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)
 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         movq $-ENOSYS,%rcx
 330         cmova %rcx,%rax
 331         ja  1f
 332         movq %r10,%rcx  /* fixup for C */
 333         call *sys_call_table(,%rax,8)
 334 1:      movq %rax,RAX-ARGOFFSET(%rsp)
 335         /* Use IRET because user could have changed frame */
 336
 337 /*
 338  * Syscall return path ending with IRET.
 339  * Has correct top of stack, but partial stack frame.
 340  */
 341         .globl int_ret_from_sys_call
 342 int_ret_from_sys_call:
 343         DISABLE_INTERRUPTS(CLBR_NONE)
 344         TRACE_IRQS_OFF
 345         testl $3,CS-ARGOFFSET(%rsp)
 346         je retint_restore_args
 347         movl $_TIF_ALLWORK_MASK,%edi
 348         /* edi: mask to check */
 349 int_with_check:
 350         LOCKDEP_SYS_EXIT_IRQ
 351         GET_THREAD_INFO(%rcx)
 352         movl threadinfo_flags(%rcx),%edx
 353         andl %edi,%edx
 354         jnz   int_careful
 355         andl    $~TS_COMPAT,threadinfo_status(%rcx)
 356         jmp   retint_swapgs
 357
 358         /* Either reschedule or signal or syscall exit tracking needed. */
 359         /* First do a reschedule test. */
 360         /* edx: work, edi: workmask */
 361 int_careful:
 362         bt $TIF_NEED_RESCHED,%edx
 363         jnc  int_very_careful
 364         TRACE_IRQS_ON
 365         ENABLE_INTERRUPTS(CLBR_NONE)
 366         pushq %rdi
 367         CFI_ADJUST_CFA_OFFSET 8
 368         call schedule
 369         popq %rdi
 370         CFI_ADJUST_CFA_OFFSET -8
 371         DISABLE_INTERRUPTS(CLBR_NONE)
 372         TRACE_IRQS_OFF
 373         jmp int_with_check
 374
 375         /* handle signals and tracing -- both require a full stack frame */
 376 int_very_careful:
 377         TRACE_IRQS_ON
 378         ENABLE_INTERRUPTS(CLBR_NONE)
 379         SAVE_REST
 380         /* Check for syscall exit trace */
 381         testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx
 382         jz int_signal
 383         pushq %rdi
 384         CFI_ADJUST_CFA_OFFSET 8
 385         leaq 8(%rsp),%rdi       # &ptregs -> arg1
 386         call syscall_trace_leave
 387         popq %rdi
 388         CFI_ADJUST_CFA_OFFSET -8
 389         andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi
 390         jmp int_restore_rest
 391
 392 int_signal:
 393         testl $_TIF_DO_NOTIFY_MASK,%edx
 394         jz 1f
 395         movq %rsp,%rdi          # &ptregs -> arg1
 396         xorl %esi,%esi          # oldset -> arg2
 397         call do_notify_resume
 398 1:      movl $_TIF_NEED_RESCHED,%edi
 399 int_restore_rest:
 400         RESTORE_REST
 401         DISABLE_INTERRUPTS(CLBR_NONE)
 402         TRACE_IRQS_OFF
 403         jmp int_with_check
 404         CFI_ENDPROC
 405 END(system_call)
 406
 407 /*
 408  * Certain special system calls that need to save a complete full stack frame.
 409  */
 410
 411         .macro PTREGSCALL label,func,arg
 412         .globl \label
 413 \label:
 414         leaq    \func(%rip),%rax
 415         leaq    -ARGOFFSET+8(%rsp),\arg /* 8 for return address */
 416         jmp     ptregscall_common
 417 END(\label)
 418         .endm
 419
 420         CFI_STARTPROC
 421
 422         PTREGSCALL stub_clone, sys_clone, %r8
 423         PTREGSCALL stub_fork, sys_fork, %rdi
 424         PTREGSCALL stub_vfork, sys_vfork, %rdi
 425         PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx
 426         PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 427         PTREGSCALL stub_iopl, sys_iopl, %rsi
 428
 429 ENTRY(ptregscall_common)
 430         popq %r11
 431         CFI_ADJUST_CFA_OFFSET -8
 432         CFI_REGISTER rip, r11
 433         SAVE_REST
 434         movq %r11, %r15
 435         CFI_REGISTER rip, r15
 436         FIXUP_TOP_OF_STACK %r11
 437         call *%rax
 438         RESTORE_TOP_OF_STACK %r11
 439         movq %r15, %r11
 440         CFI_REGISTER rip, r11
 441         RESTORE_REST
 442         pushq %r11
 443         CFI_ADJUST_CFA_OFFSET 8
 444         CFI_REL_OFFSET rip, 0
 445         ret
 446         CFI_ENDPROC
 447 END(ptregscall_common)
 448
 449 ENTRY(stub_execve)
 450         CFI_STARTPROC
 451         popq %r11
 452         CFI_ADJUST_CFA_OFFSET -8
 453         CFI_REGISTER rip, r11
 454         SAVE_REST
 455         FIXUP_TOP_OF_STACK %r11
 456         movq %rsp, %rcx
 457         call sys_execve
 458         RESTORE_TOP_OF_STACK %r11
 459         movq %rax,RAX(%rsp)
 460         RESTORE_REST
 461         jmp int_ret_from_sys_call
 462         CFI_ENDPROC
 463 END(stub_execve)
 464
 465 /*
 466  * sigreturn is special because it needs to restore all registers on return.
 467  * This cannot be done with SYSRET, so use the IRET return path instead.
 468  */
 469 ENTRY(stub_rt_sigreturn)
 470         CFI_STARTPROC
 471         addq $8, %rsp
 472         CFI_ADJUST_CFA_OFFSET   -8
 473         SAVE_REST
 474         movq %rsp,%rdi
 475         FIXUP_TOP_OF_STACK %r11
 476         call sys_rt_sigreturn
 477         movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 478         RESTORE_REST
 479         jmp int_ret_from_sys_call
 480         CFI_ENDPROC
 481 END(stub_rt_sigreturn)
 482
 483 /*
 484  * initial frame state for interrupts and exceptions
 485  */
 486         .macro _frame ref
 487         CFI_STARTPROC simple
 488         CFI_SIGNAL_FRAME
 489         CFI_DEF_CFA rsp,SS+8-\ref
 490         /*CFI_REL_OFFSET ss,SS-\ref*/
 491         CFI_REL_OFFSET rsp,RSP-\ref
 492         /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/
 493         /*CFI_REL_OFFSET cs,CS-\ref*/
 494         CFI_REL_OFFSET rip,RIP-\ref
 495         .endm
 496
 497 /* initial frame state for interrupts (and exceptions without error code) */
 498 #define INTR_FRAME _frame RIP
 499 /* initial frame state for exceptions with error code (and interrupts with
 500    vector already pushed) */
 501 #define XCPT_FRAME _frame ORIG_RAX
 502
 503 /*
 504  * Interrupt entry/exit.
 505  *
 506  * Interrupt entry points save only callee clobbered registers in fast path.
 507  *
 508  * Entry runs with interrupts off.
 509  */
 510
 511 /* 0(%rsp): interrupt number */
 512         .macro interrupt func
 513         cld
 514         SAVE_ARGS
 515         leaq -ARGOFFSET(%rsp),%rdi      # arg1 for handler
 516         pushq %rbp
 517         CFI_ADJUST_CFA_OFFSET   8
 518         CFI_REL_OFFSET          rbp, 0
 519         movq %rsp,%rbp
 520         CFI_DEF_CFA_REGISTER    rbp
 521         testl $3,CS(%rdi)
 522         je 1f
 523         SWAPGS
 524         /* irqcount is used to check if a CPU is already on an interrupt
 525            stack or not. While this is essentially redundant with preempt_count
 526            it is a little cheaper to use a separate counter in the PDA
 527            (short of moving irq_enter into assembly, which would be too
 528             much work) */
 529 1:      incl    %gs:pda_irqcount
 530         cmoveq %gs:pda_irqstackptr,%rsp
 531         push    %rbp                    # backlink for old unwinder
 532         /*
 533          * We entered an interrupt context - irqs are off:
 534          */
 535         TRACE_IRQS_OFF
 536         call \func
 537         .endm
 538
 539 ENTRY(common_interrupt)
 540         XCPT_FRAME
 541         interrupt do_IRQ
 542         /* 0(%rsp): oldrsp-ARGOFFSET */
 543 ret_from_intr:
 544         DISABLE_INTERRUPTS(CLBR_NONE)
 545         TRACE_IRQS_OFF
 546         decl %gs:pda_irqcount
 547         leaveq
 548         CFI_DEF_CFA_REGISTER    rsp
 549         CFI_ADJUST_CFA_OFFSET   -8
 550 exit_intr:
 551         GET_THREAD_INFO(%rcx)
 552         testl $3,CS-ARGOFFSET(%rsp)
 553         je retint_kernel
 554
 555         /* Interrupt came from user space */
 556         /*
 557          * Has a correct top of stack, but a partial stack frame
 558          * %rcx: thread info. Interrupts off.
 559          */
 560 retint_with_reschedule:
 561         movl $_TIF_WORK_MASK,%edi
 562 retint_check:
 563         LOCKDEP_SYS_EXIT_IRQ
 564         movl threadinfo_flags(%rcx),%edx
 565         andl %edi,%edx
 566         CFI_REMEMBER_STATE
 567         jnz  retint_careful
 568
 569 retint_swapgs:          /* return to user-space */
 570         /*
 571          * The iretq could re-enable interrupts:
 572          */
 573         DISABLE_INTERRUPTS(CLBR_ANY)
 574         TRACE_IRQS_IRETQ
 575         SWAPGS
 576         jmp restore_args
 577
 578 retint_restore_args:    /* return to kernel space */
 579         DISABLE_INTERRUPTS(CLBR_ANY)
 580         /*
 581          * The iretq could re-enable interrupts:
 582          */
 583         TRACE_IRQS_IRETQ
 584 restore_args:
 585         RESTORE_ARGS 0,8,0
 586
 587 irq_return:
 588         INTERRUPT_RETURN
 589
 590         .section __ex_table, "a"
 591         .quad irq_return, bad_iret
 592         .previous
 593
 594 #ifdef CONFIG_PARAVIRT
 595 ENTRY(native_iret)
 596         iretq
 597
 598         .section __ex_table,"a"
 599         .quad native_iret, bad_iret
 600         .previous
 601 #endif
 602
 603         .section .fixup,"ax"
 604 bad_iret:
 605         /*
 606          * The iret traps when the %cs or %ss being restored is bogus.
 607          * We've lost the original trap vector and error code.
 608          * #GPF is the most likely one to get for an invalid selector.
 609          * So pretend we completed the iret and took the #GPF in user mode.
 610          *
 611          * We are now running with the kernel GS after exception recovery.
 612          * But error_entry expects us to have user GS to match the user %cs,
 613          * so swap back.
 614          */
 615         pushq $0
 616
 617         SWAPGS
 618         jmp general_protection
 619
 620         .previous
 621
 622         /* edi: workmask, edx: work */
 623 retint_careful:
 624         CFI_RESTORE_STATE
 625         bt    $TIF_NEED_RESCHED,%edx
 626         jnc   retint_signal
 627         TRACE_IRQS_ON
 628         ENABLE_INTERRUPTS(CLBR_NONE)
 629         pushq %rdi
 630         CFI_ADJUST_CFA_OFFSET   8
 631         call  schedule
 632         popq %rdi
 633         CFI_ADJUST_CFA_OFFSET   -8
 634         GET_THREAD_INFO(%rcx)
 635         DISABLE_INTERRUPTS(CLBR_NONE)
 636         TRACE_IRQS_OFF
 637         jmp retint_check
 638
 639 retint_signal:
 640         testl $_TIF_DO_NOTIFY_MASK,%edx
 641         jz    retint_swapgs
 642         TRACE_IRQS_ON
 643         ENABLE_INTERRUPTS(CLBR_NONE)
 644         SAVE_REST
 645         movq $-1,ORIG_RAX(%rsp)
 646         xorl %esi,%esi          # oldset
 647         movq %rsp,%rdi          # &pt_regs
 648         call do_notify_resume
 649         RESTORE_REST
 650         DISABLE_INTERRUPTS(CLBR_NONE)
 651         TRACE_IRQS_OFF
 652         movl $_TIF_NEED_RESCHED,%edi
 653         GET_THREAD_INFO(%rcx)
 654         jmp retint_check
 655
 656 #ifdef CONFIG_PREEMPT
 657         /* Returning to kernel space. Check if we need preemption */
 658         /* rcx:  threadinfo. interrupts off. */
 659 ENTRY(retint_kernel)
 660         cmpl $0,threadinfo_preempt_count(%rcx)
 661         jnz  retint_restore_args
 662         bt  $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
 663         jnc  retint_restore_args
 664         bt   $9,EFLAGS-ARGOFFSET(%rsp)  /* interrupts off? */
 665         jnc  retint_restore_args
 666         call preempt_schedule_irq
 667         jmp exit_intr
 668 #endif
 669
 670         CFI_ENDPROC
 671 END(common_interrupt)
 672
 673 /*
 674  * APIC interrupts.
 675  */
 676         .macro apicinterrupt num,func
 677         INTR_FRAME
 678         pushq $~(\num)
 679         CFI_ADJUST_CFA_OFFSET 8
 680         interrupt \func
 681         jmp ret_from_intr
 682         CFI_ENDPROC
 683         .endm
 684
 685 ENTRY(thermal_interrupt)
 686         apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
 687 END(thermal_interrupt)
 688
 689 ENTRY(threshold_interrupt)
 690         apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
 691 END(threshold_interrupt)
 692
 693 #ifdef CONFIG_SMP
 694 ENTRY(reschedule_interrupt)
 695         apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
 696 END(reschedule_interrupt)
 697
 698         .macro INVALIDATE_ENTRY num
 699 ENTRY(invalidate_interrupt\num)
 700         apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
 701 END(invalidate_interrupt\num)
 702         .endm
 703
 704         INVALIDATE_ENTRY 0
 705         INVALIDATE_ENTRY 1
 706         INVALIDATE_ENTRY 2
 707         INVALIDATE_ENTRY 3
 708         INVALIDATE_ENTRY 4
 709         INVALIDATE_ENTRY 5
 710         INVALIDATE_ENTRY 6
 711         INVALIDATE_ENTRY 7
 712
 713 ENTRY(call_function_interrupt)
 714         apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
 715 END(call_function_interrupt)
 716 ENTRY(irq_move_cleanup_interrupt)
 717         apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
 718 END(irq_move_cleanup_interrupt)
 719 #endif
 720
 721 ENTRY(apic_timer_interrupt)
 722         apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
 723 END(apic_timer_interrupt)
 724
 725 ENTRY(error_interrupt)
 726         apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
 727 END(error_interrupt)
 728
 729 ENTRY(spurious_interrupt)
 730         apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
 731 END(spurious_interrupt)
 732
 733 /*
 734  * Exception entry points.
 735  */
 736         .macro zeroentry sym
 737         INTR_FRAME
 738         pushq $0        /* push error code/oldrax */
 739         CFI_ADJUST_CFA_OFFSET 8
 740         pushq %rax      /* push real oldrax to the rdi slot */
 741         CFI_ADJUST_CFA_OFFSET 8
 742         CFI_REL_OFFSET rax,0
 743         leaq  \sym(%rip),%rax
 744         jmp error_entry
 745         CFI_ENDPROC
 746         .endm
 747
 748         .macro errorentry sym
 749         XCPT_FRAME
 750         pushq %rax
 751         CFI_ADJUST_CFA_OFFSET 8
 752         CFI_REL_OFFSET rax,0
 753         leaq  \sym(%rip),%rax
 754         jmp error_entry
 755         CFI_ENDPROC
 756         .endm
 757
 758         /* error code is on the stack already */
 759         /* handle NMI like exceptions that can happen everywhere */
 760         .macro paranoidentry sym, ist=0, irqtrace=1
 761         SAVE_ALL
 762         cld
 763         movl $1,%ebx
 764         movl  $MSR_GS_BASE,%ecx
 765         rdmsr
 766         testl %edx,%edx
 767         js    1f
 768         SWAPGS
 769         xorl  %ebx,%ebx
 770 1:
 771         .if \ist
 772         movq    %gs:pda_data_offset, %rbp
 773         .endif
 774         movq %rsp,%rdi
 775         movq ORIG_RAX(%rsp),%rsi
 776         movq $-1,ORIG_RAX(%rsp)
 777         .if \ist
 778         subq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 779         .endif
 780         call \sym
 781         .if \ist
 782         addq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 783         .endif
 784         DISABLE_INTERRUPTS(CLBR_NONE)
 785         .if \irqtrace
 786         TRACE_IRQS_OFF
 787         .endif
 788         .endm
 789
 790         /*
 791          * "Paranoid" exit path from exception stack.
 792          * Paranoid because this is used by NMIs and cannot take
 793          * any kernel state for granted.
 794          * We don't do kernel preemption checks here, because only
 795          * NMI should be common and it does not enable IRQs and
 796          * cannot get reschedule ticks.
 797          *
 798          * "trace" is 0 for the NMI handler only, because irq-tracing
 799          * is fundamentally NMI-unsafe. (we cannot change the soft and
 800          * hard flags at once, atomically)
 801          */
 802         .macro paranoidexit trace=1
 803         /* ebx: no swapgs flag */
 804 paranoid_exit\trace:
 805         testl %ebx,%ebx                         /* swapgs needed? */
 806         jnz paranoid_restore\trace
 807         testl $3,CS(%rsp)
 808         jnz   paranoid_userspace\trace
 809 paranoid_swapgs\trace:
 810         .if \trace
 811         TRACE_IRQS_IRETQ 0
 812         .endif
 813         SWAPGS_UNSAFE_STACK
 814 paranoid_restore\trace:
 815         RESTORE_ALL 8
 816         jmp irq_return
 817 paranoid_userspace\trace:
 818         GET_THREAD_INFO(%rcx)
 819         movl threadinfo_flags(%rcx),%ebx
 820         andl $_TIF_WORK_MASK,%ebx
 821         jz paranoid_swapgs\trace
 822         movq %rsp,%rdi                  /* &pt_regs */
 823         call sync_regs
 824         movq %rax,%rsp                  /* switch stack for scheduling */
 825         testl $_TIF_NEED_RESCHED,%ebx
 826         jnz paranoid_schedule\trace
 827         movl %ebx,%edx                  /* arg3: thread flags */
 828         .if \trace
 829         TRACE_IRQS_ON
 830         .endif
 831         ENABLE_INTERRUPTS(CLBR_NONE)
 832         xorl %esi,%esi                  /* arg2: oldset */
 833         movq %rsp,%rdi                  /* arg1: &pt_regs */
 834         call do_notify_resume
 835         DISABLE_INTERRUPTS(CLBR_NONE)
 836         .if \trace
 837         TRACE_IRQS_OFF
 838         .endif
 839         jmp paranoid_userspace\trace
 840 paranoid_schedule\trace:
 841         .if \trace
 842         TRACE_IRQS_ON
 843         .endif
 844         ENABLE_INTERRUPTS(CLBR_ANY)
 845         call schedule
 846         DISABLE_INTERRUPTS(CLBR_ANY)
 847         .if \trace
 848         TRACE_IRQS_OFF
 849         .endif
 850         jmp paranoid_userspace\trace
 851         CFI_ENDPROC
 852         .endm
 853
 854 /*
 855  * Exception entry point. This expects an error code/orig_rax on the stack
 856  * and the exception handler in %rax.
 857  */
 858 KPROBE_ENTRY(error_entry)
 859         _frame RDI
 860         CFI_REL_OFFSET rax,0
 861         /* rdi slot contains rax, oldrax contains error code */
 862         cld
 863         subq  $14*8,%rsp
 864         CFI_ADJUST_CFA_OFFSET   (14*8)
 865         movq %rsi,13*8(%rsp)
 866         CFI_REL_OFFSET  rsi,RSI
 867         movq 14*8(%rsp),%rsi    /* load rax from rdi slot */
 868         CFI_REGISTER    rax,rsi
 869         movq %rdx,12*8(%rsp)
 870         CFI_REL_OFFSET  rdx,RDX
 871         movq %rcx,11*8(%rsp)
 872         CFI_REL_OFFSET  rcx,RCX
 873         movq %rsi,10*8(%rsp)    /* store rax */
 874         CFI_REL_OFFSET  rax,RAX
 875         movq %r8, 9*8(%rsp)
 876         CFI_REL_OFFSET  r8,R8
 877         movq %r9, 8*8(%rsp)
 878         CFI_REL_OFFSET  r9,R9
 879         movq %r10,7*8(%rsp)
 880         CFI_REL_OFFSET  r10,R10
 881         movq %r11,6*8(%rsp)
 882         CFI_REL_OFFSET  r11,R11
 883         movq %rbx,5*8(%rsp)
 884         CFI_REL_OFFSET  rbx,RBX
 885         movq %rbp,4*8(%rsp)
 886         CFI_REL_OFFSET  rbp,RBP
 887         movq %r12,3*8(%rsp)
 888         CFI_REL_OFFSET  r12,R12
 889         movq %r13,2*8(%rsp)
 890         CFI_REL_OFFSET  r13,R13
 891         movq %r14,1*8(%rsp)
 892         CFI_REL_OFFSET  r14,R14
 893         movq %r15,(%rsp)
 894         CFI_REL_OFFSET  r15,R15
 895         xorl %ebx,%ebx
 896         testl $3,CS(%rsp)
 897         je  error_kernelspace
 898 error_swapgs:
 899         SWAPGS
 900 error_sti:
 901         movq %rdi,RDI(%rsp)
 902         CFI_REL_OFFSET  rdi,RDI
 903         movq %rsp,%rdi
 904         movq ORIG_RAX(%rsp),%rsi        /* get error code */
 905         movq $-1,ORIG_RAX(%rsp)
 906         call *%rax
 907         /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
 908 error_exit:
 909         movl %ebx,%eax
 910         RESTORE_REST
 911         DISABLE_INTERRUPTS(CLBR_NONE)
 912         TRACE_IRQS_OFF
 913         GET_THREAD_INFO(%rcx)
 914         testl %eax,%eax
 915         jne  retint_kernel
 916         LOCKDEP_SYS_EXIT_IRQ
 917         movl  threadinfo_flags(%rcx),%edx
 918         movl  $_TIF_WORK_MASK,%edi
 919         andl  %edi,%edx
 920         jnz  retint_careful
 921         jmp retint_swapgs
 922         CFI_ENDPROC
 923
 924 error_kernelspace:
 925         incl %ebx
 926        /* There are two places in the kernel that can potentially fault with
 927           usergs. Handle them here. The exception handlers after
 928            iret run with kernel gs again, so don't set the user space flag.
 929            B stepping K8s sometimes report an truncated RIP for IRET
 930            exceptions returning to compat mode. Check for these here too. */
 931         leaq irq_return(%rip),%rbp
 932         cmpq %rbp,RIP(%rsp)
 933         je   error_swapgs
 934         movl %ebp,%ebp  /* zero extend */
 935         cmpq %rbp,RIP(%rsp)
 936         je   error_swapgs
 937         cmpq $gs_change,RIP(%rsp)
 938         je   error_swapgs
 939         jmp  error_sti
 940 KPROBE_END(error_entry)
 941
 942        /* Reload gs selector with exception handling */
 943        /* edi:  new selector */
 944 ENTRY(load_gs_index)
 945         CFI_STARTPROC
 946         pushf
 947         CFI_ADJUST_CFA_OFFSET 8
 948         DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
 949         SWAPGS
 950 gs_change:
 951         movl %edi,%gs
 952 2:      mfence          /* workaround */
 953         SWAPGS
 954         popf
 955         CFI_ADJUST_CFA_OFFSET -8
 956         ret
 957         CFI_ENDPROC
 958 ENDPROC(load_gs_index)
 959
 960         .section __ex_table,"a"
 961         .align 8
 962         .quad gs_change,bad_gs
 963         .previous
 964         .section .fixup,"ax"
 965         /* running with kernelgs */
 966 bad_gs:
 967         SWAPGS                  /* switch back to user gs */
 968         xorl %eax,%eax
 969         movl %eax,%gs
 970         jmp  2b
 971         .previous
 972
 973 /*
 974  * Create a kernel thread.
 975  *
 976  * C extern interface:
 977  *      extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 978  *
 979  * asm input arguments:
 980  *      rdi: fn, rsi: arg, rdx: flags
 981  */
 982 ENTRY(kernel_thread)
 983         CFI_STARTPROC
 984         FAKE_STACK_FRAME $child_rip
 985         SAVE_ALL
 986
 987         # rdi: flags, rsi: usp, rdx: will be &pt_regs
 988         movq %rdx,%rdi
 989         orq  kernel_thread_flags(%rip),%rdi
 990         movq $-1, %rsi
 991         movq %rsp, %rdx
 992
 993         xorl %r8d,%r8d
 994         xorl %r9d,%r9d
 995
 996         # clone now
 997         call do_fork
 998         movq %rax,RAX(%rsp)
 999         xorl %edi,%edi
1000
1001         /*
1002          * It isn't worth to check for reschedule here,
1003          * so internally to the x86_64 port you can rely on kernel_thread()
1004          * not to reschedule the child before returning, this avoids the need
1005          * of hacks for example to fork off the per-CPU idle tasks.
1006          * [Hopefully no generic code relies on the reschedule -AK]
1007          */
1008         RESTORE_ALL
1009         UNFAKE_STACK_FRAME
1010         ret
1011         CFI_ENDPROC
1012 ENDPROC(kernel_thread)
1013
1014 child_rip:
1015         pushq $0                # fake return address
1016         CFI_STARTPROC
1017         /*
1018          * Here we are in the child and the registers are set as they were
1019          * at kernel_thread() invocation in the parent.
1020          */
1021         movq %rdi, %rax
1022         movq %rsi, %rdi
1023         call *%rax
1024         # exit
1025         mov %eax, %edi
1026         call do_exit
1027         CFI_ENDPROC
1028 ENDPROC(child_rip)
1029
1030 /*
1031  * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1032  *
1033  * C extern interface:
1034  *       extern long execve(char *name, char **argv, char **envp)
1035  *
1036  * asm input arguments:
1037  *      rdi: name, rsi: argv, rdx: envp
1038  *
1039  * We want to fallback into:
1040  *      extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1041  *
1042  * do_sys_execve asm fallback arguments:
1043  *      rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1044  */
1045 ENTRY(kernel_execve)
1046         CFI_STARTPROC
1047         FAKE_STACK_FRAME $0
1048         SAVE_ALL
1049         movq %rsp,%rcx
1050         call sys_execve
1051         movq %rax, RAX(%rsp)
1052         RESTORE_REST
1053         testq %rax,%rax
1054         je int_ret_from_sys_call
1055         RESTORE_ARGS
1056         UNFAKE_STACK_FRAME
1057         ret
1058         CFI_ENDPROC
1059 ENDPROC(kernel_execve)
1060
1061 KPROBE_ENTRY(page_fault)
1062         errorentry do_page_fault
1063 KPROBE_END(page_fault)
1064
1065 ENTRY(coprocessor_error)
1066         zeroentry do_coprocessor_error
1067 END(coprocessor_error)
1068
1069 ENTRY(simd_coprocessor_error)
1070         zeroentry do_simd_coprocessor_error
1071 END(simd_coprocessor_error)
1072
1073 ENTRY(device_not_available)
1074         zeroentry math_state_restore
1075 END(device_not_available)
1076
1077         /* runs on exception stack */
1078 KPROBE_ENTRY(debug)
1079         INTR_FRAME
1080         pushq $0
1081         CFI_ADJUST_CFA_OFFSET 8
1082         paranoidentry do_debug, DEBUG_STACK
1083         paranoidexit
1084 KPROBE_END(debug)
1085
1086         /* runs on exception stack */
1087 KPROBE_ENTRY(nmi)
1088         INTR_FRAME
1089         pushq $-1
1090         CFI_ADJUST_CFA_OFFSET 8
1091         paranoidentry do_nmi, 0, 0
1092 #ifdef CONFIG_TRACE_IRQFLAGS
1093         paranoidexit 0
1094 #else
1095         jmp paranoid_exit1
1096         CFI_ENDPROC
1097 #endif
1098 KPROBE_END(nmi)
1099
1100 KPROBE_ENTRY(int3)
1101         INTR_FRAME
1102         pushq $0
1103         CFI_ADJUST_CFA_OFFSET 8
1104         paranoidentry do_int3, DEBUG_STACK
1105         jmp paranoid_exit1
1106         CFI_ENDPROC
1107 KPROBE_END(int3)
1108
1109 ENTRY(overflow)
1110         zeroentry do_overflow
1111 END(overflow)
1112
1113 ENTRY(bounds)
1114         zeroentry do_bounds
1115 END(bounds)
1116
1117 ENTRY(invalid_op)
1118         zeroentry do_invalid_op
1119 END(invalid_op)
1120
1121 ENTRY(coprocessor_segment_overrun)
1122         zeroentry do_coprocessor_segment_overrun
1123 END(coprocessor_segment_overrun)
1124
1125 ENTRY(reserved)
1126         zeroentry do_reserved
1127 END(reserved)
1128
1129         /* runs on exception stack */
1130 ENTRY(double_fault)
1131         XCPT_FRAME
1132         paranoidentry do_double_fault
1133         jmp paranoid_exit1
1134         CFI_ENDPROC
1135 END(double_fault)
1136
1137 ENTRY(invalid_TSS)
1138         errorentry do_invalid_TSS
1139 END(invalid_TSS)
1140
1141 ENTRY(segment_not_present)
1142         errorentry do_segment_not_present
1143 END(segment_not_present)
1144
1145         /* runs on exception stack */
1146 ENTRY(stack_segment)
1147         XCPT_FRAME
1148         paranoidentry do_stack_segment
1149         jmp paranoid_exit1
1150         CFI_ENDPROC
1151 END(stack_segment)
1152
1153 KPROBE_ENTRY(general_protection)
1154         errorentry do_general_protection
1155 KPROBE_END(general_protection)
1156
1157 ENTRY(alignment_check)
1158         errorentry do_alignment_check
1159 END(alignment_check)
1160
1161 ENTRY(divide_error)
1162         zeroentry do_divide_error
1163 END(divide_error)
1164
1165 ENTRY(spurious_interrupt_bug)
1166         zeroentry do_spurious_interrupt_bug
1167 END(spurious_interrupt_bug)
1168
1169 #ifdef CONFIG_X86_MCE
1170         /* runs on exception stack */
1171 ENTRY(machine_check)
1172         INTR_FRAME
1173         pushq $0
1174         CFI_ADJUST_CFA_OFFSET 8
1175         paranoidentry do_machine_check
1176         jmp paranoid_exit1
1177         CFI_ENDPROC
1178 END(machine_check)
1179 #endif
1180
1181 /* Call softirq on interrupt stack. Interrupts are off. */
1182 ENTRY(call_softirq)
1183         CFI_STARTPROC
1184         push %rbp
1185         CFI_ADJUST_CFA_OFFSET   8
1186         CFI_REL_OFFSET rbp,0
1187         mov  %rsp,%rbp
1188         CFI_DEF_CFA_REGISTER rbp
1189         incl %gs:pda_irqcount
1190         cmove %gs:pda_irqstackptr,%rsp
1191         push  %rbp                      # backlink for old unwinder
1192         call __do_softirq
1193         leaveq
1194         CFI_DEF_CFA_REGISTER    rsp
1195         CFI_ADJUST_CFA_OFFSET   -8
1196         decl %gs:pda_irqcount
1197         ret
1198         CFI_ENDPROC
1199 ENDPROC(call_softirq)
1200
1201 KPROBE_ENTRY(ignore_sysret)
1202         CFI_STARTPROC
1203         mov $-ENOSYS,%eax
1204         sysret
1205         CFI_ENDPROC
1206 ENDPROC(ignore_sysret)