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