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 #include <asm/ftrace.h>
  55 #include <asm/percpu.h>
  56
  57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
  58 #include <linux/elf-em.h>
  59 #define AUDIT_ARCH_X86_64       (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
  60 #define __AUDIT_ARCH_64BIT 0x80000000
  61 #define __AUDIT_ARCH_LE    0x40000000
  62
  63         .code64
  64 #ifdef CONFIG_FUNCTION_TRACER
  65 #ifdef CONFIG_DYNAMIC_FTRACE
  66 ENTRY(mcount)
  67         retq
  68 END(mcount)
  69
  70 ENTRY(ftrace_caller)
  71         cmpl $0, function_trace_stop
  72         jne  ftrace_stub
  73
  74         MCOUNT_SAVE_FRAME
  75
  76         movq 0x38(%rsp), %rdi
  77         movq 8(%rbp), %rsi
  78         subq $MCOUNT_INSN_SIZE, %rdi
  79
  80 .globl ftrace_call
  81 ftrace_call:
  82         call ftrace_stub
  83
  84         MCOUNT_RESTORE_FRAME
  85
  86 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
  87 .globl ftrace_graph_call
  88 ftrace_graph_call:
  89         jmp ftrace_stub
  90 #endif
  91
  92 .globl ftrace_stub
  93 ftrace_stub:
  94         retq
  95 END(ftrace_caller)
  96
  97 #else /* ! CONFIG_DYNAMIC_FTRACE */
  98 ENTRY(mcount)
  99         cmpl $0, function_trace_stop
 100         jne  ftrace_stub
 101
 102         cmpq $ftrace_stub, ftrace_trace_function
 103         jnz trace
 104
 105 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 106         cmpq $ftrace_stub, ftrace_graph_return
 107         jnz ftrace_graph_caller
 108
 109         cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
 110         jnz ftrace_graph_caller
 111 #endif
 112
 113 .globl ftrace_stub
 114 ftrace_stub:
 115         retq
 116
 117 trace:
 118         MCOUNT_SAVE_FRAME
 119
 120         movq 0x38(%rsp), %rdi
 121         movq 8(%rbp), %rsi
 122         subq $MCOUNT_INSN_SIZE, %rdi
 123
 124         call   *ftrace_trace_function
 125
 126         MCOUNT_RESTORE_FRAME
 127
 128         jmp ftrace_stub
 129 END(mcount)
 130 #endif /* CONFIG_DYNAMIC_FTRACE */
 131 #endif /* CONFIG_FUNCTION_TRACER */
 132
 133 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 134 ENTRY(ftrace_graph_caller)
 135         cmpl $0, function_trace_stop
 136         jne ftrace_stub
 137
 138         MCOUNT_SAVE_FRAME
 139
 140         leaq 8(%rbp), %rdi
 141         movq 0x38(%rsp), %rsi
 142         subq $MCOUNT_INSN_SIZE, %rsi
 143
 144         call    prepare_ftrace_return
 145
 146         MCOUNT_RESTORE_FRAME
 147
 148         retq
 149 END(ftrace_graph_caller)
 150
 151
 152 .globl return_to_handler
 153 return_to_handler:
 154         subq  $80, %rsp
 155
 156         movq %rax, (%rsp)
 157         movq %rcx, 8(%rsp)
 158         movq %rdx, 16(%rsp)
 159         movq %rsi, 24(%rsp)
 160         movq %rdi, 32(%rsp)
 161         movq %r8, 40(%rsp)
 162         movq %r9, 48(%rsp)
 163         movq %r10, 56(%rsp)
 164         movq %r11, 64(%rsp)
 165
 166         call ftrace_return_to_handler
 167
 168         movq %rax, 72(%rsp)
 169         movq 64(%rsp), %r11
 170         movq 56(%rsp), %r10
 171         movq 48(%rsp), %r9
 172         movq 40(%rsp), %r8
 173         movq 32(%rsp), %rdi
 174         movq 24(%rsp), %rsi
 175         movq 16(%rsp), %rdx
 176         movq 8(%rsp), %rcx
 177         movq (%rsp), %rax
 178         addq $72, %rsp
 179         retq
 180 #endif
 181
 182
 183 #ifndef CONFIG_PREEMPT
 184 #define retint_kernel retint_restore_args
 185 #endif
 186
 187 #ifdef CONFIG_PARAVIRT
 188 ENTRY(native_usergs_sysret64)
 189         swapgs
 190         sysretq
 191 #endif /* CONFIG_PARAVIRT */
 192
 193
 194 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
 195 #ifdef CONFIG_TRACE_IRQFLAGS
 196         bt   $9,EFLAGS-\offset(%rsp)    /* interrupts off? */
 197         jnc  1f
 198         TRACE_IRQS_ON
 199 1:
 200 #endif
 201 .endm
 202
 203 /*
 204  * C code is not supposed to know about undefined top of stack. Every time
 205  * a C function with an pt_regs argument is called from the SYSCALL based
 206  * fast path FIXUP_TOP_OF_STACK is needed.
 207  * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
 208  * manipulation.
 209  */
 210
 211         /* %rsp:at FRAMEEND */
 212         .macro FIXUP_TOP_OF_STACK tmp offset=0
 213         movq PER_CPU_VAR(old_rsp),\tmp
 214         movq \tmp,RSP+\offset(%rsp)
 215         movq $__USER_DS,SS+\offset(%rsp)
 216         movq $__USER_CS,CS+\offset(%rsp)
 217         movq $-1,RCX+\offset(%rsp)
 218         movq R11+\offset(%rsp),\tmp  /* get eflags */
 219         movq \tmp,EFLAGS+\offset(%rsp)
 220         .endm
 221
 222         .macro RESTORE_TOP_OF_STACK tmp offset=0
 223         movq RSP+\offset(%rsp),\tmp
 224         movq \tmp,PER_CPU_VAR(old_rsp)
 225         movq EFLAGS+\offset(%rsp),\tmp
 226         movq \tmp,R11+\offset(%rsp)
 227         .endm
 228
 229         .macro FAKE_STACK_FRAME child_rip
 230         /* push in order ss, rsp, eflags, cs, rip */
 231         xorl %eax, %eax
 232         pushq $__KERNEL_DS /* ss */
 233         CFI_ADJUST_CFA_OFFSET   8
 234         /*CFI_REL_OFFSET        ss,0*/
 235         pushq %rax /* rsp */
 236         CFI_ADJUST_CFA_OFFSET   8
 237         CFI_REL_OFFSET  rsp,0
 238         pushq $X86_EFLAGS_IF /* eflags - interrupts on */
 239         CFI_ADJUST_CFA_OFFSET   8
 240         /*CFI_REL_OFFSET        rflags,0*/
 241         pushq $__KERNEL_CS /* cs */
 242         CFI_ADJUST_CFA_OFFSET   8
 243         /*CFI_REL_OFFSET        cs,0*/
 244         pushq \child_rip /* rip */
 245         CFI_ADJUST_CFA_OFFSET   8
 246         CFI_REL_OFFSET  rip,0
 247         pushq   %rax /* orig rax */
 248         CFI_ADJUST_CFA_OFFSET   8
 249         .endm
 250
 251         .macro UNFAKE_STACK_FRAME
 252         addq $8*6, %rsp
 253         CFI_ADJUST_CFA_OFFSET   -(6*8)
 254         .endm
 255
 256 /*
 257  * initial frame state for interrupts (and exceptions without error code)
 258  */
 259         .macro EMPTY_FRAME start=1 offset=0
 260         .if \start
 261         CFI_STARTPROC simple
 262         CFI_SIGNAL_FRAME
 263         CFI_DEF_CFA rsp,8+\offset
 264         .else
 265         CFI_DEF_CFA_OFFSET 8+\offset
 266         .endif
 267         .endm
 268
 269 /*
 270  * initial frame state for interrupts (and exceptions without error code)
 271  */
 272         .macro INTR_FRAME start=1 offset=0
 273         EMPTY_FRAME \start, SS+8+\offset-RIP
 274         /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
 275         CFI_REL_OFFSET rsp, RSP+\offset-RIP
 276         /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
 277         /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
 278         CFI_REL_OFFSET rip, RIP+\offset-RIP
 279         .endm
 280
 281 /*
 282  * initial frame state for exceptions with error code (and interrupts
 283  * with vector already pushed)
 284  */
 285         .macro XCPT_FRAME start=1 offset=0
 286         INTR_FRAME \start, RIP+\offset-ORIG_RAX
 287         /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
 288         .endm
 289
 290 /*
 291  * frame that enables calling into C.
 292  */
 293         .macro PARTIAL_FRAME start=1 offset=0
 294         XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
 295         CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
 296         CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
 297         CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
 298         CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
 299         CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
 300         CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
 301         CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
 302         CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
 303         CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
 304         .endm
 305
 306 /*
 307  * frame that enables passing a complete pt_regs to a C function.
 308  */
 309         .macro DEFAULT_FRAME start=1 offset=0
 310         PARTIAL_FRAME \start, R11+\offset-R15
 311         CFI_REL_OFFSET rbx, RBX+\offset
 312         CFI_REL_OFFSET rbp, RBP+\offset
 313         CFI_REL_OFFSET r12, R12+\offset
 314         CFI_REL_OFFSET r13, R13+\offset
 315         CFI_REL_OFFSET r14, R14+\offset
 316         CFI_REL_OFFSET r15, R15+\offset
 317         .endm
 318
 319 /* save partial stack frame */
 320 ENTRY(save_args)
 321         XCPT_FRAME
 322         cld
 323         movq_cfi rdi, RDI+16-ARGOFFSET
 324         movq_cfi rsi, RSI+16-ARGOFFSET
 325         movq_cfi rdx, RDX+16-ARGOFFSET
 326         movq_cfi rcx, RCX+16-ARGOFFSET
 327         movq_cfi rax, RAX+16-ARGOFFSET
 328         movq_cfi  r8,  R8+16-ARGOFFSET
 329         movq_cfi  r9,  R9+16-ARGOFFSET
 330         movq_cfi r10, R10+16-ARGOFFSET
 331         movq_cfi r11, R11+16-ARGOFFSET
 332
 333         leaq -ARGOFFSET+16(%rsp),%rdi   /* arg1 for handler */
 334         movq_cfi rbp, 8         /* push %rbp */
 335         leaq 8(%rsp), %rbp              /* mov %rsp, %ebp */
 336         testl $3, CS(%rdi)
 337         je 1f
 338         SWAPGS
 339         /*
 340          * irq_count is used to check if a CPU is already on an interrupt stack
 341          * or not. While this is essentially redundant with preempt_count it is
 342          * a little cheaper to use a separate counter in the PDA (short of
 343          * moving irq_enter into assembly, which would be too much work)
 344          */
 345 1:      incl PER_CPU_VAR(irq_count)
 346         jne 2f
 347         popq_cfi %rax                   /* move return address... */
 348         mov PER_CPU_VAR(irq_stack_ptr),%rsp
 349         EMPTY_FRAME 0
 350         pushq_cfi %rbp                  /* backlink for unwinder */
 351         pushq_cfi %rax                  /* ... to the new stack */
 352         /*
 353          * We entered an interrupt context - irqs are off:
 354          */
 355 2:      TRACE_IRQS_OFF
 356         ret
 357         CFI_ENDPROC
 358 END(save_args)
 359
 360 ENTRY(save_rest)
 361         PARTIAL_FRAME 1 REST_SKIP+8
 362         movq 5*8+16(%rsp), %r11 /* save return address */
 363         movq_cfi rbx, RBX+16
 364         movq_cfi rbp, RBP+16
 365         movq_cfi r12, R12+16
 366         movq_cfi r13, R13+16
 367         movq_cfi r14, R14+16
 368         movq_cfi r15, R15+16
 369         movq %r11, 8(%rsp)      /* return address */
 370         FIXUP_TOP_OF_STACK %r11, 16
 371         ret
 372         CFI_ENDPROC
 373 END(save_rest)
 374
 375 /* save complete stack frame */
 376 ENTRY(save_paranoid)
 377         XCPT_FRAME 1 RDI+8
 378         cld
 379         movq_cfi rdi, RDI+8
 380         movq_cfi rsi, RSI+8
 381         movq_cfi rdx, RDX+8
 382         movq_cfi rcx, RCX+8
 383         movq_cfi rax, RAX+8
 384         movq_cfi r8, R8+8
 385         movq_cfi r9, R9+8
 386         movq_cfi r10, R10+8
 387         movq_cfi r11, R11+8
 388         movq_cfi rbx, RBX+8
 389         movq_cfi rbp, RBP+8
 390         movq_cfi r12, R12+8
 391         movq_cfi r13, R13+8
 392         movq_cfi r14, R14+8
 393         movq_cfi r15, R15+8
 394         movl $1,%ebx
 395         movl $MSR_GS_BASE,%ecx
 396         rdmsr
 397         testl %edx,%edx
 398         js 1f   /* negative -> in kernel */
 399         SWAPGS
 400         xorl %ebx,%ebx
 401 1:      ret
 402         CFI_ENDPROC
 403 END(save_paranoid)
 404
 405 /*
 406  * A newly forked process directly context switches into this address.
 407  *
 408  * rdi: prev task we switched from
 409  */
 410 ENTRY(ret_from_fork)
 411         DEFAULT_FRAME
 412
 413         LOCK ; btr $TIF_FORK,TI_flags(%r8)
 414
 415         push kernel_eflags(%rip)
 416         CFI_ADJUST_CFA_OFFSET 8
 417         popf                                    # reset kernel eflags
 418         CFI_ADJUST_CFA_OFFSET -8
 419
 420         call schedule_tail                      # rdi: 'prev' task parameter
 421
 422         GET_THREAD_INFO(%rcx)
 423
 424         CFI_REMEMBER_STATE
 425         RESTORE_REST
 426
 427         testl $3, CS-ARGOFFSET(%rsp)            # from kernel_thread?
 428         je   int_ret_from_sys_call
 429
 430         testl $_TIF_IA32, TI_flags(%rcx)        # 32-bit compat task needs IRET
 431         jnz  int_ret_from_sys_call
 432
 433         RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
 434         jmp ret_from_sys_call                   # go to the SYSRET fastpath
 435
 436         CFI_RESTORE_STATE
 437         CFI_ENDPROC
 438 END(ret_from_fork)
 439
 440 /*
 441  * System call entry. Upto 6 arguments in registers are supported.
 442  *
 443  * SYSCALL does not save anything on the stack and does not change the
 444  * stack pointer.
 445  */
 446
 447 /*
 448  * Register setup:
 449  * rax  system call number
 450  * rdi  arg0
 451  * rcx  return address for syscall/sysret, C arg3
 452  * rsi  arg1
 453  * rdx  arg2
 454  * r10  arg3    (--> moved to rcx for C)
 455  * r8   arg4
 456  * r9   arg5
 457  * r11  eflags for syscall/sysret, temporary for C
 458  * r12-r15,rbp,rbx saved by C code, not touched.
 459  *
 460  * Interrupts are off on entry.
 461  * Only called from user space.
 462  *
 463  * XXX  if we had a free scratch register we could save the RSP into the stack frame
 464  *      and report it properly in ps. Unfortunately we haven't.
 465  *
 466  * When user can change the frames always force IRET. That is because
 467  * it deals with uncanonical addresses better. SYSRET has trouble
 468  * with them due to bugs in both AMD and Intel CPUs.
 469  */
 470
 471 ENTRY(system_call)
 472         CFI_STARTPROC   simple
 473         CFI_SIGNAL_FRAME
 474         CFI_DEF_CFA     rsp,KERNEL_STACK_OFFSET
 475         CFI_REGISTER    rip,rcx
 476         /*CFI_REGISTER  rflags,r11*/
 477         SWAPGS_UNSAFE_STACK
 478         /*
 479          * A hypervisor implementation might want to use a label
 480          * after the swapgs, so that it can do the swapgs
 481          * for the guest and jump here on syscall.
 482          */
 483 ENTRY(system_call_after_swapgs)
 484
 485         movq    %rsp,PER_CPU_VAR(old_rsp)
 486         movq    PER_CPU_VAR(kernel_stack),%rsp
 487         /*
 488          * No need to follow this irqs off/on section - it's straight
 489          * and short:
 490          */
 491         ENABLE_INTERRUPTS(CLBR_NONE)
 492         SAVE_ARGS 8,1
 493         movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
 494         movq  %rcx,RIP-ARGOFFSET(%rsp)
 495         CFI_REL_OFFSET rip,RIP-ARGOFFSET
 496         GET_THREAD_INFO(%rcx)
 497         testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
 498         jnz tracesys
 499 system_call_fastpath:
 500         cmpq $__NR_syscall_max,%rax
 501         ja badsys
 502         movq %r10,%rcx
 503         call *sys_call_table(,%rax,8)  # XXX:    rip relative
 504         movq %rax,RAX-ARGOFFSET(%rsp)
 505 /*
 506  * Syscall return path ending with SYSRET (fast path)
 507  * Has incomplete stack frame and undefined top of stack.
 508  */
 509 ret_from_sys_call:
 510         movl $_TIF_ALLWORK_MASK,%edi
 511         /* edi: flagmask */
 512 sysret_check:
 513         LOCKDEP_SYS_EXIT
 514         GET_THREAD_INFO(%rcx)
 515         DISABLE_INTERRUPTS(CLBR_NONE)
 516         TRACE_IRQS_OFF
 517         movl TI_flags(%rcx),%edx
 518         andl %edi,%edx
 519         jnz  sysret_careful
 520         CFI_REMEMBER_STATE
 521         /*
 522          * sysretq will re-enable interrupts:
 523          */
 524         TRACE_IRQS_ON
 525         movq RIP-ARGOFFSET(%rsp),%rcx
 526         CFI_REGISTER    rip,rcx
 527         RESTORE_ARGS 0,-ARG_SKIP,1
 528         /*CFI_REGISTER  rflags,r11*/
 529         movq    PER_CPU_VAR(old_rsp), %rsp
 530         USERGS_SYSRET64
 531
 532         CFI_RESTORE_STATE
 533         /* Handle reschedules */
 534         /* edx: work, edi: workmask */
 535 sysret_careful:
 536         bt $TIF_NEED_RESCHED,%edx
 537         jnc sysret_signal
 538         TRACE_IRQS_ON
 539         ENABLE_INTERRUPTS(CLBR_NONE)
 540         pushq %rdi
 541         CFI_ADJUST_CFA_OFFSET 8
 542         call schedule
 543         popq  %rdi
 544         CFI_ADJUST_CFA_OFFSET -8
 545         jmp sysret_check
 546
 547         /* Handle a signal */
 548 sysret_signal:
 549         TRACE_IRQS_ON
 550         ENABLE_INTERRUPTS(CLBR_NONE)
 551 #ifdef CONFIG_AUDITSYSCALL
 552         bt $TIF_SYSCALL_AUDIT,%edx
 553         jc sysret_audit
 554 #endif
 555         /* edx: work flags (arg3) */
 556         leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1
 557         xorl %esi,%esi # oldset -> arg2
 558         SAVE_REST
 559         FIXUP_TOP_OF_STACK %r11
 560         call do_notify_resume
 561         RESTORE_TOP_OF_STACK %r11
 562         RESTORE_REST
 563         movl $_TIF_WORK_MASK,%edi
 564         /* Use IRET because user could have changed frame. This
 565            works because ptregscall_common has called FIXUP_TOP_OF_STACK. */
 566         DISABLE_INTERRUPTS(CLBR_NONE)
 567         TRACE_IRQS_OFF
 568         jmp int_with_check
 569
 570 badsys:
 571         movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
 572         jmp ret_from_sys_call
 573
 574 #ifdef CONFIG_AUDITSYSCALL
 575         /*
 576          * Fast path for syscall audit without full syscall trace.
 577          * We just call audit_syscall_entry() directly, and then
 578          * jump back to the normal fast path.
 579          */
 580 auditsys:
 581         movq %r10,%r9                   /* 6th arg: 4th syscall arg */
 582         movq %rdx,%r8                   /* 5th arg: 3rd syscall arg */
 583         movq %rsi,%rcx                  /* 4th arg: 2nd syscall arg */
 584         movq %rdi,%rdx                  /* 3rd arg: 1st syscall arg */
 585         movq %rax,%rsi                  /* 2nd arg: syscall number */
 586         movl $AUDIT_ARCH_X86_64,%edi    /* 1st arg: audit arch */
 587         call audit_syscall_entry
 588         LOAD_ARGS 0             /* reload call-clobbered registers */
 589         jmp system_call_fastpath
 590
 591         /*
 592          * Return fast path for syscall audit.  Call audit_syscall_exit()
 593          * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
 594          * masked off.
 595          */
 596 sysret_audit:
 597         movq %rax,%rsi          /* second arg, syscall return value */
 598         cmpq $0,%rax            /* is it < 0? */
 599         setl %al                /* 1 if so, 0 if not */
 600         movzbl %al,%edi         /* zero-extend that into %edi */
 601         inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
 602         call audit_syscall_exit
 603         movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
 604         jmp sysret_check
 605 #endif  /* CONFIG_AUDITSYSCALL */
 606
 607         /* Do syscall tracing */
 608 tracesys:
 609 #ifdef CONFIG_AUDITSYSCALL
 610         testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
 611         jz auditsys
 612 #endif
 613         SAVE_REST
 614         movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
 615         FIXUP_TOP_OF_STACK %rdi
 616         movq %rsp,%rdi
 617         call syscall_trace_enter
 618         /*
 619          * Reload arg registers from stack in case ptrace changed them.
 620          * We don't reload %rax because syscall_trace_enter() returned
 621          * the value it wants us to use in the table lookup.
 622          */
 623         LOAD_ARGS ARGOFFSET, 1
 624         RESTORE_REST
 625         cmpq $__NR_syscall_max,%rax
 626         ja   int_ret_from_sys_call      /* RAX(%rsp) set to -ENOSYS above */
 627         movq %r10,%rcx  /* fixup for C */
 628         call *sys_call_table(,%rax,8)
 629         movq %rax,RAX-ARGOFFSET(%rsp)
 630         /* Use IRET because user could have changed frame */
 631
 632 /*
 633  * Syscall return path ending with IRET.
 634  * Has correct top of stack, but partial stack frame.
 635  */
 636         .globl int_ret_from_sys_call
 637         .globl int_with_check
 638 int_ret_from_sys_call:
 639         DISABLE_INTERRUPTS(CLBR_NONE)
 640         TRACE_IRQS_OFF
 641         testl $3,CS-ARGOFFSET(%rsp)
 642         je retint_restore_args
 643         movl $_TIF_ALLWORK_MASK,%edi
 644         /* edi: mask to check */
 645 int_with_check:
 646         LOCKDEP_SYS_EXIT_IRQ
 647         GET_THREAD_INFO(%rcx)
 648         movl TI_flags(%rcx),%edx
 649         andl %edi,%edx
 650         jnz   int_careful
 651         andl    $~TS_COMPAT,TI_status(%rcx)
 652         jmp   retint_swapgs
 653
 654         /* Either reschedule or signal or syscall exit tracking needed. */
 655         /* First do a reschedule test. */
 656         /* edx: work, edi: workmask */
 657 int_careful:
 658         bt $TIF_NEED_RESCHED,%edx
 659         jnc  int_very_careful
 660         TRACE_IRQS_ON
 661         ENABLE_INTERRUPTS(CLBR_NONE)
 662         pushq %rdi
 663         CFI_ADJUST_CFA_OFFSET 8
 664         call schedule
 665         popq %rdi
 666         CFI_ADJUST_CFA_OFFSET -8
 667         DISABLE_INTERRUPTS(CLBR_NONE)
 668         TRACE_IRQS_OFF
 669         jmp int_with_check
 670
 671         /* handle signals and tracing -- both require a full stack frame */
 672 int_very_careful:
 673         TRACE_IRQS_ON
 674         ENABLE_INTERRUPTS(CLBR_NONE)
 675         SAVE_REST
 676         /* Check for syscall exit trace */
 677         testl $_TIF_WORK_SYSCALL_EXIT,%edx
 678         jz int_signal
 679         pushq %rdi
 680         CFI_ADJUST_CFA_OFFSET 8
 681         leaq 8(%rsp),%rdi       # &ptregs -> arg1
 682         call syscall_trace_leave
 683         popq %rdi
 684         CFI_ADJUST_CFA_OFFSET -8
 685         andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
 686         jmp int_restore_rest
 687
 688 int_signal:
 689         testl $_TIF_DO_NOTIFY_MASK,%edx
 690         jz 1f
 691         movq %rsp,%rdi          # &ptregs -> arg1
 692         xorl %esi,%esi          # oldset -> arg2
 693         call do_notify_resume
 694 1:      movl $_TIF_WORK_MASK,%edi
 695 int_restore_rest:
 696         RESTORE_REST
 697         DISABLE_INTERRUPTS(CLBR_NONE)
 698         TRACE_IRQS_OFF
 699         jmp int_with_check
 700         CFI_ENDPROC
 701 END(system_call)
 702
 703 /*
 704  * Certain special system calls that need to save a complete full stack frame.
 705  */
 706         .macro PTREGSCALL label,func,arg
 707 ENTRY(\label)
 708         PARTIAL_FRAME 1 8               /* offset 8: return address */
 709         subq $REST_SKIP, %rsp
 710         CFI_ADJUST_CFA_OFFSET REST_SKIP
 711         call save_rest
 712         DEFAULT_FRAME 0 8               /* offset 8: return address */
 713         leaq 8(%rsp), \arg      /* pt_regs pointer */
 714         call \func
 715         jmp ptregscall_common
 716         CFI_ENDPROC
 717 END(\label)
 718         .endm
 719
 720         PTREGSCALL stub_clone, sys_clone, %r8
 721         PTREGSCALL stub_fork, sys_fork, %rdi
 722         PTREGSCALL stub_vfork, sys_vfork, %rdi
 723         PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
 724         PTREGSCALL stub_iopl, sys_iopl, %rsi
 725
 726 ENTRY(ptregscall_common)
 727         DEFAULT_FRAME 1 8       /* offset 8: return address */
 728         RESTORE_TOP_OF_STACK %r11, 8
 729         movq_cfi_restore R15+8, r15
 730         movq_cfi_restore R14+8, r14
 731         movq_cfi_restore R13+8, r13
 732         movq_cfi_restore R12+8, r12
 733         movq_cfi_restore RBP+8, rbp
 734         movq_cfi_restore RBX+8, rbx
 735         ret $REST_SKIP          /* pop extended registers */
 736         CFI_ENDPROC
 737 END(ptregscall_common)
 738
 739 ENTRY(stub_execve)
 740         CFI_STARTPROC
 741         popq %r11
 742         CFI_ADJUST_CFA_OFFSET -8
 743         CFI_REGISTER rip, r11
 744         SAVE_REST
 745         FIXUP_TOP_OF_STACK %r11
 746         movq %rsp, %rcx
 747         call sys_execve
 748         RESTORE_TOP_OF_STACK %r11
 749         movq %rax,RAX(%rsp)
 750         RESTORE_REST
 751         jmp int_ret_from_sys_call
 752         CFI_ENDPROC
 753 END(stub_execve)
 754
 755 /*
 756  * sigreturn is special because it needs to restore all registers on return.
 757  * This cannot be done with SYSRET, so use the IRET return path instead.
 758  */
 759 ENTRY(stub_rt_sigreturn)
 760         CFI_STARTPROC
 761         addq $8, %rsp
 762         CFI_ADJUST_CFA_OFFSET   -8
 763         SAVE_REST
 764         movq %rsp,%rdi
 765         FIXUP_TOP_OF_STACK %r11
 766         call sys_rt_sigreturn
 767         movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
 768         RESTORE_REST
 769         jmp int_ret_from_sys_call
 770         CFI_ENDPROC
 771 END(stub_rt_sigreturn)
 772
 773 /*
 774  * Build the entry stubs and pointer table with some assembler magic.
 775  * We pack 7 stubs into a single 32-byte chunk, which will fit in a
 776  * single cache line on all modern x86 implementations.
 777  */
 778         .section .init.rodata,"a"
 779 ENTRY(interrupt)
 780         .text
 781         .p2align 5
 782         .p2align CONFIG_X86_L1_CACHE_SHIFT
 783 ENTRY(irq_entries_start)
 784         INTR_FRAME
 785 vector=FIRST_EXTERNAL_VECTOR
 786 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
 787         .balign 32
 788   .rept 7
 789     .if vector < NR_VECTORS
 790       .if vector <> FIRST_EXTERNAL_VECTOR
 791         CFI_ADJUST_CFA_OFFSET -8
 792       .endif
 793 1:      pushq $(~vector+0x80)   /* Note: always in signed byte range */
 794         CFI_ADJUST_CFA_OFFSET 8
 795       .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
 796         jmp 2f
 797       .endif
 798       .previous
 799         .quad 1b
 800       .text
 801 vector=vector+1
 802     .endif
 803   .endr
 804 2:      jmp common_interrupt
 805 .endr
 806         CFI_ENDPROC
 807 END(irq_entries_start)
 808
 809 .previous
 810 END(interrupt)
 811 .previous
 812
 813 /*
 814  * Interrupt entry/exit.
 815  *
 816  * Interrupt entry points save only callee clobbered registers in fast path.
 817  *
 818  * Entry runs with interrupts off.
 819  */
 820
 821 /* 0(%rsp): ~(interrupt number) */
 822         .macro interrupt func
 823         subq $10*8, %rsp
 824         CFI_ADJUST_CFA_OFFSET 10*8
 825         call save_args
 826         PARTIAL_FRAME 0
 827         call \func
 828         .endm
 829
 830         /*
 831          * The interrupt stubs push (~vector+0x80) onto the stack and
 832          * then jump to common_interrupt.
 833          */
 834         .p2align CONFIG_X86_L1_CACHE_SHIFT
 835 common_interrupt:
 836         XCPT_FRAME
 837         addq $-0x80,(%rsp)              /* Adjust vector to [-256,-1] range */
 838         interrupt do_IRQ
 839         /* 0(%rsp): old_rsp-ARGOFFSET */
 840 ret_from_intr:
 841         DISABLE_INTERRUPTS(CLBR_NONE)
 842         TRACE_IRQS_OFF
 843         decl PER_CPU_VAR(irq_count)
 844         leaveq
 845         CFI_DEF_CFA_REGISTER    rsp
 846         CFI_ADJUST_CFA_OFFSET   -8
 847 exit_intr:
 848         GET_THREAD_INFO(%rcx)
 849         testl $3,CS-ARGOFFSET(%rsp)
 850         je retint_kernel
 851
 852         /* Interrupt came from user space */
 853         /*
 854          * Has a correct top of stack, but a partial stack frame
 855          * %rcx: thread info. Interrupts off.
 856          */
 857 retint_with_reschedule:
 858         movl $_TIF_WORK_MASK,%edi
 859 retint_check:
 860         LOCKDEP_SYS_EXIT_IRQ
 861         movl TI_flags(%rcx),%edx
 862         andl %edi,%edx
 863         CFI_REMEMBER_STATE
 864         jnz  retint_careful
 865
 866 retint_swapgs:          /* return to user-space */
 867         /*
 868          * The iretq could re-enable interrupts:
 869          */
 870         DISABLE_INTERRUPTS(CLBR_ANY)
 871         TRACE_IRQS_IRETQ
 872         SWAPGS
 873         jmp restore_args
 874
 875 retint_restore_args:    /* return to kernel space */
 876         DISABLE_INTERRUPTS(CLBR_ANY)
 877         /*
 878          * The iretq could re-enable interrupts:
 879          */
 880         TRACE_IRQS_IRETQ
 881 restore_args:
 882         RESTORE_ARGS 0,8,0
 883
 884 irq_return:
 885         INTERRUPT_RETURN
 886
 887         .section __ex_table, "a"
 888         .quad irq_return, bad_iret
 889         .previous
 890
 891 #ifdef CONFIG_PARAVIRT
 892 ENTRY(native_iret)
 893         iretq
 894
 895         .section __ex_table,"a"
 896         .quad native_iret, bad_iret
 897         .previous
 898 #endif
 899
 900         .section .fixup,"ax"
 901 bad_iret:
 902         /*
 903          * The iret traps when the %cs or %ss being restored is bogus.
 904          * We've lost the original trap vector and error code.
 905          * #GPF is the most likely one to get for an invalid selector.
 906          * So pretend we completed the iret and took the #GPF in user mode.
 907          *
 908          * We are now running with the kernel GS after exception recovery.
 909          * But error_entry expects us to have user GS to match the user %cs,
 910          * so swap back.
 911          */
 912         pushq $0
 913
 914         SWAPGS
 915         jmp general_protection
 916
 917         .previous
 918
 919         /* edi: workmask, edx: work */
 920 retint_careful:
 921         CFI_RESTORE_STATE
 922         bt    $TIF_NEED_RESCHED,%edx
 923         jnc   retint_signal
 924         TRACE_IRQS_ON
 925         ENABLE_INTERRUPTS(CLBR_NONE)
 926         pushq %rdi
 927         CFI_ADJUST_CFA_OFFSET   8
 928         call  schedule
 929         popq %rdi
 930         CFI_ADJUST_CFA_OFFSET   -8
 931         GET_THREAD_INFO(%rcx)
 932         DISABLE_INTERRUPTS(CLBR_NONE)
 933         TRACE_IRQS_OFF
 934         jmp retint_check
 935
 936 retint_signal:
 937         testl $_TIF_DO_NOTIFY_MASK,%edx
 938         jz    retint_swapgs
 939         TRACE_IRQS_ON
 940         ENABLE_INTERRUPTS(CLBR_NONE)
 941         SAVE_REST
 942         movq $-1,ORIG_RAX(%rsp)
 943         xorl %esi,%esi          # oldset
 944         movq %rsp,%rdi          # &pt_regs
 945         call do_notify_resume
 946         RESTORE_REST
 947         DISABLE_INTERRUPTS(CLBR_NONE)
 948         TRACE_IRQS_OFF
 949         GET_THREAD_INFO(%rcx)
 950         jmp retint_with_reschedule
 951
 952 #ifdef CONFIG_PREEMPT
 953         /* Returning to kernel space. Check if we need preemption */
 954         /* rcx:  threadinfo. interrupts off. */
 955 ENTRY(retint_kernel)
 956         cmpl $0,TI_preempt_count(%rcx)
 957         jnz  retint_restore_args
 958         bt  $TIF_NEED_RESCHED,TI_flags(%rcx)
 959         jnc  retint_restore_args
 960         bt   $9,EFLAGS-ARGOFFSET(%rsp)  /* interrupts off? */
 961         jnc  retint_restore_args
 962         call preempt_schedule_irq
 963         jmp exit_intr
 964 #endif
 965
 966         CFI_ENDPROC
 967 END(common_interrupt)
 968
 969 /*
 970  * APIC interrupts.
 971  */
 972 .macro apicinterrupt num sym do_sym
 973 ENTRY(\sym)
 974         INTR_FRAME
 975         pushq $~(\num)
 976         CFI_ADJUST_CFA_OFFSET 8
 977         interrupt \do_sym
 978         jmp ret_from_intr
 979         CFI_ENDPROC
 980 END(\sym)
 981 .endm
 982
 983 #ifdef CONFIG_SMP
 984 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
 985         irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
 986 #endif
 987
 988 #ifdef CONFIG_X86_UV
 989 apicinterrupt UV_BAU_MESSAGE \
 990         uv_bau_message_intr1 uv_bau_message_interrupt
 991 #endif
 992 apicinterrupt LOCAL_TIMER_VECTOR \
 993         apic_timer_interrupt smp_apic_timer_interrupt
 994
 995 #ifdef CONFIG_SMP
 996 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
 997         invalidate_interrupt0 smp_invalidate_interrupt
 998 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
 999         invalidate_interrupt1 smp_invalidate_interrupt
1000 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
1001         invalidate_interrupt2 smp_invalidate_interrupt
1002 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
1003         invalidate_interrupt3 smp_invalidate_interrupt
1004 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
1005         invalidate_interrupt4 smp_invalidate_interrupt
1006 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
1007         invalidate_interrupt5 smp_invalidate_interrupt
1008 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
1009         invalidate_interrupt6 smp_invalidate_interrupt
1010 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
1011         invalidate_interrupt7 smp_invalidate_interrupt
1012 #endif
1013
1014 apicinterrupt THRESHOLD_APIC_VECTOR \
1015         threshold_interrupt mce_threshold_interrupt
1016 apicinterrupt THERMAL_APIC_VECTOR \
1017         thermal_interrupt smp_thermal_interrupt
1018
1019 #ifdef CONFIG_SMP
1020 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1021         call_function_single_interrupt smp_call_function_single_interrupt
1022 apicinterrupt CALL_FUNCTION_VECTOR \
1023         call_function_interrupt smp_call_function_interrupt
1024 apicinterrupt RESCHEDULE_VECTOR \
1025         reschedule_interrupt smp_reschedule_interrupt
1026 #endif
1027
1028 apicinterrupt ERROR_APIC_VECTOR \
1029         error_interrupt smp_error_interrupt
1030 apicinterrupt SPURIOUS_APIC_VECTOR \
1031         spurious_interrupt smp_spurious_interrupt
1032
1033 /*
1034  * Exception entry points.
1035  */
1036 .macro zeroentry sym do_sym
1037 ENTRY(\sym)
1038         INTR_FRAME
1039         PARAVIRT_ADJUST_EXCEPTION_FRAME
1040         pushq_cfi $-1           /* ORIG_RAX: no syscall to restart */
1041         subq $15*8,%rsp
1042         CFI_ADJUST_CFA_OFFSET 15*8
1043         call error_entry
1044         DEFAULT_FRAME 0
1045         movq %rsp,%rdi          /* pt_regs pointer */
1046         xorl %esi,%esi          /* no error code */
1047         call \do_sym
1048         jmp error_exit          /* %ebx: no swapgs flag */
1049         CFI_ENDPROC
1050 END(\sym)
1051 .endm
1052
1053 .macro paranoidzeroentry sym do_sym
1054 ENTRY(\sym)
1055         INTR_FRAME
1056         PARAVIRT_ADJUST_EXCEPTION_FRAME
1057         pushq $-1               /* ORIG_RAX: no syscall to restart */
1058         CFI_ADJUST_CFA_OFFSET 8
1059         subq $15*8, %rsp
1060         call save_paranoid
1061         TRACE_IRQS_OFF
1062         movq %rsp,%rdi          /* pt_regs pointer */
1063         xorl %esi,%esi          /* no error code */
1064         call \do_sym
1065         jmp paranoid_exit       /* %ebx: no swapgs flag */
1066         CFI_ENDPROC
1067 END(\sym)
1068 .endm
1069
1070 .macro paranoidzeroentry_ist sym do_sym ist
1071 ENTRY(\sym)
1072         INTR_FRAME
1073         PARAVIRT_ADJUST_EXCEPTION_FRAME
1074         pushq $-1               /* ORIG_RAX: no syscall to restart */
1075         CFI_ADJUST_CFA_OFFSET 8
1076         subq $15*8, %rsp
1077         call save_paranoid
1078         TRACE_IRQS_OFF
1079         movq %rsp,%rdi          /* pt_regs pointer */
1080         xorl %esi,%esi          /* no error code */
1081         PER_CPU(init_tss, %rbp)
1082         subq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1083         call \do_sym
1084         addq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1085         jmp paranoid_exit       /* %ebx: no swapgs flag */
1086         CFI_ENDPROC
1087 END(\sym)
1088 .endm
1089
1090 .macro errorentry sym do_sym
1091 ENTRY(\sym)
1092         XCPT_FRAME
1093         PARAVIRT_ADJUST_EXCEPTION_FRAME
1094         subq $15*8,%rsp
1095         CFI_ADJUST_CFA_OFFSET 15*8
1096         call error_entry
1097         DEFAULT_FRAME 0
1098         movq %rsp,%rdi                  /* pt_regs pointer */
1099         movq ORIG_RAX(%rsp),%rsi        /* get error code */
1100         movq $-1,ORIG_RAX(%rsp)         /* no syscall to restart */
1101         call \do_sym
1102         jmp error_exit                  /* %ebx: no swapgs flag */
1103         CFI_ENDPROC
1104 END(\sym)
1105 .endm
1106
1107         /* error code is on the stack already */
1108 .macro paranoiderrorentry sym do_sym
1109 ENTRY(\sym)
1110         XCPT_FRAME
1111         PARAVIRT_ADJUST_EXCEPTION_FRAME
1112         subq $15*8,%rsp
1113         CFI_ADJUST_CFA_OFFSET 15*8
1114         call save_paranoid
1115         DEFAULT_FRAME 0
1116         TRACE_IRQS_OFF
1117         movq %rsp,%rdi                  /* pt_regs pointer */
1118         movq ORIG_RAX(%rsp),%rsi        /* get error code */
1119         movq $-1,ORIG_RAX(%rsp)         /* no syscall to restart */
1120         call \do_sym
1121         jmp paranoid_exit               /* %ebx: no swapgs flag */
1122         CFI_ENDPROC
1123 END(\sym)
1124 .endm
1125
1126 zeroentry divide_error do_divide_error
1127 zeroentry overflow do_overflow
1128 zeroentry bounds do_bounds
1129 zeroentry invalid_op do_invalid_op
1130 zeroentry device_not_available do_device_not_available
1131 paranoiderrorentry double_fault do_double_fault
1132 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1133 errorentry invalid_TSS do_invalid_TSS
1134 errorentry segment_not_present do_segment_not_present
1135 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1136 zeroentry coprocessor_error do_coprocessor_error
1137 errorentry alignment_check do_alignment_check
1138 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1139
1140         /* Reload gs selector with exception handling */
1141         /* edi:  new selector */
1142 ENTRY(native_load_gs_index)
1143         CFI_STARTPROC
1144         pushf
1145         CFI_ADJUST_CFA_OFFSET 8
1146         DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1147         SWAPGS
1148 gs_change:
1149         movl %edi,%gs
1150 2:      mfence          /* workaround */
1151         SWAPGS
1152         popf
1153         CFI_ADJUST_CFA_OFFSET -8
1154         ret
1155         CFI_ENDPROC
1156 END(native_load_gs_index)
1157
1158         .section __ex_table,"a"
1159         .align 8
1160         .quad gs_change,bad_gs
1161         .previous
1162         .section .fixup,"ax"
1163         /* running with kernelgs */
1164 bad_gs:
1165         SWAPGS                  /* switch back to user gs */
1166         xorl %eax,%eax
1167         movl %eax,%gs
1168         jmp  2b
1169         .previous
1170
1171 /*
1172  * Create a kernel thread.
1173  *
1174  * C extern interface:
1175  *      extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1176  *
1177  * asm input arguments:
1178  *      rdi: fn, rsi: arg, rdx: flags
1179  */
1180 ENTRY(kernel_thread)
1181         CFI_STARTPROC
1182         FAKE_STACK_FRAME $child_rip
1183         SAVE_ALL
1184
1185         # rdi: flags, rsi: usp, rdx: will be &pt_regs
1186         movq %rdx,%rdi
1187         orq  kernel_thread_flags(%rip),%rdi
1188         movq $-1, %rsi
1189         movq %rsp, %rdx
1190
1191         xorl %r8d,%r8d
1192         xorl %r9d,%r9d
1193
1194         # clone now
1195         call do_fork
1196         movq %rax,RAX(%rsp)
1197         xorl %edi,%edi
1198
1199         /*
1200          * It isn't worth to check for reschedule here,
1201          * so internally to the x86_64 port you can rely on kernel_thread()
1202          * not to reschedule the child before returning, this avoids the need
1203          * of hacks for example to fork off the per-CPU idle tasks.
1204          * [Hopefully no generic code relies on the reschedule -AK]
1205          */
1206         RESTORE_ALL
1207         UNFAKE_STACK_FRAME
1208         ret
1209         CFI_ENDPROC
1210 END(kernel_thread)
1211
1212 ENTRY(child_rip)
1213         pushq $0                # fake return address
1214         CFI_STARTPROC
1215         /*
1216          * Here we are in the child and the registers are set as they were
1217          * at kernel_thread() invocation in the parent.
1218          */
1219         movq %rdi, %rax
1220         movq %rsi, %rdi
1221         call *%rax
1222         # exit
1223         mov %eax, %edi
1224         call do_exit
1225         ud2                     # padding for call trace
1226         CFI_ENDPROC
1227 END(child_rip)
1228
1229 /*
1230  * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1231  *
1232  * C extern interface:
1233  *       extern long execve(char *name, char **argv, char **envp)
1234  *
1235  * asm input arguments:
1236  *      rdi: name, rsi: argv, rdx: envp
1237  *
1238  * We want to fallback into:
1239  *      extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1240  *
1241  * do_sys_execve asm fallback arguments:
1242  *      rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1243  */
1244 ENTRY(kernel_execve)
1245         CFI_STARTPROC
1246         FAKE_STACK_FRAME $0
1247         SAVE_ALL
1248         movq %rsp,%rcx
1249         call sys_execve
1250         movq %rax, RAX(%rsp)
1251         RESTORE_REST
1252         testq %rax,%rax
1253         je int_ret_from_sys_call
1254         RESTORE_ARGS
1255         UNFAKE_STACK_FRAME
1256         ret
1257         CFI_ENDPROC
1258 END(kernel_execve)
1259
1260 /* Call softirq on interrupt stack. Interrupts are off. */
1261 ENTRY(call_softirq)
1262         CFI_STARTPROC
1263         push %rbp
1264         CFI_ADJUST_CFA_OFFSET   8
1265         CFI_REL_OFFSET rbp,0
1266         mov  %rsp,%rbp
1267         CFI_DEF_CFA_REGISTER rbp
1268         incl PER_CPU_VAR(irq_count)
1269         cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1270         push  %rbp                      # backlink for old unwinder
1271         call __do_softirq
1272         leaveq
1273         CFI_DEF_CFA_REGISTER    rsp
1274         CFI_ADJUST_CFA_OFFSET   -8
1275         decl PER_CPU_VAR(irq_count)
1276         ret
1277         CFI_ENDPROC
1278 END(call_softirq)
1279
1280 #ifdef CONFIG_XEN
1281 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1282
1283 /*
1284  * A note on the "critical region" in our callback handler.
1285  * We want to avoid stacking callback handlers due to events occurring
1286  * during handling of the last event. To do this, we keep events disabled
1287  * until we've done all processing. HOWEVER, we must enable events before
1288  * popping the stack frame (can't be done atomically) and so it would still
1289  * be possible to get enough handler activations to overflow the stack.
1290  * Although unlikely, bugs of that kind are hard to track down, so we'd
1291  * like to avoid the possibility.
1292  * So, on entry to the handler we detect whether we interrupted an
1293  * existing activation in its critical region -- if so, we pop the current
1294  * activation and restart the handler using the previous one.
1295  */
1296 ENTRY(xen_do_hypervisor_callback)   # do_hypervisor_callback(struct *pt_regs)
1297         CFI_STARTPROC
1298 /*
1299  * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1300  * see the correct pointer to the pt_regs
1301  */
1302         movq %rdi, %rsp            # we don't return, adjust the stack frame
1303         CFI_ENDPROC
1304         DEFAULT_FRAME
1305 11:     incl PER_CPU_VAR(irq_count)
1306         movq %rsp,%rbp
1307         CFI_DEF_CFA_REGISTER rbp
1308         cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1309         pushq %rbp                      # backlink for old unwinder
1310         call xen_evtchn_do_upcall
1311         popq %rsp
1312         CFI_DEF_CFA_REGISTER rsp
1313         decl PER_CPU_VAR(irq_count)
1314         jmp  error_exit
1315         CFI_ENDPROC
1316 END(do_hypervisor_callback)
1317
1318 /*
1319  * Hypervisor uses this for application faults while it executes.
1320  * We get here for two reasons:
1321  *  1. Fault while reloading DS, ES, FS or GS
1322  *  2. Fault while executing IRET
1323  * Category 1 we do not need to fix up as Xen has already reloaded all segment
1324  * registers that could be reloaded and zeroed the others.
1325  * Category 2 we fix up by killing the current process. We cannot use the
1326  * normal Linux return path in this case because if we use the IRET hypercall
1327  * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1328  * We distinguish between categories by comparing each saved segment register
1329  * with its current contents: any discrepancy means we in category 1.
1330  */
1331 ENTRY(xen_failsafe_callback)
1332         INTR_FRAME 1 (6*8)
1333         /*CFI_REL_OFFSET gs,GS*/
1334         /*CFI_REL_OFFSET fs,FS*/
1335         /*CFI_REL_OFFSET es,ES*/
1336         /*CFI_REL_OFFSET ds,DS*/
1337         CFI_REL_OFFSET r11,8
1338         CFI_REL_OFFSET rcx,0
1339         movw %ds,%cx
1340         cmpw %cx,0x10(%rsp)
1341         CFI_REMEMBER_STATE
1342         jne 1f
1343         movw %es,%cx
1344         cmpw %cx,0x18(%rsp)
1345         jne 1f
1346         movw %fs,%cx
1347         cmpw %cx,0x20(%rsp)
1348         jne 1f
1349         movw %gs,%cx
1350         cmpw %cx,0x28(%rsp)
1351         jne 1f
1352         /* All segments match their saved values => Category 2 (Bad IRET). */
1353         movq (%rsp),%rcx
1354         CFI_RESTORE rcx
1355         movq 8(%rsp),%r11
1356         CFI_RESTORE r11
1357         addq $0x30,%rsp
1358         CFI_ADJUST_CFA_OFFSET -0x30
1359         pushq_cfi $0    /* RIP */
1360         pushq_cfi %r11
1361         pushq_cfi %rcx
1362         jmp general_protection
1363         CFI_RESTORE_STATE
1364 1:      /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1365         movq (%rsp),%rcx
1366         CFI_RESTORE rcx
1367         movq 8(%rsp),%r11
1368         CFI_RESTORE r11
1369         addq $0x30,%rsp
1370         CFI_ADJUST_CFA_OFFSET -0x30
1371         pushq_cfi $0
1372         SAVE_ALL
1373         jmp error_exit
1374         CFI_ENDPROC
1375 END(xen_failsafe_callback)
1376
1377 #endif /* CONFIG_XEN */
1378
1379 /*
1380  * Some functions should be protected against kprobes
1381  */
1382         .pushsection .kprobes.text, "ax"
1383
1384 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1385 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1386 paranoiderrorentry stack_segment do_stack_segment
1387 errorentry general_protection do_general_protection
1388 errorentry page_fault do_page_fault
1389 #ifdef CONFIG_X86_MCE
1390 paranoidzeroentry machine_check do_machine_check
1391 #endif
1392
1393         /*
1394          * "Paranoid" exit path from exception stack.
1395          * Paranoid because this is used by NMIs and cannot take
1396          * any kernel state for granted.
1397          * We don't do kernel preemption checks here, because only
1398          * NMI should be common and it does not enable IRQs and
1399          * cannot get reschedule ticks.
1400          *
1401          * "trace" is 0 for the NMI handler only, because irq-tracing
1402          * is fundamentally NMI-unsafe. (we cannot change the soft and
1403          * hard flags at once, atomically)
1404          */
1405
1406         /* ebx: no swapgs flag */
1407 ENTRY(paranoid_exit)
1408         INTR_FRAME
1409         DISABLE_INTERRUPTS(CLBR_NONE)
1410         TRACE_IRQS_OFF
1411         testl %ebx,%ebx                         /* swapgs needed? */
1412         jnz paranoid_restore
1413         testl $3,CS(%rsp)
1414         jnz   paranoid_userspace
1415 paranoid_swapgs:
1416         TRACE_IRQS_IRETQ 0
1417         SWAPGS_UNSAFE_STACK
1418 paranoid_restore:
1419         RESTORE_ALL 8
1420         jmp irq_return
1421 paranoid_userspace:
1422         GET_THREAD_INFO(%rcx)
1423         movl TI_flags(%rcx),%ebx
1424         andl $_TIF_WORK_MASK,%ebx
1425         jz paranoid_swapgs
1426         movq %rsp,%rdi                  /* &pt_regs */
1427         call sync_regs
1428         movq %rax,%rsp                  /* switch stack for scheduling */
1429         testl $_TIF_NEED_RESCHED,%ebx
1430         jnz paranoid_schedule
1431         movl %ebx,%edx                  /* arg3: thread flags */
1432         TRACE_IRQS_ON
1433         ENABLE_INTERRUPTS(CLBR_NONE)
1434         xorl %esi,%esi                  /* arg2: oldset */
1435         movq %rsp,%rdi                  /* arg1: &pt_regs */
1436         call do_notify_resume
1437         DISABLE_INTERRUPTS(CLBR_NONE)
1438         TRACE_IRQS_OFF
1439         jmp paranoid_userspace
1440 paranoid_schedule:
1441         TRACE_IRQS_ON
1442         ENABLE_INTERRUPTS(CLBR_ANY)
1443         call schedule
1444         DISABLE_INTERRUPTS(CLBR_ANY)
1445         TRACE_IRQS_OFF
1446         jmp paranoid_userspace
1447         CFI_ENDPROC
1448 END(paranoid_exit)
1449
1450 /*
1451  * Exception entry point. This expects an error code/orig_rax on the stack.
1452  * returns in "no swapgs flag" in %ebx.
1453  */
1454 ENTRY(error_entry)
1455         XCPT_FRAME
1456         CFI_ADJUST_CFA_OFFSET 15*8
1457         /* oldrax contains error code */
1458         cld
1459         movq_cfi rdi, RDI+8
1460         movq_cfi rsi, RSI+8
1461         movq_cfi rdx, RDX+8
1462         movq_cfi rcx, RCX+8
1463         movq_cfi rax, RAX+8
1464         movq_cfi  r8,  R8+8
1465         movq_cfi  r9,  R9+8
1466         movq_cfi r10, R10+8
1467         movq_cfi r11, R11+8
1468         movq_cfi rbx, RBX+8
1469         movq_cfi rbp, RBP+8
1470         movq_cfi r12, R12+8
1471         movq_cfi r13, R13+8
1472         movq_cfi r14, R14+8
1473         movq_cfi r15, R15+8
1474         xorl %ebx,%ebx
1475         testl $3,CS+8(%rsp)
1476         je error_kernelspace
1477 error_swapgs:
1478         SWAPGS
1479 error_sti:
1480         TRACE_IRQS_OFF
1481         ret
1482         CFI_ENDPROC
1483
1484 /*
1485  * There are two places in the kernel that can potentially fault with
1486  * usergs. Handle them here. The exception handlers after iret run with
1487  * kernel gs again, so don't set the user space flag. B stepping K8s
1488  * sometimes report an truncated RIP for IRET exceptions returning to
1489  * compat mode. Check for these here too.
1490  */
1491 error_kernelspace:
1492         incl %ebx
1493         leaq irq_return(%rip),%rcx
1494         cmpq %rcx,RIP+8(%rsp)
1495         je error_swapgs
1496         movl %ecx,%ecx  /* zero extend */
1497         cmpq %rcx,RIP+8(%rsp)
1498         je error_swapgs
1499         cmpq $gs_change,RIP+8(%rsp)
1500         je error_swapgs
1501         jmp error_sti
1502 END(error_entry)
1503
1504
1505 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1506 ENTRY(error_exit)
1507         DEFAULT_FRAME
1508         movl %ebx,%eax
1509         RESTORE_REST
1510         DISABLE_INTERRUPTS(CLBR_NONE)
1511         TRACE_IRQS_OFF
1512         GET_THREAD_INFO(%rcx)
1513         testl %eax,%eax
1514         jne retint_kernel
1515         LOCKDEP_SYS_EXIT_IRQ
1516         movl TI_flags(%rcx),%edx
1517         movl $_TIF_WORK_MASK,%edi
1518         andl %edi,%edx
1519         jnz retint_careful
1520         jmp retint_swapgs
1521         CFI_ENDPROC
1522 END(error_exit)
1523
1524
1525         /* runs on exception stack */
1526 ENTRY(nmi)
1527         INTR_FRAME
1528         PARAVIRT_ADJUST_EXCEPTION_FRAME
1529         pushq_cfi $-1
1530         subq $15*8, %rsp
1531         CFI_ADJUST_CFA_OFFSET 15*8
1532         call save_paranoid
1533         DEFAULT_FRAME 0
1534         /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1535         movq %rsp,%rdi
1536         movq $-1,%rsi
1537         call do_nmi
1538 #ifdef CONFIG_TRACE_IRQFLAGS
1539         /* paranoidexit; without TRACE_IRQS_OFF */
1540         /* ebx: no swapgs flag */
1541         DISABLE_INTERRUPTS(CLBR_NONE)
1542         testl %ebx,%ebx                         /* swapgs needed? */
1543         jnz nmi_restore
1544         testl $3,CS(%rsp)
1545         jnz nmi_userspace
1546 nmi_swapgs:
1547         SWAPGS_UNSAFE_STACK
1548 nmi_restore:
1549         RESTORE_ALL 8
1550         jmp irq_return
1551 nmi_userspace:
1552         GET_THREAD_INFO(%rcx)
1553         movl TI_flags(%rcx),%ebx
1554         andl $_TIF_WORK_MASK,%ebx
1555         jz nmi_swapgs
1556         movq %rsp,%rdi                  /* &pt_regs */
1557         call sync_regs
1558         movq %rax,%rsp                  /* switch stack for scheduling */
1559         testl $_TIF_NEED_RESCHED,%ebx
1560         jnz nmi_schedule
1561         movl %ebx,%edx                  /* arg3: thread flags */
1562         ENABLE_INTERRUPTS(CLBR_NONE)
1563         xorl %esi,%esi                  /* arg2: oldset */
1564         movq %rsp,%rdi                  /* arg1: &pt_regs */
1565         call do_notify_resume
1566         DISABLE_INTERRUPTS(CLBR_NONE)
1567         jmp nmi_userspace
1568 nmi_schedule:
1569         ENABLE_INTERRUPTS(CLBR_ANY)
1570         call schedule
1571         DISABLE_INTERRUPTS(CLBR_ANY)
1572         jmp nmi_userspace
1573         CFI_ENDPROC
1574 #else
1575         jmp paranoid_exit
1576         CFI_ENDPROC
1577 #endif
1578 END(nmi)
1579
1580 ENTRY(ignore_sysret)
1581         CFI_STARTPROC
1582         mov $-ENOSYS,%eax
1583         sysret
1584         CFI_ENDPROC
1585 END(ignore_sysret)
1586
1587 /*
1588  * End of kprobes section
1589  */
1590         .popsection