entry_32.S 34 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497
  1. /*
  2. *
  3. * Copyright (C) 1991, 1992 Linus Torvalds
  4. */
  5. /*
  6. * entry.S contains the system-call and fault low-level handling routines.
  7. * This also contains the timer-interrupt handler, as well as all interrupts
  8. * and faults that can result in a task-switch.
  9. *
  10. * NOTE: This code handles signal-recognition, which happens every time
  11. * after a timer-interrupt and after each system call.
  12. *
  13. * I changed all the .align's to 4 (16 byte alignment), as that's faster
  14. * on a 486.
  15. *
  16. * Stack layout in 'syscall_exit':
  17. * ptrace needs to have all regs on the stack.
  18. * if the order here is changed, it needs to be
  19. * updated in fork.c:copy_process, signal.c:do_signal,
  20. * ptrace.c and ptrace.h
  21. *
  22. * 0(%esp) - %ebx
  23. * 4(%esp) - %ecx
  24. * 8(%esp) - %edx
  25. * C(%esp) - %esi
  26. * 10(%esp) - %edi
  27. * 14(%esp) - %ebp
  28. * 18(%esp) - %eax
  29. * 1C(%esp) - %ds
  30. * 20(%esp) - %es
  31. * 24(%esp) - %fs
  32. * 28(%esp) - %gs saved iff !CONFIG_X86_32_LAZY_GS
  33. * 2C(%esp) - orig_eax
  34. * 30(%esp) - %eip
  35. * 34(%esp) - %cs
  36. * 38(%esp) - %eflags
  37. * 3C(%esp) - %oldesp
  38. * 40(%esp) - %oldss
  39. *
  40. * "current" is in register %ebx during any slow entries.
  41. */
  42. #include <linux/linkage.h>
  43. #include <linux/err.h>
  44. #include <asm/thread_info.h>
  45. #include <asm/irqflags.h>
  46. #include <asm/errno.h>
  47. #include <asm/segment.h>
  48. #include <asm/smp.h>
  49. #include <asm/page_types.h>
  50. #include <asm/percpu.h>
  51. #include <asm/dwarf2.h>
  52. #include <asm/processor-flags.h>
  53. #include <asm/ftrace.h>
  54. #include <asm/irq_vectors.h>
  55. #include <asm/cpufeature.h>
  56. #include <asm/alternative-asm.h>
  57. #include <asm/asm.h>
  58. #include <asm/smap.h>
  59. /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
  60. #include <linux/elf-em.h>
  61. #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
  62. #define __AUDIT_ARCH_LE 0x40000000
  63. #ifndef CONFIG_AUDITSYSCALL
  64. #define sysenter_audit syscall_trace_entry
  65. #define sysexit_audit syscall_exit_work
  66. #endif
  67. .section .entry.text, "ax"
  68. /*
  69. * We use macros for low-level operations which need to be overridden
  70. * for paravirtualization. The following will never clobber any registers:
  71. * INTERRUPT_RETURN (aka. "iret")
  72. * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  73. * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  74. *
  75. * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  76. * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  77. * Allowing a register to be clobbered can shrink the paravirt replacement
  78. * enough to patch inline, increasing performance.
  79. */
  80. #ifdef CONFIG_PREEMPT
  81. #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  82. #else
  83. #define preempt_stop(clobbers)
  84. #define resume_kernel restore_all
  85. #endif
  86. .macro TRACE_IRQS_IRET
  87. #ifdef CONFIG_TRACE_IRQFLAGS
  88. testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
  89. jz 1f
  90. TRACE_IRQS_ON
  91. 1:
  92. #endif
  93. .endm
  94. /*
  95. * User gs save/restore
  96. *
  97. * %gs is used for userland TLS and kernel only uses it for stack
  98. * canary which is required to be at %gs:20 by gcc. Read the comment
  99. * at the top of stackprotector.h for more info.
  100. *
  101. * Local labels 98 and 99 are used.
  102. */
  103. #ifdef CONFIG_X86_32_LAZY_GS
  104. /* unfortunately push/pop can't be no-op */
  105. .macro PUSH_GS
  106. pushl_cfi $0
  107. .endm
  108. .macro POP_GS pop=0
  109. addl $(4 + \pop), %esp
  110. CFI_ADJUST_CFA_OFFSET -(4 + \pop)
  111. .endm
  112. .macro POP_GS_EX
  113. .endm
  114. /* all the rest are no-op */
  115. .macro PTGS_TO_GS
  116. .endm
  117. .macro PTGS_TO_GS_EX
  118. .endm
  119. .macro GS_TO_REG reg
  120. .endm
  121. .macro REG_TO_PTGS reg
  122. .endm
  123. .macro SET_KERNEL_GS reg
  124. .endm
  125. #else /* CONFIG_X86_32_LAZY_GS */
  126. .macro PUSH_GS
  127. pushl_cfi %gs
  128. /*CFI_REL_OFFSET gs, 0*/
  129. .endm
  130. .macro POP_GS pop=0
  131. 98: popl_cfi %gs
  132. /*CFI_RESTORE gs*/
  133. .if \pop <> 0
  134. add $\pop, %esp
  135. CFI_ADJUST_CFA_OFFSET -\pop
  136. .endif
  137. .endm
  138. .macro POP_GS_EX
  139. .pushsection .fixup, "ax"
  140. 99: movl $0, (%esp)
  141. jmp 98b
  142. .popsection
  143. _ASM_EXTABLE(98b,99b)
  144. .endm
  145. .macro PTGS_TO_GS
  146. 98: mov PT_GS(%esp), %gs
  147. .endm
  148. .macro PTGS_TO_GS_EX
  149. .pushsection .fixup, "ax"
  150. 99: movl $0, PT_GS(%esp)
  151. jmp 98b
  152. .popsection
  153. _ASM_EXTABLE(98b,99b)
  154. .endm
  155. .macro GS_TO_REG reg
  156. movl %gs, \reg
  157. /*CFI_REGISTER gs, \reg*/
  158. .endm
  159. .macro REG_TO_PTGS reg
  160. movl \reg, PT_GS(%esp)
  161. /*CFI_REL_OFFSET gs, PT_GS*/
  162. .endm
  163. .macro SET_KERNEL_GS reg
  164. movl $(__KERNEL_STACK_CANARY), \reg
  165. movl \reg, %gs
  166. .endm
  167. #endif /* CONFIG_X86_32_LAZY_GS */
  168. .macro SAVE_ALL
  169. cld
  170. PUSH_GS
  171. pushl_cfi %fs
  172. /*CFI_REL_OFFSET fs, 0;*/
  173. pushl_cfi %es
  174. /*CFI_REL_OFFSET es, 0;*/
  175. pushl_cfi %ds
  176. /*CFI_REL_OFFSET ds, 0;*/
  177. pushl_cfi %eax
  178. CFI_REL_OFFSET eax, 0
  179. pushl_cfi %ebp
  180. CFI_REL_OFFSET ebp, 0
  181. pushl_cfi %edi
  182. CFI_REL_OFFSET edi, 0
  183. pushl_cfi %esi
  184. CFI_REL_OFFSET esi, 0
  185. pushl_cfi %edx
  186. CFI_REL_OFFSET edx, 0
  187. pushl_cfi %ecx
  188. CFI_REL_OFFSET ecx, 0
  189. pushl_cfi %ebx
  190. CFI_REL_OFFSET ebx, 0
  191. movl $(__USER_DS), %edx
  192. movl %edx, %ds
  193. movl %edx, %es
  194. movl $(__KERNEL_PERCPU), %edx
  195. movl %edx, %fs
  196. SET_KERNEL_GS %edx
  197. .endm
  198. .macro RESTORE_INT_REGS
  199. popl_cfi %ebx
  200. CFI_RESTORE ebx
  201. popl_cfi %ecx
  202. CFI_RESTORE ecx
  203. popl_cfi %edx
  204. CFI_RESTORE edx
  205. popl_cfi %esi
  206. CFI_RESTORE esi
  207. popl_cfi %edi
  208. CFI_RESTORE edi
  209. popl_cfi %ebp
  210. CFI_RESTORE ebp
  211. popl_cfi %eax
  212. CFI_RESTORE eax
  213. .endm
  214. .macro RESTORE_REGS pop=0
  215. RESTORE_INT_REGS
  216. 1: popl_cfi %ds
  217. /*CFI_RESTORE ds;*/
  218. 2: popl_cfi %es
  219. /*CFI_RESTORE es;*/
  220. 3: popl_cfi %fs
  221. /*CFI_RESTORE fs;*/
  222. POP_GS \pop
  223. .pushsection .fixup, "ax"
  224. 4: movl $0, (%esp)
  225. jmp 1b
  226. 5: movl $0, (%esp)
  227. jmp 2b
  228. 6: movl $0, (%esp)
  229. jmp 3b
  230. .popsection
  231. _ASM_EXTABLE(1b,4b)
  232. _ASM_EXTABLE(2b,5b)
  233. _ASM_EXTABLE(3b,6b)
  234. POP_GS_EX
  235. .endm
  236. .macro RING0_INT_FRAME
  237. CFI_STARTPROC simple
  238. CFI_SIGNAL_FRAME
  239. CFI_DEF_CFA esp, 3*4
  240. /*CFI_OFFSET cs, -2*4;*/
  241. CFI_OFFSET eip, -3*4
  242. .endm
  243. .macro RING0_EC_FRAME
  244. CFI_STARTPROC simple
  245. CFI_SIGNAL_FRAME
  246. CFI_DEF_CFA esp, 4*4
  247. /*CFI_OFFSET cs, -2*4;*/
  248. CFI_OFFSET eip, -3*4
  249. .endm
  250. .macro RING0_PTREGS_FRAME
  251. CFI_STARTPROC simple
  252. CFI_SIGNAL_FRAME
  253. CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
  254. /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
  255. CFI_OFFSET eip, PT_EIP-PT_OLDESP
  256. /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
  257. /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
  258. CFI_OFFSET eax, PT_EAX-PT_OLDESP
  259. CFI_OFFSET ebp, PT_EBP-PT_OLDESP
  260. CFI_OFFSET edi, PT_EDI-PT_OLDESP
  261. CFI_OFFSET esi, PT_ESI-PT_OLDESP
  262. CFI_OFFSET edx, PT_EDX-PT_OLDESP
  263. CFI_OFFSET ecx, PT_ECX-PT_OLDESP
  264. CFI_OFFSET ebx, PT_EBX-PT_OLDESP
  265. .endm
  266. ENTRY(ret_from_fork)
  267. CFI_STARTPROC
  268. pushl_cfi %eax
  269. call schedule_tail
  270. GET_THREAD_INFO(%ebp)
  271. popl_cfi %eax
  272. pushl_cfi $0x0202 # Reset kernel eflags
  273. popfl_cfi
  274. jmp syscall_exit
  275. CFI_ENDPROC
  276. END(ret_from_fork)
  277. /*
  278. * Interrupt exit functions should be protected against kprobes
  279. */
  280. .pushsection .kprobes.text, "ax"
  281. /*
  282. * Return to user mode is not as complex as all this looks,
  283. * but we want the default path for a system call return to
  284. * go as quickly as possible which is why some of this is
  285. * less clear than it otherwise should be.
  286. */
  287. # userspace resumption stub bypassing syscall exit tracing
  288. ALIGN
  289. RING0_PTREGS_FRAME
  290. ret_from_exception:
  291. preempt_stop(CLBR_ANY)
  292. ret_from_intr:
  293. GET_THREAD_INFO(%ebp)
  294. #ifdef CONFIG_VM86
  295. movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
  296. movb PT_CS(%esp), %al
  297. andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
  298. #else
  299. /*
  300. * We can be coming here from a syscall done in the kernel space,
  301. * e.g. a failed kernel_execve().
  302. */
  303. movl PT_CS(%esp), %eax
  304. andl $SEGMENT_RPL_MASK, %eax
  305. #endif
  306. cmpl $USER_RPL, %eax
  307. jb resume_kernel # not returning to v8086 or userspace
  308. ENTRY(resume_userspace)
  309. LOCKDEP_SYS_EXIT
  310. DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
  311. # setting need_resched or sigpending
  312. # between sampling and the iret
  313. TRACE_IRQS_OFF
  314. movl TI_flags(%ebp), %ecx
  315. andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
  316. # int/exception return?
  317. jne work_pending
  318. jmp restore_all
  319. END(ret_from_exception)
  320. #ifdef CONFIG_PREEMPT
  321. ENTRY(resume_kernel)
  322. DISABLE_INTERRUPTS(CLBR_ANY)
  323. cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
  324. jnz restore_all
  325. need_resched:
  326. movl TI_flags(%ebp), %ecx # need_resched set ?
  327. testb $_TIF_NEED_RESCHED, %cl
  328. jz restore_all
  329. testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
  330. jz restore_all
  331. call preempt_schedule_irq
  332. jmp need_resched
  333. END(resume_kernel)
  334. #endif
  335. CFI_ENDPROC
  336. /*
  337. * End of kprobes section
  338. */
  339. .popsection
  340. /* SYSENTER_RETURN points to after the "sysenter" instruction in
  341. the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
  342. # sysenter call handler stub
  343. ENTRY(ia32_sysenter_target)
  344. CFI_STARTPROC simple
  345. CFI_SIGNAL_FRAME
  346. CFI_DEF_CFA esp, 0
  347. CFI_REGISTER esp, ebp
  348. movl TSS_sysenter_sp0(%esp),%esp
  349. sysenter_past_esp:
  350. /*
  351. * Interrupts are disabled here, but we can't trace it until
  352. * enough kernel state to call TRACE_IRQS_OFF can be called - but
  353. * we immediately enable interrupts at that point anyway.
  354. */
  355. pushl_cfi $__USER_DS
  356. /*CFI_REL_OFFSET ss, 0*/
  357. pushl_cfi %ebp
  358. CFI_REL_OFFSET esp, 0
  359. pushfl_cfi
  360. orl $X86_EFLAGS_IF, (%esp)
  361. pushl_cfi $__USER_CS
  362. /*CFI_REL_OFFSET cs, 0*/
  363. /*
  364. * Push current_thread_info()->sysenter_return to the stack.
  365. * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
  366. * pushed above; +8 corresponds to copy_thread's esp0 setting.
  367. */
  368. pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
  369. CFI_REL_OFFSET eip, 0
  370. pushl_cfi %eax
  371. SAVE_ALL
  372. ENABLE_INTERRUPTS(CLBR_NONE)
  373. /*
  374. * Load the potential sixth argument from user stack.
  375. * Careful about security.
  376. */
  377. cmpl $__PAGE_OFFSET-3,%ebp
  378. jae syscall_fault
  379. ASM_STAC
  380. 1: movl (%ebp),%ebp
  381. ASM_CLAC
  382. movl %ebp,PT_EBP(%esp)
  383. _ASM_EXTABLE(1b,syscall_fault)
  384. GET_THREAD_INFO(%ebp)
  385. testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
  386. jnz sysenter_audit
  387. sysenter_do_call:
  388. cmpl $(NR_syscalls), %eax
  389. jae syscall_badsys
  390. call *sys_call_table(,%eax,4)
  391. movl %eax,PT_EAX(%esp)
  392. LOCKDEP_SYS_EXIT
  393. DISABLE_INTERRUPTS(CLBR_ANY)
  394. TRACE_IRQS_OFF
  395. movl TI_flags(%ebp), %ecx
  396. testl $_TIF_ALLWORK_MASK, %ecx
  397. jne sysexit_audit
  398. sysenter_exit:
  399. /* if something modifies registers it must also disable sysexit */
  400. movl PT_EIP(%esp), %edx
  401. movl PT_OLDESP(%esp), %ecx
  402. xorl %ebp,%ebp
  403. TRACE_IRQS_ON
  404. 1: mov PT_FS(%esp), %fs
  405. PTGS_TO_GS
  406. ENABLE_INTERRUPTS_SYSEXIT
  407. #ifdef CONFIG_AUDITSYSCALL
  408. sysenter_audit:
  409. testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
  410. jnz syscall_trace_entry
  411. addl $4,%esp
  412. CFI_ADJUST_CFA_OFFSET -4
  413. /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
  414. /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
  415. /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
  416. movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
  417. movl %eax,%edx /* 2nd arg: syscall number */
  418. movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
  419. call __audit_syscall_entry
  420. pushl_cfi %ebx
  421. movl PT_EAX(%esp),%eax /* reload syscall number */
  422. jmp sysenter_do_call
  423. sysexit_audit:
  424. testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
  425. jne syscall_exit_work
  426. TRACE_IRQS_ON
  427. ENABLE_INTERRUPTS(CLBR_ANY)
  428. movl %eax,%edx /* second arg, syscall return value */
  429. cmpl $-MAX_ERRNO,%eax /* is it an error ? */
  430. setbe %al /* 1 if so, 0 if not */
  431. movzbl %al,%eax /* zero-extend that */
  432. call __audit_syscall_exit
  433. DISABLE_INTERRUPTS(CLBR_ANY)
  434. TRACE_IRQS_OFF
  435. movl TI_flags(%ebp), %ecx
  436. testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
  437. jne syscall_exit_work
  438. movl PT_EAX(%esp),%eax /* reload syscall return value */
  439. jmp sysenter_exit
  440. #endif
  441. CFI_ENDPROC
  442. .pushsection .fixup,"ax"
  443. 2: movl $0,PT_FS(%esp)
  444. jmp 1b
  445. .popsection
  446. _ASM_EXTABLE(1b,2b)
  447. PTGS_TO_GS_EX
  448. ENDPROC(ia32_sysenter_target)
  449. /*
  450. * syscall stub including irq exit should be protected against kprobes
  451. */
  452. .pushsection .kprobes.text, "ax"
  453. # system call handler stub
  454. ENTRY(system_call)
  455. RING0_INT_FRAME # can't unwind into user space anyway
  456. ASM_CLAC
  457. pushl_cfi %eax # save orig_eax
  458. SAVE_ALL
  459. GET_THREAD_INFO(%ebp)
  460. # system call tracing in operation / emulation
  461. testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
  462. jnz syscall_trace_entry
  463. cmpl $(NR_syscalls), %eax
  464. jae syscall_badsys
  465. syscall_call:
  466. call *sys_call_table(,%eax,4)
  467. movl %eax,PT_EAX(%esp) # store the return value
  468. syscall_exit:
  469. LOCKDEP_SYS_EXIT
  470. DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
  471. # setting need_resched or sigpending
  472. # between sampling and the iret
  473. TRACE_IRQS_OFF
  474. movl TI_flags(%ebp), %ecx
  475. testl $_TIF_ALLWORK_MASK, %ecx # current->work
  476. jne syscall_exit_work
  477. restore_all:
  478. TRACE_IRQS_IRET
  479. restore_all_notrace:
  480. movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
  481. # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
  482. # are returning to the kernel.
  483. # See comments in process.c:copy_thread() for details.
  484. movb PT_OLDSS(%esp), %ah
  485. movb PT_CS(%esp), %al
  486. andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
  487. cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
  488. CFI_REMEMBER_STATE
  489. je ldt_ss # returning to user-space with LDT SS
  490. restore_nocheck:
  491. RESTORE_REGS 4 # skip orig_eax/error_code
  492. irq_return:
  493. INTERRUPT_RETURN
  494. .section .fixup,"ax"
  495. ENTRY(iret_exc)
  496. pushl $0 # no error code
  497. pushl $do_iret_error
  498. jmp error_code
  499. .previous
  500. _ASM_EXTABLE(irq_return,iret_exc)
  501. CFI_RESTORE_STATE
  502. ldt_ss:
  503. larl PT_OLDSS(%esp), %eax
  504. jnz restore_nocheck
  505. testl $0x00400000, %eax # returning to 32bit stack?
  506. jnz restore_nocheck # allright, normal return
  507. #ifdef CONFIG_PARAVIRT
  508. /*
  509. * The kernel can't run on a non-flat stack if paravirt mode
  510. * is active. Rather than try to fixup the high bits of
  511. * ESP, bypass this code entirely. This may break DOSemu
  512. * and/or Wine support in a paravirt VM, although the option
  513. * is still available to implement the setting of the high
  514. * 16-bits in the INTERRUPT_RETURN paravirt-op.
  515. */
  516. cmpl $0, pv_info+PARAVIRT_enabled
  517. jne restore_nocheck
  518. #endif
  519. /*
  520. * Setup and switch to ESPFIX stack
  521. *
  522. * We're returning to userspace with a 16 bit stack. The CPU will not
  523. * restore the high word of ESP for us on executing iret... This is an
  524. * "official" bug of all the x86-compatible CPUs, which we can work
  525. * around to make dosemu and wine happy. We do this by preloading the
  526. * high word of ESP with the high word of the userspace ESP while
  527. * compensating for the offset by changing to the ESPFIX segment with
  528. * a base address that matches for the difference.
  529. */
  530. #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
  531. mov %esp, %edx /* load kernel esp */
  532. mov PT_OLDESP(%esp), %eax /* load userspace esp */
  533. mov %dx, %ax /* eax: new kernel esp */
  534. sub %eax, %edx /* offset (low word is 0) */
  535. shr $16, %edx
  536. mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
  537. mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
  538. pushl_cfi $__ESPFIX_SS
  539. pushl_cfi %eax /* new kernel esp */
  540. /* Disable interrupts, but do not irqtrace this section: we
  541. * will soon execute iret and the tracer was already set to
  542. * the irqstate after the iret */
  543. DISABLE_INTERRUPTS(CLBR_EAX)
  544. lss (%esp), %esp /* switch to espfix segment */
  545. CFI_ADJUST_CFA_OFFSET -8
  546. jmp restore_nocheck
  547. CFI_ENDPROC
  548. ENDPROC(system_call)
  549. # perform work that needs to be done immediately before resumption
  550. ALIGN
  551. RING0_PTREGS_FRAME # can't unwind into user space anyway
  552. work_pending:
  553. testb $_TIF_NEED_RESCHED, %cl
  554. jz work_notifysig
  555. work_resched:
  556. call schedule
  557. LOCKDEP_SYS_EXIT
  558. DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
  559. # setting need_resched or sigpending
  560. # between sampling and the iret
  561. TRACE_IRQS_OFF
  562. movl TI_flags(%ebp), %ecx
  563. andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
  564. # than syscall tracing?
  565. jz restore_all
  566. testb $_TIF_NEED_RESCHED, %cl
  567. jnz work_resched
  568. work_notifysig: # deal with pending signals and
  569. # notify-resume requests
  570. #ifdef CONFIG_VM86
  571. testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
  572. movl %esp, %eax
  573. jne work_notifysig_v86 # returning to kernel-space or
  574. # vm86-space
  575. TRACE_IRQS_ON
  576. ENABLE_INTERRUPTS(CLBR_NONE)
  577. movb PT_CS(%esp), %bl
  578. andb $SEGMENT_RPL_MASK, %bl
  579. cmpb $USER_RPL, %bl
  580. jb resume_kernel
  581. xorl %edx, %edx
  582. call do_notify_resume
  583. jmp resume_userspace
  584. ALIGN
  585. work_notifysig_v86:
  586. pushl_cfi %ecx # save ti_flags for do_notify_resume
  587. call save_v86_state # %eax contains pt_regs pointer
  588. popl_cfi %ecx
  589. movl %eax, %esp
  590. #else
  591. movl %esp, %eax
  592. #endif
  593. TRACE_IRQS_ON
  594. ENABLE_INTERRUPTS(CLBR_NONE)
  595. movb PT_CS(%esp), %bl
  596. andb $SEGMENT_RPL_MASK, %bl
  597. cmpb $USER_RPL, %bl
  598. jb resume_kernel
  599. xorl %edx, %edx
  600. call do_notify_resume
  601. jmp resume_userspace
  602. END(work_pending)
  603. # perform syscall exit tracing
  604. ALIGN
  605. syscall_trace_entry:
  606. movl $-ENOSYS,PT_EAX(%esp)
  607. movl %esp, %eax
  608. call syscall_trace_enter
  609. /* What it returned is what we'll actually use. */
  610. cmpl $(NR_syscalls), %eax
  611. jnae syscall_call
  612. jmp syscall_exit
  613. END(syscall_trace_entry)
  614. # perform syscall exit tracing
  615. ALIGN
  616. syscall_exit_work:
  617. testl $_TIF_WORK_SYSCALL_EXIT, %ecx
  618. jz work_pending
  619. TRACE_IRQS_ON
  620. ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
  621. # schedule() instead
  622. movl %esp, %eax
  623. call syscall_trace_leave
  624. jmp resume_userspace
  625. END(syscall_exit_work)
  626. CFI_ENDPROC
  627. RING0_INT_FRAME # can't unwind into user space anyway
  628. syscall_fault:
  629. ASM_CLAC
  630. GET_THREAD_INFO(%ebp)
  631. movl $-EFAULT,PT_EAX(%esp)
  632. jmp resume_userspace
  633. END(syscall_fault)
  634. syscall_badsys:
  635. movl $-ENOSYS,PT_EAX(%esp)
  636. jmp resume_userspace
  637. END(syscall_badsys)
  638. CFI_ENDPROC
  639. /*
  640. * End of kprobes section
  641. */
  642. .popsection
  643. /*
  644. * System calls that need a pt_regs pointer.
  645. */
  646. #define PTREGSCALL0(name) \
  647. ENTRY(ptregs_##name) ; \
  648. leal 4(%esp),%eax; \
  649. jmp sys_##name; \
  650. ENDPROC(ptregs_##name)
  651. #define PTREGSCALL1(name) \
  652. ENTRY(ptregs_##name) ; \
  653. leal 4(%esp),%edx; \
  654. movl (PT_EBX+4)(%esp),%eax; \
  655. jmp sys_##name; \
  656. ENDPROC(ptregs_##name)
  657. #define PTREGSCALL2(name) \
  658. ENTRY(ptregs_##name) ; \
  659. leal 4(%esp),%ecx; \
  660. movl (PT_ECX+4)(%esp),%edx; \
  661. movl (PT_EBX+4)(%esp),%eax; \
  662. jmp sys_##name; \
  663. ENDPROC(ptregs_##name)
  664. #define PTREGSCALL3(name) \
  665. ENTRY(ptregs_##name) ; \
  666. CFI_STARTPROC; \
  667. leal 4(%esp),%eax; \
  668. pushl_cfi %eax; \
  669. movl PT_EDX(%eax),%ecx; \
  670. movl PT_ECX(%eax),%edx; \
  671. movl PT_EBX(%eax),%eax; \
  672. call sys_##name; \
  673. addl $4,%esp; \
  674. CFI_ADJUST_CFA_OFFSET -4; \
  675. ret; \
  676. CFI_ENDPROC; \
  677. ENDPROC(ptregs_##name)
  678. PTREGSCALL1(iopl)
  679. PTREGSCALL0(fork)
  680. PTREGSCALL0(vfork)
  681. PTREGSCALL3(execve)
  682. PTREGSCALL2(sigaltstack)
  683. PTREGSCALL0(sigreturn)
  684. PTREGSCALL0(rt_sigreturn)
  685. PTREGSCALL2(vm86)
  686. PTREGSCALL1(vm86old)
  687. /* Clone is an oddball. The 4th arg is in %edi */
  688. ENTRY(ptregs_clone)
  689. CFI_STARTPROC
  690. leal 4(%esp),%eax
  691. pushl_cfi %eax
  692. pushl_cfi PT_EDI(%eax)
  693. movl PT_EDX(%eax),%ecx
  694. movl PT_ECX(%eax),%edx
  695. movl PT_EBX(%eax),%eax
  696. call sys_clone
  697. addl $8,%esp
  698. CFI_ADJUST_CFA_OFFSET -8
  699. ret
  700. CFI_ENDPROC
  701. ENDPROC(ptregs_clone)
  702. .macro FIXUP_ESPFIX_STACK
  703. /*
  704. * Switch back for ESPFIX stack to the normal zerobased stack
  705. *
  706. * We can't call C functions using the ESPFIX stack. This code reads
  707. * the high word of the segment base from the GDT and swiches to the
  708. * normal stack and adjusts ESP with the matching offset.
  709. */
  710. /* fixup the stack */
  711. mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
  712. mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
  713. shl $16, %eax
  714. addl %esp, %eax /* the adjusted stack pointer */
  715. pushl_cfi $__KERNEL_DS
  716. pushl_cfi %eax
  717. lss (%esp), %esp /* switch to the normal stack segment */
  718. CFI_ADJUST_CFA_OFFSET -8
  719. .endm
  720. .macro UNWIND_ESPFIX_STACK
  721. movl %ss, %eax
  722. /* see if on espfix stack */
  723. cmpw $__ESPFIX_SS, %ax
  724. jne 27f
  725. movl $__KERNEL_DS, %eax
  726. movl %eax, %ds
  727. movl %eax, %es
  728. /* switch to normal stack */
  729. FIXUP_ESPFIX_STACK
  730. 27:
  731. .endm
  732. /*
  733. * Build the entry stubs and pointer table with some assembler magic.
  734. * We pack 7 stubs into a single 32-byte chunk, which will fit in a
  735. * single cache line on all modern x86 implementations.
  736. */
  737. .section .init.rodata,"a"
  738. ENTRY(interrupt)
  739. .section .entry.text, "ax"
  740. .p2align 5
  741. .p2align CONFIG_X86_L1_CACHE_SHIFT
  742. ENTRY(irq_entries_start)
  743. RING0_INT_FRAME
  744. vector=FIRST_EXTERNAL_VECTOR
  745. .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
  746. .balign 32
  747. .rept 7
  748. .if vector < NR_VECTORS
  749. .if vector <> FIRST_EXTERNAL_VECTOR
  750. CFI_ADJUST_CFA_OFFSET -4
  751. .endif
  752. 1: pushl_cfi $(~vector+0x80) /* Note: always in signed byte range */
  753. .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
  754. jmp 2f
  755. .endif
  756. .previous
  757. .long 1b
  758. .section .entry.text, "ax"
  759. vector=vector+1
  760. .endif
  761. .endr
  762. 2: jmp common_interrupt
  763. .endr
  764. END(irq_entries_start)
  765. .previous
  766. END(interrupt)
  767. .previous
  768. /*
  769. * the CPU automatically disables interrupts when executing an IRQ vector,
  770. * so IRQ-flags tracing has to follow that:
  771. */
  772. .p2align CONFIG_X86_L1_CACHE_SHIFT
  773. common_interrupt:
  774. ASM_CLAC
  775. addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
  776. SAVE_ALL
  777. TRACE_IRQS_OFF
  778. movl %esp,%eax
  779. call do_IRQ
  780. jmp ret_from_intr
  781. ENDPROC(common_interrupt)
  782. CFI_ENDPROC
  783. /*
  784. * Irq entries should be protected against kprobes
  785. */
  786. .pushsection .kprobes.text, "ax"
  787. #define BUILD_INTERRUPT3(name, nr, fn) \
  788. ENTRY(name) \
  789. RING0_INT_FRAME; \
  790. ASM_CLAC; \
  791. pushl_cfi $~(nr); \
  792. SAVE_ALL; \
  793. TRACE_IRQS_OFF \
  794. movl %esp,%eax; \
  795. call fn; \
  796. jmp ret_from_intr; \
  797. CFI_ENDPROC; \
  798. ENDPROC(name)
  799. #define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
  800. /* The include is where all of the SMP etc. interrupts come from */
  801. #include <asm/entry_arch.h>
  802. ENTRY(coprocessor_error)
  803. RING0_INT_FRAME
  804. ASM_CLAC
  805. pushl_cfi $0
  806. pushl_cfi $do_coprocessor_error
  807. jmp error_code
  808. CFI_ENDPROC
  809. END(coprocessor_error)
  810. ENTRY(simd_coprocessor_error)
  811. RING0_INT_FRAME
  812. ASM_CLAC
  813. pushl_cfi $0
  814. #ifdef CONFIG_X86_INVD_BUG
  815. /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
  816. 661: pushl_cfi $do_general_protection
  817. 662:
  818. .section .altinstructions,"a"
  819. altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
  820. .previous
  821. .section .altinstr_replacement,"ax"
  822. 663: pushl $do_simd_coprocessor_error
  823. 664:
  824. .previous
  825. #else
  826. pushl_cfi $do_simd_coprocessor_error
  827. #endif
  828. jmp error_code
  829. CFI_ENDPROC
  830. END(simd_coprocessor_error)
  831. ENTRY(device_not_available)
  832. RING0_INT_FRAME
  833. ASM_CLAC
  834. pushl_cfi $-1 # mark this as an int
  835. pushl_cfi $do_device_not_available
  836. jmp error_code
  837. CFI_ENDPROC
  838. END(device_not_available)
  839. #ifdef CONFIG_PARAVIRT
  840. ENTRY(native_iret)
  841. iret
  842. _ASM_EXTABLE(native_iret, iret_exc)
  843. END(native_iret)
  844. ENTRY(native_irq_enable_sysexit)
  845. sti
  846. sysexit
  847. END(native_irq_enable_sysexit)
  848. #endif
  849. ENTRY(overflow)
  850. RING0_INT_FRAME
  851. ASM_CLAC
  852. pushl_cfi $0
  853. pushl_cfi $do_overflow
  854. jmp error_code
  855. CFI_ENDPROC
  856. END(overflow)
  857. ENTRY(bounds)
  858. RING0_INT_FRAME
  859. ASM_CLAC
  860. pushl_cfi $0
  861. pushl_cfi $do_bounds
  862. jmp error_code
  863. CFI_ENDPROC
  864. END(bounds)
  865. ENTRY(invalid_op)
  866. RING0_INT_FRAME
  867. ASM_CLAC
  868. pushl_cfi $0
  869. pushl_cfi $do_invalid_op
  870. jmp error_code
  871. CFI_ENDPROC
  872. END(invalid_op)
  873. ENTRY(coprocessor_segment_overrun)
  874. RING0_INT_FRAME
  875. ASM_CLAC
  876. pushl_cfi $0
  877. pushl_cfi $do_coprocessor_segment_overrun
  878. jmp error_code
  879. CFI_ENDPROC
  880. END(coprocessor_segment_overrun)
  881. ENTRY(invalid_TSS)
  882. RING0_EC_FRAME
  883. ASM_CLAC
  884. pushl_cfi $do_invalid_TSS
  885. jmp error_code
  886. CFI_ENDPROC
  887. END(invalid_TSS)
  888. ENTRY(segment_not_present)
  889. RING0_EC_FRAME
  890. ASM_CLAC
  891. pushl_cfi $do_segment_not_present
  892. jmp error_code
  893. CFI_ENDPROC
  894. END(segment_not_present)
  895. ENTRY(stack_segment)
  896. RING0_EC_FRAME
  897. ASM_CLAC
  898. pushl_cfi $do_stack_segment
  899. jmp error_code
  900. CFI_ENDPROC
  901. END(stack_segment)
  902. ENTRY(alignment_check)
  903. RING0_EC_FRAME
  904. ASM_CLAC
  905. pushl_cfi $do_alignment_check
  906. jmp error_code
  907. CFI_ENDPROC
  908. END(alignment_check)
  909. ENTRY(divide_error)
  910. RING0_INT_FRAME
  911. ASM_CLAC
  912. pushl_cfi $0 # no error code
  913. pushl_cfi $do_divide_error
  914. jmp error_code
  915. CFI_ENDPROC
  916. END(divide_error)
  917. #ifdef CONFIG_X86_MCE
  918. ENTRY(machine_check)
  919. RING0_INT_FRAME
  920. ASM_CLAC
  921. pushl_cfi $0
  922. pushl_cfi machine_check_vector
  923. jmp error_code
  924. CFI_ENDPROC
  925. END(machine_check)
  926. #endif
  927. ENTRY(spurious_interrupt_bug)
  928. RING0_INT_FRAME
  929. ASM_CLAC
  930. pushl_cfi $0
  931. pushl_cfi $do_spurious_interrupt_bug
  932. jmp error_code
  933. CFI_ENDPROC
  934. END(spurious_interrupt_bug)
  935. /*
  936. * End of kprobes section
  937. */
  938. .popsection
  939. ENTRY(kernel_thread_helper)
  940. pushl $0 # fake return address for unwinder
  941. CFI_STARTPROC
  942. movl %edi,%eax
  943. call *%esi
  944. call do_exit
  945. ud2 # padding for call trace
  946. CFI_ENDPROC
  947. ENDPROC(kernel_thread_helper)
  948. #ifdef CONFIG_XEN
  949. /* Xen doesn't set %esp to be precisely what the normal sysenter
  950. entrypoint expects, so fix it up before using the normal path. */
  951. ENTRY(xen_sysenter_target)
  952. RING0_INT_FRAME
  953. addl $5*4, %esp /* remove xen-provided frame */
  954. CFI_ADJUST_CFA_OFFSET -5*4
  955. jmp sysenter_past_esp
  956. CFI_ENDPROC
  957. ENTRY(xen_hypervisor_callback)
  958. CFI_STARTPROC
  959. pushl_cfi $0
  960. SAVE_ALL
  961. TRACE_IRQS_OFF
  962. /* Check to see if we got the event in the critical
  963. region in xen_iret_direct, after we've reenabled
  964. events and checked for pending events. This simulates
  965. iret instruction's behaviour where it delivers a
  966. pending interrupt when enabling interrupts. */
  967. movl PT_EIP(%esp),%eax
  968. cmpl $xen_iret_start_crit,%eax
  969. jb 1f
  970. cmpl $xen_iret_end_crit,%eax
  971. jae 1f
  972. jmp xen_iret_crit_fixup
  973. ENTRY(xen_do_upcall)
  974. 1: mov %esp, %eax
  975. call xen_evtchn_do_upcall
  976. jmp ret_from_intr
  977. CFI_ENDPROC
  978. ENDPROC(xen_hypervisor_callback)
  979. # Hypervisor uses this for application faults while it executes.
  980. # We get here for two reasons:
  981. # 1. Fault while reloading DS, ES, FS or GS
  982. # 2. Fault while executing IRET
  983. # Category 1 we fix up by reattempting the load, and zeroing the segment
  984. # register if the load fails.
  985. # Category 2 we fix up by jumping to do_iret_error. We cannot use the
  986. # normal Linux return path in this case because if we use the IRET hypercall
  987. # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
  988. # We distinguish between categories by maintaining a status value in EAX.
  989. ENTRY(xen_failsafe_callback)
  990. CFI_STARTPROC
  991. pushl_cfi %eax
  992. movl $1,%eax
  993. 1: mov 4(%esp),%ds
  994. 2: mov 8(%esp),%es
  995. 3: mov 12(%esp),%fs
  996. 4: mov 16(%esp),%gs
  997. testl %eax,%eax
  998. popl_cfi %eax
  999. lea 16(%esp),%esp
  1000. CFI_ADJUST_CFA_OFFSET -16
  1001. jz 5f
  1002. addl $16,%esp
  1003. jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
  1004. 5: pushl_cfi $0 # EAX == 0 => Category 1 (Bad segment)
  1005. SAVE_ALL
  1006. jmp ret_from_exception
  1007. CFI_ENDPROC
  1008. .section .fixup,"ax"
  1009. 6: xorl %eax,%eax
  1010. movl %eax,4(%esp)
  1011. jmp 1b
  1012. 7: xorl %eax,%eax
  1013. movl %eax,8(%esp)
  1014. jmp 2b
  1015. 8: xorl %eax,%eax
  1016. movl %eax,12(%esp)
  1017. jmp 3b
  1018. 9: xorl %eax,%eax
  1019. movl %eax,16(%esp)
  1020. jmp 4b
  1021. .previous
  1022. _ASM_EXTABLE(1b,6b)
  1023. _ASM_EXTABLE(2b,7b)
  1024. _ASM_EXTABLE(3b,8b)
  1025. _ASM_EXTABLE(4b,9b)
  1026. ENDPROC(xen_failsafe_callback)
  1027. BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
  1028. xen_evtchn_do_upcall)
  1029. #endif /* CONFIG_XEN */
  1030. #ifdef CONFIG_FUNCTION_TRACER
  1031. #ifdef CONFIG_DYNAMIC_FTRACE
  1032. ENTRY(mcount)
  1033. ret
  1034. END(mcount)
  1035. ENTRY(ftrace_caller)
  1036. cmpl $0, function_trace_stop
  1037. jne ftrace_stub
  1038. pushl %eax
  1039. pushl %ecx
  1040. pushl %edx
  1041. pushl $0 /* Pass NULL as regs pointer */
  1042. movl 4*4(%esp), %eax
  1043. movl 0x4(%ebp), %edx
  1044. leal function_trace_op, %ecx
  1045. subl $MCOUNT_INSN_SIZE, %eax
  1046. .globl ftrace_call
  1047. ftrace_call:
  1048. call ftrace_stub
  1049. addl $4,%esp /* skip NULL pointer */
  1050. popl %edx
  1051. popl %ecx
  1052. popl %eax
  1053. ftrace_ret:
  1054. #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  1055. .globl ftrace_graph_call
  1056. ftrace_graph_call:
  1057. jmp ftrace_stub
  1058. #endif
  1059. .globl ftrace_stub
  1060. ftrace_stub:
  1061. ret
  1062. END(ftrace_caller)
  1063. ENTRY(ftrace_regs_caller)
  1064. pushf /* push flags before compare (in cs location) */
  1065. cmpl $0, function_trace_stop
  1066. jne ftrace_restore_flags
  1067. /*
  1068. * i386 does not save SS and ESP when coming from kernel.
  1069. * Instead, to get sp, &regs->sp is used (see ptrace.h).
  1070. * Unfortunately, that means eflags must be at the same location
  1071. * as the current return ip is. We move the return ip into the
  1072. * ip location, and move flags into the return ip location.
  1073. */
  1074. pushl 4(%esp) /* save return ip into ip slot */
  1075. pushl $0 /* Load 0 into orig_ax */
  1076. pushl %gs
  1077. pushl %fs
  1078. pushl %es
  1079. pushl %ds
  1080. pushl %eax
  1081. pushl %ebp
  1082. pushl %edi
  1083. pushl %esi
  1084. pushl %edx
  1085. pushl %ecx
  1086. pushl %ebx
  1087. movl 13*4(%esp), %eax /* Get the saved flags */
  1088. movl %eax, 14*4(%esp) /* Move saved flags into regs->flags location */
  1089. /* clobbering return ip */
  1090. movl $__KERNEL_CS,13*4(%esp)
  1091. movl 12*4(%esp), %eax /* Load ip (1st parameter) */
  1092. subl $MCOUNT_INSN_SIZE, %eax /* Adjust ip */
  1093. movl 0x4(%ebp), %edx /* Load parent ip (2nd parameter) */
  1094. leal function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
  1095. pushl %esp /* Save pt_regs as 4th parameter */
  1096. GLOBAL(ftrace_regs_call)
  1097. call ftrace_stub
  1098. addl $4, %esp /* Skip pt_regs */
  1099. movl 14*4(%esp), %eax /* Move flags back into cs */
  1100. movl %eax, 13*4(%esp) /* Needed to keep addl from modifying flags */
  1101. movl 12*4(%esp), %eax /* Get return ip from regs->ip */
  1102. movl %eax, 14*4(%esp) /* Put return ip back for ret */
  1103. popl %ebx
  1104. popl %ecx
  1105. popl %edx
  1106. popl %esi
  1107. popl %edi
  1108. popl %ebp
  1109. popl %eax
  1110. popl %ds
  1111. popl %es
  1112. popl %fs
  1113. popl %gs
  1114. addl $8, %esp /* Skip orig_ax and ip */
  1115. popf /* Pop flags at end (no addl to corrupt flags) */
  1116. jmp ftrace_ret
  1117. ftrace_restore_flags:
  1118. popf
  1119. jmp ftrace_stub
  1120. #else /* ! CONFIG_DYNAMIC_FTRACE */
  1121. ENTRY(mcount)
  1122. cmpl $0, function_trace_stop
  1123. jne ftrace_stub
  1124. cmpl $ftrace_stub, ftrace_trace_function
  1125. jnz trace
  1126. #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  1127. cmpl $ftrace_stub, ftrace_graph_return
  1128. jnz ftrace_graph_caller
  1129. cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
  1130. jnz ftrace_graph_caller
  1131. #endif
  1132. .globl ftrace_stub
  1133. ftrace_stub:
  1134. ret
  1135. /* taken from glibc */
  1136. trace:
  1137. pushl %eax
  1138. pushl %ecx
  1139. pushl %edx
  1140. movl 0xc(%esp), %eax
  1141. movl 0x4(%ebp), %edx
  1142. subl $MCOUNT_INSN_SIZE, %eax
  1143. call *ftrace_trace_function
  1144. popl %edx
  1145. popl %ecx
  1146. popl %eax
  1147. jmp ftrace_stub
  1148. END(mcount)
  1149. #endif /* CONFIG_DYNAMIC_FTRACE */
  1150. #endif /* CONFIG_FUNCTION_TRACER */
  1151. #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  1152. ENTRY(ftrace_graph_caller)
  1153. pushl %eax
  1154. pushl %ecx
  1155. pushl %edx
  1156. movl 0xc(%esp), %edx
  1157. lea 0x4(%ebp), %eax
  1158. movl (%ebp), %ecx
  1159. subl $MCOUNT_INSN_SIZE, %edx
  1160. call prepare_ftrace_return
  1161. popl %edx
  1162. popl %ecx
  1163. popl %eax
  1164. ret
  1165. END(ftrace_graph_caller)
  1166. .globl return_to_handler
  1167. return_to_handler:
  1168. pushl %eax
  1169. pushl %edx
  1170. movl %ebp, %eax
  1171. call ftrace_return_to_handler
  1172. movl %eax, %ecx
  1173. popl %edx
  1174. popl %eax
  1175. jmp *%ecx
  1176. #endif
  1177. /*
  1178. * Some functions should be protected against kprobes
  1179. */
  1180. .pushsection .kprobes.text, "ax"
  1181. ENTRY(page_fault)
  1182. RING0_EC_FRAME
  1183. ASM_CLAC
  1184. pushl_cfi $do_page_fault
  1185. ALIGN
  1186. error_code:
  1187. /* the function address is in %gs's slot on the stack */
  1188. pushl_cfi %fs
  1189. /*CFI_REL_OFFSET fs, 0*/
  1190. pushl_cfi %es
  1191. /*CFI_REL_OFFSET es, 0*/
  1192. pushl_cfi %ds
  1193. /*CFI_REL_OFFSET ds, 0*/
  1194. pushl_cfi %eax
  1195. CFI_REL_OFFSET eax, 0
  1196. pushl_cfi %ebp
  1197. CFI_REL_OFFSET ebp, 0
  1198. pushl_cfi %edi
  1199. CFI_REL_OFFSET edi, 0
  1200. pushl_cfi %esi
  1201. CFI_REL_OFFSET esi, 0
  1202. pushl_cfi %edx
  1203. CFI_REL_OFFSET edx, 0
  1204. pushl_cfi %ecx
  1205. CFI_REL_OFFSET ecx, 0
  1206. pushl_cfi %ebx
  1207. CFI_REL_OFFSET ebx, 0
  1208. cld
  1209. movl $(__KERNEL_PERCPU), %ecx
  1210. movl %ecx, %fs
  1211. UNWIND_ESPFIX_STACK
  1212. GS_TO_REG %ecx
  1213. movl PT_GS(%esp), %edi # get the function address
  1214. movl PT_ORIG_EAX(%esp), %edx # get the error code
  1215. movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
  1216. REG_TO_PTGS %ecx
  1217. SET_KERNEL_GS %ecx
  1218. movl $(__USER_DS), %ecx
  1219. movl %ecx, %ds
  1220. movl %ecx, %es
  1221. TRACE_IRQS_OFF
  1222. movl %esp,%eax # pt_regs pointer
  1223. call *%edi
  1224. jmp ret_from_exception
  1225. CFI_ENDPROC
  1226. END(page_fault)
  1227. /*
  1228. * Debug traps and NMI can happen at the one SYSENTER instruction
  1229. * that sets up the real kernel stack. Check here, since we can't
  1230. * allow the wrong stack to be used.
  1231. *
  1232. * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
  1233. * already pushed 3 words if it hits on the sysenter instruction:
  1234. * eflags, cs and eip.
  1235. *
  1236. * We just load the right stack, and push the three (known) values
  1237. * by hand onto the new stack - while updating the return eip past
  1238. * the instruction that would have done it for sysenter.
  1239. */
  1240. .macro FIX_STACK offset ok label
  1241. cmpw $__KERNEL_CS, 4(%esp)
  1242. jne \ok
  1243. \label:
  1244. movl TSS_sysenter_sp0 + \offset(%esp), %esp
  1245. CFI_DEF_CFA esp, 0
  1246. CFI_UNDEFINED eip
  1247. pushfl_cfi
  1248. pushl_cfi $__KERNEL_CS
  1249. pushl_cfi $sysenter_past_esp
  1250. CFI_REL_OFFSET eip, 0
  1251. .endm
  1252. ENTRY(debug)
  1253. RING0_INT_FRAME
  1254. ASM_CLAC
  1255. cmpl $ia32_sysenter_target,(%esp)
  1256. jne debug_stack_correct
  1257. FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
  1258. debug_stack_correct:
  1259. pushl_cfi $-1 # mark this as an int
  1260. SAVE_ALL
  1261. TRACE_IRQS_OFF
  1262. xorl %edx,%edx # error code 0
  1263. movl %esp,%eax # pt_regs pointer
  1264. call do_debug
  1265. jmp ret_from_exception
  1266. CFI_ENDPROC
  1267. END(debug)
  1268. /*
  1269. * NMI is doubly nasty. It can happen _while_ we're handling
  1270. * a debug fault, and the debug fault hasn't yet been able to
  1271. * clear up the stack. So we first check whether we got an
  1272. * NMI on the sysenter entry path, but after that we need to
  1273. * check whether we got an NMI on the debug path where the debug
  1274. * fault happened on the sysenter path.
  1275. */
  1276. ENTRY(nmi)
  1277. RING0_INT_FRAME
  1278. ASM_CLAC
  1279. pushl_cfi %eax
  1280. movl %ss, %eax
  1281. cmpw $__ESPFIX_SS, %ax
  1282. popl_cfi %eax
  1283. je nmi_espfix_stack
  1284. cmpl $ia32_sysenter_target,(%esp)
  1285. je nmi_stack_fixup
  1286. pushl_cfi %eax
  1287. movl %esp,%eax
  1288. /* Do not access memory above the end of our stack page,
  1289. * it might not exist.
  1290. */
  1291. andl $(THREAD_SIZE-1),%eax
  1292. cmpl $(THREAD_SIZE-20),%eax
  1293. popl_cfi %eax
  1294. jae nmi_stack_correct
  1295. cmpl $ia32_sysenter_target,12(%esp)
  1296. je nmi_debug_stack_check
  1297. nmi_stack_correct:
  1298. /* We have a RING0_INT_FRAME here */
  1299. pushl_cfi %eax
  1300. SAVE_ALL
  1301. xorl %edx,%edx # zero error code
  1302. movl %esp,%eax # pt_regs pointer
  1303. call do_nmi
  1304. jmp restore_all_notrace
  1305. CFI_ENDPROC
  1306. nmi_stack_fixup:
  1307. RING0_INT_FRAME
  1308. FIX_STACK 12, nmi_stack_correct, 1
  1309. jmp nmi_stack_correct
  1310. nmi_debug_stack_check:
  1311. /* We have a RING0_INT_FRAME here */
  1312. cmpw $__KERNEL_CS,16(%esp)
  1313. jne nmi_stack_correct
  1314. cmpl $debug,(%esp)
  1315. jb nmi_stack_correct
  1316. cmpl $debug_esp_fix_insn,(%esp)
  1317. ja nmi_stack_correct
  1318. FIX_STACK 24, nmi_stack_correct, 1
  1319. jmp nmi_stack_correct
  1320. nmi_espfix_stack:
  1321. /* We have a RING0_INT_FRAME here.
  1322. *
  1323. * create the pointer to lss back
  1324. */
  1325. pushl_cfi %ss
  1326. pushl_cfi %esp
  1327. addl $4, (%esp)
  1328. /* copy the iret frame of 12 bytes */
  1329. .rept 3
  1330. pushl_cfi 16(%esp)
  1331. .endr
  1332. pushl_cfi %eax
  1333. SAVE_ALL
  1334. FIXUP_ESPFIX_STACK # %eax == %esp
  1335. xorl %edx,%edx # zero error code
  1336. call do_nmi
  1337. RESTORE_REGS
  1338. lss 12+4(%esp), %esp # back to espfix stack
  1339. CFI_ADJUST_CFA_OFFSET -24
  1340. jmp irq_return
  1341. CFI_ENDPROC
  1342. END(nmi)
  1343. ENTRY(int3)
  1344. RING0_INT_FRAME
  1345. ASM_CLAC
  1346. pushl_cfi $-1 # mark this as an int
  1347. SAVE_ALL
  1348. TRACE_IRQS_OFF
  1349. xorl %edx,%edx # zero error code
  1350. movl %esp,%eax # pt_regs pointer
  1351. call do_int3
  1352. jmp ret_from_exception
  1353. CFI_ENDPROC
  1354. END(int3)
  1355. ENTRY(general_protection)
  1356. RING0_EC_FRAME
  1357. pushl_cfi $do_general_protection
  1358. jmp error_code
  1359. CFI_ENDPROC
  1360. END(general_protection)
  1361. #ifdef CONFIG_KVM_GUEST
  1362. ENTRY(async_page_fault)
  1363. RING0_EC_FRAME
  1364. ASM_CLAC
  1365. pushl_cfi $do_async_page_fault
  1366. jmp error_code
  1367. CFI_ENDPROC
  1368. END(async_page_fault)
  1369. #endif
  1370. /*
  1371. * End of kprobes section
  1372. */
  1373. .popsection