• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <linux/err.h>
46#include <asm/thread_info.h>
47#include <asm/irqflags.h>
48#include <asm/errno.h>
49#include <asm/segment.h>
50#include <asm/smp.h>
51#include <asm/page_types.h>
52#include <asm/percpu.h>
53#include <asm/dwarf2.h>
54#include <asm/processor-flags.h>
55#include <asm/ftrace.h>
56#include <asm/irq_vectors.h>
57#include <asm/cpufeature.h>
58#include <asm/alternative-asm.h>
59#include <asm/asm.h>
60#include <asm/smap.h>
61
62/* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
63#include <linux/elf-em.h>
64#define AUDIT_ARCH_I386		(EM_386|__AUDIT_ARCH_LE)
65#define __AUDIT_ARCH_LE	   0x40000000
66
67#ifndef CONFIG_AUDITSYSCALL
68#define sysenter_audit	syscall_trace_entry
69#define sysexit_audit	syscall_exit_work
70#endif
71
72	.section .entry.text, "ax"
73
74/*
75 * We use macros for low-level operations which need to be overridden
76 * for paravirtualization.  The following will never clobber any registers:
77 *   INTERRUPT_RETURN (aka. "iret")
78 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
79 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
80 *
81 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
82 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
83 * Allowing a register to be clobbered can shrink the paravirt replacement
84 * enough to patch inline, increasing performance.
85 */
86
87#ifdef CONFIG_PREEMPT
88#define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
89#else
90#define preempt_stop(clobbers)
91#define resume_kernel		restore_all
92#endif
93
94.macro TRACE_IRQS_IRET
95#ifdef CONFIG_TRACE_IRQFLAGS
96	testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)     # interrupts off?
97	jz 1f
98	TRACE_IRQS_ON
991:
100#endif
101.endm
102
103/*
104 * User gs save/restore
105 *
106 * %gs is used for userland TLS and kernel only uses it for stack
107 * canary which is required to be at %gs:20 by gcc.  Read the comment
108 * at the top of stackprotector.h for more info.
109 *
110 * Local labels 98 and 99 are used.
111 */
112#ifdef CONFIG_X86_32_LAZY_GS
113
114 /* unfortunately push/pop can't be no-op */
115.macro PUSH_GS
116	pushl_cfi $0
117.endm
118.macro POP_GS pop=0
119	addl $(4 + \pop), %esp
120	CFI_ADJUST_CFA_OFFSET -(4 + \pop)
121.endm
122.macro POP_GS_EX
123.endm
124
125 /* all the rest are no-op */
126.macro PTGS_TO_GS
127.endm
128.macro PTGS_TO_GS_EX
129.endm
130.macro GS_TO_REG reg
131.endm
132.macro REG_TO_PTGS reg
133.endm
134.macro SET_KERNEL_GS reg
135.endm
136
137#else	/* CONFIG_X86_32_LAZY_GS */
138
139.macro PUSH_GS
140	pushl_cfi %gs
141	/*CFI_REL_OFFSET gs, 0*/
142.endm
143
144.macro POP_GS pop=0
14598:	popl_cfi %gs
146	/*CFI_RESTORE gs*/
147  .if \pop <> 0
148	add $\pop, %esp
149	CFI_ADJUST_CFA_OFFSET -\pop
150  .endif
151.endm
152.macro POP_GS_EX
153.pushsection .fixup, "ax"
15499:	movl $0, (%esp)
155	jmp 98b
156.popsection
157	_ASM_EXTABLE(98b,99b)
158.endm
159
160.macro PTGS_TO_GS
16198:	mov PT_GS(%esp), %gs
162.endm
163.macro PTGS_TO_GS_EX
164.pushsection .fixup, "ax"
16599:	movl $0, PT_GS(%esp)
166	jmp 98b
167.popsection
168	_ASM_EXTABLE(98b,99b)
169.endm
170
171.macro GS_TO_REG reg
172	movl %gs, \reg
173	/*CFI_REGISTER gs, \reg*/
174.endm
175.macro REG_TO_PTGS reg
176	movl \reg, PT_GS(%esp)
177	/*CFI_REL_OFFSET gs, PT_GS*/
178.endm
179.macro SET_KERNEL_GS reg
180	movl $(__KERNEL_STACK_CANARY), \reg
181	movl \reg, %gs
182.endm
183
184#endif	/* CONFIG_X86_32_LAZY_GS */
185
186.macro SAVE_ALL
187	cld
188	PUSH_GS
189	pushl_cfi %fs
190	/*CFI_REL_OFFSET fs, 0;*/
191	pushl_cfi %es
192	/*CFI_REL_OFFSET es, 0;*/
193	pushl_cfi %ds
194	/*CFI_REL_OFFSET ds, 0;*/
195	pushl_cfi %eax
196	CFI_REL_OFFSET eax, 0
197	pushl_cfi %ebp
198	CFI_REL_OFFSET ebp, 0
199	pushl_cfi %edi
200	CFI_REL_OFFSET edi, 0
201	pushl_cfi %esi
202	CFI_REL_OFFSET esi, 0
203	pushl_cfi %edx
204	CFI_REL_OFFSET edx, 0
205	pushl_cfi %ecx
206	CFI_REL_OFFSET ecx, 0
207	pushl_cfi %ebx
208	CFI_REL_OFFSET ebx, 0
209	movl $(__USER_DS), %edx
210	movl %edx, %ds
211	movl %edx, %es
212	movl $(__KERNEL_PERCPU), %edx
213	movl %edx, %fs
214	SET_KERNEL_GS %edx
215.endm
216
217.macro RESTORE_INT_REGS
218	popl_cfi %ebx
219	CFI_RESTORE ebx
220	popl_cfi %ecx
221	CFI_RESTORE ecx
222	popl_cfi %edx
223	CFI_RESTORE edx
224	popl_cfi %esi
225	CFI_RESTORE esi
226	popl_cfi %edi
227	CFI_RESTORE edi
228	popl_cfi %ebp
229	CFI_RESTORE ebp
230	popl_cfi %eax
231	CFI_RESTORE eax
232.endm
233
234.macro RESTORE_REGS pop=0
235	RESTORE_INT_REGS
2361:	popl_cfi %ds
237	/*CFI_RESTORE ds;*/
2382:	popl_cfi %es
239	/*CFI_RESTORE es;*/
2403:	popl_cfi %fs
241	/*CFI_RESTORE fs;*/
242	POP_GS \pop
243.pushsection .fixup, "ax"
2444:	movl $0, (%esp)
245	jmp 1b
2465:	movl $0, (%esp)
247	jmp 2b
2486:	movl $0, (%esp)
249	jmp 3b
250.popsection
251	_ASM_EXTABLE(1b,4b)
252	_ASM_EXTABLE(2b,5b)
253	_ASM_EXTABLE(3b,6b)
254	POP_GS_EX
255.endm
256
257.macro RING0_INT_FRAME
258	CFI_STARTPROC simple
259	CFI_SIGNAL_FRAME
260	CFI_DEF_CFA esp, 3*4
261	/*CFI_OFFSET cs, -2*4;*/
262	CFI_OFFSET eip, -3*4
263.endm
264
265.macro RING0_EC_FRAME
266	CFI_STARTPROC simple
267	CFI_SIGNAL_FRAME
268	CFI_DEF_CFA esp, 4*4
269	/*CFI_OFFSET cs, -2*4;*/
270	CFI_OFFSET eip, -3*4
271.endm
272
273.macro RING0_PTREGS_FRAME
274	CFI_STARTPROC simple
275	CFI_SIGNAL_FRAME
276	CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
277	/*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
278	CFI_OFFSET eip, PT_EIP-PT_OLDESP
279	/*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
280	/*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
281	CFI_OFFSET eax, PT_EAX-PT_OLDESP
282	CFI_OFFSET ebp, PT_EBP-PT_OLDESP
283	CFI_OFFSET edi, PT_EDI-PT_OLDESP
284	CFI_OFFSET esi, PT_ESI-PT_OLDESP
285	CFI_OFFSET edx, PT_EDX-PT_OLDESP
286	CFI_OFFSET ecx, PT_ECX-PT_OLDESP
287	CFI_OFFSET ebx, PT_EBX-PT_OLDESP
288.endm
289
290ENTRY(ret_from_fork)
291	CFI_STARTPROC
292	pushl_cfi %eax
293	call schedule_tail
294	GET_THREAD_INFO(%ebp)
295	popl_cfi %eax
296	pushl_cfi $0x0202		# Reset kernel eflags
297	popfl_cfi
298	jmp syscall_exit
299	CFI_ENDPROC
300END(ret_from_fork)
301
302ENTRY(ret_from_kernel_thread)
303	CFI_STARTPROC
304	pushl_cfi %eax
305	call schedule_tail
306	GET_THREAD_INFO(%ebp)
307	popl_cfi %eax
308	pushl_cfi $0x0202		# Reset kernel eflags
309	popfl_cfi
310	movl PT_EBP(%esp),%eax
311	call *PT_EBX(%esp)
312	movl $0,PT_EAX(%esp)
313	jmp syscall_exit
314	CFI_ENDPROC
315ENDPROC(ret_from_kernel_thread)
316
317/*
318 * Interrupt exit functions should be protected against kprobes
319 */
320	.pushsection .kprobes.text, "ax"
321/*
322 * Return to user mode is not as complex as all this looks,
323 * but we want the default path for a system call return to
324 * go as quickly as possible which is why some of this is
325 * less clear than it otherwise should be.
326 */
327
328	# userspace resumption stub bypassing syscall exit tracing
329	ALIGN
330	RING0_PTREGS_FRAME
331ret_from_exception:
332	preempt_stop(CLBR_ANY)
333ret_from_intr:
334	GET_THREAD_INFO(%ebp)
335#ifdef CONFIG_VM86
336	movl PT_EFLAGS(%esp), %eax	# mix EFLAGS and CS
337	movb PT_CS(%esp), %al
338	andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
339#else
340	/*
341	 * We can be coming here from child spawned by kernel_thread().
342	 */
343	movl PT_CS(%esp), %eax
344	andl $SEGMENT_RPL_MASK, %eax
345#endif
346	cmpl $USER_RPL, %eax
347	jb resume_kernel		# not returning to v8086 or userspace
348
349ENTRY(resume_userspace)
350	LOCKDEP_SYS_EXIT
351 	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
352					# setting need_resched or sigpending
353					# between sampling and the iret
354	TRACE_IRQS_OFF
355	movl TI_flags(%ebp), %ecx
356	andl $_TIF_WORK_MASK, %ecx	# is there any work to be done on
357					# int/exception return?
358	jne work_pending
359	jmp restore_all
360END(ret_from_exception)
361
362#ifdef CONFIG_PREEMPT
363ENTRY(resume_kernel)
364	DISABLE_INTERRUPTS(CLBR_ANY)
365	cmpl $0,TI_preempt_count(%ebp)	# non-zero preempt_count ?
366	jnz restore_all
367need_resched:
368	movl TI_flags(%ebp), %ecx	# need_resched set ?
369	testb $_TIF_NEED_RESCHED, %cl
370	jz restore_all
371	testl $X86_EFLAGS_IF,PT_EFLAGS(%esp)	# interrupts off (exception path) ?
372	jz restore_all
373	call preempt_schedule_irq
374	jmp need_resched
375END(resume_kernel)
376#endif
377	CFI_ENDPROC
378/*
379 * End of kprobes section
380 */
381	.popsection
382
383/* SYSENTER_RETURN points to after the "sysenter" instruction in
384   the vsyscall page.  See vsyscall-sysentry.S, which defines the symbol.  */
385
386	# sysenter call handler stub
387ENTRY(ia32_sysenter_target)
388	CFI_STARTPROC simple
389	CFI_SIGNAL_FRAME
390	CFI_DEF_CFA esp, 0
391	CFI_REGISTER esp, ebp
392	movl TSS_sysenter_sp0(%esp),%esp
393sysenter_past_esp:
394	/*
395	 * Interrupts are disabled here, but we can't trace it until
396	 * enough kernel state to call TRACE_IRQS_OFF can be called - but
397	 * we immediately enable interrupts at that point anyway.
398	 */
399	pushl_cfi $__USER_DS
400	/*CFI_REL_OFFSET ss, 0*/
401	pushl_cfi %ebp
402	CFI_REL_OFFSET esp, 0
403	pushfl_cfi
404	orl $X86_EFLAGS_IF, (%esp)
405	pushl_cfi $__USER_CS
406	/*CFI_REL_OFFSET cs, 0*/
407	/*
408	 * Push current_thread_info()->sysenter_return to the stack.
409	 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
410	 * pushed above; +8 corresponds to copy_thread's esp0 setting.
411	 */
412	pushl_cfi ((TI_sysenter_return)-THREAD_SIZE+8+4*4)(%esp)
413	CFI_REL_OFFSET eip, 0
414
415	pushl_cfi %eax
416	SAVE_ALL
417	ENABLE_INTERRUPTS(CLBR_NONE)
418
419/*
420 * Load the potential sixth argument from user stack.
421 * Careful about security.
422 */
423	cmpl $__PAGE_OFFSET-3,%ebp
424	jae syscall_fault
425	ASM_STAC
4261:	movl (%ebp),%ebp
427	ASM_CLAC
428	movl %ebp,PT_EBP(%esp)
429	_ASM_EXTABLE(1b,syscall_fault)
430
431	GET_THREAD_INFO(%ebp)
432
433	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
434	jnz sysenter_audit
435sysenter_do_call:
436	cmpl $(NR_syscalls), %eax
437	jae syscall_badsys
438	call *sys_call_table(,%eax,4)
439	movl %eax,PT_EAX(%esp)
440	LOCKDEP_SYS_EXIT
441	DISABLE_INTERRUPTS(CLBR_ANY)
442	TRACE_IRQS_OFF
443	movl TI_flags(%ebp), %ecx
444	testl $_TIF_ALLWORK_MASK, %ecx
445	jne sysexit_audit
446sysenter_exit:
447/* if something modifies registers it must also disable sysexit */
448	movl PT_EIP(%esp), %edx
449	movl PT_OLDESP(%esp), %ecx
450	xorl %ebp,%ebp
451	TRACE_IRQS_ON
4521:	mov  PT_FS(%esp), %fs
453	PTGS_TO_GS
454	ENABLE_INTERRUPTS_SYSEXIT
455
456#ifdef CONFIG_AUDITSYSCALL
457sysenter_audit:
458	testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
459	jnz syscall_trace_entry
460	addl $4,%esp
461	CFI_ADJUST_CFA_OFFSET -4
462	/* %esi already in 8(%esp)	   6th arg: 4th syscall arg */
463	/* %edx already in 4(%esp)	   5th arg: 3rd syscall arg */
464	/* %ecx already in 0(%esp)	   4th arg: 2nd syscall arg */
465	movl %ebx,%ecx			/* 3rd arg: 1st syscall arg */
466	movl %eax,%edx			/* 2nd arg: syscall number */
467	movl $AUDIT_ARCH_I386,%eax	/* 1st arg: audit arch */
468	call __audit_syscall_entry
469	pushl_cfi %ebx
470	movl PT_EAX(%esp),%eax		/* reload syscall number */
471	jmp sysenter_do_call
472
473sysexit_audit:
474	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
475	jne syscall_exit_work
476	TRACE_IRQS_ON
477	ENABLE_INTERRUPTS(CLBR_ANY)
478	movl %eax,%edx		/* second arg, syscall return value */
479	cmpl $-MAX_ERRNO,%eax	/* is it an error ? */
480	setbe %al		/* 1 if so, 0 if not */
481	movzbl %al,%eax		/* zero-extend that */
482	call __audit_syscall_exit
483	DISABLE_INTERRUPTS(CLBR_ANY)
484	TRACE_IRQS_OFF
485	movl TI_flags(%ebp), %ecx
486	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %ecx
487	jne syscall_exit_work
488	movl PT_EAX(%esp),%eax	/* reload syscall return value */
489	jmp sysenter_exit
490#endif
491
492	CFI_ENDPROC
493.pushsection .fixup,"ax"
4942:	movl $0,PT_FS(%esp)
495	jmp 1b
496.popsection
497	_ASM_EXTABLE(1b,2b)
498	PTGS_TO_GS_EX
499ENDPROC(ia32_sysenter_target)
500
501/*
502 * syscall stub including irq exit should be protected against kprobes
503 */
504	.pushsection .kprobes.text, "ax"
505	# system call handler stub
506ENTRY(system_call)
507	RING0_INT_FRAME			# can't unwind into user space anyway
508	ASM_CLAC
509	pushl_cfi %eax			# save orig_eax
510	SAVE_ALL
511	GET_THREAD_INFO(%ebp)
512					# system call tracing in operation / emulation
513	testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
514	jnz syscall_trace_entry
515	cmpl $(NR_syscalls), %eax
516	jae syscall_badsys
517syscall_call:
518	call *sys_call_table(,%eax,4)
519	movl %eax,PT_EAX(%esp)		# store the return value
520syscall_exit:
521	LOCKDEP_SYS_EXIT
522	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
523					# setting need_resched or sigpending
524					# between sampling and the iret
525	TRACE_IRQS_OFF
526	movl TI_flags(%ebp), %ecx
527	testl $_TIF_ALLWORK_MASK, %ecx	# current->work
528	jne syscall_exit_work
529
530restore_all:
531	TRACE_IRQS_IRET
532restore_all_notrace:
533#ifdef CONFIG_X86_ESPFIX32
534	movl PT_EFLAGS(%esp), %eax	# mix EFLAGS, SS and CS
535	# Warning: PT_OLDSS(%esp) contains the wrong/random values if we
536	# are returning to the kernel.
537	# See comments in process.c:copy_thread() for details.
538	movb PT_OLDSS(%esp), %ah
539	movb PT_CS(%esp), %al
540	andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
541	cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
542	CFI_REMEMBER_STATE
543	je ldt_ss			# returning to user-space with LDT SS
544#endif
545restore_nocheck:
546	RESTORE_REGS 4			# skip orig_eax/error_code
547irq_return:
548	INTERRUPT_RETURN
549.section .fixup,"ax"
550ENTRY(iret_exc)
551	pushl $0			# no error code
552	pushl $do_iret_error
553	jmp error_code
554.previous
555	_ASM_EXTABLE(irq_return,iret_exc)
556
557#ifdef CONFIG_X86_ESPFIX32
558	CFI_RESTORE_STATE
559ldt_ss:
560#ifdef CONFIG_PARAVIRT
561	/*
562	 * The kernel can't run on a non-flat stack if paravirt mode
563	 * is active.  Rather than try to fixup the high bits of
564	 * ESP, bypass this code entirely.  This may break DOSemu
565	 * and/or Wine support in a paravirt VM, although the option
566	 * is still available to implement the setting of the high
567	 * 16-bits in the INTERRUPT_RETURN paravirt-op.
568	 */
569	cmpl $0, pv_info+PARAVIRT_enabled
570	jne restore_nocheck
571#endif
572
573/*
574 * Setup and switch to ESPFIX stack
575 *
576 * We're returning to userspace with a 16 bit stack. The CPU will not
577 * restore the high word of ESP for us on executing iret... This is an
578 * "official" bug of all the x86-compatible CPUs, which we can work
579 * around to make dosemu and wine happy. We do this by preloading the
580 * high word of ESP with the high word of the userspace ESP while
581 * compensating for the offset by changing to the ESPFIX segment with
582 * a base address that matches for the difference.
583 */
584#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
585	mov %esp, %edx			/* load kernel esp */
586	mov PT_OLDESP(%esp), %eax	/* load userspace esp */
587	mov %dx, %ax			/* eax: new kernel esp */
588	sub %eax, %edx			/* offset (low word is 0) */
589	shr $16, %edx
590	mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
591	mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
592	pushl_cfi $__ESPFIX_SS
593	pushl_cfi %eax			/* new kernel esp */
594	/* Disable interrupts, but do not irqtrace this section: we
595	 * will soon execute iret and the tracer was already set to
596	 * the irqstate after the iret */
597	DISABLE_INTERRUPTS(CLBR_EAX)
598	lss (%esp), %esp		/* switch to espfix segment */
599	CFI_ADJUST_CFA_OFFSET -8
600	jmp restore_nocheck
601#endif
602	CFI_ENDPROC
603ENDPROC(system_call)
604
605	# perform work that needs to be done immediately before resumption
606	ALIGN
607	RING0_PTREGS_FRAME		# can't unwind into user space anyway
608work_pending:
609	testb $_TIF_NEED_RESCHED, %cl
610	jz work_notifysig
611work_resched:
612	call schedule
613	LOCKDEP_SYS_EXIT
614	DISABLE_INTERRUPTS(CLBR_ANY)	# make sure we don't miss an interrupt
615					# setting need_resched or sigpending
616					# between sampling and the iret
617	TRACE_IRQS_OFF
618	movl TI_flags(%ebp), %ecx
619	andl $_TIF_WORK_MASK, %ecx	# is there any work to be done other
620					# than syscall tracing?
621	jz restore_all
622	testb $_TIF_NEED_RESCHED, %cl
623	jnz work_resched
624
625work_notifysig:				# deal with pending signals and
626					# notify-resume requests
627#ifdef CONFIG_VM86
628	testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
629	movl %esp, %eax
630	jne work_notifysig_v86		# returning to kernel-space or
631					# vm86-space
6321:
633#else
634	movl %esp, %eax
635#endif
636	TRACE_IRQS_ON
637	ENABLE_INTERRUPTS(CLBR_NONE)
638	movb PT_CS(%esp), %bl
639	andb $SEGMENT_RPL_MASK, %bl
640	cmpb $USER_RPL, %bl
641	jb resume_kernel
642	xorl %edx, %edx
643	call do_notify_resume
644	jmp resume_userspace
645
646#ifdef CONFIG_VM86
647	ALIGN
648work_notifysig_v86:
649	pushl_cfi %ecx			# save ti_flags for do_notify_resume
650	call save_v86_state		# %eax contains pt_regs pointer
651	popl_cfi %ecx
652	movl %eax, %esp
653	jmp 1b
654#endif
655END(work_pending)
656
657	# perform syscall exit tracing
658	ALIGN
659syscall_trace_entry:
660	movl $-ENOSYS,PT_EAX(%esp)
661	movl %esp, %eax
662	call syscall_trace_enter
663	/* What it returned is what we'll actually use.  */
664	cmpl $(NR_syscalls), %eax
665	jnae syscall_call
666	jmp syscall_exit
667END(syscall_trace_entry)
668
669	# perform syscall exit tracing
670	ALIGN
671syscall_exit_work:
672	testl $_TIF_WORK_SYSCALL_EXIT, %ecx
673	jz work_pending
674	TRACE_IRQS_ON
675	ENABLE_INTERRUPTS(CLBR_ANY)	# could let syscall_trace_leave() call
676					# schedule() instead
677	movl %esp, %eax
678	call syscall_trace_leave
679	jmp resume_userspace
680END(syscall_exit_work)
681	CFI_ENDPROC
682
683	RING0_INT_FRAME			# can't unwind into user space anyway
684syscall_fault:
685	ASM_CLAC
686	GET_THREAD_INFO(%ebp)
687	movl $-EFAULT,PT_EAX(%esp)
688	jmp resume_userspace
689END(syscall_fault)
690
691syscall_badsys:
692	movl $-ENOSYS,PT_EAX(%esp)
693	jmp resume_userspace
694END(syscall_badsys)
695	CFI_ENDPROC
696/*
697 * End of kprobes section
698 */
699	.popsection
700
701.macro FIXUP_ESPFIX_STACK
702/*
703 * Switch back for ESPFIX stack to the normal zerobased stack
704 *
705 * We can't call C functions using the ESPFIX stack. This code reads
706 * the high word of the segment base from the GDT and swiches to the
707 * normal stack and adjusts ESP with the matching offset.
708 */
709#ifdef CONFIG_X86_ESPFIX32
710	/* fixup the stack */
711	mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
712	mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
713	shl $16, %eax
714	addl %esp, %eax			/* the adjusted stack pointer */
715	pushl_cfi $__KERNEL_DS
716	pushl_cfi %eax
717	lss (%esp), %esp		/* switch to the normal stack segment */
718	CFI_ADJUST_CFA_OFFSET -8
719#endif
720.endm
721.macro UNWIND_ESPFIX_STACK
722#ifdef CONFIG_X86_ESPFIX32
723	movl %ss, %eax
724	/* see if on espfix stack */
725	cmpw $__ESPFIX_SS, %ax
726	jne 27f
727	movl $__KERNEL_DS, %eax
728	movl %eax, %ds
729	movl %eax, %es
730	/* switch to normal stack */
731	FIXUP_ESPFIX_STACK
73227:
733#endif
734.endm
735
736/*
737 * Build the entry stubs and pointer table with some assembler magic.
738 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
739 * single cache line on all modern x86 implementations.
740 */
741.section .init.rodata,"a"
742ENTRY(interrupt)
743.section .entry.text, "ax"
744	.p2align 5
745	.p2align CONFIG_X86_L1_CACHE_SHIFT
746ENTRY(irq_entries_start)
747	RING0_INT_FRAME
748vector=FIRST_EXTERNAL_VECTOR
749.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
750	.balign 32
751  .rept	7
752    .if vector < NR_VECTORS
753      .if vector <> FIRST_EXTERNAL_VECTOR
754	CFI_ADJUST_CFA_OFFSET -4
755      .endif
7561:	pushl_cfi $(~vector+0x80)	/* Note: always in signed byte range */
757      .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
758	jmp 2f
759      .endif
760      .previous
761	.long 1b
762      .section .entry.text, "ax"
763vector=vector+1
764    .endif
765  .endr
7662:	jmp common_interrupt
767.endr
768END(irq_entries_start)
769
770.previous
771END(interrupt)
772.previous
773
774/*
775 * the CPU automatically disables interrupts when executing an IRQ vector,
776 * so IRQ-flags tracing has to follow that:
777 */
778	.p2align CONFIG_X86_L1_CACHE_SHIFT
779common_interrupt:
780	ASM_CLAC
781	addl $-0x80,(%esp)	/* Adjust vector into the [-256,-1] range */
782	SAVE_ALL
783	TRACE_IRQS_OFF
784	movl %esp,%eax
785	call do_IRQ
786	jmp ret_from_intr
787ENDPROC(common_interrupt)
788	CFI_ENDPROC
789
790/*
791 *  Irq entries should be protected against kprobes
792 */
793	.pushsection .kprobes.text, "ax"
794#define BUILD_INTERRUPT3(name, nr, fn)	\
795ENTRY(name)				\
796	RING0_INT_FRAME;		\
797	ASM_CLAC;			\
798	pushl_cfi $~(nr);		\
799	SAVE_ALL;			\
800	TRACE_IRQS_OFF			\
801	movl %esp,%eax;			\
802	call fn;			\
803	jmp ret_from_intr;		\
804	CFI_ENDPROC;			\
805ENDPROC(name)
806
807#define BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(name, nr, smp_##name)
808
809/* The include is where all of the SMP etc. interrupts come from */
810#include <asm/entry_arch.h>
811
812ENTRY(coprocessor_error)
813	RING0_INT_FRAME
814	ASM_CLAC
815	pushl_cfi $0
816	pushl_cfi $do_coprocessor_error
817	jmp error_code
818	CFI_ENDPROC
819END(coprocessor_error)
820
821ENTRY(simd_coprocessor_error)
822	RING0_INT_FRAME
823	ASM_CLAC
824	pushl_cfi $0
825#ifdef CONFIG_X86_INVD_BUG
826	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
827661:	pushl_cfi $do_general_protection
828662:
829.section .altinstructions,"a"
830	altinstruction_entry 661b, 663f, X86_FEATURE_XMM, 662b-661b, 664f-663f
831.previous
832.section .altinstr_replacement,"ax"
833663:	pushl $do_simd_coprocessor_error
834664:
835.previous
836#else
837	pushl_cfi $do_simd_coprocessor_error
838#endif
839	jmp error_code
840	CFI_ENDPROC
841END(simd_coprocessor_error)
842
843ENTRY(device_not_available)
844	RING0_INT_FRAME
845	ASM_CLAC
846	pushl_cfi $-1			# mark this as an int
847	pushl_cfi $do_device_not_available
848	jmp error_code
849	CFI_ENDPROC
850END(device_not_available)
851
852#ifdef CONFIG_PARAVIRT
853ENTRY(native_iret)
854	iret
855	_ASM_EXTABLE(native_iret, iret_exc)
856END(native_iret)
857
858ENTRY(native_irq_enable_sysexit)
859	sti
860	sysexit
861END(native_irq_enable_sysexit)
862#endif
863
864ENTRY(overflow)
865	RING0_INT_FRAME
866	ASM_CLAC
867	pushl_cfi $0
868	pushl_cfi $do_overflow
869	jmp error_code
870	CFI_ENDPROC
871END(overflow)
872
873ENTRY(bounds)
874	RING0_INT_FRAME
875	ASM_CLAC
876	pushl_cfi $0
877	pushl_cfi $do_bounds
878	jmp error_code
879	CFI_ENDPROC
880END(bounds)
881
882ENTRY(invalid_op)
883	RING0_INT_FRAME
884	ASM_CLAC
885	pushl_cfi $0
886	pushl_cfi $do_invalid_op
887	jmp error_code
888	CFI_ENDPROC
889END(invalid_op)
890
891ENTRY(coprocessor_segment_overrun)
892	RING0_INT_FRAME
893	ASM_CLAC
894	pushl_cfi $0
895	pushl_cfi $do_coprocessor_segment_overrun
896	jmp error_code
897	CFI_ENDPROC
898END(coprocessor_segment_overrun)
899
900ENTRY(invalid_TSS)
901	RING0_EC_FRAME
902	ASM_CLAC
903	pushl_cfi $do_invalid_TSS
904	jmp error_code
905	CFI_ENDPROC
906END(invalid_TSS)
907
908ENTRY(segment_not_present)
909	RING0_EC_FRAME
910	ASM_CLAC
911	pushl_cfi $do_segment_not_present
912	jmp error_code
913	CFI_ENDPROC
914END(segment_not_present)
915
916ENTRY(stack_segment)
917	RING0_EC_FRAME
918	ASM_CLAC
919	pushl_cfi $do_stack_segment
920	jmp error_code
921	CFI_ENDPROC
922END(stack_segment)
923
924ENTRY(alignment_check)
925	RING0_EC_FRAME
926	ASM_CLAC
927	pushl_cfi $do_alignment_check
928	jmp error_code
929	CFI_ENDPROC
930END(alignment_check)
931
932ENTRY(divide_error)
933	RING0_INT_FRAME
934	ASM_CLAC
935	pushl_cfi $0			# no error code
936	pushl_cfi $do_divide_error
937	jmp error_code
938	CFI_ENDPROC
939END(divide_error)
940
941#ifdef CONFIG_X86_MCE
942ENTRY(machine_check)
943	RING0_INT_FRAME
944	ASM_CLAC
945	pushl_cfi $0
946	pushl_cfi machine_check_vector
947	jmp error_code
948	CFI_ENDPROC
949END(machine_check)
950#endif
951
952ENTRY(spurious_interrupt_bug)
953	RING0_INT_FRAME
954	ASM_CLAC
955	pushl_cfi $0
956	pushl_cfi $do_spurious_interrupt_bug
957	jmp error_code
958	CFI_ENDPROC
959END(spurious_interrupt_bug)
960/*
961 * End of kprobes section
962 */
963	.popsection
964
965#ifdef CONFIG_XEN
966/* Xen doesn't set %esp to be precisely what the normal sysenter
967   entrypoint expects, so fix it up before using the normal path. */
968ENTRY(xen_sysenter_target)
969	RING0_INT_FRAME
970	addl $5*4, %esp		/* remove xen-provided frame */
971	CFI_ADJUST_CFA_OFFSET -5*4
972	jmp sysenter_past_esp
973	CFI_ENDPROC
974
975ENTRY(xen_hypervisor_callback)
976	CFI_STARTPROC
977	pushl_cfi $-1 /* orig_ax = -1 => not a system call */
978	SAVE_ALL
979	TRACE_IRQS_OFF
980
981	/* Check to see if we got the event in the critical
982	   region in xen_iret_direct, after we've reenabled
983	   events and checked for pending events.  This simulates
984	   iret instruction's behaviour where it delivers a
985	   pending interrupt when enabling interrupts. */
986	movl PT_EIP(%esp),%eax
987	cmpl $xen_iret_start_crit,%eax
988	jb   1f
989	cmpl $xen_iret_end_crit,%eax
990	jae  1f
991
992	jmp  xen_iret_crit_fixup
993
994ENTRY(xen_do_upcall)
9951:	mov %esp, %eax
996	call xen_evtchn_do_upcall
997	jmp  ret_from_intr
998	CFI_ENDPROC
999ENDPROC(xen_hypervisor_callback)
1000
1001# Hypervisor uses this for application faults while it executes.
1002# We get here for two reasons:
1003#  1. Fault while reloading DS, ES, FS or GS
1004#  2. Fault while executing IRET
1005# Category 1 we fix up by reattempting the load, and zeroing the segment
1006# register if the load fails.
1007# Category 2 we fix up by jumping to do_iret_error. We cannot use the
1008# normal Linux return path in this case because if we use the IRET hypercall
1009# to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1010# We distinguish between categories by maintaining a status value in EAX.
1011ENTRY(xen_failsafe_callback)
1012	CFI_STARTPROC
1013	pushl_cfi %eax
1014	movl $1,%eax
10151:	mov 4(%esp),%ds
10162:	mov 8(%esp),%es
10173:	mov 12(%esp),%fs
10184:	mov 16(%esp),%gs
1019	/* EAX == 0 => Category 1 (Bad segment)
1020	   EAX != 0 => Category 2 (Bad IRET) */
1021	testl %eax,%eax
1022	popl_cfi %eax
1023	lea 16(%esp),%esp
1024	CFI_ADJUST_CFA_OFFSET -16
1025	jz 5f
1026	jmp iret_exc
10275:	pushl_cfi $-1 /* orig_ax = -1 => not a system call */
1028	SAVE_ALL
1029	jmp ret_from_exception
1030	CFI_ENDPROC
1031
1032.section .fixup,"ax"
10336:	xorl %eax,%eax
1034	movl %eax,4(%esp)
1035	jmp 1b
10367:	xorl %eax,%eax
1037	movl %eax,8(%esp)
1038	jmp 2b
10398:	xorl %eax,%eax
1040	movl %eax,12(%esp)
1041	jmp 3b
10429:	xorl %eax,%eax
1043	movl %eax,16(%esp)
1044	jmp 4b
1045.previous
1046	_ASM_EXTABLE(1b,6b)
1047	_ASM_EXTABLE(2b,7b)
1048	_ASM_EXTABLE(3b,8b)
1049	_ASM_EXTABLE(4b,9b)
1050ENDPROC(xen_failsafe_callback)
1051
1052BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1053		xen_evtchn_do_upcall)
1054
1055#endif	/* CONFIG_XEN */
1056
1057#if IS_ENABLED(CONFIG_HYPERV)
1058
1059BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1060	hyperv_vector_handler)
1061
1062#endif /* CONFIG_HYPERV */
1063
1064#ifdef CONFIG_FUNCTION_TRACER
1065#ifdef CONFIG_DYNAMIC_FTRACE
1066
1067ENTRY(mcount)
1068	ret
1069END(mcount)
1070
1071ENTRY(ftrace_caller)
1072	cmpl $0, function_trace_stop
1073	jne  ftrace_stub
1074
1075	pushl %eax
1076	pushl %ecx
1077	pushl %edx
1078	pushl $0	/* Pass NULL as regs pointer */
1079	movl 4*4(%esp), %eax
1080	movl 0x4(%ebp), %edx
1081	leal function_trace_op, %ecx
1082	subl $MCOUNT_INSN_SIZE, %eax
1083
1084.globl ftrace_call
1085ftrace_call:
1086	call ftrace_stub
1087
1088	addl $4,%esp	/* skip NULL pointer */
1089	popl %edx
1090	popl %ecx
1091	popl %eax
1092ftrace_ret:
1093#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1094.globl ftrace_graph_call
1095ftrace_graph_call:
1096	jmp ftrace_stub
1097#endif
1098
1099.globl ftrace_stub
1100ftrace_stub:
1101	ret
1102END(ftrace_caller)
1103
1104ENTRY(ftrace_regs_caller)
1105	pushf	/* push flags before compare (in cs location) */
1106	cmpl $0, function_trace_stop
1107	jne ftrace_restore_flags
1108
1109	/*
1110	 * i386 does not save SS and ESP when coming from kernel.
1111	 * Instead, to get sp, &regs->sp is used (see ptrace.h).
1112	 * Unfortunately, that means eflags must be at the same location
1113	 * as the current return ip is. We move the return ip into the
1114	 * ip location, and move flags into the return ip location.
1115	 */
1116	pushl 4(%esp)	/* save return ip into ip slot */
1117
1118	pushl $0	/* Load 0 into orig_ax */
1119	pushl %gs
1120	pushl %fs
1121	pushl %es
1122	pushl %ds
1123	pushl %eax
1124	pushl %ebp
1125	pushl %edi
1126	pushl %esi
1127	pushl %edx
1128	pushl %ecx
1129	pushl %ebx
1130
1131	movl 13*4(%esp), %eax	/* Get the saved flags */
1132	movl %eax, 14*4(%esp)	/* Move saved flags into regs->flags location */
1133				/* clobbering return ip */
1134	movl $__KERNEL_CS,13*4(%esp)
1135
1136	movl 12*4(%esp), %eax	/* Load ip (1st parameter) */
1137	subl $MCOUNT_INSN_SIZE, %eax	/* Adjust ip */
1138	movl 0x4(%ebp), %edx	/* Load parent ip (2nd parameter) */
1139	leal function_trace_op, %ecx /* Save ftrace_pos in 3rd parameter */
1140	pushl %esp		/* Save pt_regs as 4th parameter */
1141
1142GLOBAL(ftrace_regs_call)
1143	call ftrace_stub
1144
1145	addl $4, %esp		/* Skip pt_regs */
1146	movl 14*4(%esp), %eax	/* Move flags back into cs */
1147	movl %eax, 13*4(%esp)	/* Needed to keep addl from modifying flags */
1148	movl 12*4(%esp), %eax	/* Get return ip from regs->ip */
1149	movl %eax, 14*4(%esp)	/* Put return ip back for ret */
1150
1151	popl %ebx
1152	popl %ecx
1153	popl %edx
1154	popl %esi
1155	popl %edi
1156	popl %ebp
1157	popl %eax
1158	popl %ds
1159	popl %es
1160	popl %fs
1161	popl %gs
1162	addl $8, %esp		/* Skip orig_ax and ip */
1163	popf			/* Pop flags at end (no addl to corrupt flags) */
1164	jmp ftrace_ret
1165
1166ftrace_restore_flags:
1167	popf
1168	jmp  ftrace_stub
1169#else /* ! CONFIG_DYNAMIC_FTRACE */
1170
1171ENTRY(mcount)
1172	cmpl $0, function_trace_stop
1173	jne  ftrace_stub
1174
1175	cmpl $ftrace_stub, ftrace_trace_function
1176	jnz trace
1177#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1178	cmpl $ftrace_stub, ftrace_graph_return
1179	jnz ftrace_graph_caller
1180
1181	cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1182	jnz ftrace_graph_caller
1183#endif
1184.globl ftrace_stub
1185ftrace_stub:
1186	ret
1187
1188	/* taken from glibc */
1189trace:
1190	pushl %eax
1191	pushl %ecx
1192	pushl %edx
1193	movl 0xc(%esp), %eax
1194	movl 0x4(%ebp), %edx
1195	subl $MCOUNT_INSN_SIZE, %eax
1196
1197	call *ftrace_trace_function
1198
1199	popl %edx
1200	popl %ecx
1201	popl %eax
1202	jmp ftrace_stub
1203END(mcount)
1204#endif /* CONFIG_DYNAMIC_FTRACE */
1205#endif /* CONFIG_FUNCTION_TRACER */
1206
1207#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1208ENTRY(ftrace_graph_caller)
1209	pushl %eax
1210	pushl %ecx
1211	pushl %edx
1212	movl 0xc(%esp), %edx
1213	lea 0x4(%ebp), %eax
1214	movl (%ebp), %ecx
1215	subl $MCOUNT_INSN_SIZE, %edx
1216	call prepare_ftrace_return
1217	popl %edx
1218	popl %ecx
1219	popl %eax
1220	ret
1221END(ftrace_graph_caller)
1222
1223.globl return_to_handler
1224return_to_handler:
1225	pushl %eax
1226	pushl %edx
1227	movl %ebp, %eax
1228	call ftrace_return_to_handler
1229	movl %eax, %ecx
1230	popl %edx
1231	popl %eax
1232	jmp *%ecx
1233#endif
1234
1235/*
1236 * Some functions should be protected against kprobes
1237 */
1238	.pushsection .kprobes.text, "ax"
1239
1240ENTRY(page_fault)
1241	RING0_EC_FRAME
1242	ASM_CLAC
1243	pushl_cfi $do_page_fault
1244	ALIGN
1245error_code:
1246	/* the function address is in %gs's slot on the stack */
1247	pushl_cfi %fs
1248	/*CFI_REL_OFFSET fs, 0*/
1249	pushl_cfi %es
1250	/*CFI_REL_OFFSET es, 0*/
1251	pushl_cfi %ds
1252	/*CFI_REL_OFFSET ds, 0*/
1253	pushl_cfi %eax
1254	CFI_REL_OFFSET eax, 0
1255	pushl_cfi %ebp
1256	CFI_REL_OFFSET ebp, 0
1257	pushl_cfi %edi
1258	CFI_REL_OFFSET edi, 0
1259	pushl_cfi %esi
1260	CFI_REL_OFFSET esi, 0
1261	pushl_cfi %edx
1262	CFI_REL_OFFSET edx, 0
1263	pushl_cfi %ecx
1264	CFI_REL_OFFSET ecx, 0
1265	pushl_cfi %ebx
1266	CFI_REL_OFFSET ebx, 0
1267	cld
1268	movl $(__KERNEL_PERCPU), %ecx
1269	movl %ecx, %fs
1270	UNWIND_ESPFIX_STACK
1271	GS_TO_REG %ecx
1272	movl PT_GS(%esp), %edi		# get the function address
1273	movl PT_ORIG_EAX(%esp), %edx	# get the error code
1274	movl $-1, PT_ORIG_EAX(%esp)	# no syscall to restart
1275	REG_TO_PTGS %ecx
1276	SET_KERNEL_GS %ecx
1277	movl $(__USER_DS), %ecx
1278	movl %ecx, %ds
1279	movl %ecx, %es
1280	TRACE_IRQS_OFF
1281	movl %esp,%eax			# pt_regs pointer
1282	call *%edi
1283	jmp ret_from_exception
1284	CFI_ENDPROC
1285END(page_fault)
1286
1287/*
1288 * Debug traps and NMI can happen at the one SYSENTER instruction
1289 * that sets up the real kernel stack. Check here, since we can't
1290 * allow the wrong stack to be used.
1291 *
1292 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1293 * already pushed 3 words if it hits on the sysenter instruction:
1294 * eflags, cs and eip.
1295 *
1296 * We just load the right stack, and push the three (known) values
1297 * by hand onto the new stack - while updating the return eip past
1298 * the instruction that would have done it for sysenter.
1299 */
1300.macro FIX_STACK offset ok label
1301	cmpw $__KERNEL_CS, 4(%esp)
1302	jne \ok
1303\label:
1304	movl TSS_sysenter_sp0 + \offset(%esp), %esp
1305	CFI_DEF_CFA esp, 0
1306	CFI_UNDEFINED eip
1307	pushfl_cfi
1308	pushl_cfi $__KERNEL_CS
1309	pushl_cfi $sysenter_past_esp
1310	CFI_REL_OFFSET eip, 0
1311.endm
1312
1313ENTRY(debug)
1314	RING0_INT_FRAME
1315	ASM_CLAC
1316	cmpl $ia32_sysenter_target,(%esp)
1317	jne debug_stack_correct
1318	FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
1319debug_stack_correct:
1320	pushl_cfi $-1			# mark this as an int
1321	SAVE_ALL
1322	TRACE_IRQS_OFF
1323	xorl %edx,%edx			# error code 0
1324	movl %esp,%eax			# pt_regs pointer
1325	call do_debug
1326	jmp ret_from_exception
1327	CFI_ENDPROC
1328END(debug)
1329
1330/*
1331 * NMI is doubly nasty. It can happen _while_ we're handling
1332 * a debug fault, and the debug fault hasn't yet been able to
1333 * clear up the stack. So we first check whether we got  an
1334 * NMI on the sysenter entry path, but after that we need to
1335 * check whether we got an NMI on the debug path where the debug
1336 * fault happened on the sysenter path.
1337 */
1338ENTRY(nmi)
1339	RING0_INT_FRAME
1340	ASM_CLAC
1341#ifdef CONFIG_X86_ESPFIX32
1342	pushl_cfi %eax
1343	movl %ss, %eax
1344	cmpw $__ESPFIX_SS, %ax
1345	popl_cfi %eax
1346	je nmi_espfix_stack
1347#endif
1348	cmpl $ia32_sysenter_target,(%esp)
1349	je nmi_stack_fixup
1350	pushl_cfi %eax
1351	movl %esp,%eax
1352	/* Do not access memory above the end of our stack page,
1353	 * it might not exist.
1354	 */
1355	andl $(THREAD_SIZE-1),%eax
1356	cmpl $(THREAD_SIZE-20),%eax
1357	popl_cfi %eax
1358	jae nmi_stack_correct
1359	cmpl $ia32_sysenter_target,12(%esp)
1360	je nmi_debug_stack_check
1361nmi_stack_correct:
1362	/* We have a RING0_INT_FRAME here */
1363	pushl_cfi %eax
1364	SAVE_ALL
1365	xorl %edx,%edx		# zero error code
1366	movl %esp,%eax		# pt_regs pointer
1367	call do_nmi
1368	jmp restore_all_notrace
1369	CFI_ENDPROC
1370
1371nmi_stack_fixup:
1372	RING0_INT_FRAME
1373	FIX_STACK 12, nmi_stack_correct, 1
1374	jmp nmi_stack_correct
1375
1376nmi_debug_stack_check:
1377	/* We have a RING0_INT_FRAME here */
1378	cmpw $__KERNEL_CS,16(%esp)
1379	jne nmi_stack_correct
1380	cmpl $debug,(%esp)
1381	jb nmi_stack_correct
1382	cmpl $debug_esp_fix_insn,(%esp)
1383	ja nmi_stack_correct
1384	FIX_STACK 24, nmi_stack_correct, 1
1385	jmp nmi_stack_correct
1386
1387#ifdef CONFIG_X86_ESPFIX32
1388nmi_espfix_stack:
1389	/* We have a RING0_INT_FRAME here.
1390	 *
1391	 * create the pointer to lss back
1392	 */
1393	pushl_cfi %ss
1394	pushl_cfi %esp
1395	addl $4, (%esp)
1396	/* copy the iret frame of 12 bytes */
1397	.rept 3
1398	pushl_cfi 16(%esp)
1399	.endr
1400	pushl_cfi %eax
1401	SAVE_ALL
1402	FIXUP_ESPFIX_STACK		# %eax == %esp
1403	xorl %edx,%edx			# zero error code
1404	call do_nmi
1405	RESTORE_REGS
1406	lss 12+4(%esp), %esp		# back to espfix stack
1407	CFI_ADJUST_CFA_OFFSET -24
1408	jmp irq_return
1409#endif
1410	CFI_ENDPROC
1411END(nmi)
1412
1413ENTRY(int3)
1414	RING0_INT_FRAME
1415	ASM_CLAC
1416	pushl_cfi $-1			# mark this as an int
1417	SAVE_ALL
1418	TRACE_IRQS_OFF
1419	xorl %edx,%edx		# zero error code
1420	movl %esp,%eax		# pt_regs pointer
1421	call do_int3
1422	jmp ret_from_exception
1423	CFI_ENDPROC
1424END(int3)
1425
1426ENTRY(general_protection)
1427	RING0_EC_FRAME
1428	pushl_cfi $do_general_protection
1429	jmp error_code
1430	CFI_ENDPROC
1431END(general_protection)
1432
1433#ifdef CONFIG_KVM_GUEST
1434ENTRY(async_page_fault)
1435	RING0_EC_FRAME
1436	ASM_CLAC
1437	pushl_cfi $do_async_page_fault
1438	jmp error_code
1439	CFI_ENDPROC
1440END(async_page_fault)
1441#endif
1442
1443/*
1444 * End of kprobes section
1445 */
1446	.popsection
1447