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