arch/x86/kernel/entry_32.S

   1 /*
   2  *
   3  *  Copyright (C) 1991, 1992  Linus Torvalds
   4  */
   5
   6 /*
   7  * entry.S contains the system-call and fault low-level handling routines.
   8  * This also contains the timer-interrupt handler, as well as all interrupts
   9  * and faults that can result in a task-switch.
  10  *
  11  * NOTE: This code handles signal-recognition, which happens every time
  12  * after a timer-interrupt and after each system call.
  13  *
  14  * I changed all the .align's to 4 (16 byte alignment), as that's faster
  15  * on a 486.
  16  *
  17  * Stack layout in 'syscall_exit':
  18  *      ptrace needs to have all regs on the stack.
  19  *      if the order here is changed, it needs to be
  20  *      updated in fork.c:copy_process, signal.c:do_signal,
  21  *      ptrace.c and ptrace.h
  22  *
  23  *       0(%esp) - %ebx
  24  *       4(%esp) - %ecx
  25  *       8(%esp) - %edx
  26  *       C(%esp) - %esi
  27  *      10(%esp) - %edi
  28  *      14(%esp) - %ebp
  29  *      18(%esp) - %eax
  30  *      1C(%esp) - %ds
  31  *      20(%esp) - %es
  32  *      24(%esp) - %fs
  33  *      28(%esp) - %gs          saved iff !CONFIG_X86_32_LAZY_GS
  34  *      2C(%esp) - orig_eax
  35  *      30(%esp) - %eip
  36  *      34(%esp) - %cs
  37  *      38(%esp) - %eflags
  38  *      3C(%esp) - %oldesp
  39  *      40(%esp) - %oldss
  40  *
  41  * "current" is in register %ebx during any slow entries.
  42  */
  43
  44 #include <linux/linkage.h>
  45 #include <linux/err.h>
  46 #include <asm/thread_info.h>
  47 #include <asm/irqflags.h>
  48 #include <asm/errno.h>
  49 #include <asm/segment.h>
  50 #include <asm/smp.h>
  51 #include <asm/page_types.h>
  52 #include <asm/percpu.h>
  53 #include <asm/dwarf2.h>
  54 #include <asm/processor-flags.h>
  55 #include <asm/ftrace.h>
  56 #include <asm/irq_vectors.h>
  57 #include <asm/cpufeature.h>
  58 #include <asm/alternative-asm.h>
  59
  60 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
  61 #include <linux/elf-em.h>
  62 #define AUDIT_ARCH_I386         (EM_386|__AUDIT_ARCH_LE)
  63 #define __AUDIT_ARCH_LE    0x40000000
  64
  65 #ifndef CONFIG_AUDITSYSCALL
  66 #define sysenter_audit  syscall_trace_entry
  67 #define sysexit_audit   syscall_exit_work
  68 #endif
  69
  70         .section .entry.text, "ax"
  71
  72 /*
  73  * We use macros for low-level operations which need to be overridden
  74  * for paravirtualization.  The following will never clobber any registers:
  75  *   INTERRUPT_RETURN (aka. "iret")
  76  *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  77  *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  78  *
  79  * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  80  * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  81  * Allowing a register to be clobbered can shrink the paravirt replacement
  82  * enough to patch inline, increasing performance.
  83  */
  84
  85 #define nr_syscalls ((syscall_table_size)/4)
  86
  87 #ifdef CONFIG_PREEMPT
  88 #define preempt_stop(clobbers)  DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  89 #else
  90 #define preempt_stop(clobbers)
  91 #define resume_kernel           restore_all
  92 #endif
  93
  94 .macro TRACE_IRQS_IRET
  95 #ifdef CONFIG_TRACE_IRQFLAGS
  96         testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)     # interrupts off?
  97         jz 1f
  98         TRACE_IRQS_ON
  99 1:
 100 #endif
 101 .endm
 102
 103 #ifdef CONFIG_VM86
 104 #define resume_userspace_sig    check_userspace
 105 #else
 106 #define resume_userspace_sig    resume_userspace
 107 #endif
 108
 109 /*
 110  * User gs save/restore
 111  *
 112  * %gs is used for userland TLS and kernel only uses it for stack
 113  * canary which is required to be at %gs:20 by gcc.  Read the comment
 114  * at the top of stackprotector.h for more info.
 115  *
 116  * Local labels 98 and 99 are used.
 117  */
 118 #ifdef CONFIG_X86_32_LAZY_GS
 119
 120  /* unfortunately push/pop can't be no-op */
 121 .macro PUSH_GS
 122         pushl_cfi $0
 123 .endm
 124 .macro POP_GS pop=0
 125         addl $(4 + \pop), %esp
 126         CFI_ADJUST_CFA_OFFSET -(4 + \pop)
 127 .endm
 128 .macro POP_GS_EX
 129 .endm
 130
 131  /* all the rest are no-op */
 132 .macro PTGS_TO_GS
 133 .endm
 134 .macro PTGS_TO_GS_EX
 135 .endm
 136 .macro GS_TO_REG reg
 137 .endm
 138 .macro REG_TO_PTGS reg
 139 .endm
 140 .macro SET_KERNEL_GS reg
 141 .endm
 142
 143 #else   /* CONFIG_X86_32_LAZY_GS */
 144
 145 .macro PUSH_GS
 146         pushl_cfi %gs
 147         /*CFI_REL_OFFSET gs, 0*/
 148 .endm
 149
 150 .macro POP_GS pop=0
 151 98:     popl_cfi %gs
 152         /*CFI_RESTORE gs*/
 153   .if \pop <> 0
 154         add $\pop, %esp
 155         CFI_ADJUST_CFA_OFFSET -\pop
 156   .endif
 157 .endm
 158 .macro POP_GS_EX
 159 .pushsection .fixup, "ax"
 160 99:     movl $0, (%esp)
 161         jmp 98b
 162 .section __ex_table, "a"
 163         .align 4
 164         .long 98b, 99b
 165 .popsection
 166 .endm
 167
 168 .macro PTGS_TO_GS
 169 98:     mov PT_GS(%esp), %gs
 170 .endm
 171 .macro PTGS_TO_GS_EX
 172 .pushsection .fixup, "ax"
 173 99:     movl $0, PT_GS(%esp)
 174         jmp 98b
 175 .section __ex_table, "a"
 176         .align 4
 177         .long 98b, 99b
 178 .popsection
 179 .endm
 180
 181 .macro GS_TO_REG reg
 182         movl %gs, \reg
 183         /*CFI_REGISTER gs, \reg*/
 184 .endm
 185 .macro REG_TO_PTGS reg
 186         movl \reg, PT_GS(%esp)
 187         /*CFI_REL_OFFSET gs, PT_GS*/
 188 .endm
 189 .macro SET_KERNEL_GS reg
 190         movl $(__KERNEL_STACK_CANARY), \reg
 191         movl \reg, %gs
 192 .endm
 193
 194 #endif  /* CONFIG_X86_32_LAZY_GS */
 195
 196 .macro SAVE_ALL
 197         cld
 198         PUSH_GS
 199         pushl_cfi %fs
 200         /*CFI_REL_OFFSET fs, 0;*/
 201         pushl_cfi %es
 202         /*CFI_REL_OFFSET es, 0;*/
 203         pushl_cfi %ds
 204         /*CFI_REL_OFFSET ds, 0;*/
 205         pushl_cfi %eax
 206         CFI_REL_OFFSET eax, 0
 207         pushl_cfi %ebp
 208         CFI_REL_OFFSET ebp, 0
 209         pushl_cfi %edi
 210         CFI_REL_OFFSET edi, 0
 211         pushl_cfi %esi
 212         CFI_REL_OFFSET esi, 0
 213         pushl_cfi %edx
 214         CFI_REL_OFFSET edx, 0
 215         pushl_cfi %ecx
 216         CFI_REL_OFFSET ecx, 0
 217         pushl_cfi %ebx
 218         CFI_REL_OFFSET ebx, 0
 219         movl $(__USER_DS), %edx
 220         movl %edx, %ds
 221         movl %edx, %es
 222         movl $(__KERNEL_PERCPU), %edx
 223         movl %edx, %fs
 224         SET_KERNEL_GS %edx
 225 .endm
 226
 227 .macro RESTORE_INT_REGS
 228         popl_cfi %ebx
 229         CFI_RESTORE ebx
 230         popl_cfi %ecx
 231         CFI_RESTORE ecx
 232         popl_cfi %edx
 233         CFI_RESTORE edx
 234         popl_cfi %esi
 235         CFI_RESTORE esi
 236         popl_cfi %edi
 237         CFI_RESTORE edi
 238         popl_cfi %ebp
 239         CFI_RESTORE ebp
 240         popl_cfi %eax
 241         CFI_RESTORE eax
 242 .endm
 243
 244 .macro RESTORE_REGS pop=0
 245         RESTORE_INT_REGS
 246 1:      popl_cfi %ds
 247         /*CFI_RESTORE ds;*/
 248 2:      popl_cfi %es
 249         /*CFI_RESTORE es;*/
 250 3:      popl_cfi %fs
 251         /*CFI_RESTORE fs;*/
 252         POP_GS \pop
 253 .pushsection .fixup, "ax"
 254 4:      movl $0, (%esp)
 255         jmp 1b
 256 5:      movl $0, (%esp)
 257         jmp 2b
 258 6:      movl $0, (%esp)
 259         jmp 3b
 260 .section __ex_table, "a"
 261         .align 4
 262         .long 1b, 4b
 263         .long 2b, 5b
 264         .long 3b, 6b
 265 .popsection
 266         POP_GS_EX
 267 .endm
 268
 269 .macro RING0_INT_FRAME
 270         CFI_STARTPROC simple
 271         CFI_SIGNAL_FRAME
 272         CFI_DEF_CFA esp, 3*4
 273         /*CFI_OFFSET cs, -2*4;*/
 274         CFI_OFFSET eip, -3*4
 275 .endm
 276
 277 .macro RING0_EC_FRAME
 278         CFI_STARTPROC simple
 279         CFI_SIGNAL_FRAME
 280         CFI_DEF_CFA esp, 4*4
 281         /*CFI_OFFSET cs, -2*4;*/
 282         CFI_OFFSET eip, -3*4
 283 .endm
 284
 285 .macro RING0_PTREGS_FRAME
 286         CFI_STARTPROC simple
 287         CFI_SIGNAL_FRAME
 288         CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
 289         /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
 290         CFI_OFFSET eip, PT_EIP-PT_OLDESP
 291         /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
 292         /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
 293         CFI_OFFSET eax, PT_EAX-PT_OLDESP
 294         CFI_OFFSET ebp, PT_EBP-PT_OLDESP
 295         CFI_OFFSET edi, PT_EDI-PT_OLDESP
 296         CFI_OFFSET esi, PT_ESI-PT_OLDESP
 297         CFI_OFFSET edx, PT_EDX-PT_OLDESP
 298         CFI_OFFSET ecx, PT_ECX-PT_OLDESP
 299         CFI_OFFSET ebx, PT_EBX-PT_OLDESP
 300 .endm
 301
 302 ENTRY(ret_from_fork)
 303         CFI_STARTPROC
 304         pushl_cfi %eax
 305         call schedule_tail
 306         GET_THREAD_INFO(%ebp)
 307         popl_cfi %eax
 308         pushl_cfi $0x0202               # Reset kernel eflags
 309         popfl_cfi
 310         jmp syscall_exit
 311         CFI_ENDPROC
 312 END(ret_from_fork)
 313
 314 /*
 315  * Interrupt exit functions should be protected against kprobes
 316  */
 317         .pushsection .kprobes.text, "ax"
 318 /*
 319  * Return to user mode is not as complex as all this looks,
 320  * but we want the default path for a system call return to
 321  * go as quickly as possible which is why some of this is
 322  * less clear than it otherwise should be.
 323  */
 324
 325         # userspace resumption stub bypassing syscall exit tracing
 326         ALIGN
 327         RING0_PTREGS_FRAME
 328 ret_from_exception:
 329         preempt_stop(CLBR_ANY)
 330 ret_from_intr:
 331         GET_THREAD_INFO(%ebp)
 332 check_userspace:
 333         movl PT_EFLAGS(%esp), %eax      # mix EFLAGS and CS
 334         movb PT_CS(%esp), %al
 335         andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 336         cmpl $USER_RPL, %eax
 337         jb resume_kernel                # not returning to v8086 or userspace
 338
 339 ENTRY(resume_userspace)
 340         LOCKDEP_SYS_EXIT
 341         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 342                                         # setting need_resched or sigpending
 343                                         # between sampling and the iret
 344         TRACE_IRQS_OFF
 345         movl TI_flags(%ebp), %ecx
 346         andl $_TIF_WORK_MASK, %ecx      # is there any work to be done on
 347                                         # int/exception return?
 348         jne work_pending
 349         jmp restore_all
 350 END(ret_from_exception)
 351
 352 #ifdef CONFIG_PREEMPT
 353 ENTRY(resume_kernel)
 354         DISABLE_INTERRUPTS(CLBR_ANY)
 355         cmpl $0,TI_preempt_count(%ebp)  # non-zero preempt_count ?
 356         jnz restore_all
 357 need_resched:
 358         movl TI_flags(%ebp), %ecx       # need_resched set ?
 359         testb $_TIF_NEED_RESCHED, %cl
 360         jz restore_all
 361         testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)    # interrupts off (exception path) ?
 362         jz restore_all
 363         call preempt_schedule_irq
 364         jmp need_resched
 365 END(resume_kernel)
 366 #endif
 367         CFI_ENDPROC
 368 /*
 369  * End of kprobes section
 370  */
 371         .popsection
 372
 373 /* SYSENTER_RETURN points to after the "sysenter" instruction in
 374    the vsyscall page.  See vsyscall-sysentry.S, which defines the symbol.  */
 375
 376         # sysenter call handler stub
 377 ENTRY(ia32_sysenter_target)
 378         CFI_STARTPROC simple
 379         CFI_SIGNAL_FRAME
 380         CFI_DEF_CFA esp, 0
 381         CFI_REGISTER esp, ebp
 382         movl TSS_sysenter_sp0(%esp),%esp
 383 sysenter_past_esp:
 384         /*
 385          * Interrupts are disabled here, but we can't trace it until
 386          * enough kernel state to call TRACE_IRQS_OFF can be called - but
 387          * we immediately enable interrupts at that point anyway.
 388          */
 389         pushl_cfi $__USER_DS
 390         /*CFI_REL_OFFSET ss, 0*/
 391         pushl_cfi %ebp
 392         CFI_REL_OFFSET esp, 0
 393         pushfl_cfi
 394         orl $X86_EFLAGS_IF, (%esp)
 395         pushl_cfi $__USER_CS
 396         /*CFI_REL_OFFSET cs, 0*/
 397         /*
 398          * Push current_thread_info()->sysenter_return to the stack.
 399          * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
 400          * pushed above; +8 corresponds to copy_thread's esp0 setting.
 401          */
 402         pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
 403         CFI_REL_OFFSET eip, 0
 404
 405         pushl_cfi %eax
 406         SAVE_ALL
 407         ENABLE_INTERRUPTS(CLBR_NONE)
 408
 409 /*
 410  * Load the potential sixth argument from user stack.
 411  * Careful about security.
 412  */
 413         cmpl $__PAGE_OFFSET-3,%ebp
 414         jae syscall_fault
 415 1:      movl (%ebp),%ebp
 416         movl %ebp,PT_EBP(%esp)
 417 .section __ex_table,"a"
 418         .align 4
 419         .long 1b,syscall_fault
 420 .previous
 421
 422         GET_THREAD_INFO(%ebp)
 423
 424         testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
 425         jnz sysenter_audit
 426 sysenter_do_call:
 427         cmpl $(nr_syscalls), %eax
 428         jae syscall_badsys
 429         call *sys_call_table(,%eax,4)
 430         movl %eax,PT_EAX(%esp)
 431         LOCKDEP_SYS_EXIT
 432         DISABLE_INTERRUPTS(CLBR_ANY)
 433         TRACE_IRQS_OFF
 434         movl TI_flags(%ebp), %ecx
 435         testl $_TIF_ALLWORK_MASK, %ecx
 436         jne sysexit_audit
 437 sysenter_exit:
 438 /* if something modifies registers it must also disable sysexit */
 439         movl PT_EIP(%esp), %edx
 440         movl PT_OLDESP(%esp), %ecx
 441         xorl %ebp,%ebp
 442         TRACE_IRQS_ON
 443 1:      mov  PT_FS(%esp), %fs
 444         PTGS_TO_GS
 445         ENABLE_INTERRUPTS_SYSEXIT
 446
 447 #ifdef CONFIG_AUDITSYSCALL
 448 sysenter_audit:
 449         testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
 450         jnz syscall_trace_entry
 451         addl $4,%esp
 452         CFI_ADJUST_CFA_OFFSET -4
 453         /* %esi already in 8(%esp)         6th arg: 4th syscall arg */
 454         /* %edx already in 4(%esp)         5th arg: 3rd syscall arg */
 455         /* %ecx already in 0(%esp)         4th arg: 2nd syscall arg */
 456         movl %ebx,%ecx                  /* 3rd arg: 1st syscall arg */
 457         movl %eax,%edx                  /* 2nd arg: syscall number */
 458         movl $AUDIT_ARCH_I386,%eax      /* 1st arg: audit arch */
 459         call __audit_syscall_entry
 460         pushl_cfi %ebx
 461         movl PT_EAX(%esp),%eax          /* reload syscall number */
 462         jmp sysenter_do_call
 463
 464 sysexit_audit:
 465         testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 466         jne syscall_exit_work
 467         TRACE_IRQS_ON
 468         ENABLE_INTERRUPTS(CLBR_ANY)
 469         movl %eax,%edx          /* second arg, syscall return value */
 470         cmpl $-MAX_ERRNO,%eax   /* is it an error ? */
 471         setbe %al               /* 1 if so, 0 if not */
 472         movzbl %al,%eax         /* zero-extend that */
 473         call __audit_syscall_exit
 474         DISABLE_INTERRUPTS(CLBR_ANY)
 475         TRACE_IRQS_OFF
 476         movl TI_flags(%ebp), %ecx
 477         testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 478         jne syscall_exit_work
 479         movl PT_EAX(%esp),%eax  /* reload syscall return value */
 480         jmp sysenter_exit
 481 #endif
 482
 483         CFI_ENDPROC
 484 .pushsection .fixup,"ax"
 485 2:      movl $0,PT_FS(%esp)
 486         jmp 1b
 487 .section __ex_table,"a"
 488         .align 4
 489         .long 1b,2b
 490 .popsection
 491         PTGS_TO_GS_EX
 492 ENDPROC(ia32_sysenter_target)
 493
 494 /*
 495  * syscall stub including irq exit should be protected against kprobes
 496  */
 497         .pushsection .kprobes.text, "ax"
 498         # system call handler stub
 499 ENTRY(system_call)
 500         RING0_INT_FRAME                 # can't unwind into user space anyway
 501         pushl_cfi %eax                  # save orig_eax
 502         SAVE_ALL
 503         GET_THREAD_INFO(%ebp)
 504                                         # system call tracing in operation / emulation
 505         testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
 506         jnz syscall_trace_entry
 507         cmpl $(nr_syscalls), %eax
 508         jae syscall_badsys
 509 syscall_call:
 510         call *sys_call_table(,%eax,4)
 511         movl %eax,PT_EAX(%esp)          # store the return value
 512 syscall_exit:
 513         LOCKDEP_SYS_EXIT
 514         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 515                                         # setting need_resched or sigpending
 516                                         # between sampling and the iret
 517         TRACE_IRQS_OFF
 518         movl TI_flags(%ebp), %ecx
 519         testl $_TIF_ALLWORK_MASK, %ecx  # current->work
 520         jne syscall_exit_work
 521
 522 restore_all:
 523         TRACE_IRQS_IRET
 524 restore_all_notrace:
 525         movl PT_EFLAGS(%esp), %eax      # mix EFLAGS, SS and CS
 526         # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 527         # are returning to the kernel.
 528         # See comments in process.c:copy_thread() for details.
 529         movb PT_OLDSS(%esp), %ah
 530         movb PT_CS(%esp), %al
 531         andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 532         cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
 533         CFI_REMEMBER_STATE
 534         je ldt_ss                       # returning to user-space with LDT SS
 535 restore_nocheck:
 536         RESTORE_REGS 4                  # skip orig_eax/error_code
 537 irq_return:
 538         INTERRUPT_RETURN
 539 .section .fixup,"ax"
 540 ENTRY(iret_exc)
 541         pushl $0                        # no error code
 542         pushl $do_iret_error
 543         jmp error_code
 544 .previous
 545 .section __ex_table,"a"
 546         .align 4
 547         .long irq_return,iret_exc
 548 .previous
 549
 550         CFI_RESTORE_STATE
 551 ldt_ss:
 552         larl PT_OLDSS(%esp), %eax
 553         jnz restore_nocheck
 554         testl $0x00400000, %eax         # returning to 32bit stack?
 555         jnz restore_nocheck             # allright, normal return
 556
 557 #ifdef CONFIG_PARAVIRT
 558         /*
 559          * The kernel can't run on a non-flat stack if paravirt mode
 560          * is active.  Rather than try to fixup the high bits of
 561          * ESP, bypass this code entirely.  This may break DOSemu
 562          * and/or Wine support in a paravirt VM, although the option
 563          * is still available to implement the setting of the high
 564          * 16-bits in the INTERRUPT_RETURN paravirt-op.
 565          */
 566         cmpl $0, pv_info+PARAVIRT_enabled
 567         jne restore_nocheck
 568 #endif
 569
 570 /*
 571  * Setup and switch to ESPFIX stack
 572  *
 573  * We're returning to userspace with a 16 bit stack. The CPU will not
 574  * restore the high word of ESP for us on executing iret... This is an
 575  * "official" bug of all the x86-compatible CPUs, which we can work
 576  * around to make dosemu and wine happy. We do this by preloading the
 577  * high word of ESP with the high word of the userspace ESP while
 578  * compensating for the offset by changing to the ESPFIX segment with
 579  * a base address that matches for the difference.
 580  */
 581 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 582         mov %esp, %edx                  /* load kernel esp */
 583         mov PT_OLDESP(%esp), %eax       /* load userspace esp */
 584         mov %dx, %ax                    /* eax: new kernel esp */
 585         sub %eax, %edx                  /* offset (low word is 0) */
 586         shr $16, %edx
 587         mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
 588         mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
 589         pushl_cfi $__ESPFIX_SS
 590         pushl_cfi %eax                  /* new kernel esp */
 591         /* Disable interrupts, but do not irqtrace this section: we
 592          * will soon execute iret and the tracer was already set to
 593          * the irqstate after the iret */
 594         DISABLE_INTERRUPTS(CLBR_EAX)
 595         lss (%esp), %esp                /* switch to espfix segment */
 596         CFI_ADJUST_CFA_OFFSET -8
 597         jmp restore_nocheck
 598         CFI_ENDPROC
 599 ENDPROC(system_call)
 600
 601         # perform work that needs to be done immediately before resumption
 602         ALIGN
 603         RING0_PTREGS_FRAME              # can't unwind into user space anyway
 604 work_pending:
 605         testb $_TIF_NEED_RESCHED, %cl
 606         jz work_notifysig
 607 work_resched:
 608         call schedule
 609         LOCKDEP_SYS_EXIT
 610         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 611                                         # setting need_resched or sigpending
 612                                         # between sampling and the iret
 613         TRACE_IRQS_OFF
 614         movl TI_flags(%ebp), %ecx
 615         andl $_TIF_WORK_MASK, %ecx      # is there any work to be done other
 616                                         # than syscall tracing?
 617         jz restore_all
 618         testb $_TIF_NEED_RESCHED, %cl
 619         jnz work_resched
 620
 621 work_notifysig:                         # deal with pending signals and
 622                                         # notify-resume requests
 623 #ifdef CONFIG_VM86
 624         testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
 625         movl %esp, %eax
 626         jne work_notifysig_v86          # returning to kernel-space or
 627                                         # vm86-space
 628         TRACE_IRQS_ON
 629         ENABLE_INTERRUPTS(CLBR_NONE)
 630         xorl %edx, %edx
 631         call do_notify_resume
 632         jmp resume_userspace_sig
 633
 634         ALIGN
 635 work_notifysig_v86:
 636         pushl_cfi %ecx                  # save ti_flags for do_notify_resume
 637         call save_v86_state             # %eax contains pt_regs pointer
 638         popl_cfi %ecx
 639         movl %eax, %esp
 640 #else
 641         movl %esp, %eax
 642 #endif
 643         TRACE_IRQS_ON
 644         ENABLE_INTERRUPTS(CLBR_NONE)
 645         xorl %edx, %edx
 646         call do_notify_resume
 647         jmp resume_userspace_sig
 648 END(work_pending)
 649
 650         # perform syscall exit tracing
 651         ALIGN
 652 syscall_trace_entry:
 653         movl $-ENOSYS,PT_EAX(%esp)
 654         movl %esp, %eax
 655         call syscall_trace_enter
 656         /* What it returned is what we'll actually use.  */
 657         cmpl $(nr_syscalls), %eax
 658         jnae syscall_call
 659         jmp syscall_exit
 660 END(syscall_trace_entry)
 661
 662         # perform syscall exit tracing
 663         ALIGN
 664 syscall_exit_work:
 665         testl $_TIF_WORK_SYSCALL_EXIT, %ecx
 666         jz work_pending
 667         TRACE_IRQS_ON
 668         ENABLE_INTERRUPTS(CLBR_ANY)     # could let syscall_trace_leave() call
 669                                         # schedule() instead
 670         movl %esp, %eax
 671         call syscall_trace_leave
 672         jmp resume_userspace
 673 END(syscall_exit_work)
 674         CFI_ENDPROC
 675
 676         RING0_INT_FRAME                 # can't unwind into user space anyway
 677 syscall_fault:
 678         GET_THREAD_INFO(%ebp)
 679         movl $-EFAULT,PT_EAX(%esp)
 680         jmp resume_userspace
 681 END(syscall_fault)
 682
 683 syscall_badsys:
 684         movl $-ENOSYS,PT_EAX(%esp)
 685         jmp resume_userspace
 686 END(syscall_badsys)
 687         CFI_ENDPROC
 688 /*
 689  * End of kprobes section
 690  */
 691         .popsection
 692
 693 /*
 694  * System calls that need a pt_regs pointer.
 695  */
 696 #define PTREGSCALL0(name) \
 697         ALIGN; \
 698 ptregs_##name: \
 699         leal 4(%esp),%eax; \
 700         jmp sys_##name;
 701
 702 #define PTREGSCALL1(name) \
 703         ALIGN; \
 704 ptregs_##name: \
 705         leal 4(%esp),%edx; \
 706         movl (PT_EBX+4)(%esp),%eax; \
 707         jmp sys_##name;
 708
 709 #define PTREGSCALL2(name) \
 710         ALIGN; \
 711 ptregs_##name: \
 712         leal 4(%esp),%ecx; \
 713         movl (PT_ECX+4)(%esp),%edx; \
 714         movl (PT_EBX+4)(%esp),%eax; \
 715         jmp sys_##name;
 716
 717 #define PTREGSCALL3(name) \
 718         ALIGN; \
 719 ptregs_##name: \
 720         CFI_STARTPROC; \
 721         leal 4(%esp),%eax; \
 722         pushl_cfi %eax; \
 723         movl PT_EDX(%eax),%ecx; \
 724         movl PT_ECX(%eax),%edx; \
 725         movl PT_EBX(%eax),%eax; \
 726         call sys_##name; \
 727         addl $4,%esp; \
 728         CFI_ADJUST_CFA_OFFSET -4; \
 729         ret; \
 730         CFI_ENDPROC; \
 731 ENDPROC(ptregs_##name)
 732
 733 PTREGSCALL1(iopl)
 734 PTREGSCALL0(fork)
 735 PTREGSCALL0(vfork)
 736 PTREGSCALL3(execve)
 737 PTREGSCALL2(sigaltstack)
 738 PTREGSCALL0(sigreturn)
 739 PTREGSCALL0(rt_sigreturn)
 740 PTREGSCALL2(vm86)
 741 PTREGSCALL1(vm86old)
 742
 743 /* Clone is an oddball.  The 4th arg is in %edi */
 744         ALIGN;
 745 ptregs_clone:
 746         CFI_STARTPROC
 747         leal 4(%esp),%eax
 748         pushl_cfi %eax
 749         pushl_cfi PT_EDI(%eax)
 750         movl PT_EDX(%eax),%ecx
 751         movl PT_ECX(%eax),%edx
 752         movl PT_EBX(%eax),%eax
 753         call sys_clone
 754         addl $8,%esp
 755         CFI_ADJUST_CFA_OFFSET -8
 756         ret
 757         CFI_ENDPROC
 758 ENDPROC(ptregs_clone)
 759
 760 .macro FIXUP_ESPFIX_STACK
 761 /*
 762  * Switch back for ESPFIX stack to the normal zerobased stack
 763  *
 764  * We can't call C functions using the ESPFIX stack. This code reads
 765  * the high word of the segment base from the GDT and swiches to the
 766  * normal stack and adjusts ESP with the matching offset.
 767  */
 768         /* fixup the stack */
 769         mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 770         mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 771         shl $16, %eax
 772         addl %esp, %eax                 /* the adjusted stack pointer */
 773         pushl_cfi $__KERNEL_DS
 774         pushl_cfi %eax
 775         lss (%esp), %esp                /* switch to the normal stack segment */
 776         CFI_ADJUST_CFA_OFFSET -8
 777 .endm
 778 .macro UNWIND_ESPFIX_STACK
 779         movl %ss, %eax
 780         /* see if on espfix stack */
 781         cmpw $__ESPFIX_SS, %ax
 782         jne 27f
 783         movl $__KERNEL_DS, %eax
 784         movl %eax, %ds
 785         movl %eax, %es
 786         /* switch to normal stack */
 787         FIXUP_ESPFIX_STACK
 788 27:
 789 .endm
 790
 791 /*
 792  * Build the entry stubs and pointer table with some assembler magic.
 793  * We pack 7 stubs into a single 32-byte chunk, which will fit in a
 794  * single cache line on all modern x86 implementations.
 795  */
 796 .section .init.rodata,"a"
 797 ENTRY(interrupt)
 798 .section .entry.text, "ax"
 799         .p2align 5
 800         .p2align CONFIG_X86_L1_CACHE_SHIFT
 801 ENTRY(irq_entries_start)
 802         RING0_INT_FRAME
 803 vector=FIRST_EXTERNAL_VECTOR
 804 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
 805         .balign 32
 806   .rept 7
 807     .if vector < NR_VECTORS
 808       .if vector <> FIRST_EXTERNAL_VECTOR
 809         CFI_ADJUST_CFA_OFFSET -4
 810       .endif
 811 1:      pushl_cfi $(~vector+0x80)       /* Note: always in signed byte range */
 812       .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
 813         jmp 2f
 814       .endif
 815       .previous
 816         .long 1b
 817       .section .entry.text, "ax"
 818 vector=vector+1
 819     .endif
 820   .endr
 821 2:      jmp common_interrupt
 822 .endr
 823 END(irq_entries_start)
 824
 825 .previous
 826 END(interrupt)
 827 .previous
 828
 829 /*
 830  * the CPU automatically disables interrupts when executing an IRQ vector,
 831  * so IRQ-flags tracing has to follow that:
 832  */
 833         .p2align CONFIG_X86_L1_CACHE_SHIFT
 834 common_interrupt:
 835         addl $-0x80,(%esp)      /* Adjust vector into the [-256,-1] range */
 836         SAVE_ALL
 837         TRACE_IRQS_OFF
 838         movl %esp,%eax
 839         call do_IRQ
 840         jmp ret_from_intr
 841 ENDPROC(common_interrupt)
 842         CFI_ENDPROC
 843
 844 /*
 845  *  Irq entries should be protected against kprobes
 846  */
 847         .pushsection .kprobes.text, "ax"
 848 #define BUILD_INTERRUPT3(name, nr, fn)  \
 849 ENTRY(name)                             \
 850         RING0_INT_FRAME;                \
 851         pushl_cfi $~(nr);               \
 852         SAVE_ALL;                       \
 853         TRACE_IRQS_OFF                  \
 854         movl %esp,%eax;                 \
 855         call fn;                        \
 856         jmp ret_from_intr;              \
 857         CFI_ENDPROC;                    \
 858 ENDPROC(name)
 859
 860 #define BUILD_INTERRUPT(name, nr)       BUILD_INTERRUPT3(name, nr, smp_##name)
 861
 862 /* The include is where all of the SMP etc. interrupts come from */
 863 #include <asm/entry_arch.h>
 864
 865 ENTRY(coprocessor_error)
 866         RING0_INT_FRAME
 867         pushl_cfi $0
 868         pushl_cfi $do_coprocessor_error
 869         jmp error_code
 870         CFI_ENDPROC
 871 END(coprocessor_error)
 872
 873 ENTRY(simd_coprocessor_error)
 874         RING0_INT_FRAME
 875         pushl_cfi $0
 876 #ifdef CONFIG_X86_INVD_BUG
 877         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 878 661:    pushl_cfi $do_general_protection
 879 662:
 880 .section .altinstructions,"a"
 881         altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
 882 .previous
 883 .section .altinstr_replacement,"ax"
 884 663:    pushl $do_simd_coprocessor_error
 885 664:
 886 .previous
 887 #else
 888         pushl_cfi $do_simd_coprocessor_error
 889 #endif
 890         jmp error_code
 891         CFI_ENDPROC
 892 END(simd_coprocessor_error)
 893
 894 ENTRY(device_not_available)
 895         RING0_INT_FRAME
 896         pushl_cfi $-1                   # mark this as an int
 897         pushl_cfi $do_device_not_available
 898         jmp error_code
 899         CFI_ENDPROC
 900 END(device_not_available)
 901
 902 #ifdef CONFIG_PARAVIRT
 903 ENTRY(native_iret)
 904         iret
 905 .section __ex_table,"a"
 906         .align 4
 907         .long native_iret, iret_exc
 908 .previous
 909 END(native_iret)
 910
 911 ENTRY(native_irq_enable_sysexit)
 912         sti
 913         sysexit
 914 END(native_irq_enable_sysexit)
 915 #endif
 916
 917 ENTRY(overflow)
 918         RING0_INT_FRAME
 919         pushl_cfi $0
 920         pushl_cfi $do_overflow
 921         jmp error_code
 922         CFI_ENDPROC
 923 END(overflow)
 924
 925 ENTRY(bounds)
 926         RING0_INT_FRAME
 927         pushl_cfi $0
 928         pushl_cfi $do_bounds
 929         jmp error_code
 930         CFI_ENDPROC
 931 END(bounds)
 932
 933 ENTRY(invalid_op)
 934         RING0_INT_FRAME
 935         pushl_cfi $0
 936         pushl_cfi $do_invalid_op
 937         jmp error_code
 938         CFI_ENDPROC
 939 END(invalid_op)
 940
 941 ENTRY(coprocessor_segment_overrun)
 942         RING0_INT_FRAME
 943         pushl_cfi $0
 944         pushl_cfi $do_coprocessor_segment_overrun
 945         jmp error_code
 946         CFI_ENDPROC
 947 END(coprocessor_segment_overrun)
 948
 949 ENTRY(invalid_TSS)
 950         RING0_EC_FRAME
 951         pushl_cfi $do_invalid_TSS
 952         jmp error_code
 953         CFI_ENDPROC
 954 END(invalid_TSS)
 955
 956 ENTRY(segment_not_present)
 957         RING0_EC_FRAME
 958         pushl_cfi $do_segment_not_present
 959         jmp error_code
 960         CFI_ENDPROC
 961 END(segment_not_present)
 962
 963 ENTRY(stack_segment)
 964         RING0_EC_FRAME
 965         pushl_cfi $do_stack_segment
 966         jmp error_code
 967         CFI_ENDPROC
 968 END(stack_segment)
 969
 970 ENTRY(alignment_check)
 971         RING0_EC_FRAME
 972         pushl_cfi $do_alignment_check
 973         jmp error_code
 974         CFI_ENDPROC
 975 END(alignment_check)
 976
 977 ENTRY(divide_error)
 978         RING0_INT_FRAME
 979         pushl_cfi $0                    # no error code
 980         pushl_cfi $do_divide_error
 981         jmp error_code
 982         CFI_ENDPROC
 983 END(divide_error)
 984
 985 #ifdef CONFIG_X86_MCE
 986 ENTRY(machine_check)
 987         RING0_INT_FRAME
 988         pushl_cfi $0
 989         pushl_cfi machine_check_vector
 990         jmp error_code
 991         CFI_ENDPROC
 992 END(machine_check)
 993 #endif
 994
 995 ENTRY(spurious_interrupt_bug)
 996         RING0_INT_FRAME
 997         pushl_cfi $0
 998         pushl_cfi $do_spurious_interrupt_bug
 999         jmp error_code
1000         CFI_ENDPROC
1001 END(spurious_interrupt_bug)
1002 /*
1003  * End of kprobes section
1004  */
1005         .popsection
1006
1007 ENTRY(kernel_thread_helper)
1008         pushl $0                # fake return address for unwinder
1009         CFI_STARTPROC
1010         movl %edi,%eax
1011         call *%esi
1012         call do_exit
1013         ud2                     # padding for call trace
1014         CFI_ENDPROC
1015 ENDPROC(kernel_thread_helper)
1016
1017 #ifdef CONFIG_XEN
1018 /* Xen doesn't set %esp to be precisely what the normal sysenter
1019    entrypoint expects, so fix it up before using the normal path. */
1020 ENTRY(xen_sysenter_target)
1021         RING0_INT_FRAME
1022         addl $5*4, %esp         /* remove xen-provided frame */
1023         CFI_ADJUST_CFA_OFFSET -5*4
1024         jmp sysenter_past_esp
1025         CFI_ENDPROC
1026
1027 ENTRY(xen_hypervisor_callback)
1028         CFI_STARTPROC
1029         pushl_cfi $0
1030         SAVE_ALL
1031         TRACE_IRQS_OFF
1032
1033         /* Check to see if we got the event in the critical
1034            region in xen_iret_direct, after we've reenabled
1035            events and checked for pending events.  This simulates
1036            iret instruction's behaviour where it delivers a
1037            pending interrupt when enabling interrupts. */
1038         movl PT_EIP(%esp),%eax
1039         cmpl $xen_iret_start_crit,%eax
1040         jb   1f
1041         cmpl $xen_iret_end_crit,%eax
1042         jae  1f
1043
1044         jmp  xen_iret_crit_fixup
1045
1046 ENTRY(xen_do_upcall)
1047 1:      mov %esp, %eax
1048         call xen_evtchn_do_upcall
1049         jmp  ret_from_intr
1050         CFI_ENDPROC
1051 ENDPROC(xen_hypervisor_callback)
1052
1053 # Hypervisor uses this for application faults while it executes.
1054 # We get here for two reasons:
1055 #  1. Fault while reloading DS, ES, FS or GS
1056 #  2. Fault while executing IRET
1057 # Category 1 we fix up by reattempting the load, and zeroing the segment
1058 # register if the load fails.
1059 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1060 # normal Linux return path in this case because if we use the IRET hypercall
1061 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1062 # We distinguish between categories by maintaining a status value in EAX.
1063 ENTRY(xen_failsafe_callback)
1064         CFI_STARTPROC
1065         pushl_cfi %eax
1066         movl $1,%eax
1067 1:      mov 4(%esp),%ds
1068 2:      mov 8(%esp),%es
1069 3:      mov 12(%esp),%fs
1070 4:      mov 16(%esp),%gs
1071         testl %eax,%eax
1072         popl_cfi %eax
1073         lea 16(%esp),%esp
1074         CFI_ADJUST_CFA_OFFSET -16
1075         jz 5f
1076         addl $16,%esp
1077         jmp iret_exc            # EAX != 0 => Category 2 (Bad IRET)
1078 5:      pushl_cfi $0            # EAX == 0 => Category 1 (Bad segment)
1079         SAVE_ALL
1080         jmp ret_from_exception
1081         CFI_ENDPROC
1082
1083 .section .fixup,"ax"
1084 6:      xorl %eax,%eax
1085         movl %eax,4(%esp)
1086         jmp 1b
1087 7:      xorl %eax,%eax
1088         movl %eax,8(%esp)
1089         jmp 2b
1090 8:      xorl %eax,%eax
1091         movl %eax,12(%esp)
1092         jmp 3b
1093 9:      xorl %eax,%eax
1094         movl %eax,16(%esp)
1095         jmp 4b
1096 .previous
1097 .section __ex_table,"a"
1098         .align 4
1099         .long 1b,6b
1100         .long 2b,7b
1101         .long 3b,8b
1102         .long 4b,9b
1103 .previous
1104 ENDPROC(xen_failsafe_callback)
1105
1106 BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1107                 xen_evtchn_do_upcall)
1108
1109 #endif  /* CONFIG_XEN */
1110
1111 #ifdef CONFIG_FUNCTION_TRACER
1112 #ifdef CONFIG_DYNAMIC_FTRACE
1113
1114 ENTRY(mcount)
1115         ret
1116 END(mcount)
1117
1118 ENTRY(ftrace_caller)
1119         cmpl $0, function_trace_stop
1120         jne  ftrace_stub
1121
1122         pushl %eax
1123         pushl %ecx
1124         pushl %edx
1125         movl 0xc(%esp), %eax
1126         movl 0x4(%ebp), %edx
1127         subl $MCOUNT_INSN_SIZE, %eax
1128
1129 .globl ftrace_call
1130 ftrace_call:
1131         call ftrace_stub
1132
1133         popl %edx
1134         popl %ecx
1135         popl %eax
1136 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1137 .globl ftrace_graph_call
1138 ftrace_graph_call:
1139         jmp ftrace_stub
1140 #endif
1141
1142 .globl ftrace_stub
1143 ftrace_stub:
1144         ret
1145 END(ftrace_caller)
1146
1147 #else /* ! CONFIG_DYNAMIC_FTRACE */
1148
1149 ENTRY(mcount)
1150         cmpl $0, function_trace_stop
1151         jne  ftrace_stub
1152
1153         cmpl $ftrace_stub, ftrace_trace_function
1154         jnz trace
1155 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1156         cmpl $ftrace_stub, ftrace_graph_return
1157         jnz ftrace_graph_caller
1158
1159         cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1160         jnz ftrace_graph_caller
1161 #endif
1162 .globl ftrace_stub
1163 ftrace_stub:
1164         ret
1165
1166         /* taken from glibc */
1167 trace:
1168         pushl %eax
1169         pushl %ecx
1170         pushl %edx
1171         movl 0xc(%esp), %eax
1172         movl 0x4(%ebp), %edx
1173         subl $MCOUNT_INSN_SIZE, %eax
1174
1175         call *ftrace_trace_function
1176
1177         popl %edx
1178         popl %ecx
1179         popl %eax
1180         jmp ftrace_stub
1181 END(mcount)
1182 #endif /* CONFIG_DYNAMIC_FTRACE */
1183 #endif /* CONFIG_FUNCTION_TRACER */
1184
1185 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1186 ENTRY(ftrace_graph_caller)
1187         cmpl $0, function_trace_stop
1188         jne ftrace_stub
1189
1190         pushl %eax
1191         pushl %ecx
1192         pushl %edx
1193         movl 0xc(%esp), %edx
1194         lea 0x4(%ebp), %eax
1195         movl (%ebp), %ecx
1196         subl $MCOUNT_INSN_SIZE, %edx
1197         call prepare_ftrace_return
1198         popl %edx
1199         popl %ecx
1200         popl %eax
1201         ret
1202 END(ftrace_graph_caller)
1203
1204 .globl return_to_handler
1205 return_to_handler:
1206         pushl %eax
1207         pushl %edx
1208         movl %ebp, %eax
1209         call ftrace_return_to_handler
1210         movl %eax, %ecx
1211         popl %edx
1212         popl %eax
1213         jmp *%ecx
1214 #endif
1215
1216 .section .rodata,"a"
1217 #include "syscall_table_32.S"
1218
1219 syscall_table_size=(.-sys_call_table)
1220
1221 /*
1222  * Some functions should be protected against kprobes
1223  */
1224         .pushsection .kprobes.text, "ax"
1225
1226 ENTRY(page_fault)
1227         RING0_EC_FRAME
1228         pushl_cfi $do_page_fault
1229         ALIGN
1230 error_code:
1231         /* the function address is in %gs's slot on the stack */
1232         pushl_cfi %fs
1233         /*CFI_REL_OFFSET fs, 0*/
1234         pushl_cfi %es
1235         /*CFI_REL_OFFSET es, 0*/
1236         pushl_cfi %ds
1237         /*CFI_REL_OFFSET ds, 0*/
1238         pushl_cfi %eax
1239         CFI_REL_OFFSET eax, 0
1240         pushl_cfi %ebp
1241         CFI_REL_OFFSET ebp, 0
1242         pushl_cfi %edi
1243         CFI_REL_OFFSET edi, 0
1244         pushl_cfi %esi
1245         CFI_REL_OFFSET esi, 0
1246         pushl_cfi %edx
1247         CFI_REL_OFFSET edx, 0
1248         pushl_cfi %ecx
1249         CFI_REL_OFFSET ecx, 0
1250         pushl_cfi %ebx
1251         CFI_REL_OFFSET ebx, 0
1252         cld
1253         movl $(__KERNEL_PERCPU), %ecx
1254         movl %ecx, %fs
1255         UNWIND_ESPFIX_STACK
1256         GS_TO_REG %ecx
1257         movl PT_GS(%esp), %edi          # get the function address
1258         movl PT_ORIG_EAX(%esp), %edx    # get the error code
1259         movl $-1, PT_ORIG_EAX(%esp)     # no syscall to restart
1260         REG_TO_PTGS %ecx
1261         SET_KERNEL_GS %ecx
1262         movl $(__USER_DS), %ecx
1263         movl %ecx, %ds
1264         movl %ecx, %es
1265         TRACE_IRQS_OFF
1266         movl %esp,%eax                  # pt_regs pointer
1267         call *%edi
1268         jmp ret_from_exception
1269         CFI_ENDPROC
1270 END(page_fault)
1271
1272 /*
1273  * Debug traps and NMI can happen at the one SYSENTER instruction
1274  * that sets up the real kernel stack. Check here, since we can't
1275  * allow the wrong stack to be used.
1276  *
1277  * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1278  * already pushed 3 words if it hits on the sysenter instruction:
1279  * eflags, cs and eip.
1280  *
1281  * We just load the right stack, and push the three (known) values
1282  * by hand onto the new stack - while updating the return eip past
1283  * the instruction that would have done it for sysenter.
1284  */
1285 .macro FIX_STACK offset ok label
1286         cmpw $__KERNEL_CS, 4(%esp)
1287         jne \ok
1288 \label:
1289         movl TSS_sysenter_sp0 + \offset(%esp), %esp
1290         CFI_DEF_CFA esp, 0
1291         CFI_UNDEFINED eip
1292         pushfl_cfi
1293         pushl_cfi $__KERNEL_CS
1294         pushl_cfi $sysenter_past_esp
1295         CFI_REL_OFFSET eip, 0
1296 .endm
1297
1298 ENTRY(debug)
1299         RING0_INT_FRAME
1300         cmpl $ia32_sysenter_target,(%esp)
1301         jne debug_stack_correct
1302         FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1303 debug_stack_correct:
1304         pushl_cfi $-1                   # mark this as an int
1305         SAVE_ALL
1306         TRACE_IRQS_OFF
1307         xorl %edx,%edx                  # error code 0
1308         movl %esp,%eax                  # pt_regs pointer
1309         call do_debug
1310         jmp ret_from_exception
1311         CFI_ENDPROC
1312 END(debug)
1313
1314 /*
1315  * NMI is doubly nasty. It can happen _while_ we're handling
1316  * a debug fault, and the debug fault hasn't yet been able to
1317  * clear up the stack. So we first check whether we got  an
1318  * NMI on the sysenter entry path, but after that we need to
1319  * check whether we got an NMI on the debug path where the debug
1320  * fault happened on the sysenter path.
1321  */
1322 ENTRY(nmi)
1323         RING0_INT_FRAME
1324         pushl_cfi %eax
1325         movl %ss, %eax
1326         cmpw $__ESPFIX_SS, %ax
1327         popl_cfi %eax
1328         je nmi_espfix_stack
1329         cmpl $ia32_sysenter_target,(%esp)
1330         je nmi_stack_fixup
1331         pushl_cfi %eax
1332         movl %esp,%eax
1333         /* Do not access memory above the end of our stack page,
1334          * it might not exist.
1335          */
1336         andl $(THREAD_SIZE-1),%eax
1337         cmpl $(THREAD_SIZE-20),%eax
1338         popl_cfi %eax
1339         jae nmi_stack_correct
1340         cmpl $ia32_sysenter_target,12(%esp)
1341         je nmi_debug_stack_check
1342 nmi_stack_correct:
1343         /* We have a RING0_INT_FRAME here */
1344         pushl_cfi %eax
1345         SAVE_ALL
1346         xorl %edx,%edx          # zero error code
1347         movl %esp,%eax          # pt_regs pointer
1348         call do_nmi
1349         jmp restore_all_notrace
1350         CFI_ENDPROC
1351
1352 nmi_stack_fixup:
1353         RING0_INT_FRAME
1354         FIX_STACK 12, nmi_stack_correct, 1
1355         jmp nmi_stack_correct
1356
1357 nmi_debug_stack_check:
1358         /* We have a RING0_INT_FRAME here */
1359         cmpw $__KERNEL_CS,16(%esp)
1360         jne nmi_stack_correct
1361         cmpl $debug,(%esp)
1362         jb nmi_stack_correct
1363         cmpl $debug_esp_fix_insn,(%esp)
1364         ja nmi_stack_correct
1365         FIX_STACK 24, nmi_stack_correct, 1
1366         jmp nmi_stack_correct
1367
1368 nmi_espfix_stack:
1369         /* We have a RING0_INT_FRAME here.
1370          *
1371          * create the pointer to lss back
1372          */
1373         pushl_cfi %ss
1374         pushl_cfi %esp
1375         addl $4, (%esp)
1376         /* copy the iret frame of 12 bytes */
1377         .rept 3
1378         pushl_cfi 16(%esp)
1379         .endr
1380         pushl_cfi %eax
1381         SAVE_ALL
1382         FIXUP_ESPFIX_STACK              # %eax == %esp
1383         xorl %edx,%edx                  # zero error code
1384         call do_nmi
1385         RESTORE_REGS
1386         lss 12+4(%esp), %esp            # back to espfix stack
1387         CFI_ADJUST_CFA_OFFSET -24
1388         jmp irq_return
1389         CFI_ENDPROC
1390 END(nmi)
1391
1392 ENTRY(int3)
1393         RING0_INT_FRAME
1394         pushl_cfi $-1                   # mark this as an int
1395         SAVE_ALL
1396         TRACE_IRQS_OFF
1397         xorl %edx,%edx          # zero error code
1398         movl %esp,%eax          # pt_regs pointer
1399         call do_int3
1400         jmp ret_from_exception
1401         CFI_ENDPROC
1402 END(int3)
1403
1404 ENTRY(general_protection)
1405         RING0_EC_FRAME
1406         pushl_cfi $do_general_protection
1407         jmp error_code
1408         CFI_ENDPROC
1409 END(general_protection)
1410
1411 #ifdef CONFIG_KVM_GUEST
1412 ENTRY(async_page_fault)
1413         RING0_EC_FRAME
1414         pushl_cfi $do_async_page_fault
1415         jmp error_code
1416         CFI_ENDPROC
1417 END(async_page_fault)
1418 #endif
1419
1420 /*
1421  * End of kprobes section
1422  */
1423         .popsection