head_64.S 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420
  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. * Copyright (C) 2005 Eric Biederman <ebiederm@xmission.com>
  9. */
  10. #include <linux/linkage.h>
  11. #include <linux/threads.h>
  12. #include <linux/init.h>
  13. #include <asm/desc.h>
  14. #include <asm/segment.h>
  15. #include <asm/pgtable.h>
  16. #include <asm/page.h>
  17. #include <asm/msr.h>
  18. #include <asm/cache.h>
  19. #include <asm/processor-flags.h>
  20. #ifdef CONFIG_PARAVIRT
  21. #include <asm/asm-offsets.h>
  22. #include <asm/paravirt.h>
  23. #else
  24. #define GET_CR2_INTO_RCX movq %cr2, %rcx
  25. #endif
  26. /* we are not able to switch in one step to the final KERNEL ADRESS SPACE
  27. * because we need identity-mapped pages.
  28. *
  29. */
  30. #define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
  31. L4_PAGE_OFFSET = pgd_index(__PAGE_OFFSET)
  32. L3_PAGE_OFFSET = pud_index(__PAGE_OFFSET)
  33. L4_START_KERNEL = pgd_index(__START_KERNEL_map)
  34. L3_START_KERNEL = pud_index(__START_KERNEL_map)
  35. .text
  36. .section .text.head
  37. .code64
  38. .globl startup_64
  39. startup_64:
  40. /*
  41. * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
  42. * and someone has loaded an identity mapped page table
  43. * for us. These identity mapped page tables map all of the
  44. * kernel pages and possibly all of memory.
  45. *
  46. * %esi holds a physical pointer to real_mode_data.
  47. *
  48. * We come here either directly from a 64bit bootloader, or from
  49. * arch/x86_64/boot/compressed/head.S.
  50. *
  51. * We only come here initially at boot nothing else comes here.
  52. *
  53. * Since we may be loaded at an address different from what we were
  54. * compiled to run at we first fixup the physical addresses in our page
  55. * tables and then reload them.
  56. */
  57. /* Compute the delta between the address I am compiled to run at and the
  58. * address I am actually running at.
  59. */
  60. leaq _text(%rip), %rbp
  61. subq $_text - __START_KERNEL_map, %rbp
  62. /* Is the address not 2M aligned? */
  63. movq %rbp, %rax
  64. andl $~PMD_PAGE_MASK, %eax
  65. testl %eax, %eax
  66. jnz bad_address
  67. /* Is the address too large? */
  68. leaq _text(%rip), %rdx
  69. movq $PGDIR_SIZE, %rax
  70. cmpq %rax, %rdx
  71. jae bad_address
  72. /* Fixup the physical addresses in the page table
  73. */
  74. addq %rbp, init_level4_pgt + 0(%rip)
  75. addq %rbp, init_level4_pgt + (L4_PAGE_OFFSET*8)(%rip)
  76. addq %rbp, init_level4_pgt + (L4_START_KERNEL*8)(%rip)
  77. addq %rbp, level3_ident_pgt + 0(%rip)
  78. addq %rbp, level3_kernel_pgt + (510*8)(%rip)
  79. addq %rbp, level3_kernel_pgt + (511*8)(%rip)
  80. addq %rbp, level2_fixmap_pgt + (506*8)(%rip)
  81. /* Add an Identity mapping if I am above 1G */
  82. leaq _text(%rip), %rdi
  83. andq $PMD_PAGE_MASK, %rdi
  84. movq %rdi, %rax
  85. shrq $PUD_SHIFT, %rax
  86. andq $(PTRS_PER_PUD - 1), %rax
  87. jz ident_complete
  88. leaq (level2_spare_pgt - __START_KERNEL_map + _KERNPG_TABLE)(%rbp), %rdx
  89. leaq level3_ident_pgt(%rip), %rbx
  90. movq %rdx, 0(%rbx, %rax, 8)
  91. movq %rdi, %rax
  92. shrq $PMD_SHIFT, %rax
  93. andq $(PTRS_PER_PMD - 1), %rax
  94. leaq __PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
  95. leaq level2_spare_pgt(%rip), %rbx
  96. movq %rdx, 0(%rbx, %rax, 8)
  97. ident_complete:
  98. /*
  99. * Fixup the kernel text+data virtual addresses. Note that
  100. * we might write invalid pmds, when the kernel is relocated
  101. * cleanup_highmap() fixes this up along with the mappings
  102. * beyond _end.
  103. */
  104. leaq level2_kernel_pgt(%rip), %rdi
  105. leaq 4096(%rdi), %r8
  106. /* See if it is a valid page table entry */
  107. 1: testq $1, 0(%rdi)
  108. jz 2f
  109. addq %rbp, 0(%rdi)
  110. /* Go to the next page */
  111. 2: addq $8, %rdi
  112. cmp %r8, %rdi
  113. jne 1b
  114. /* Fixup phys_base */
  115. addq %rbp, phys_base(%rip)
  116. #ifdef CONFIG_X86_TRAMPOLINE
  117. addq %rbp, trampoline_level4_pgt + 0(%rip)
  118. addq %rbp, trampoline_level4_pgt + (511*8)(%rip)
  119. #endif
  120. /* Due to ENTRY(), sometimes the empty space gets filled with
  121. * zeros. Better take a jmp than relying on empty space being
  122. * filled with 0x90 (nop)
  123. */
  124. jmp secondary_startup_64
  125. ENTRY(secondary_startup_64)
  126. /*
  127. * At this point the CPU runs in 64bit mode CS.L = 1 CS.D = 1,
  128. * and someone has loaded a mapped page table.
  129. *
  130. * %esi holds a physical pointer to real_mode_data.
  131. *
  132. * We come here either from startup_64 (using physical addresses)
  133. * or from trampoline.S (using virtual addresses).
  134. *
  135. * Using virtual addresses from trampoline.S removes the need
  136. * to have any identity mapped pages in the kernel page table
  137. * after the boot processor executes this code.
  138. */
  139. /* Enable PAE mode and PGE */
  140. movl $(X86_CR4_PAE | X86_CR4_PGE), %eax
  141. movq %rax, %cr4
  142. /* Setup early boot stage 4 level pagetables. */
  143. movq $(init_level4_pgt - __START_KERNEL_map), %rax
  144. addq phys_base(%rip), %rax
  145. movq %rax, %cr3
  146. /* Ensure I am executing from virtual addresses */
  147. movq $1f, %rax
  148. jmp *%rax
  149. 1:
  150. /* Check if nx is implemented */
  151. movl $0x80000001, %eax
  152. cpuid
  153. movl %edx,%edi
  154. /* Setup EFER (Extended Feature Enable Register) */
  155. movl $MSR_EFER, %ecx
  156. rdmsr
  157. btsl $_EFER_SCE, %eax /* Enable System Call */
  158. btl $20,%edi /* No Execute supported? */
  159. jnc 1f
  160. btsl $_EFER_NX, %eax
  161. 1: wrmsr /* Make changes effective */
  162. /* Setup cr0 */
  163. #define CR0_STATE (X86_CR0_PE | X86_CR0_MP | X86_CR0_ET | \
  164. X86_CR0_NE | X86_CR0_WP | X86_CR0_AM | \
  165. X86_CR0_PG)
  166. movl $CR0_STATE, %eax
  167. /* Make changes effective */
  168. movq %rax, %cr0
  169. /* Setup a boot time stack */
  170. movq stack_start(%rip),%rsp
  171. /* zero EFLAGS after setting rsp */
  172. pushq $0
  173. popfq
  174. /*
  175. * We must switch to a new descriptor in kernel space for the GDT
  176. * because soon the kernel won't have access anymore to the userspace
  177. * addresses where we're currently running on. We have to do that here
  178. * because in 32bit we couldn't load a 64bit linear address.
  179. */
  180. lgdt early_gdt_descr(%rip)
  181. /* set up data segments. actually 0 would do too */
  182. movl $__KERNEL_DS,%eax
  183. movl %eax,%ds
  184. movl %eax,%ss
  185. movl %eax,%es
  186. /*
  187. * We don't really need to load %fs or %gs, but load them anyway
  188. * to kill any stale realmode selectors. This allows execution
  189. * under VT hardware.
  190. */
  191. movl %eax,%fs
  192. movl %eax,%gs
  193. /*
  194. * Setup up a dummy PDA. this is just for some early bootup code
  195. * that does in_interrupt()
  196. */
  197. movl $MSR_GS_BASE,%ecx
  198. movq $empty_zero_page,%rax
  199. movq %rax,%rdx
  200. shrq $32,%rdx
  201. wrmsr
  202. /* esi is pointer to real mode structure with interesting info.
  203. pass it to C */
  204. movl %esi, %edi
  205. /* Finally jump to run C code and to be on real kernel address
  206. * Since we are running on identity-mapped space we have to jump
  207. * to the full 64bit address, this is only possible as indirect
  208. * jump. In addition we need to ensure %cs is set so we make this
  209. * a far return.
  210. */
  211. movq initial_code(%rip),%rax
  212. pushq $0 # fake return address to stop unwinder
  213. pushq $__KERNEL_CS # set correct cs
  214. pushq %rax # target address in negative space
  215. lretq
  216. /* SMP bootup changes these two */
  217. __REFDATA
  218. .align 8
  219. ENTRY(initial_code)
  220. .quad x86_64_start_kernel
  221. __FINITDATA
  222. ENTRY(stack_start)
  223. .quad init_thread_union+THREAD_SIZE-8
  224. .word 0
  225. bad_address:
  226. jmp bad_address
  227. .section ".init.text","ax"
  228. #ifdef CONFIG_EARLY_PRINTK
  229. .globl early_idt_handlers
  230. early_idt_handlers:
  231. i = 0
  232. .rept NUM_EXCEPTION_VECTORS
  233. movl $i, %esi
  234. jmp early_idt_handler
  235. i = i + 1
  236. .endr
  237. #endif
  238. ENTRY(early_idt_handler)
  239. #ifdef CONFIG_EARLY_PRINTK
  240. cmpl $2,early_recursion_flag(%rip)
  241. jz 1f
  242. incl early_recursion_flag(%rip)
  243. GET_CR2_INTO_RCX
  244. movq %rcx,%r9
  245. xorl %r8d,%r8d # zero for error code
  246. movl %esi,%ecx # get vector number
  247. # Test %ecx against mask of vectors that push error code.
  248. cmpl $31,%ecx
  249. ja 0f
  250. movl $1,%eax
  251. salq %cl,%rax
  252. testl $0x27d00,%eax
  253. je 0f
  254. popq %r8 # get error code
  255. 0: movq 0(%rsp),%rcx # get ip
  256. movq 8(%rsp),%rdx # get cs
  257. xorl %eax,%eax
  258. leaq early_idt_msg(%rip),%rdi
  259. call early_printk
  260. cmpl $2,early_recursion_flag(%rip)
  261. jz 1f
  262. call dump_stack
  263. #ifdef CONFIG_KALLSYMS
  264. leaq early_idt_ripmsg(%rip),%rdi
  265. movq 8(%rsp),%rsi # get rip again
  266. call __print_symbol
  267. #endif
  268. #endif /* EARLY_PRINTK */
  269. 1: hlt
  270. jmp 1b
  271. #ifdef CONFIG_EARLY_PRINTK
  272. early_recursion_flag:
  273. .long 0
  274. early_idt_msg:
  275. .asciz "PANIC: early exception %02lx rip %lx:%lx error %lx cr2 %lx\n"
  276. early_idt_ripmsg:
  277. .asciz "RIP %s\n"
  278. #endif /* CONFIG_EARLY_PRINTK */
  279. .previous
  280. .balign PAGE_SIZE
  281. #define NEXT_PAGE(name) \
  282. .balign PAGE_SIZE; \
  283. ENTRY(name)
  284. /* Automate the creation of 1 to 1 mapping pmd entries */
  285. #define PMDS(START, PERM, COUNT) \
  286. i = 0 ; \
  287. .rept (COUNT) ; \
  288. .quad (START) + (i << PMD_SHIFT) + (PERM) ; \
  289. i = i + 1 ; \
  290. .endr
  291. /*
  292. * This default setting generates an ident mapping at address 0x100000
  293. * and a mapping for the kernel that precisely maps virtual address
  294. * 0xffffffff80000000 to physical address 0x000000. (always using
  295. * 2Mbyte large pages provided by PAE mode)
  296. */
  297. NEXT_PAGE(init_level4_pgt)
  298. .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
  299. .org init_level4_pgt + L4_PAGE_OFFSET*8, 0
  300. .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
  301. .org init_level4_pgt + L4_START_KERNEL*8, 0
  302. /* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
  303. .quad level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE
  304. NEXT_PAGE(level3_ident_pgt)
  305. .quad level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE
  306. .fill 511,8,0
  307. NEXT_PAGE(level3_kernel_pgt)
  308. .fill L3_START_KERNEL,8,0
  309. /* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
  310. .quad level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE
  311. .quad level2_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
  312. NEXT_PAGE(level2_fixmap_pgt)
  313. .fill 506,8,0
  314. .quad level1_fixmap_pgt - __START_KERNEL_map + _PAGE_TABLE
  315. /* 8MB reserved for vsyscalls + a 2MB hole = 4 + 1 entries */
  316. .fill 5,8,0
  317. NEXT_PAGE(level1_fixmap_pgt)
  318. .fill 512,8,0
  319. NEXT_PAGE(level2_ident_pgt)
  320. /* Since I easily can, map the first 1G.
  321. * Don't set NX because code runs from these pages.
  322. */
  323. PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
  324. NEXT_PAGE(level2_kernel_pgt)
  325. /*
  326. * 512 MB kernel mapping. We spend a full page on this pagetable
  327. * anyway.
  328. *
  329. * The kernel code+data+bss must not be bigger than that.
  330. *
  331. * (NOTE: at +512MB starts the module area, see MODULES_VADDR.
  332. * If you want to increase this then increase MODULES_VADDR
  333. * too.)
  334. */
  335. PMDS(0, __PAGE_KERNEL_LARGE_EXEC,
  336. KERNEL_IMAGE_SIZE/PMD_SIZE)
  337. NEXT_PAGE(level2_spare_pgt)
  338. .fill 512, 8, 0
  339. #undef PMDS
  340. #undef NEXT_PAGE
  341. .data
  342. .align 16
  343. .globl early_gdt_descr
  344. early_gdt_descr:
  345. .word GDT_ENTRIES*8-1
  346. .quad per_cpu__gdt_page
  347. ENTRY(phys_base)
  348. /* This must match the first entry in level2_kernel_pgt */
  349. .quad 0x0000000000000000
  350. #include "../../x86/xen/xen-head.S"
  351. .section .bss, "aw", @nobits
  352. .align L1_CACHE_BYTES
  353. ENTRY(idt_table)
  354. .skip 256 * 16
  355. .section .bss.page_aligned, "aw", @nobits
  356. .align PAGE_SIZE
  357. ENTRY(empty_zero_page)
  358. .skip PAGE_SIZE