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 in 'syscall_exit':
   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 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
  49 #include <linux/elf-em.h>
  50 #define AUDIT_ARCH_I386         (EM_386|__AUDIT_ARCH_LE)
  51 #define __AUDIT_ARCH_LE         0x40000000
  52
  53 #ifndef CONFIG_AUDITSYSCALL
  54 # define sysenter_audit         syscall_trace_entry
  55 # define sysexit_audit          syscall_exit_work
  56 #endif
  57
  58         .section .entry.text, "ax"
  59
  60 /*
  61  * We use macros for low-level operations which need to be overridden
  62  * for paravirtualization.  The following will never clobber any registers:
  63  *   INTERRUPT_RETURN (aka. "iret")
  64  *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  65  *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  66  *
  67  * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  68  * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  69  * Allowing a register to be clobbered can shrink the paravirt replacement
  70  * enough to patch inline, increasing performance.
  71  */
  72
  73 #ifdef CONFIG_PREEMPT
  74 # define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  75 #else
  76 # define preempt_stop(clobbers)
  77 # define resume_kernel          restore_all
  78 #endif
  79
  80 .macro TRACE_IRQS_IRET
  81 #ifdef CONFIG_TRACE_IRQFLAGS
  82         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
  83         jz      1f
  84         TRACE_IRQS_ON
  85 1:
  86 #endif
  87 .endm
  88
  89 /*
  90  * User gs save/restore
  91  *
  92  * %gs is used for userland TLS and kernel only uses it for stack
  93  * canary which is required to be at %gs:20 by gcc.  Read the comment
  94  * at the top of stackprotector.h for more info.
  95  *
  96  * Local labels 98 and 99 are used.
  97  */
  98 #ifdef CONFIG_X86_32_LAZY_GS
  99
 100  /* unfortunately push/pop can't be no-op */
 101 .macro PUSH_GS
 102         pushl   $0
 103 .endm
 104 .macro POP_GS pop=0
 105         addl    $(4 + \pop), %esp
 106 .endm
 107 .macro POP_GS_EX
 108 .endm
 109
 110  /* all the rest are no-op */
 111 .macro PTGS_TO_GS
 112 .endm
 113 .macro PTGS_TO_GS_EX
 114 .endm
 115 .macro GS_TO_REG reg
 116 .endm
 117 .macro REG_TO_PTGS reg
 118 .endm
 119 .macro SET_KERNEL_GS reg
 120 .endm
 121
 122 #else   /* CONFIG_X86_32_LAZY_GS */
 123
 124 .macro PUSH_GS
 125         pushl   %gs
 126 .endm
 127
 128 .macro POP_GS pop=0
 129 98:     popl    %gs
 130   .if \pop <> 0
 131         add     $\pop, %esp
 132   .endif
 133 .endm
 134 .macro POP_GS_EX
 135 .pushsection .fixup, "ax"
 136 99:     movl    $0, (%esp)
 137         jmp     98b
 138 .popsection
 139         _ASM_EXTABLE(98b, 99b)
 140 .endm
 141
 142 .macro PTGS_TO_GS
 143 98:     mov     PT_GS(%esp), %gs
 144 .endm
 145 .macro PTGS_TO_GS_EX
 146 .pushsection .fixup, "ax"
 147 99:     movl    $0, PT_GS(%esp)
 148         jmp     98b
 149 .popsection
 150         _ASM_EXTABLE(98b, 99b)
 151 .endm
 152
 153 .macro GS_TO_REG reg
 154         movl    %gs, \reg
 155 .endm
 156 .macro REG_TO_PTGS reg
 157         movl    \reg, PT_GS(%esp)
 158 .endm
 159 .macro SET_KERNEL_GS reg
 160         movl    $(__KERNEL_STACK_CANARY), \reg
 161         movl    \reg, %gs
 162 .endm
 163
 164 #endif /* CONFIG_X86_32_LAZY_GS */
 165
 166 .macro SAVE_ALL
 167         cld
 168         PUSH_GS
 169         pushl   %fs
 170         pushl   %es
 171         pushl   %ds
 172         pushl   %eax
 173         pushl   %ebp
 174         pushl   %edi
 175         pushl   %esi
 176         pushl   %edx
 177         pushl   %ecx
 178         pushl   %ebx
 179         movl    $(__USER_DS), %edx
 180         movl    %edx, %ds
 181         movl    %edx, %es
 182         movl    $(__KERNEL_PERCPU), %edx
 183         movl    %edx, %fs
 184         SET_KERNEL_GS %edx
 185 .endm
 186
 187 .macro RESTORE_INT_REGS
 188         popl    %ebx
 189         popl    %ecx
 190         popl    %edx
 191         popl    %esi
 192         popl    %edi
 193         popl    %ebp
 194         popl    %eax
 195 .endm
 196
 197 .macro RESTORE_REGS pop=0
 198         RESTORE_INT_REGS
 199 1:      popl    %ds
 200 2:      popl    %es
 201 3:      popl    %fs
 202         POP_GS \pop
 203 .pushsection .fixup, "ax"
 204 4:      movl    $0, (%esp)
 205         jmp     1b
 206 5:      movl    $0, (%esp)
 207         jmp     2b
 208 6:      movl    $0, (%esp)
 209         jmp     3b
 210 .popsection
 211         _ASM_EXTABLE(1b, 4b)
 212         _ASM_EXTABLE(2b, 5b)
 213         _ASM_EXTABLE(3b, 6b)
 214         POP_GS_EX
 215 .endm
 216
 217 ENTRY(ret_from_fork)
 218         pushl   %eax
 219         call    schedule_tail
 220         GET_THREAD_INFO(%ebp)
 221         popl    %eax
 222         pushl   $0x0202                         # Reset kernel eflags
 223         popfl
 224         jmp     syscall_exit
 225 END(ret_from_fork)
 226
 227 ENTRY(ret_from_kernel_thread)
 228         pushl   %eax
 229         call    schedule_tail
 230         GET_THREAD_INFO(%ebp)
 231         popl    %eax
 232         pushl   $0x0202                         # Reset kernel eflags
 233         popfl
 234         movl    PT_EBP(%esp), %eax
 235         call    *PT_EBX(%esp)
 236         movl    $0, PT_EAX(%esp)
 237         jmp     syscall_exit
 238 ENDPROC(ret_from_kernel_thread)
 239
 240 /*
 241  * Return to user mode is not as complex as all this looks,
 242  * but we want the default path for a system call return to
 243  * go as quickly as possible which is why some of this is
 244  * less clear than it otherwise should be.
 245  */
 246
 247         # userspace resumption stub bypassing syscall exit tracing
 248         ALIGN
 249 ret_from_exception:
 250         preempt_stop(CLBR_ANY)
 251 ret_from_intr:
 252         GET_THREAD_INFO(%ebp)
 253 #ifdef CONFIG_VM86
 254         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS and CS
 255         movb    PT_CS(%esp), %al
 256         andl    $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 257 #else
 258         /*
 259          * We can be coming here from child spawned by kernel_thread().
 260          */
 261         movl    PT_CS(%esp), %eax
 262         andl    $SEGMENT_RPL_MASK, %eax
 263 #endif
 264         cmpl    $USER_RPL, %eax
 265         jb      resume_kernel                   # not returning to v8086 or userspace
 266
 267 ENTRY(resume_userspace)
 268         LOCKDEP_SYS_EXIT
 269         DISABLE_INTERRUPTS(CLBR_ANY)            # make sure we don't miss an interrupt
 270                                                 # setting need_resched or sigpending
 271                                                 # between sampling and the iret
 272         TRACE_IRQS_OFF
 273         movl    TI_flags(%ebp), %ecx
 274         andl    $_TIF_WORK_MASK, %ecx           # is there any work to be done on
 275                                                 # int/exception return?
 276         jne     work_pending
 277         jmp     restore_all
 278 END(ret_from_exception)
 279
 280 #ifdef CONFIG_PREEMPT
 281 ENTRY(resume_kernel)
 282         DISABLE_INTERRUPTS(CLBR_ANY)
 283 need_resched:
 284         cmpl    $0, PER_CPU_VAR(__preempt_count)
 285         jnz     restore_all
 286         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
 287         jz      restore_all
 288         call    preempt_schedule_irq
 289         jmp     need_resched
 290 END(resume_kernel)
 291 #endif
 292
 293 /*
 294  * SYSENTER_RETURN points to after the SYSENTER instruction
 295  * in the vsyscall page.  See vsyscall-sysentry.S, which defines
 296  * the symbol.
 297  */
 298
 299         # SYSENTER  call handler stub
 300 ENTRY(entry_SYSENTER_32)
 301         movl    TSS_sysenter_sp0(%esp), %esp
 302 sysenter_past_esp:
 303         /*
 304          * Interrupts are disabled here, but we can't trace it until
 305          * enough kernel state to call TRACE_IRQS_OFF can be called - but
 306          * we immediately enable interrupts at that point anyway.
 307          */
 308         pushl   $__USER_DS
 309         pushl   %ebp
 310         pushfl
 311         orl     $X86_EFLAGS_IF, (%esp)
 312         pushl   $__USER_CS
 313         /*
 314          * Push current_thread_info()->sysenter_return to the stack.
 315          * A tiny bit of offset fixup is necessary: TI_sysenter_return
 316          * is relative to thread_info, which is at the bottom of the
 317          * kernel stack page.  4*4 means the 4 words pushed above;
 318          * TOP_OF_KERNEL_STACK_PADDING takes us to the top of the stack;
 319          * and THREAD_SIZE takes us to the bottom.
 320          */
 321         pushl   ((TI_sysenter_return) - THREAD_SIZE + TOP_OF_KERNEL_STACK_PADDING + 4*4)(%esp)
 322
 323         pushl   %eax
 324         SAVE_ALL
 325         ENABLE_INTERRUPTS(CLBR_NONE)
 326
 327 /*
 328  * Load the potential sixth argument from user stack.
 329  * Careful about security.
 330  */
 331         cmpl    $__PAGE_OFFSET-3, %ebp
 332         jae     syscall_fault
 333         ASM_STAC
 334 1:      movl    (%ebp), %ebp
 335         ASM_CLAC
 336         movl    %ebp, PT_EBP(%esp)
 337         _ASM_EXTABLE(1b, syscall_fault)
 338
 339         GET_THREAD_INFO(%ebp)
 340
 341         testl   $_TIF_WORK_SYSCALL_ENTRY, TI_flags(%ebp)
 342         jnz     sysenter_audit
 343 sysenter_do_call:
 344         cmpl    $(NR_syscalls), %eax
 345         jae     sysenter_badsys
 346         call    *sys_call_table(, %eax, 4)
 347 sysenter_after_call:
 348         movl    %eax, PT_EAX(%esp)
 349         LOCKDEP_SYS_EXIT
 350         DISABLE_INTERRUPTS(CLBR_ANY)
 351         TRACE_IRQS_OFF
 352         movl    TI_flags(%ebp), %ecx
 353         testl   $_TIF_ALLWORK_MASK, %ecx
 354         jnz     sysexit_audit
 355 sysenter_exit:
 356 /* if something modifies registers it must also disable sysexit */
 357         movl    PT_EIP(%esp), %edx
 358         movl    PT_OLDESP(%esp), %ecx
 359         xorl    %ebp, %ebp
 360         TRACE_IRQS_ON
 361 1:      mov     PT_FS(%esp), %fs
 362         PTGS_TO_GS
 363         ENABLE_INTERRUPTS_SYSEXIT
 364
 365 #ifdef CONFIG_AUDITSYSCALL
 366 sysenter_audit:
 367         testl   $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT), TI_flags(%ebp)
 368         jnz     syscall_trace_entry
 369         /* movl PT_EAX(%esp), %eax already set, syscall number: 1st arg to audit */
 370         movl    PT_EBX(%esp), %edx              /* ebx/a0: 2nd arg to audit */
 371         /* movl PT_ECX(%esp), %ecx already set, a1: 3nd arg to audit */
 372         pushl   PT_ESI(%esp)                    /* a3: 5th arg */
 373         pushl   PT_EDX+4(%esp)                  /* a2: 4th arg */
 374         call    __audit_syscall_entry
 375         popl    %ecx                            /* get that remapped edx off the stack */
 376         popl    %ecx                            /* get that remapped esi off the stack */
 377         movl    PT_EAX(%esp), %eax              /* reload syscall number */
 378         jmp     sysenter_do_call
 379
 380 sysexit_audit:
 381         testl   $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 382         jnz     syscall_exit_work
 383         TRACE_IRQS_ON
 384         ENABLE_INTERRUPTS(CLBR_ANY)
 385         movl    %eax, %edx                      /* second arg, syscall return value */
 386         cmpl    $-MAX_ERRNO, %eax               /* is it an error ? */
 387         setbe %al                               /* 1 if so, 0 if not */
 388         movzbl %al, %eax                        /* zero-extend that */
 389         call    __audit_syscall_exit
 390         DISABLE_INTERRUPTS(CLBR_ANY)
 391         TRACE_IRQS_OFF
 392         movl    TI_flags(%ebp), %ecx
 393         testl   $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 394         jnz     syscall_exit_work
 395         movl    PT_EAX(%esp), %eax              /* reload syscall return value */
 396         jmp     sysenter_exit
 397 #endif
 398
 399 .pushsection .fixup, "ax"
 400 2:      movl    $0, PT_FS(%esp)
 401         jmp     1b
 402 .popsection
 403         _ASM_EXTABLE(1b, 2b)
 404         PTGS_TO_GS_EX
 405 ENDPROC(entry_SYSENTER_32)
 406
 407         # system call handler stub
 408 ENTRY(entry_INT80_32)
 409         ASM_CLAC
 410         pushl   %eax                            # save orig_eax
 411         SAVE_ALL
 412         GET_THREAD_INFO(%ebp)
 413                                                 # system call tracing in operation / emulation
 414         testl   $_TIF_WORK_SYSCALL_ENTRY, TI_flags(%ebp)
 415         jnz     syscall_trace_entry
 416         cmpl    $(NR_syscalls), %eax
 417         jae     syscall_badsys
 418 syscall_call:
 419         call    *sys_call_table(, %eax, 4)
 420 syscall_after_call:
 421         movl    %eax, PT_EAX(%esp)              # store the return value
 422 syscall_exit:
 423         LOCKDEP_SYS_EXIT
 424         DISABLE_INTERRUPTS(CLBR_ANY)            # make sure we don't miss an interrupt
 425                                                 # setting need_resched or sigpending
 426                                                 # between sampling and the iret
 427         TRACE_IRQS_OFF
 428         movl    TI_flags(%ebp), %ecx
 429         testl   $_TIF_ALLWORK_MASK, %ecx        # current->work
 430         jnz     syscall_exit_work
 431
 432 restore_all:
 433         TRACE_IRQS_IRET
 434 restore_all_notrace:
 435 #ifdef CONFIG_X86_ESPFIX32
 436         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS, SS and CS
 437         /*
 438          * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 439          * are returning to the kernel.
 440          * See comments in process.c:copy_thread() for details.
 441          */
 442         movb    PT_OLDSS(%esp), %ah
 443         movb    PT_CS(%esp), %al
 444         andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 445         cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
 446         je ldt_ss                               # returning to user-space with LDT SS
 447 #endif
 448 restore_nocheck:
 449         RESTORE_REGS 4                          # skip orig_eax/error_code
 450 irq_return:
 451         INTERRUPT_RETURN
 452 .section .fixup, "ax"
 453 ENTRY(iret_exc  )
 454         pushl   $0                              # no error code
 455         pushl   $do_iret_error
 456         jmp     error_code
 457 .previous
 458         _ASM_EXTABLE(irq_return, iret_exc)
 459
 460 #ifdef CONFIG_X86_ESPFIX32
 461 ldt_ss:
 462 #ifdef CONFIG_PARAVIRT
 463         /*
 464          * The kernel can't run on a non-flat stack if paravirt mode
 465          * is active.  Rather than try to fixup the high bits of
 466          * ESP, bypass this code entirely.  This may break DOSemu
 467          * and/or Wine support in a paravirt VM, although the option
 468          * is still available to implement the setting of the high
 469          * 16-bits in the INTERRUPT_RETURN paravirt-op.
 470          */
 471         cmpl    $0, pv_info+PARAVIRT_enabled
 472         jne     restore_nocheck
 473 #endif
 474
 475 /*
 476  * Setup and switch to ESPFIX stack
 477  *
 478  * We're returning to userspace with a 16 bit stack. The CPU will not
 479  * restore the high word of ESP for us on executing iret... This is an
 480  * "official" bug of all the x86-compatible CPUs, which we can work
 481  * around to make dosemu and wine happy. We do this by preloading the
 482  * high word of ESP with the high word of the userspace ESP while
 483  * compensating for the offset by changing to the ESPFIX segment with
 484  * a base address that matches for the difference.
 485  */
 486 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 487         mov     %esp, %edx                      /* load kernel esp */
 488         mov     PT_OLDESP(%esp), %eax           /* load userspace esp */
 489         mov     %dx, %ax                        /* eax: new kernel esp */
 490         sub     %eax, %edx                      /* offset (low word is 0) */
 491         shr     $16, %edx
 492         mov     %dl, GDT_ESPFIX_SS + 4          /* bits 16..23 */
 493         mov     %dh, GDT_ESPFIX_SS + 7          /* bits 24..31 */
 494         pushl   $__ESPFIX_SS
 495         pushl   %eax                            /* new kernel esp */
 496         /*
 497          * Disable interrupts, but do not irqtrace this section: we
 498          * will soon execute iret and the tracer was already set to
 499          * the irqstate after the IRET:
 500          */
 501         DISABLE_INTERRUPTS(CLBR_EAX)
 502         lss     (%esp), %esp                    /* switch to espfix segment */
 503         jmp     restore_nocheck
 504 #endif
 505 ENDPROC(entry_INT80_32)
 506
 507         # perform work that needs to be done immediately before resumption
 508         ALIGN
 509 work_pending:
 510         testb   $_TIF_NEED_RESCHED, %cl
 511         jz      work_notifysig
 512 work_resched:
 513         call    schedule
 514         LOCKDEP_SYS_EXIT
 515         DISABLE_INTERRUPTS(CLBR_ANY)            # make sure we don't miss an interrupt
 516                                                 # setting need_resched or sigpending
 517                                                 # between sampling and the iret
 518         TRACE_IRQS_OFF
 519         movl    TI_flags(%ebp), %ecx
 520         andl    $_TIF_WORK_MASK, %ecx           # is there any work to be done other
 521                                                 # than syscall tracing?
 522         jz      restore_all
 523         testb   $_TIF_NEED_RESCHED, %cl
 524         jnz     work_resched
 525
 526 work_notifysig:                                 # deal with pending signals and
 527                                                 # notify-resume requests
 528         TRACE_IRQS_ON
 529         ENABLE_INTERRUPTS(CLBR_NONE)
 530         movl    %esp, %eax
 531         xorl    %edx, %edx
 532         call    do_notify_resume
 533         jmp     resume_userspace
 534 END(work_pending)
 535
 536         # perform syscall exit tracing
 537         ALIGN
 538 syscall_trace_entry:
 539         movl    $-ENOSYS, PT_EAX(%esp)
 540         movl    %esp, %eax
 541         call    syscall_trace_enter
 542         /* What it returned is what we'll actually use.  */
 543         cmpl    $(NR_syscalls), %eax
 544         jnae    syscall_call
 545         jmp     syscall_exit
 546 END(syscall_trace_entry)
 547
 548         # perform syscall exit tracing
 549         ALIGN
 550 syscall_exit_work:
 551         testl   $_TIF_WORK_SYSCALL_EXIT, %ecx
 552         jz      work_pending
 553         TRACE_IRQS_ON
 554         ENABLE_INTERRUPTS(CLBR_ANY)             # could let syscall_trace_leave() call
 555                                                 # schedule() instead
 556         movl    %esp, %eax
 557         call    syscall_trace_leave
 558         jmp     resume_userspace
 559 END(syscall_exit_work)
 560
 561 syscall_fault:
 562         ASM_CLAC
 563         GET_THREAD_INFO(%ebp)
 564         movl    $-EFAULT, PT_EAX(%esp)
 565         jmp     resume_userspace
 566 END(syscall_fault)
 567
 568 syscall_badsys:
 569         movl    $-ENOSYS, %eax
 570         jmp     syscall_after_call
 571 END(syscall_badsys)
 572
 573 sysenter_badsys:
 574         movl    $-ENOSYS, %eax
 575         jmp     sysenter_after_call
 576 END(sysenter_badsys)
 577
 578 .macro FIXUP_ESPFIX_STACK
 579 /*
 580  * Switch back for ESPFIX stack to the normal zerobased stack
 581  *
 582  * We can't call C functions using the ESPFIX stack. This code reads
 583  * the high word of the segment base from the GDT and swiches to the
 584  * normal stack and adjusts ESP with the matching offset.
 585  */
 586 #ifdef CONFIG_X86_ESPFIX32
 587         /* fixup the stack */
 588         mov     GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 589         mov     GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 590         shl     $16, %eax
 591         addl    %esp, %eax                      /* the adjusted stack pointer */
 592         pushl   $__KERNEL_DS
 593         pushl   %eax
 594         lss     (%esp), %esp                    /* switch to the normal stack segment */
 595 #endif
 596 .endm
 597 .macro UNWIND_ESPFIX_STACK
 598 #ifdef CONFIG_X86_ESPFIX32
 599         movl    %ss, %eax
 600         /* see if on espfix stack */
 601         cmpw    $__ESPFIX_SS, %ax
 602         jne     27f
 603         movl    $__KERNEL_DS, %eax
 604         movl    %eax, %ds
 605         movl    %eax, %es
 606         /* switch to normal stack */
 607         FIXUP_ESPFIX_STACK
 608 27:
 609 #endif
 610 .endm
 611
 612 /*
 613  * Build the entry stubs with some assembler magic.
 614  * We pack 1 stub into every 8-byte block.
 615  */
 616         .align 8
 617 ENTRY(irq_entries_start)
 618     vector=FIRST_EXTERNAL_VECTOR
 619     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 620         pushl   $(~vector+0x80)                 /* Note: always in signed byte range */
 621     vector=vector+1
 622         jmp     common_interrupt
 623         .align  8
 624     .endr
 625 END(irq_entries_start)
 626
 627 /*
 628  * the CPU automatically disables interrupts when executing an IRQ vector,
 629  * so IRQ-flags tracing has to follow that:
 630  */
 631         .p2align CONFIG_X86_L1_CACHE_SHIFT
 632 common_interrupt:
 633         ASM_CLAC
 634         addl    $-0x80, (%esp)                  /* Adjust vector into the [-256, -1] range */
 635         SAVE_ALL
 636         TRACE_IRQS_OFF
 637         movl    %esp, %eax
 638         call    do_IRQ
 639         jmp     ret_from_intr
 640 ENDPROC(common_interrupt)
 641
 642 #define BUILD_INTERRUPT3(name, nr, fn)  \
 643 ENTRY(name)                             \
 644         ASM_CLAC;                       \
 645         pushl   $~(nr);                 \
 646         SAVE_ALL;                       \
 647         TRACE_IRQS_OFF                  \
 648         movl    %esp, %eax;             \
 649         call    fn;                     \
 650         jmp     ret_from_intr;          \
 651 ENDPROC(name)
 652
 653
 654 #ifdef CONFIG_TRACING
 655 # define TRACE_BUILD_INTERRUPT(name, nr)        BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
 656 #else
 657 # define TRACE_BUILD_INTERRUPT(name, nr)
 658 #endif
 659
 660 #define BUILD_INTERRUPT(name, nr)               \
 661         BUILD_INTERRUPT3(name, nr, smp_##name); \
 662         TRACE_BUILD_INTERRUPT(name, nr)
 663
 664 /* The include is where all of the SMP etc. interrupts come from */
 665 #include <asm/entry_arch.h>
 666
 667 ENTRY(coprocessor_error)
 668         ASM_CLAC
 669         pushl   $0
 670         pushl   $do_coprocessor_error
 671         jmp     error_code
 672 END(coprocessor_error)
 673
 674 ENTRY(simd_coprocessor_error)
 675         ASM_CLAC
 676         pushl   $0
 677 #ifdef CONFIG_X86_INVD_BUG
 678         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 679         ALTERNATIVE "pushl      $do_general_protection",        \
 680                     "pushl      $do_simd_coprocessor_error",    \
 681                     X86_FEATURE_XMM
 682 #else
 683         pushl   $do_simd_coprocessor_error
 684 #endif
 685         jmp     error_code
 686 END(simd_coprocessor_error)
 687
 688 ENTRY(device_not_available)
 689         ASM_CLAC
 690         pushl   $-1                             # mark this as an int
 691         pushl   $do_device_not_available
 692         jmp     error_code
 693 END(device_not_available)
 694
 695 #ifdef CONFIG_PARAVIRT
 696 ENTRY(native_iret)
 697         iret
 698         _ASM_EXTABLE(native_iret, iret_exc)
 699 END(native_iret)
 700
 701 ENTRY(native_irq_enable_sysexit)
 702         sti
 703         sysexit
 704 END(native_irq_enable_sysexit)
 705 #endif
 706
 707 ENTRY(overflow)
 708         ASM_CLAC
 709         pushl   $0
 710         pushl   $do_overflow
 711         jmp     error_code
 712 END(overflow)
 713
 714 ENTRY(bounds)
 715         ASM_CLAC
 716         pushl   $0
 717         pushl   $do_bounds
 718         jmp     error_code
 719 END(bounds)
 720
 721 ENTRY(invalid_op)
 722         ASM_CLAC
 723         pushl   $0
 724         pushl   $do_invalid_op
 725         jmp     error_code
 726 END(invalid_op)
 727
 728 ENTRY(coprocessor_segment_overrun)
 729         ASM_CLAC
 730         pushl   $0
 731         pushl   $do_coprocessor_segment_overrun
 732         jmp     error_code
 733 END(coprocessor_segment_overrun)
 734
 735 ENTRY(invalid_TSS)
 736         ASM_CLAC
 737         pushl   $do_invalid_TSS
 738         jmp     error_code
 739 END(invalid_TSS)
 740
 741 ENTRY(segment_not_present)
 742         ASM_CLAC
 743         pushl   $do_segment_not_present
 744         jmp     error_code
 745 END(segment_not_present)
 746
 747 ENTRY(stack_segment)
 748         ASM_CLAC
 749         pushl   $do_stack_segment
 750         jmp     error_code
 751 END(stack_segment)
 752
 753 ENTRY(alignment_check)
 754         ASM_CLAC
 755         pushl   $do_alignment_check
 756         jmp     error_code
 757 END(alignment_check)
 758
 759 ENTRY(divide_error)
 760         ASM_CLAC
 761         pushl   $0                              # no error code
 762         pushl   $do_divide_error
 763         jmp     error_code
 764 END(divide_error)
 765
 766 #ifdef CONFIG_X86_MCE
 767 ENTRY(machine_check)
 768         ASM_CLAC
 769         pushl   $0
 770         pushl   machine_check_vector
 771         jmp     error_code
 772 END(machine_check)
 773 #endif
 774
 775 ENTRY(spurious_interrupt_bug)
 776         ASM_CLAC
 777         pushl   $0
 778         pushl   $do_spurious_interrupt_bug
 779         jmp     error_code
 780 END(spurious_interrupt_bug)
 781
 782 #ifdef CONFIG_XEN
 783 /*
 784  * Xen doesn't set %esp to be precisely what the normal SYSENTER
 785  * entry point expects, so fix it up before using the normal path.
 786  */
 787 ENTRY(xen_sysenter_target)
 788         addl    $5*4, %esp                      /* remove xen-provided frame */
 789         jmp     sysenter_past_esp
 790
 791 ENTRY(xen_hypervisor_callback)
 792         pushl   $-1                             /* orig_ax = -1 => not a system call */
 793         SAVE_ALL
 794         TRACE_IRQS_OFF
 795
 796         /*
 797          * Check to see if we got the event in the critical
 798          * region in xen_iret_direct, after we've reenabled
 799          * events and checked for pending events.  This simulates
 800          * iret instruction's behaviour where it delivers a
 801          * pending interrupt when enabling interrupts:
 802          */
 803         movl    PT_EIP(%esp), %eax
 804         cmpl    $xen_iret_start_crit, %eax
 805         jb      1f
 806         cmpl    $xen_iret_end_crit, %eax
 807         jae     1f
 808
 809         jmp     xen_iret_crit_fixup
 810
 811 ENTRY(xen_do_upcall)
 812 1:      mov     %esp, %eax
 813         call    xen_evtchn_do_upcall
 814 #ifndef CONFIG_PREEMPT
 815         call    xen_maybe_preempt_hcall
 816 #endif
 817         jmp     ret_from_intr
 818 ENDPROC(xen_hypervisor_callback)
 819
 820 /*
 821  * Hypervisor uses this for application faults while it executes.
 822  * We get here for two reasons:
 823  *  1. Fault while reloading DS, ES, FS or GS
 824  *  2. Fault while executing IRET
 825  * Category 1 we fix up by reattempting the load, and zeroing the segment
 826  * register if the load fails.
 827  * Category 2 we fix up by jumping to do_iret_error. We cannot use the
 828  * normal Linux return path in this case because if we use the IRET hypercall
 829  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
 830  * We distinguish between categories by maintaining a status value in EAX.
 831  */
 832 ENTRY(xen_failsafe_callback)
 833         pushl   %eax
 834         movl    $1, %eax
 835 1:      mov     4(%esp), %ds
 836 2:      mov     8(%esp), %es
 837 3:      mov     12(%esp), %fs
 838 4:      mov     16(%esp), %gs
 839         /* EAX == 0 => Category 1 (Bad segment)
 840            EAX != 0 => Category 2 (Bad IRET) */
 841         testl   %eax, %eax
 842         popl    %eax
 843         lea     16(%esp), %esp
 844         jz      5f
 845         jmp     iret_exc
 846 5:      pushl   $-1                             /* orig_ax = -1 => not a system call */
 847         SAVE_ALL
 848         jmp     ret_from_exception
 849
 850 .section .fixup, "ax"
 851 6:      xorl    %eax, %eax
 852         movl    %eax, 4(%esp)
 853         jmp     1b
 854 7:      xorl    %eax, %eax
 855         movl    %eax, 8(%esp)
 856         jmp     2b
 857 8:      xorl    %eax, %eax
 858         movl    %eax, 12(%esp)
 859         jmp     3b
 860 9:      xorl    %eax, %eax
 861         movl    %eax, 16(%esp)
 862         jmp     4b
 863 .previous
 864         _ASM_EXTABLE(1b, 6b)
 865         _ASM_EXTABLE(2b, 7b)
 866         _ASM_EXTABLE(3b, 8b)
 867         _ASM_EXTABLE(4b, 9b)
 868 ENDPROC(xen_failsafe_callback)
 869
 870 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 871                 xen_evtchn_do_upcall)
 872
 873 #endif /* CONFIG_XEN */
 874
 875 #if IS_ENABLED(CONFIG_HYPERV)
 876
 877 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 878         hyperv_vector_handler)
 879
 880 #endif /* CONFIG_HYPERV */
 881
 882 #ifdef CONFIG_FUNCTION_TRACER
 883 #ifdef CONFIG_DYNAMIC_FTRACE
 884
 885 ENTRY(mcount)
 886         ret
 887 END(mcount)
 888
 889 ENTRY(ftrace_caller)
 890         pushl   %eax
 891         pushl   %ecx
 892         pushl   %edx
 893         pushl   $0                              /* Pass NULL as regs pointer */
 894         movl    4*4(%esp), %eax
 895         movl    0x4(%ebp), %edx
 896         movl    function_trace_op, %ecx
 897         subl    $MCOUNT_INSN_SIZE, %eax
 898
 899 .globl ftrace_call
 900 ftrace_call:
 901         call    ftrace_stub
 902
 903         addl    $4, %esp                        /* skip NULL pointer */
 904         popl    %edx
 905         popl    %ecx
 906         popl    %eax
 907 ftrace_ret:
 908 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 909 .globl ftrace_graph_call
 910 ftrace_graph_call:
 911         jmp     ftrace_stub
 912 #endif
 913
 914 .globl ftrace_stub
 915 ftrace_stub:
 916         ret
 917 END(ftrace_caller)
 918
 919 ENTRY(ftrace_regs_caller)
 920         pushf   /* push flags before compare (in cs location) */
 921
 922         /*
 923          * i386 does not save SS and ESP when coming from kernel.
 924          * Instead, to get sp, &regs->sp is used (see ptrace.h).
 925          * Unfortunately, that means eflags must be at the same location
 926          * as the current return ip is. We move the return ip into the
 927          * ip location, and move flags into the return ip location.
 928          */
 929         pushl   4(%esp)                         /* save return ip into ip slot */
 930
 931         pushl   $0                              /* Load 0 into orig_ax */
 932         pushl   %gs
 933         pushl   %fs
 934         pushl   %es
 935         pushl   %ds
 936         pushl   %eax
 937         pushl   %ebp
 938         pushl   %edi
 939         pushl   %esi
 940         pushl   %edx
 941         pushl   %ecx
 942         pushl   %ebx
 943
 944         movl    13*4(%esp), %eax                /* Get the saved flags */
 945         movl    %eax, 14*4(%esp)                /* Move saved flags into regs->flags location */
 946                                                 /* clobbering return ip */
 947         movl    $__KERNEL_CS, 13*4(%esp)
 948
 949         movl    12*4(%esp), %eax                /* Load ip (1st parameter) */
 950         subl    $MCOUNT_INSN_SIZE, %eax         /* Adjust ip */
 951         movl    0x4(%ebp), %edx                 /* Load parent ip (2nd parameter) */
 952         movl    function_trace_op, %ecx         /* Save ftrace_pos in 3rd parameter */
 953         pushl   %esp                            /* Save pt_regs as 4th parameter */
 954
 955 GLOBAL(ftrace_regs_call)
 956         call    ftrace_stub
 957
 958         addl    $4, %esp                        /* Skip pt_regs */
 959         movl    14*4(%esp), %eax                /* Move flags back into cs */
 960         movl    %eax, 13*4(%esp)                /* Needed to keep addl  from modifying flags */
 961         movl    12*4(%esp), %eax                /* Get return ip from regs->ip */
 962         movl    %eax, 14*4(%esp)                /* Put return ip back for ret */
 963
 964         popl    %ebx
 965         popl    %ecx
 966         popl    %edx
 967         popl    %esi
 968         popl    %edi
 969         popl    %ebp
 970         popl    %eax
 971         popl    %ds
 972         popl    %es
 973         popl    %fs
 974         popl    %gs
 975         addl    $8, %esp                        /* Skip orig_ax and ip */
 976         popf                                    /* Pop flags at end (no addl to corrupt flags) */
 977         jmp     ftrace_ret
 978
 979         popf
 980         jmp     ftrace_stub
 981 #else /* ! CONFIG_DYNAMIC_FTRACE */
 982
 983 ENTRY(mcount)
 984         cmpl    $__PAGE_OFFSET, %esp
 985         jb      ftrace_stub                     /* Paging not enabled yet? */
 986
 987         cmpl    $ftrace_stub, ftrace_trace_function
 988         jnz     trace
 989 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 990         cmpl    $ftrace_stub, ftrace_graph_return
 991         jnz     ftrace_graph_caller
 992
 993         cmpl    $ftrace_graph_entry_stub, ftrace_graph_entry
 994         jnz     ftrace_graph_caller
 995 #endif
 996 .globl ftrace_stub
 997 ftrace_stub:
 998         ret
 999
1000         /* taken from glibc */
1001 trace:
1002         pushl   %eax
1003         pushl   %ecx
1004         pushl   %edx
1005         movl    0xc(%esp), %eax
1006         movl    0x4(%ebp), %edx
1007         subl    $MCOUNT_INSN_SIZE, %eax
1008
1009         call    *ftrace_trace_function
1010
1011         popl    %edx
1012         popl    %ecx
1013         popl    %eax
1014         jmp     ftrace_stub
1015 END(mcount)
1016 #endif /* CONFIG_DYNAMIC_FTRACE */
1017 #endif /* CONFIG_FUNCTION_TRACER */
1018
1019 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1020 ENTRY(ftrace_graph_caller)
1021         pushl   %eax
1022         pushl   %ecx
1023         pushl   %edx
1024         movl    0xc(%esp), %eax
1025         lea     0x4(%ebp), %edx
1026         movl    (%ebp), %ecx
1027         subl    $MCOUNT_INSN_SIZE, %eax
1028         call    prepare_ftrace_return
1029         popl    %edx
1030         popl    %ecx
1031         popl    %eax
1032         ret
1033 END(ftrace_graph_caller)
1034
1035 .globl return_to_handler
1036 return_to_handler:
1037         pushl   %eax
1038         pushl   %edx
1039         movl    %ebp, %eax
1040         call    ftrace_return_to_handler
1041         movl    %eax, %ecx
1042         popl    %edx
1043         popl    %eax
1044         jmp     *%ecx
1045 #endif
1046
1047 #ifdef CONFIG_TRACING
1048 ENTRY(trace_page_fault)
1049         ASM_CLAC
1050         pushl   $trace_do_page_fault
1051         jmp     error_code
1052 END(trace_page_fault)
1053 #endif
1054
1055 ENTRY(page_fault)
1056         ASM_CLAC
1057         pushl   $do_page_fault
1058         ALIGN
1059 error_code:
1060         /* the function address is in %gs's slot on the stack */
1061         pushl   %fs
1062         pushl   %es
1063         pushl   %ds
1064         pushl   %eax
1065         pushl   %ebp
1066         pushl   %edi
1067         pushl   %esi
1068         pushl   %edx
1069         pushl   %ecx
1070         pushl   %ebx
1071         cld
1072         movl    $(__KERNEL_PERCPU), %ecx
1073         movl    %ecx, %fs
1074         UNWIND_ESPFIX_STACK
1075         GS_TO_REG %ecx
1076         movl    PT_GS(%esp), %edi               # get the function address
1077         movl    PT_ORIG_EAX(%esp), %edx         # get the error code
1078         movl    $-1, PT_ORIG_EAX(%esp)          # no syscall to restart
1079         REG_TO_PTGS %ecx
1080         SET_KERNEL_GS %ecx
1081         movl    $(__USER_DS), %ecx
1082         movl    %ecx, %ds
1083         movl    %ecx, %es
1084         TRACE_IRQS_OFF
1085         movl    %esp, %eax                      # pt_regs pointer
1086         call    *%edi
1087         jmp     ret_from_exception
1088 END(page_fault)
1089
1090 /*
1091  * Debug traps and NMI can happen at the one SYSENTER instruction
1092  * that sets up the real kernel stack. Check here, since we can't
1093  * allow the wrong stack to be used.
1094  *
1095  * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1096  * already pushed 3 words if it hits on the sysenter instruction:
1097  * eflags, cs and eip.
1098  *
1099  * We just load the right stack, and push the three (known) values
1100  * by hand onto the new stack - while updating the return eip past
1101  * the instruction that would have done it for sysenter.
1102  */
1103 .macro FIX_STACK offset ok label
1104         cmpw    $__KERNEL_CS, 4(%esp)
1105         jne     \ok
1106 \label:
1107         movl    TSS_sysenter_sp0 + \offset(%esp), %esp
1108         pushfl
1109         pushl   $__KERNEL_CS
1110         pushl   $sysenter_past_esp
1111 .endm
1112
1113 ENTRY(debug)
1114         ASM_CLAC
1115         cmpl    $entry_SYSENTER_32, (%esp)
1116         jne     debug_stack_correct
1117         FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1118 debug_stack_correct:
1119         pushl   $-1                             # mark this as an int
1120         SAVE_ALL
1121         TRACE_IRQS_OFF
1122         xorl    %edx, %edx                      # error code 0
1123         movl    %esp, %eax                      # pt_regs pointer
1124         call    do_debug
1125         jmp     ret_from_exception
1126 END(debug)
1127
1128 /*
1129  * NMI is doubly nasty. It can happen _while_ we're handling
1130  * a debug fault, and the debug fault hasn't yet been able to
1131  * clear up the stack. So we first check whether we got  an
1132  * NMI on the sysenter entry path, but after that we need to
1133  * check whether we got an NMI on the debug path where the debug
1134  * fault happened on the sysenter path.
1135  */
1136 ENTRY(nmi)
1137         ASM_CLAC
1138 #ifdef CONFIG_X86_ESPFIX32
1139         pushl   %eax
1140         movl    %ss, %eax
1141         cmpw    $__ESPFIX_SS, %ax
1142         popl    %eax
1143         je      nmi_espfix_stack
1144 #endif
1145         cmpl    $entry_SYSENTER_32, (%esp)
1146         je      nmi_stack_fixup
1147         pushl   %eax
1148         movl    %esp, %eax
1149         /*
1150          * Do not access memory above the end of our stack page,
1151          * it might not exist.
1152          */
1153         andl    $(THREAD_SIZE-1), %eax
1154         cmpl    $(THREAD_SIZE-20), %eax
1155         popl    %eax
1156         jae     nmi_stack_correct
1157         cmpl    $entry_SYSENTER_32, 12(%esp)
1158         je      nmi_debug_stack_check
1159 nmi_stack_correct:
1160         pushl   %eax
1161         SAVE_ALL
1162         xorl    %edx, %edx                      # zero error code
1163         movl    %esp, %eax                      # pt_regs pointer
1164         call    do_nmi
1165         jmp     restore_all_notrace
1166
1167 nmi_stack_fixup:
1168         FIX_STACK 12, nmi_stack_correct, 1
1169         jmp     nmi_stack_correct
1170
1171 nmi_debug_stack_check:
1172         cmpw    $__KERNEL_CS, 16(%esp)
1173         jne     nmi_stack_correct
1174         cmpl    $debug, (%esp)
1175         jb      nmi_stack_correct
1176         cmpl    $debug_esp_fix_insn, (%esp)
1177         ja      nmi_stack_correct
1178         FIX_STACK 24, nmi_stack_correct, 1
1179         jmp     nmi_stack_correct
1180
1181 #ifdef CONFIG_X86_ESPFIX32
1182 nmi_espfix_stack:
1183         /*
1184          * create the pointer to lss back
1185          */
1186         pushl   %ss
1187         pushl   %esp
1188         addl    $4, (%esp)
1189         /* copy the iret frame of 12 bytes */
1190         .rept 3
1191         pushl   16(%esp)
1192         .endr
1193         pushl   %eax
1194         SAVE_ALL
1195         FIXUP_ESPFIX_STACK                      # %eax == %esp
1196         xorl    %edx, %edx                      # zero error code
1197         call    do_nmi
1198         RESTORE_REGS
1199         lss     12+4(%esp), %esp                # back to espfix stack
1200         jmp     irq_return
1201 #endif
1202 END(nmi)
1203
1204 ENTRY(int3)
1205         ASM_CLAC
1206         pushl   $-1                             # mark this as an int
1207         SAVE_ALL
1208         TRACE_IRQS_OFF
1209         xorl    %edx, %edx                      # zero error code
1210         movl    %esp, %eax                      # pt_regs pointer
1211         call    do_int3
1212         jmp     ret_from_exception
1213 END(int3)
1214
1215 ENTRY(general_protection)
1216         pushl   $do_general_protection
1217         jmp     error_code
1218 END(general_protection)
1219
1220 #ifdef CONFIG_KVM_GUEST
1221 ENTRY(async_page_fault)
1222         ASM_CLAC
1223         pushl   $do_async_page_fault
1224         jmp     error_code
1225 END(async_page_fault)
1226 #endif