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