arch/x86/kernel/entry_64.S

   1 /*
   2  *  linux/arch/x86_64/entry.S
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2000, 2001, 2002  Andi Kleen SuSE Labs
   6  *  Copyright (C) 2000  Pavel Machek <pavel@suse.cz>
   7  */
   8
   9 /*
  10  * entry.S contains the system-call and fault low-level handling routines.
  11  *
  12  * NOTE: This code handles signal-recognition, which happens every time
  13  * after an interrupt and after each system call.
  14  *
  15  * Normal syscalls and interrupts don't save a full stack frame, this is
  16  * only done for syscall tracing, signals or fork/exec et.al.
  17  *
  18  * A note on terminology:
  19  * - top of stack: Architecture defined interrupt frame from SS to RIP
  20  * at the top of the kernel process stack.
  21  * - partial stack frame: partially saved registers upto R11.
  22  * - full stack frame: Like partial stack frame, but all register saved.
  23  *
  24  * Some macro usage:
  25  * - CFI macros are used to generate dwarf2 unwind information for better
  26  * backtraces. They don't change any code.
  27  * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
  28  * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
  29  * There are unfortunately lots of special cases where some registers
  30  * not touched. The macro is a big mess that should be cleaned up.
  31  * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
  32  * Gives a full stack frame.
  33  * - ENTRY/END Define functions in the symbol table.
  34  * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
  35  * frame that is otherwise undefined after a SYSCALL
  36  * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
  37  * - errorentry/paranoidentry/zeroentry - Define exception entry points.
  38  */
  39
  40 #include <linux/linkage.h>
  41 #include <asm/segment.h>
  42 #include <asm/cache.h>
  43 #include <asm/errno.h>
  44 #include <asm/dwarf2.h>
  45 #include <asm/calling.h>
  46 #include <asm/asm-offsets.h>
  47 #include <asm/msr.h>
  48 #include <asm/unistd.h>
  49 #include <asm/thread_info.h>
  50 #include <asm/hw_irq.h>
  51 #include <asm/page.h>
  52 #include <asm/irqflags.h>
  53 #include <asm/paravirt.h>
  54
  55         .code64
  56
  57 #ifndef CONFIG_PREEMPT
  58 #define retint_kernel retint_restore_args
  59 #endif
  60
  61 #ifdef CONFIG_PARAVIRT
  62 ENTRY(native_irq_enable_syscall_ret)
  63         movq    %gs:pda_oldrsp,%rsp
  64         swapgs
  65         sysretq
  66 #endif /* CONFIG_PARAVIRT */
  67
  68
  69 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
  70 #ifdef CONFIG_TRACE_IRQFLAGS
  71         bt   $9,EFLAGS-\offset(%rsp)    /* interrupts off? */
  72         jnc  1f
  73         TRACE_IRQS_ON
  74 1:
  75 #endif
  76 .endm
  77
  78 /*
  79  * C code is not supposed to know about undefined top of stack. Every time
  80  * a C function with an pt_regs argument is called from the SYSCALL based
  81  * fast path FIXUP_TOP_OF_STACK is needed.
  82  * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
  83  * manipulation.
  84  */
  85
  86         /* %rsp:at FRAMEEND */
  87         .macro FIXUP_TOP_OF_STACK tmp
  88         movq    %gs:pda_oldrsp,\tmp
  89         movq    \tmp,RSP(%rsp)
  90         movq    $__USER_DS,SS(%rsp)
  91         movq    $__USER_CS,CS(%rsp)
  92         movq    $-1,RCX(%rsp)
  93         movq    R11(%rsp),\tmp  /* get eflags */
  94         movq    \tmp,EFLAGS(%rsp)
  95         .endm
  96
  97         .macro RESTORE_TOP_OF_STACK tmp,offset=0
  98         movq   RSP-\offset(%rsp),\tmp
  99         movq   \tmp,%gs:pda_oldrsp
 100         movq   EFLAGS-\offset(%rsp),\tmp
 101         movq   \tmp,R11-\offset(%rsp)
 102         .endm
 103
 104         .macro FAKE_STACK_FRAME child_rip
 105         /* push in order ss, rsp, eflags, cs, rip */
 106         xorl %eax, %eax
 107         pushq %rax /* ss */
 108         CFI_ADJUST_CFA_OFFSET   8
 109         /*CFI_REL_OFFSET        ss,0*/
 110         pushq %rax /* rsp */
 111         CFI_ADJUST_CFA_OFFSET   8
 112         CFI_REL_OFFSET  rsp,0
 113         pushq $(1<<9) /* eflags - interrupts on */
 114         CFI_ADJUST_CFA_OFFSET   8
 115         /*CFI_REL_OFFSET        rflags,0*/
 116         pushq $__KERNEL_CS /* cs */
 117         CFI_ADJUST_CFA_OFFSET   8
 118         /*CFI_REL_OFFSET        cs,0*/
 119         pushq \child_rip /* rip */
 120         CFI_ADJUST_CFA_OFFSET   8
 121         CFI_REL_OFFSET  rip,0
 122         pushq   %rax /* orig rax */
 123         CFI_ADJUST_CFA_OFFSET   8
 124         .endm
 125
 126         .macro UNFAKE_STACK_FRAME
 127         addq $8*6, %rsp
 128         CFI_ADJUST_CFA_OFFSET   -(6*8)
 129         .endm
 130
 131         .macro  CFI_DEFAULT_STACK start=1
 132         .if \start
 133         CFI_STARTPROC   simple
 134         CFI_SIGNAL_FRAME
 135         CFI_DEF_CFA     rsp,SS+8
 136         .else
 137         CFI_DEF_CFA_OFFSET SS+8
 138         .endif
 139         CFI_REL_OFFSET  r15,R15
 140         CFI_REL_OFFSET  r14,R14
 141         CFI_REL_OFFSET  r13,R13
 142         CFI_REL_OFFSET  r12,R12
 143         CFI_REL_OFFSET  rbp,RBP
 144         CFI_REL_OFFSET  rbx,RBX
 145         CFI_REL_OFFSET  r11,R11
 146         CFI_REL_OFFSET  r10,R10
 147         CFI_REL_OFFSET  r9,R9
 148         CFI_REL_OFFSET  r8,R8
 149         CFI_REL_OFFSET  rax,RAX
 150         CFI_REL_OFFSET  rcx,RCX
 151         CFI_REL_OFFSET  rdx,RDX
 152         CFI_REL_OFFSET  rsi,RSI
 153         CFI_REL_OFFSET  rdi,RDI
 154         CFI_REL_OFFSET  rip,RIP
 155         /*CFI_REL_OFFSET        cs,CS*/
 156         /*CFI_REL_OFFSET        rflags,EFLAGS*/
 157         CFI_REL_OFFSET  rsp,RSP
 158         /*CFI_REL_OFFSET        ss,SS*/
 159         .endm
 160 /*
 161  * A newly forked process directly context switches into this.
 162  */
 163 /* rdi: prev */
 164 ENTRY(ret_from_fork)
 165         CFI_DEFAULT_STACK
 166         push kernel_eflags(%rip)
 167         CFI_ADJUST_CFA_OFFSET 4
 168         popf                            # reset kernel eflags
 169         CFI_ADJUST_CFA_OFFSET -4
 170         call schedule_tail
 171         GET_THREAD_INFO(%rcx)
 172         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),threadinfo_flags(%rcx)
 173         jnz rff_trace
 174 rff_action:
 175         RESTORE_REST
 176         testl $3,CS-ARGOFFSET(%rsp)     # from kernel_thread?
 177         je   int_ret_from_sys_call
 178         testl $_TIF_IA32,threadinfo_flags(%rcx)
 179         jnz  int_ret_from_sys_call
 180         RESTORE_TOP_OF_STACK %rdi,ARGOFFSET
 181         jmp ret_from_sys_call
 182 rff_trace:
 183         movq %rsp,%rdi
 184         call syscall_trace_leave
 185         GET_THREAD_INFO(%rcx)
 186         jmp rff_action
 187         CFI_ENDPROC
 188 END(ret_from_fork)
 189
 190 /*
 191  * System call entry. Upto 6 arguments in registers are supported.
 192  *
 193  * SYSCALL does not save anything on the stack and does not change the
 194  * stack pointer.
 195  */
 196
 197 /*
 198  * Register setup:
 199  * rax  system call number
 200  * rdi  arg0
 201  * rcx  return address for syscall/sysret, C arg3
 202  * rsi  arg1
 203  * rdx  arg2
 204  * r10  arg3    (--> moved to rcx for C)
 205  * r8   arg4
 206  * r9   arg5
 207  * r11  eflags for syscall/sysret, temporary for C
 208  * r12-r15,rbp,rbx saved by C code, not touched.
 209  *
 210  * Interrupts are off on entry.
 211  * Only called from user space.
 212  *
 213  * XXX  if we had a free scratch register we could save the RSP into the stack frame
 214  *      and report it properly in ps. Unfortunately we haven't.
 215  *
 216  * When user can change the frames always force IRET. That is because
 217  * it deals with uncanonical addresses better. SYSRET has trouble
 218  * with them due to bugs in both AMD and Intel CPUs.
 219  */
 220
 221 ENTRY(system_call)
 222         CFI_STARTPROC   simple
 223         CFI_SIGNAL_FRAME
 224         CFI_DEF_CFA     rsp,PDA_STACKOFFSET
 225         CFI_REGISTER    rip,rcx
 226         /*CFI_REGISTER  rflags,r11*/
 227         SWAPGS_UNSAFE_STACK
 228         /*
 229          * A hypervisor implementation might want to use a label
 230          * after the swapgs, so that it can do the swapgs
 231          * for the guest and jump here on syscall.
 232          */
 233 ENTRY(system_call_after_swapgs)
 234
 235         movq    %rsp,%gs:pda_oldrsp
 236         movq    %gs:pda_kernelstack,%rsp
 237         /*
 238          * No need to follow this irqs off/on section - it's straight
 239          * and short:
 240          */
 241         ENABLE_INTERRUPTS(CLBR_NONE)
 242         SAVE_ARGS 8,1
 243         movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
 244         movq  %rcx,RIP-ARGOFFSET(%rsp)
 245         CFI_REL_OFFSET rip,RIP-ARGOFFSET
 246         GET_THREAD_INFO(%rcx)
 247         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP),threadinfo_flags(%rcx)
 248         jnz tracesys
 249         cmpq $__NR_syscall_max,%rax
 250         ja badsys
 251         movq %r10,%rcx
 252         call *sys_call_table(,%rax,8)  # XXX:    rip relative
 253         movq %rax,RAX-ARGOFFSET(%rsp)
 254 /*
 255  * Syscall return path ending with SYSRET (fast path)
 256  * Has incomplete stack frame and undefined top of stack.
 257  */
 258 ret_from_sys_call:
 259         movl $_TIF_ALLWORK_MASK,%edi
 260         /* edi: flagmask */
 261 sysret_check:
 262         LOCKDEP_SYS_EXIT
 263         GET_THREAD_INFO(%rcx)
 264         DISABLE_INTERRUPTS(CLBR_NONE)
 265         TRACE_IRQS_OFF
 266         movl threadinfo_flags(%rcx),%edx
 267         andl %edi,%edx
 268         jnz  sysret_careful
 269         CFI_REMEMBER_STATE
 270         /*
 271          * sysretq will re-enable interrupts:
 272          */
 273         TRACE_IRQS_ON
 274         movq RIP-ARGOFFSET(%rsp),%rcx
 275         CFI_REGISTER    rip,rcx
 276         RESTORE_ARGS 0,-ARG_SKIP,1
 277         /*CFI_REGISTER  rflags,r11*/
 278         ENABLE_INTERRUPTS_SYSCALL_RET
 279
 280         CFI_RESTORE_STATE
 281         /* Handle reschedules */
 282         /* edx: work, edi: workmask */
 283 sysret_careful:
 284         bt $TIF_NEED_RESCHED,%edx
 285         jnc sysret_signal
 286         TRACE_IRQS_ON
 287         ENABLE_INTERRUPTS(CLBR_NONE)
 288         pushq %rdi
 289         CFI_ADJUST_CFA_OFFSET 8
 290         call schedule
 291         popq  %rdi
 292         CFI_ADJUST_CFA_OFFSET -8
 293         jmp sysret_check
 294
 295         /* Handle a signal */
 296 sysret_signal:
 297         TRACE_IRQS_ON
 298         ENABLE_INTERRUPTS(CLBR_NONE)
 299         testl $_TIF_DO_NOTIFY_MASK,%edx
 300         jz    1f
 301
 302         /* Really a signal */
 303         /* edx: work flags (arg3) */
 304         leaq do_notify_resume(%rip),%rax
 305         leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
 306         xorl %esi,%esi # oldset -> arg2
 307         call ptregscall_common
 308 1:      movl $_TIF_NEED_RESCHED,%edi
 309         /* Use IRET because user could have changed frame. This
 310            works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
 311         DISABLE_INTERRUPTS(CLBR_NONE)
 312         TRACE_IRQS_OFF
 313         jmp int_with_check
 314
 315 badsys:
 316         movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
 317         jmp ret_from_sys_call
 318
 319         /* Do syscall tracing */
 320 tracesys:
 321         SAVE_REST
 322         movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
 323         FIXUP_TOP_OF_STACK %rdi
 324         movq %rsp,%rdi
 325         call syscall_trace_enter
 326         LOAD_ARGS ARGOFFSET  /* reload args from stack in case ptrace changed it */
 327         RESTORE_REST
 328         cmpq $__NR_syscall_max,%rax
 329         ja   int_ret_from_sys_call      /* RAX(%rsp) set to -ENOSYS above */
 330         movq %r10,%rcx  /* fixup for C */
 331         call *sys_call_table(,%rax,8)
 332         movq %rax,RAX-ARGOFFSET(%rsp)
 333         /* Use IRET because user could have changed frame */
 334
 335 /*
 336  * Syscall return path ending with IRET.
 337  * Has correct top of stack, but partial stack frame.
 338  */
 339         .globl int_ret_from_sys_call
 340 int_ret_from_sys_call:
 341         DISABLE_INTERRUPTS(CLBR_NONE)
 342         TRACE_IRQS_OFF
 343         testl $3,CS-ARGOFFSET(%rsp)
 344         je retint_restore_args
 345         movl $_TIF_ALLWORK_MASK,%edi
 346         /* edi: mask to check */
 347 int_with_check:
 348         LOCKDEP_SYS_EXIT_IRQ
 349         GET_THREAD_INFO(%rcx)
 350         movl threadinfo_flags(%rcx),%edx
 351         andl %edi,%edx
 352         jnz   int_careful
 353         andl    $~TS_COMPAT,threadinfo_status(%rcx)
 354         jmp   retint_swapgs
 355
 356         /* Either reschedule or signal or syscall exit tracking needed. */
 357         /* First do a reschedule test. */
 358         /* edx: work, edi: workmask */
 359 int_careful:
 360         bt $TIF_NEED_RESCHED,%edx
 361         jnc  int_very_careful
 362         TRACE_IRQS_ON
 363         ENABLE_INTERRUPTS(CLBR_NONE)
 364         pushq %rdi
 365         CFI_ADJUST_CFA_OFFSET 8
 366         call schedule
 367         popq %rdi
 368         CFI_ADJUST_CFA_OFFSET -8
 369         DISABLE_INTERRUPTS(CLBR_NONE)
 370         TRACE_IRQS_OFF
 371         jmp int_with_check
 372
 373         /* handle signals and tracing -- both require a full stack frame */
 374 int_very_careful:
 375         TRACE_IRQS_ON
 376         ENABLE_INTERRUPTS(CLBR_NONE)
 377         SAVE_REST
 378         /* Check for syscall exit trace */
 379         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx
 380         jz int_signal
 381         pushq %rdi
 382         CFI_ADJUST_CFA_OFFSET 8
 383         leaq 8(%rsp),%rdi       # &ptregs -> arg1
 384         call syscall_trace_leave
 385         popq %rdi
 386         CFI_ADJUST_CFA_OFFSET -8
 387         andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi
 388         jmp int_restore_rest
 389
 390 int_signal:
 391         testl $_TIF_DO_NOTIFY_MASK,%edx
 392         jz 1f
 393         movq %rsp,%rdi          # &ptregs -> arg1
 394         xorl %esi,%esi          # oldset -> arg2
 395         call do_notify_resume
 396 1:      movl $_TIF_NEED_RESCHED,%edi
 397 int_restore_rest:
 398         RESTORE_REST
 399         DISABLE_INTERRUPTS(CLBR_NONE)
 400         TRACE_IRQS_OFF
 401         jmp int_with_check
 402         CFI_ENDPROC
 403 END(system_call)
 404
 405 /*
 406  * Certain special system calls that need to save a complete full stack frame.
 407  */
 408
 409         .macro PTREGSCALL label,func,arg
 410         .globl \label
 411 \label:
 412         leaq    \func(%rip),%rax
 413         leaq    -ARGOFFSET+8(%rsp),\arg /* 8 for return address */
 414         jmp     ptregscall_common
 415 END(\label)
 416         .endm
 417
 418         CFI_STARTPROC
 419
 420         PTREGSCALL stub_clone, sys_clone, %r8
 421         PTREGSCALL stub_fork, sys_fork, %rdi
 422         PTREGSCALL stub_vfork, sys_vfork, %rdi
 423         PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx
 424         PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 425         PTREGSCALL stub_iopl, sys_iopl, %rsi
 426
 427 ENTRY(ptregscall_common)
 428         popq %r11
 429         CFI_ADJUST_CFA_OFFSET -8
 430         CFI_REGISTER rip, r11
 431         SAVE_REST
 432         movq %r11, %r15
 433         CFI_REGISTER rip, r15
 434         FIXUP_TOP_OF_STACK %r11
 435         call *%rax
 436         RESTORE_TOP_OF_STACK %r11
 437         movq %r15, %r11
 438         CFI_REGISTER rip, r11
 439         RESTORE_REST
 440         pushq %r11
 441         CFI_ADJUST_CFA_OFFSET 8
 442         CFI_REL_OFFSET rip, 0
 443         ret
 444         CFI_ENDPROC
 445 END(ptregscall_common)
 446
 447 ENTRY(stub_execve)
 448         CFI_STARTPROC
 449         popq %r11
 450         CFI_ADJUST_CFA_OFFSET -8
 451         CFI_REGISTER rip, r11
 452         SAVE_REST
 453         FIXUP_TOP_OF_STACK %r11
 454         movq %rsp, %rcx
 455         call sys_execve
 456         RESTORE_TOP_OF_STACK %r11
 457         movq %rax,RAX(%rsp)
 458         RESTORE_REST
 459         jmp int_ret_from_sys_call
 460         CFI_ENDPROC
 461 END(stub_execve)
 462
 463 /*
 464  * sigreturn is special because it needs to restore all registers on return.
 465  * This cannot be done with SYSRET, so use the IRET return path instead.
 466  */
 467 ENTRY(stub_rt_sigreturn)
 468         CFI_STARTPROC
 469         addq $8, %rsp
 470         CFI_ADJUST_CFA_OFFSET   -8
 471         SAVE_REST
 472         movq %rsp,%rdi
 473         FIXUP_TOP_OF_STACK %r11
 474         call sys_rt_sigreturn
 475         movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 476         RESTORE_REST
 477         jmp int_ret_from_sys_call
 478         CFI_ENDPROC
 479 END(stub_rt_sigreturn)
 480
 481 /*
 482  * initial frame state for interrupts and exceptions
 483  */
 484         .macro _frame ref
 485         CFI_STARTPROC simple
 486         CFI_SIGNAL_FRAME
 487         CFI_DEF_CFA rsp,SS+8-\ref
 488         /*CFI_REL_OFFSET ss,SS-\ref*/
 489         CFI_REL_OFFSET rsp,RSP-\ref
 490         /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/
 491         /*CFI_REL_OFFSET cs,CS-\ref*/
 492         CFI_REL_OFFSET rip,RIP-\ref
 493         .endm
 494
 495 /* initial frame state for interrupts (and exceptions without error code) */
 496 #define INTR_FRAME _frame RIP
 497 /* initial frame state for exceptions with error code (and interrupts with
 498    vector already pushed) */
 499 #define XCPT_FRAME _frame ORIG_RAX
 500
 501 /*
 502  * Interrupt entry/exit.
 503  *
 504  * Interrupt entry points save only callee clobbered registers in fast path.
 505  *
 506  * Entry runs with interrupts off.
 507  */
 508
 509 /* 0(%rsp): interrupt number */
 510         .macro interrupt func
 511         cld
 512         SAVE_ARGS
 513         leaq -ARGOFFSET(%rsp),%rdi      # arg1 for handler
 514         pushq %rbp
 515         CFI_ADJUST_CFA_OFFSET   8
 516         CFI_REL_OFFSET          rbp, 0
 517         movq %rsp,%rbp
 518         CFI_DEF_CFA_REGISTER    rbp
 519         testl $3,CS(%rdi)
 520         je 1f
 521         SWAPGS
 522         /* irqcount is used to check if a CPU is already on an interrupt
 523            stack or not. While this is essentially redundant with preempt_count
 524            it is a little cheaper to use a separate counter in the PDA
 525            (short of moving irq_enter into assembly, which would be too
 526             much work) */
 527 1:      incl    %gs:pda_irqcount
 528         cmoveq %gs:pda_irqstackptr,%rsp
 529         push    %rbp                    # backlink for old unwinder
 530         /*
 531          * We entered an interrupt context - irqs are off:
 532          */
 533         TRACE_IRQS_OFF
 534         call \func
 535         .endm
 536
 537 ENTRY(common_interrupt)
 538         XCPT_FRAME
 539         interrupt do_IRQ
 540         /* 0(%rsp): oldrsp-ARGOFFSET */
 541 ret_from_intr:
 542         DISABLE_INTERRUPTS(CLBR_NONE)
 543         TRACE_IRQS_OFF
 544         decl %gs:pda_irqcount
 545         leaveq
 546         CFI_DEF_CFA_REGISTER    rsp
 547         CFI_ADJUST_CFA_OFFSET   -8
 548 exit_intr:
 549         GET_THREAD_INFO(%rcx)
 550         testl $3,CS-ARGOFFSET(%rsp)
 551         je retint_kernel
 552
 553         /* Interrupt came from user space */
 554         /*
 555          * Has a correct top of stack, but a partial stack frame
 556          * %rcx: thread info. Interrupts off.
 557          */
 558 retint_with_reschedule:
 559         movl $_TIF_WORK_MASK,%edi
 560 retint_check:
 561         LOCKDEP_SYS_EXIT_IRQ
 562         movl threadinfo_flags(%rcx),%edx
 563         andl %edi,%edx
 564         CFI_REMEMBER_STATE
 565         jnz  retint_careful
 566
 567 retint_swapgs:          /* return to user-space */
 568         /*
 569          * The iretq could re-enable interrupts:
 570          */
 571         DISABLE_INTERRUPTS(CLBR_ANY)
 572         TRACE_IRQS_IRETQ
 573         SWAPGS
 574         jmp restore_args
 575
 576 retint_restore_args:    /* return to kernel space */
 577         DISABLE_INTERRUPTS(CLBR_ANY)
 578         /*
 579          * The iretq could re-enable interrupts:
 580          */
 581         TRACE_IRQS_IRETQ
 582 restore_args:
 583         RESTORE_ARGS 0,8,0
 584
 585 irq_return:
 586         INTERRUPT_RETURN
 587
 588         .section __ex_table, "a"
 589         .quad irq_return, bad_iret
 590         .previous
 591
 592 #ifdef CONFIG_PARAVIRT
 593 ENTRY(native_iret)
 594         iretq
 595
 596         .section __ex_table,"a"
 597         .quad native_iret, bad_iret
 598         .previous
 599 #endif
 600
 601         .section .fixup,"ax"
 602 bad_iret:
 603         /*
 604          * The iret traps when the %cs or %ss being restored is bogus.
 605          * We've lost the original trap vector and error code.
 606          * #GPF is the most likely one to get for an invalid selector.
 607          * So pretend we completed the iret and took the #GPF in user mode.
 608          *
 609          * We are now running with the kernel GS after exception recovery.
 610          * But error_entry expects us to have user GS to match the user %cs,
 611          * so swap back.
 612          */
 613         pushq $0
 614
 615         SWAPGS
 616         jmp general_protection
 617
 618         .previous
 619
 620         /* edi: workmask, edx: work */
 621 retint_careful:
 622         CFI_RESTORE_STATE
 623         bt    $TIF_NEED_RESCHED,%edx
 624         jnc   retint_signal
 625         TRACE_IRQS_ON
 626         ENABLE_INTERRUPTS(CLBR_NONE)
 627         pushq %rdi
 628         CFI_ADJUST_CFA_OFFSET   8
 629         call  schedule
 630         popq %rdi
 631         CFI_ADJUST_CFA_OFFSET   -8
 632         GET_THREAD_INFO(%rcx)
 633         DISABLE_INTERRUPTS(CLBR_NONE)
 634         TRACE_IRQS_OFF
 635         jmp retint_check
 636
 637 retint_signal:
 638         testl $_TIF_DO_NOTIFY_MASK,%edx
 639         jz    retint_swapgs
 640         TRACE_IRQS_ON
 641         ENABLE_INTERRUPTS(CLBR_NONE)
 642         SAVE_REST
 643         movq $-1,ORIG_RAX(%rsp)
 644         xorl %esi,%esi          # oldset
 645         movq %rsp,%rdi          # &pt_regs
 646         call do_notify_resume
 647         RESTORE_REST
 648         DISABLE_INTERRUPTS(CLBR_NONE)
 649         TRACE_IRQS_OFF
 650         movl $_TIF_NEED_RESCHED,%edi
 651         GET_THREAD_INFO(%rcx)
 652         jmp retint_check
 653
 654 #ifdef CONFIG_PREEMPT
 655         /* Returning to kernel space. Check if we need preemption */
 656         /* rcx:  threadinfo. interrupts off. */
 657 ENTRY(retint_kernel)
 658         cmpl $0,threadinfo_preempt_count(%rcx)
 659         jnz  retint_restore_args
 660         bt  $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
 661         jnc  retint_restore_args
 662         bt   $9,EFLAGS-ARGOFFSET(%rsp)  /* interrupts off? */
 663         jnc  retint_restore_args
 664         call preempt_schedule_irq
 665         jmp exit_intr
 666 #endif
 667
 668         CFI_ENDPROC
 669 END(common_interrupt)
 670
 671 /*
 672  * APIC interrupts.
 673  */
 674         .macro apicinterrupt num,func
 675         INTR_FRAME
 676         pushq $~(\num)
 677         CFI_ADJUST_CFA_OFFSET 8
 678         interrupt \func
 679         jmp ret_from_intr
 680         CFI_ENDPROC
 681         .endm
 682
 683 ENTRY(thermal_interrupt)
 684         apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
 685 END(thermal_interrupt)
 686
 687 ENTRY(threshold_interrupt)
 688         apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
 689 END(threshold_interrupt)
 690
 691 #ifdef CONFIG_SMP
 692 ENTRY(reschedule_interrupt)
 693         apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
 694 END(reschedule_interrupt)
 695
 696         .macro INVALIDATE_ENTRY num
 697 ENTRY(invalidate_interrupt\num)
 698         apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
 699 END(invalidate_interrupt\num)
 700         .endm
 701
 702         INVALIDATE_ENTRY 0
 703         INVALIDATE_ENTRY 1
 704         INVALIDATE_ENTRY 2
 705         INVALIDATE_ENTRY 3
 706         INVALIDATE_ENTRY 4
 707         INVALIDATE_ENTRY 5
 708         INVALIDATE_ENTRY 6
 709         INVALIDATE_ENTRY 7
 710
 711 ENTRY(call_function_interrupt)
 712         apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
 713 END(call_function_interrupt)
 714 ENTRY(irq_move_cleanup_interrupt)
 715         apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
 716 END(irq_move_cleanup_interrupt)
 717 #endif
 718
 719 ENTRY(apic_timer_interrupt)
 720         apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
 721 END(apic_timer_interrupt)
 722
 723 ENTRY(error_interrupt)
 724         apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
 725 END(error_interrupt)
 726
 727 ENTRY(spurious_interrupt)
 728         apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
 729 END(spurious_interrupt)
 730
 731 /*
 732  * Exception entry points.
 733  */
 734         .macro zeroentry sym
 735         INTR_FRAME
 736         pushq $0        /* push error code/oldrax */
 737         CFI_ADJUST_CFA_OFFSET 8
 738         pushq %rax      /* push real oldrax to the rdi slot */
 739         CFI_ADJUST_CFA_OFFSET 8
 740         CFI_REL_OFFSET rax,0
 741         leaq  \sym(%rip),%rax
 742         jmp error_entry
 743         CFI_ENDPROC
 744         .endm
 745
 746         .macro errorentry sym
 747         XCPT_FRAME
 748         pushq %rax
 749         CFI_ADJUST_CFA_OFFSET 8
 750         CFI_REL_OFFSET rax,0
 751         leaq  \sym(%rip),%rax
 752         jmp error_entry
 753         CFI_ENDPROC
 754         .endm
 755
 756         /* error code is on the stack already */
 757         /* handle NMI like exceptions that can happen everywhere */
 758         .macro paranoidentry sym, ist=0, irqtrace=1
 759         SAVE_ALL
 760         cld
 761         movl $1,%ebx
 762         movl  $MSR_GS_BASE,%ecx
 763         rdmsr
 764         testl %edx,%edx
 765         js    1f
 766         SWAPGS
 767         xorl  %ebx,%ebx
 768 1:
 769         .if \ist
 770         movq    %gs:pda_data_offset, %rbp
 771         .endif
 772         movq %rsp,%rdi
 773         movq ORIG_RAX(%rsp),%rsi
 774         movq $-1,ORIG_RAX(%rsp)
 775         .if \ist
 776         subq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 777         .endif
 778         call \sym
 779         .if \ist
 780         addq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 781         .endif
 782         DISABLE_INTERRUPTS(CLBR_NONE)
 783         .if \irqtrace
 784         TRACE_IRQS_OFF
 785         .endif
 786         .endm
 787
 788         /*
 789          * "Paranoid" exit path from exception stack.
 790          * Paranoid because this is used by NMIs and cannot take
 791          * any kernel state for granted.
 792          * We don't do kernel preemption checks here, because only
 793          * NMI should be common and it does not enable IRQs and
 794          * cannot get reschedule ticks.
 795          *
 796          * "trace" is 0 for the NMI handler only, because irq-tracing
 797          * is fundamentally NMI-unsafe. (we cannot change the soft and
 798          * hard flags at once, atomically)
 799          */
 800         .macro paranoidexit trace=1
 801         /* ebx: no swapgs flag */
 802 paranoid_exit\trace:
 803         testl %ebx,%ebx                         /* swapgs needed? */
 804         jnz paranoid_restore\trace
 805         testl $3,CS(%rsp)
 806         jnz   paranoid_userspace\trace
 807 paranoid_swapgs\trace:
 808         .if \trace
 809         TRACE_IRQS_IRETQ 0
 810         .endif
 811         SWAPGS_UNSAFE_STACK
 812 paranoid_restore\trace:
 813         RESTORE_ALL 8
 814         jmp irq_return
 815 paranoid_userspace\trace:
 816         GET_THREAD_INFO(%rcx)
 817         movl threadinfo_flags(%rcx),%ebx
 818         andl $_TIF_WORK_MASK,%ebx
 819         jz paranoid_swapgs\trace
 820         movq %rsp,%rdi                  /* &pt_regs */
 821         call sync_regs
 822         movq %rax,%rsp                  /* switch stack for scheduling */
 823         testl $_TIF_NEED_RESCHED,%ebx
 824         jnz paranoid_schedule\trace
 825         movl %ebx,%edx                  /* arg3: thread flags */
 826         .if \trace
 827         TRACE_IRQS_ON
 828         .endif
 829         ENABLE_INTERRUPTS(CLBR_NONE)
 830         xorl %esi,%esi                  /* arg2: oldset */
 831         movq %rsp,%rdi                  /* arg1: &pt_regs */
 832         call do_notify_resume
 833         DISABLE_INTERRUPTS(CLBR_NONE)
 834         .if \trace
 835         TRACE_IRQS_OFF
 836         .endif
 837         jmp paranoid_userspace\trace
 838 paranoid_schedule\trace:
 839         .if \trace
 840         TRACE_IRQS_ON
 841         .endif
 842         ENABLE_INTERRUPTS(CLBR_ANY)
 843         call schedule
 844         DISABLE_INTERRUPTS(CLBR_ANY)
 845         .if \trace
 846         TRACE_IRQS_OFF
 847         .endif
 848         jmp paranoid_userspace\trace
 849         CFI_ENDPROC
 850         .endm
 851
 852 /*
 853  * Exception entry point. This expects an error code/orig_rax on the stack
 854  * and the exception handler in %rax.
 855  */
 856 KPROBE_ENTRY(error_entry)
 857         _frame RDI
 858         CFI_REL_OFFSET rax,0
 859         /* rdi slot contains rax, oldrax contains error code */
 860         cld
 861         subq  $14*8,%rsp
 862         CFI_ADJUST_CFA_OFFSET   (14*8)
 863         movq %rsi,13*8(%rsp)
 864         CFI_REL_OFFSET  rsi,RSI
 865         movq 14*8(%rsp),%rsi    /* load rax from rdi slot */
 866         CFI_REGISTER    rax,rsi
 867         movq %rdx,12*8(%rsp)
 868         CFI_REL_OFFSET  rdx,RDX
 869         movq %rcx,11*8(%rsp)
 870         CFI_REL_OFFSET  rcx,RCX
 871         movq %rsi,10*8(%rsp)    /* store rax */
 872         CFI_REL_OFFSET  rax,RAX
 873         movq %r8, 9*8(%rsp)
 874         CFI_REL_OFFSET  r8,R8
 875         movq %r9, 8*8(%rsp)
 876         CFI_REL_OFFSET  r9,R9
 877         movq %r10,7*8(%rsp)
 878         CFI_REL_OFFSET  r10,R10
 879         movq %r11,6*8(%rsp)
 880         CFI_REL_OFFSET  r11,R11
 881         movq %rbx,5*8(%rsp)
 882         CFI_REL_OFFSET  rbx,RBX
 883         movq %rbp,4*8(%rsp)
 884         CFI_REL_OFFSET  rbp,RBP
 885         movq %r12,3*8(%rsp)
 886         CFI_REL_OFFSET  r12,R12
 887         movq %r13,2*8(%rsp)
 888         CFI_REL_OFFSET  r13,R13
 889         movq %r14,1*8(%rsp)
 890         CFI_REL_OFFSET  r14,R14
 891         movq %r15,(%rsp)
 892         CFI_REL_OFFSET  r15,R15
 893         xorl %ebx,%ebx
 894         testl $3,CS(%rsp)
 895         je  error_kernelspace
 896 error_swapgs:
 897         SWAPGS
 898 error_sti:
 899         movq %rdi,RDI(%rsp)
 900         CFI_REL_OFFSET  rdi,RDI
 901         movq %rsp,%rdi
 902         movq ORIG_RAX(%rsp),%rsi        /* get error code */
 903         movq $-1,ORIG_RAX(%rsp)
 904         call *%rax
 905         /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
 906 error_exit:
 907         movl %ebx,%eax
 908         RESTORE_REST
 909         DISABLE_INTERRUPTS(CLBR_NONE)
 910         TRACE_IRQS_OFF
 911         GET_THREAD_INFO(%rcx)
 912         testl %eax,%eax
 913         jne  retint_kernel
 914         LOCKDEP_SYS_EXIT_IRQ
 915         movl  threadinfo_flags(%rcx),%edx
 916         movl  $_TIF_WORK_MASK,%edi
 917         andl  %edi,%edx
 918         jnz  retint_careful
 919         jmp retint_swapgs
 920         CFI_ENDPROC
 921
 922 error_kernelspace:
 923         incl %ebx
 924        /* There are two places in the kernel that can potentially fault with
 925           usergs. Handle them here. The exception handlers after
 926            iret run with kernel gs again, so don't set the user space flag.
 927            B stepping K8s sometimes report an truncated RIP for IRET
 928            exceptions returning to compat mode. Check for these here too. */
 929         leaq irq_return(%rip),%rcx
 930         cmpq %rcx,RIP(%rsp)
 931         je   error_swapgs
 932         movl %ecx,%ecx  /* zero extend */
 933         cmpq %rcx,RIP(%rsp)
 934         je   error_swapgs
 935         cmpq $gs_change,RIP(%rsp)
 936         je   error_swapgs
 937         jmp  error_sti
 938 KPROBE_END(error_entry)
 939
 940        /* Reload gs selector with exception handling */
 941        /* edi:  new selector */
 942 ENTRY(load_gs_index)
 943         CFI_STARTPROC
 944         pushf
 945         CFI_ADJUST_CFA_OFFSET 8
 946         DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
 947         SWAPGS
 948 gs_change:
 949         movl %edi,%gs
 950 2:      mfence          /* workaround */
 951         SWAPGS
 952         popf
 953         CFI_ADJUST_CFA_OFFSET -8
 954         ret
 955         CFI_ENDPROC
 956 ENDPROC(load_gs_index)
 957
 958         .section __ex_table,"a"
 959         .align 8
 960         .quad gs_change,bad_gs
 961         .previous
 962         .section .fixup,"ax"
 963         /* running with kernelgs */
 964 bad_gs:
 965         SWAPGS                  /* switch back to user gs */
 966         xorl %eax,%eax
 967         movl %eax,%gs
 968         jmp  2b
 969         .previous
 970
 971 /*
 972  * Create a kernel thread.
 973  *
 974  * C extern interface:
 975  *      extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 976  *
 977  * asm input arguments:
 978  *      rdi: fn, rsi: arg, rdx: flags
 979  */
 980 ENTRY(kernel_thread)
 981         CFI_STARTPROC
 982         FAKE_STACK_FRAME $child_rip
 983         SAVE_ALL
 984
 985         # rdi: flags, rsi: usp, rdx: will be &pt_regs
 986         movq %rdx,%rdi
 987         orq  kernel_thread_flags(%rip),%rdi
 988         movq $-1, %rsi
 989         movq %rsp, %rdx
 990
 991         xorl %r8d,%r8d
 992         xorl %r9d,%r9d
 993
 994         # clone now
 995         call do_fork
 996         movq %rax,RAX(%rsp)
 997         xorl %edi,%edi
 998
 999         /*
1000          * It isn't worth to check for reschedule here,
1001          * so internally to the x86_64 port you can rely on kernel_thread()
1002          * not to reschedule the child before returning, this avoids the need
1003          * of hacks for example to fork off the per-CPU idle tasks.
1004          * [Hopefully no generic code relies on the reschedule -AK]
1005          */
1006         RESTORE_ALL
1007         UNFAKE_STACK_FRAME
1008         ret
1009         CFI_ENDPROC
1010 ENDPROC(kernel_thread)
1011
1012 child_rip:
1013         pushq $0                # fake return address
1014         CFI_STARTPROC
1015         /*
1016          * Here we are in the child and the registers are set as they were
1017          * at kernel_thread() invocation in the parent.
1018          */
1019         movq %rdi, %rax
1020         movq %rsi, %rdi
1021         call *%rax
1022         # exit
1023         mov %eax, %edi
1024         call do_exit
1025         CFI_ENDPROC
1026 ENDPROC(child_rip)
1027
1028 /*
1029  * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1030  *
1031  * C extern interface:
1032  *       extern long execve(char *name, char **argv, char **envp)
1033  *
1034  * asm input arguments:
1035  *      rdi: name, rsi: argv, rdx: envp
1036  *
1037  * We want to fallback into:
1038  *      extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1039  *
1040  * do_sys_execve asm fallback arguments:
1041  *      rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1042  */
1043 ENTRY(kernel_execve)
1044         CFI_STARTPROC
1045         FAKE_STACK_FRAME $0
1046         SAVE_ALL
1047         movq %rsp,%rcx
1048         call sys_execve
1049         movq %rax, RAX(%rsp)
1050         RESTORE_REST
1051         testq %rax,%rax
1052         je int_ret_from_sys_call
1053         RESTORE_ARGS
1054         UNFAKE_STACK_FRAME
1055         ret
1056         CFI_ENDPROC
1057 ENDPROC(kernel_execve)
1058
1059 KPROBE_ENTRY(page_fault)
1060         errorentry do_page_fault
1061 KPROBE_END(page_fault)
1062
1063 ENTRY(coprocessor_error)
1064         zeroentry do_coprocessor_error
1065 END(coprocessor_error)
1066
1067 ENTRY(simd_coprocessor_error)
1068         zeroentry do_simd_coprocessor_error
1069 END(simd_coprocessor_error)
1070
1071 ENTRY(device_not_available)
1072         zeroentry math_state_restore
1073 END(device_not_available)
1074
1075         /* runs on exception stack */
1076 KPROBE_ENTRY(debug)
1077         INTR_FRAME
1078         pushq $0
1079         CFI_ADJUST_CFA_OFFSET 8
1080         paranoidentry do_debug, DEBUG_STACK
1081         paranoidexit
1082 KPROBE_END(debug)
1083
1084         /* runs on exception stack */
1085 KPROBE_ENTRY(nmi)
1086         INTR_FRAME
1087         pushq $-1
1088         CFI_ADJUST_CFA_OFFSET 8
1089         paranoidentry do_nmi, 0, 0
1090 #ifdef CONFIG_TRACE_IRQFLAGS
1091         paranoidexit 0
1092 #else
1093         jmp paranoid_exit1
1094         CFI_ENDPROC
1095 #endif
1096 KPROBE_END(nmi)
1097
1098 KPROBE_ENTRY(int3)
1099         INTR_FRAME
1100         pushq $0
1101         CFI_ADJUST_CFA_OFFSET 8
1102         paranoidentry do_int3, DEBUG_STACK
1103         jmp paranoid_exit1
1104         CFI_ENDPROC
1105 KPROBE_END(int3)
1106
1107 ENTRY(overflow)
1108         zeroentry do_overflow
1109 END(overflow)
1110
1111 ENTRY(bounds)
1112         zeroentry do_bounds
1113 END(bounds)
1114
1115 ENTRY(invalid_op)
1116         zeroentry do_invalid_op
1117 END(invalid_op)
1118
1119 ENTRY(coprocessor_segment_overrun)
1120         zeroentry do_coprocessor_segment_overrun
1121 END(coprocessor_segment_overrun)
1122
1123 ENTRY(reserved)
1124         zeroentry do_reserved
1125 END(reserved)
1126
1127         /* runs on exception stack */
1128 ENTRY(double_fault)
1129         XCPT_FRAME
1130         paranoidentry do_double_fault
1131         jmp paranoid_exit1
1132         CFI_ENDPROC
1133 END(double_fault)
1134
1135 ENTRY(invalid_TSS)
1136         errorentry do_invalid_TSS
1137 END(invalid_TSS)
1138
1139 ENTRY(segment_not_present)
1140         errorentry do_segment_not_present
1141 END(segment_not_present)
1142
1143         /* runs on exception stack */
1144 ENTRY(stack_segment)
1145         XCPT_FRAME
1146         paranoidentry do_stack_segment
1147         jmp paranoid_exit1
1148         CFI_ENDPROC
1149 END(stack_segment)
1150
1151 KPROBE_ENTRY(general_protection)
1152         errorentry do_general_protection
1153 KPROBE_END(general_protection)
1154
1155 ENTRY(alignment_check)
1156         errorentry do_alignment_check
1157 END(alignment_check)
1158
1159 ENTRY(divide_error)
1160         zeroentry do_divide_error
1161 END(divide_error)
1162
1163 ENTRY(spurious_interrupt_bug)
1164         zeroentry do_spurious_interrupt_bug
1165 END(spurious_interrupt_bug)
1166
1167 #ifdef CONFIG_X86_MCE
1168         /* runs on exception stack */
1169 ENTRY(machine_check)
1170         INTR_FRAME
1171         pushq $0
1172         CFI_ADJUST_CFA_OFFSET 8
1173         paranoidentry do_machine_check
1174         jmp paranoid_exit1
1175         CFI_ENDPROC
1176 END(machine_check)
1177 #endif
1178
1179 /* Call softirq on interrupt stack. Interrupts are off. */
1180 ENTRY(call_softirq)
1181         CFI_STARTPROC
1182         push %rbp
1183         CFI_ADJUST_CFA_OFFSET   8
1184         CFI_REL_OFFSET rbp,0
1185         mov  %rsp,%rbp
1186         CFI_DEF_CFA_REGISTER rbp
1187         incl %gs:pda_irqcount
1188         cmove %gs:pda_irqstackptr,%rsp
1189         push  %rbp                      # backlink for old unwinder
1190         call __do_softirq
1191         leaveq
1192         CFI_DEF_CFA_REGISTER    rsp
1193         CFI_ADJUST_CFA_OFFSET   -8
1194         decl %gs:pda_irqcount
1195         ret
1196         CFI_ENDPROC
1197 ENDPROC(call_softirq)
1198
1199 KPROBE_ENTRY(ignore_sysret)
1200         CFI_STARTPROC
1201         mov $-ENOSYS,%eax
1202         sysret
1203         CFI_ENDPROC
1204 ENDPROC(ignore_sysret)