_CC_STACKPROTECTOR
533 * Configure the stack canary. The linker can't handle this by
534 * relocation. Manually set base address in stack canary
535 * segment descriptor.
538 movl $stack_canary,%ecx
539 movw %cx, 8 * GDT_ENTRY_STACK_CANARY + 2(%eax)
541 movb %cl, 8 * GDT_ENTRY_STACK_CANARY + 4(%eax)
542 movb %ch, 8 * GDT_ENTRY_STACK_CANARY + 7(%eax)
545 andl $0,setup_once_ref /* Once is enough, thanks */
548 ENTRY(early_idt_handlers)
552 # 24(%rsp) error code
554 .rept NUM_EXCEPTION_VECTORS
555 .if (EXCEPTION_ERRCODE_MASK >> i) & 1
558 pushl $0 # Dummy error code, to make stack frame uniform
560 pushl $i # 20(%esp) Vector number
561 jmp early_idt_handler
564 ENDPROC(early_idt_handlers)
566 /* This is global to keep gas from relaxing the jumps */
567 ENTRY(early_idt_handler)
570 cmpl $2,(%esp) # X86_TRAP_NMI
571 je is_nmi # Ignore NMI
573 cmpl $2,%ss:early_recursion_flag
575 incl %ss:early_recursion_flag
582 movl $(__KERNEL_DS),%eax
586 cmpl $(__KERNEL_CS),32(%esp)
589 leal 28(%esp),%eax # Pointer to %eip
590 call early_fixup_exception
592 jnz ex_entry /* found an exception entry */
597 movw %ax,2(%esp) /* clean up the segment values on some cpus */
601 pushl %eax /* %esp before the exception */
608 pushl (20+6*4)(%esp) /* trapno */
623 decl %ss:early_recursion_flag
625 addl $8,%esp /* drop vector number and error code */
627 ENDPROC(early_idt_handler)
629 /* This is the default interrupt "handler" :-) */
639 movl $(__KERNEL_DS),%eax
642 cmpl $2,early_recursion_flag
644 incl early_recursion_flag
665 early_recursion_flag:
671 .long i386_start_kernel
672 ENTRY(setup_once_ref)
680 #ifdef CONFIG_X86_PAE
684 ENTRY(initial_page_table)
689 ENTRY(empty_zero_page)
691 ENTRY(swapper_pg_dir)
695 * This starts the data section.
697 #ifdef CONFIG_X86_PAE
699 /* Page-aligned for the benefit of paravirt? */
701 ENTRY(initial_page_table)
702 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */
704 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
705 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
706 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0
709 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
710 .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0
714 .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0
716 # error "Kernel PMDs should be 1, 2 or 3"
718 .align PAGE_SIZE /* needs to be page-sized too */
724 .long init_thread_union+THREAD_SIZE
728 .asciz "Unknown interrupt or fault at: %p %p %p\n"
732 .ascii "BUG: Int %d: CR2 %p\n"
733 /* regs pushed in early_idt_handler: */
734 .ascii " EDI %p ESI %p EBP %p EBX %p\n"
735 .ascii " ESP %p ES %p DS %p\n"
736 .ascii " EDX %p ECX %p EAX %p\n"
738 .ascii " vec %p err %p EIP %p CS %p flg %p\n"
739 .ascii "Stack: %p %p %p %p %p %p %p %p\n"
740 .ascii " %p %p %p %p %p %p %p %p\n"
741 .asciz " %p %p %p %p %p %p %p %p\n"
743 #include "../../x86/xen/xen-head.S"
746 * The IDT and GDT 'descriptors' are a strange 48-bit object
747 * only used by the lidt and lgdt instructions. They are not
748 * like usual segment descriptors - they consist of a 16-bit
749 * segment size, and 32-bit linear address value:
753 .globl boot_gdt_descr
757 # early boot GDT descriptor (must use 1:1 address mapping)
758 .word 0 # 32 bit align gdt_desc.address
761 .long boot_gdt - __PAGE_OFFSET
763 .word 0 # 32-bit align idt_desc.address
765 .word IDT_ENTRIES*8-1 # idt contains 256 entries
768 # boot GDT descriptor (later on used by CPU#0):
769 .word 0 # 32 bit align gdt_desc.address
770 ENTRY(early_gdt_descr)
771 .word GDT_ENTRIES*8-1
772 .long gdt_page /* Overwritten for secondary CPUs */
775 * The boot_gdt must mirror the equivalent in setup.S and is
776 * used only for booting.
778 .align L1_CACHE_BYTES
780 .fill GDT_ENTRY_BOOT_CS,8,0
781 .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */
782 .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */