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