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
  54         .code64
  55
  56 #ifndef CONFIG_PREEMPT
  57 #define retint_kernel retint_restore_args
  58 #endif
  59
  60
  61 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
  62 #ifdef CONFIG_TRACE_IRQFLAGS
  63         bt   $9,EFLAGS-\offset(%rsp)    /* interrupts off? */
  64         jnc  1f
  65         TRACE_IRQS_ON
  66 1:
  67 #endif
  68 .endm
  69
  70 /*
  71  * C code is not supposed to know about undefined top of stack. Every time
  72  * a C function with an pt_regs argument is called from the SYSCALL based
  73  * fast path FIXUP_TOP_OF_STACK is needed.
  74  * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
  75  * manipulation.
  76  */
  77
  78         /* %rsp:at FRAMEEND */
  79         .macro FIXUP_TOP_OF_STACK tmp
  80         movq    %gs:pda_oldrsp,\tmp
  81         movq    \tmp,RSP(%rsp)
  82         movq    $__USER_DS,SS(%rsp)
  83         movq    $__USER_CS,CS(%rsp)
  84         movq    $-1,RCX(%rsp)
  85         movq    R11(%rsp),\tmp  /* get eflags */
  86         movq    \tmp,EFLAGS(%rsp)
  87         .endm
  88
  89         .macro RESTORE_TOP_OF_STACK tmp,offset=0
  90         movq   RSP-\offset(%rsp),\tmp
  91         movq   \tmp,%gs:pda_oldrsp
  92         movq   EFLAGS-\offset(%rsp),\tmp
  93         movq   \tmp,R11-\offset(%rsp)
  94         .endm
  95
  96         .macro FAKE_STACK_FRAME child_rip
  97         /* push in order ss, rsp, eflags, cs, rip */
  98         xorl %eax, %eax
  99         pushq %rax /* ss */
 100         CFI_ADJUST_CFA_OFFSET   8
 101         /*CFI_REL_OFFSET        ss,0*/
 102         pushq %rax /* rsp */
 103         CFI_ADJUST_CFA_OFFSET   8
 104         CFI_REL_OFFSET  rsp,0
 105         pushq $(1<<9) /* eflags - interrupts on */
 106         CFI_ADJUST_CFA_OFFSET   8
 107         /*CFI_REL_OFFSET        rflags,0*/
 108         pushq $__KERNEL_CS /* cs */
 109         CFI_ADJUST_CFA_OFFSET   8
 110         /*CFI_REL_OFFSET        cs,0*/
 111         pushq \child_rip /* rip */
 112         CFI_ADJUST_CFA_OFFSET   8
 113         CFI_REL_OFFSET  rip,0
 114         pushq   %rax /* orig rax */
 115         CFI_ADJUST_CFA_OFFSET   8
 116         .endm
 117
 118         .macro UNFAKE_STACK_FRAME
 119         addq $8*6, %rsp
 120         CFI_ADJUST_CFA_OFFSET   -(6*8)
 121         .endm
 122
 123         .macro  CFI_DEFAULT_STACK start=1
 124         .if \start
 125         CFI_STARTPROC   simple
 126         CFI_SIGNAL_FRAME
 127         CFI_DEF_CFA     rsp,SS+8
 128         .else
 129         CFI_DEF_CFA_OFFSET SS+8
 130         .endif
 131         CFI_REL_OFFSET  r15,R15
 132         CFI_REL_OFFSET  r14,R14
 133         CFI_REL_OFFSET  r13,R13
 134         CFI_REL_OFFSET  r12,R12
 135         CFI_REL_OFFSET  rbp,RBP
 136         CFI_REL_OFFSET  rbx,RBX
 137         CFI_REL_OFFSET  r11,R11
 138         CFI_REL_OFFSET  r10,R10
 139         CFI_REL_OFFSET  r9,R9
 140         CFI_REL_OFFSET  r8,R8
 141         CFI_REL_OFFSET  rax,RAX
 142         CFI_REL_OFFSET  rcx,RCX
 143         CFI_REL_OFFSET  rdx,RDX
 144         CFI_REL_OFFSET  rsi,RSI
 145         CFI_REL_OFFSET  rdi,RDI
 146         CFI_REL_OFFSET  rip,RIP
 147         /*CFI_REL_OFFSET        cs,CS*/
 148         /*CFI_REL_OFFSET        rflags,EFLAGS*/
 149         CFI_REL_OFFSET  rsp,RSP
 150         /*CFI_REL_OFFSET        ss,SS*/
 151         .endm
 152 /*
 153  * A newly forked process directly context switches into this.
 154  */
 155 /* rdi: prev */
 156 ENTRY(ret_from_fork)
 157         CFI_DEFAULT_STACK
 158         push kernel_eflags(%rip)
 159         CFI_ADJUST_CFA_OFFSET 4
 160         popf                            # reset kernel eflags
 161         CFI_ADJUST_CFA_OFFSET -4
 162         call schedule_tail
 163         GET_THREAD_INFO(%rcx)
 164         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),threadinfo_flags(%rcx)
 165         jnz rff_trace
 166 rff_action:
 167         RESTORE_REST
 168         testl $3,CS-ARGOFFSET(%rsp)     # from kernel_thread?
 169         je   int_ret_from_sys_call
 170         testl $_TIF_IA32,threadinfo_flags(%rcx)
 171         jnz  int_ret_from_sys_call
 172         RESTORE_TOP_OF_STACK %rdi,ARGOFFSET
 173         jmp ret_from_sys_call
 174 rff_trace:
 175         movq %rsp,%rdi
 176         call syscall_trace_leave
 177         GET_THREAD_INFO(%rcx)
 178         jmp rff_action
 179         CFI_ENDPROC
 180 END(ret_from_fork)
 181
 182 /*
 183  * System call entry. Upto 6 arguments in registers are supported.
 184  *
 185  * SYSCALL does not save anything on the stack and does not change the
 186  * stack pointer.
 187  */
 188
 189 /*
 190  * Register setup:
 191  * rax  system call number
 192  * rdi  arg0
 193  * rcx  return address for syscall/sysret, C arg3
 194  * rsi  arg1
 195  * rdx  arg2
 196  * r10  arg3    (--> moved to rcx for C)
 197  * r8   arg4
 198  * r9   arg5
 199  * r11  eflags for syscall/sysret, temporary for C
 200  * r12-r15,rbp,rbx saved by C code, not touched.
 201  *
 202  * Interrupts are off on entry.
 203  * Only called from user space.
 204  *
 205  * XXX  if we had a free scratch register we could save the RSP into the stack frame
 206  *      and report it properly in ps. Unfortunately we haven't.
 207  *
 208  * When user can change the frames always force IRET. That is because
 209  * it deals with uncanonical addresses better. SYSRET has trouble
 210  * with them due to bugs in both AMD and Intel CPUs.
 211  */
 212
 213 ENTRY(system_call)
 214         CFI_STARTPROC   simple
 215         CFI_SIGNAL_FRAME
 216         CFI_DEF_CFA     rsp,PDA_STACKOFFSET
 217         CFI_REGISTER    rip,rcx
 218         /*CFI_REGISTER  rflags,r11*/
 219         swapgs
 220         movq    %rsp,%gs:pda_oldrsp
 221         movq    %gs:pda_kernelstack,%rsp
 222         /*
 223          * No need to follow this irqs off/on section - it's straight
 224          * and short:
 225          */
 226         sti
 227         SAVE_ARGS 8,1
 228         movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
 229         movq  %rcx,RIP-ARGOFFSET(%rsp)
 230         CFI_REL_OFFSET rip,RIP-ARGOFFSET
 231         GET_THREAD_INFO(%rcx)
 232         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP),threadinfo_flags(%rcx)
 233         jnz tracesys
 234         cmpq $__NR_syscall_max,%rax
 235         ja badsys
 236         movq %r10,%rcx
 237         call *sys_call_table(,%rax,8)  # XXX:    rip relative
 238         movq %rax,RAX-ARGOFFSET(%rsp)
 239 /*
 240  * Syscall return path ending with SYSRET (fast path)
 241  * Has incomplete stack frame and undefined top of stack.
 242  */
 243 ret_from_sys_call:
 244         movl $_TIF_ALLWORK_MASK,%edi
 245         /* edi: flagmask */
 246 sysret_check:
 247         GET_THREAD_INFO(%rcx)
 248         cli
 249         TRACE_IRQS_OFF
 250         movl threadinfo_flags(%rcx),%edx
 251         andl %edi,%edx
 252         jnz  sysret_careful
 253         CFI_REMEMBER_STATE
 254         /*
 255          * sysretq will re-enable interrupts:
 256          */
 257         TRACE_IRQS_ON
 258         movq RIP-ARGOFFSET(%rsp),%rcx
 259         CFI_REGISTER    rip,rcx
 260         RESTORE_ARGS 0,-ARG_SKIP,1
 261         /*CFI_REGISTER  rflags,r11*/
 262         movq    %gs:pda_oldrsp,%rsp
 263         swapgs
 264         sysretq
 265
 266         CFI_RESTORE_STATE
 267         /* Handle reschedules */
 268         /* edx: work, edi: workmask */
 269 sysret_careful:
 270         bt $TIF_NEED_RESCHED,%edx
 271         jnc sysret_signal
 272         TRACE_IRQS_ON
 273         sti
 274         pushq %rdi
 275         CFI_ADJUST_CFA_OFFSET 8
 276         call schedule
 277         popq  %rdi
 278         CFI_ADJUST_CFA_OFFSET -8
 279         jmp sysret_check
 280
 281         /* Handle a signal */
 282 sysret_signal:
 283         TRACE_IRQS_ON
 284         sti
 285         testl $(_TIF_SIGPENDING|_TIF_SINGLESTEP|_TIF_MCE_NOTIFY),%edx
 286         jz    1f
 287
 288         /* Really a signal */
 289         /* edx: work flags (arg3) */
 290         leaq do_notify_resume(%rip),%rax
 291         leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
 292         xorl %esi,%esi # oldset -> arg2
 293         call ptregscall_common
 294 1:      movl $_TIF_NEED_RESCHED,%edi
 295         /* Use IRET because user could have changed frame. This
 296            works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
 297         cli
 298         TRACE_IRQS_OFF
 299         jmp int_with_check
 300
 301 badsys:
 302         movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
 303         jmp ret_from_sys_call
 304
 305         /* Do syscall tracing */
 306 tracesys:
 307         SAVE_REST
 308         movq $-ENOSYS,RAX(%rsp)
 309         FIXUP_TOP_OF_STACK %rdi
 310         movq %rsp,%rdi
 311         call syscall_trace_enter
 312         LOAD_ARGS ARGOFFSET  /* reload args from stack in case ptrace changed it */
 313         RESTORE_REST
 314         cmpq $__NR_syscall_max,%rax
 315         movq $-ENOSYS,%rcx
 316         cmova %rcx,%rax
 317         ja  1f
 318         movq %r10,%rcx  /* fixup for C */
 319         call *sys_call_table(,%rax,8)
 320 1:      movq %rax,RAX-ARGOFFSET(%rsp)
 321         /* Use IRET because user could have changed frame */
 322
 323 /*
 324  * Syscall return path ending with IRET.
 325  * Has correct top of stack, but partial stack frame.
 326  */
 327         .globl int_ret_from_sys_call
 328 int_ret_from_sys_call:
 329         cli
 330         TRACE_IRQS_OFF
 331         testl $3,CS-ARGOFFSET(%rsp)
 332         je retint_restore_args
 333         movl $_TIF_ALLWORK_MASK,%edi
 334         /* edi: mask to check */
 335 int_with_check:
 336         GET_THREAD_INFO(%rcx)
 337         movl threadinfo_flags(%rcx),%edx
 338         andl %edi,%edx
 339         jnz   int_careful
 340         andl    $~TS_COMPAT,threadinfo_status(%rcx)
 341         jmp   retint_swapgs
 342
 343         /* Either reschedule or signal or syscall exit tracking needed. */
 344         /* First do a reschedule test. */
 345         /* edx: work, edi: workmask */
 346 int_careful:
 347         bt $TIF_NEED_RESCHED,%edx
 348         jnc  int_very_careful
 349         TRACE_IRQS_ON
 350         sti
 351         pushq %rdi
 352         CFI_ADJUST_CFA_OFFSET 8
 353         call schedule
 354         popq %rdi
 355         CFI_ADJUST_CFA_OFFSET -8
 356         cli
 357         TRACE_IRQS_OFF
 358         jmp int_with_check
 359
 360         /* handle signals and tracing -- both require a full stack frame */
 361 int_very_careful:
 362         TRACE_IRQS_ON
 363         sti
 364         SAVE_REST
 365         /* Check for syscall exit trace */
 366         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx
 367         jz int_signal
 368         pushq %rdi
 369         CFI_ADJUST_CFA_OFFSET 8
 370         leaq 8(%rsp),%rdi       # &ptregs -> arg1
 371         call syscall_trace_leave
 372         popq %rdi
 373         CFI_ADJUST_CFA_OFFSET -8
 374         andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi
 375         jmp int_restore_rest
 376
 377 int_signal:
 378         testl $(_TIF_SIGPENDING|_TIF_SINGLESTEP|_TIF_MCE_NOTIFY),%edx
 379         jz 1f
 380         movq %rsp,%rdi          # &ptregs -> arg1
 381         xorl %esi,%esi          # oldset -> arg2
 382         call do_notify_resume
 383 1:      movl $_TIF_NEED_RESCHED,%edi
 384 int_restore_rest:
 385         RESTORE_REST
 386         cli
 387         TRACE_IRQS_OFF
 388         jmp int_with_check
 389         CFI_ENDPROC
 390 END(system_call)
 391
 392 /*
 393  * Certain special system calls that need to save a complete full stack frame.
 394  */
 395
 396         .macro PTREGSCALL label,func,arg
 397         .globl \label
 398 \label:
 399         leaq    \func(%rip),%rax
 400         leaq    -ARGOFFSET+8(%rsp),\arg /* 8 for return address */
 401         jmp     ptregscall_common
 402 END(\label)
 403         .endm
 404
 405         CFI_STARTPROC
 406
 407         PTREGSCALL stub_clone, sys_clone, %r8
 408         PTREGSCALL stub_fork, sys_fork, %rdi
 409         PTREGSCALL stub_vfork, sys_vfork, %rdi
 410         PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx
 411         PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 412         PTREGSCALL stub_iopl, sys_iopl, %rsi
 413
 414 ENTRY(ptregscall_common)
 415         popq %r11
 416         CFI_ADJUST_CFA_OFFSET -8
 417         CFI_REGISTER rip, r11
 418         SAVE_REST
 419         movq %r11, %r15
 420         CFI_REGISTER rip, r15
 421         FIXUP_TOP_OF_STACK %r11
 422         call *%rax
 423         RESTORE_TOP_OF_STACK %r11
 424         movq %r15, %r11
 425         CFI_REGISTER rip, r11
 426         RESTORE_REST
 427         pushq %r11
 428         CFI_ADJUST_CFA_OFFSET 8
 429         CFI_REL_OFFSET rip, 0
 430         ret
 431         CFI_ENDPROC
 432 END(ptregscall_common)
 433
 434 ENTRY(stub_execve)
 435         CFI_STARTPROC
 436         popq %r11
 437         CFI_ADJUST_CFA_OFFSET -8
 438         CFI_REGISTER rip, r11
 439         SAVE_REST
 440         FIXUP_TOP_OF_STACK %r11
 441         call sys_execve
 442         RESTORE_TOP_OF_STACK %r11
 443         movq %rax,RAX(%rsp)
 444         RESTORE_REST
 445         jmp int_ret_from_sys_call
 446         CFI_ENDPROC
 447 END(stub_execve)
 448
 449 /*
 450  * sigreturn is special because it needs to restore all registers on return.
 451  * This cannot be done with SYSRET, so use the IRET return path instead.
 452  */
 453 ENTRY(stub_rt_sigreturn)
 454         CFI_STARTPROC
 455         addq $8, %rsp
 456         CFI_ADJUST_CFA_OFFSET   -8
 457         SAVE_REST
 458         movq %rsp,%rdi
 459         FIXUP_TOP_OF_STACK %r11
 460         call sys_rt_sigreturn
 461         movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 462         RESTORE_REST
 463         jmp int_ret_from_sys_call
 464         CFI_ENDPROC
 465 END(stub_rt_sigreturn)
 466
 467 /*
 468  * initial frame state for interrupts and exceptions
 469  */
 470         .macro _frame ref
 471         CFI_STARTPROC simple
 472         CFI_SIGNAL_FRAME
 473         CFI_DEF_CFA rsp,SS+8-\ref
 474         /*CFI_REL_OFFSET ss,SS-\ref*/
 475         CFI_REL_OFFSET rsp,RSP-\ref
 476         /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/
 477         /*CFI_REL_OFFSET cs,CS-\ref*/
 478         CFI_REL_OFFSET rip,RIP-\ref
 479         .endm
 480
 481 /* initial frame state for interrupts (and exceptions without error code) */
 482 #define INTR_FRAME _frame RIP
 483 /* initial frame state for exceptions with error code (and interrupts with
 484    vector already pushed) */
 485 #define XCPT_FRAME _frame ORIG_RAX
 486
 487 /*
 488  * Interrupt entry/exit.
 489  *
 490  * Interrupt entry points save only callee clobbered registers in fast path.
 491  *
 492  * Entry runs with interrupts off.
 493  */
 494
 495 /* 0(%rsp): interrupt number */
 496         .macro interrupt func
 497         cld
 498         SAVE_ARGS
 499         leaq -ARGOFFSET(%rsp),%rdi      # arg1 for handler
 500         pushq %rbp
 501         CFI_ADJUST_CFA_OFFSET   8
 502         CFI_REL_OFFSET          rbp, 0
 503         movq %rsp,%rbp
 504         CFI_DEF_CFA_REGISTER    rbp
 505         testl $3,CS(%rdi)
 506         je 1f
 507         swapgs
 508         /* irqcount is used to check if a CPU is already on an interrupt
 509            stack or not. While this is essentially redundant with preempt_count
 510            it is a little cheaper to use a separate counter in the PDA
 511            (short of moving irq_enter into assembly, which would be too
 512             much work) */
 513 1:      incl    %gs:pda_irqcount
 514         cmoveq %gs:pda_irqstackptr,%rsp
 515         push    %rbp                    # backlink for old unwinder
 516         /*
 517          * We entered an interrupt context - irqs are off:
 518          */
 519         TRACE_IRQS_OFF
 520         call \func
 521         .endm
 522
 523 ENTRY(common_interrupt)
 524         XCPT_FRAME
 525         interrupt do_IRQ
 526         /* 0(%rsp): oldrsp-ARGOFFSET */
 527 ret_from_intr:
 528         cli
 529         TRACE_IRQS_OFF
 530         decl %gs:pda_irqcount
 531         leaveq
 532         CFI_DEF_CFA_REGISTER    rsp
 533         CFI_ADJUST_CFA_OFFSET   -8
 534 exit_intr:
 535         GET_THREAD_INFO(%rcx)
 536         testl $3,CS-ARGOFFSET(%rsp)
 537         je retint_kernel
 538
 539         /* Interrupt came from user space */
 540         /*
 541          * Has a correct top of stack, but a partial stack frame
 542          * %rcx: thread info. Interrupts off.
 543          */
 544 retint_with_reschedule:
 545         movl $_TIF_WORK_MASK,%edi
 546 retint_check:
 547         movl threadinfo_flags(%rcx),%edx
 548         andl %edi,%edx
 549         CFI_REMEMBER_STATE
 550         jnz  retint_careful
 551 retint_swapgs:
 552         /*
 553          * The iretq could re-enable interrupts:
 554          */
 555         cli
 556         TRACE_IRQS_IRETQ
 557         swapgs
 558         jmp restore_args
 559
 560 retint_restore_args:
 561         cli
 562         /*
 563          * The iretq could re-enable interrupts:
 564          */
 565         TRACE_IRQS_IRETQ
 566 restore_args:
 567         RESTORE_ARGS 0,8,0
 568 iret_label:
 569         iretq
 570
 571         .section __ex_table,"a"
 572         .quad iret_label,bad_iret
 573         .previous
 574         .section .fixup,"ax"
 575         /* force a signal here? this matches i386 behaviour */
 576         /* running with kernel gs */
 577 bad_iret:
 578         movq $11,%rdi   /* SIGSEGV */
 579         TRACE_IRQS_ON
 580         sti
 581         jmp do_exit
 582         .previous
 583
 584         /* edi: workmask, edx: work */
 585 retint_careful:
 586         CFI_RESTORE_STATE
 587         bt    $TIF_NEED_RESCHED,%edx
 588         jnc   retint_signal
 589         TRACE_IRQS_ON
 590         sti
 591         pushq %rdi
 592         CFI_ADJUST_CFA_OFFSET   8
 593         call  schedule
 594         popq %rdi
 595         CFI_ADJUST_CFA_OFFSET   -8
 596         GET_THREAD_INFO(%rcx)
 597         cli
 598         TRACE_IRQS_OFF
 599         jmp retint_check
 600
 601 retint_signal:
 602         testl $(_TIF_SIGPENDING|_TIF_SINGLESTEP|_TIF_MCE_NOTIFY),%edx
 603         jz    retint_swapgs
 604         TRACE_IRQS_ON
 605         sti
 606         SAVE_REST
 607         movq $-1,ORIG_RAX(%rsp)
 608         xorl %esi,%esi          # oldset
 609         movq %rsp,%rdi          # &pt_regs
 610         call do_notify_resume
 611         RESTORE_REST
 612         cli
 613         TRACE_IRQS_OFF
 614         movl $_TIF_NEED_RESCHED,%edi
 615         GET_THREAD_INFO(%rcx)
 616         jmp retint_check
 617
 618 #ifdef CONFIG_PREEMPT
 619         /* Returning to kernel space. Check if we need preemption */
 620         /* rcx:  threadinfo. interrupts off. */
 621 ENTRY(retint_kernel)
 622         cmpl $0,threadinfo_preempt_count(%rcx)
 623         jnz  retint_restore_args
 624         bt  $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
 625         jnc  retint_restore_args
 626         bt   $9,EFLAGS-ARGOFFSET(%rsp)  /* interrupts off? */
 627         jnc  retint_restore_args
 628         call preempt_schedule_irq
 629         jmp exit_intr
 630 #endif
 631
 632         CFI_ENDPROC
 633 END(common_interrupt)
 634
 635 /*
 636  * APIC interrupts.
 637  */
 638         .macro apicinterrupt num,func
 639         INTR_FRAME
 640         pushq $~(\num)
 641         CFI_ADJUST_CFA_OFFSET 8
 642         interrupt \func
 643         jmp ret_from_intr
 644         CFI_ENDPROC
 645         .endm
 646
 647 ENTRY(thermal_interrupt)
 648         apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
 649 END(thermal_interrupt)
 650
 651 ENTRY(threshold_interrupt)
 652         apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
 653 END(threshold_interrupt)
 654
 655 #ifdef CONFIG_SMP
 656 ENTRY(reschedule_interrupt)
 657         apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
 658 END(reschedule_interrupt)
 659
 660         .macro INVALIDATE_ENTRY num
 661 ENTRY(invalidate_interrupt\num)
 662         apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
 663 END(invalidate_interrupt\num)
 664         .endm
 665
 666         INVALIDATE_ENTRY 0
 667         INVALIDATE_ENTRY 1
 668         INVALIDATE_ENTRY 2
 669         INVALIDATE_ENTRY 3
 670         INVALIDATE_ENTRY 4
 671         INVALIDATE_ENTRY 5
 672         INVALIDATE_ENTRY 6
 673         INVALIDATE_ENTRY 7
 674
 675 ENTRY(call_function_interrupt)
 676         apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
 677 END(call_function_interrupt)
 678 ENTRY(irq_move_cleanup_interrupt)
 679         apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
 680 END(irq_move_cleanup_interrupt)
 681 #endif
 682
 683 ENTRY(apic_timer_interrupt)
 684         apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
 685 END(apic_timer_interrupt)
 686
 687 ENTRY(error_interrupt)
 688         apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
 689 END(error_interrupt)
 690
 691 ENTRY(spurious_interrupt)
 692         apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
 693 END(spurious_interrupt)
 694
 695 /*
 696  * Exception entry points.
 697  */
 698         .macro zeroentry sym
 699         INTR_FRAME
 700         pushq $0        /* push error code/oldrax */
 701         CFI_ADJUST_CFA_OFFSET 8
 702         pushq %rax      /* push real oldrax to the rdi slot */
 703         CFI_ADJUST_CFA_OFFSET 8
 704         CFI_REL_OFFSET rax,0
 705         leaq  \sym(%rip),%rax
 706         jmp error_entry
 707         CFI_ENDPROC
 708         .endm
 709
 710         .macro errorentry sym
 711         XCPT_FRAME
 712         pushq %rax
 713         CFI_ADJUST_CFA_OFFSET 8
 714         CFI_REL_OFFSET rax,0
 715         leaq  \sym(%rip),%rax
 716         jmp error_entry
 717         CFI_ENDPROC
 718         .endm
 719
 720         /* error code is on the stack already */
 721         /* handle NMI like exceptions that can happen everywhere */
 722         .macro paranoidentry sym, ist=0, irqtrace=1
 723         SAVE_ALL
 724         cld
 725         movl $1,%ebx
 726         movl  $MSR_GS_BASE,%ecx
 727         rdmsr
 728         testl %edx,%edx
 729         js    1f
 730         swapgs
 731         xorl  %ebx,%ebx
 732 1:
 733         .if \ist
 734         movq    %gs:pda_data_offset, %rbp
 735         .endif
 736         movq %rsp,%rdi
 737         movq ORIG_RAX(%rsp),%rsi
 738         movq $-1,ORIG_RAX(%rsp)
 739         .if \ist
 740         subq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 741         .endif
 742         call \sym
 743         .if \ist
 744         addq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 745         .endif
 746         cli
 747         .if \irqtrace
 748         TRACE_IRQS_OFF
 749         .endif
 750         .endm
 751
 752         /*
 753          * "Paranoid" exit path from exception stack.
 754          * Paranoid because this is used by NMIs and cannot take
 755          * any kernel state for granted.
 756          * We don't do kernel preemption checks here, because only
 757          * NMI should be common and it does not enable IRQs and
 758          * cannot get reschedule ticks.
 759          *
 760          * "trace" is 0 for the NMI handler only, because irq-tracing
 761          * is fundamentally NMI-unsafe. (we cannot change the soft and
 762          * hard flags at once, atomically)
 763          */
 764         .macro paranoidexit trace=1
 765         /* ebx: no swapgs flag */
 766 paranoid_exit\trace:
 767         testl %ebx,%ebx                         /* swapgs needed? */
 768         jnz paranoid_restore\trace
 769         testl $3,CS(%rsp)
 770         jnz   paranoid_userspace\trace
 771 paranoid_swapgs\trace:
 772         .if \trace
 773         TRACE_IRQS_IRETQ 0
 774         .endif
 775         swapgs
 776 paranoid_restore\trace:
 777         RESTORE_ALL 8
 778         iretq
 779 paranoid_userspace\trace:
 780         GET_THREAD_INFO(%rcx)
 781         movl threadinfo_flags(%rcx),%ebx
 782         andl $_TIF_WORK_MASK,%ebx
 783         jz paranoid_swapgs\trace
 784         movq %rsp,%rdi                  /* &pt_regs */
 785         call sync_regs
 786         movq %rax,%rsp                  /* switch stack for scheduling */
 787         testl $_TIF_NEED_RESCHED,%ebx
 788         jnz paranoid_schedule\trace
 789         movl %ebx,%edx                  /* arg3: thread flags */
 790         .if \trace
 791         TRACE_IRQS_ON
 792         .endif
 793         sti
 794         xorl %esi,%esi                  /* arg2: oldset */
 795         movq %rsp,%rdi                  /* arg1: &pt_regs */
 796         call do_notify_resume
 797         cli
 798         .if \trace
 799         TRACE_IRQS_OFF
 800         .endif
 801         jmp paranoid_userspace\trace
 802 paranoid_schedule\trace:
 803         .if \trace
 804         TRACE_IRQS_ON
 805         .endif
 806         sti
 807         call schedule
 808         cli
 809         .if \trace
 810         TRACE_IRQS_OFF
 811         .endif
 812         jmp paranoid_userspace\trace
 813         CFI_ENDPROC
 814         .endm
 815
 816 /*
 817  * Exception entry point. This expects an error code/orig_rax on the stack
 818  * and the exception handler in %rax.
 819  */
 820 KPROBE_ENTRY(error_entry)
 821         _frame RDI
 822         CFI_REL_OFFSET rax,0
 823         /* rdi slot contains rax, oldrax contains error code */
 824         cld
 825         subq  $14*8,%rsp
 826         CFI_ADJUST_CFA_OFFSET   (14*8)
 827         movq %rsi,13*8(%rsp)
 828         CFI_REL_OFFSET  rsi,RSI
 829         movq 14*8(%rsp),%rsi    /* load rax from rdi slot */
 830         CFI_REGISTER    rax,rsi
 831         movq %rdx,12*8(%rsp)
 832         CFI_REL_OFFSET  rdx,RDX
 833         movq %rcx,11*8(%rsp)
 834         CFI_REL_OFFSET  rcx,RCX
 835         movq %rsi,10*8(%rsp)    /* store rax */
 836         CFI_REL_OFFSET  rax,RAX
 837         movq %r8, 9*8(%rsp)
 838         CFI_REL_OFFSET  r8,R8
 839         movq %r9, 8*8(%rsp)
 840         CFI_REL_OFFSET  r9,R9
 841         movq %r10,7*8(%rsp)
 842         CFI_REL_OFFSET  r10,R10
 843         movq %r11,6*8(%rsp)
 844         CFI_REL_OFFSET  r11,R11
 845         movq %rbx,5*8(%rsp)
 846         CFI_REL_OFFSET  rbx,RBX
 847         movq %rbp,4*8(%rsp)
 848         CFI_REL_OFFSET  rbp,RBP
 849         movq %r12,3*8(%rsp)
 850         CFI_REL_OFFSET  r12,R12
 851         movq %r13,2*8(%rsp)
 852         CFI_REL_OFFSET  r13,R13
 853         movq %r14,1*8(%rsp)
 854         CFI_REL_OFFSET  r14,R14
 855         movq %r15,(%rsp)
 856         CFI_REL_OFFSET  r15,R15
 857         xorl %ebx,%ebx
 858         testl $3,CS(%rsp)
 859         je  error_kernelspace
 860 error_swapgs:
 861         swapgs
 862 error_sti:
 863         movq %rdi,RDI(%rsp)
 864         CFI_REL_OFFSET  rdi,RDI
 865         movq %rsp,%rdi
 866         movq ORIG_RAX(%rsp),%rsi        /* get error code */
 867         movq $-1,ORIG_RAX(%rsp)
 868         call *%rax
 869         /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
 870 error_exit:
 871         movl %ebx,%eax
 872         RESTORE_REST
 873         cli
 874         TRACE_IRQS_OFF
 875         GET_THREAD_INFO(%rcx)
 876         testl %eax,%eax
 877         jne  retint_kernel
 878         movl  threadinfo_flags(%rcx),%edx
 879         movl  $_TIF_WORK_MASK,%edi
 880         andl  %edi,%edx
 881         jnz  retint_careful
 882         /*
 883          * The iret might restore flags:
 884          */
 885         TRACE_IRQS_IRETQ
 886         swapgs
 887         RESTORE_ARGS 0,8,0
 888         jmp iret_label
 889         CFI_ENDPROC
 890
 891 error_kernelspace:
 892         incl %ebx
 893        /* There are two places in the kernel that can potentially fault with
 894           usergs. Handle them here. The exception handlers after
 895            iret run with kernel gs again, so don't set the user space flag.
 896            B stepping K8s sometimes report an truncated RIP for IRET
 897            exceptions returning to compat mode. Check for these here too. */
 898         leaq iret_label(%rip),%rbp
 899         cmpq %rbp,RIP(%rsp)
 900         je   error_swapgs
 901         movl %ebp,%ebp  /* zero extend */
 902         cmpq %rbp,RIP(%rsp)
 903         je   error_swapgs
 904         cmpq $gs_change,RIP(%rsp)
 905         je   error_swapgs
 906         jmp  error_sti
 907 KPROBE_END(error_entry)
 908
 909        /* Reload gs selector with exception handling */
 910        /* edi:  new selector */
 911 ENTRY(load_gs_index)
 912         CFI_STARTPROC
 913         pushf
 914         CFI_ADJUST_CFA_OFFSET 8
 915         cli
 916         swapgs
 917 gs_change:
 918         movl %edi,%gs
 919 2:      mfence          /* workaround */
 920         swapgs
 921         popf
 922         CFI_ADJUST_CFA_OFFSET -8
 923         ret
 924         CFI_ENDPROC
 925 ENDPROC(load_gs_index)
 926
 927         .section __ex_table,"a"
 928         .align 8
 929         .quad gs_change,bad_gs
 930         .previous
 931         .section .fixup,"ax"
 932         /* running with kernelgs */
 933 bad_gs:
 934         swapgs                  /* switch back to user gs */
 935         xorl %eax,%eax
 936         movl %eax,%gs
 937         jmp  2b
 938         .previous
 939
 940 /*
 941  * Create a kernel thread.
 942  *
 943  * C extern interface:
 944  *      extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 945  *
 946  * asm input arguments:
 947  *      rdi: fn, rsi: arg, rdx: flags
 948  */
 949 ENTRY(kernel_thread)
 950         CFI_STARTPROC
 951         FAKE_STACK_FRAME $child_rip
 952         SAVE_ALL
 953
 954         # rdi: flags, rsi: usp, rdx: will be &pt_regs
 955         movq %rdx,%rdi
 956         orq  kernel_thread_flags(%rip),%rdi
 957         movq $-1, %rsi
 958         movq %rsp, %rdx
 959
 960         xorl %r8d,%r8d
 961         xorl %r9d,%r9d
 962
 963         # clone now
 964         call do_fork
 965         movq %rax,RAX(%rsp)
 966         xorl %edi,%edi
 967
 968         /*
 969          * It isn't worth to check for reschedule here,
 970          * so internally to the x86_64 port you can rely on kernel_thread()
 971          * not to reschedule the child before returning, this avoids the need
 972          * of hacks for example to fork off the per-CPU idle tasks.
 973          * [Hopefully no generic code relies on the reschedule -AK]
 974          */
 975         RESTORE_ALL
 976         UNFAKE_STACK_FRAME
 977         ret
 978         CFI_ENDPROC
 979 ENDPROC(kernel_thread)
 980
 981 child_rip:
 982         pushq $0                # fake return address
 983         CFI_STARTPROC
 984         /*
 985          * Here we are in the child and the registers are set as they were
 986          * at kernel_thread() invocation in the parent.
 987          */
 988         movq %rdi, %rax
 989         movq %rsi, %rdi
 990         call *%rax
 991         # exit
 992         xorl %edi, %edi
 993         call do_exit
 994         CFI_ENDPROC
 995 ENDPROC(child_rip)
 996
 997 /*
 998  * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
 999  *
1000  * C extern interface:
1001  *       extern long execve(char *name, char **argv, char **envp)
1002  *
1003  * asm input arguments:
1004  *      rdi: name, rsi: argv, rdx: envp
1005  *
1006  * We want to fallback into:
1007  *      extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs)
1008  *
1009  * do_sys_execve asm fallback arguments:
1010  *      rdi: name, rsi: argv, rdx: envp, fake frame on the stack
1011  */
1012 ENTRY(kernel_execve)
1013         CFI_STARTPROC
1014         FAKE_STACK_FRAME $0
1015         SAVE_ALL
1016         call sys_execve
1017         movq %rax, RAX(%rsp)
1018         RESTORE_REST
1019         testq %rax,%rax
1020         je int_ret_from_sys_call
1021         RESTORE_ARGS
1022         UNFAKE_STACK_FRAME
1023         ret
1024         CFI_ENDPROC
1025 ENDPROC(kernel_execve)
1026
1027 KPROBE_ENTRY(page_fault)
1028         errorentry do_page_fault
1029 KPROBE_END(page_fault)
1030
1031 ENTRY(coprocessor_error)
1032         zeroentry do_coprocessor_error
1033 END(coprocessor_error)
1034
1035 ENTRY(simd_coprocessor_error)
1036         zeroentry do_simd_coprocessor_error
1037 END(simd_coprocessor_error)
1038
1039 ENTRY(device_not_available)
1040         zeroentry math_state_restore
1041 END(device_not_available)
1042
1043         /* runs on exception stack */
1044 KPROBE_ENTRY(debug)
1045         INTR_FRAME
1046         pushq $0
1047         CFI_ADJUST_CFA_OFFSET 8
1048         paranoidentry do_debug, DEBUG_STACK
1049         paranoidexit
1050 KPROBE_END(debug)
1051
1052         /* runs on exception stack */
1053 KPROBE_ENTRY(nmi)
1054         INTR_FRAME
1055         pushq $-1
1056         CFI_ADJUST_CFA_OFFSET 8
1057         paranoidentry do_nmi, 0, 0
1058 #ifdef CONFIG_TRACE_IRQFLAGS
1059         paranoidexit 0
1060 #else
1061         jmp paranoid_exit1
1062         CFI_ENDPROC
1063 #endif
1064 KPROBE_END(nmi)
1065
1066 KPROBE_ENTRY(int3)
1067         INTR_FRAME
1068         pushq $0
1069         CFI_ADJUST_CFA_OFFSET 8
1070         paranoidentry do_int3, DEBUG_STACK
1071         jmp paranoid_exit1
1072         CFI_ENDPROC
1073 KPROBE_END(int3)
1074
1075 ENTRY(overflow)
1076         zeroentry do_overflow
1077 END(overflow)
1078
1079 ENTRY(bounds)
1080         zeroentry do_bounds
1081 END(bounds)
1082
1083 ENTRY(invalid_op)
1084         zeroentry do_invalid_op
1085 END(invalid_op)
1086
1087 ENTRY(coprocessor_segment_overrun)
1088         zeroentry do_coprocessor_segment_overrun
1089 END(coprocessor_segment_overrun)
1090
1091 ENTRY(reserved)
1092         zeroentry do_reserved
1093 END(reserved)
1094
1095         /* runs on exception stack */
1096 ENTRY(double_fault)
1097         XCPT_FRAME
1098         paranoidentry do_double_fault
1099         jmp paranoid_exit1
1100         CFI_ENDPROC
1101 END(double_fault)
1102
1103 ENTRY(invalid_TSS)
1104         errorentry do_invalid_TSS
1105 END(invalid_TSS)
1106
1107 ENTRY(segment_not_present)
1108         errorentry do_segment_not_present
1109 END(segment_not_present)
1110
1111         /* runs on exception stack */
1112 ENTRY(stack_segment)
1113         XCPT_FRAME
1114         paranoidentry do_stack_segment
1115         jmp paranoid_exit1
1116         CFI_ENDPROC
1117 END(stack_segment)
1118
1119 KPROBE_ENTRY(general_protection)
1120         errorentry do_general_protection
1121 KPROBE_END(general_protection)
1122
1123 ENTRY(alignment_check)
1124         errorentry do_alignment_check
1125 END(alignment_check)
1126
1127 ENTRY(divide_error)
1128         zeroentry do_divide_error
1129 END(divide_error)
1130
1131 ENTRY(spurious_interrupt_bug)
1132         zeroentry do_spurious_interrupt_bug
1133 END(spurious_interrupt_bug)
1134
1135 #ifdef CONFIG_X86_MCE
1136         /* runs on exception stack */
1137 ENTRY(machine_check)
1138         INTR_FRAME
1139         pushq $0
1140         CFI_ADJUST_CFA_OFFSET 8
1141         paranoidentry do_machine_check
1142         jmp paranoid_exit1
1143         CFI_ENDPROC
1144 END(machine_check)
1145 #endif
1146
1147 /* Call softirq on interrupt stack. Interrupts are off. */
1148 ENTRY(call_softirq)
1149         CFI_STARTPROC
1150         push %rbp
1151         CFI_ADJUST_CFA_OFFSET   8
1152         CFI_REL_OFFSET rbp,0
1153         mov  %rsp,%rbp
1154         CFI_DEF_CFA_REGISTER rbp
1155         incl %gs:pda_irqcount
1156         cmove %gs:pda_irqstackptr,%rsp
1157         push  %rbp                      # backlink for old unwinder
1158         call __do_softirq
1159         leaveq
1160         CFI_DEF_CFA_REGISTER    rsp
1161         CFI_ADJUST_CFA_OFFSET   -8
1162         decl %gs:pda_irqcount
1163         ret
1164         CFI_ENDPROC
1165 ENDPROC(call_softirq)
1166
1167 KPROBE_ENTRY(ignore_sysret)
1168         CFI_STARTPROC
1169         mov $-ENOSYS,%eax
1170         sysret
1171         CFI_ENDPROC
1172 ENDPROC(ignore_sysret)