head.S 8.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379
  1. /*
  2. * linux/arch/x86_64/kernel/head.S -- start in 32bit and switch to 64bit
  3. *
  4. * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
  5. * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
  6. * Copyright (C) 2000 Karsten Keil <kkeil@suse.de>
  7. * Copyright (C) 2001,2002 Andi Kleen <ak@suse.de>
  8. *
  9. * $Id: head.S,v 1.49 2002/03/19 17:39:25 ak Exp $
  10. */
  11. #include <linux/linkage.h>
  12. #include <linux/threads.h>
  13. #include <linux/init.h>
  14. #include <asm/desc.h>
  15. #include <asm/segment.h>
  16. #include <asm/page.h>
  17. #include <asm/msr.h>
  18. #include <asm/cache.h>
  19. /* we are not able to switch in one step to the final KERNEL ADRESS SPACE
  20. * because we need identity-mapped pages on setup so define __START_KERNEL to
  21. * 0x100000 for this stage
  22. *
  23. */
  24. .text
  25. .code32
  26. .globl startup_32
  27. /* %bx: 1 if coming from smp trampoline on secondary cpu */
  28. startup_32:
  29. /*
  30. * At this point the CPU runs in 32bit protected mode (CS.D = 1) with
  31. * paging disabled and the point of this file is to switch to 64bit
  32. * long mode with a kernel mapping for kerneland to jump into the
  33. * kernel virtual addresses.
  34. * There is no stack until we set one up.
  35. */
  36. /* Initialize the %ds segment register */
  37. movl $__KERNEL_DS,%eax
  38. movl %eax,%ds
  39. /* Load new GDT with the 64bit segments using 32bit descriptor */
  40. lgdt pGDT32 - __START_KERNEL_map
  41. /* If the CPU doesn't support CPUID this will double fault.
  42. * Unfortunately it is hard to check for CPUID without a stack.
  43. */
  44. /* Check if extended functions are implemented */
  45. movl $0x80000000, %eax
  46. cpuid
  47. cmpl $0x80000000, %eax
  48. jbe no_long_mode
  49. /* Check if long mode is implemented */
  50. mov $0x80000001, %eax
  51. cpuid
  52. btl $29, %edx
  53. jnc no_long_mode
  54. /*
  55. * Prepare for entering 64bits mode
  56. */
  57. /* Enable PAE mode */
  58. xorl %eax, %eax
  59. btsl $5, %eax
  60. movl %eax, %cr4
  61. /* Setup early boot stage 4 level pagetables */
  62. movl $(boot_level4_pgt - __START_KERNEL_map), %eax
  63. movl %eax, %cr3
  64. /* Setup EFER (Extended Feature Enable Register) */
  65. movl $MSR_EFER, %ecx
  66. rdmsr
  67. /* Enable Long Mode */
  68. btsl $_EFER_LME, %eax
  69. /* Make changes effective */
  70. wrmsr
  71. xorl %eax, %eax
  72. btsl $31, %eax /* Enable paging and in turn activate Long Mode */
  73. btsl $0, %eax /* Enable protected mode */
  74. /* Make changes effective */
  75. movl %eax, %cr0
  76. /*
  77. * At this point we're in long mode but in 32bit compatibility mode
  78. * with EFER.LME = 1, CS.L = 0, CS.D = 1 (and in turn
  79. * EFER.LMA = 1). Now we want to jump in 64bit mode, to do that we use
  80. * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
  81. */
  82. ljmp $__KERNEL_CS, $(startup_64 - __START_KERNEL_map)
  83. .code64
  84. .org 0x100
  85. .globl startup_64
  86. startup_64:
  87. /* We come here either from startup_32
  88. * or directly from a 64bit bootloader.
  89. * Since we may have come directly from a bootloader we
  90. * reload the page tables here.
  91. */
  92. /* Enable PAE mode and PGE */
  93. xorq %rax, %rax
  94. btsq $5, %rax
  95. btsq $7, %rax
  96. movq %rax, %cr4
  97. /* Setup early boot stage 4 level pagetables. */
  98. movq $(boot_level4_pgt - __START_KERNEL_map), %rax
  99. movq %rax, %cr3
  100. /* Check if nx is implemented */
  101. movl $0x80000001, %eax
  102. cpuid
  103. movl %edx,%edi
  104. /* Setup EFER (Extended Feature Enable Register) */
  105. movl $MSR_EFER, %ecx
  106. rdmsr
  107. /* Enable System Call */
  108. btsl $_EFER_SCE, %eax
  109. /* No Execute supported? */
  110. btl $20,%edi
  111. jnc 1f
  112. btsl $_EFER_NX, %eax
  113. 1:
  114. /* Make changes effective */
  115. wrmsr
  116. /* Setup cr0 */
  117. #define CR0_PM 1 /* protected mode */
  118. #define CR0_MP (1<<1)
  119. #define CR0_ET (1<<4)
  120. #define CR0_NE (1<<5)
  121. #define CR0_WP (1<<16)
  122. #define CR0_AM (1<<18)
  123. #define CR0_PAGING (1<<31)
  124. movl $CR0_PM|CR0_MP|CR0_ET|CR0_NE|CR0_WP|CR0_AM|CR0_PAGING,%eax
  125. /* Make changes effective */
  126. movq %rax, %cr0
  127. /* Setup a boot time stack */
  128. movq init_rsp(%rip),%rsp
  129. /* zero EFLAGS after setting rsp */
  130. pushq $0
  131. popfq
  132. /*
  133. * We must switch to a new descriptor in kernel space for the GDT
  134. * because soon the kernel won't have access anymore to the userspace
  135. * addresses where we're currently running on. We have to do that here
  136. * because in 32bit we couldn't load a 64bit linear address.
  137. */
  138. lgdt cpu_gdt_descr
  139. /*
  140. * Setup up a dummy PDA. this is just for some early bootup code
  141. * that does in_interrupt()
  142. */
  143. movl $MSR_GS_BASE,%ecx
  144. movq $empty_zero_page,%rax
  145. movq %rax,%rdx
  146. shrq $32,%rdx
  147. wrmsr
  148. /* set up data segments. actually 0 would do too */
  149. movl $__KERNEL_DS,%eax
  150. movl %eax,%ds
  151. movl %eax,%ss
  152. movl %eax,%es
  153. /* esi is pointer to real mode structure with interesting info.
  154. pass it to C */
  155. movl %esi, %edi
  156. /* Finally jump to run C code and to be on real kernel address
  157. * Since we are running on identity-mapped space we have to jump
  158. * to the full 64bit address , this is only possible as indirect
  159. * jump
  160. */
  161. movq initial_code(%rip),%rax
  162. jmp *%rax
  163. /* SMP bootup changes these two */
  164. .globl initial_code
  165. initial_code:
  166. .quad x86_64_start_kernel
  167. .globl init_rsp
  168. init_rsp:
  169. .quad init_thread_union+THREAD_SIZE-8
  170. ENTRY(early_idt_handler)
  171. cmpl $2,early_recursion_flag(%rip)
  172. jz 1f
  173. incl early_recursion_flag(%rip)
  174. xorl %eax,%eax
  175. movq 8(%rsp),%rsi # get rip
  176. movq (%rsp),%rdx
  177. movq %cr2,%rcx
  178. leaq early_idt_msg(%rip),%rdi
  179. call early_printk
  180. cmpl $2,early_recursion_flag(%rip)
  181. jz 1f
  182. call dump_stack
  183. 1: hlt
  184. jmp 1b
  185. early_recursion_flag:
  186. .long 0
  187. early_idt_msg:
  188. .asciz "PANIC: early exception rip %lx error %lx cr2 %lx\n"
  189. .code32
  190. ENTRY(no_long_mode)
  191. /* This isn't an x86-64 CPU so hang */
  192. 1:
  193. jmp 1b
  194. .org 0xf00
  195. .globl pGDT32
  196. pGDT32:
  197. .word gdt_end-cpu_gdt_table
  198. .long cpu_gdt_table-__START_KERNEL_map
  199. .org 0xf10
  200. ljumpvector:
  201. .long startup_64-__START_KERNEL_map
  202. .word __KERNEL_CS
  203. ENTRY(stext)
  204. ENTRY(_stext)
  205. $page = 0
  206. #define NEXT_PAGE(name) \
  207. $page = $page + 1; \
  208. .org $page * 0x1000; \
  209. phys_/**/name = $page * 0x1000 + __PHYSICAL_START; \
  210. ENTRY(name)
  211. NEXT_PAGE(init_level4_pgt)
  212. /* This gets initialized in x86_64_start_kernel */
  213. .fill 512,8,0
  214. NEXT_PAGE(level3_ident_pgt)
  215. .quad phys_level2_ident_pgt | 0x007
  216. .fill 511,8,0
  217. NEXT_PAGE(level3_kernel_pgt)
  218. .fill 510,8,0
  219. /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
  220. .quad phys_level2_kernel_pgt | 0x007
  221. .fill 1,8,0
  222. NEXT_PAGE(level2_ident_pgt)
  223. /* 40MB for bootup. */
  224. i = 0
  225. .rept 20
  226. .quad i << 21 | 0x083
  227. i = i + 1
  228. .endr
  229. /* Temporary mappings for the super early allocator in arch/x86_64/mm/init.c */
  230. .globl temp_boot_pmds
  231. temp_boot_pmds:
  232. .fill 492,8,0
  233. NEXT_PAGE(level2_kernel_pgt)
  234. /* 40MB kernel mapping. The kernel code cannot be bigger than that.
  235. When you change this change KERNEL_TEXT_SIZE in page.h too. */
  236. /* (2^48-(2*1024*1024*1024)-((2^39)*511)-((2^30)*510)) = 0 */
  237. i = 0
  238. .rept 20
  239. .quad i << 21 | 0x183
  240. i = i + 1
  241. .endr
  242. /* Module mapping starts here */
  243. .fill 492,8,0
  244. NEXT_PAGE(empty_zero_page)
  245. NEXT_PAGE(level3_physmem_pgt)
  246. .quad phys_level2_kernel_pgt | 0x007 /* so that __va works even before pagetable_init */
  247. .fill 511,8,0
  248. #undef NEXT_PAGE
  249. .data
  250. #ifdef CONFIG_ACPI_SLEEP
  251. .align PAGE_SIZE
  252. ENTRY(wakeup_level4_pgt)
  253. .quad phys_level3_ident_pgt | 0x007
  254. .fill 255,8,0
  255. .quad phys_level3_physmem_pgt | 0x007
  256. .fill 254,8,0
  257. /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
  258. .quad phys_level3_kernel_pgt | 0x007
  259. #endif
  260. #ifndef CONFIG_HOTPLUG_CPU
  261. __INITDATA
  262. #endif
  263. /*
  264. * This default setting generates an ident mapping at address 0x100000
  265. * and a mapping for the kernel that precisely maps virtual address
  266. * 0xffffffff80000000 to physical address 0x000000. (always using
  267. * 2Mbyte large pages provided by PAE mode)
  268. */
  269. .align PAGE_SIZE
  270. ENTRY(boot_level4_pgt)
  271. .quad phys_level3_ident_pgt | 0x007
  272. .fill 255,8,0
  273. .quad phys_level3_physmem_pgt | 0x007
  274. .fill 254,8,0
  275. /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
  276. .quad phys_level3_kernel_pgt | 0x007
  277. .data
  278. .align 16
  279. .globl cpu_gdt_descr
  280. cpu_gdt_descr:
  281. .word gdt_end-cpu_gdt_table
  282. gdt:
  283. .quad cpu_gdt_table
  284. #ifdef CONFIG_SMP
  285. .rept NR_CPUS-1
  286. .word 0
  287. .quad 0
  288. .endr
  289. #endif
  290. /* We need valid kernel segments for data and code in long mode too
  291. * IRET will check the segment types kkeil 2000/10/28
  292. * Also sysret mandates a special GDT layout
  293. */
  294. .align PAGE_SIZE
  295. /* The TLS descriptors are currently at a different place compared to i386.
  296. Hopefully nobody expects them at a fixed place (Wine?) */
  297. ENTRY(cpu_gdt_table)
  298. .quad 0x0000000000000000 /* NULL descriptor */
  299. .quad 0x0 /* unused */
  300. .quad 0x00af9a000000ffff /* __KERNEL_CS */
  301. .quad 0x00cf92000000ffff /* __KERNEL_DS */
  302. .quad 0x00cffa000000ffff /* __USER32_CS */
  303. .quad 0x00cff2000000ffff /* __USER_DS, __USER32_DS */
  304. .quad 0x00affa000000ffff /* __USER_CS */
  305. .quad 0x00cf9a000000ffff /* __KERNEL32_CS */
  306. .quad 0,0 /* TSS */
  307. .quad 0,0 /* LDT */
  308. .quad 0,0,0 /* three TLS descriptors */
  309. .quad 0 /* unused */
  310. gdt_end:
  311. /* asm/segment.h:GDT_ENTRIES must match this */
  312. /* This should be a multiple of the cache line size */
  313. /* GDTs of other CPUs are now dynamically allocated */
  314. /* zero the remaining page */
  315. .fill PAGE_SIZE / 8 - GDT_ENTRIES,8,0
  316. ENTRY(idt_table)
  317. .rept 256
  318. .quad 0
  319. .quad 0
  320. .endr