1/*
   2 *
   3 *  Copyright (C) 1991, 1992  Linus Torvalds
   4 */
   5
   6/*
   7 * entry.S contains the system-call and fault low-level handling routines.
   8 * This also contains the timer-interrupt handler, as well as all interrupts
   9 * and faults that can result in a task-switch.
  10 *
  11 * NOTE: This code handles signal-recognition, which happens every time
  12 * after a timer-interrupt and after each system call.
  13 *
  14 * I changed all the .align's to 4 (16 byte alignment), as that's faster
  15 * on a 486.
  16 *
  17 * Stack layout in 'syscall_exit':
  18 * 	ptrace needs to have all regs on the stack.
  19 *	if the order here is changed, it needs to be
  20 *	updated in fork.c:copy_process, signal.c:do_signal,
  21 *	ptrace.c and ptrace.h
  22 *
  23 *	 0(%esp) - %ebx
  24 *	 4(%esp) - %ecx
  25 *	 8(%esp) - %edx
  26 *       C(%esp) - %esi
  27 *	10(%esp) - %edi
  28 *	14(%esp) - %ebp
  29 *	18(%esp) - %eax
  30 *	1C(%esp) - %ds
  31 *	20(%esp) - %es
  32 *	24(%esp) - %fs
  33 *	28(%esp) - %gs		saved iff !CONFIG_X86_32_LAZY_GS
  34 *	2C(%esp) - orig_eax
  35 *	30(%esp) - %eip
  36 *	34(%esp) - %cs
  37 *	38(%esp) - %eflags
  38 *	3C(%esp) - %oldesp
  39 *	40(%esp) - %oldss
  40 *
  41 * "current" is in register %ebx during any slow entries.
  42 */
  43
  44#include <linux/linkage.h>
 
  45#include <asm/thread_info.h>
  46#include <asm/irqflags.h>
  47#include <asm/errno.h>
  48#include <asm/segment.h>
  49#include <asm/smp.h>
  50#include <asm/page_types.h>
  51#include <asm/percpu.h>
  52#include <asm/dwarf2.h>
  53#include <asm/processor-flags.h>
  54#include <asm/ftrace.h>
  55#include <asm/irq_vectors.h>
  56#include <asm/cpufeature.h>
  57#include <asm/alternative-asm.h>
 
  58
  59/* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
  60#include <linux/elf-em.h>
  61#define AUDIT_ARCH_I386		(EM_386|__AUDIT_ARCH_LE)
  62#define __AUDIT_ARCH_LE	   0x40000000
  63
  64#ifndef CONFIG_AUDITSYSCALL
  65#define sysenter_audit	syscall_trace_entry
  66#define sysexit_audit	syscall_exit_work
  67#endif
  68
  69	.section .entry.text, "ax"
  70
  71/*
  72 * We use macros for low-level operations which need to be overridden
  73 * for paravirtualization.  The following will never clobber any registers:
  74 *   INTERRUPT_RETURN (aka. "iret")
  75 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
  76 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
  77 *
  78 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
  79 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
  80 * Allowing a register to be clobbered can shrink the paravirt replacement
  81 * enough to patch inline, increasing performance.
  82 */
  83
  84#define nr_syscalls ((syscall_table_size)/4)
  85
  86#ifdef CONFIG_PREEMPT
  87#define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
  88#else
  89#define preempt_stop(clobbers)
  90#define resume_kernel		restore_all
  91#endif
  92
  93.macro TRACE_IRQS_IRET
  94#ifdef CONFIG_TRACE_IRQFLAGS
  95	testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)     # interrupts off?
  96	jz 1f
  97	TRACE_IRQS_ON
  981:
  99#endif
 100.endm
 101
 102#ifdef CONFIG_VM86
 103#define resume_userspace_sig	check_userspace
 104#else
 105#define resume_userspace_sig	resume_userspace
 106#endif
 107
 108/*
 109 * User gs save/restore
 110 *
 111 * %gs is used for userland TLS and kernel only uses it for stack
 112 * canary which is required to be at %gs:20 by gcc.  Read the comment
 113 * at the top of stackprotector.h for more info.
 114 *
 115 * Local labels 98 and 99 are used.
 116 */
 117#ifdef CONFIG_X86_32_LAZY_GS
 118
 119 /* unfortunately push/pop can't be no-op */
 120.macro PUSH_GS
 121	pushl_cfi $0
 122.endm
 123.macro POP_GS pop=0
 124	addl $(4 + \pop), %esp
 125	CFI_ADJUST_CFA_OFFSET -(4 + \pop)
 126.endm
 127.macro POP_GS_EX
 128.endm
 129
 130 /* all the rest are no-op */
 131.macro PTGS_TO_GS
 132.endm
 133.macro PTGS_TO_GS_EX
 134.endm
 135.macro GS_TO_REG reg
 136.endm
 137.macro REG_TO_PTGS reg
 138.endm
 139.macro SET_KERNEL_GS reg
 140.endm
 141
 142#else	/* CONFIG_X86_32_LAZY_GS */
 143
 144.macro PUSH_GS
 145	pushl_cfi %gs
 146	/*CFI_REL_OFFSET gs, 0*/
 147.endm
 148
 149.macro POP_GS pop=0
 15098:	popl_cfi %gs
 151	/*CFI_RESTORE gs*/
 152  .if \pop <> 0
 153	add $\pop, %esp
 154	CFI_ADJUST_CFA_OFFSET -\pop
 155  .endif
 156.endm
 157.macro POP_GS_EX
 158.pushsection .fixup, "ax"
 15999:	movl $0, (%esp)
 160	jmp 98b
 161.section __ex_table, "a"
 162	.align 4
 163	.long 98b, 99b
 164.popsection
 
 165.endm
 166
 167.macro PTGS_TO_GS
 16898:	mov PT_GS(%esp), %gs
 169.endm
 170.macro PTGS_TO_GS_EX
 171.pushsection .fixup, "ax"
 17299:	movl $0, PT_GS(%esp)
 173	jmp 98b
 174.section __ex_table, "a"
 175	.align 4
 176	.long 98b, 99b
 177.popsection
 
 178.endm
 179
 180.macro GS_TO_REG reg
 181	movl %gs, \reg
 182	/*CFI_REGISTER gs, \reg*/
 183.endm
 184.macro REG_TO_PTGS reg
 185	movl \reg, PT_GS(%esp)
 186	/*CFI_REL_OFFSET gs, PT_GS*/
 187.endm
 188.macro SET_KERNEL_GS reg
 189	movl $(__KERNEL_STACK_CANARY), \reg
 190	movl \reg, %gs
 191.endm
 192
 193#endif	/* CONFIG_X86_32_LAZY_GS */
 194
 195.macro SAVE_ALL
 196	cld
 197	PUSH_GS
 198	pushl_cfi %fs
 199	/*CFI_REL_OFFSET fs, 0;*/
 200	pushl_cfi %es
 201	/*CFI_REL_OFFSET es, 0;*/
 202	pushl_cfi %ds
 203	/*CFI_REL_OFFSET ds, 0;*/
 204	pushl_cfi %eax
 205	CFI_REL_OFFSET eax, 0
 206	pushl_cfi %ebp
 207	CFI_REL_OFFSET ebp, 0
 208	pushl_cfi %edi
 209	CFI_REL_OFFSET edi, 0
 210	pushl_cfi %esi
 211	CFI_REL_OFFSET esi, 0
 212	pushl_cfi %edx
 213	CFI_REL_OFFSET edx, 0
 214	pushl_cfi %ecx
 215	CFI_REL_OFFSET ecx, 0
 216	pushl_cfi %ebx
 217	CFI_REL_OFFSET ebx, 0
 218	movl $(__USER_DS), %edx
 219	movl %edx, %ds
 220	movl %edx, %es
 221	movl $(__KERNEL_PERCPU), %edx
 222	movl %edx, %fs
 223	SET_KERNEL_GS %edx
 224.endm
 225
 226.macro RESTORE_INT_REGS
 227	popl_cfi %ebx
 228	CFI_RESTORE ebx
 229	popl_cfi %ecx
 230	CFI_RESTORE ecx
 231	popl_cfi %edx
 232	CFI_RESTORE edx
 233	popl_cfi %esi
 234	CFI_RESTORE esi
 235	popl_cfi %edi
 236	CFI_RESTORE edi
 237	popl_cfi %ebp
 238	CFI_RESTORE ebp
 239	popl_cfi %eax
 240	CFI_RESTORE eax
 241.endm
 242
 243.macro RESTORE_REGS pop=0
 244	RESTORE_INT_REGS
 2451:	popl_cfi %ds
 246	/*CFI_RESTORE ds;*/
 2472:	popl_cfi %es
 248	/*CFI_RESTORE es;*/
 2493:	popl_cfi %fs
 250	/*CFI_RESTORE fs;*/
 251	POP_GS \pop
 252.pushsection .fixup, "ax"
 2534:	movl $0, (%esp)
 254	jmp 1b
 2555:	movl $0, (%esp)
 256	jmp 2b
 2576:	movl $0, (%esp)
 258	jmp 3b
 259.section __ex_table, "a"
 260	.align 4
 261	.long 1b, 4b
 262	.long 2b, 5b
 263	.long 3b, 6b
 264.popsection
 
 
 
 265	POP_GS_EX
 266.endm
 267
 268.macro RING0_INT_FRAME
 269	CFI_STARTPROC simple
 270	CFI_SIGNAL_FRAME
 271	CFI_DEF_CFA esp, 3*4
 272	/*CFI_OFFSET cs, -2*4;*/
 273	CFI_OFFSET eip, -3*4
 274.endm
 275
 276.macro RING0_EC_FRAME
 277	CFI_STARTPROC simple
 278	CFI_SIGNAL_FRAME
 279	CFI_DEF_CFA esp, 4*4
 280	/*CFI_OFFSET cs, -2*4;*/
 281	CFI_OFFSET eip, -3*4
 282.endm
 283
 284.macro RING0_PTREGS_FRAME
 285	CFI_STARTPROC simple
 286	CFI_SIGNAL_FRAME
 287	CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
 288	/*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
 289	CFI_OFFSET eip, PT_EIP-PT_OLDESP
 290	/*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
 291	/*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
 292	CFI_OFFSET eax, PT_EAX-PT_OLDESP
 293	CFI_OFFSET ebp, PT_EBP-PT_OLDESP
 294	CFI_OFFSET edi, PT_EDI-PT_OLDESP
 295	CFI_OFFSET esi, PT_ESI-PT_OLDESP
 296	CFI_OFFSET edx, PT_EDX-PT_OLDESP
 297	CFI_OFFSET ecx, PT_ECX-PT_OLDESP
 298	CFI_OFFSET ebx, PT_EBX-PT_OLDESP
 299.endm
 300
 301ENTRY(ret_from_fork)
 302	CFI_STARTPROC
 303	pushl_cfi %eax
 304	call schedule_tail
 305	GET_THREAD_INFO(%ebp)
 306	popl_cfi %eax
 307	pushl_cfi $0x0202		# Reset kernel eflags
 308	popfl_cfi
 309	jmp syscall_exit
 310	CFI_ENDPROC
 311END(ret_from_fork)
 312
 313/*
 314 * Interrupt exit functions should be protected against kprobes
 315 */
 316	.pushsection .kprobes.text, "ax"
 317/*
 318 * Return to user mode is not as complex as all this looks,
 319 * but we want the default path for a system call return to
 320 * go as quickly as possible which is why some of this is
 321 * less clear than it otherwise should be.
 322 */
 323
 324	# userspace resumption stub bypassing syscall exit tracing
 325	ALIGN
 326	RING0_PTREGS_FRAME
 327ret_from_exception:
 328	preempt_stop(CLBR_ANY)
 329ret_from_intr:
 330	GET_THREAD_INFO(%ebp)
 331check_userspace:
 332	movl PT_EFLAGS(%esp), %eax	# mix EFLAGS and CS
 333	movb PT_CS(%esp), %al
 334	andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
 
 
 
 
 
 
 
 
 335	cmpl $USER_RPL, %eax
 336	jb resume_kernel		# not returning to v8086 or userspace
 337
 338ENTRY(resume_userspace)
 339	LOCKDEP_SYS_EXIT
 340 	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
 341					# setting need_resched or sigpending
 342					# between sampling and the iret
 343	TRACE_IRQS_OFF
 344	movl TI_flags(%ebp), %ecx
 345	andl $_TIF_WORK_MASK, %ecx	# is there any work to be done on
 346					# int/exception return?
 347	jne work_pending
 348	jmp restore_all
 349END(ret_from_exception)
 350
 351#ifdef CONFIG_PREEMPT
 352ENTRY(resume_kernel)
 353	DISABLE_INTERRUPTS(CLBR_ANY)
 354	cmpl $0,TI_preempt_count(%ebp)	# non-zero preempt_count ?
 355	jnz restore_all
 356need_resched:
 357	movl TI_flags(%ebp), %ecx	# need_resched set ?
 358	testb $_TIF_NEED_RESCHED, %cl
 359	jz restore_all
 360	testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)	# interrupts off (exception path) ?
 361	jz restore_all
 362	call preempt_schedule_irq
 363	jmp need_resched
 364END(resume_kernel)
 365#endif
 366	CFI_ENDPROC
 367/*
 368 * End of kprobes section
 369 */
 370	.popsection
 371
 372/* SYSENTER_RETURN points to after the "sysenter" instruction in
 373   the vsyscall page.  See vsyscall-sysentry.S, which defines the symbol.  */
 374
 375	# sysenter call handler stub
 376ENTRY(ia32_sysenter_target)
 377	CFI_STARTPROC simple
 378	CFI_SIGNAL_FRAME
 379	CFI_DEF_CFA esp, 0
 380	CFI_REGISTER esp, ebp
 381	movl TSS_sysenter_sp0(%esp),%esp
 382sysenter_past_esp:
 383	/*
 384	 * Interrupts are disabled here, but we can't trace it until
 385	 * enough kernel state to call TRACE_IRQS_OFF can be called - but
 386	 * we immediately enable interrupts at that point anyway.
 387	 */
 388	pushl_cfi $__USER_DS
 389	/*CFI_REL_OFFSET ss, 0*/
 390	pushl_cfi %ebp
 391	CFI_REL_OFFSET esp, 0
 392	pushfl_cfi
 393	orl $X86_EFLAGS_IF, (%esp)
 394	pushl_cfi $__USER_CS
 395	/*CFI_REL_OFFSET cs, 0*/
 396	/*
 397	 * Push current_thread_info()->sysenter_return to the stack.
 398	 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
 399	 * pushed above; +8 corresponds to copy_thread's esp0 setting.
 400	 */
 401	pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
 402	CFI_REL_OFFSET eip, 0
 403
 404	pushl_cfi %eax
 405	SAVE_ALL
 406	ENABLE_INTERRUPTS(CLBR_NONE)
 407
 408/*
 409 * Load the potential sixth argument from user stack.
 410 * Careful about security.
 411 */
 412	cmpl $__PAGE_OFFSET-3,%ebp
 413	jae syscall_fault
 4141:	movl (%ebp),%ebp
 415	movl %ebp,PT_EBP(%esp)
 416.section __ex_table,"a"
 417	.align 4
 418	.long 1b,syscall_fault
 419.previous
 420
 421	GET_THREAD_INFO(%ebp)
 422
 423	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
 424	jnz sysenter_audit
 425sysenter_do_call:
 426	cmpl $(nr_syscalls), %eax
 427	jae syscall_badsys
 428	call *sys_call_table(,%eax,4)
 429	movl %eax,PT_EAX(%esp)
 430	LOCKDEP_SYS_EXIT
 431	DISABLE_INTERRUPTS(CLBR_ANY)
 432	TRACE_IRQS_OFF
 433	movl TI_flags(%ebp), %ecx
 434	testl $_TIF_ALLWORK_MASK, %ecx
 435	jne sysexit_audit
 436sysenter_exit:
 437/* if something modifies registers it must also disable sysexit */
 438	movl PT_EIP(%esp), %edx
 439	movl PT_OLDESP(%esp), %ecx
 440	xorl %ebp,%ebp
 441	TRACE_IRQS_ON
 4421:	mov  PT_FS(%esp), %fs
 443	PTGS_TO_GS
 444	ENABLE_INTERRUPTS_SYSEXIT
 445
 446#ifdef CONFIG_AUDITSYSCALL
 447sysenter_audit:
 448	testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
 449	jnz syscall_trace_entry
 450	addl $4,%esp
 451	CFI_ADJUST_CFA_OFFSET -4
 452	/* %esi already in 8(%esp)	   6th arg: 4th syscall arg */
 453	/* %edx already in 4(%esp)	   5th arg: 3rd syscall arg */
 454	/* %ecx already in 0(%esp)	   4th arg: 2nd syscall arg */
 455	movl %ebx,%ecx			/* 3rd arg: 1st syscall arg */
 456	movl %eax,%edx			/* 2nd arg: syscall number */
 457	movl $AUDIT_ARCH_I386,%eax	/* 1st arg: audit arch */
 458	call audit_syscall_entry
 459	pushl_cfi %ebx
 460	movl PT_EAX(%esp),%eax		/* reload syscall number */
 461	jmp sysenter_do_call
 462
 463sysexit_audit:
 464	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 465	jne syscall_exit_work
 466	TRACE_IRQS_ON
 467	ENABLE_INTERRUPTS(CLBR_ANY)
 468	movl %eax,%edx		/* second arg, syscall return value */
 469	cmpl $0,%eax		/* is it < 0? */
 470	setl %al		/* 1 if so, 0 if not */
 471	movzbl %al,%eax		/* zero-extend that */
 472	inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
 473	call audit_syscall_exit
 474	DISABLE_INTERRUPTS(CLBR_ANY)
 475	TRACE_IRQS_OFF
 476	movl TI_flags(%ebp), %ecx
 477	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
 478	jne syscall_exit_work
 479	movl PT_EAX(%esp),%eax	/* reload syscall return value */
 480	jmp sysenter_exit
 481#endif
 482
 483	CFI_ENDPROC
 484.pushsection .fixup,"ax"
 4852:	movl $0,PT_FS(%esp)
 486	jmp 1b
 487.section __ex_table,"a"
 488	.align 4
 489	.long 1b,2b
 490.popsection
 
 491	PTGS_TO_GS_EX
 492ENDPROC(ia32_sysenter_target)
 493
 494/*
 495 * syscall stub including irq exit should be protected against kprobes
 496 */
 497	.pushsection .kprobes.text, "ax"
 498	# system call handler stub
 499ENTRY(system_call)
 500	RING0_INT_FRAME			# can't unwind into user space anyway
 501	pushl_cfi %eax			# save orig_eax
 502	SAVE_ALL
 503	GET_THREAD_INFO(%ebp)
 504					# system call tracing in operation / emulation
 505	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
 506	jnz syscall_trace_entry
 507	cmpl $(nr_syscalls), %eax
 508	jae syscall_badsys
 509syscall_call:
 510	call *sys_call_table(,%eax,4)
 511	movl %eax,PT_EAX(%esp)		# store the return value
 512syscall_exit:
 513	LOCKDEP_SYS_EXIT
 514	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
 515					# setting need_resched or sigpending
 516					# between sampling and the iret
 517	TRACE_IRQS_OFF
 518	movl TI_flags(%ebp), %ecx
 519	testl $_TIF_ALLWORK_MASK, %ecx	# current->work
 520	jne syscall_exit_work
 521
 522restore_all:
 523	TRACE_IRQS_IRET
 524restore_all_notrace:
 525	movl PT_EFLAGS(%esp), %eax	# mix EFLAGS, SS and CS
 526	# Warning: PT_OLDSS(%esp) contains the wrong/random values if we
 527	# are returning to the kernel.
 528	# See comments in process.c:copy_thread() for details.
 529	movb PT_OLDSS(%esp), %ah
 530	movb PT_CS(%esp), %al
 531	andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
 532	cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
 533	CFI_REMEMBER_STATE
 534	je ldt_ss			# returning to user-space with LDT SS
 535restore_nocheck:
 536	RESTORE_REGS 4			# skip orig_eax/error_code
 537irq_return:
 538	INTERRUPT_RETURN
 539.section .fixup,"ax"
 540ENTRY(iret_exc)
 541	pushl $0			# no error code
 542	pushl $do_iret_error
 543	jmp error_code
 544.previous
 545.section __ex_table,"a"
 546	.align 4
 547	.long irq_return,iret_exc
 548.previous
 549
 550	CFI_RESTORE_STATE
 551ldt_ss:
 552	larl PT_OLDSS(%esp), %eax
 553	jnz restore_nocheck
 554	testl $0x00400000, %eax		# returning to 32bit stack?
 555	jnz restore_nocheck		# allright, normal return
 556
 557#ifdef CONFIG_PARAVIRT
 558	/*
 559	 * The kernel can't run on a non-flat stack if paravirt mode
 560	 * is active.  Rather than try to fixup the high bits of
 561	 * ESP, bypass this code entirely.  This may break DOSemu
 562	 * and/or Wine support in a paravirt VM, although the option
 563	 * is still available to implement the setting of the high
 564	 * 16-bits in the INTERRUPT_RETURN paravirt-op.
 565	 */
 566	cmpl $0, pv_info+PARAVIRT_enabled
 567	jne restore_nocheck
 568#endif
 569
 570/*
 571 * Setup and switch to ESPFIX stack
 572 *
 573 * We're returning to userspace with a 16 bit stack. The CPU will not
 574 * restore the high word of ESP for us on executing iret... This is an
 575 * "official" bug of all the x86-compatible CPUs, which we can work
 576 * around to make dosemu and wine happy. We do this by preloading the
 577 * high word of ESP with the high word of the userspace ESP while
 578 * compensating for the offset by changing to the ESPFIX segment with
 579 * a base address that matches for the difference.
 580 */
 581#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
 582	mov %esp, %edx			/* load kernel esp */
 583	mov PT_OLDESP(%esp), %eax	/* load userspace esp */
 584	mov %dx, %ax			/* eax: new kernel esp */
 585	sub %eax, %edx			/* offset (low word is 0) */
 586	shr $16, %edx
 587	mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
 588	mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
 589	pushl_cfi $__ESPFIX_SS
 590	pushl_cfi %eax			/* new kernel esp */
 591	/* Disable interrupts, but do not irqtrace this section: we
 592	 * will soon execute iret and the tracer was already set to
 593	 * the irqstate after the iret */
 594	DISABLE_INTERRUPTS(CLBR_EAX)
 595	lss (%esp), %esp		/* switch to espfix segment */
 596	CFI_ADJUST_CFA_OFFSET -8
 597	jmp restore_nocheck
 598	CFI_ENDPROC
 599ENDPROC(system_call)
 600
 601	# perform work that needs to be done immediately before resumption
 602	ALIGN
 603	RING0_PTREGS_FRAME		# can't unwind into user space anyway
 604work_pending:
 605	testb $_TIF_NEED_RESCHED, %cl
 606	jz work_notifysig
 607work_resched:
 608	call schedule
 609	LOCKDEP_SYS_EXIT
 610	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
 611					# setting need_resched or sigpending
 612					# between sampling and the iret
 613	TRACE_IRQS_OFF
 614	movl TI_flags(%ebp), %ecx
 615	andl $_TIF_WORK_MASK, %ecx	# is there any work to be done other
 616					# than syscall tracing?
 617	jz restore_all
 618	testb $_TIF_NEED_RESCHED, %cl
 619	jnz work_resched
 620
 621work_notifysig:				# deal with pending signals and
 622					# notify-resume requests
 623#ifdef CONFIG_VM86
 624	testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
 625	movl %esp, %eax
 626	jne work_notifysig_v86		# returning to kernel-space or
 627					# vm86-space
 
 
 
 
 
 
 628	xorl %edx, %edx
 629	call do_notify_resume
 630	jmp resume_userspace_sig
 631
 632	ALIGN
 633work_notifysig_v86:
 634	pushl_cfi %ecx			# save ti_flags for do_notify_resume
 635	call save_v86_state		# %eax contains pt_regs pointer
 636	popl_cfi %ecx
 637	movl %eax, %esp
 638#else
 639	movl %esp, %eax
 640#endif
 
 
 
 
 
 
 641	xorl %edx, %edx
 642	call do_notify_resume
 643	jmp resume_userspace_sig
 644END(work_pending)
 645
 646	# perform syscall exit tracing
 647	ALIGN
 648syscall_trace_entry:
 649	movl $-ENOSYS,PT_EAX(%esp)
 650	movl %esp, %eax
 651	call syscall_trace_enter
 652	/* What it returned is what we'll actually use.  */
 653	cmpl $(nr_syscalls), %eax
 654	jnae syscall_call
 655	jmp syscall_exit
 656END(syscall_trace_entry)
 657
 658	# perform syscall exit tracing
 659	ALIGN
 660syscall_exit_work:
 661	testl $_TIF_WORK_SYSCALL_EXIT, %ecx
 662	jz work_pending
 663	TRACE_IRQS_ON
 664	ENABLE_INTERRUPTS(CLBR_ANY)	# could let syscall_trace_leave() call
 665					# schedule() instead
 666	movl %esp, %eax
 667	call syscall_trace_leave
 668	jmp resume_userspace
 669END(syscall_exit_work)
 670	CFI_ENDPROC
 671
 672	RING0_INT_FRAME			# can't unwind into user space anyway
 673syscall_fault:
 674	GET_THREAD_INFO(%ebp)
 675	movl $-EFAULT,PT_EAX(%esp)
 676	jmp resume_userspace
 677END(syscall_fault)
 678
 679syscall_badsys:
 680	movl $-ENOSYS,PT_EAX(%esp)
 681	jmp resume_userspace
 682END(syscall_badsys)
 683	CFI_ENDPROC
 684/*
 685 * End of kprobes section
 686 */
 687	.popsection
 688
 689/*
 690 * System calls that need a pt_regs pointer.
 691 */
 692#define PTREGSCALL0(name) \
 693	ALIGN; \
 694ptregs_##name: \
 695	leal 4(%esp),%eax; \
 696	jmp sys_##name;
 
 697
 698#define PTREGSCALL1(name) \
 699	ALIGN; \
 700ptregs_##name: \
 701	leal 4(%esp),%edx; \
 702	movl (PT_EBX+4)(%esp),%eax; \
 703	jmp sys_##name;
 
 704
 705#define PTREGSCALL2(name) \
 706	ALIGN; \
 707ptregs_##name: \
 708	leal 4(%esp),%ecx; \
 709	movl (PT_ECX+4)(%esp),%edx; \
 710	movl (PT_EBX+4)(%esp),%eax; \
 711	jmp sys_##name;
 
 712
 713#define PTREGSCALL3(name) \
 714	ALIGN; \
 715ptregs_##name: \
 716	CFI_STARTPROC; \
 717	leal 4(%esp),%eax; \
 718	pushl_cfi %eax; \
 719	movl PT_EDX(%eax),%ecx; \
 720	movl PT_ECX(%eax),%edx; \
 721	movl PT_EBX(%eax),%eax; \
 722	call sys_##name; \
 723	addl $4,%esp; \
 724	CFI_ADJUST_CFA_OFFSET -4; \
 725	ret; \
 726	CFI_ENDPROC; \
 727ENDPROC(ptregs_##name)
 728
 729PTREGSCALL1(iopl)
 730PTREGSCALL0(fork)
 731PTREGSCALL0(vfork)
 732PTREGSCALL3(execve)
 733PTREGSCALL2(sigaltstack)
 734PTREGSCALL0(sigreturn)
 735PTREGSCALL0(rt_sigreturn)
 736PTREGSCALL2(vm86)
 737PTREGSCALL1(vm86old)
 738
 739/* Clone is an oddball.  The 4th arg is in %edi */
 740	ALIGN;
 741ptregs_clone:
 742	CFI_STARTPROC
 743	leal 4(%esp),%eax
 744	pushl_cfi %eax
 745	pushl_cfi PT_EDI(%eax)
 746	movl PT_EDX(%eax),%ecx
 747	movl PT_ECX(%eax),%edx
 748	movl PT_EBX(%eax),%eax
 749	call sys_clone
 750	addl $8,%esp
 751	CFI_ADJUST_CFA_OFFSET -8
 752	ret
 753	CFI_ENDPROC
 754ENDPROC(ptregs_clone)
 755
 756.macro FIXUP_ESPFIX_STACK
 757/*
 758 * Switch back for ESPFIX stack to the normal zerobased stack
 759 *
 760 * We can't call C functions using the ESPFIX stack. This code reads
 761 * the high word of the segment base from the GDT and swiches to the
 762 * normal stack and adjusts ESP with the matching offset.
 763 */
 764	/* fixup the stack */
 765	mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
 766	mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
 767	shl $16, %eax
 768	addl %esp, %eax			/* the adjusted stack pointer */
 769	pushl_cfi $__KERNEL_DS
 770	pushl_cfi %eax
 771	lss (%esp), %esp		/* switch to the normal stack segment */
 772	CFI_ADJUST_CFA_OFFSET -8
 773.endm
 774.macro UNWIND_ESPFIX_STACK
 775	movl %ss, %eax
 776	/* see if on espfix stack */
 777	cmpw $__ESPFIX_SS, %ax
 778	jne 27f
 779	movl $__KERNEL_DS, %eax
 780	movl %eax, %ds
 781	movl %eax, %es
 782	/* switch to normal stack */
 783	FIXUP_ESPFIX_STACK
 78427:
 785.endm
 786
 787/*
 788 * Build the entry stubs and pointer table with some assembler magic.
 789 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
 790 * single cache line on all modern x86 implementations.
 791 */
 792.section .init.rodata,"a"
 793ENTRY(interrupt)
 794.section .entry.text, "ax"
 795	.p2align 5
 796	.p2align CONFIG_X86_L1_CACHE_SHIFT
 797ENTRY(irq_entries_start)
 798	RING0_INT_FRAME
 799vector=FIRST_EXTERNAL_VECTOR
 800.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
 801	.balign 32
 802  .rept	7
 803    .if vector < NR_VECTORS
 804      .if vector <> FIRST_EXTERNAL_VECTOR
 805	CFI_ADJUST_CFA_OFFSET -4
 806      .endif
 8071:	pushl_cfi $(~vector+0x80)	/* Note: always in signed byte range */
 808      .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
 809	jmp 2f
 810      .endif
 811      .previous
 812	.long 1b
 813      .section .entry.text, "ax"
 814vector=vector+1
 815    .endif
 816  .endr
 8172:	jmp common_interrupt
 818.endr
 819END(irq_entries_start)
 820
 821.previous
 822END(interrupt)
 823.previous
 824
 825/*
 826 * the CPU automatically disables interrupts when executing an IRQ vector,
 827 * so IRQ-flags tracing has to follow that:
 828 */
 829	.p2align CONFIG_X86_L1_CACHE_SHIFT
 830common_interrupt:
 831	addl $-0x80,(%esp)	/* Adjust vector into the [-256,-1] range */
 832	SAVE_ALL
 833	TRACE_IRQS_OFF
 834	movl %esp,%eax
 835	call do_IRQ
 836	jmp ret_from_intr
 837ENDPROC(common_interrupt)
 838	CFI_ENDPROC
 839
 840/*
 841 *  Irq entries should be protected against kprobes
 842 */
 843	.pushsection .kprobes.text, "ax"
 844#define BUILD_INTERRUPT3(name, nr, fn)	\
 845ENTRY(name)				\
 846	RING0_INT_FRAME;		\
 847	pushl_cfi $~(nr);		\
 848	SAVE_ALL;			\
 849	TRACE_IRQS_OFF			\
 850	movl %esp,%eax;			\
 851	call fn;			\
 852	jmp ret_from_intr;		\
 853	CFI_ENDPROC;			\
 854ENDPROC(name)
 855
 856#define BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(name, nr, smp_##name)
 857
 858/* The include is where all of the SMP etc. interrupts come from */
 859#include <asm/entry_arch.h>
 860
 861ENTRY(coprocessor_error)
 862	RING0_INT_FRAME
 863	pushl_cfi $0
 864	pushl_cfi $do_coprocessor_error
 865	jmp error_code
 866	CFI_ENDPROC
 867END(coprocessor_error)
 868
 869ENTRY(simd_coprocessor_error)
 870	RING0_INT_FRAME
 871	pushl_cfi $0
 872#ifdef CONFIG_X86_INVD_BUG
 873	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
 874661:	pushl_cfi $do_general_protection
 875662:
 876.section .altinstructions,"a"
 877	altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
 878.previous
 879.section .altinstr_replacement,"ax"
 880663:	pushl $do_simd_coprocessor_error
 881664:
 882.previous
 883#else
 884	pushl_cfi $do_simd_coprocessor_error
 885#endif
 886	jmp error_code
 887	CFI_ENDPROC
 888END(simd_coprocessor_error)
 889
 890ENTRY(device_not_available)
 891	RING0_INT_FRAME
 892	pushl_cfi $-1			# mark this as an int
 893	pushl_cfi $do_device_not_available
 894	jmp error_code
 895	CFI_ENDPROC
 896END(device_not_available)
 897
 898#ifdef CONFIG_PARAVIRT
 899ENTRY(native_iret)
 900	iret
 901.section __ex_table,"a"
 902	.align 4
 903	.long native_iret, iret_exc
 904.previous
 905END(native_iret)
 906
 907ENTRY(native_irq_enable_sysexit)
 908	sti
 909	sysexit
 910END(native_irq_enable_sysexit)
 911#endif
 912
 913ENTRY(overflow)
 914	RING0_INT_FRAME
 915	pushl_cfi $0
 916	pushl_cfi $do_overflow
 917	jmp error_code
 918	CFI_ENDPROC
 919END(overflow)
 920
 921ENTRY(bounds)
 922	RING0_INT_FRAME
 923	pushl_cfi $0
 924	pushl_cfi $do_bounds
 925	jmp error_code
 926	CFI_ENDPROC
 927END(bounds)
 928
 929ENTRY(invalid_op)
 930	RING0_INT_FRAME
 931	pushl_cfi $0
 932	pushl_cfi $do_invalid_op
 933	jmp error_code
 934	CFI_ENDPROC
 935END(invalid_op)
 936
 937ENTRY(coprocessor_segment_overrun)
 938	RING0_INT_FRAME
 939	pushl_cfi $0
 940	pushl_cfi $do_coprocessor_segment_overrun
 941	jmp error_code
 942	CFI_ENDPROC
 943END(coprocessor_segment_overrun)
 944
 945ENTRY(invalid_TSS)
 946	RING0_EC_FRAME
 947	pushl_cfi $do_invalid_TSS
 948	jmp error_code
 949	CFI_ENDPROC
 950END(invalid_TSS)
 951
 952ENTRY(segment_not_present)
 953	RING0_EC_FRAME
 954	pushl_cfi $do_segment_not_present
 955	jmp error_code
 956	CFI_ENDPROC
 957END(segment_not_present)
 958
 959ENTRY(stack_segment)
 960	RING0_EC_FRAME
 961	pushl_cfi $do_stack_segment
 962	jmp error_code
 963	CFI_ENDPROC
 964END(stack_segment)
 965
 966ENTRY(alignment_check)
 967	RING0_EC_FRAME
 968	pushl_cfi $do_alignment_check
 969	jmp error_code
 970	CFI_ENDPROC
 971END(alignment_check)
 972
 973ENTRY(divide_error)
 974	RING0_INT_FRAME
 975	pushl_cfi $0			# no error code
 976	pushl_cfi $do_divide_error
 977	jmp error_code
 978	CFI_ENDPROC
 979END(divide_error)
 980
 981#ifdef CONFIG_X86_MCE
 982ENTRY(machine_check)
 983	RING0_INT_FRAME
 984	pushl_cfi $0
 985	pushl_cfi machine_check_vector
 986	jmp error_code
 987	CFI_ENDPROC
 988END(machine_check)
 989#endif
 990
 991ENTRY(spurious_interrupt_bug)
 992	RING0_INT_FRAME
 993	pushl_cfi $0
 994	pushl_cfi $do_spurious_interrupt_bug
 995	jmp error_code
 996	CFI_ENDPROC
 997END(spurious_interrupt_bug)
 998/*
 999 * End of kprobes section
1000 */
1001	.popsection
1002
1003ENTRY(kernel_thread_helper)
1004	pushl $0		# fake return address for unwinder
1005	CFI_STARTPROC
1006	movl %edi,%eax
1007	call *%esi
1008	call do_exit
1009	ud2			# padding for call trace
1010	CFI_ENDPROC
1011ENDPROC(kernel_thread_helper)
1012
1013#ifdef CONFIG_XEN
1014/* Xen doesn't set %esp to be precisely what the normal sysenter
1015   entrypoint expects, so fix it up before using the normal path. */
1016ENTRY(xen_sysenter_target)
1017	RING0_INT_FRAME
1018	addl $5*4, %esp		/* remove xen-provided frame */
1019	CFI_ADJUST_CFA_OFFSET -5*4
1020	jmp sysenter_past_esp
1021	CFI_ENDPROC
1022
1023ENTRY(xen_hypervisor_callback)
1024	CFI_STARTPROC
1025	pushl_cfi $0
1026	SAVE_ALL
1027	TRACE_IRQS_OFF
1028
1029	/* Check to see if we got the event in the critical
1030	   region in xen_iret_direct, after we've reenabled
1031	   events and checked for pending events.  This simulates
1032	   iret instruction's behaviour where it delivers a
1033	   pending interrupt when enabling interrupts. */
1034	movl PT_EIP(%esp),%eax
1035	cmpl $xen_iret_start_crit,%eax
1036	jb   1f
1037	cmpl $xen_iret_end_crit,%eax
1038	jae  1f
1039
1040	jmp  xen_iret_crit_fixup
1041
1042ENTRY(xen_do_upcall)
10431:	mov %esp, %eax
1044	call xen_evtchn_do_upcall
1045	jmp  ret_from_intr
1046	CFI_ENDPROC
1047ENDPROC(xen_hypervisor_callback)
1048
1049# Hypervisor uses this for application faults while it executes.
1050# We get here for two reasons:
1051#  1. Fault while reloading DS, ES, FS or GS
1052#  2. Fault while executing IRET
1053# Category 1 we fix up by reattempting the load, and zeroing the segment
1054# register if the load fails.
1055# Category 2 we fix up by jumping to do_iret_error. We cannot use the
1056# normal Linux return path in this case because if we use the IRET hypercall
1057# to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1058# We distinguish between categories by maintaining a status value in EAX.
1059ENTRY(xen_failsafe_callback)
1060	CFI_STARTPROC
1061	pushl_cfi %eax
1062	movl $1,%eax
10631:	mov 4(%esp),%ds
10642:	mov 8(%esp),%es
10653:	mov 12(%esp),%fs
10664:	mov 16(%esp),%gs
1067	testl %eax,%eax
1068	popl_cfi %eax
1069	lea 16(%esp),%esp
1070	CFI_ADJUST_CFA_OFFSET -16
1071	jz 5f
1072	addl $16,%esp
1073	jmp iret_exc		# EAX != 0 => Category 2 (Bad IRET)
10745:	pushl_cfi $0		# EAX == 0 => Category 1 (Bad segment)
1075	SAVE_ALL
1076	jmp ret_from_exception
1077	CFI_ENDPROC
1078
1079.section .fixup,"ax"
10806:	xorl %eax,%eax
1081	movl %eax,4(%esp)
1082	jmp 1b
10837:	xorl %eax,%eax
1084	movl %eax,8(%esp)
1085	jmp 2b
10868:	xorl %eax,%eax
1087	movl %eax,12(%esp)
1088	jmp 3b
10899:	xorl %eax,%eax
1090	movl %eax,16(%esp)
1091	jmp 4b
1092.previous
1093.section __ex_table,"a"
1094	.align 4
1095	.long 1b,6b
1096	.long 2b,7b
1097	.long 3b,8b
1098	.long 4b,9b
1099.previous
1100ENDPROC(xen_failsafe_callback)
1101
1102BUILD_INTERRUPT3(xen_hvm_callback_vector, XEN_HVM_EVTCHN_CALLBACK,
1103		xen_evtchn_do_upcall)
1104
1105#endif	/* CONFIG_XEN */
1106
1107#ifdef CONFIG_FUNCTION_TRACER
1108#ifdef CONFIG_DYNAMIC_FTRACE
1109
1110ENTRY(mcount)
1111	ret
1112END(mcount)
1113
1114ENTRY(ftrace_caller)
1115	cmpl $0, function_trace_stop
1116	jne  ftrace_stub
1117
1118	pushl %eax
1119	pushl %ecx
1120	pushl %edx
1121	movl 0xc(%esp), %eax
1122	movl 0x4(%ebp), %edx
1123	subl $MCOUNT_INSN_SIZE, %eax
1124
1125.globl ftrace_call
1126ftrace_call:
1127	call ftrace_stub
1128
1129	popl %edx
1130	popl %ecx
1131	popl %eax
1132#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1133.globl ftrace_graph_call
1134ftrace_graph_call:
1135	jmp ftrace_stub
1136#endif
1137
1138.globl ftrace_stub
1139ftrace_stub:
1140	ret
1141END(ftrace_caller)
1142
1143#else /* ! CONFIG_DYNAMIC_FTRACE */
1144
1145ENTRY(mcount)
1146	cmpl $0, function_trace_stop
1147	jne  ftrace_stub
1148
1149	cmpl $ftrace_stub, ftrace_trace_function
1150	jnz trace
1151#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1152	cmpl $ftrace_stub, ftrace_graph_return
1153	jnz ftrace_graph_caller
1154
1155	cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1156	jnz ftrace_graph_caller
1157#endif
1158.globl ftrace_stub
1159ftrace_stub:
1160	ret
1161
1162	/* taken from glibc */
1163trace:
1164	pushl %eax
1165	pushl %ecx
1166	pushl %edx
1167	movl 0xc(%esp), %eax
1168	movl 0x4(%ebp), %edx
1169	subl $MCOUNT_INSN_SIZE, %eax
1170
1171	call *ftrace_trace_function
1172
1173	popl %edx
1174	popl %ecx
1175	popl %eax
1176	jmp ftrace_stub
1177END(mcount)
1178#endif /* CONFIG_DYNAMIC_FTRACE */
1179#endif /* CONFIG_FUNCTION_TRACER */
1180
1181#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1182ENTRY(ftrace_graph_caller)
1183	cmpl $0, function_trace_stop
1184	jne ftrace_stub
1185
1186	pushl %eax
1187	pushl %ecx
1188	pushl %edx
1189	movl 0xc(%esp), %edx
1190	lea 0x4(%ebp), %eax
1191	movl (%ebp), %ecx
1192	subl $MCOUNT_INSN_SIZE, %edx
1193	call prepare_ftrace_return
1194	popl %edx
1195	popl %ecx
1196	popl %eax
1197	ret
1198END(ftrace_graph_caller)
1199
1200.globl return_to_handler
1201return_to_handler:
1202	pushl %eax
1203	pushl %edx
1204	movl %ebp, %eax
1205	call ftrace_return_to_handler
1206	movl %eax, %ecx
1207	popl %edx
1208	popl %eax
1209	jmp *%ecx
1210#endif
1211
1212.section .rodata,"a"
1213#include "syscall_table_32.S"
1214
1215syscall_table_size=(.-sys_call_table)
1216
1217/*
1218 * Some functions should be protected against kprobes
1219 */
1220	.pushsection .kprobes.text, "ax"
1221
1222ENTRY(page_fault)
1223	RING0_EC_FRAME
1224	pushl_cfi $do_page_fault
1225	ALIGN
1226error_code:
1227	/* the function address is in %gs's slot on the stack */
1228	pushl_cfi %fs
1229	/*CFI_REL_OFFSET fs, 0*/
1230	pushl_cfi %es
1231	/*CFI_REL_OFFSET es, 0*/
1232	pushl_cfi %ds
1233	/*CFI_REL_OFFSET ds, 0*/
1234	pushl_cfi %eax
1235	CFI_REL_OFFSET eax, 0
1236	pushl_cfi %ebp
1237	CFI_REL_OFFSET ebp, 0
1238	pushl_cfi %edi
1239	CFI_REL_OFFSET edi, 0
1240	pushl_cfi %esi
1241	CFI_REL_OFFSET esi, 0
1242	pushl_cfi %edx
1243	CFI_REL_OFFSET edx, 0
1244	pushl_cfi %ecx
1245	CFI_REL_OFFSET ecx, 0
1246	pushl_cfi %ebx
1247	CFI_REL_OFFSET ebx, 0
1248	cld
1249	movl $(__KERNEL_PERCPU), %ecx
1250	movl %ecx, %fs
1251	UNWIND_ESPFIX_STACK
1252	GS_TO_REG %ecx
1253	movl PT_GS(%esp), %edi		# get the function address
1254	movl PT_ORIG_EAX(%esp), %edx	# get the error code
1255	movl $-1, PT_ORIG_EAX(%esp)	# no syscall to restart
1256	REG_TO_PTGS %ecx
1257	SET_KERNEL_GS %ecx
1258	movl $(__USER_DS), %ecx
1259	movl %ecx, %ds
1260	movl %ecx, %es
1261	TRACE_IRQS_OFF
1262	movl %esp,%eax			# pt_regs pointer
1263	call *%edi
1264	jmp ret_from_exception
1265	CFI_ENDPROC
1266END(page_fault)
1267
1268/*
1269 * Debug traps and NMI can happen at the one SYSENTER instruction
1270 * that sets up the real kernel stack. Check here, since we can't
1271 * allow the wrong stack to be used.
1272 *
1273 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1274 * already pushed 3 words if it hits on the sysenter instruction:
1275 * eflags, cs and eip.
1276 *
1277 * We just load the right stack, and push the three (known) values
1278 * by hand onto the new stack - while updating the return eip past
1279 * the instruction that would have done it for sysenter.
1280 */
1281.macro FIX_STACK offset ok label
1282	cmpw $__KERNEL_CS, 4(%esp)
1283	jne \ok
1284\label:
1285	movl TSS_sysenter_sp0 + \offset(%esp), %esp
1286	CFI_DEF_CFA esp, 0
1287	CFI_UNDEFINED eip
1288	pushfl_cfi
1289	pushl_cfi $__KERNEL_CS
1290	pushl_cfi $sysenter_past_esp
1291	CFI_REL_OFFSET eip, 0
1292.endm
1293
1294ENTRY(debug)
1295	RING0_INT_FRAME
1296	cmpl $ia32_sysenter_target,(%esp)
1297	jne debug_stack_correct
1298	FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1299debug_stack_correct:
1300	pushl_cfi $-1			# mark this as an int
1301	SAVE_ALL
1302	TRACE_IRQS_OFF
1303	xorl %edx,%edx			# error code 0
1304	movl %esp,%eax			# pt_regs pointer
1305	call do_debug
1306	jmp ret_from_exception
1307	CFI_ENDPROC
1308END(debug)
1309
1310/*
1311 * NMI is doubly nasty. It can happen _while_ we're handling
1312 * a debug fault, and the debug fault hasn't yet been able to
1313 * clear up the stack. So we first check whether we got  an
1314 * NMI on the sysenter entry path, but after that we need to
1315 * check whether we got an NMI on the debug path where the debug
1316 * fault happened on the sysenter path.
1317 */
1318ENTRY(nmi)
1319	RING0_INT_FRAME
1320	pushl_cfi %eax
1321	movl %ss, %eax
1322	cmpw $__ESPFIX_SS, %ax
1323	popl_cfi %eax
1324	je nmi_espfix_stack
1325	cmpl $ia32_sysenter_target,(%esp)
1326	je nmi_stack_fixup
1327	pushl_cfi %eax
1328	movl %esp,%eax
1329	/* Do not access memory above the end of our stack page,
1330	 * it might not exist.
1331	 */
1332	andl $(THREAD_SIZE-1),%eax
1333	cmpl $(THREAD_SIZE-20),%eax
1334	popl_cfi %eax
1335	jae nmi_stack_correct
1336	cmpl $ia32_sysenter_target,12(%esp)
1337	je nmi_debug_stack_check
1338nmi_stack_correct:
1339	/* We have a RING0_INT_FRAME here */
1340	pushl_cfi %eax
1341	SAVE_ALL
1342	xorl %edx,%edx		# zero error code
1343	movl %esp,%eax		# pt_regs pointer
1344	call do_nmi
1345	jmp restore_all_notrace
1346	CFI_ENDPROC
1347
1348nmi_stack_fixup:
1349	RING0_INT_FRAME
1350	FIX_STACK 12, nmi_stack_correct, 1
1351	jmp nmi_stack_correct
1352
1353nmi_debug_stack_check:
1354	/* We have a RING0_INT_FRAME here */
1355	cmpw $__KERNEL_CS,16(%esp)
1356	jne nmi_stack_correct
1357	cmpl $debug,(%esp)
1358	jb nmi_stack_correct
1359	cmpl $debug_esp_fix_insn,(%esp)
1360	ja nmi_stack_correct
1361	FIX_STACK 24, nmi_stack_correct, 1
1362	jmp nmi_stack_correct
1363
1364nmi_espfix_stack:
1365	/* We have a RING0_INT_FRAME here.
1366	 *
1367	 * create the pointer to lss back
1368	 */
1369	pushl_cfi %ss
1370	pushl_cfi %esp
1371	addl $4, (%esp)
1372	/* copy the iret frame of 12 bytes */
1373	.rept 3
1374	pushl_cfi 16(%esp)
1375	.endr
1376	pushl_cfi %eax
1377	SAVE_ALL
1378	FIXUP_ESPFIX_STACK		# %eax == %esp
1379	xorl %edx,%edx			# zero error code
1380	call do_nmi
1381	RESTORE_REGS
1382	lss 12+4(%esp), %esp		# back to espfix stack
1383	CFI_ADJUST_CFA_OFFSET -24
1384	jmp irq_return
1385	CFI_ENDPROC
1386END(nmi)
1387
1388ENTRY(int3)
1389	RING0_INT_FRAME
1390	pushl_cfi $-1			# mark this as an int
1391	SAVE_ALL
1392	TRACE_IRQS_OFF
1393	xorl %edx,%edx		# zero error code
1394	movl %esp,%eax		# pt_regs pointer
1395	call do_int3
1396	jmp ret_from_exception
1397	CFI_ENDPROC
1398END(int3)
1399
1400ENTRY(general_protection)
1401	RING0_EC_FRAME
1402	pushl_cfi $do_general_protection
1403	jmp error_code
1404	CFI_ENDPROC
1405END(general_protection)
1406
1407#ifdef CONFIG_KVM_GUEST
1408ENTRY(async_page_fault)
1409	RING0_EC_FRAME
1410	pushl_cfi $do_async_page_fault
1411	jmp error_code
1412	CFI_ENDPROC
1413END(async_page_fault)
1414#endif
1415
1416/*
1417 * End of kprobes section
1418 */
1419	.popsection