arch/x86/entry/entry_32.S

   1 /*
   2  *  Copyright (C) 1991,1992  Linus Torvalds
   3  *
   4  * entry_32.S contains the system-call and low-level fault and trap handling routines.
   5  *
   6  * Stack layout while running C code:
   7  *      ptrace needs to have all registers on the stack.
   8  *      If the order here is changed, it needs to be
   9  *      updated in fork.c:copy_process(), signal.c:do_signal(),
  10  *      ptrace.c and ptrace.h
  11  *
  12  *       0(%esp) - %ebx
  13  *       4(%esp) - %ecx
  14  *       8(%esp) - %edx
  15  *       C(%esp) - %esi
  16  *      10(%esp) - %edi
  17  *      14(%esp) - %ebp
  18  *      18(%esp) - %eax
  19  *      1C(%esp) - %ds
  20  *      20(%esp) - %es
  21  *      24(%esp) - %fs
  22  *      28(%esp) - %gs          saved iff !CONFIG_X86_32_LAZY_GS
  23  *      2C(%esp) - orig_eax
  24  *      30(%esp) - %eip
  25  *      34(%esp) - %cs
  26  *      38(%esp) - %eflags
  27  *      3C(%esp) - %oldesp
  28  *      40(%esp) - %oldss
  29  */
  30
  31 #include <linux/linkage.h>
  32 #include <linux/err.h>
  33 #include <asm/thread_info.h>
  34 #include <asm/irqflags.h>
  35 #include <asm/errno.h>
  36 #include <asm/segment.h>
  37 #include <asm/smp.h>
  38 #include <asm/page_types.h>
  39 #include <asm/percpu.h>
  40 #include <asm/processor-flags.h>
  41 #include <asm/ftrace.h>
  42 #include <asm/irq_vectors.h>
  43 #include <asm/cpufeatures.h>
  44 #include <asm/alternative-asm.h>
  45 #include <asm/asm.h>
  46 #include <asm/smap.h>
  47
  48         .section .entry.text, "ax"
  49
  50 /*
  51  * We use macros for low-level operations which need to be overridden
  52  * for paravirtualization.  The following will never clobber any registers:
  53  *   INTERRUPT_RETURN (aka. "iret")
  54  *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  55  *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  56  *
  57  * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  58  * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  59  * Allowing a register to be clobbered can shrink the paravirt replacement
  60  * enough to patch inline, increasing performance.
  61  */
  62
  63 #ifdef CONFIG_PREEMPT
  64 # define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  65 #else
  66 # define preempt_stop(clobbers)
  67 # define resume_kernel          restore_all
  68 #endif
  69
  70 .macro TRACE_IRQS_IRET
  71 #ifdef CONFIG_TRACE_IRQFLAGS
  72         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
  73         jz      1f
  74         TRACE_IRQS_ON
  75 1:
  76 #endif
  77 .endm
  78
  79 /*
  80  * User gs save/restore
  81  *
  82  * %gs is used for userland TLS and kernel only uses it for stack
  83  * canary which is required to be at %gs:20 by gcc.  Read the comment
  84  * at the top of stackprotector.h for more info.
  85  *
  86  * Local labels 98 and 99 are used.
  87  */
  88 #ifdef CONFIG_X86_32_LAZY_GS
  89
  90  /* unfortunately push/pop can't be no-op */
  91 .macro PUSH_GS
  92         pushl   $0
  93 .endm
  94 .macro POP_GS pop=0
  95         addl    $(4 + \pop), %esp
  96 .endm
  97 .macro POP_GS_EX
  98 .endm
  99
 100  /* all the rest are no-op */
 101 .macro PTGS_TO_GS
 102 .endm
 103 .macro PTGS_TO_GS_EX
 104 .endm
 105 .macro GS_TO_REG reg
 106 .endm
 107 .macro REG_TO_PTGS reg
 108 .endm
 109 .macro SET_KERNEL_GS reg
 110 .endm
 111
 112 #else   /* CONFIG_X86_32_LAZY_GS */
 113
 114 .macro PUSH_GS
 115         pushl   %gs
 116 .endm
 117
 118 .macro POP_GS pop=0
 119 98:     popl    %gs
 120   .if \pop <> 0
 121         add     $\pop, %esp
 122   .endif
 123 .endm
 124 .macro POP_GS_EX
 125 .pushsection .fixup, "ax"
 126 99:     movl    $0, (%esp)
 127         jmp     98b
 128 .popsection
 129         _ASM_EXTABLE(98b, 99b)
 130 .endm
 131
 132 .macro PTGS_TO_GS
 133 98:     mov     PT_GS(%esp), %gs
 134 .endm
 135 .macro PTGS_TO_GS_EX
 136 .pushsection .fixup, "ax"
 137 99:     movl    $0, PT_GS(%esp)
 138         jmp     98b
 139 .popsection
 140         _ASM_EXTABLE(98b, 99b)
 141 .endm
 142
 143 .macro GS_TO_REG reg
 144         movl    %gs, \reg
 145 .endm
 146 .macro REG_TO_PTGS reg
 147         movl    \reg, PT_GS(%esp)
 148 .endm
 149 .macro SET_KERNEL_GS reg
 150         movl    $(__KERNEL_STACK_CANARY), \reg
 151         movl    \reg, %gs
 152 .endm
 153
 154 #endif /* CONFIG_X86_32_LAZY_GS */
 155
 156 .macro SAVE_ALL pt_regs_ax=%eax
 157         cld
 158         PUSH_GS
 159         pushl   %fs
 160         pushl   %es
 161         pushl   %ds
 162         pushl   \pt_regs_ax
 163         pushl   %ebp
 164         pushl   %edi
 165         pushl   %esi
 166         pushl   %edx
 167         pushl   %ecx
 168         pushl   %ebx
 169         movl    $(__USER_DS), %edx
 170         movl    %edx, %ds
 171         movl    %edx, %es
 172         movl    $(__KERNEL_PERCPU), %edx
 173         movl    %edx, %fs
 174         SET_KERNEL_GS %edx
 175 .endm
 176
 177 .macro RESTORE_INT_REGS
 178         popl    %ebx
 179         popl    %ecx
 180         popl    %edx
 181         popl    %esi
 182         popl    %edi
 183         popl    %ebp
 184         popl    %eax
 185 .endm
 186
 187 .macro RESTORE_REGS pop=0
 188         RESTORE_INT_REGS
 189 1:      popl    %ds
 190 2:      popl    %es
 191 3:      popl    %fs
 192         POP_GS \pop
 193 .pushsection .fixup, "ax"
 194 4:      movl    $0, (%esp)
 195         jmp     1b
 196 5:      movl    $0, (%esp)
 197         jmp     2b
 198 6:      movl    $0, (%esp)
 199         jmp     3b
 200 .popsection
 201         _ASM_EXTABLE(1b, 4b)
 202         _ASM_EXTABLE(2b, 5b)
 203         _ASM_EXTABLE(3b, 6b)
 204         POP_GS_EX
 205 .endm
 206
 207 ENTRY(ret_from_fork)
 208         pushl   %eax
 209         call    schedule_tail
 210         popl    %eax
 211
 212         /* When we fork, we trace the syscall return in the child, too. */
 213         movl    %esp, %eax
 214         call    syscall_return_slowpath
 215         jmp     restore_all
 216 END(ret_from_fork)
 217
 218 ENTRY(ret_from_kernel_thread)
 219         pushl   %eax
 220         call    schedule_tail
 221         popl    %eax
 222         movl    PT_EBP(%esp), %eax
 223         call    *PT_EBX(%esp)
 224         movl    $0, PT_EAX(%esp)
 225
 226         /*
 227          * Kernel threads return to userspace as if returning from a syscall.
 228          * We should check whether anything actually uses this path and, if so,
 229          * consider switching it over to ret_from_fork.
 230          */
 231         movl    %esp, %eax
 232         call    syscall_return_slowpath
 233         jmp     restore_all
 234 ENDPROC(ret_from_kernel_thread)
 235
 236 /*
 237  * Return to user mode is not as complex as all this looks,
 238  * but we want the default path for a system call return to
 239  * go as quickly as possible which is why some of this is
 240  * less clear than it otherwise should be.
 241  */
 242
 243         # userspace resumption stub bypassing syscall exit tracing
 244         ALIGN
 245 ret_from_exception:
 246         preempt_stop(CLBR_ANY)
 247 ret_from_intr:
 248 #ifdef CONFIG_VM86
 249         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS and CS
 250         movb    PT_CS(%esp), %al
 251         andl    $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 252 #else
 253         /*
 254          * We can be coming here from child spawned by kernel_thread().
 255          */
 256         movl    PT_CS(%esp), %eax
 257         andl    $SEGMENT_RPL_MASK, %eax
 258 #endif
 259         cmpl    $USER_RPL, %eax
 260         jb      resume_kernel                   # not returning to v8086 or userspace
 261
 262 ENTRY(resume_userspace)
 263         DISABLE_INTERRUPTS(CLBR_ANY)
 264         TRACE_IRQS_OFF
 265         movl    %esp, %eax
 266         call    prepare_exit_to_usermode
 267         jmp     restore_all
 268 END(ret_from_exception)
 269
 270 #ifdef CONFIG_PREEMPT
 271 ENTRY(resume_kernel)
 272         DISABLE_INTERRUPTS(CLBR_ANY)
 273 need_resched:
 274         cmpl    $0, PER_CPU_VAR(__preempt_count)
 275         jnz     restore_all
 276         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
 277         jz      restore_all
 278         call    preempt_schedule_irq
 279         jmp     need_resched
 280 END(resume_kernel)
 281 #endif
 282
 283 GLOBAL(__begin_SYSENTER_singlestep_region)
 284 /*
 285  * All code from here through __end_SYSENTER_singlestep_region is subject
 286  * to being single-stepped if a user program sets TF and executes SYSENTER.
 287  * There is absolutely nothing that we can do to prevent this from happening
 288  * (thanks Intel!).  To keep our handling of this situation as simple as
 289  * possible, we handle TF just like AC and NT, except that our #DB handler
 290  * will ignore all of the single-step traps generated in this range.
 291  */
 292
 293 #ifdef CONFIG_XEN
 294 /*
 295  * Xen doesn't set %esp to be precisely what the normal SYSENTER
 296  * entry point expects, so fix it up before using the normal path.
 297  */
 298 ENTRY(xen_sysenter_target)
 299         addl    $5*4, %esp                      /* remove xen-provided frame */
 300         jmp     sysenter_past_esp
 301 #endif
 302
 303 /*
 304  * 32-bit SYSENTER entry.
 305  *
 306  * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
 307  * if X86_FEATURE_SEP is available.  This is the preferred system call
 308  * entry on 32-bit systems.
 309  *
 310  * The SYSENTER instruction, in principle, should *only* occur in the
 311  * vDSO.  In practice, a small number of Android devices were shipped
 312  * with a copy of Bionic that inlined a SYSENTER instruction.  This
 313  * never happened in any of Google's Bionic versions -- it only happened
 314  * in a narrow range of Intel-provided versions.
 315  *
 316  * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
 317  * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
 318  * SYSENTER does not save anything on the stack,
 319  * and does not save old EIP (!!!), ESP, or EFLAGS.
 320  *
 321  * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
 322  * user and/or vm86 state), we explicitly disable the SYSENTER
 323  * instruction in vm86 mode by reprogramming the MSRs.
 324  *
 325  * Arguments:
 326  * eax  system call number
 327  * ebx  arg1
 328  * ecx  arg2
 329  * edx  arg3
 330  * esi  arg4
 331  * edi  arg5
 332  * ebp  user stack
 333  * 0(%ebp) arg6
 334  */
 335 ENTRY(entry_SYSENTER_32)
 336         movl    TSS_sysenter_sp0(%esp), %esp
 337 sysenter_past_esp:
 338         pushl   $__USER_DS              /* pt_regs->ss */
 339         pushl   %ebp                    /* pt_regs->sp (stashed in bp) */
 340         pushfl                          /* pt_regs->flags (except IF = 0) */
 341         orl     $X86_EFLAGS_IF, (%esp)  /* Fix IF */
 342         pushl   $__USER_CS              /* pt_regs->cs */
 343         pushl   $0                      /* pt_regs->ip = 0 (placeholder) */
 344         pushl   %eax                    /* pt_regs->orig_ax */
 345         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 346
 347         /*
 348          * SYSENTER doesn't filter flags, so we need to clear NT, AC
 349          * and TF ourselves.  To save a few cycles, we can check whether
 350          * either was set instead of doing an unconditional popfq.
 351          * This needs to happen before enabling interrupts so that
 352          * we don't get preempted with NT set.
 353          *
 354          * If TF is set, we will single-step all the way to here -- do_debug
 355          * will ignore all the traps.  (Yes, this is slow, but so is
 356          * single-stepping in general.  This allows us to avoid having
 357          * a more complicated code to handle the case where a user program
 358          * forces us to single-step through the SYSENTER entry code.)
 359          *
 360          * NB.: .Lsysenter_fix_flags is a label with the code under it moved
 361          * out-of-line as an optimization: NT is unlikely to be set in the
 362          * majority of the cases and instead of polluting the I$ unnecessarily,
 363          * we're keeping that code behind a branch which will predict as
 364          * not-taken and therefore its instructions won't be fetched.
 365          */
 366         testl   $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
 367         jnz     .Lsysenter_fix_flags
 368 .Lsysenter_flags_fixed:
 369
 370         /*
 371          * User mode is traced as though IRQs are on, and SYSENTER
 372          * turned them off.
 373          */
 374         TRACE_IRQS_OFF
 375
 376         movl    %esp, %eax
 377         call    do_fast_syscall_32
 378         /* XEN PV guests always use IRET path */
 379         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 380                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 381
 382 /* Opportunistic SYSEXIT */
 383         TRACE_IRQS_ON                   /* User mode traces as IRQs on. */
 384         movl    PT_EIP(%esp), %edx      /* pt_regs->ip */
 385         movl    PT_OLDESP(%esp), %ecx   /* pt_regs->sp */
 386 1:      mov     PT_FS(%esp), %fs
 387         PTGS_TO_GS
 388         popl    %ebx                    /* pt_regs->bx */
 389         addl    $2*4, %esp              /* skip pt_regs->cx and pt_regs->dx */
 390         popl    %esi                    /* pt_regs->si */
 391         popl    %edi                    /* pt_regs->di */
 392         popl    %ebp                    /* pt_regs->bp */
 393         popl    %eax                    /* pt_regs->ax */
 394
 395         /*
 396          * Restore all flags except IF. (We restore IF separately because
 397          * STI gives a one-instruction window in which we won't be interrupted,
 398          * whereas POPF does not.)
 399          */
 400         addl    $PT_EFLAGS-PT_DS, %esp  /* point esp at pt_regs->flags */
 401         btr     $X86_EFLAGS_IF_BIT, (%esp)
 402         popfl
 403
 404         /*
 405          * Return back to the vDSO, which will pop ecx and edx.
 406          * Don't bother with DS and ES (they already contain __USER_DS).
 407          */
 408         sti
 409         sysexit
 410
 411 .pushsection .fixup, "ax"
 412 2:      movl    $0, PT_FS(%esp)
 413         jmp     1b
 414 .popsection
 415         _ASM_EXTABLE(1b, 2b)
 416         PTGS_TO_GS_EX
 417
 418 .Lsysenter_fix_flags:
 419         pushl   $X86_EFLAGS_FIXED
 420         popfl
 421         jmp     .Lsysenter_flags_fixed
 422 GLOBAL(__end_SYSENTER_singlestep_region)
 423 ENDPROC(entry_SYSENTER_32)
 424
 425 /*
 426  * 32-bit legacy system call entry.
 427  *
 428  * 32-bit x86 Linux system calls traditionally used the INT $0x80
 429  * instruction.  INT $0x80 lands here.
 430  *
 431  * This entry point can be used by any 32-bit perform system calls.
 432  * Instances of INT $0x80 can be found inline in various programs and
 433  * libraries.  It is also used by the vDSO's __kernel_vsyscall
 434  * fallback for hardware that doesn't support a faster entry method.
 435  * Restarted 32-bit system calls also fall back to INT $0x80
 436  * regardless of what instruction was originally used to do the system
 437  * call.  (64-bit programs can use INT $0x80 as well, but they can
 438  * only run on 64-bit kernels and therefore land in
 439  * entry_INT80_compat.)
 440  *
 441  * This is considered a slow path.  It is not used by most libc
 442  * implementations on modern hardware except during process startup.
 443  *
 444  * Arguments:
 445  * eax  system call number
 446  * ebx  arg1
 447  * ecx  arg2
 448  * edx  arg3
 449  * esi  arg4
 450  * edi  arg5
 451  * ebp  arg6
 452  */
 453 ENTRY(entry_INT80_32)
 454         ASM_CLAC
 455         pushl   %eax                    /* pt_regs->orig_ax */
 456         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 457
 458         /*
 459          * User mode is traced as though IRQs are on, and the interrupt gate
 460          * turned them off.
 461          */
 462         TRACE_IRQS_OFF
 463
 464         movl    %esp, %eax
 465         call    do_int80_syscall_32
 466 .Lsyscall_32_done:
 467
 468 restore_all:
 469         TRACE_IRQS_IRET
 470 restore_all_notrace:
 471 #ifdef CONFIG_X86_ESPFIX32
 472         ALTERNATIVE     "jmp restore_nocheck", "", X86_BUG_ESPFIX
 473
 474         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS, SS and CS
 475         /*
 476          * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 477          * are returning to the kernel.
 478          * See comments in process.c:copy_thread() for details.
 479          */
 480         movb    PT_OLDSS(%esp), %ah
 481         movb    PT_CS(%esp), %al
 482         andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 483         cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
 484         je ldt_ss                               # returning to user-space with LDT SS
 485 #endif
 486 restore_nocheck:
 487         RESTORE_REGS 4                          # skip orig_eax/error_code
 488 irq_return:
 489         INTERRUPT_RETURN
 490 .section .fixup, "ax"
 491 ENTRY(iret_exc  )
 492         pushl   $0                              # no error code
 493         pushl   $do_iret_error
 494         jmp     error_code
 495 .previous
 496         _ASM_EXTABLE(irq_return, iret_exc)
 497
 498 #ifdef CONFIG_X86_ESPFIX32
 499 ldt_ss:
 500 /*
 501  * Setup and switch to ESPFIX stack
 502  *
 503  * We're returning to userspace with a 16 bit stack. The CPU will not
 504  * restore the high word of ESP for us on executing iret... This is an
 505  * "official" bug of all the x86-compatible CPUs, which we can work
 506  * around to make dosemu and wine happy. We do this by preloading the
 507  * high word of ESP with the high word of the userspace ESP while
 508  * compensating for the offset by changing to the ESPFIX segment with
 509  * a base address that matches for the difference.
 510  */
 511 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 512         mov     %esp, %edx                      /* load kernel esp */
 513         mov     PT_OLDESP(%esp), %eax           /* load userspace esp */
 514         mov     %dx, %ax                        /* eax: new kernel esp */
 515         sub     %eax, %edx                      /* offset (low word is 0) */
 516         shr     $16, %edx
 517         mov     %dl, GDT_ESPFIX_SS + 4          /* bits 16..23 */
 518         mov     %dh, GDT_ESPFIX_SS + 7          /* bits 24..31 */
 519         pushl   $__ESPFIX_SS
 520         pushl   %eax                            /* new kernel esp */
 521         /*
 522          * Disable interrupts, but do not irqtrace this section: we
 523          * will soon execute iret and the tracer was already set to
 524          * the irqstate after the IRET:
 525          */
 526         DISABLE_INTERRUPTS(CLBR_EAX)
 527         lss     (%esp), %esp                    /* switch to espfix segment */
 528         jmp     restore_nocheck
 529 #endif
 530 ENDPROC(entry_INT80_32)
 531
 532 .macro FIXUP_ESPFIX_STACK
 533 /*
 534  * Switch back for ESPFIX stack to the normal zerobased stack
 535  *
 536  * We can't call C functions using the ESPFIX stack. This code reads
 537  * the high word of the segment base from the GDT and swiches to the
 538  * normal stack and adjusts ESP with the matching offset.
 539  */
 540 #ifdef CONFIG_X86_ESPFIX32
 541         /* fixup the stack */
 542         mov     GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 543         mov     GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 544         shl     $16, %eax
 545         addl    %esp, %eax                      /* the adjusted stack pointer */
 546         pushl   $__KERNEL_DS
 547         pushl   %eax
 548         lss     (%esp), %esp                    /* switch to the normal stack segment */
 549 #endif
 550 .endm
 551 .macro UNWIND_ESPFIX_STACK
 552 #ifdef CONFIG_X86_ESPFIX32
 553         movl    %ss, %eax
 554         /* see if on espfix stack */
 555         cmpw    $__ESPFIX_SS, %ax
 556         jne     27f
 557         movl    $__KERNEL_DS, %eax
 558         movl    %eax, %ds
 559         movl    %eax, %es
 560         /* switch to normal stack */
 561         FIXUP_ESPFIX_STACK
 562 27:
 563 #endif
 564 .endm
 565
 566 /*
 567  * Build the entry stubs with some assembler magic.
 568  * We pack 1 stub into every 8-byte block.
 569  */
 570         .align 8
 571 ENTRY(irq_entries_start)
 572     vector=FIRST_EXTERNAL_VECTOR
 573     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 574         pushl   $(~vector+0x80)                 /* Note: always in signed byte range */
 575     vector=vector+1
 576         jmp     common_interrupt
 577         .align  8
 578     .endr
 579 END(irq_entries_start)
 580
 581 /*
 582  * the CPU automatically disables interrupts when executing an IRQ vector,
 583  * so IRQ-flags tracing has to follow that:
 584  */
 585         .p2align CONFIG_X86_L1_CACHE_SHIFT
 586 common_interrupt:
 587         ASM_CLAC
 588         addl    $-0x80, (%esp)                  /* Adjust vector into the [-256, -1] range */
 589         SAVE_ALL
 590         TRACE_IRQS_OFF
 591         movl    %esp, %eax
 592         call    do_IRQ
 593         jmp     ret_from_intr
 594 ENDPROC(common_interrupt)
 595
 596 #define BUILD_INTERRUPT3(name, nr, fn)  \
 597 ENTRY(name)                             \
 598         ASM_CLAC;                       \
 599         pushl   $~(nr);                 \
 600         SAVE_ALL;                       \
 601         TRACE_IRQS_OFF                  \
 602         movl    %esp, %eax;             \
 603         call    fn;                     \
 604         jmp     ret_from_intr;          \
 605 ENDPROC(name)
 606
 607
 608 #ifdef CONFIG_TRACING
 609 # define TRACE_BUILD_INTERRUPT(name, nr)        BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
 610 #else
 611 # define TRACE_BUILD_INTERRUPT(name, nr)
 612 #endif
 613
 614 #define BUILD_INTERRUPT(name, nr)               \
 615         BUILD_INTERRUPT3(name, nr, smp_##name); \
 616         TRACE_BUILD_INTERRUPT(name, nr)
 617
 618 /* The include is where all of the SMP etc. interrupts come from */
 619 #include <asm/entry_arch.h>
 620
 621 ENTRY(coprocessor_error)
 622         ASM_CLAC
 623         pushl   $0
 624         pushl   $do_coprocessor_error
 625         jmp     error_code
 626 END(coprocessor_error)
 627
 628 ENTRY(simd_coprocessor_error)
 629         ASM_CLAC
 630         pushl   $0
 631 #ifdef CONFIG_X86_INVD_BUG
 632         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 633         ALTERNATIVE "pushl      $do_general_protection",        \
 634                     "pushl      $do_simd_coprocessor_error",    \
 635                     X86_FEATURE_XMM
 636 #else
 637         pushl   $do_simd_coprocessor_error
 638 #endif
 639         jmp     error_code
 640 END(simd_coprocessor_error)
 641
 642 ENTRY(device_not_available)
 643         ASM_CLAC
 644         pushl   $-1                             # mark this as an int
 645         pushl   $do_device_not_available
 646         jmp     error_code
 647 END(device_not_available)
 648
 649 #ifdef CONFIG_PARAVIRT
 650 ENTRY(native_iret)
 651         iret
 652         _ASM_EXTABLE(native_iret, iret_exc)
 653 END(native_iret)
 654 #endif
 655
 656 ENTRY(overflow)
 657         ASM_CLAC
 658         pushl   $0
 659         pushl   $do_overflow
 660         jmp     error_code
 661 END(overflow)
 662
 663 ENTRY(bounds)
 664         ASM_CLAC
 665         pushl   $0
 666         pushl   $do_bounds
 667         jmp     error_code
 668 END(bounds)
 669
 670 ENTRY(invalid_op)
 671         ASM_CLAC
 672         pushl   $0
 673         pushl   $do_invalid_op
 674         jmp     error_code
 675 END(invalid_op)
 676
 677 ENTRY(coprocessor_segment_overrun)
 678         ASM_CLAC
 679         pushl   $0
 680         pushl   $do_coprocessor_segment_overrun
 681         jmp     error_code
 682 END(coprocessor_segment_overrun)
 683
 684 ENTRY(invalid_TSS)
 685         ASM_CLAC
 686         pushl   $do_invalid_TSS
 687         jmp     error_code
 688 END(invalid_TSS)
 689
 690 ENTRY(segment_not_present)
 691         ASM_CLAC
 692         pushl   $do_segment_not_present
 693         jmp     error_code
 694 END(segment_not_present)
 695
 696 ENTRY(stack_segment)
 697         ASM_CLAC
 698         pushl   $do_stack_segment
 699         jmp     error_code
 700 END(stack_segment)
 701
 702 ENTRY(alignment_check)
 703         ASM_CLAC
 704         pushl   $do_alignment_check
 705         jmp     error_code
 706 END(alignment_check)
 707
 708 ENTRY(divide_error)
 709         ASM_CLAC
 710         pushl   $0                              # no error code
 711         pushl   $do_divide_error
 712         jmp     error_code
 713 END(divide_error)
 714
 715 #ifdef CONFIG_X86_MCE
 716 ENTRY(machine_check)
 717         ASM_CLAC
 718         pushl   $0
 719         pushl   machine_check_vector
 720         jmp     error_code
 721 END(machine_check)
 722 #endif
 723
 724 ENTRY(spurious_interrupt_bug)
 725         ASM_CLAC
 726         pushl   $0
 727         pushl   $do_spurious_interrupt_bug
 728         jmp     error_code
 729 END(spurious_interrupt_bug)
 730
 731 #ifdef CONFIG_XEN
 732 ENTRY(xen_hypervisor_callback)
 733         pushl   $-1                             /* orig_ax = -1 => not a system call */
 734         SAVE_ALL
 735         TRACE_IRQS_OFF
 736
 737         /*
 738          * Check to see if we got the event in the critical
 739          * region in xen_iret_direct, after we've reenabled
 740          * events and checked for pending events.  This simulates
 741          * iret instruction's behaviour where it delivers a
 742          * pending interrupt when enabling interrupts:
 743          */
 744         movl    PT_EIP(%esp), %eax
 745         cmpl    $xen_iret_start_crit, %eax
 746         jb      1f
 747         cmpl    $xen_iret_end_crit, %eax
 748         jae     1f
 749
 750         jmp     xen_iret_crit_fixup
 751
 752 ENTRY(xen_do_upcall)
 753 1:      mov     %esp, %eax
 754         call    xen_evtchn_do_upcall
 755 #ifndef CONFIG_PREEMPT
 756         call    xen_maybe_preempt_hcall
 757 #endif
 758         jmp     ret_from_intr
 759 ENDPROC(xen_hypervisor_callback)
 760
 761 /*
 762  * Hypervisor uses this for application faults while it executes.
 763  * We get here for two reasons:
 764  *  1. Fault while reloading DS, ES, FS or GS
 765  *  2. Fault while executing IRET
 766  * Category 1 we fix up by reattempting the load, and zeroing the segment
 767  * register if the load fails.
 768  * Category 2 we fix up by jumping to do_iret_error. We cannot use the
 769  * normal Linux return path in this case because if we use the IRET hypercall
 770  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
 771  * We distinguish between categories by maintaining a status value in EAX.
 772  */
 773 ENTRY(xen_failsafe_callback)
 774         pushl   %eax
 775         movl    $1, %eax
 776 1:      mov     4(%esp), %ds
 777 2:      mov     8(%esp), %es
 778 3:      mov     12(%esp), %fs
 779 4:      mov     16(%esp), %gs
 780         /* EAX == 0 => Category 1 (Bad segment)
 781            EAX != 0 => Category 2 (Bad IRET) */
 782         testl   %eax, %eax
 783         popl    %eax
 784         lea     16(%esp), %esp
 785         jz      5f
 786         jmp     iret_exc
 787 5:      pushl   $-1                             /* orig_ax = -1 => not a system call */
 788         SAVE_ALL
 789         jmp     ret_from_exception
 790
 791 .section .fixup, "ax"
 792 6:      xorl    %eax, %eax
 793         movl    %eax, 4(%esp)
 794         jmp     1b
 795 7:      xorl    %eax, %eax
 796         movl    %eax, 8(%esp)
 797         jmp     2b
 798 8:      xorl    %eax, %eax
 799         movl    %eax, 12(%esp)
 800         jmp     3b
 801 9:      xorl    %eax, %eax
 802         movl    %eax, 16(%esp)
 803         jmp     4b
 804 .previous
 805         _ASM_EXTABLE(1b, 6b)
 806         _ASM_EXTABLE(2b, 7b)
 807         _ASM_EXTABLE(3b, 8b)
 808         _ASM_EXTABLE(4b, 9b)
 809 ENDPROC(xen_failsafe_callback)
 810
 811 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 812                 xen_evtchn_do_upcall)
 813
 814 #endif /* CONFIG_XEN */
 815
 816 #if IS_ENABLED(CONFIG_HYPERV)
 817
 818 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 819         hyperv_vector_handler)
 820
 821 #endif /* CONFIG_HYPERV */
 822
 823 #ifdef CONFIG_FUNCTION_TRACER
 824 #ifdef CONFIG_DYNAMIC_FTRACE
 825
 826 ENTRY(mcount)
 827         ret
 828 END(mcount)
 829
 830 ENTRY(ftrace_caller)
 831         pushl   %eax
 832         pushl   %ecx
 833         pushl   %edx
 834         pushl   $0                              /* Pass NULL as regs pointer */
 835         movl    4*4(%esp), %eax
 836         movl    0x4(%ebp), %edx
 837         movl    function_trace_op, %ecx
 838         subl    $MCOUNT_INSN_SIZE, %eax
 839
 840 .globl ftrace_call
 841 ftrace_call:
 842         call    ftrace_stub
 843
 844         addl    $4, %esp                        /* skip NULL pointer */
 845         popl    %edx
 846         popl    %ecx
 847         popl    %eax
 848 ftrace_ret:
 849 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 850 .globl ftrace_graph_call
 851 ftrace_graph_call:
 852         jmp     ftrace_stub
 853 #endif
 854
 855 .globl ftrace_stub
 856 ftrace_stub:
 857         ret
 858 END(ftrace_caller)
 859
 860 ENTRY(ftrace_regs_caller)
 861         pushf   /* push flags before compare (in cs location) */
 862
 863         /*
 864          * i386 does not save SS and ESP when coming from kernel.
 865          * Instead, to get sp, &regs->sp is used (see ptrace.h).
 866          * Unfortunately, that means eflags must be at the same location
 867          * as the current return ip is. We move the return ip into the
 868          * ip location, and move flags into the return ip location.
 869          */
 870         pushl   4(%esp)                         /* save return ip into ip slot */
 871
 872         pushl   $0                              /* Load 0 into orig_ax */
 873         pushl   %gs
 874         pushl   %fs
 875         pushl   %es
 876         pushl   %ds
 877         pushl   %eax
 878         pushl   %ebp
 879         pushl   %edi
 880         pushl   %esi
 881         pushl   %edx
 882         pushl   %ecx
 883         pushl   %ebx
 884
 885         movl    13*4(%esp), %eax                /* Get the saved flags */
 886         movl    %eax, 14*4(%esp)                /* Move saved flags into regs->flags location */
 887                                                 /* clobbering return ip */
 888         movl    $__KERNEL_CS, 13*4(%esp)
 889
 890         movl    12*4(%esp), %eax                /* Load ip (1st parameter) */
 891         subl    $MCOUNT_INSN_SIZE, %eax         /* Adjust ip */
 892         movl    0x4(%ebp), %edx                 /* Load parent ip (2nd parameter) */
 893         movl    function_trace_op, %ecx         /* Save ftrace_pos in 3rd parameter */
 894         pushl   %esp                            /* Save pt_regs as 4th parameter */
 895
 896 GLOBAL(ftrace_regs_call)
 897         call    ftrace_stub
 898
 899         addl    $4, %esp                        /* Skip pt_regs */
 900         movl    14*4(%esp), %eax                /* Move flags back into cs */
 901         movl    %eax, 13*4(%esp)                /* Needed to keep addl  from modifying flags */
 902         movl    12*4(%esp), %eax                /* Get return ip from regs->ip */
 903         movl    %eax, 14*4(%esp)                /* Put return ip back for ret */
 904
 905         popl    %ebx
 906         popl    %ecx
 907         popl    %edx
 908         popl    %esi
 909         popl    %edi
 910         popl    %ebp
 911         popl    %eax
 912         popl    %ds
 913         popl    %es
 914         popl    %fs
 915         popl    %gs
 916         addl    $8, %esp                        /* Skip orig_ax and ip */
 917         popf                                    /* Pop flags at end (no addl to corrupt flags) */
 918         jmp     ftrace_ret
 919
 920         popf
 921         jmp     ftrace_stub
 922 #else /* ! CONFIG_DYNAMIC_FTRACE */
 923
 924 ENTRY(mcount)
 925         cmpl    $__PAGE_OFFSET, %esp
 926         jb      ftrace_stub                     /* Paging not enabled yet? */
 927
 928         cmpl    $ftrace_stub, ftrace_trace_function
 929         jnz     trace
 930 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 931         cmpl    $ftrace_stub, ftrace_graph_return
 932         jnz     ftrace_graph_caller
 933
 934         cmpl    $ftrace_graph_entry_stub, ftrace_graph_entry
 935         jnz     ftrace_graph_caller
 936 #endif
 937 .globl ftrace_stub
 938 ftrace_stub:
 939         ret
 940
 941         /* taken from glibc */
 942 trace:
 943         pushl   %eax
 944         pushl   %ecx
 945         pushl   %edx
 946         movl    0xc(%esp), %eax
 947         movl    0x4(%ebp), %edx
 948         subl    $MCOUNT_INSN_SIZE, %eax
 949
 950         call    *ftrace_trace_function
 951
 952         popl    %edx
 953         popl    %ecx
 954         popl    %eax
 955         jmp     ftrace_stub
 956 END(mcount)
 957 #endif /* CONFIG_DYNAMIC_FTRACE */
 958 #endif /* CONFIG_FUNCTION_TRACER */
 959
 960 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 961 ENTRY(ftrace_graph_caller)
 962         pushl   %eax
 963         pushl   %ecx
 964         pushl   %edx
 965         movl    0xc(%esp), %eax
 966         lea     0x4(%ebp), %edx
 967         movl    (%ebp), %ecx
 968         subl    $MCOUNT_INSN_SIZE, %eax
 969         call    prepare_ftrace_return
 970         popl    %edx
 971         popl    %ecx
 972         popl    %eax
 973         ret
 974 END(ftrace_graph_caller)
 975
 976 .globl return_to_handler
 977 return_to_handler:
 978         pushl   %eax
 979         pushl   %edx
 980         movl    %ebp, %eax
 981         call    ftrace_return_to_handler
 982         movl    %eax, %ecx
 983         popl    %edx
 984         popl    %eax
 985         jmp     *%ecx
 986 #endif
 987
 988 #ifdef CONFIG_TRACING
 989 ENTRY(trace_page_fault)
 990         ASM_CLAC
 991         pushl   $trace_do_page_fault
 992         jmp     error_code
 993 END(trace_page_fault)
 994 #endif
 995
 996 ENTRY(page_fault)
 997         ASM_CLAC
 998         pushl   $do_page_fault
 999         ALIGN
1000 error_code:
1001         /* the function address is in %gs's slot on the stack */
1002         pushl   %fs
1003         pushl   %es
1004         pushl   %ds
1005         pushl   %eax
1006         pushl   %ebp
1007         pushl   %edi
1008         pushl   %esi
1009         pushl   %edx
1010         pushl   %ecx
1011         pushl   %ebx
1012         cld
1013         movl    $(__KERNEL_PERCPU), %ecx
1014         movl    %ecx, %fs
1015         UNWIND_ESPFIX_STACK
1016         GS_TO_REG %ecx
1017         movl    PT_GS(%esp), %edi               # get the function address
1018         movl    PT_ORIG_EAX(%esp), %edx         # get the error code
1019         movl    $-1, PT_ORIG_EAX(%esp)          # no syscall to restart
1020         REG_TO_PTGS %ecx
1021         SET_KERNEL_GS %ecx
1022         movl    $(__USER_DS), %ecx
1023         movl    %ecx, %ds
1024         movl    %ecx, %es
1025         TRACE_IRQS_OFF
1026         movl    %esp, %eax                      # pt_regs pointer
1027         call    *%edi
1028         jmp     ret_from_exception
1029 END(page_fault)
1030
1031 ENTRY(debug)
1032         /*
1033          * #DB can happen at the first instruction of
1034          * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
1035          * happens, then we will be running on a very small stack.  We
1036          * need to detect this condition and switch to the thread
1037          * stack before calling any C code at all.
1038          *
1039          * If you edit this code, keep in mind that NMIs can happen in here.
1040          */
1041         ASM_CLAC
1042         pushl   $-1                             # mark this as an int
1043         SAVE_ALL
1044         xorl    %edx, %edx                      # error code 0
1045         movl    %esp, %eax                      # pt_regs pointer
1046
1047         /* Are we currently on the SYSENTER stack? */
1048         PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1049         subl    %eax, %ecx      /* ecx = (end of SYSENTER_stack) - esp */
1050         cmpl    $SIZEOF_SYSENTER_stack, %ecx
1051         jb      .Ldebug_from_sysenter_stack
1052
1053         TRACE_IRQS_OFF
1054         call    do_debug
1055         jmp     ret_from_exception
1056
1057 .Ldebug_from_sysenter_stack:
1058         /* We're on the SYSENTER stack.  Switch off. */
1059         movl    %esp, %ebp
1060         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esp
1061         TRACE_IRQS_OFF
1062         call    do_debug
1063         movl    %ebp, %esp
1064         jmp     ret_from_exception
1065 END(debug)
1066
1067 /*
1068  * NMI is doubly nasty.  It can happen on the first instruction of
1069  * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1070  * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1071  * switched stacks.  We handle both conditions by simply checking whether we
1072  * interrupted kernel code running on the SYSENTER stack.
1073  */
1074 ENTRY(nmi)
1075         ASM_CLAC
1076 #ifdef CONFIG_X86_ESPFIX32
1077         pushl   %eax
1078         movl    %ss, %eax
1079         cmpw    $__ESPFIX_SS, %ax
1080         popl    %eax
1081         je      nmi_espfix_stack
1082 #endif
1083
1084         pushl   %eax                            # pt_regs->orig_ax
1085         SAVE_ALL
1086         xorl    %edx, %edx                      # zero error code
1087         movl    %esp, %eax                      # pt_regs pointer
1088
1089         /* Are we currently on the SYSENTER stack? */
1090         PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1091         subl    %eax, %ecx      /* ecx = (end of SYSENTER_stack) - esp */
1092         cmpl    $SIZEOF_SYSENTER_stack, %ecx
1093         jb      .Lnmi_from_sysenter_stack
1094
1095         /* Not on SYSENTER stack. */
1096         call    do_nmi
1097         jmp     restore_all_notrace
1098
1099 .Lnmi_from_sysenter_stack:
1100         /*
1101          * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1102          * is using the thread stack right now, so it's safe for us to use it.
1103          */
1104         movl    %esp, %ebp
1105         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esp
1106         call    do_nmi
1107         movl    %ebp, %esp
1108         jmp     restore_all_notrace
1109
1110 #ifdef CONFIG_X86_ESPFIX32
1111 nmi_espfix_stack:
1112         /*
1113          * create the pointer to lss back
1114          */
1115         pushl   %ss
1116         pushl   %esp
1117         addl    $4, (%esp)
1118         /* copy the iret frame of 12 bytes */
1119         .rept 3
1120         pushl   16(%esp)
1121         .endr
1122         pushl   %eax
1123         SAVE_ALL
1124         FIXUP_ESPFIX_STACK                      # %eax == %esp
1125         xorl    %edx, %edx                      # zero error code
1126         call    do_nmi
1127         RESTORE_REGS
1128         lss     12+4(%esp), %esp                # back to espfix stack
1129         jmp     irq_return
1130 #endif
1131 END(nmi)
1132
1133 ENTRY(int3)
1134         ASM_CLAC
1135         pushl   $-1                             # mark this as an int
1136         SAVE_ALL
1137         TRACE_IRQS_OFF
1138         xorl    %edx, %edx                      # zero error code
1139         movl    %esp, %eax                      # pt_regs pointer
1140         call    do_int3
1141         jmp     ret_from_exception
1142 END(int3)
1143
1144 ENTRY(general_protection)
1145         pushl   $do_general_protection
1146         jmp     error_code
1147 END(general_protection)
1148
1149 #ifdef CONFIG_KVM_GUEST
1150 ENTRY(async_page_fault)
1151         ASM_CLAC
1152         pushl   $do_async_page_fault
1153         jmp     error_code
1154 END(async_page_fault)
1155 #endif
1156
1157 ENTRY(rewind_stack_do_exit)
1158         /* Prevent any naive code from trying to unwind to our caller. */
1159         xorl    %ebp, %ebp
1160
1161         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esi
1162         leal    -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1163
1164         call    do_exit
1165 1:      jmp 1b
1166 END(rewind_stack_do_exit)