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/cpufeature.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         GET_THREAD_INFO(%ebp)
 211         popl    %eax
 212         pushl   $0x0202                         # Reset kernel eflags
 213         popfl
 214
 215         /* When we fork, we trace the syscall return in the child, too. */
 216         movl    %esp, %eax
 217         call    syscall_return_slowpath
 218         jmp     restore_all
 219 END(ret_from_fork)
 220
 221 ENTRY(ret_from_kernel_thread)
 222         pushl   %eax
 223         call    schedule_tail
 224         GET_THREAD_INFO(%ebp)
 225         popl    %eax
 226         pushl   $0x0202                         # Reset kernel eflags
 227         popfl
 228         movl    PT_EBP(%esp), %eax
 229         call    *PT_EBX(%esp)
 230         movl    $0, PT_EAX(%esp)
 231
 232         /*
 233          * Kernel threads return to userspace as if returning from a syscall.
 234          * We should check whether anything actually uses this path and, if so,
 235          * consider switching it over to ret_from_fork.
 236          */
 237         movl    %esp, %eax
 238         call    syscall_return_slowpath
 239         jmp     restore_all
 240 ENDPROC(ret_from_kernel_thread)
 241
 242 /*
 243  * Return to user mode is not as complex as all this looks,
 244  * but we want the default path for a system call return to
 245  * go as quickly as possible which is why some of this is
 246  * less clear than it otherwise should be.
 247  */
 248
 249         # userspace resumption stub bypassing syscall exit tracing
 250         ALIGN
 251 ret_from_exception:
 252         preempt_stop(CLBR_ANY)
 253 ret_from_intr:
 254         GET_THREAD_INFO(%ebp)
 255 #ifdef CONFIG_VM86
 256         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS and CS
 257         movb    PT_CS(%esp), %al
 258         andl    $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 259 #else
 260         /*
 261          * We can be coming here from child spawned by kernel_thread().
 262          */
 263         movl    PT_CS(%esp), %eax
 264         andl    $SEGMENT_RPL_MASK, %eax
 265 #endif
 266         cmpl    $USER_RPL, %eax
 267         jb      resume_kernel                   # not returning to v8086 or userspace
 268
 269 ENTRY(resume_userspace)
 270         DISABLE_INTERRUPTS(CLBR_ANY)
 271         TRACE_IRQS_OFF
 272         movl    %esp, %eax
 273         call    prepare_exit_to_usermode
 274         jmp     restore_all
 275 END(ret_from_exception)
 276
 277 #ifdef CONFIG_PREEMPT
 278 ENTRY(resume_kernel)
 279         DISABLE_INTERRUPTS(CLBR_ANY)
 280 need_resched:
 281         cmpl    $0, PER_CPU_VAR(__preempt_count)
 282         jnz     restore_all
 283         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
 284         jz      restore_all
 285         call    preempt_schedule_irq
 286         jmp     need_resched
 287 END(resume_kernel)
 288 #endif
 289
 290         # SYSENTER  call handler stub
 291 ENTRY(entry_SYSENTER_32)
 292         movl    TSS_sysenter_sp0(%esp), %esp
 293 sysenter_past_esp:
 294         pushl   $__USER_DS              /* pt_regs->ss */
 295         pushl   %ebp                    /* pt_regs->sp (stashed in bp) */
 296         pushfl                          /* pt_regs->flags (except IF = 0) */
 297         orl     $X86_EFLAGS_IF, (%esp)  /* Fix IF */
 298         pushl   $__USER_CS              /* pt_regs->cs */
 299         pushl   $0                      /* pt_regs->ip = 0 (placeholder) */
 300         pushl   %eax                    /* pt_regs->orig_ax */
 301         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 302
 303         /*
 304          * User mode is traced as though IRQs are on, and SYSENTER
 305          * turned them off.
 306          */
 307         TRACE_IRQS_OFF
 308
 309         movl    %esp, %eax
 310         call    do_fast_syscall_32
 311         /* XEN PV guests always use IRET path */
 312         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 313                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 314
 315 /* Opportunistic SYSEXIT */
 316         TRACE_IRQS_ON                   /* User mode traces as IRQs on. */
 317         movl    PT_EIP(%esp), %edx      /* pt_regs->ip */
 318         movl    PT_OLDESP(%esp), %ecx   /* pt_regs->sp */
 319 1:      mov     PT_FS(%esp), %fs
 320         PTGS_TO_GS
 321         popl    %ebx                    /* pt_regs->bx */
 322         addl    $2*4, %esp              /* skip pt_regs->cx and pt_regs->dx */
 323         popl    %esi                    /* pt_regs->si */
 324         popl    %edi                    /* pt_regs->di */
 325         popl    %ebp                    /* pt_regs->bp */
 326         popl    %eax                    /* pt_regs->ax */
 327
 328         /*
 329          * Return back to the vDSO, which will pop ecx and edx.
 330          * Don't bother with DS and ES (they already contain __USER_DS).
 331          */
 332         sti
 333         sysexit
 334
 335 .pushsection .fixup, "ax"
 336 2:      movl    $0, PT_FS(%esp)
 337         jmp     1b
 338 .popsection
 339         _ASM_EXTABLE(1b, 2b)
 340         PTGS_TO_GS_EX
 341 ENDPROC(entry_SYSENTER_32)
 342
 343         # system call handler stub
 344 ENTRY(entry_INT80_32)
 345         ASM_CLAC
 346         pushl   %eax                    /* pt_regs->orig_ax */
 347         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 348
 349         /*
 350          * User mode is traced as though IRQs are on.  Unlike the 64-bit
 351          * case, INT80 is a trap gate on 32-bit kernels, so interrupts
 352          * are already on (unless user code is messing around with iopl).
 353          */
 354
 355         movl    %esp, %eax
 356         call    do_syscall_32_irqs_on
 357 .Lsyscall_32_done:
 358
 359 restore_all:
 360         TRACE_IRQS_IRET
 361 restore_all_notrace:
 362 #ifdef CONFIG_X86_ESPFIX32
 363         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS, SS and CS
 364         /*
 365          * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 366          * are returning to the kernel.
 367          * See comments in process.c:copy_thread() for details.
 368          */
 369         movb    PT_OLDSS(%esp), %ah
 370         movb    PT_CS(%esp), %al
 371         andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 372         cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
 373         je ldt_ss                               # returning to user-space with LDT SS
 374 #endif
 375 restore_nocheck:
 376         RESTORE_REGS 4                          # skip orig_eax/error_code
 377 irq_return:
 378         INTERRUPT_RETURN
 379 .section .fixup, "ax"
 380 ENTRY(iret_exc  )
 381         pushl   $0                              # no error code
 382         pushl   $do_iret_error
 383         jmp     error_code
 384 .previous
 385         _ASM_EXTABLE(irq_return, iret_exc)
 386
 387 #ifdef CONFIG_X86_ESPFIX32
 388 ldt_ss:
 389 #ifdef CONFIG_PARAVIRT
 390         /*
 391          * The kernel can't run on a non-flat stack if paravirt mode
 392          * is active.  Rather than try to fixup the high bits of
 393          * ESP, bypass this code entirely.  This may break DOSemu
 394          * and/or Wine support in a paravirt VM, although the option
 395          * is still available to implement the setting of the high
 396          * 16-bits in the INTERRUPT_RETURN paravirt-op.
 397          */
 398         cmpl    $0, pv_info+PARAVIRT_enabled
 399         jne     restore_nocheck
 400 #endif
 401
 402 /*
 403  * Setup and switch to ESPFIX stack
 404  *
 405  * We're returning to userspace with a 16 bit stack. The CPU will not
 406  * restore the high word of ESP for us on executing iret... This is an
 407  * "official" bug of all the x86-compatible CPUs, which we can work
 408  * around to make dosemu and wine happy. We do this by preloading the
 409  * high word of ESP with the high word of the userspace ESP while
 410  * compensating for the offset by changing to the ESPFIX segment with
 411  * a base address that matches for the difference.
 412  */
 413 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 414         mov     %esp, %edx                      /* load kernel esp */
 415         mov     PT_OLDESP(%esp), %eax           /* load userspace esp */
 416         mov     %dx, %ax                        /* eax: new kernel esp */
 417         sub     %eax, %edx                      /* offset (low word is 0) */
 418         shr     $16, %edx
 419         mov     %dl, GDT_ESPFIX_SS + 4          /* bits 16..23 */
 420         mov     %dh, GDT_ESPFIX_SS + 7          /* bits 24..31 */
 421         pushl   $__ESPFIX_SS
 422         pushl   %eax                            /* new kernel esp */
 423         /*
 424          * Disable interrupts, but do not irqtrace this section: we
 425          * will soon execute iret and the tracer was already set to
 426          * the irqstate after the IRET:
 427          */
 428         DISABLE_INTERRUPTS(CLBR_EAX)
 429         lss     (%esp), %esp                    /* switch to espfix segment */
 430         jmp     restore_nocheck
 431 #endif
 432 ENDPROC(entry_INT80_32)
 433
 434 .macro FIXUP_ESPFIX_STACK
 435 /*
 436  * Switch back for ESPFIX stack to the normal zerobased stack
 437  *
 438  * We can't call C functions using the ESPFIX stack. This code reads
 439  * the high word of the segment base from the GDT and swiches to the
 440  * normal stack and adjusts ESP with the matching offset.
 441  */
 442 #ifdef CONFIG_X86_ESPFIX32
 443         /* fixup the stack */
 444         mov     GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 445         mov     GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 446         shl     $16, %eax
 447         addl    %esp, %eax                      /* the adjusted stack pointer */
 448         pushl   $__KERNEL_DS
 449         pushl   %eax
 450         lss     (%esp), %esp                    /* switch to the normal stack segment */
 451 #endif
 452 .endm
 453 .macro UNWIND_ESPFIX_STACK
 454 #ifdef CONFIG_X86_ESPFIX32
 455         movl    %ss, %eax
 456         /* see if on espfix stack */
 457         cmpw    $__ESPFIX_SS, %ax
 458         jne     27f
 459         movl    $__KERNEL_DS, %eax
 460         movl    %eax, %ds
 461         movl    %eax, %es
 462         /* switch to normal stack */
 463         FIXUP_ESPFIX_STACK
 464 27:
 465 #endif
 466 .endm
 467
 468 /*
 469  * Build the entry stubs with some assembler magic.
 470  * We pack 1 stub into every 8-byte block.
 471  */
 472         .align 8
 473 ENTRY(irq_entries_start)
 474     vector=FIRST_EXTERNAL_VECTOR
 475     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 476         pushl   $(~vector+0x80)                 /* Note: always in signed byte range */
 477     vector=vector+1
 478         jmp     common_interrupt
 479         .align  8
 480     .endr
 481 END(irq_entries_start)
 482
 483 /*
 484  * the CPU automatically disables interrupts when executing an IRQ vector,
 485  * so IRQ-flags tracing has to follow that:
 486  */
 487         .p2align CONFIG_X86_L1_CACHE_SHIFT
 488 common_interrupt:
 489         ASM_CLAC
 490         addl    $-0x80, (%esp)                  /* Adjust vector into the [-256, -1] range */
 491         SAVE_ALL
 492         TRACE_IRQS_OFF
 493         movl    %esp, %eax
 494         call    do_IRQ
 495         jmp     ret_from_intr
 496 ENDPROC(common_interrupt)
 497
 498 #define BUILD_INTERRUPT3(name, nr, fn)  \
 499 ENTRY(name)                             \
 500         ASM_CLAC;                       \
 501         pushl   $~(nr);                 \
 502         SAVE_ALL;                       \
 503         TRACE_IRQS_OFF                  \
 504         movl    %esp, %eax;             \
 505         call    fn;                     \
 506         jmp     ret_from_intr;          \
 507 ENDPROC(name)
 508
 509
 510 #ifdef CONFIG_TRACING
 511 # define TRACE_BUILD_INTERRUPT(name, nr)        BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
 512 #else
 513 # define TRACE_BUILD_INTERRUPT(name, nr)
 514 #endif
 515
 516 #define BUILD_INTERRUPT(name, nr)               \
 517         BUILD_INTERRUPT3(name, nr, smp_##name); \
 518         TRACE_BUILD_INTERRUPT(name, nr)
 519
 520 /* The include is where all of the SMP etc. interrupts come from */
 521 #include <asm/entry_arch.h>
 522
 523 ENTRY(coprocessor_error)
 524         ASM_CLAC
 525         pushl   $0
 526         pushl   $do_coprocessor_error
 527         jmp     error_code
 528 END(coprocessor_error)
 529
 530 ENTRY(simd_coprocessor_error)
 531         ASM_CLAC
 532         pushl   $0
 533 #ifdef CONFIG_X86_INVD_BUG
 534         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 535         ALTERNATIVE "pushl      $do_general_protection",        \
 536                     "pushl      $do_simd_coprocessor_error",    \
 537                     X86_FEATURE_XMM
 538 #else
 539         pushl   $do_simd_coprocessor_error
 540 #endif
 541         jmp     error_code
 542 END(simd_coprocessor_error)
 543
 544 ENTRY(device_not_available)
 545         ASM_CLAC
 546         pushl   $-1                             # mark this as an int
 547         pushl   $do_device_not_available
 548         jmp     error_code
 549 END(device_not_available)
 550
 551 #ifdef CONFIG_PARAVIRT
 552 ENTRY(native_iret)
 553         iret
 554         _ASM_EXTABLE(native_iret, iret_exc)
 555 END(native_iret)
 556 #endif
 557
 558 ENTRY(overflow)
 559         ASM_CLAC
 560         pushl   $0
 561         pushl   $do_overflow
 562         jmp     error_code
 563 END(overflow)
 564
 565 ENTRY(bounds)
 566         ASM_CLAC
 567         pushl   $0
 568         pushl   $do_bounds
 569         jmp     error_code
 570 END(bounds)
 571
 572 ENTRY(invalid_op)
 573         ASM_CLAC
 574         pushl   $0
 575         pushl   $do_invalid_op
 576         jmp     error_code
 577 END(invalid_op)
 578
 579 ENTRY(coprocessor_segment_overrun)
 580         ASM_CLAC
 581         pushl   $0
 582         pushl   $do_coprocessor_segment_overrun
 583         jmp     error_code
 584 END(coprocessor_segment_overrun)
 585
 586 ENTRY(invalid_TSS)
 587         ASM_CLAC
 588         pushl   $do_invalid_TSS
 589         jmp     error_code
 590 END(invalid_TSS)
 591
 592 ENTRY(segment_not_present)
 593         ASM_CLAC
 594         pushl   $do_segment_not_present
 595         jmp     error_code
 596 END(segment_not_present)
 597
 598 ENTRY(stack_segment)
 599         ASM_CLAC
 600         pushl   $do_stack_segment
 601         jmp     error_code
 602 END(stack_segment)
 603
 604 ENTRY(alignment_check)
 605         ASM_CLAC
 606         pushl   $do_alignment_check
 607         jmp     error_code
 608 END(alignment_check)
 609
 610 ENTRY(divide_error)
 611         ASM_CLAC
 612         pushl   $0                              # no error code
 613         pushl   $do_divide_error
 614         jmp     error_code
 615 END(divide_error)
 616
 617 #ifdef CONFIG_X86_MCE
 618 ENTRY(machine_check)
 619         ASM_CLAC
 620         pushl   $0
 621         pushl   machine_check_vector
 622         jmp     error_code
 623 END(machine_check)
 624 #endif
 625
 626 ENTRY(spurious_interrupt_bug)
 627         ASM_CLAC
 628         pushl   $0
 629         pushl   $do_spurious_interrupt_bug
 630         jmp     error_code
 631 END(spurious_interrupt_bug)
 632
 633 #ifdef CONFIG_XEN
 634 /*
 635  * Xen doesn't set %esp to be precisely what the normal SYSENTER
 636  * entry point expects, so fix it up before using the normal path.
 637  */
 638 ENTRY(xen_sysenter_target)
 639         addl    $5*4, %esp                      /* remove xen-provided frame */
 640         jmp     sysenter_past_esp
 641
 642 ENTRY(xen_hypervisor_callback)
 643         pushl   $-1                             /* orig_ax = -1 => not a system call */
 644         SAVE_ALL
 645         TRACE_IRQS_OFF
 646
 647         /*
 648          * Check to see if we got the event in the critical
 649          * region in xen_iret_direct, after we've reenabled
 650          * events and checked for pending events.  This simulates
 651          * iret instruction's behaviour where it delivers a
 652          * pending interrupt when enabling interrupts:
 653          */
 654         movl    PT_EIP(%esp), %eax
 655         cmpl    $xen_iret_start_crit, %eax
 656         jb      1f
 657         cmpl    $xen_iret_end_crit, %eax
 658         jae     1f
 659
 660         jmp     xen_iret_crit_fixup
 661
 662 ENTRY(xen_do_upcall)
 663 1:      mov     %esp, %eax
 664         call    xen_evtchn_do_upcall
 665 #ifndef CONFIG_PREEMPT
 666         call    xen_maybe_preempt_hcall
 667 #endif
 668         jmp     ret_from_intr
 669 ENDPROC(xen_hypervisor_callback)
 670
 671 /*
 672  * Hypervisor uses this for application faults while it executes.
 673  * We get here for two reasons:
 674  *  1. Fault while reloading DS, ES, FS or GS
 675  *  2. Fault while executing IRET
 676  * Category 1 we fix up by reattempting the load, and zeroing the segment
 677  * register if the load fails.
 678  * Category 2 we fix up by jumping to do_iret_error. We cannot use the
 679  * normal Linux return path in this case because if we use the IRET hypercall
 680  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
 681  * We distinguish between categories by maintaining a status value in EAX.
 682  */
 683 ENTRY(xen_failsafe_callback)
 684         pushl   %eax
 685         movl    $1, %eax
 686 1:      mov     4(%esp), %ds
 687 2:      mov     8(%esp), %es
 688 3:      mov     12(%esp), %fs
 689 4:      mov     16(%esp), %gs
 690         /* EAX == 0 => Category 1 (Bad segment)
 691            EAX != 0 => Category 2 (Bad IRET) */
 692         testl   %eax, %eax
 693         popl    %eax
 694         lea     16(%esp), %esp
 695         jz      5f
 696         jmp     iret_exc
 697 5:      pushl   $-1                             /* orig_ax = -1 => not a system call */
 698         SAVE_ALL
 699         jmp     ret_from_exception
 700
 701 .section .fixup, "ax"
 702 6:      xorl    %eax, %eax
 703         movl    %eax, 4(%esp)
 704         jmp     1b
 705 7:      xorl    %eax, %eax
 706         movl    %eax, 8(%esp)
 707         jmp     2b
 708 8:      xorl    %eax, %eax
 709         movl    %eax, 12(%esp)
 710         jmp     3b
 711 9:      xorl    %eax, %eax
 712         movl    %eax, 16(%esp)
 713         jmp     4b
 714 .previous
 715         _ASM_EXTABLE(1b, 6b)
 716         _ASM_EXTABLE(2b, 7b)
 717         _ASM_EXTABLE(3b, 8b)
 718         _ASM_EXTABLE(4b, 9b)
 719 ENDPROC(xen_failsafe_callback)
 720
 721 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 722                 xen_evtchn_do_upcall)
 723
 724 #endif /* CONFIG_XEN */
 725
 726 #if IS_ENABLED(CONFIG_HYPERV)
 727
 728 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 729         hyperv_vector_handler)
 730
 731 #endif /* CONFIG_HYPERV */
 732
 733 #ifdef CONFIG_FUNCTION_TRACER
 734 #ifdef CONFIG_DYNAMIC_FTRACE
 735
 736 ENTRY(mcount)
 737         ret
 738 END(mcount)
 739
 740 ENTRY(ftrace_caller)
 741         pushl   %eax
 742         pushl   %ecx
 743         pushl   %edx
 744         pushl   $0                              /* Pass NULL as regs pointer */
 745         movl    4*4(%esp), %eax
 746         movl    0x4(%ebp), %edx
 747         movl    function_trace_op, %ecx
 748         subl    $MCOUNT_INSN_SIZE, %eax
 749
 750 .globl ftrace_call
 751 ftrace_call:
 752         call    ftrace_stub
 753
 754         addl    $4, %esp                        /* skip NULL pointer */
 755         popl    %edx
 756         popl    %ecx
 757         popl    %eax
 758 ftrace_ret:
 759 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 760 .globl ftrace_graph_call
 761 ftrace_graph_call:
 762         jmp     ftrace_stub
 763 #endif
 764
 765 .globl ftrace_stub
 766 ftrace_stub:
 767         ret
 768 END(ftrace_caller)
 769
 770 ENTRY(ftrace_regs_caller)
 771         pushf   /* push flags before compare (in cs location) */
 772
 773         /*
 774          * i386 does not save SS and ESP when coming from kernel.
 775          * Instead, to get sp, &regs->sp is used (see ptrace.h).
 776          * Unfortunately, that means eflags must be at the same location
 777          * as the current return ip is. We move the return ip into the
 778          * ip location, and move flags into the return ip location.
 779          */
 780         pushl   4(%esp)                         /* save return ip into ip slot */
 781
 782         pushl   $0                              /* Load 0 into orig_ax */
 783         pushl   %gs
 784         pushl   %fs
 785         pushl   %es
 786         pushl   %ds
 787         pushl   %eax
 788         pushl   %ebp
 789         pushl   %edi
 790         pushl   %esi
 791         pushl   %edx
 792         pushl   %ecx
 793         pushl   %ebx
 794
 795         movl    13*4(%esp), %eax                /* Get the saved flags */
 796         movl    %eax, 14*4(%esp)                /* Move saved flags into regs->flags location */
 797                                                 /* clobbering return ip */
 798         movl    $__KERNEL_CS, 13*4(%esp)
 799
 800         movl    12*4(%esp), %eax                /* Load ip (1st parameter) */
 801         subl    $MCOUNT_INSN_SIZE, %eax         /* Adjust ip */
 802         movl    0x4(%ebp), %edx                 /* Load parent ip (2nd parameter) */
 803         movl    function_trace_op, %ecx         /* Save ftrace_pos in 3rd parameter */
 804         pushl   %esp                            /* Save pt_regs as 4th parameter */
 805
 806 GLOBAL(ftrace_regs_call)
 807         call    ftrace_stub
 808
 809         addl    $4, %esp                        /* Skip pt_regs */
 810         movl    14*4(%esp), %eax                /* Move flags back into cs */
 811         movl    %eax, 13*4(%esp)                /* Needed to keep addl  from modifying flags */
 812         movl    12*4(%esp), %eax                /* Get return ip from regs->ip */
 813         movl    %eax, 14*4(%esp)                /* Put return ip back for ret */
 814
 815         popl    %ebx
 816         popl    %ecx
 817         popl    %edx
 818         popl    %esi
 819         popl    %edi
 820         popl    %ebp
 821         popl    %eax
 822         popl    %ds
 823         popl    %es
 824         popl    %fs
 825         popl    %gs
 826         addl    $8, %esp                        /* Skip orig_ax and ip */
 827         popf                                    /* Pop flags at end (no addl to corrupt flags) */
 828         jmp     ftrace_ret
 829
 830         popf
 831         jmp     ftrace_stub
 832 #else /* ! CONFIG_DYNAMIC_FTRACE */
 833
 834 ENTRY(mcount)
 835         cmpl    $__PAGE_OFFSET, %esp
 836         jb      ftrace_stub                     /* Paging not enabled yet? */
 837
 838         cmpl    $ftrace_stub, ftrace_trace_function
 839         jnz     trace
 840 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 841         cmpl    $ftrace_stub, ftrace_graph_return
 842         jnz     ftrace_graph_caller
 843
 844         cmpl    $ftrace_graph_entry_stub, ftrace_graph_entry
 845         jnz     ftrace_graph_caller
 846 #endif
 847 .globl ftrace_stub
 848 ftrace_stub:
 849         ret
 850
 851         /* taken from glibc */
 852 trace:
 853         pushl   %eax
 854         pushl   %ecx
 855         pushl   %edx
 856         movl    0xc(%esp), %eax
 857         movl    0x4(%ebp), %edx
 858         subl    $MCOUNT_INSN_SIZE, %eax
 859
 860         call    *ftrace_trace_function
 861
 862         popl    %edx
 863         popl    %ecx
 864         popl    %eax
 865         jmp     ftrace_stub
 866 END(mcount)
 867 #endif /* CONFIG_DYNAMIC_FTRACE */
 868 #endif /* CONFIG_FUNCTION_TRACER */
 869
 870 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 871 ENTRY(ftrace_graph_caller)
 872         pushl   %eax
 873         pushl   %ecx
 874         pushl   %edx
 875         movl    0xc(%esp), %eax
 876         lea     0x4(%ebp), %edx
 877         movl    (%ebp), %ecx
 878         subl    $MCOUNT_INSN_SIZE, %eax
 879         call    prepare_ftrace_return
 880         popl    %edx
 881         popl    %ecx
 882         popl    %eax
 883         ret
 884 END(ftrace_graph_caller)
 885
 886 .globl return_to_handler
 887 return_to_handler:
 888         pushl   %eax
 889         pushl   %edx
 890         movl    %ebp, %eax
 891         call    ftrace_return_to_handler
 892         movl    %eax, %ecx
 893         popl    %edx
 894         popl    %eax
 895         jmp     *%ecx
 896 #endif
 897
 898 #ifdef CONFIG_TRACING
 899 ENTRY(trace_page_fault)
 900         ASM_CLAC
 901         pushl   $trace_do_page_fault
 902         jmp     error_code
 903 END(trace_page_fault)
 904 #endif
 905
 906 ENTRY(page_fault)
 907         ASM_CLAC
 908         pushl   $do_page_fault
 909         ALIGN
 910 error_code:
 911         /* the function address is in %gs's slot on the stack */
 912         pushl   %fs
 913         pushl   %es
 914         pushl   %ds
 915         pushl   %eax
 916         pushl   %ebp
 917         pushl   %edi
 918         pushl   %esi
 919         pushl   %edx
 920         pushl   %ecx
 921         pushl   %ebx
 922         cld
 923         movl    $(__KERNEL_PERCPU), %ecx
 924         movl    %ecx, %fs
 925         UNWIND_ESPFIX_STACK
 926         GS_TO_REG %ecx
 927         movl    PT_GS(%esp), %edi               # get the function address
 928         movl    PT_ORIG_EAX(%esp), %edx         # get the error code
 929         movl    $-1, PT_ORIG_EAX(%esp)          # no syscall to restart
 930         REG_TO_PTGS %ecx
 931         SET_KERNEL_GS %ecx
 932         movl    $(__USER_DS), %ecx
 933         movl    %ecx, %ds
 934         movl    %ecx, %es
 935         TRACE_IRQS_OFF
 936         movl    %esp, %eax                      # pt_regs pointer
 937         call    *%edi
 938         jmp     ret_from_exception
 939 END(page_fault)
 940
 941 /*
 942  * Debug traps and NMI can happen at the one SYSENTER instruction
 943  * that sets up the real kernel stack. Check here, since we can't
 944  * allow the wrong stack to be used.
 945  *
 946  * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
 947  * already pushed 3 words if it hits on the sysenter instruction:
 948  * eflags, cs and eip.
 949  *
 950  * We just load the right stack, and push the three (known) values
 951  * by hand onto the new stack - while updating the return eip past
 952  * the instruction that would have done it for sysenter.
 953  */
 954 .macro FIX_STACK offset ok label
 955         cmpw    $__KERNEL_CS, 4(%esp)
 956         jne     \ok
 957 \label:
 958         movl    TSS_sysenter_sp0 + \offset(%esp), %esp
 959         pushfl
 960         pushl   $__KERNEL_CS
 961         pushl   $sysenter_past_esp
 962 .endm
 963
 964 ENTRY(debug)
 965         ASM_CLAC
 966         cmpl    $entry_SYSENTER_32, (%esp)
 967         jne     debug_stack_correct
 968         FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
 969 debug_stack_correct:
 970         pushl   $-1                             # mark this as an int
 971         SAVE_ALL
 972         TRACE_IRQS_OFF
 973         xorl    %edx, %edx                      # error code 0
 974         movl    %esp, %eax                      # pt_regs pointer
 975         call    do_debug
 976         jmp     ret_from_exception
 977 END(debug)
 978
 979 /*
 980  * NMI is doubly nasty. It can happen _while_ we're handling
 981  * a debug fault, and the debug fault hasn't yet been able to
 982  * clear up the stack. So we first check whether we got  an
 983  * NMI on the sysenter entry path, but after that we need to
 984  * check whether we got an NMI on the debug path where the debug
 985  * fault happened on the sysenter path.
 986  */
 987 ENTRY(nmi)
 988         ASM_CLAC
 989 #ifdef CONFIG_X86_ESPFIX32
 990         pushl   %eax
 991         movl    %ss, %eax
 992         cmpw    $__ESPFIX_SS, %ax
 993         popl    %eax
 994         je      nmi_espfix_stack
 995 #endif
 996         cmpl    $entry_SYSENTER_32, (%esp)
 997         je      nmi_stack_fixup
 998         pushl   %eax
 999         movl    %esp, %eax
1000         /*
1001          * Do not access memory above the end of our stack page,
1002          * it might not exist.
1003          */
1004         andl    $(THREAD_SIZE-1), %eax
1005         cmpl    $(THREAD_SIZE-20), %eax
1006         popl    %eax
1007         jae     nmi_stack_correct
1008         cmpl    $entry_SYSENTER_32, 12(%esp)
1009         je      nmi_debug_stack_check
1010 nmi_stack_correct:
1011         pushl   %eax
1012         SAVE_ALL
1013         xorl    %edx, %edx                      # zero error code
1014         movl    %esp, %eax                      # pt_regs pointer
1015         call    do_nmi
1016         jmp     restore_all_notrace
1017
1018 nmi_stack_fixup:
1019         FIX_STACK 12, nmi_stack_correct, 1
1020         jmp     nmi_stack_correct
1021
1022 nmi_debug_stack_check:
1023         cmpw    $__KERNEL_CS, 16(%esp)
1024         jne     nmi_stack_correct
1025         cmpl    $debug, (%esp)
1026         jb      nmi_stack_correct
1027         cmpl    $debug_esp_fix_insn, (%esp)
1028         ja      nmi_stack_correct
1029         FIX_STACK 24, nmi_stack_correct, 1
1030         jmp     nmi_stack_correct
1031
1032 #ifdef CONFIG_X86_ESPFIX32
1033 nmi_espfix_stack:
1034         /*
1035          * create the pointer to lss back
1036          */
1037         pushl   %ss
1038         pushl   %esp
1039         addl    $4, (%esp)
1040         /* copy the iret frame of 12 bytes */
1041         .rept 3
1042         pushl   16(%esp)
1043         .endr
1044         pushl   %eax
1045         SAVE_ALL
1046         FIXUP_ESPFIX_STACK                      # %eax == %esp
1047         xorl    %edx, %edx                      # zero error code
1048         call    do_nmi
1049         RESTORE_REGS
1050         lss     12+4(%esp), %esp                # back to espfix stack
1051         jmp     irq_return
1052 #endif
1053 END(nmi)
1054
1055 ENTRY(int3)
1056         ASM_CLAC
1057         pushl   $-1                             # mark this as an int
1058         SAVE_ALL
1059         TRACE_IRQS_OFF
1060         xorl    %edx, %edx                      # zero error code
1061         movl    %esp, %eax                      # pt_regs pointer
1062         call    do_int3
1063         jmp     ret_from_exception
1064 END(int3)
1065
1066 ENTRY(general_protection)
1067         pushl   $do_general_protection
1068         jmp     error_code
1069 END(general_protection)
1070
1071 #ifdef CONFIG_KVM_GUEST
1072 ENTRY(async_page_fault)
1073         ASM_CLAC
1074         pushl   $do_async_page_fault
1075         jmp     error_code
1076 END(async_page_fault)
1077 #endif