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         call sys_execve
 457         RESTORE_TOP_OF_STACK %r11
 458         movq %rax,RAX(%rsp)
 459         RESTORE_REST
 460         jmp int_ret_from_sys_call
 461         CFI_ENDPROC
 462 END(stub_execve)
 463
 464 /*
 465  * sigreturn is special because it needs to restore all registers on return.
 466  * This cannot be done with SYSRET, so use the IRET return path instead.
 467  */
 468 ENTRY(stub_rt_sigreturn)
 469         CFI_STARTPROC
 470         addq $8, %rsp
 471         CFI_ADJUST_CFA_OFFSET   -8
 472         SAVE_REST
 473         movq %rsp,%rdi
 474         FIXUP_TOP_OF_STACK %r11
 475         call sys_rt_sigreturn
 476         movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 477         RESTORE_REST
 478         jmp int_ret_from_sys_call
 479         CFI_ENDPROC
 480 END(stub_rt_sigreturn)
 481
 482 /*
 483  * initial frame state for interrupts and exceptions
 484  */
 485         .macro _frame ref
 486         CFI_STARTPROC simple
 487         CFI_SIGNAL_FRAME
 488         CFI_DEF_CFA rsp,SS+8-\ref
 489         /*CFI_REL_OFFSET ss,SS-\ref*/
 490         CFI_REL_OFFSET rsp,RSP-\ref
 491         /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/
 492         /*CFI_REL_OFFSET cs,CS-\ref*/
 493         CFI_REL_OFFSET rip,RIP-\ref
 494         .endm
 495
 496 /* initial frame state for interrupts (and exceptions without error code) */
 497 #define INTR_FRAME _frame RIP
 498 /* initial frame state for exceptions with error code (and interrupts with
 499    vector already pushed) */
 500 #define XCPT_FRAME _frame ORIG_RAX
 501
 502 /*
 503  * Interrupt entry/exit.
 504  *
 505  * Interrupt entry points save only callee clobbered registers in fast path.
 506  *
 507  * Entry runs with interrupts off.
 508  */
 509
 510 /* 0(%rsp): interrupt number */
 511         .macro interrupt func
 512         cld
 513         SAVE_ARGS
 514         leaq -ARGOFFSET(%rsp),%rdi      # arg1 for handler
 515         pushq %rbp
 516         CFI_ADJUST_CFA_OFFSET   8
 517         CFI_REL_OFFSET          rbp, 0
 518         movq %rsp,%rbp
 519         CFI_DEF_CFA_REGISTER    rbp
 520         testl $3,CS(%rdi)
 521         je 1f
 522         SWAPGS
 523         /* irqcount is used to check if a CPU is already on an interrupt
 524            stack or not. While this is essentially redundant with preempt_count
 525            it is a little cheaper to use a separate counter in the PDA
 526            (short of moving irq_enter into assembly, which would be too
 527             much work) */
 528 1:      incl    %gs:pda_irqcount
 529         cmoveq %gs:pda_irqstackptr,%rsp
 530         push    %rbp                    # backlink for old unwinder
 531         /*
 532          * We entered an interrupt context - irqs are off:
 533          */
 534         TRACE_IRQS_OFF
 535         call \func
 536         .endm
 537
 538 ENTRY(common_interrupt)
 539         XCPT_FRAME
 540         interrupt do_IRQ
 541         /* 0(%rsp): oldrsp-ARGOFFSET */
 542 ret_from_intr:
 543         DISABLE_INTERRUPTS(CLBR_NONE)
 544         TRACE_IRQS_OFF
 545         decl %gs:pda_irqcount
 546         leaveq
 547         CFI_DEF_CFA_REGISTER    rsp
 548         CFI_ADJUST_CFA_OFFSET   -8
 549 exit_intr:
 550         GET_THREAD_INFO(%rcx)
 551         testl $3,CS-ARGOFFSET(%rsp)
 552         je retint_kernel
 553
 554         /* Interrupt came from user space */
 555         /*
 556          * Has a correct top of stack, but a partial stack frame
 557          * %rcx: thread info. Interrupts off.
 558          */
 559 retint_with_reschedule:
 560         movl $_TIF_WORK_MASK,%edi
 561 retint_check:
 562         LOCKDEP_SYS_EXIT_IRQ
 563         movl threadinfo_flags(%rcx),%edx
 564         andl %edi,%edx
 565         CFI_REMEMBER_STATE
 566         jnz  retint_careful
 567
 568 retint_swapgs:          /* return to user-space */
 569         /*
 570          * The iretq could re-enable interrupts:
 571          */
 572         DISABLE_INTERRUPTS(CLBR_ANY)
 573         TRACE_IRQS_IRETQ
 574         SWAPGS
 575         jmp restore_args
 576
 577 retint_restore_args:    /* return to kernel space */
 578         DISABLE_INTERRUPTS(CLBR_ANY)
 579         /*
 580          * The iretq could re-enable interrupts:
 581          */
 582         TRACE_IRQS_IRETQ
 583 restore_args:
 584         RESTORE_ARGS 0,8,0
 585
 586 irq_return:
 587         INTERRUPT_RETURN
 588
 589         .section __ex_table, "a"
 590         .quad irq_return, bad_iret
 591         .previous
 592
 593 #ifdef CONFIG_PARAVIRT
 594 ENTRY(native_iret)
 595         iretq
 596
 597         .section __ex_table,"a"
 598         .quad native_iret, bad_iret
 599         .previous
 600 #endif
 601
 602         .section .fixup,"ax"
 603 bad_iret:
 604         /*
 605          * The iret traps when the %cs or %ss being restored is bogus.
 606          * We've lost the original trap vector and error code.
 607          * #GPF is the most likely one to get for an invalid selector.
 608          * So pretend we completed the iret and took the #GPF in user mode.
 609          *
 610          * We are now running with the kernel GS after exception recovery.
 611          * But error_entry expects us to have user GS to match the user %cs,
 612          * so swap back.
 613          */
 614         pushq $0
 615
 616         SWAPGS
 617         jmp general_protection
 618
 619         .previous
 620
 621         /* edi: workmask, edx: work */
 622 retint_careful:
 623         CFI_RESTORE_STATE
 624         bt    $TIF_NEED_RESCHED,%edx
 625         jnc   retint_signal
 626         TRACE_IRQS_ON
 627         ENABLE_INTERRUPTS(CLBR_NONE)
 628         pushq %rdi
 629         CFI_ADJUST_CFA_OFFSET   8
 630         call  schedule
 631         popq %rdi
 632         CFI_ADJUST_CFA_OFFSET   -8
 633         GET_THREAD_INFO(%rcx)
 634         DISABLE_INTERRUPTS(CLBR_NONE)
 635         TRACE_IRQS_OFF
 636         jmp retint_check
 637
 638 retint_signal:
 639         testl $_TIF_DO_NOTIFY_MASK,%edx
 640         jz    retint_swapgs
 641         TRACE_IRQS_ON
 642         ENABLE_INTERRUPTS(CLBR_NONE)
 643         SAVE_REST
 644         movq $-1,ORIG_RAX(%rsp)
 645         xorl %esi,%esi          # oldset
 646         movq %rsp,%rdi          # &pt_regs
 647         call do_notify_resume
 648         RESTORE_REST
 649         DISABLE_INTERRUPTS(CLBR_NONE)
 650         TRACE_IRQS_OFF
 651         movl $_TIF_NEED_RESCHED,%edi
 652         GET_THREAD_INFO(%rcx)
 653         jmp retint_check
 654
 655 #ifdef CONFIG_PREEMPT
 656         /* Returning to kernel space. Check if we need preemption */
 657         /* rcx:  threadinfo. interrupts off. */
 658 ENTRY(retint_kernel)
 659         cmpl $0,threadinfo_preempt_count(%rcx)
 660         jnz  retint_restore_args
 661         bt  $TIF_NEED_RESCHED,threadinfo_flags(%rcx)
 662         jnc  retint_restore_args
 663         bt   $9,EFLAGS-ARGOFFSET(%rsp)  /* interrupts off? */
 664         jnc  retint_restore_args
 665         call preempt_schedule_irq
 666         jmp exit_intr
 667 #endif
 668
 669         CFI_ENDPROC
 670 END(common_interrupt)
 671
 672 /*
 673  * APIC interrupts.
 674  */
 675         .macro apicinterrupt num,func
 676         INTR_FRAME
 677         pushq $~(\num)
 678         CFI_ADJUST_CFA_OFFSET 8
 679         interrupt \func
 680         jmp ret_from_intr
 681         CFI_ENDPROC
 682         .endm
 683
 684 ENTRY(thermal_interrupt)
 685         apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt
 686 END(thermal_interrupt)
 687
 688 ENTRY(threshold_interrupt)
 689         apicinterrupt THRESHOLD_APIC_VECTOR,mce_threshold_interrupt
 690 END(threshold_interrupt)
 691
 692 #ifdef CONFIG_SMP
 693 ENTRY(reschedule_interrupt)
 694         apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt
 695 END(reschedule_interrupt)
 696
 697         .macro INVALIDATE_ENTRY num
 698 ENTRY(invalidate_interrupt\num)
 699         apicinterrupt INVALIDATE_TLB_VECTOR_START+\num,smp_invalidate_interrupt
 700 END(invalidate_interrupt\num)
 701         .endm
 702
 703         INVALIDATE_ENTRY 0
 704         INVALIDATE_ENTRY 1
 705         INVALIDATE_ENTRY 2
 706         INVALIDATE_ENTRY 3
 707         INVALIDATE_ENTRY 4
 708         INVALIDATE_ENTRY 5
 709         INVALIDATE_ENTRY 6
 710         INVALIDATE_ENTRY 7
 711
 712 ENTRY(call_function_interrupt)
 713         apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt
 714 END(call_function_interrupt)
 715 ENTRY(irq_move_cleanup_interrupt)
 716         apicinterrupt IRQ_MOVE_CLEANUP_VECTOR,smp_irq_move_cleanup_interrupt
 717 END(irq_move_cleanup_interrupt)
 718 #endif
 719
 720 ENTRY(apic_timer_interrupt)
 721         apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt
 722 END(apic_timer_interrupt)
 723
 724 ENTRY(error_interrupt)
 725         apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt
 726 END(error_interrupt)
 727
 728 ENTRY(spurious_interrupt)
 729         apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt
 730 END(spurious_interrupt)
 731
 732 /*
 733  * Exception entry points.
 734  */
 735         .macro zeroentry sym
 736         INTR_FRAME
 737         pushq $0        /* push error code/oldrax */
 738         CFI_ADJUST_CFA_OFFSET 8
 739         pushq %rax      /* push real oldrax to the rdi slot */
 740         CFI_ADJUST_CFA_OFFSET 8
 741         CFI_REL_OFFSET rax,0
 742         leaq  \sym(%rip),%rax
 743         jmp error_entry
 744         CFI_ENDPROC
 745         .endm
 746
 747         .macro errorentry sym
 748         XCPT_FRAME
 749         pushq %rax
 750         CFI_ADJUST_CFA_OFFSET 8
 751         CFI_REL_OFFSET rax,0
 752         leaq  \sym(%rip),%rax
 753         jmp error_entry
 754         CFI_ENDPROC
 755         .endm
 756
 757         /* error code is on the stack already */
 758         /* handle NMI like exceptions that can happen everywhere */
 759         .macro paranoidentry sym, ist=0, irqtrace=1
 760         SAVE_ALL
 761         cld
 762         movl $1,%ebx
 763         movl  $MSR_GS_BASE,%ecx
 764         rdmsr
 765         testl %edx,%edx
 766         js    1f
 767         SWAPGS
 768         xorl  %ebx,%ebx
 769 1:
 770         .if \ist
 771         movq    %gs:pda_data_offset, %rbp
 772         .endif
 773         movq %rsp,%rdi
 774         movq ORIG_RAX(%rsp),%rsi
 775         movq $-1,ORIG_RAX(%rsp)
 776         .if \ist
 777         subq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 778         .endif
 779         call \sym
 780         .if \ist
 781         addq    $EXCEPTION_STKSZ, per_cpu__init_tss + TSS_ist + (\ist - 1) * 8(%rbp)
 782         .endif
 783         DISABLE_INTERRUPTS(CLBR_NONE)
 784         .if \irqtrace
 785         TRACE_IRQS_OFF
 786         .endif
 787         .endm
 788
 789         /*
 790          * "Paranoid" exit path from exception stack.
 791          * Paranoid because this is used by NMIs and cannot take
 792          * any kernel state for granted.
 793          * We don't do kernel preemption checks here, because only
 794          * NMI should be common and it does not enable IRQs and
 795          * cannot get reschedule ticks.
 796          *
 797          * "trace" is 0 for the NMI handler only, because irq-tracing
 798          * is fundamentally NMI-unsafe. (we cannot change the soft and
 799          * hard flags at once, atomically)
 800          */
 801         .macro paranoidexit trace=1
 802         /* ebx: no swapgs flag */
 803 paranoid_exit\trace:
 804         testl %ebx,%ebx                         /* swapgs needed? */
 805         jnz paranoid_restore\trace
 806         testl $3,CS(%rsp)
 807         jnz   paranoid_userspace\trace
 808 paranoid_swapgs\trace:
 809         .if \trace
 810         TRACE_IRQS_IRETQ 0
 811         .endif
 812         SWAPGS_UNSAFE_STACK
 813 paranoid_restore\trace:
 814         RESTORE_ALL 8
 815         jmp irq_return
 816 paranoid_userspace\trace:
 817         GET_THREAD_INFO(%rcx)
 818         movl threadinfo_flags(%rcx),%ebx
 819         andl $_TIF_WORK_MASK,%ebx
 820         jz paranoid_swapgs\trace
 821         movq %rsp,%rdi                  /* &pt_regs */
 822         call sync_regs
 823         movq %rax,%rsp                  /* switch stack for scheduling */
 824         testl $_TIF_NEED_RESCHED,%ebx
 825         jnz paranoid_schedule\trace
 826         movl %ebx,%edx                  /* arg3: thread flags */
 827         .if \trace
 828         TRACE_IRQS_ON
 829         .endif
 830         ENABLE_INTERRUPTS(CLBR_NONE)
 831         xorl %esi,%esi                  /* arg2: oldset */
 832         movq %rsp,%rdi                  /* arg1: &pt_regs */
 833         call do_notify_resume
 834         DISABLE_INTERRUPTS(CLBR_NONE)
 835         .if \trace
 836         TRACE_IRQS_OFF
 837         .endif
 838         jmp paranoid_userspace\trace
 839 paranoid_schedule\trace:
 840         .if \trace
 841         TRACE_IRQS_ON
 842         .endif
 843         ENABLE_INTERRUPTS(CLBR_ANY)
 844         call schedule
 845         DISABLE_INTERRUPTS(CLBR_ANY)
 846         .if \trace
 847         TRACE_IRQS_OFF
 848         .endif
 849         jmp paranoid_userspace\trace
 850         CFI_ENDPROC
 851         .endm
 852
 853 /*
 854  * Exception entry point. This expects an error code/orig_rax on the stack
 855  * and the exception handler in %rax.
 856  */
 857 KPROBE_ENTRY(error_entry)
 858         _frame RDI
 859         CFI_REL_OFFSET rax,0
 860         /* rdi slot contains rax, oldrax contains error code */
 861         cld
 862         subq  $14*8,%rsp
 863         CFI_ADJUST_CFA_OFFSET   (14*8)
 864         movq %rsi,13*8(%rsp)
 865         CFI_REL_OFFSET  rsi,RSI
 866         movq 14*8(%rsp),%rsi    /* load rax from rdi slot */
 867         CFI_REGISTER    rax,rsi
 868         movq %rdx,12*8(%rsp)
 869         CFI_REL_OFFSET  rdx,RDX
 870         movq %rcx,11*8(%rsp)
 871         CFI_REL_OFFSET  rcx,RCX
 872         movq %rsi,10*8(%rsp)    /* store rax */
 873         CFI_REL_OFFSET  rax,RAX
 874         movq %r8, 9*8(%rsp)
 875         CFI_REL_OFFSET  r8,R8
 876         movq %r9, 8*8(%rsp)
 877         CFI_REL_OFFSET  r9,R9
 878         movq %r10,7*8(%rsp)
 879         CFI_REL_OFFSET  r10,R10
 880         movq %r11,6*8(%rsp)
 881         CFI_REL_OFFSET  r11,R11
 882         movq %rbx,5*8(%rsp)
 883         CFI_REL_OFFSET  rbx,RBX
 884         movq %rbp,4*8(%rsp)
 885         CFI_REL_OFFSET  rbp,RBP
 886         movq %r12,3*8(%rsp)
 887         CFI_REL_OFFSET  r12,R12
 888         movq %r13,2*8(%rsp)
 889         CFI_REL_OFFSET  r13,R13
 890         movq %r14,1*8(%rsp)
 891         CFI_REL_OFFSET  r14,R14
 892         movq %r15,(%rsp)
 893         CFI_REL_OFFSET  r15,R15
 894         xorl %ebx,%ebx
 895         testl $3,CS(%rsp)
 896         je  error_kernelspace
 897 error_swapgs:
 898         SWAPGS
 899 error_sti:
 900         movq %rdi,RDI(%rsp)
 901         CFI_REL_OFFSET  rdi,RDI
 902         movq %rsp,%rdi
 903         movq ORIG_RAX(%rsp),%rsi        /* get error code */
 904         movq $-1,ORIG_RAX(%rsp)
 905         call *%rax
 906         /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
 907 error_exit:
 908         movl %ebx,%eax
 909         RESTORE_REST
 910         DISABLE_INTERRUPTS(CLBR_NONE)
 911         TRACE_IRQS_OFF
 912         GET_THREAD_INFO(%rcx)
 913         testl %eax,%eax
 914         jne  retint_kernel
 915         LOCKDEP_SYS_EXIT_IRQ
 916         movl  threadinfo_flags(%rcx),%edx
 917         movl  $_TIF_WORK_MASK,%edi
 918         andl  %edi,%edx
 919         jnz  retint_careful
 920         jmp retint_swapgs
 921         CFI_ENDPROC
 922
 923 error_kernelspace:
 924         incl %ebx
 925        /* There are two places in the kernel that can potentially fault with
 926           usergs. Handle them here. The exception handlers after
 927            iret run with kernel gs again, so don't set the user space flag.
 928            B stepping K8s sometimes report an truncated RIP for IRET
 929            exceptions returning to compat mode. Check for these here too. */
 930         leaq irq_return(%rip),%rbp
 931         cmpq %rbp,RIP(%rsp)
 932         je   error_swapgs
 933         movl %ebp,%ebp  /* zero extend */
 934         cmpq %rbp,RIP(%rsp)
 935         je   error_swapgs
 936         cmpq $gs_change,RIP(%rsp)
 937         je   error_swapgs
 938         jmp  error_sti
 939 KPROBE_END(error_entry)
 940
 941        /* Reload gs selector with exception handling */
 942        /* edi:  new selector */
 943 ENTRY(load_gs_index)
 944         CFI_STARTPROC
 945         pushf
 946         CFI_ADJUST_CFA_OFFSET 8
 947         DISABLE_INTERRUPTS(CLBR_ANY | ~(CLBR_RDI))
 948         SWAPGS
 949 gs_change:
 950         movl %edi,%gs
 951 2:      mfence          /* workaround */
 952         SWAPGS
 953         popf
 954         CFI_ADJUST_CFA_OFFSET -8
 955         ret
 956         CFI_ENDPROC
 957 ENDPROC(load_gs_index)
 958
 959         .section __ex_table,"a"
 960         .align 8
 961         .quad gs_change,bad_gs
 962         .previous
 963         .section .fixup,"ax"
 964         /* running with kernelgs */
 965 bad_gs:
 966         SWAPGS                  /* switch back to user gs */
 967         xorl %eax,%eax
 968         movl %eax,%gs
 969         jmp  2b
 970         .previous
 971
 972 /*
 973  * Create a kernel thread.
 974  *
 975  * C extern interface:
 976  *      extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 977  *
 978  * asm input arguments:
 979  *      rdi: fn, rsi: arg, rdx: flags
 980  */
 981 ENTRY(kernel_thread)
 982         CFI_STARTPROC
 983         FAKE_STACK_FRAME $child_rip
 984         SAVE_ALL
 985
 986         # rdi: flags, rsi: usp, rdx: will be &pt_regs
 987         movq %rdx,%rdi
 988         orq  kernel_thread_flags(%rip),%rdi
 989         movq $-1, %rsi
 990         movq %rsp, %rdx
 991
 992         xorl %r8d,%r8d
 993         xorl %r9d,%r9d
 994
 995         # clone now
 996         call do_fork
 997         movq %rax,RAX(%rsp)
 998         xorl %edi,%edi
 999
1000         /*
1001          * It isn't worth to check for reschedule here,
1002          * so internally to the x86_64 port you can rely on kernel_thread()
1003          * not to reschedule the child before returning, this avoids the need
1004          * of hacks for example to fork off the per-CPU idle tasks.
1005          * [Hopefully no generic code relies on the reschedule -AK]
1006          */
1007         RESTORE_ALL
1008         UNFAKE_STACK_FRAME
1009         ret
1010         CFI_ENDPROC
1011 ENDPROC(kernel_thread)
1012
1013 child_rip:
1014         pushq $0                # fake return address
1015         CFI_STARTPROC
1016         /*
1017          * Here we are in the child and the registers are set as they were
1018          * at kernel_thread() invocation in the parent.
1019          */
1020         movq %rdi, %rax
1021         movq %rsi, %rdi
1022         call *%rax
1023         # exit
1024         mov %eax, %edi
1025         call do_exit
1026         CFI_ENDPROC
1027 ENDPROC(child_rip)
1028
1029 /*
1030  * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1031  *
1032  * C extern interface:
1033  *       extern long execve(char *name, char **argv, char **envp)
1034  *
1035  * asm input arguments:
1036  *      rdi: name, rsi: argv, rdx: envp
1037  *
1038  * We want to fallback into:
1039  *      extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs)
1040  *
1041  * do_sys_execve asm fallback arguments:
1042  *      rdi: name, rsi: argv, rdx: envp, fake frame on the stack
1043  */
1044 ENTRY(kernel_execve)
1045         CFI_STARTPROC
1046         FAKE_STACK_FRAME $0
1047         SAVE_ALL
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)