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) - orig_eax
  34  *      2C(%esp) - %eip
  35  *      30(%esp) - %cs
  36  *      34(%esp) - %eflags
  37  *      38(%esp) - %oldesp
  38  *      3C(%esp) - %oldss
  39  *
  40  * "current" is in register %ebx during any slow entries.
  41  */
  42
  43 #include <linux/linkage.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.h>
  50 #include <asm/desc.h>
  51 #include <asm/percpu.h>
  52 #include <asm/dwarf2.h>
  53 #include <asm/processor-flags.h>
  54 #include "irq_vectors.h"
  55
  56 /*
  57  * We use macros for low-level operations which need to be overridden
  58  * for paravirtualization.  The following will never clobber any registers:
  59  *   INTERRUPT_RETURN (aka. "iret")
  60  *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  61  *   ENABLE_INTERRUPTS_SYSCALL_RET (aka "sti; sysexit").
  62  *
  63  * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  64  * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  65  * Allowing a register to be clobbered can shrink the paravirt replacement
  66  * enough to patch inline, increasing performance.
  67  */
  68
  69 #define nr_syscalls ((syscall_table_size)/4)
  70
  71 #ifdef CONFIG_PREEMPT
  72 #define preempt_stop(clobbers)  DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  73 #else
  74 #define preempt_stop(clobbers)
  75 #define resume_kernel           restore_nocheck
  76 #endif
  77
  78 .macro TRACE_IRQS_IRET
  79 #ifdef CONFIG_TRACE_IRQFLAGS
  80         testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)     # interrupts off?
  81         jz 1f
  82         TRACE_IRQS_ON
  83 1:
  84 #endif
  85 .endm
  86
  87 #ifdef CONFIG_VM86
  88 #define resume_userspace_sig    check_userspace
  89 #else
  90 #define resume_userspace_sig    resume_userspace
  91 #endif
  92
  93 #define SAVE_ALL \
  94         cld; \
  95         pushl %fs; \
  96         CFI_ADJUST_CFA_OFFSET 4;\
  97         /*CFI_REL_OFFSET fs, 0;*/\
  98         pushl %es; \
  99         CFI_ADJUST_CFA_OFFSET 4;\
 100         /*CFI_REL_OFFSET es, 0;*/\
 101         pushl %ds; \
 102         CFI_ADJUST_CFA_OFFSET 4;\
 103         /*CFI_REL_OFFSET ds, 0;*/\
 104         pushl %eax; \
 105         CFI_ADJUST_CFA_OFFSET 4;\
 106         CFI_REL_OFFSET eax, 0;\
 107         pushl %ebp; \
 108         CFI_ADJUST_CFA_OFFSET 4;\
 109         CFI_REL_OFFSET ebp, 0;\
 110         pushl %edi; \
 111         CFI_ADJUST_CFA_OFFSET 4;\
 112         CFI_REL_OFFSET edi, 0;\
 113         pushl %esi; \
 114         CFI_ADJUST_CFA_OFFSET 4;\
 115         CFI_REL_OFFSET esi, 0;\
 116         pushl %edx; \
 117         CFI_ADJUST_CFA_OFFSET 4;\
 118         CFI_REL_OFFSET edx, 0;\
 119         pushl %ecx; \
 120         CFI_ADJUST_CFA_OFFSET 4;\
 121         CFI_REL_OFFSET ecx, 0;\
 122         pushl %ebx; \
 123         CFI_ADJUST_CFA_OFFSET 4;\
 124         CFI_REL_OFFSET ebx, 0;\
 125         movl $(__USER_DS), %edx; \
 126         movl %edx, %ds; \
 127         movl %edx, %es; \
 128         movl $(__KERNEL_PERCPU), %edx; \
 129         movl %edx, %fs
 130
 131 #define RESTORE_INT_REGS \
 132         popl %ebx;      \
 133         CFI_ADJUST_CFA_OFFSET -4;\
 134         CFI_RESTORE ebx;\
 135         popl %ecx;      \
 136         CFI_ADJUST_CFA_OFFSET -4;\
 137         CFI_RESTORE ecx;\
 138         popl %edx;      \
 139         CFI_ADJUST_CFA_OFFSET -4;\
 140         CFI_RESTORE edx;\
 141         popl %esi;      \
 142         CFI_ADJUST_CFA_OFFSET -4;\
 143         CFI_RESTORE esi;\
 144         popl %edi;      \
 145         CFI_ADJUST_CFA_OFFSET -4;\
 146         CFI_RESTORE edi;\
 147         popl %ebp;      \
 148         CFI_ADJUST_CFA_OFFSET -4;\
 149         CFI_RESTORE ebp;\
 150         popl %eax;      \
 151         CFI_ADJUST_CFA_OFFSET -4;\
 152         CFI_RESTORE eax
 153
 154 #define RESTORE_REGS    \
 155         RESTORE_INT_REGS; \
 156 1:      popl %ds;       \
 157         CFI_ADJUST_CFA_OFFSET -4;\
 158         /*CFI_RESTORE ds;*/\
 159 2:      popl %es;       \
 160         CFI_ADJUST_CFA_OFFSET -4;\
 161         /*CFI_RESTORE es;*/\
 162 3:      popl %fs;       \
 163         CFI_ADJUST_CFA_OFFSET -4;\
 164         /*CFI_RESTORE fs;*/\
 165 .pushsection .fixup,"ax";       \
 166 4:      movl $0,(%esp); \
 167         jmp 1b;         \
 168 5:      movl $0,(%esp); \
 169         jmp 2b;         \
 170 6:      movl $0,(%esp); \
 171         jmp 3b;         \
 172 .section __ex_table,"a";\
 173         .align 4;       \
 174         .long 1b,4b;    \
 175         .long 2b,5b;    \
 176         .long 3b,6b;    \
 177 .popsection
 178
 179 #define RING0_INT_FRAME \
 180         CFI_STARTPROC simple;\
 181         CFI_SIGNAL_FRAME;\
 182         CFI_DEF_CFA esp, 3*4;\
 183         /*CFI_OFFSET cs, -2*4;*/\
 184         CFI_OFFSET eip, -3*4
 185
 186 #define RING0_EC_FRAME \
 187         CFI_STARTPROC simple;\
 188         CFI_SIGNAL_FRAME;\
 189         CFI_DEF_CFA esp, 4*4;\
 190         /*CFI_OFFSET cs, -2*4;*/\
 191         CFI_OFFSET eip, -3*4
 192
 193 #define RING0_PTREGS_FRAME \
 194         CFI_STARTPROC simple;\
 195         CFI_SIGNAL_FRAME;\
 196         CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\
 197         /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\
 198         CFI_OFFSET eip, PT_EIP-PT_OLDESP;\
 199         /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\
 200         /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\
 201         CFI_OFFSET eax, PT_EAX-PT_OLDESP;\
 202         CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\
 203         CFI_OFFSET edi, PT_EDI-PT_OLDESP;\
 204         CFI_OFFSET esi, PT_ESI-PT_OLDESP;\
 205         CFI_OFFSET edx, PT_EDX-PT_OLDESP;\
 206         CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\
 207         CFI_OFFSET ebx, PT_EBX-PT_OLDESP
 208
 209 ENTRY(ret_from_fork)
 210         CFI_STARTPROC
 211         pushl %eax
 212         CFI_ADJUST_CFA_OFFSET 4
 213         call schedule_tail
 214         GET_THREAD_INFO(%ebp)
 215         popl %eax
 216         CFI_ADJUST_CFA_OFFSET -4
 217         pushl $0x0202                   # Reset kernel eflags
 218         CFI_ADJUST_CFA_OFFSET 4
 219         popfl
 220         CFI_ADJUST_CFA_OFFSET -4
 221         jmp syscall_exit
 222         CFI_ENDPROC
 223 END(ret_from_fork)
 224
 225 /*
 226  * Return to user mode is not as complex as all this looks,
 227  * but we want the default path for a system call return to
 228  * go as quickly as possible which is why some of this is
 229  * less clear than it otherwise should be.
 230  */
 231
 232         # userspace resumption stub bypassing syscall exit tracing
 233         ALIGN
 234         RING0_PTREGS_FRAME
 235 ret_from_exception:
 236         preempt_stop(CLBR_ANY)
 237 ret_from_intr:
 238         GET_THREAD_INFO(%ebp)
 239 check_userspace:
 240         movl PT_EFLAGS(%esp), %eax      # mix EFLAGS and CS
 241         movb PT_CS(%esp), %al
 242         andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 243         cmpl $USER_RPL, %eax
 244         jb resume_kernel                # not returning to v8086 or userspace
 245
 246 ENTRY(resume_userspace)
 247         LOCKDEP_SYS_EXIT
 248         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 249                                         # setting need_resched or sigpending
 250                                         # between sampling and the iret
 251         TRACE_IRQS_OFF
 252         movl TI_flags(%ebp), %ecx
 253         andl $_TIF_WORK_MASK, %ecx      # is there any work to be done on
 254                                         # int/exception return?
 255         jne work_pending
 256         jmp restore_all
 257 END(ret_from_exception)
 258
 259 #ifdef CONFIG_PREEMPT
 260 ENTRY(resume_kernel)
 261         DISABLE_INTERRUPTS(CLBR_ANY)
 262         cmpl $0,TI_preempt_count(%ebp)  # non-zero preempt_count ?
 263         jnz restore_nocheck
 264 need_resched:
 265         movl TI_flags(%ebp), %ecx       # need_resched set ?
 266         testb $_TIF_NEED_RESCHED, %cl
 267         jz restore_all
 268         testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)    # interrupts off (exception path) ?
 269         jz restore_all
 270         call preempt_schedule_irq
 271         jmp need_resched
 272 END(resume_kernel)
 273 #endif
 274         CFI_ENDPROC
 275
 276 /* SYSENTER_RETURN points to after the "sysenter" instruction in
 277    the vsyscall page.  See vsyscall-sysentry.S, which defines the symbol.  */
 278
 279         # sysenter call handler stub
 280 ENTRY(ia32_sysenter_target)
 281         CFI_STARTPROC simple
 282         CFI_SIGNAL_FRAME
 283         CFI_DEF_CFA esp, 0
 284         CFI_REGISTER esp, ebp
 285         movl TSS_sysenter_sp0(%esp),%esp
 286 sysenter_past_esp:
 287         /*
 288          * Interrupts are disabled here, but we can't trace it until
 289          * enough kernel state to call TRACE_IRQS_OFF can be called - but
 290          * we immediately enable interrupts at that point anyway.
 291          */
 292         pushl $(__USER_DS)
 293         CFI_ADJUST_CFA_OFFSET 4
 294         /*CFI_REL_OFFSET ss, 0*/
 295         pushl %ebp
 296         CFI_ADJUST_CFA_OFFSET 4
 297         CFI_REL_OFFSET esp, 0
 298         pushfl
 299         orl $X86_EFLAGS_IF, (%esp)
 300         CFI_ADJUST_CFA_OFFSET 4
 301         pushl $(__USER_CS)
 302         CFI_ADJUST_CFA_OFFSET 4
 303         /*CFI_REL_OFFSET cs, 0*/
 304         /*
 305          * Push current_thread_info()->sysenter_return to the stack.
 306          * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
 307          * pushed above; +8 corresponds to copy_thread's esp0 setting.
 308          */
 309         pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
 310         CFI_ADJUST_CFA_OFFSET 4
 311         CFI_REL_OFFSET eip, 0
 312
 313         pushl %eax
 314         CFI_ADJUST_CFA_OFFSET 4
 315         SAVE_ALL
 316         ENABLE_INTERRUPTS(CLBR_NONE)
 317
 318 /*
 319  * Load the potential sixth argument from user stack.
 320  * Careful about security.
 321  */
 322         cmpl $__PAGE_OFFSET-3,%ebp
 323         jae syscall_fault
 324 1:      movl (%ebp),%ebp
 325         movl %ebp,PT_EBP(%esp)
 326 .section __ex_table,"a"
 327         .align 4
 328         .long 1b,syscall_fault
 329 .previous
 330
 331         GET_THREAD_INFO(%ebp)
 332
 333         /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
 334         testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
 335         jnz syscall_trace_entry
 336         cmpl $(nr_syscalls), %eax
 337         jae syscall_badsys
 338         call *sys_call_table(,%eax,4)
 339         movl %eax,PT_EAX(%esp)
 340         LOCKDEP_SYS_EXIT
 341         DISABLE_INTERRUPTS(CLBR_ANY)
 342         TRACE_IRQS_OFF
 343         movl TI_flags(%ebp), %ecx
 344         testw $_TIF_ALLWORK_MASK, %cx
 345         jne syscall_exit_work
 346 /* if something modifies registers it must also disable sysexit */
 347         movl PT_EIP(%esp), %edx
 348         movl PT_OLDESP(%esp), %ecx
 349         xorl %ebp,%ebp
 350         TRACE_IRQS_ON
 351 1:      mov  PT_FS(%esp), %fs
 352         ENABLE_INTERRUPTS_SYSCALL_RET
 353         CFI_ENDPROC
 354 .pushsection .fixup,"ax"
 355 2:      movl $0,PT_FS(%esp)
 356         jmp 1b
 357 .section __ex_table,"a"
 358         .align 4
 359         .long 1b,2b
 360 .popsection
 361 ENDPROC(ia32_sysenter_target)
 362
 363         # system call handler stub
 364 ENTRY(system_call)
 365         RING0_INT_FRAME                 # can't unwind into user space anyway
 366         pushl %eax                      # save orig_eax
 367         CFI_ADJUST_CFA_OFFSET 4
 368         SAVE_ALL
 369         GET_THREAD_INFO(%ebp)
 370                                         # system call tracing in operation / emulation
 371         /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
 372         testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
 373         jnz syscall_trace_entry
 374         cmpl $(nr_syscalls), %eax
 375         jae syscall_badsys
 376 syscall_call:
 377         call *sys_call_table(,%eax,4)
 378         movl %eax,PT_EAX(%esp)          # store the return value
 379 syscall_exit:
 380         LOCKDEP_SYS_EXIT
 381         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 382                                         # setting need_resched or sigpending
 383                                         # between sampling and the iret
 384         TRACE_IRQS_OFF
 385         testl $X86_EFLAGS_TF,PT_EFLAGS(%esp)    # If tracing set singlestep flag on exit
 386         jz no_singlestep
 387         orl $_TIF_SINGLESTEP,TI_flags(%ebp)
 388 no_singlestep:
 389         movl TI_flags(%ebp), %ecx
 390         testw $_TIF_ALLWORK_MASK, %cx   # current->work
 391         jne syscall_exit_work
 392
 393 restore_all:
 394         movl PT_EFLAGS(%esp), %eax      # mix EFLAGS, SS and CS
 395         # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 396         # are returning to the kernel.
 397         # See comments in process.c:copy_thread() for details.
 398         movb PT_OLDSS(%esp), %ah
 399         movb PT_CS(%esp), %al
 400         andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 401         cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
 402         CFI_REMEMBER_STATE
 403         je ldt_ss                       # returning to user-space with LDT SS
 404 restore_nocheck:
 405         TRACE_IRQS_IRET
 406 restore_nocheck_notrace:
 407         RESTORE_REGS
 408         addl $4, %esp                   # skip orig_eax/error_code
 409         CFI_ADJUST_CFA_OFFSET -4
 410 irq_return:
 411         INTERRUPT_RETURN
 412 .section .fixup,"ax"
 413 ENTRY(iret_exc)
 414         pushl $0                        # no error code
 415         pushl $do_iret_error
 416         jmp error_code
 417 .previous
 418 .section __ex_table,"a"
 419         .align 4
 420         .long irq_return,iret_exc
 421 .previous
 422
 423         CFI_RESTORE_STATE
 424 ldt_ss:
 425         larl PT_OLDSS(%esp), %eax
 426         jnz restore_nocheck
 427         testl $0x00400000, %eax         # returning to 32bit stack?
 428         jnz restore_nocheck             # allright, normal return
 429
 430 #ifdef CONFIG_PARAVIRT
 431         /*
 432          * The kernel can't run on a non-flat stack if paravirt mode
 433          * is active.  Rather than try to fixup the high bits of
 434          * ESP, bypass this code entirely.  This may break DOSemu
 435          * and/or Wine support in a paravirt VM, although the option
 436          * is still available to implement the setting of the high
 437          * 16-bits in the INTERRUPT_RETURN paravirt-op.
 438          */
 439         cmpl $0, pv_info+PARAVIRT_enabled
 440         jne restore_nocheck
 441 #endif
 442
 443         /* If returning to userspace with 16bit stack,
 444          * try to fix the higher word of ESP, as the CPU
 445          * won't restore it.
 446          * This is an "official" bug of all the x86-compatible
 447          * CPUs, which we can try to work around to make
 448          * dosemu and wine happy. */
 449         movl PT_OLDESP(%esp), %eax
 450         movl %esp, %edx
 451         call patch_espfix_desc
 452         pushl $__ESPFIX_SS
 453         CFI_ADJUST_CFA_OFFSET 4
 454         pushl %eax
 455         CFI_ADJUST_CFA_OFFSET 4
 456         DISABLE_INTERRUPTS(CLBR_EAX)
 457         TRACE_IRQS_OFF
 458         lss (%esp), %esp
 459         CFI_ADJUST_CFA_OFFSET -8
 460         jmp restore_nocheck
 461         CFI_ENDPROC
 462 ENDPROC(system_call)
 463
 464         # perform work that needs to be done immediately before resumption
 465         ALIGN
 466         RING0_PTREGS_FRAME              # can't unwind into user space anyway
 467 work_pending:
 468         testb $_TIF_NEED_RESCHED, %cl
 469         jz work_notifysig
 470 work_resched:
 471         call schedule
 472         LOCKDEP_SYS_EXIT
 473         DISABLE_INTERRUPTS(CLBR_ANY)    # make sure we don't miss an interrupt
 474                                         # setting need_resched or sigpending
 475                                         # between sampling and the iret
 476         TRACE_IRQS_OFF
 477         movl TI_flags(%ebp), %ecx
 478         andl $_TIF_WORK_MASK, %ecx      # is there any work to be done other
 479                                         # than syscall tracing?
 480         jz restore_all
 481         testb $_TIF_NEED_RESCHED, %cl
 482         jnz work_resched
 483
 484 work_notifysig:                         # deal with pending signals and
 485                                         # notify-resume requests
 486 #ifdef CONFIG_VM86
 487         testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
 488         movl %esp, %eax
 489         jne work_notifysig_v86          # returning to kernel-space or
 490                                         # vm86-space
 491         xorl %edx, %edx
 492         call do_notify_resume
 493         jmp resume_userspace_sig
 494
 495         ALIGN
 496 work_notifysig_v86:
 497         pushl %ecx                      # save ti_flags for do_notify_resume
 498         CFI_ADJUST_CFA_OFFSET 4
 499         call save_v86_state             # %eax contains pt_regs pointer
 500         popl %ecx
 501         CFI_ADJUST_CFA_OFFSET -4
 502         movl %eax, %esp
 503 #else
 504         movl %esp, %eax
 505 #endif
 506         xorl %edx, %edx
 507         call do_notify_resume
 508         jmp resume_userspace_sig
 509 END(work_pending)
 510
 511         # perform syscall exit tracing
 512         ALIGN
 513 syscall_trace_entry:
 514         movl $-ENOSYS,PT_EAX(%esp)
 515         movl %esp, %eax
 516         xorl %edx,%edx
 517         call do_syscall_trace
 518         cmpl $0, %eax
 519         jne resume_userspace            # ret != 0 -> running under PTRACE_SYSEMU,
 520                                         # so must skip actual syscall
 521         movl PT_ORIG_EAX(%esp), %eax
 522         cmpl $(nr_syscalls), %eax
 523         jnae syscall_call
 524         jmp syscall_exit
 525 END(syscall_trace_entry)
 526
 527         # perform syscall exit tracing
 528         ALIGN
 529 syscall_exit_work:
 530         testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl
 531         jz work_pending
 532         TRACE_IRQS_ON
 533         ENABLE_INTERRUPTS(CLBR_ANY)     # could let do_syscall_trace() call
 534                                         # schedule() instead
 535         movl %esp, %eax
 536         movl $1, %edx
 537         call do_syscall_trace
 538         jmp resume_userspace
 539 END(syscall_exit_work)
 540         CFI_ENDPROC
 541
 542         RING0_INT_FRAME                 # can't unwind into user space anyway
 543 syscall_fault:
 544         GET_THREAD_INFO(%ebp)
 545         movl $-EFAULT,PT_EAX(%esp)
 546         jmp resume_userspace
 547 END(syscall_fault)
 548
 549 syscall_badsys:
 550         movl $-ENOSYS,PT_EAX(%esp)
 551         jmp resume_userspace
 552 END(syscall_badsys)
 553         CFI_ENDPROC
 554
 555 #define FIXUP_ESPFIX_STACK \
 556         /* since we are on a wrong stack, we cant make it a C code :( */ \
 557         PER_CPU(gdt_page, %ebx); \
 558         GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \
 559         addl %esp, %eax; \
 560         pushl $__KERNEL_DS; \
 561         CFI_ADJUST_CFA_OFFSET 4; \
 562         pushl %eax; \
 563         CFI_ADJUST_CFA_OFFSET 4; \
 564         lss (%esp), %esp; \
 565         CFI_ADJUST_CFA_OFFSET -8;
 566 #define UNWIND_ESPFIX_STACK \
 567         movl %ss, %eax; \
 568         /* see if on espfix stack */ \
 569         cmpw $__ESPFIX_SS, %ax; \
 570         jne 27f; \
 571         movl $__KERNEL_DS, %eax; \
 572         movl %eax, %ds; \
 573         movl %eax, %es; \
 574         /* switch to normal stack */ \
 575         FIXUP_ESPFIX_STACK; \
 576 27:;
 577
 578 /*
 579  * Build the entry stubs and pointer table with
 580  * some assembler magic.
 581  */
 582 .section .rodata,"a"
 583 ENTRY(interrupt)
 584 .text
 585
 586 ENTRY(irq_entries_start)
 587         RING0_INT_FRAME
 588 vector=0
 589 .rept NR_IRQS
 590         ALIGN
 591  .if vector
 592         CFI_ADJUST_CFA_OFFSET -4
 593  .endif
 594 1:      pushl $~(vector)
 595         CFI_ADJUST_CFA_OFFSET 4
 596         jmp common_interrupt
 597  .previous
 598         .long 1b
 599  .text
 600 vector=vector+1
 601 .endr
 602 END(irq_entries_start)
 603
 604 .previous
 605 END(interrupt)
 606 .previous
 607
 608 /*
 609  * the CPU automatically disables interrupts when executing an IRQ vector,
 610  * so IRQ-flags tracing has to follow that:
 611  */
 612         ALIGN
 613 common_interrupt:
 614         SAVE_ALL
 615         TRACE_IRQS_OFF
 616         movl %esp,%eax
 617         call do_IRQ
 618         jmp ret_from_intr
 619 ENDPROC(common_interrupt)
 620         CFI_ENDPROC
 621
 622 #define BUILD_INTERRUPT(name, nr)       \
 623 ENTRY(name)                             \
 624         RING0_INT_FRAME;                \
 625         pushl $~(nr);                   \
 626         CFI_ADJUST_CFA_OFFSET 4;        \
 627         SAVE_ALL;                       \
 628         TRACE_IRQS_OFF                  \
 629         movl %esp,%eax;                 \
 630         call smp_##name;                \
 631         jmp ret_from_intr;              \
 632         CFI_ENDPROC;                    \
 633 ENDPROC(name)
 634
 635 /* The include is where all of the SMP etc. interrupts come from */
 636 #include "entry_arch.h"
 637
 638 KPROBE_ENTRY(page_fault)
 639         RING0_EC_FRAME
 640         pushl $do_page_fault
 641         CFI_ADJUST_CFA_OFFSET 4
 642         ALIGN
 643 error_code:
 644         /* the function address is in %fs's slot on the stack */
 645         pushl %es
 646         CFI_ADJUST_CFA_OFFSET 4
 647         /*CFI_REL_OFFSET es, 0*/
 648         pushl %ds
 649         CFI_ADJUST_CFA_OFFSET 4
 650         /*CFI_REL_OFFSET ds, 0*/
 651         pushl %eax
 652         CFI_ADJUST_CFA_OFFSET 4
 653         CFI_REL_OFFSET eax, 0
 654         pushl %ebp
 655         CFI_ADJUST_CFA_OFFSET 4
 656         CFI_REL_OFFSET ebp, 0
 657         pushl %edi
 658         CFI_ADJUST_CFA_OFFSET 4
 659         CFI_REL_OFFSET edi, 0
 660         pushl %esi
 661         CFI_ADJUST_CFA_OFFSET 4
 662         CFI_REL_OFFSET esi, 0
 663         pushl %edx
 664         CFI_ADJUST_CFA_OFFSET 4
 665         CFI_REL_OFFSET edx, 0
 666         pushl %ecx
 667         CFI_ADJUST_CFA_OFFSET 4
 668         CFI_REL_OFFSET ecx, 0
 669         pushl %ebx
 670         CFI_ADJUST_CFA_OFFSET 4
 671         CFI_REL_OFFSET ebx, 0
 672         cld
 673         pushl %fs
 674         CFI_ADJUST_CFA_OFFSET 4
 675         /*CFI_REL_OFFSET fs, 0*/
 676         movl $(__KERNEL_PERCPU), %ecx
 677         movl %ecx, %fs
 678         UNWIND_ESPFIX_STACK
 679         popl %ecx
 680         CFI_ADJUST_CFA_OFFSET -4
 681         /*CFI_REGISTER es, ecx*/
 682         movl PT_FS(%esp), %edi          # get the function address
 683         movl PT_ORIG_EAX(%esp), %edx    # get the error code
 684         movl $-1, PT_ORIG_EAX(%esp)     # no syscall to restart
 685         mov  %ecx, PT_FS(%esp)
 686         /*CFI_REL_OFFSET fs, ES*/
 687         movl $(__USER_DS), %ecx
 688         movl %ecx, %ds
 689         movl %ecx, %es
 690         movl %esp,%eax                  # pt_regs pointer
 691         call *%edi
 692         jmp ret_from_exception
 693         CFI_ENDPROC
 694 KPROBE_END(page_fault)
 695
 696 ENTRY(coprocessor_error)
 697         RING0_INT_FRAME
 698         pushl $0
 699         CFI_ADJUST_CFA_OFFSET 4
 700         pushl $do_coprocessor_error
 701         CFI_ADJUST_CFA_OFFSET 4
 702         jmp error_code
 703         CFI_ENDPROC
 704 END(coprocessor_error)
 705
 706 ENTRY(simd_coprocessor_error)
 707         RING0_INT_FRAME
 708         pushl $0
 709         CFI_ADJUST_CFA_OFFSET 4
 710         pushl $do_simd_coprocessor_error
 711         CFI_ADJUST_CFA_OFFSET 4
 712         jmp error_code
 713         CFI_ENDPROC
 714 END(simd_coprocessor_error)
 715
 716 ENTRY(device_not_available)
 717         RING0_INT_FRAME
 718         pushl $-1                       # mark this as an int
 719         CFI_ADJUST_CFA_OFFSET 4
 720         SAVE_ALL
 721         GET_CR0_INTO_EAX
 722         testl $0x4, %eax                # EM (math emulation bit)
 723         jne device_not_available_emulate
 724         preempt_stop(CLBR_ANY)
 725         call math_state_restore
 726         jmp ret_from_exception
 727 device_not_available_emulate:
 728         pushl $0                        # temporary storage for ORIG_EIP
 729         CFI_ADJUST_CFA_OFFSET 4
 730         call math_emulate
 731         addl $4, %esp
 732         CFI_ADJUST_CFA_OFFSET -4
 733         jmp ret_from_exception
 734         CFI_ENDPROC
 735 END(device_not_available)
 736
 737 /*
 738  * Debug traps and NMI can happen at the one SYSENTER instruction
 739  * that sets up the real kernel stack. Check here, since we can't
 740  * allow the wrong stack to be used.
 741  *
 742  * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
 743  * already pushed 3 words if it hits on the sysenter instruction:
 744  * eflags, cs and eip.
 745  *
 746  * We just load the right stack, and push the three (known) values
 747  * by hand onto the new stack - while updating the return eip past
 748  * the instruction that would have done it for sysenter.
 749  */
 750 #define FIX_STACK(offset, ok, label)            \
 751         cmpw $__KERNEL_CS,4(%esp);              \
 752         jne ok;                                 \
 753 label:                                          \
 754         movl TSS_sysenter_sp0+offset(%esp),%esp;        \
 755         CFI_DEF_CFA esp, 0;                     \
 756         CFI_UNDEFINED eip;                      \
 757         pushfl;                                 \
 758         CFI_ADJUST_CFA_OFFSET 4;                \
 759         pushl $__KERNEL_CS;                     \
 760         CFI_ADJUST_CFA_OFFSET 4;                \
 761         pushl $sysenter_past_esp;               \
 762         CFI_ADJUST_CFA_OFFSET 4;                \
 763         CFI_REL_OFFSET eip, 0
 764
 765 KPROBE_ENTRY(debug)
 766         RING0_INT_FRAME
 767         cmpl $ia32_sysenter_target,(%esp)
 768         jne debug_stack_correct
 769         FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
 770 debug_stack_correct:
 771         pushl $-1                       # mark this as an int
 772         CFI_ADJUST_CFA_OFFSET 4
 773         SAVE_ALL
 774         xorl %edx,%edx                  # error code 0
 775         movl %esp,%eax                  # pt_regs pointer
 776         call do_debug
 777         jmp ret_from_exception
 778         CFI_ENDPROC
 779 KPROBE_END(debug)
 780
 781 /*
 782  * NMI is doubly nasty. It can happen _while_ we're handling
 783  * a debug fault, and the debug fault hasn't yet been able to
 784  * clear up the stack. So we first check whether we got  an
 785  * NMI on the sysenter entry path, but after that we need to
 786  * check whether we got an NMI on the debug path where the debug
 787  * fault happened on the sysenter path.
 788  */
 789 KPROBE_ENTRY(nmi)
 790         RING0_INT_FRAME
 791         pushl %eax
 792         CFI_ADJUST_CFA_OFFSET 4
 793         movl %ss, %eax
 794         cmpw $__ESPFIX_SS, %ax
 795         popl %eax
 796         CFI_ADJUST_CFA_OFFSET -4
 797         je nmi_espfix_stack
 798         cmpl $ia32_sysenter_target,(%esp)
 799         je nmi_stack_fixup
 800         pushl %eax
 801         CFI_ADJUST_CFA_OFFSET 4
 802         movl %esp,%eax
 803         /* Do not access memory above the end of our stack page,
 804          * it might not exist.
 805          */
 806         andl $(THREAD_SIZE-1),%eax
 807         cmpl $(THREAD_SIZE-20),%eax
 808         popl %eax
 809         CFI_ADJUST_CFA_OFFSET -4
 810         jae nmi_stack_correct
 811         cmpl $ia32_sysenter_target,12(%esp)
 812         je nmi_debug_stack_check
 813 nmi_stack_correct:
 814         /* We have a RING0_INT_FRAME here */
 815         pushl %eax
 816         CFI_ADJUST_CFA_OFFSET 4
 817         SAVE_ALL
 818         xorl %edx,%edx          # zero error code
 819         movl %esp,%eax          # pt_regs pointer
 820         call do_nmi
 821         jmp restore_nocheck_notrace
 822         CFI_ENDPROC
 823
 824 nmi_stack_fixup:
 825         RING0_INT_FRAME
 826         FIX_STACK(12,nmi_stack_correct, 1)
 827         jmp nmi_stack_correct
 828
 829 nmi_debug_stack_check:
 830         /* We have a RING0_INT_FRAME here */
 831         cmpw $__KERNEL_CS,16(%esp)
 832         jne nmi_stack_correct
 833         cmpl $debug,(%esp)
 834         jb nmi_stack_correct
 835         cmpl $debug_esp_fix_insn,(%esp)
 836         ja nmi_stack_correct
 837         FIX_STACK(24,nmi_stack_correct, 1)
 838         jmp nmi_stack_correct
 839
 840 nmi_espfix_stack:
 841         /* We have a RING0_INT_FRAME here.
 842          *
 843          * create the pointer to lss back
 844          */
 845         pushl %ss
 846         CFI_ADJUST_CFA_OFFSET 4
 847         pushl %esp
 848         CFI_ADJUST_CFA_OFFSET 4
 849         addw $4, (%esp)
 850         /* copy the iret frame of 12 bytes */
 851         .rept 3
 852         pushl 16(%esp)
 853         CFI_ADJUST_CFA_OFFSET 4
 854         .endr
 855         pushl %eax
 856         CFI_ADJUST_CFA_OFFSET 4
 857         SAVE_ALL
 858         FIXUP_ESPFIX_STACK              # %eax == %esp
 859         xorl %edx,%edx                  # zero error code
 860         call do_nmi
 861         RESTORE_REGS
 862         lss 12+4(%esp), %esp            # back to espfix stack
 863         CFI_ADJUST_CFA_OFFSET -24
 864         jmp irq_return
 865         CFI_ENDPROC
 866 KPROBE_END(nmi)
 867
 868 #ifdef CONFIG_PARAVIRT
 869 ENTRY(native_iret)
 870         iret
 871 .section __ex_table,"a"
 872         .align 4
 873         .long native_iret, iret_exc
 874 .previous
 875 END(native_iret)
 876
 877 ENTRY(native_irq_enable_syscall_ret)
 878         sti
 879         sysexit
 880 END(native_irq_enable_syscall_ret)
 881 #endif
 882
 883 KPROBE_ENTRY(int3)
 884         RING0_INT_FRAME
 885         pushl $-1                       # mark this as an int
 886         CFI_ADJUST_CFA_OFFSET 4
 887         SAVE_ALL
 888         xorl %edx,%edx          # zero error code
 889         movl %esp,%eax          # pt_regs pointer
 890         call do_int3
 891         jmp ret_from_exception
 892         CFI_ENDPROC
 893 KPROBE_END(int3)
 894
 895 ENTRY(overflow)
 896         RING0_INT_FRAME
 897         pushl $0
 898         CFI_ADJUST_CFA_OFFSET 4
 899         pushl $do_overflow
 900         CFI_ADJUST_CFA_OFFSET 4
 901         jmp error_code
 902         CFI_ENDPROC
 903 END(overflow)
 904
 905 ENTRY(bounds)
 906         RING0_INT_FRAME
 907         pushl $0
 908         CFI_ADJUST_CFA_OFFSET 4
 909         pushl $do_bounds
 910         CFI_ADJUST_CFA_OFFSET 4
 911         jmp error_code
 912         CFI_ENDPROC
 913 END(bounds)
 914
 915 ENTRY(invalid_op)
 916         RING0_INT_FRAME
 917         pushl $0
 918         CFI_ADJUST_CFA_OFFSET 4
 919         pushl $do_invalid_op
 920         CFI_ADJUST_CFA_OFFSET 4
 921         jmp error_code
 922         CFI_ENDPROC
 923 END(invalid_op)
 924
 925 ENTRY(coprocessor_segment_overrun)
 926         RING0_INT_FRAME
 927         pushl $0
 928         CFI_ADJUST_CFA_OFFSET 4
 929         pushl $do_coprocessor_segment_overrun
 930         CFI_ADJUST_CFA_OFFSET 4
 931         jmp error_code
 932         CFI_ENDPROC
 933 END(coprocessor_segment_overrun)
 934
 935 ENTRY(invalid_TSS)
 936         RING0_EC_FRAME
 937         pushl $do_invalid_TSS
 938         CFI_ADJUST_CFA_OFFSET 4
 939         jmp error_code
 940         CFI_ENDPROC
 941 END(invalid_TSS)
 942
 943 ENTRY(segment_not_present)
 944         RING0_EC_FRAME
 945         pushl $do_segment_not_present
 946         CFI_ADJUST_CFA_OFFSET 4
 947         jmp error_code
 948         CFI_ENDPROC
 949 END(segment_not_present)
 950
 951 ENTRY(stack_segment)
 952         RING0_EC_FRAME
 953         pushl $do_stack_segment
 954         CFI_ADJUST_CFA_OFFSET 4
 955         jmp error_code
 956         CFI_ENDPROC
 957 END(stack_segment)
 958
 959 KPROBE_ENTRY(general_protection)
 960         RING0_EC_FRAME
 961         pushl $do_general_protection
 962         CFI_ADJUST_CFA_OFFSET 4
 963         jmp error_code
 964         CFI_ENDPROC
 965 KPROBE_END(general_protection)
 966
 967 ENTRY(alignment_check)
 968         RING0_EC_FRAME
 969         pushl $do_alignment_check
 970         CFI_ADJUST_CFA_OFFSET 4
 971         jmp error_code
 972         CFI_ENDPROC
 973 END(alignment_check)
 974
 975 ENTRY(divide_error)
 976         RING0_INT_FRAME
 977         pushl $0                        # no error code
 978         CFI_ADJUST_CFA_OFFSET 4
 979         pushl $do_divide_error
 980         CFI_ADJUST_CFA_OFFSET 4
 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 $0
 989         CFI_ADJUST_CFA_OFFSET 4
 990         pushl machine_check_vector
 991         CFI_ADJUST_CFA_OFFSET 4
 992         jmp error_code
 993         CFI_ENDPROC
 994 END(machine_check)
 995 #endif
 996
 997 ENTRY(spurious_interrupt_bug)
 998         RING0_INT_FRAME
 999         pushl $0
1000         CFI_ADJUST_CFA_OFFSET 4
1001         pushl $do_spurious_interrupt_bug
1002         CFI_ADJUST_CFA_OFFSET 4
1003         jmp error_code
1004         CFI_ENDPROC
1005 END(spurious_interrupt_bug)
1006
1007 ENTRY(kernel_thread_helper)
1008         pushl $0                # fake return address for unwinder
1009         CFI_STARTPROC
1010         movl %edx,%eax
1011         push %edx
1012         CFI_ADJUST_CFA_OFFSET 4
1013         call *%ebx
1014         push %eax
1015         CFI_ADJUST_CFA_OFFSET 4
1016         call do_exit
1017         CFI_ENDPROC
1018 ENDPROC(kernel_thread_helper)
1019
1020 #ifdef CONFIG_XEN
1021 /* Xen doesn't set %esp to be precisely what the normal sysenter
1022    entrypoint expects, so fix it up before using the normal path. */
1023 ENTRY(xen_sysenter_target)
1024         RING0_INT_FRAME
1025         addl $5*4, %esp         /* remove xen-provided frame */
1026         jmp sysenter_past_esp
1027
1028 ENTRY(xen_hypervisor_callback)
1029         CFI_STARTPROC
1030         pushl $0
1031         CFI_ADJUST_CFA_OFFSET 4
1032         SAVE_ALL
1033         TRACE_IRQS_OFF
1034
1035         /* Check to see if we got the event in the critical
1036            region in xen_iret_direct, after we've reenabled
1037            events and checked for pending events.  This simulates
1038            iret instruction's behaviour where it delivers a
1039            pending interrupt when enabling interrupts. */
1040         movl PT_EIP(%esp),%eax
1041         cmpl $xen_iret_start_crit,%eax
1042         jb   1f
1043         cmpl $xen_iret_end_crit,%eax
1044         jae  1f
1045
1046         jmp  xen_iret_crit_fixup
1047
1048 ENTRY(xen_do_upcall)
1049 1:      mov %esp, %eax
1050         call xen_evtchn_do_upcall
1051         jmp  ret_from_intr
1052         CFI_ENDPROC
1053 ENDPROC(xen_hypervisor_callback)
1054
1055 # Hypervisor uses this for application faults while it executes.
1056 # We get here for two reasons:
1057 #  1. Fault while reloading DS, ES, FS or GS
1058 #  2. Fault while executing IRET
1059 # Category 1 we fix up by reattempting the load, and zeroing the segment
1060 # register if the load fails.
1061 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
1062 # normal Linux return path in this case because if we use the IRET hypercall
1063 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1064 # We distinguish between categories by maintaining a status value in EAX.
1065 ENTRY(xen_failsafe_callback)
1066         CFI_STARTPROC
1067         pushl %eax
1068         CFI_ADJUST_CFA_OFFSET 4
1069         movl $1,%eax
1070 1:      mov 4(%esp),%ds
1071 2:      mov 8(%esp),%es
1072 3:      mov 12(%esp),%fs
1073 4:      mov 16(%esp),%gs
1074         testl %eax,%eax
1075         popl %eax
1076         CFI_ADJUST_CFA_OFFSET -4
1077         lea 16(%esp),%esp
1078         CFI_ADJUST_CFA_OFFSET -16
1079         jz 5f
1080         addl $16,%esp
1081         jmp iret_exc            # EAX != 0 => Category 2 (Bad IRET)
1082 5:      pushl $0                # EAX == 0 => Category 1 (Bad segment)
1083         CFI_ADJUST_CFA_OFFSET 4
1084         SAVE_ALL
1085         jmp ret_from_exception
1086         CFI_ENDPROC
1087
1088 .section .fixup,"ax"
1089 6:      xorl %eax,%eax
1090         movl %eax,4(%esp)
1091         jmp 1b
1092 7:      xorl %eax,%eax
1093         movl %eax,8(%esp)
1094         jmp 2b
1095 8:      xorl %eax,%eax
1096         movl %eax,12(%esp)
1097         jmp 3b
1098 9:      xorl %eax,%eax
1099         movl %eax,16(%esp)
1100         jmp 4b
1101 .previous
1102 .section __ex_table,"a"
1103         .align 4
1104         .long 1b,6b
1105         .long 2b,7b
1106         .long 3b,8b
1107         .long 4b,9b
1108 .previous
1109 ENDPROC(xen_failsafe_callback)
1110
1111 #endif  /* CONFIG_XEN */
1112
1113 .section .rodata,"a"
1114 #include "syscall_table_32.S"
1115
1116 syscall_table_size=(.-sys_call_table)