arch/x86/entry/entry_32.S

   1 /*
   2  *  Copyright (C) 1991,1992  Linus Torvalds
   3  *
   4  * entry_32.S contains the system-call and low-level fault and trap handling routines.
   5  *
   6  * Stack layout while running C code:
   7  *      ptrace needs to have all registers on the stack.
   8  *      If the order here is changed, it needs to be
   9  *      updated in fork.c:copy_process(), signal.c:do_signal(),
  10  *      ptrace.c and ptrace.h
  11  *
  12  *       0(%esp) - %ebx
  13  *       4(%esp) - %ecx
  14  *       8(%esp) - %edx
  15  *       C(%esp) - %esi
  16  *      10(%esp) - %edi
  17  *      14(%esp) - %ebp
  18  *      18(%esp) - %eax
  19  *      1C(%esp) - %ds
  20  *      20(%esp) - %es
  21  *      24(%esp) - %fs
  22  *      28(%esp) - %gs          saved iff !CONFIG_X86_32_LAZY_GS
  23  *      2C(%esp) - orig_eax
  24  *      30(%esp) - %eip
  25  *      34(%esp) - %cs
  26  *      38(%esp) - %eflags
  27  *      3C(%esp) - %oldesp
  28  *      40(%esp) - %oldss
  29  */
  30
  31 #include <linux/linkage.h>
  32 #include <linux/err.h>
  33 #include <asm/thread_info.h>
  34 #include <asm/irqflags.h>
  35 #include <asm/errno.h>
  36 #include <asm/segment.h>
  37 #include <asm/smp.h>
  38 #include <asm/page_types.h>
  39 #include <asm/percpu.h>
  40 #include <asm/processor-flags.h>
  41 #include <asm/ftrace.h>
  42 #include <asm/irq_vectors.h>
  43 #include <asm/cpufeatures.h>
  44 #include <asm/alternative-asm.h>
  45 #include <asm/asm.h>
  46 #include <asm/smap.h>
  47 #include <asm/export.h>
  48 #include <asm/frame.h>
  49
  50         .section .entry.text, "ax"
  51
  52 /*
  53  * We use macros for low-level operations which need to be overridden
  54  * for paravirtualization.  The following will never clobber any registers:
  55  *   INTERRUPT_RETURN (aka. "iret")
  56  *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  57  *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  58  *
  59  * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  60  * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  61  * Allowing a register to be clobbered can shrink the paravirt replacement
  62  * enough to patch inline, increasing performance.
  63  */
  64
  65 #ifdef CONFIG_PREEMPT
  66 # define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  67 #else
  68 # define preempt_stop(clobbers)
  69 # define resume_kernel          restore_all
  70 #endif
  71
  72 .macro TRACE_IRQS_IRET
  73 #ifdef CONFIG_TRACE_IRQFLAGS
  74         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
  75         jz      1f
  76         TRACE_IRQS_ON
  77 1:
  78 #endif
  79 .endm
  80
  81 /*
  82  * User gs save/restore
  83  *
  84  * %gs is used for userland TLS and kernel only uses it for stack
  85  * canary which is required to be at %gs:20 by gcc.  Read the comment
  86  * at the top of stackprotector.h for more info.
  87  *
  88  * Local labels 98 and 99 are used.
  89  */
  90 #ifdef CONFIG_X86_32_LAZY_GS
  91
  92  /* unfortunately push/pop can't be no-op */
  93 .macro PUSH_GS
  94         pushl   $0
  95 .endm
  96 .macro POP_GS pop=0
  97         addl    $(4 + \pop), %esp
  98 .endm
  99 .macro POP_GS_EX
 100 .endm
 101
 102  /* all the rest are no-op */
 103 .macro PTGS_TO_GS
 104 .endm
 105 .macro PTGS_TO_GS_EX
 106 .endm
 107 .macro GS_TO_REG reg
 108 .endm
 109 .macro REG_TO_PTGS reg
 110 .endm
 111 .macro SET_KERNEL_GS reg
 112 .endm
 113
 114 #else   /* CONFIG_X86_32_LAZY_GS */
 115
 116 .macro PUSH_GS
 117         pushl   %gs
 118 .endm
 119
 120 .macro POP_GS pop=0
 121 98:     popl    %gs
 122   .if \pop <> 0
 123         add     $\pop, %esp
 124   .endif
 125 .endm
 126 .macro POP_GS_EX
 127 .pushsection .fixup, "ax"
 128 99:     movl    $0, (%esp)
 129         jmp     98b
 130 .popsection
 131         _ASM_EXTABLE(98b, 99b)
 132 .endm
 133
 134 .macro PTGS_TO_GS
 135 98:     mov     PT_GS(%esp), %gs
 136 .endm
 137 .macro PTGS_TO_GS_EX
 138 .pushsection .fixup, "ax"
 139 99:     movl    $0, PT_GS(%esp)
 140         jmp     98b
 141 .popsection
 142         _ASM_EXTABLE(98b, 99b)
 143 .endm
 144
 145 .macro GS_TO_REG reg
 146         movl    %gs, \reg
 147 .endm
 148 .macro REG_TO_PTGS reg
 149         movl    \reg, PT_GS(%esp)
 150 .endm
 151 .macro SET_KERNEL_GS reg
 152         movl    $(__KERNEL_STACK_CANARY), \reg
 153         movl    \reg, %gs
 154 .endm
 155
 156 #endif /* CONFIG_X86_32_LAZY_GS */
 157
 158 .macro SAVE_ALL pt_regs_ax=%eax
 159         cld
 160         PUSH_GS
 161         pushl   %fs
 162         pushl   %es
 163         pushl   %ds
 164         pushl   \pt_regs_ax
 165         pushl   %ebp
 166         pushl   %edi
 167         pushl   %esi
 168         pushl   %edx
 169         pushl   %ecx
 170         pushl   %ebx
 171         movl    $(__USER_DS), %edx
 172         movl    %edx, %ds
 173         movl    %edx, %es
 174         movl    $(__KERNEL_PERCPU), %edx
 175         movl    %edx, %fs
 176         SET_KERNEL_GS %edx
 177 .endm
 178
 179 /*
 180  * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
 181  * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
 182  * is just setting the LSB, which makes it an invalid stack address and is also
 183  * a signal to the unwinder that it's a pt_regs pointer in disguise.
 184  *
 185  * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
 186  * original rbp.
 187  */
 188 .macro ENCODE_FRAME_POINTER
 189 #ifdef CONFIG_FRAME_POINTER
 190         mov %esp, %ebp
 191         orl $0x1, %ebp
 192 #endif
 193 .endm
 194
 195 .macro RESTORE_INT_REGS
 196         popl    %ebx
 197         popl    %ecx
 198         popl    %edx
 199         popl    %esi
 200         popl    %edi
 201         popl    %ebp
 202         popl    %eax
 203 .endm
 204
 205 .macro RESTORE_REGS pop=0
 206         RESTORE_INT_REGS
 207 1:      popl    %ds
 208 2:      popl    %es
 209 3:      popl    %fs
 210         POP_GS \pop
 211 .pushsection .fixup, "ax"
 212 4:      movl    $0, (%esp)
 213         jmp     1b
 214 5:      movl    $0, (%esp)
 215         jmp     2b
 216 6:      movl    $0, (%esp)
 217         jmp     3b
 218 .popsection
 219         _ASM_EXTABLE(1b, 4b)
 220         _ASM_EXTABLE(2b, 5b)
 221         _ASM_EXTABLE(3b, 6b)
 222         POP_GS_EX
 223 .endm
 224
 225 /*
 226  * %eax: prev task
 227  * %edx: next task
 228  */
 229 ENTRY(__switch_to_asm)
 230         /*
 231          * Save callee-saved registers
 232          * This must match the order in struct inactive_task_frame
 233          */
 234         pushl   %ebp
 235         pushl   %ebx
 236         pushl   %edi
 237         pushl   %esi
 238
 239         /* switch stack */
 240         movl    %esp, TASK_threadsp(%eax)
 241         movl    TASK_threadsp(%edx), %esp
 242
 243 #ifdef CONFIG_CC_STACKPROTECTOR
 244         movl    TASK_stack_canary(%edx), %ebx
 245         movl    %ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
 246 #endif
 247
 248         /* restore callee-saved registers */
 249         popl    %esi
 250         popl    %edi
 251         popl    %ebx
 252         popl    %ebp
 253
 254         jmp     __switch_to
 255 END(__switch_to_asm)
 256
 257 /*
 258  * A newly forked process directly context switches into this address.
 259  *
 260  * eax: prev task we switched from
 261  * ebx: kernel thread func (NULL for user thread)
 262  * edi: kernel thread arg
 263  */
 264 ENTRY(ret_from_fork)
 265         FRAME_BEGIN             /* help unwinder find end of stack */
 266
 267         /*
 268          * schedule_tail() is asmlinkage so we have to put its 'prev' argument
 269          * on the stack.
 270          */
 271         pushl   %eax
 272         call    schedule_tail
 273         popl    %eax
 274
 275         testl   %ebx, %ebx
 276         jnz     1f              /* kernel threads are uncommon */
 277
 278 2:
 279         /* When we fork, we trace the syscall return in the child, too. */
 280         leal    FRAME_OFFSET(%esp), %eax
 281         call    syscall_return_slowpath
 282         FRAME_END
 283         jmp     restore_all
 284
 285         /* kernel thread */
 286 1:      movl    %edi, %eax
 287         call    *%ebx
 288         /*
 289          * A kernel thread is allowed to return here after successfully
 290          * calling do_execve().  Exit to userspace to complete the execve()
 291          * syscall.
 292          */
 293         movl    $0, PT_EAX(%esp)
 294         jmp     2b
 295 END(ret_from_fork)
 296
 297 /*
 298  * Return to user mode is not as complex as all this looks,
 299  * but we want the default path for a system call return to
 300  * go as quickly as possible which is why some of this is
 301  * less clear than it otherwise should be.
 302  */
 303
 304         # userspace resumption stub bypassing syscall exit tracing
 305         ALIGN
 306 ret_from_exception:
 307         preempt_stop(CLBR_ANY)
 308 ret_from_intr:
 309 #ifdef CONFIG_VM86
 310         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS and CS
 311         movb    PT_CS(%esp), %al
 312         andl    $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 313 #else
 314         /*
 315          * We can be coming here from child spawned by kernel_thread().
 316          */
 317         movl    PT_CS(%esp), %eax
 318         andl    $SEGMENT_RPL_MASK, %eax
 319 #endif
 320         cmpl    $USER_RPL, %eax
 321         jb      resume_kernel                   # not returning to v8086 or userspace
 322
 323 ENTRY(resume_userspace)
 324         DISABLE_INTERRUPTS(CLBR_ANY)
 325         TRACE_IRQS_OFF
 326         movl    %esp, %eax
 327         call    prepare_exit_to_usermode
 328         jmp     restore_all
 329 END(ret_from_exception)
 330
 331 #ifdef CONFIG_PREEMPT
 332 ENTRY(resume_kernel)
 333         DISABLE_INTERRUPTS(CLBR_ANY)
 334 .Lneed_resched:
 335         cmpl    $0, PER_CPU_VAR(__preempt_count)
 336         jnz     restore_all
 337         testl   $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
 338         jz      restore_all
 339         call    preempt_schedule_irq
 340         jmp     .Lneed_resched
 341 END(resume_kernel)
 342 #endif
 343
 344 GLOBAL(__begin_SYSENTER_singlestep_region)
 345 /*
 346  * All code from here through __end_SYSENTER_singlestep_region is subject
 347  * to being single-stepped if a user program sets TF and executes SYSENTER.
 348  * There is absolutely nothing that we can do to prevent this from happening
 349  * (thanks Intel!).  To keep our handling of this situation as simple as
 350  * possible, we handle TF just like AC and NT, except that our #DB handler
 351  * will ignore all of the single-step traps generated in this range.
 352  */
 353
 354 #ifdef CONFIG_XEN
 355 /*
 356  * Xen doesn't set %esp to be precisely what the normal SYSENTER
 357  * entry point expects, so fix it up before using the normal path.
 358  */
 359 ENTRY(xen_sysenter_target)
 360         addl    $5*4, %esp                      /* remove xen-provided frame */
 361         jmp     .Lsysenter_past_esp
 362 #endif
 363
 364 /*
 365  * 32-bit SYSENTER entry.
 366  *
 367  * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
 368  * if X86_FEATURE_SEP is available.  This is the preferred system call
 369  * entry on 32-bit systems.
 370  *
 371  * The SYSENTER instruction, in principle, should *only* occur in the
 372  * vDSO.  In practice, a small number of Android devices were shipped
 373  * with a copy of Bionic that inlined a SYSENTER instruction.  This
 374  * never happened in any of Google's Bionic versions -- it only happened
 375  * in a narrow range of Intel-provided versions.
 376  *
 377  * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
 378  * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
 379  * SYSENTER does not save anything on the stack,
 380  * and does not save old EIP (!!!), ESP, or EFLAGS.
 381  *
 382  * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
 383  * user and/or vm86 state), we explicitly disable the SYSENTER
 384  * instruction in vm86 mode by reprogramming the MSRs.
 385  *
 386  * Arguments:
 387  * eax  system call number
 388  * ebx  arg1
 389  * ecx  arg2
 390  * edx  arg3
 391  * esi  arg4
 392  * edi  arg5
 393  * ebp  user stack
 394  * 0(%ebp) arg6
 395  */
 396 ENTRY(entry_SYSENTER_32)
 397         movl    TSS_sysenter_sp0(%esp), %esp
 398 .Lsysenter_past_esp:
 399         pushl   $__USER_DS              /* pt_regs->ss */
 400         pushl   %ebp                    /* pt_regs->sp (stashed in bp) */
 401         pushfl                          /* pt_regs->flags (except IF = 0) */
 402         orl     $X86_EFLAGS_IF, (%esp)  /* Fix IF */
 403         pushl   $__USER_CS              /* pt_regs->cs */
 404         pushl   $0                      /* pt_regs->ip = 0 (placeholder) */
 405         pushl   %eax                    /* pt_regs->orig_ax */
 406         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 407
 408         /*
 409          * SYSENTER doesn't filter flags, so we need to clear NT, AC
 410          * and TF ourselves.  To save a few cycles, we can check whether
 411          * either was set instead of doing an unconditional popfq.
 412          * This needs to happen before enabling interrupts so that
 413          * we don't get preempted with NT set.
 414          *
 415          * If TF is set, we will single-step all the way to here -- do_debug
 416          * will ignore all the traps.  (Yes, this is slow, but so is
 417          * single-stepping in general.  This allows us to avoid having
 418          * a more complicated code to handle the case where a user program
 419          * forces us to single-step through the SYSENTER entry code.)
 420          *
 421          * NB.: .Lsysenter_fix_flags is a label with the code under it moved
 422          * out-of-line as an optimization: NT is unlikely to be set in the
 423          * majority of the cases and instead of polluting the I$ unnecessarily,
 424          * we're keeping that code behind a branch which will predict as
 425          * not-taken and therefore its instructions won't be fetched.
 426          */
 427         testl   $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
 428         jnz     .Lsysenter_fix_flags
 429 .Lsysenter_flags_fixed:
 430
 431         /*
 432          * User mode is traced as though IRQs are on, and SYSENTER
 433          * turned them off.
 434          */
 435         TRACE_IRQS_OFF
 436
 437         movl    %esp, %eax
 438         call    do_fast_syscall_32
 439         /* XEN PV guests always use IRET path */
 440         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 441                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 442
 443 /* Opportunistic SYSEXIT */
 444         TRACE_IRQS_ON                   /* User mode traces as IRQs on. */
 445         movl    PT_EIP(%esp), %edx      /* pt_regs->ip */
 446         movl    PT_OLDESP(%esp), %ecx   /* pt_regs->sp */
 447 1:      mov     PT_FS(%esp), %fs
 448         PTGS_TO_GS
 449         popl    %ebx                    /* pt_regs->bx */
 450         addl    $2*4, %esp              /* skip pt_regs->cx and pt_regs->dx */
 451         popl    %esi                    /* pt_regs->si */
 452         popl    %edi                    /* pt_regs->di */
 453         popl    %ebp                    /* pt_regs->bp */
 454         popl    %eax                    /* pt_regs->ax */
 455
 456         /*
 457          * Restore all flags except IF. (We restore IF separately because
 458          * STI gives a one-instruction window in which we won't be interrupted,
 459          * whereas POPF does not.)
 460          */
 461         addl    $PT_EFLAGS-PT_DS, %esp  /* point esp at pt_regs->flags */
 462         btr     $X86_EFLAGS_IF_BIT, (%esp)
 463         popfl
 464
 465         /*
 466          * Return back to the vDSO, which will pop ecx and edx.
 467          * Don't bother with DS and ES (they already contain __USER_DS).
 468          */
 469         sti
 470         sysexit
 471
 472 .pushsection .fixup, "ax"
 473 2:      movl    $0, PT_FS(%esp)
 474         jmp     1b
 475 .popsection
 476         _ASM_EXTABLE(1b, 2b)
 477         PTGS_TO_GS_EX
 478
 479 .Lsysenter_fix_flags:
 480         pushl   $X86_EFLAGS_FIXED
 481         popfl
 482         jmp     .Lsysenter_flags_fixed
 483 GLOBAL(__end_SYSENTER_singlestep_region)
 484 ENDPROC(entry_SYSENTER_32)
 485
 486 /*
 487  * 32-bit legacy system call entry.
 488  *
 489  * 32-bit x86 Linux system calls traditionally used the INT $0x80
 490  * instruction.  INT $0x80 lands here.
 491  *
 492  * This entry point can be used by any 32-bit perform system calls.
 493  * Instances of INT $0x80 can be found inline in various programs and
 494  * libraries.  It is also used by the vDSO's __kernel_vsyscall
 495  * fallback for hardware that doesn't support a faster entry method.
 496  * Restarted 32-bit system calls also fall back to INT $0x80
 497  * regardless of what instruction was originally used to do the system
 498  * call.  (64-bit programs can use INT $0x80 as well, but they can
 499  * only run on 64-bit kernels and therefore land in
 500  * entry_INT80_compat.)
 501  *
 502  * This is considered a slow path.  It is not used by most libc
 503  * implementations on modern hardware except during process startup.
 504  *
 505  * Arguments:
 506  * eax  system call number
 507  * ebx  arg1
 508  * ecx  arg2
 509  * edx  arg3
 510  * esi  arg4
 511  * edi  arg5
 512  * ebp  arg6
 513  */
 514 ENTRY(entry_INT80_32)
 515         ASM_CLAC
 516         pushl   %eax                    /* pt_regs->orig_ax */
 517         SAVE_ALL pt_regs_ax=$-ENOSYS    /* save rest */
 518
 519         /*
 520          * User mode is traced as though IRQs are on, and the interrupt gate
 521          * turned them off.
 522          */
 523         TRACE_IRQS_OFF
 524
 525         movl    %esp, %eax
 526         call    do_int80_syscall_32
 527 .Lsyscall_32_done:
 528
 529 restore_all:
 530         TRACE_IRQS_IRET
 531 .Lrestore_all_notrace:
 532 #ifdef CONFIG_X86_ESPFIX32
 533         ALTERNATIVE     "jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
 534
 535         movl    PT_EFLAGS(%esp), %eax           # mix EFLAGS, SS and CS
 536         /*
 537          * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 538          * are returning to the kernel.
 539          * See comments in process.c:copy_thread() for details.
 540          */
 541         movb    PT_OLDSS(%esp), %ah
 542         movb    PT_CS(%esp), %al
 543         andl    $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 544         cmpl    $((SEGMENT_LDT << 8) | USER_RPL), %eax
 545         je .Lldt_ss                             # returning to user-space with LDT SS
 546 #endif
 547 .Lrestore_nocheck:
 548         RESTORE_REGS 4                          # skip orig_eax/error_code
 549 .Lirq_return:
 550         INTERRUPT_RETURN
 551
 552 .section .fixup, "ax"
 553 ENTRY(iret_exc  )
 554         pushl   $0                              # no error code
 555         pushl   $do_iret_error
 556         jmp     common_exception
 557 .previous
 558         _ASM_EXTABLE(.Lirq_return, iret_exc)
 559
 560 #ifdef CONFIG_X86_ESPFIX32
 561 .Lldt_ss:
 562 /*
 563  * Setup and switch to ESPFIX stack
 564  *
 565  * We're returning to userspace with a 16 bit stack. The CPU will not
 566  * restore the high word of ESP for us on executing iret... This is an
 567  * "official" bug of all the x86-compatible CPUs, which we can work
 568  * around to make dosemu and wine happy. We do this by preloading the
 569  * high word of ESP with the high word of the userspace ESP while
 570  * compensating for the offset by changing to the ESPFIX segment with
 571  * a base address that matches for the difference.
 572  */
 573 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 574         mov     %esp, %edx                      /* load kernel esp */
 575         mov     PT_OLDESP(%esp), %eax           /* load userspace esp */
 576         mov     %dx, %ax                        /* eax: new kernel esp */
 577         sub     %eax, %edx                      /* offset (low word is 0) */
 578         shr     $16, %edx
 579         mov     %dl, GDT_ESPFIX_SS + 4          /* bits 16..23 */
 580         mov     %dh, GDT_ESPFIX_SS + 7          /* bits 24..31 */
 581         pushl   $__ESPFIX_SS
 582         pushl   %eax                            /* new kernel esp */
 583         /*
 584          * Disable interrupts, but do not irqtrace this section: we
 585          * will soon execute iret and the tracer was already set to
 586          * the irqstate after the IRET:
 587          */
 588         DISABLE_INTERRUPTS(CLBR_EAX)
 589         lss     (%esp), %esp                    /* switch to espfix segment */
 590         jmp     .Lrestore_nocheck
 591 #endif
 592 ENDPROC(entry_INT80_32)
 593
 594 .macro FIXUP_ESPFIX_STACK
 595 /*
 596  * Switch back for ESPFIX stack to the normal zerobased stack
 597  *
 598  * We can't call C functions using the ESPFIX stack. This code reads
 599  * the high word of the segment base from the GDT and swiches to the
 600  * normal stack and adjusts ESP with the matching offset.
 601  */
 602 #ifdef CONFIG_X86_ESPFIX32
 603         /* fixup the stack */
 604         mov     GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 605         mov     GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 606         shl     $16, %eax
 607         addl    %esp, %eax                      /* the adjusted stack pointer */
 608         pushl   $__KERNEL_DS
 609         pushl   %eax
 610         lss     (%esp), %esp                    /* switch to the normal stack segment */
 611 #endif
 612 .endm
 613 .macro UNWIND_ESPFIX_STACK
 614 #ifdef CONFIG_X86_ESPFIX32
 615         movl    %ss, %eax
 616         /* see if on espfix stack */
 617         cmpw    $__ESPFIX_SS, %ax
 618         jne     27f
 619         movl    $__KERNEL_DS, %eax
 620         movl    %eax, %ds
 621         movl    %eax, %es
 622         /* switch to normal stack */
 623         FIXUP_ESPFIX_STACK
 624 27:
 625 #endif
 626 .endm
 627
 628 /*
 629  * Build the entry stubs with some assembler magic.
 630  * We pack 1 stub into every 8-byte block.
 631  */
 632         .align 8
 633 ENTRY(irq_entries_start)
 634     vector=FIRST_EXTERNAL_VECTOR
 635     .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
 636         pushl   $(~vector+0x80)                 /* Note: always in signed byte range */
 637     vector=vector+1
 638         jmp     common_interrupt
 639         .align  8
 640     .endr
 641 END(irq_entries_start)
 642
 643 /*
 644  * the CPU automatically disables interrupts when executing an IRQ vector,
 645  * so IRQ-flags tracing has to follow that:
 646  */
 647         .p2align CONFIG_X86_L1_CACHE_SHIFT
 648 common_interrupt:
 649         ASM_CLAC
 650         addl    $-0x80, (%esp)                  /* Adjust vector into the [-256, -1] range */
 651         SAVE_ALL
 652         ENCODE_FRAME_POINTER
 653         TRACE_IRQS_OFF
 654         movl    %esp, %eax
 655         call    do_IRQ
 656         jmp     ret_from_intr
 657 ENDPROC(common_interrupt)
 658
 659 #define BUILD_INTERRUPT3(name, nr, fn)  \
 660 ENTRY(name)                             \
 661         ASM_CLAC;                       \
 662         pushl   $~(nr);                 \
 663         SAVE_ALL;                       \
 664         ENCODE_FRAME_POINTER;           \
 665         TRACE_IRQS_OFF                  \
 666         movl    %esp, %eax;             \
 667         call    fn;                     \
 668         jmp     ret_from_intr;          \
 669 ENDPROC(name)
 670
 671
 672 #ifdef CONFIG_TRACING
 673 # define TRACE_BUILD_INTERRUPT(name, nr)        BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
 674 #else
 675 # define TRACE_BUILD_INTERRUPT(name, nr)
 676 #endif
 677
 678 #define BUILD_INTERRUPT(name, nr)               \
 679         BUILD_INTERRUPT3(name, nr, smp_##name); \
 680         TRACE_BUILD_INTERRUPT(name, nr)
 681
 682 /* The include is where all of the SMP etc. interrupts come from */
 683 #include <asm/entry_arch.h>
 684
 685 ENTRY(coprocessor_error)
 686         ASM_CLAC
 687         pushl   $0
 688         pushl   $do_coprocessor_error
 689         jmp     common_exception
 690 END(coprocessor_error)
 691
 692 ENTRY(simd_coprocessor_error)
 693         ASM_CLAC
 694         pushl   $0
 695 #ifdef CONFIG_X86_INVD_BUG
 696         /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 697         ALTERNATIVE "pushl      $do_general_protection",        \
 698                     "pushl      $do_simd_coprocessor_error",    \
 699                     X86_FEATURE_XMM
 700 #else
 701         pushl   $do_simd_coprocessor_error
 702 #endif
 703         jmp     common_exception
 704 END(simd_coprocessor_error)
 705
 706 ENTRY(device_not_available)
 707         ASM_CLAC
 708         pushl   $-1                             # mark this as an int
 709         pushl   $do_device_not_available
 710         jmp     common_exception
 711 END(device_not_available)
 712
 713 #ifdef CONFIG_PARAVIRT
 714 ENTRY(native_iret)
 715         iret
 716         _ASM_EXTABLE(native_iret, iret_exc)
 717 END(native_iret)
 718 #endif
 719
 720 ENTRY(overflow)
 721         ASM_CLAC
 722         pushl   $0
 723         pushl   $do_overflow
 724         jmp     common_exception
 725 END(overflow)
 726
 727 ENTRY(bounds)
 728         ASM_CLAC
 729         pushl   $0
 730         pushl   $do_bounds
 731         jmp     common_exception
 732 END(bounds)
 733
 734 ENTRY(invalid_op)
 735         ASM_CLAC
 736         pushl   $0
 737         pushl   $do_invalid_op
 738         jmp     common_exception
 739 END(invalid_op)
 740
 741 ENTRY(coprocessor_segment_overrun)
 742         ASM_CLAC
 743         pushl   $0
 744         pushl   $do_coprocessor_segment_overrun
 745         jmp     common_exception
 746 END(coprocessor_segment_overrun)
 747
 748 ENTRY(invalid_TSS)
 749         ASM_CLAC
 750         pushl   $do_invalid_TSS
 751         jmp     common_exception
 752 END(invalid_TSS)
 753
 754 ENTRY(segment_not_present)
 755         ASM_CLAC
 756         pushl   $do_segment_not_present
 757         jmp     common_exception
 758 END(segment_not_present)
 759
 760 ENTRY(stack_segment)
 761         ASM_CLAC
 762         pushl   $do_stack_segment
 763         jmp     common_exception
 764 END(stack_segment)
 765
 766 ENTRY(alignment_check)
 767         ASM_CLAC
 768         pushl   $do_alignment_check
 769         jmp     common_exception
 770 END(alignment_check)
 771
 772 ENTRY(divide_error)
 773         ASM_CLAC
 774         pushl   $0                              # no error code
 775         pushl   $do_divide_error
 776         jmp     common_exception
 777 END(divide_error)
 778
 779 #ifdef CONFIG_X86_MCE
 780 ENTRY(machine_check)
 781         ASM_CLAC
 782         pushl   $0
 783         pushl   machine_check_vector
 784         jmp     common_exception
 785 END(machine_check)
 786 #endif
 787
 788 ENTRY(spurious_interrupt_bug)
 789         ASM_CLAC
 790         pushl   $0
 791         pushl   $do_spurious_interrupt_bug
 792         jmp     common_exception
 793 END(spurious_interrupt_bug)
 794
 795 #ifdef CONFIG_XEN
 796 ENTRY(xen_hypervisor_callback)
 797         pushl   $-1                             /* orig_ax = -1 => not a system call */
 798         SAVE_ALL
 799         ENCODE_FRAME_POINTER
 800         TRACE_IRQS_OFF
 801
 802         /*
 803          * Check to see if we got the event in the critical
 804          * region in xen_iret_direct, after we've reenabled
 805          * events and checked for pending events.  This simulates
 806          * iret instruction's behaviour where it delivers a
 807          * pending interrupt when enabling interrupts:
 808          */
 809         movl    PT_EIP(%esp), %eax
 810         cmpl    $xen_iret_start_crit, %eax
 811         jb      1f
 812         cmpl    $xen_iret_end_crit, %eax
 813         jae     1f
 814
 815         jmp     xen_iret_crit_fixup
 816
 817 ENTRY(xen_do_upcall)
 818 1:      mov     %esp, %eax
 819         call    xen_evtchn_do_upcall
 820 #ifndef CONFIG_PREEMPT
 821         call    xen_maybe_preempt_hcall
 822 #endif
 823         jmp     ret_from_intr
 824 ENDPROC(xen_hypervisor_callback)
 825
 826 /*
 827  * Hypervisor uses this for application faults while it executes.
 828  * We get here for two reasons:
 829  *  1. Fault while reloading DS, ES, FS or GS
 830  *  2. Fault while executing IRET
 831  * Category 1 we fix up by reattempting the load, and zeroing the segment
 832  * register if the load fails.
 833  * Category 2 we fix up by jumping to do_iret_error. We cannot use the
 834  * normal Linux return path in this case because if we use the IRET hypercall
 835  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
 836  * We distinguish between categories by maintaining a status value in EAX.
 837  */
 838 ENTRY(xen_failsafe_callback)
 839         pushl   %eax
 840         movl    $1, %eax
 841 1:      mov     4(%esp), %ds
 842 2:      mov     8(%esp), %es
 843 3:      mov     12(%esp), %fs
 844 4:      mov     16(%esp), %gs
 845         /* EAX == 0 => Category 1 (Bad segment)
 846            EAX != 0 => Category 2 (Bad IRET) */
 847         testl   %eax, %eax
 848         popl    %eax
 849         lea     16(%esp), %esp
 850         jz      5f
 851         jmp     iret_exc
 852 5:      pushl   $-1                             /* orig_ax = -1 => not a system call */
 853         SAVE_ALL
 854         ENCODE_FRAME_POINTER
 855         jmp     ret_from_exception
 856
 857 .section .fixup, "ax"
 858 6:      xorl    %eax, %eax
 859         movl    %eax, 4(%esp)
 860         jmp     1b
 861 7:      xorl    %eax, %eax
 862         movl    %eax, 8(%esp)
 863         jmp     2b
 864 8:      xorl    %eax, %eax
 865         movl    %eax, 12(%esp)
 866         jmp     3b
 867 9:      xorl    %eax, %eax
 868         movl    %eax, 16(%esp)
 869         jmp     4b
 870 .previous
 871         _ASM_EXTABLE(1b, 6b)
 872         _ASM_EXTABLE(2b, 7b)
 873         _ASM_EXTABLE(3b, 8b)
 874         _ASM_EXTABLE(4b, 9b)
 875 ENDPROC(xen_failsafe_callback)
 876
 877 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 878                 xen_evtchn_do_upcall)
 879
 880 #endif /* CONFIG_XEN */
 881
 882 #if IS_ENABLED(CONFIG_HYPERV)
 883
 884 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
 885         hyperv_vector_handler)
 886
 887 #endif /* CONFIG_HYPERV */
 888
 889 #ifdef CONFIG_FUNCTION_TRACER
 890 #ifdef CONFIG_DYNAMIC_FTRACE
 891
 892 ENTRY(mcount)
 893         ret
 894 END(mcount)
 895
 896 ENTRY(ftrace_caller)
 897         pushl   %eax
 898         pushl   %ecx
 899         pushl   %edx
 900         pushl   $0                              /* Pass NULL as regs pointer */
 901         movl    4*4(%esp), %eax
 902         movl    0x4(%ebp), %edx
 903         movl    function_trace_op, %ecx
 904         subl    $MCOUNT_INSN_SIZE, %eax
 905
 906 .globl ftrace_call
 907 ftrace_call:
 908         call    ftrace_stub
 909
 910         addl    $4, %esp                        /* skip NULL pointer */
 911         popl    %edx
 912         popl    %ecx
 913         popl    %eax
 914 .Lftrace_ret:
 915 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 916 .globl ftrace_graph_call
 917 ftrace_graph_call:
 918         jmp     ftrace_stub
 919 #endif
 920
 921 /* This is weak to keep gas from relaxing the jumps */
 922 WEAK(ftrace_stub)
 923         ret
 924 END(ftrace_caller)
 925
 926 ENTRY(ftrace_regs_caller)
 927         pushf   /* push flags before compare (in cs location) */
 928
 929         /*
 930          * i386 does not save SS and ESP when coming from kernel.
 931          * Instead, to get sp, &regs->sp is used (see ptrace.h).
 932          * Unfortunately, that means eflags must be at the same location
 933          * as the current return ip is. We move the return ip into the
 934          * ip location, and move flags into the return ip location.
 935          */
 936         pushl   4(%esp)                         /* save return ip into ip slot */
 937
 938         pushl   $0                              /* Load 0 into orig_ax */
 939         pushl   %gs
 940         pushl   %fs
 941         pushl   %es
 942         pushl   %ds
 943         pushl   %eax
 944         pushl   %ebp
 945         pushl   %edi
 946         pushl   %esi
 947         pushl   %edx
 948         pushl   %ecx
 949         pushl   %ebx
 950
 951         movl    13*4(%esp), %eax                /* Get the saved flags */
 952         movl    %eax, 14*4(%esp)                /* Move saved flags into regs->flags location */
 953                                                 /* clobbering return ip */
 954         movl    $__KERNEL_CS, 13*4(%esp)
 955
 956         movl    12*4(%esp), %eax                /* Load ip (1st parameter) */
 957         subl    $MCOUNT_INSN_SIZE, %eax         /* Adjust ip */
 958         movl    0x4(%ebp), %edx                 /* Load parent ip (2nd parameter) */
 959         movl    function_trace_op, %ecx         /* Save ftrace_pos in 3rd parameter */
 960         pushl   %esp                            /* Save pt_regs as 4th parameter */
 961
 962 GLOBAL(ftrace_regs_call)
 963         call    ftrace_stub
 964
 965         addl    $4, %esp                        /* Skip pt_regs */
 966         movl    14*4(%esp), %eax                /* Move flags back into cs */
 967         movl    %eax, 13*4(%esp)                /* Needed to keep addl  from modifying flags */
 968         movl    12*4(%esp), %eax                /* Get return ip from regs->ip */
 969         movl    %eax, 14*4(%esp)                /* Put return ip back for ret */
 970
 971         popl    %ebx
 972         popl    %ecx
 973         popl    %edx
 974         popl    %esi
 975         popl    %edi
 976         popl    %ebp
 977         popl    %eax
 978         popl    %ds
 979         popl    %es
 980         popl    %fs
 981         popl    %gs
 982         addl    $8, %esp                        /* Skip orig_ax and ip */
 983         popf                                    /* Pop flags at end (no addl to corrupt flags) */
 984         jmp     .Lftrace_ret
 985
 986         popf
 987         jmp     ftrace_stub
 988 #else /* ! CONFIG_DYNAMIC_FTRACE */
 989
 990 ENTRY(mcount)
 991         cmpl    $__PAGE_OFFSET, %esp
 992         jb      ftrace_stub                     /* Paging not enabled yet? */
 993
 994         cmpl    $ftrace_stub, ftrace_trace_function
 995         jnz     .Ltrace
 996 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 997         cmpl    $ftrace_stub, ftrace_graph_return
 998         jnz     ftrace_graph_caller
 999
1000         cmpl    $ftrace_graph_entry_stub, ftrace_graph_entry
1001         jnz     ftrace_graph_caller
1002 #endif
1003 .globl ftrace_stub
1004 ftrace_stub:
1005         ret
1006
1007         /* taken from glibc */
1008 .Ltrace:
1009         pushl   %eax
1010         pushl   %ecx
1011         pushl   %edx
1012         movl    0xc(%esp), %eax
1013         movl    0x4(%ebp), %edx
1014         subl    $MCOUNT_INSN_SIZE, %eax
1015
1016         call    *ftrace_trace_function
1017
1018         popl    %edx
1019         popl    %ecx
1020         popl    %eax
1021         jmp     ftrace_stub
1022 END(mcount)
1023 #endif /* CONFIG_DYNAMIC_FTRACE */
1024 EXPORT_SYMBOL(mcount)
1025 #endif /* CONFIG_FUNCTION_TRACER */
1026
1027 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1028 ENTRY(ftrace_graph_caller)
1029         pushl   %eax
1030         pushl   %ecx
1031         pushl   %edx
1032         movl    0xc(%esp), %eax
1033         lea     0x4(%ebp), %edx
1034         movl    (%ebp), %ecx
1035         subl    $MCOUNT_INSN_SIZE, %eax
1036         call    prepare_ftrace_return
1037         popl    %edx
1038         popl    %ecx
1039         popl    %eax
1040         ret
1041 END(ftrace_graph_caller)
1042
1043 .globl return_to_handler
1044 return_to_handler:
1045         pushl   %eax
1046         pushl   %edx
1047         movl    %ebp, %eax
1048         call    ftrace_return_to_handler
1049         movl    %eax, %ecx
1050         popl    %edx
1051         popl    %eax
1052         jmp     *%ecx
1053 #endif
1054
1055 #ifdef CONFIG_TRACING
1056 ENTRY(trace_page_fault)
1057         ASM_CLAC
1058         pushl   $trace_do_page_fault
1059         jmp     common_exception
1060 END(trace_page_fault)
1061 #endif
1062
1063 ENTRY(page_fault)
1064         ASM_CLAC
1065         pushl   $do_page_fault
1066         ALIGN
1067         jmp common_exception
1068 END(page_fault)
1069
1070 common_exception:
1071         /* the function address is in %gs's slot on the stack */
1072         pushl   %fs
1073         pushl   %es
1074         pushl   %ds
1075         pushl   %eax
1076         pushl   %ebp
1077         pushl   %edi
1078         pushl   %esi
1079         pushl   %edx
1080         pushl   %ecx
1081         pushl   %ebx
1082         ENCODE_FRAME_POINTER
1083         cld
1084         movl    $(__KERNEL_PERCPU), %ecx
1085         movl    %ecx, %fs
1086         UNWIND_ESPFIX_STACK
1087         GS_TO_REG %ecx
1088         movl    PT_GS(%esp), %edi               # get the function address
1089         movl    PT_ORIG_EAX(%esp), %edx         # get the error code
1090         movl    $-1, PT_ORIG_EAX(%esp)          # no syscall to restart
1091         REG_TO_PTGS %ecx
1092         SET_KERNEL_GS %ecx
1093         movl    $(__USER_DS), %ecx
1094         movl    %ecx, %ds
1095         movl    %ecx, %es
1096         TRACE_IRQS_OFF
1097         movl    %esp, %eax                      # pt_regs pointer
1098         call    *%edi
1099         jmp     ret_from_exception
1100 END(common_exception)
1101
1102 ENTRY(debug)
1103         /*
1104          * #DB can happen at the first instruction of
1105          * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
1106          * happens, then we will be running on a very small stack.  We
1107          * need to detect this condition and switch to the thread
1108          * stack before calling any C code at all.
1109          *
1110          * If you edit this code, keep in mind that NMIs can happen in here.
1111          */
1112         ASM_CLAC
1113         pushl   $-1                             # mark this as an int
1114         SAVE_ALL
1115         ENCODE_FRAME_POINTER
1116         xorl    %edx, %edx                      # error code 0
1117         movl    %esp, %eax                      # pt_regs pointer
1118
1119         /* Are we currently on the SYSENTER stack? */
1120         PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1121         subl    %eax, %ecx      /* ecx = (end of SYSENTER_stack) - esp */
1122         cmpl    $SIZEOF_SYSENTER_stack, %ecx
1123         jb      .Ldebug_from_sysenter_stack
1124
1125         TRACE_IRQS_OFF
1126         call    do_debug
1127         jmp     ret_from_exception
1128
1129 .Ldebug_from_sysenter_stack:
1130         /* We're on the SYSENTER stack.  Switch off. */
1131         movl    %esp, %ebx
1132         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esp
1133         TRACE_IRQS_OFF
1134         call    do_debug
1135         movl    %ebx, %esp
1136         jmp     ret_from_exception
1137 END(debug)
1138
1139 /*
1140  * NMI is doubly nasty.  It can happen on the first instruction of
1141  * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1142  * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1143  * switched stacks.  We handle both conditions by simply checking whether we
1144  * interrupted kernel code running on the SYSENTER stack.
1145  */
1146 ENTRY(nmi)
1147         ASM_CLAC
1148 #ifdef CONFIG_X86_ESPFIX32
1149         pushl   %eax
1150         movl    %ss, %eax
1151         cmpw    $__ESPFIX_SS, %ax
1152         popl    %eax
1153         je      .Lnmi_espfix_stack
1154 #endif
1155
1156         pushl   %eax                            # pt_regs->orig_ax
1157         SAVE_ALL
1158         ENCODE_FRAME_POINTER
1159         xorl    %edx, %edx                      # zero error code
1160         movl    %esp, %eax                      # pt_regs pointer
1161
1162         /* Are we currently on the SYSENTER stack? */
1163         PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1164         subl    %eax, %ecx      /* ecx = (end of SYSENTER_stack) - esp */
1165         cmpl    $SIZEOF_SYSENTER_stack, %ecx
1166         jb      .Lnmi_from_sysenter_stack
1167
1168         /* Not on SYSENTER stack. */
1169         call    do_nmi
1170         jmp     .Lrestore_all_notrace
1171
1172 .Lnmi_from_sysenter_stack:
1173         /*
1174          * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1175          * is using the thread stack right now, so it's safe for us to use it.
1176          */
1177         movl    %esp, %ebx
1178         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esp
1179         call    do_nmi
1180         movl    %ebx, %esp
1181         jmp     .Lrestore_all_notrace
1182
1183 #ifdef CONFIG_X86_ESPFIX32
1184 .Lnmi_espfix_stack:
1185         /*
1186          * create the pointer to lss back
1187          */
1188         pushl   %ss
1189         pushl   %esp
1190         addl    $4, (%esp)
1191         /* copy the iret frame of 12 bytes */
1192         .rept 3
1193         pushl   16(%esp)
1194         .endr
1195         pushl   %eax
1196         SAVE_ALL
1197         ENCODE_FRAME_POINTER
1198         FIXUP_ESPFIX_STACK                      # %eax == %esp
1199         xorl    %edx, %edx                      # zero error code
1200         call    do_nmi
1201         RESTORE_REGS
1202         lss     12+4(%esp), %esp                # back to espfix stack
1203         jmp     .Lirq_return
1204 #endif
1205 END(nmi)
1206
1207 ENTRY(int3)
1208         ASM_CLAC
1209         pushl   $-1                             # mark this as an int
1210         SAVE_ALL
1211         ENCODE_FRAME_POINTER
1212         TRACE_IRQS_OFF
1213         xorl    %edx, %edx                      # zero error code
1214         movl    %esp, %eax                      # pt_regs pointer
1215         call    do_int3
1216         jmp     ret_from_exception
1217 END(int3)
1218
1219 ENTRY(general_protection)
1220         pushl   $do_general_protection
1221         jmp     common_exception
1222 END(general_protection)
1223
1224 #ifdef CONFIG_KVM_GUEST
1225 ENTRY(async_page_fault)
1226         ASM_CLAC
1227         pushl   $do_async_page_fault
1228         jmp     common_exception
1229 END(async_page_fault)
1230 #endif
1231
1232 ENTRY(rewind_stack_do_exit)
1233         /* Prevent any naive code from trying to unwind to our caller. */
1234         xorl    %ebp, %ebp
1235
1236         movl    PER_CPU_VAR(cpu_current_top_of_stack), %esi
1237         leal    -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1238
1239         call    do_exit
1240 1:      jmp 1b
1241 END(rewind_stack_do_exit)