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         ASM_CLAC                        /* Clear AC after saving FLAGS */
 298         orl     $X86_EFLAGS_IF, (%esp)  /* Fix IF */
 299         pushl   $__USER_CS              /* pt_regs->cs */
 300         pushl   $0                      /* pt_regs->ip = 0 (placeholder) */
 301         pushl   %eax                    /* pt_regs->orig_ax */
 302         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 303
 304         /*
 305          * User mode is traced as though IRQs are on, and SYSENTER
 306          * turned them off.
 307          */
 308         TRACE_IRQS_OFF
 309
 310         movl    %esp, %eax
 311         call    do_fast_syscall_32
 312         /* XEN PV guests always use IRET path */
 313         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 314                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 315
 316 /* Opportunistic SYSEXIT */
 317         TRACE_IRQS_ON                   /* User mode traces as IRQs on. */
 318         movl    PT_EIP(%esp), %edx      /* pt_regs->ip */
 319         movl    PT_OLDESP(%esp), %ecx   /* pt_regs->sp */
 320 1:      mov     PT_FS(%esp), %fs
 321         PTGS_TO_GS
 322         popl    %ebx                    /* pt_regs->bx */
 323         addl    $2*4, %esp              /* skip pt_regs->cx and pt_regs->dx */
 324         popl    %esi                    /* pt_regs->si */
 325         popl    %edi                    /* pt_regs->di */
 326         popl    %ebp                    /* pt_regs->bp */
 327         popl    %eax                    /* pt_regs->ax */
 328
 329         /*
 330          * Return back to the vDSO, which will pop ecx and edx.
 331          * Don't bother with DS and ES (they already contain __USER_DS).
 332          */
 333         sti
 334         sysexit
 335
 336 .pushsection .fixup, "ax"
 337 2:      movl    $0, PT_FS(%esp)
 338         jmp     1b
 339 .popsection
 340         _ASM_EXTABLE(1b, 2b)
 341         PTGS_TO_GS_EX
 342 ENDPROC(entry_SYSENTER_32)
 343
 344         # system call handler stub
 345 ENTRY(entry_INT80_32)
 346         ASM_CLAC
 347         pushl   %eax                    /* pt_regs->orig_ax */
 348         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 349
 350         /*
 351          * User mode is traced as though IRQs are on.  Unlike the 64-bit
 352          * case, INT80 is a trap gate on 32-bit kernels, so interrupts
 353          * are already on (unless user code is messing around with iopl).
 354          */
 355
 356         movl    %esp, %eax
 357         call    do_syscall_32_irqs_on
 358 .Lsyscall_32_done:
 359
 360 restore_all:
 361         TRACE_IRQS_IRET
 362 restore_all_notrace:
 363 #ifdef CONFIG_X86_ESPFIX32
 364         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS, SS and CS
 365         /*
 366          * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 367          * are returning to the kernel.
 368          * See comments in process.c:copy_thread() for details.
 369          */
 370         movb    PT_OLDSS(%esp), %ah
 371         movb    PT_CS(%esp), %al
 372         andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 373         cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
 374         je ldt_ss                               # returning to user-space with LDT SS
 375 #endif
 376 restore_nocheck:
 377         RESTORE_REGS 4                          # skip orig_eax/error_code
 378 irq_return:
 379         INTERRUPT_RETURN
 380 .section .fixup, "ax"
 381 ENTRY(iret_exc  )
 382         pushl   $0                              # no error code
 383         pushl   $do_iret_error
 384         jmp     error_code
 385 .previous
 386         _ASM_EXTABLE(irq_return, iret_exc)
 387
 388 #ifdef CONFIG_X86_ESPFIX32
 389 ldt_ss:
 390 #ifdef CONFIG_PARAVIRT
 391         /*
 392          * The kernel can't run on a non-flat stack if paravirt mode
 393          * is active.  Rather than try to fixup the high bits of
 394          * ESP, bypass this code entirely.  This may break DOSemu
 395          * and/or Wine support in a paravirt VM, although the option
 396          * is still available to implement the setting of the high
 397          * 16-bits in the INTERRUPT_RETURN paravirt-op.
 398          */
 399         cmpl    $0, pv_info+PARAVIRT_enabled
 400         jne     restore_nocheck
 401 #endif
 402
 403 /*
 404  * Setup and switch to ESPFIX stack
 405  *
 406  * We're returning to userspace with a 16 bit stack. The CPU will not
 407  * restore the high word of ESP for us on executing iret... This is an
 408  * "official" bug of all the x86-compatible CPUs, which we can work
 409  * around to make dosemu and wine happy. We do this by preloading the
 410  * high word of ESP with the high word of the userspace ESP while
 411  * compensating for the offset by changing to the ESPFIX segment with
 412  * a base address that matches for the difference.
 413  */
 414 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 415         mov     %esp, %edx                      /* load kernel esp */
 416         mov     PT_OLDESP(%esp), %eax           /* load userspace esp */
 417         mov     %dx, %ax                        /* eax: new kernel esp */
 418         sub     %eax, %edx                      /* offset (low word is 0) */
 419         shr     $16, %edx
 420         mov     %dl, GDT_ESPFIX_SS + 4          /* bits 16..23 */
 421         mov     %dh, GDT_ESPFIX_SS + 7          /* bits 24..31 */
 422         pushl   $__ESPFIX_SS
 423         pushl   %eax                            /* new kernel esp */
 424         /*
 425          * Disable interrupts, but do not irqtrace this section: we
 426          * will soon execute iret and the tracer was already set to
 427          * the irqstate after the IRET:
 428          */
 429         DISABLE_INTERRUPTS(CLBR_EAX)
 430         lss     (%esp), %esp                    /* switch to espfix segment */
 431         jmp     restore_nocheck
 432 #endif
 433 ENDPROC(entry_INT80_32)
 434
 435 .macro FIXUP_ESPFIX_STACK
 436 /*
 437  * Switch back for ESPFIX stack to the normal zerobased stack
 438  *
 439  * We can't call C functions using the ESPFIX stack. This code reads
 440  * the high word of the segment base from the GDT and swiches to the
 441  * normal stack and adjusts ESP with the matching offset.
 442  */
 443 #ifdef CONFIG_X86_ESPFIX32
 444         /* fixup the stack */
 445         mov     GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 446         mov     GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 447         shl     $16, %eax
 448         addl    %esp, %eax                      /* the adjusted stack pointer */
 449         pushl   $__KERNEL_DS
 450         pushl   %eax
 451         lss     (%esp), %esp                    /* switch to the normal stack segment */
 452 #endif
 453 .endm
 454 .macro UNWIND_ESPFIX_STACK
 455 #ifdef CONFIG_X86_ESPFIX32
 456         movl    %ss, %eax
 457         /* see if on espfix stack */
 458         cmpw    $__ESPFIX_SS, %ax
 459         jne     27f
 460         movl    $__KERNEL_DS, %eax
 461         movl    %eax, %ds
 462         movl    %eax, %es
 463         /* switch to normal stack */
 464         FIXUP_ESPFIX_STACK
 465 27:
 466 #endif
 467 .endm
 468
 469 /*
 470  * Build the entry stubs with some assembler magic.
 471  * We pack 1 stub into every 8-byte block.
 472  */
 473         .align 8
 474 ENTRY(irq_entries_start)
 475     vector=FIRST_EXTERNAL_VECTOR
 476     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 477         pushl   $(~vector+0x80)                 /* Note: always in signed byte range */
 478     vector=vector+1
 479         jmp     common_interrupt
 480         .align  8
 481     .endr
 482 END(irq_entries_start)
 483
 484 /*
 485  * the CPU automatically disables interrupts when executing an IRQ vector,
 486  * so IRQ-flags tracing has to follow that:
 487  */
 488         .p2align CONFIG_X86_L1_CACHE_SHIFT
 489 common_interrupt:
 490         ASM_CLAC
 491         addl    $-0x80, (%esp)                  /* Adjust vector into the [-256, -1] range */
 492         SAVE_ALL
 493         TRACE_IRQS_OFF
 494         movl    %esp, %eax
 495         call    do_IRQ
 496         jmp     ret_from_intr
 497 ENDPROC(common_interrupt)
 498
 499 #define BUILD_INTERRUPT3(name, nr, fn)  \
 500 ENTRY(name)                             \
 501         ASM_CLAC;                       \
 502         pushl   $~(nr);                 \
 503         SAVE_ALL;                       \
 504         TRACE_IRQS_OFF                  \
 505         movl    %esp, %eax;             \
 506         call    fn;                     \
 507         jmp     ret_from_intr;          \
 508 ENDPROC(name)
 509
 510
 511 #ifdef CONFIG_TRACING
 512 # define TRACE_BUILD_INTERRUPT(name, nr)        BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
 513 #else
 514 # define TRACE_BUILD_INTERRUPT(name, nr)
 515 #endif
 516
 517 #define BUILD_INTERRUPT(name, nr)               \
 518         BUILD_INTERRUPT3(name, nr, smp_##name); \
 519         TRACE_BUILD_INTERRUPT(name, nr)
 520
 521 /* The include is where all of the SMP etc. interrupts come from */
 522 #include <asm/entry_arch.h>
 523
 524 ENTRY(coprocessor_error)
 525         ASM_CLAC
 526         pushl   $0
 527         pushl   $do_coprocessor_error
 528         jmp     error_code
 529 END(coprocessor_error)
 530
 531 ENTRY(simd_coprocessor_error)
 532         ASM_CLAC
 533         pushl   $0
 534 #ifdef CONFIG_X86_INVD_BUG
 535         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 536         ALTERNATIVE "pushl      $do_general_protection",        \
 537                     "pushl      $do_simd_coprocessor_error",    \
 538                     X86_FEATURE_XMM
 539 #else
 540         pushl   $do_simd_coprocessor_error
 541 #endif
 542         jmp     error_code
 543 END(simd_coprocessor_error)
 544
 545 ENTRY(device_not_available)
 546         ASM_CLAC
 547         pushl   $-1                             # mark this as an int
 548         pushl   $do_device_not_available
 549         jmp     error_code
 550 END(device_not_available)
 551
 552 #ifdef CONFIG_PARAVIRT
 553 ENTRY(native_iret)
 554         iret
 555         _ASM_EXTABLE(native_iret, iret_exc)
 556 END(native_iret)
 557 #endif
 558
 559 ENTRY(overflow)
 560         ASM_CLAC
 561         pushl   $0
 562         pushl   $do_overflow
 563         jmp     error_code
 564 END(overflow)
 565
 566 ENTRY(bounds)
 567         ASM_CLAC
 568         pushl   $0
 569         pushl   $do_bounds
 570         jmp     error_code
 571 END(bounds)
 572
 573 ENTRY(invalid_op)
 574         ASM_CLAC
 575         pushl   $0
 576         pushl   $do_invalid_op
 577         jmp     error_code
 578 END(invalid_op)
 579
 580 ENTRY(coprocessor_segment_overrun)
 581         ASM_CLAC
 582         pushl   $0
 583         pushl   $do_coprocessor_segment_overrun
 584         jmp     error_code
 585 END(coprocessor_segment_overrun)
 586
 587 ENTRY(invalid_TSS)
 588         ASM_CLAC
 589         pushl   $do_invalid_TSS
 590         jmp     error_code
 591 END(invalid_TSS)
 592
 593 ENTRY(segment_not_present)
 594         ASM_CLAC
 595         pushl   $do_segment_not_present
 596         jmp     error_code
 597 END(segment_not_present)
 598
 599 ENTRY(stack_segment)
 600         ASM_CLAC
 601         pushl   $do_stack_segment
 602         jmp     error_code
 603 END(stack_segment)
 604
 605 ENTRY(alignment_check)
 606         ASM_CLAC
 607         pushl   $do_alignment_check
 608         jmp     error_code
 609 END(alignment_check)
 610
 611 ENTRY(divide_error)
 612         ASM_CLAC
 613         pushl   $0                              # no error code
 614         pushl   $do_divide_error
 615         jmp     error_code
 616 END(divide_error)
 617
 618 #ifdef CONFIG_X86_MCE
 619 ENTRY(machine_check)
 620         ASM_CLAC
 621         pushl   $0
 622         pushl   machine_check_vector
 623         jmp     error_code
 624 END(machine_check)
 625 #endif
 626
 627 ENTRY(spurious_interrupt_bug)
 628         ASM_CLAC
 629         pushl   $0
 630         pushl   $do_spurious_interrupt_bug
 631         jmp     error_code
 632 END(spurious_interrupt_bug)
 633
 634 #ifdef CONFIG_XEN
 635 /*
 636  * Xen doesn't set %esp to be precisely what the normal SYSENTER
 637  * entry point expects, so fix it up before using the normal path.
 638  */
 639 ENTRY(xen_sysenter_target)
 640         addl    $5*4, %esp                      /* remove xen-provided frame */
 641         jmp     sysenter_past_esp
 642
 643 ENTRY(xen_hypervisor_callback)
 644         pushl   $-1                             /* orig_ax = -1 => not a system call */
 645         SAVE_ALL
 646         TRACE_IRQS_OFF
 647
 648         /*
 649          * Check to see if we got the event in the critical
 650          * region in xen_iret_direct, after we've reenabled
 651          * events and checked for pending events.  This simulates
 652          * iret instruction's behaviour where it delivers a
 653          * pending interrupt when enabling interrupts:
 654          */
 655         movl    PT_EIP(%esp), %eax
 656         cmpl    $xen_iret_start_crit, %eax
 657         jb      1f
 658         cmpl    $xen_iret_end_crit, %eax
 659         jae     1f
 660
 661         jmp     xen_iret_crit_fixup
 662
 663 ENTRY(xen_do_upcall)
 664 1:      mov     %esp, %eax
 665         call    xen_evtchn_do_upcall
 666 #ifndef CONFIG_PREEMPT
 667         call    xen_maybe_preempt_hcall
 668 #endif
 669         jmp     ret_from_intr
 670 ENDPROC(xen_hypervisor_callback)
 671
 672 /*
 673  * Hypervisor uses this for application faults while it executes.
 674  * We get here for two reasons:
 675  *  1. Fault while reloading DS, ES, FS or GS
 676  *  2. Fault while executing IRET
 677  * Category 1 we fix up by reattempting the load, and zeroing the segment
 678  * register if the load fails.
 679  * Category 2 we fix up by jumping to do_iret_error. We cannot use the
 680  * normal Linux return path in this case because if we use the IRET hypercall
 681  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
 682  * We distinguish between categories by maintaining a status value in EAX.
 683  */
 684 ENTRY(xen_failsafe_callback)
 685         pushl   %eax
 686         movl    $1, %eax
 687 1:      mov     4(%esp), %ds
 688 2:      mov     8(%esp), %es
 689 3:      mov     12(%esp), %fs
 690 4:      mov     16(%esp), %gs
 691         /* EAX == 0 => Category 1 (Bad segment)
 692            EAX != 0 => Category 2 (Bad IRET) */
 693         testl   %eax, %eax
 694         popl    %eax
 695         lea     16(%esp), %esp
 696         jz      5f
 697         jmp     iret_exc
 698 5:      pushl   $-1                             /* orig_ax = -1 => not a system call */
 699         SAVE_ALL
 700         jmp     ret_from_exception
 701
 702 .section .fixup, "ax"
 703 6:      xorl    %eax, %eax
 704         movl    %eax, 4(%esp)
 705         jmp     1b
 706 7:      xorl    %eax, %eax
 707         movl    %eax, 8(%esp)
 708         jmp     2b
 709 8:      xorl    %eax, %eax
 710         movl    %eax, 12(%esp)
 711         jmp     3b
 712 9:      xorl    %eax, %eax
 713         movl    %eax, 16(%esp)
 714         jmp     4b
 715 .previous
 716         _ASM_EXTABLE(1b, 6b)
 717         _ASM_EXTABLE(2b, 7b)
 718         _ASM_EXTABLE(3b, 8b)
 719         _ASM_EXTABLE(4b, 9b)
 720 ENDPROC(xen_failsafe_callback)
 721
 722 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 723                 xen_evtchn_do_upcall)
 724
 725 #endif /* CONFIG_XEN */
 726
 727 #if IS_ENABLED(CONFIG_HYPERV)
 728
 729 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 730         hyperv_vector_handler)
 731
 732 #endif /* CONFIG_HYPERV */
 733
 734 #ifdef CONFIG_FUNCTION_TRACER
 735 #ifdef CONFIG_DYNAMIC_FTRACE
 736
 737 ENTRY(mcount)
 738         ret
 739 END(mcount)
 740
 741 ENTRY(ftrace_caller)
 742         pushl   %eax
 743         pushl   %ecx
 744         pushl   %edx
 745         pushl   $0                              /* Pass NULL as regs pointer */
 746         movl    4*4(%esp), %eax
 747         movl    0x4(%ebp), %edx
 748         movl    function_trace_op, %ecx
 749         subl    $MCOUNT_INSN_SIZE, %eax
 750
 751 .globl ftrace_call
 752 ftrace_call:
 753         call    ftrace_stub
 754
 755         addl    $4, %esp                        /* skip NULL pointer */
 756         popl    %edx
 757         popl    %ecx
 758         popl    %eax
 759 ftrace_ret:
 760 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 761 .globl ftrace_graph_call
 762 ftrace_graph_call:
 763         jmp     ftrace_stub
 764 #endif
 765
 766 .globl ftrace_stub
 767 ftrace_stub:
 768         ret
 769 END(ftrace_caller)
 770
 771 ENTRY(ftrace_regs_caller)
 772         pushf   /* push flags before compare (in cs location) */
 773
 774         /*
 775          * i386 does not save SS and ESP when coming from kernel.
 776          * Instead, to get sp, &regs->sp is used (see ptrace.h).
 777          * Unfortunately, that means eflags must be at the same location
 778          * as the current return ip is. We move the return ip into the
 779          * ip location, and move flags into the return ip location.
 780          */
 781         pushl   4(%esp)                         /* save return ip into ip slot */
 782
 783         pushl   $0                              /* Load 0 into orig_ax */
 784         pushl   %gs
 785         pushl   %fs
 786         pushl   %es
 787         pushl   %ds
 788         pushl   %eax
 789         pushl   %ebp
 790         pushl   %edi
 791         pushl   %esi
 792         pushl   %edx
 793         pushl   %ecx
 794         pushl   %ebx
 795
 796         movl    13*4(%esp), %eax                /* Get the saved flags */
 797         movl    %eax, 14*4(%esp)                /* Move saved flags into regs->flags location */
 798                                                 /* clobbering return ip */
 799         movl    $__KERNEL_CS, 13*4(%esp)
 800
 801         movl    12*4(%esp), %eax                /* Load ip (1st parameter) */
 802         subl    $MCOUNT_INSN_SIZE, %eax         /* Adjust ip */
 803         movl    0x4(%ebp), %edx                 /* Load parent ip (2nd parameter) */
 804         movl    function_trace_op, %ecx         /* Save ftrace_pos in 3rd parameter */
 805         pushl   %esp                            /* Save pt_regs as 4th parameter */
 806
 807 GLOBAL(ftrace_regs_call)
 808         call    ftrace_stub
 809
 810         addl    $4, %esp                        /* Skip pt_regs */
 811         movl    14*4(%esp), %eax                /* Move flags back into cs */
 812         movl    %eax, 13*4(%esp)                /* Needed to keep addl  from modifying flags */
 813         movl    12*4(%esp), %eax                /* Get return ip from regs->ip */
 814         movl    %eax, 14*4(%esp)                /* Put return ip back for ret */
 815
 816         popl    %ebx
 817         popl    %ecx
 818         popl    %edx
 819         popl    %esi
 820         popl    %edi
 821         popl    %ebp
 822         popl    %eax
 823         popl    %ds
 824         popl    %es
 825         popl    %fs
 826         popl    %gs
 827         addl    $8, %esp                        /* Skip orig_ax and ip */
 828         popf                                    /* Pop flags at end (no addl to corrupt flags) */
 829         jmp     ftrace_ret
 830
 831         popf
 832         jmp     ftrace_stub
 833 #else /* ! CONFIG_DYNAMIC_FTRACE */
 834
 835 ENTRY(mcount)
 836         cmpl    $__PAGE_OFFSET, %esp
 837         jb      ftrace_stub                     /* Paging not enabled yet? */
 838
 839         cmpl    $ftrace_stub, ftrace_trace_function
 840         jnz     trace
 841 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 842         cmpl    $ftrace_stub, ftrace_graph_return
 843         jnz     ftrace_graph_caller
 844
 845         cmpl    $ftrace_graph_entry_stub, ftrace_graph_entry
 846         jnz     ftrace_graph_caller
 847 #endif
 848 .globl ftrace_stub
 849 ftrace_stub:
 850         ret
 851
 852         /* taken from glibc */
 853 trace:
 854         pushl   %eax
 855         pushl   %ecx
 856         pushl   %edx
 857         movl    0xc(%esp), %eax
 858         movl    0x4(%ebp), %edx
 859         subl    $MCOUNT_INSN_SIZE, %eax
 860
 861         call    *ftrace_trace_function
 862
 863         popl    %edx
 864         popl    %ecx
 865         popl    %eax
 866         jmp     ftrace_stub
 867 END(mcount)
 868 #endif /* CONFIG_DYNAMIC_FTRACE */
 869 #endif /* CONFIG_FUNCTION_TRACER */
 870
 871 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 872 ENTRY(ftrace_graph_caller)
 873         pushl   %eax
 874         pushl   %ecx
 875         pushl   %edx
 876         movl    0xc(%esp), %eax
 877         lea     0x4(%ebp), %edx
 878         movl    (%ebp), %ecx
 879         subl    $MCOUNT_INSN_SIZE, %eax
 880         call    prepare_ftrace_return
 881         popl    %edx
 882         popl    %ecx
 883         popl    %eax
 884         ret
 885 END(ftrace_graph_caller)
 886
 887 .globl return_to_handler
 888 return_to_handler:
 889         pushl   %eax
 890         pushl   %edx
 891         movl    %ebp, %eax
 892         call    ftrace_return_to_handler
 893         movl    %eax, %ecx
 894         popl    %edx
 895         popl    %eax
 896         jmp     *%ecx
 897 #endif
 898
 899 #ifdef CONFIG_TRACING
 900 ENTRY(trace_page_fault)
 901         ASM_CLAC
 902         pushl   $trace_do_page_fault
 903         jmp     error_code
 904 END(trace_page_fault)
 905 #endif
 906
 907 ENTRY(page_fault)
 908         ASM_CLAC
 909         pushl   $do_page_fault
 910         ALIGN
 911 error_code:
 912         /* the function address is in %gs's slot on the stack */
 913         pushl   %fs
 914         pushl   %es
 915         pushl   %ds
 916         pushl   %eax
 917         pushl   %ebp
 918         pushl   %edi
 919         pushl   %esi
 920         pushl   %edx
 921         pushl   %ecx
 922         pushl   %ebx
 923         cld
 924         movl    $(__KERNEL_PERCPU), %ecx
 925         movl    %ecx, %fs
 926         UNWIND_ESPFIX_STACK
 927         GS_TO_REG %ecx
 928         movl    PT_GS(%esp), %edi               # get the function address
 929         movl    PT_ORIG_EAX(%esp), %edx         # get the error code
 930         movl    $-1, PT_ORIG_EAX(%esp)          # no syscall to restart
 931         REG_TO_PTGS %ecx
 932         SET_KERNEL_GS %ecx
 933         movl    $(__USER_DS), %ecx
 934         movl    %ecx, %ds
 935         movl    %ecx, %es
 936         TRACE_IRQS_OFF
 937         movl    %esp, %eax                      # pt_regs pointer
 938         call    *%edi
 939         jmp     ret_from_exception
 940 END(page_fault)
 941
 942 /*
 943  * Debug traps and NMI can happen at the one SYSENTER instruction
 944  * that sets up the real kernel stack. Check here, since we can't
 945  * allow the wrong stack to be used.
 946  *
 947  * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
 948  * already pushed 3 words if it hits on the sysenter instruction:
 949  * eflags, cs and eip.
 950  *
 951  * We just load the right stack, and push the three (known) values
 952  * by hand onto the new stack - while updating the return eip past
 953  * the instruction that would have done it for sysenter.
 954  */
 955 .macro FIX_STACK offset ok label
 956         cmpw    $__KERNEL_CS, 4(%esp)
 957         jne     \ok
 958 \label:
 959         movl    TSS_sysenter_sp0 + \offset(%esp), %esp
 960         pushfl
 961         pushl   $__KERNEL_CS
 962         pushl   $sysenter_past_esp
 963 .endm
 964
 965 ENTRY(debug)
 966         ASM_CLAC
 967         cmpl    $entry_SYSENTER_32, (%esp)
 968         jne     debug_stack_correct
 969         FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
 970 debug_stack_correct:
 971         pushl   $-1                             # mark this as an int
 972         SAVE_ALL
 973         TRACE_IRQS_OFF
 974         xorl    %edx, %edx                      # error code 0
 975         movl    %esp, %eax                      # pt_regs pointer
 976         call    do_debug
 977         jmp     ret_from_exception
 978 END(debug)
 979
 980 /*
 981  * NMI is doubly nasty. It can happen _while_ we're handling
 982  * a debug fault, and the debug fault hasn't yet been able to
 983  * clear up the stack. So we first check whether we got  an
 984  * NMI on the sysenter entry path, but after that we need to
 985  * check whether we got an NMI on the debug path where the debug
 986  * fault happened on the sysenter path.
 987  */
 988 ENTRY(nmi)
 989         ASM_CLAC
 990 #ifdef CONFIG_X86_ESPFIX32
 991         pushl   %eax
 992         movl    %ss, %eax
 993         cmpw    $__ESPFIX_SS, %ax
 994         popl    %eax
 995         je      nmi_espfix_stack
 996 #endif
 997         cmpl    $entry_SYSENTER_32, (%esp)
 998         je      nmi_stack_fixup
 999         pushl   %eax
1000         movl    %esp, %eax
1001         /*
1002          * Do not access memory above the end of our stack page,
1003          * it might not exist.
1004          */
1005         andl    $(THREAD_SIZE-1), %eax
1006         cmpl    $(THREAD_SIZE-20), %eax
1007         popl    %eax
1008         jae     nmi_stack_correct
1009         cmpl    $entry_SYSENTER_32, 12(%esp)
1010         je      nmi_debug_stack_check
1011 nmi_stack_correct:
1012         pushl   %eax
1013         SAVE_ALL
1014         xorl    %edx, %edx                      # zero error code
1015         movl    %esp, %eax                      # pt_regs pointer
1016         call    do_nmi
1017         jmp     restore_all_notrace
1018
1019 nmi_stack_fixup:
1020         FIX_STACK 12, nmi_stack_correct, 1
1021         jmp     nmi_stack_correct
1022
1023 nmi_debug_stack_check:
1024         cmpw    $__KERNEL_CS, 16(%esp)
1025         jne     nmi_stack_correct
1026         cmpl    $debug, (%esp)
1027         jb      nmi_stack_correct
1028         cmpl    $debug_esp_fix_insn, (%esp)
1029         ja      nmi_stack_correct
1030         FIX_STACK 24, nmi_stack_correct, 1
1031         jmp     nmi_stack_correct
1032
1033 #ifdef CONFIG_X86_ESPFIX32
1034 nmi_espfix_stack:
1035         /*
1036          * create the pointer to lss back
1037          */
1038         pushl   %ss
1039         pushl   %esp
1040         addl    $4, (%esp)
1041         /* copy the iret frame of 12 bytes */
1042         .rept 3
1043         pushl   16(%esp)
1044         .endr
1045         pushl   %eax
1046         SAVE_ALL
1047         FIXUP_ESPFIX_STACK                      # %eax == %esp
1048         xorl    %edx, %edx                      # zero error code
1049         call    do_nmi
1050         RESTORE_REGS
1051         lss     12+4(%esp), %esp                # back to espfix stack
1052         jmp     irq_return
1053 #endif
1054 END(nmi)
1055
1056 ENTRY(int3)
1057         ASM_CLAC
1058         pushl   $-1                             # mark this as an int
1059         SAVE_ALL
1060         TRACE_IRQS_OFF
1061         xorl    %edx, %edx                      # zero error code
1062         movl    %esp, %eax                      # pt_regs pointer
1063         call    do_int3
1064         jmp     ret_from_exception
1065 END(int3)
1066
1067 ENTRY(general_protection)
1068         pushl   $do_general_protection
1069         jmp     error_code
1070 END(general_protection)
1071
1072 #ifdef CONFIG_KVM_GUEST
1073 ENTRY(async_page_fault)
1074         ASM_CLAC
1075         pushl   $do_async_page_fault
1076         jmp     error_code
1077 END(async_page_fault)
1078 #endif