xref: /openbmc/linux/arch/x86/entry/entry_32.S (revision 9cfc5c90)
1/*
2 *  Copyright (C) 1991,1992  Linus Torvalds
3 *
4 * entry_32.S contains the system-call and low-level fault and trap handling routines.
5 *
6 * Stack layout while running C code:
7 *	ptrace needs to have all registers on the stack.
8 *	If the order here is changed, it needs to be
9 *	updated in fork.c:copy_process(), signal.c:do_signal(),
10 *	ptrace.c and ptrace.h
11 *
12 *	 0(%esp) - %ebx
13 *	 4(%esp) - %ecx
14 *	 8(%esp) - %edx
15 *	 C(%esp) - %esi
16 *	10(%esp) - %edi
17 *	14(%esp) - %ebp
18 *	18(%esp) - %eax
19 *	1C(%esp) - %ds
20 *	20(%esp) - %es
21 *	24(%esp) - %fs
22 *	28(%esp) - %gs		saved iff !CONFIG_X86_32_LAZY_GS
23 *	2C(%esp) - orig_eax
24 *	30(%esp) - %eip
25 *	34(%esp) - %cs
26 *	38(%esp) - %eflags
27 *	3C(%esp) - %oldesp
28 *	40(%esp) - %oldss
29 */
30
31#include <linux/linkage.h>
32#include <linux/err.h>
33#include <asm/thread_info.h>
34#include <asm/irqflags.h>
35#include <asm/errno.h>
36#include <asm/segment.h>
37#include <asm/smp.h>
38#include <asm/page_types.h>
39#include <asm/percpu.h>
40#include <asm/processor-flags.h>
41#include <asm/ftrace.h>
42#include <asm/irq_vectors.h>
43#include <asm/cpufeature.h>
44#include <asm/alternative-asm.h>
45#include <asm/asm.h>
46#include <asm/smap.h>
47
48	.section .entry.text, "ax"
49
50/*
51 * We use macros for low-level operations which need to be overridden
52 * for paravirtualization.  The following will never clobber any registers:
53 *   INTERRUPT_RETURN (aka. "iret")
54 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
55 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
56 *
57 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
58 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
59 * Allowing a register to be clobbered can shrink the paravirt replacement
60 * enough to patch inline, increasing performance.
61 */
62
63#ifdef CONFIG_PREEMPT
64# define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
65#else
66# define preempt_stop(clobbers)
67# define resume_kernel		restore_all
68#endif
69
70.macro TRACE_IRQS_IRET
71#ifdef CONFIG_TRACE_IRQFLAGS
72	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
73	jz	1f
74	TRACE_IRQS_ON
751:
76#endif
77.endm
78
79/*
80 * User gs save/restore
81 *
82 * %gs is used for userland TLS and kernel only uses it for stack
83 * canary which is required to be at %gs:20 by gcc.  Read the comment
84 * at the top of stackprotector.h for more info.
85 *
86 * Local labels 98 and 99 are used.
87 */
88#ifdef CONFIG_X86_32_LAZY_GS
89
90 /* unfortunately push/pop can't be no-op */
91.macro PUSH_GS
92	pushl	$0
93.endm
94.macro POP_GS pop=0
95	addl	$(4 + \pop), %esp
96.endm
97.macro POP_GS_EX
98.endm
99
100 /* all the rest are no-op */
101.macro PTGS_TO_GS
102.endm
103.macro PTGS_TO_GS_EX
104.endm
105.macro GS_TO_REG reg
106.endm
107.macro REG_TO_PTGS reg
108.endm
109.macro SET_KERNEL_GS reg
110.endm
111
112#else	/* CONFIG_X86_32_LAZY_GS */
113
114.macro PUSH_GS
115	pushl	%gs
116.endm
117
118.macro POP_GS pop=0
11998:	popl	%gs
120  .if \pop <> 0
121	add	$\pop, %esp
122  .endif
123.endm
124.macro POP_GS_EX
125.pushsection .fixup, "ax"
12699:	movl	$0, (%esp)
127	jmp	98b
128.popsection
129	_ASM_EXTABLE(98b, 99b)
130.endm
131
132.macro PTGS_TO_GS
13398:	mov	PT_GS(%esp), %gs
134.endm
135.macro PTGS_TO_GS_EX
136.pushsection .fixup, "ax"
13799:	movl	$0, PT_GS(%esp)
138	jmp	98b
139.popsection
140	_ASM_EXTABLE(98b, 99b)
141.endm
142
143.macro GS_TO_REG reg
144	movl	%gs, \reg
145.endm
146.macro REG_TO_PTGS reg
147	movl	\reg, PT_GS(%esp)
148.endm
149.macro SET_KERNEL_GS reg
150	movl	$(__KERNEL_STACK_CANARY), \reg
151	movl	\reg, %gs
152.endm
153
154#endif /* CONFIG_X86_32_LAZY_GS */
155
156.macro SAVE_ALL pt_regs_ax=%eax
157	cld
158	PUSH_GS
159	pushl	%fs
160	pushl	%es
161	pushl	%ds
162	pushl	\pt_regs_ax
163	pushl	%ebp
164	pushl	%edi
165	pushl	%esi
166	pushl	%edx
167	pushl	%ecx
168	pushl	%ebx
169	movl	$(__USER_DS), %edx
170	movl	%edx, %ds
171	movl	%edx, %es
172	movl	$(__KERNEL_PERCPU), %edx
173	movl	%edx, %fs
174	SET_KERNEL_GS %edx
175.endm
176
177.macro RESTORE_INT_REGS
178	popl	%ebx
179	popl	%ecx
180	popl	%edx
181	popl	%esi
182	popl	%edi
183	popl	%ebp
184	popl	%eax
185.endm
186
187.macro RESTORE_REGS pop=0
188	RESTORE_INT_REGS
1891:	popl	%ds
1902:	popl	%es
1913:	popl	%fs
192	POP_GS \pop
193.pushsection .fixup, "ax"
1944:	movl	$0, (%esp)
195	jmp	1b
1965:	movl	$0, (%esp)
197	jmp	2b
1986:	movl	$0, (%esp)
199	jmp	3b
200.popsection
201	_ASM_EXTABLE(1b, 4b)
202	_ASM_EXTABLE(2b, 5b)
203	_ASM_EXTABLE(3b, 6b)
204	POP_GS_EX
205.endm
206
207ENTRY(ret_from_fork)
208	pushl	%eax
209	call	schedule_tail
210	GET_THREAD_INFO(%ebp)
211	popl	%eax
212	pushl	$0x0202				# Reset kernel eflags
213	popfl
214
215	/* When we fork, we trace the syscall return in the child, too. */
216	movl    %esp, %eax
217	call    syscall_return_slowpath
218	jmp     restore_all
219END(ret_from_fork)
220
221ENTRY(ret_from_kernel_thread)
222	pushl	%eax
223	call	schedule_tail
224	GET_THREAD_INFO(%ebp)
225	popl	%eax
226	pushl	$0x0202				# Reset kernel eflags
227	popfl
228	movl	PT_EBP(%esp), %eax
229	call	*PT_EBX(%esp)
230	movl	$0, PT_EAX(%esp)
231
232	/*
233	 * Kernel threads return to userspace as if returning from a syscall.
234	 * We should check whether anything actually uses this path and, if so,
235	 * consider switching it over to ret_from_fork.
236	 */
237	movl    %esp, %eax
238	call    syscall_return_slowpath
239	jmp     restore_all
240ENDPROC(ret_from_kernel_thread)
241
242/*
243 * Return to user mode is not as complex as all this looks,
244 * but we want the default path for a system call return to
245 * go as quickly as possible which is why some of this is
246 * less clear than it otherwise should be.
247 */
248
249	# userspace resumption stub bypassing syscall exit tracing
250	ALIGN
251ret_from_exception:
252	preempt_stop(CLBR_ANY)
253ret_from_intr:
254	GET_THREAD_INFO(%ebp)
255#ifdef CONFIG_VM86
256	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
257	movb	PT_CS(%esp), %al
258	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
259#else
260	/*
261	 * We can be coming here from child spawned by kernel_thread().
262	 */
263	movl	PT_CS(%esp), %eax
264	andl	$SEGMENT_RPL_MASK, %eax
265#endif
266	cmpl	$USER_RPL, %eax
267	jb	resume_kernel			# not returning to v8086 or userspace
268
269ENTRY(resume_userspace)
270	DISABLE_INTERRUPTS(CLBR_ANY)
271	TRACE_IRQS_OFF
272	movl	%esp, %eax
273	call	prepare_exit_to_usermode
274	jmp	restore_all
275END(ret_from_exception)
276
277#ifdef CONFIG_PREEMPT
278ENTRY(resume_kernel)
279	DISABLE_INTERRUPTS(CLBR_ANY)
280need_resched:
281	cmpl	$0, PER_CPU_VAR(__preempt_count)
282	jnz	restore_all
283	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
284	jz	restore_all
285	call	preempt_schedule_irq
286	jmp	need_resched
287END(resume_kernel)
288#endif
289
290	# SYSENTER  call handler stub
291ENTRY(entry_SYSENTER_32)
292	movl	TSS_sysenter_sp0(%esp), %esp
293sysenter_past_esp:
294	pushl	$__USER_DS		/* pt_regs->ss */
295	pushl	%ecx			/* pt_regs->cx */
296	pushfl				/* pt_regs->flags (except IF = 0) */
297	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
298	pushl	$__USER_CS		/* pt_regs->cs */
299	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
300	pushl	%eax			/* pt_regs->orig_ax */
301	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
302
303	/*
304	 * User mode is traced as though IRQs are on, and SYSENTER
305	 * turned them off.
306	 */
307	TRACE_IRQS_OFF
308
309	movl	%esp, %eax
310	call	do_fast_syscall_32
311	testl	%eax, %eax
312	jz	.Lsyscall_32_done
313
314/* Opportunistic SYSEXIT */
315	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
316	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
317	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
3181:	mov	PT_FS(%esp), %fs
319	PTGS_TO_GS
320	popl	%ebx			/* pt_regs->bx */
321	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
322	popl	%esi			/* pt_regs->si */
323	popl	%edi			/* pt_regs->di */
324	popl	%ebp			/* pt_regs->bp */
325	popl	%eax			/* pt_regs->ax */
326
327	/*
328	 * Return back to the vDSO, which will pop ecx and edx.
329	 * Don't bother with DS and ES (they already contain __USER_DS).
330	 */
331	ENABLE_INTERRUPTS_SYSEXIT
332
333.pushsection .fixup, "ax"
3342:	movl	$0, PT_FS(%esp)
335	jmp	1b
336.popsection
337	_ASM_EXTABLE(1b, 2b)
338	PTGS_TO_GS_EX
339ENDPROC(entry_SYSENTER_32)
340
341	# system call handler stub
342ENTRY(entry_INT80_32)
343	ASM_CLAC
344	pushl	%eax			/* pt_regs->orig_ax */
345	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
346
347	/*
348	 * User mode is traced as though IRQs are on.  Unlike the 64-bit
349	 * case, INT80 is a trap gate on 32-bit kernels, so interrupts
350	 * are already on (unless user code is messing around with iopl).
351	 */
352
353	movl	%esp, %eax
354	call	do_syscall_32_irqs_on
355.Lsyscall_32_done:
356
357restore_all:
358	TRACE_IRQS_IRET
359restore_all_notrace:
360#ifdef CONFIG_X86_ESPFIX32
361	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
362	/*
363	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
364	 * are returning to the kernel.
365	 * See comments in process.c:copy_thread() for details.
366	 */
367	movb	PT_OLDSS(%esp), %ah
368	movb	PT_CS(%esp), %al
369	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
370	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
371	je ldt_ss				# returning to user-space with LDT SS
372#endif
373restore_nocheck:
374	RESTORE_REGS 4				# skip orig_eax/error_code
375irq_return:
376	INTERRUPT_RETURN
377.section .fixup, "ax"
378ENTRY(iret_exc	)
379	pushl	$0				# no error code
380	pushl	$do_iret_error
381	jmp	error_code
382.previous
383	_ASM_EXTABLE(irq_return, iret_exc)
384
385#ifdef CONFIG_X86_ESPFIX32
386ldt_ss:
387#ifdef CONFIG_PARAVIRT
388	/*
389	 * The kernel can't run on a non-flat stack if paravirt mode
390	 * is active.  Rather than try to fixup the high bits of
391	 * ESP, bypass this code entirely.  This may break DOSemu
392	 * and/or Wine support in a paravirt VM, although the option
393	 * is still available to implement the setting of the high
394	 * 16-bits in the INTERRUPT_RETURN paravirt-op.
395	 */
396	cmpl	$0, pv_info+PARAVIRT_enabled
397	jne	restore_nocheck
398#endif
399
400/*
401 * Setup and switch to ESPFIX stack
402 *
403 * We're returning to userspace with a 16 bit stack. The CPU will not
404 * restore the high word of ESP for us on executing iret... This is an
405 * "official" bug of all the x86-compatible CPUs, which we can work
406 * around to make dosemu and wine happy. We do this by preloading the
407 * high word of ESP with the high word of the userspace ESP while
408 * compensating for the offset by changing to the ESPFIX segment with
409 * a base address that matches for the difference.
410 */
411#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
412	mov	%esp, %edx			/* load kernel esp */
413	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
414	mov	%dx, %ax			/* eax: new kernel esp */
415	sub	%eax, %edx			/* offset (low word is 0) */
416	shr	$16, %edx
417	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
418	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
419	pushl	$__ESPFIX_SS
420	pushl	%eax				/* new kernel esp */
421	/*
422	 * Disable interrupts, but do not irqtrace this section: we
423	 * will soon execute iret and the tracer was already set to
424	 * the irqstate after the IRET:
425	 */
426	DISABLE_INTERRUPTS(CLBR_EAX)
427	lss	(%esp), %esp			/* switch to espfix segment */
428	jmp	restore_nocheck
429#endif
430ENDPROC(entry_INT80_32)
431
432.macro FIXUP_ESPFIX_STACK
433/*
434 * Switch back for ESPFIX stack to the normal zerobased stack
435 *
436 * We can't call C functions using the ESPFIX stack. This code reads
437 * the high word of the segment base from the GDT and swiches to the
438 * normal stack and adjusts ESP with the matching offset.
439 */
440#ifdef CONFIG_X86_ESPFIX32
441	/* fixup the stack */
442	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
443	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
444	shl	$16, %eax
445	addl	%esp, %eax			/* the adjusted stack pointer */
446	pushl	$__KERNEL_DS
447	pushl	%eax
448	lss	(%esp), %esp			/* switch to the normal stack segment */
449#endif
450.endm
451.macro UNWIND_ESPFIX_STACK
452#ifdef CONFIG_X86_ESPFIX32
453	movl	%ss, %eax
454	/* see if on espfix stack */
455	cmpw	$__ESPFIX_SS, %ax
456	jne	27f
457	movl	$__KERNEL_DS, %eax
458	movl	%eax, %ds
459	movl	%eax, %es
460	/* switch to normal stack */
461	FIXUP_ESPFIX_STACK
46227:
463#endif
464.endm
465
466/*
467 * Build the entry stubs with some assembler magic.
468 * We pack 1 stub into every 8-byte block.
469 */
470	.align 8
471ENTRY(irq_entries_start)
472    vector=FIRST_EXTERNAL_VECTOR
473    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
474	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
475    vector=vector+1
476	jmp	common_interrupt
477	.align	8
478    .endr
479END(irq_entries_start)
480
481/*
482 * the CPU automatically disables interrupts when executing an IRQ vector,
483 * so IRQ-flags tracing has to follow that:
484 */
485	.p2align CONFIG_X86_L1_CACHE_SHIFT
486common_interrupt:
487	ASM_CLAC
488	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
489	SAVE_ALL
490	TRACE_IRQS_OFF
491	movl	%esp, %eax
492	call	do_IRQ
493	jmp	ret_from_intr
494ENDPROC(common_interrupt)
495
496#define BUILD_INTERRUPT3(name, nr, fn)	\
497ENTRY(name)				\
498	ASM_CLAC;			\
499	pushl	$~(nr);			\
500	SAVE_ALL;			\
501	TRACE_IRQS_OFF			\
502	movl	%esp, %eax;		\
503	call	fn;			\
504	jmp	ret_from_intr;		\
505ENDPROC(name)
506
507
508#ifdef CONFIG_TRACING
509# define TRACE_BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
510#else
511# define TRACE_BUILD_INTERRUPT(name, nr)
512#endif
513
514#define BUILD_INTERRUPT(name, nr)		\
515	BUILD_INTERRUPT3(name, nr, smp_##name);	\
516	TRACE_BUILD_INTERRUPT(name, nr)
517
518/* The include is where all of the SMP etc. interrupts come from */
519#include <asm/entry_arch.h>
520
521ENTRY(coprocessor_error)
522	ASM_CLAC
523	pushl	$0
524	pushl	$do_coprocessor_error
525	jmp	error_code
526END(coprocessor_error)
527
528ENTRY(simd_coprocessor_error)
529	ASM_CLAC
530	pushl	$0
531#ifdef CONFIG_X86_INVD_BUG
532	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
533	ALTERNATIVE "pushl	$do_general_protection",	\
534		    "pushl	$do_simd_coprocessor_error",	\
535		    X86_FEATURE_XMM
536#else
537	pushl	$do_simd_coprocessor_error
538#endif
539	jmp	error_code
540END(simd_coprocessor_error)
541
542ENTRY(device_not_available)
543	ASM_CLAC
544	pushl	$-1				# mark this as an int
545	pushl	$do_device_not_available
546	jmp	error_code
547END(device_not_available)
548
549#ifdef CONFIG_PARAVIRT
550ENTRY(native_iret)
551	iret
552	_ASM_EXTABLE(native_iret, iret_exc)
553END(native_iret)
554
555ENTRY(native_irq_enable_sysexit)
556	sti
557	sysexit
558END(native_irq_enable_sysexit)
559#endif
560
561ENTRY(overflow)
562	ASM_CLAC
563	pushl	$0
564	pushl	$do_overflow
565	jmp	error_code
566END(overflow)
567
568ENTRY(bounds)
569	ASM_CLAC
570	pushl	$0
571	pushl	$do_bounds
572	jmp	error_code
573END(bounds)
574
575ENTRY(invalid_op)
576	ASM_CLAC
577	pushl	$0
578	pushl	$do_invalid_op
579	jmp	error_code
580END(invalid_op)
581
582ENTRY(coprocessor_segment_overrun)
583	ASM_CLAC
584	pushl	$0
585	pushl	$do_coprocessor_segment_overrun
586	jmp	error_code
587END(coprocessor_segment_overrun)
588
589ENTRY(invalid_TSS)
590	ASM_CLAC
591	pushl	$do_invalid_TSS
592	jmp	error_code
593END(invalid_TSS)
594
595ENTRY(segment_not_present)
596	ASM_CLAC
597	pushl	$do_segment_not_present
598	jmp	error_code
599END(segment_not_present)
600
601ENTRY(stack_segment)
602	ASM_CLAC
603	pushl	$do_stack_segment
604	jmp	error_code
605END(stack_segment)
606
607ENTRY(alignment_check)
608	ASM_CLAC
609	pushl	$do_alignment_check
610	jmp	error_code
611END(alignment_check)
612
613ENTRY(divide_error)
614	ASM_CLAC
615	pushl	$0				# no error code
616	pushl	$do_divide_error
617	jmp	error_code
618END(divide_error)
619
620#ifdef CONFIG_X86_MCE
621ENTRY(machine_check)
622	ASM_CLAC
623	pushl	$0
624	pushl	machine_check_vector
625	jmp	error_code
626END(machine_check)
627#endif
628
629ENTRY(spurious_interrupt_bug)
630	ASM_CLAC
631	pushl	$0
632	pushl	$do_spurious_interrupt_bug
633	jmp	error_code
634END(spurious_interrupt_bug)
635
636#ifdef CONFIG_XEN
637/*
638 * Xen doesn't set %esp to be precisely what the normal SYSENTER
639 * entry point expects, so fix it up before using the normal path.
640 */
641ENTRY(xen_sysenter_target)
642	addl	$5*4, %esp			/* remove xen-provided frame */
643	jmp	sysenter_past_esp
644
645ENTRY(xen_hypervisor_callback)
646	pushl	$-1				/* orig_ax = -1 => not a system call */
647	SAVE_ALL
648	TRACE_IRQS_OFF
649
650	/*
651	 * Check to see if we got the event in the critical
652	 * region in xen_iret_direct, after we've reenabled
653	 * events and checked for pending events.  This simulates
654	 * iret instruction's behaviour where it delivers a
655	 * pending interrupt when enabling interrupts:
656	 */
657	movl	PT_EIP(%esp), %eax
658	cmpl	$xen_iret_start_crit, %eax
659	jb	1f
660	cmpl	$xen_iret_end_crit, %eax
661	jae	1f
662
663	jmp	xen_iret_crit_fixup
664
665ENTRY(xen_do_upcall)
6661:	mov	%esp, %eax
667	call	xen_evtchn_do_upcall
668#ifndef CONFIG_PREEMPT
669	call	xen_maybe_preempt_hcall
670#endif
671	jmp	ret_from_intr
672ENDPROC(xen_hypervisor_callback)
673
674/*
675 * Hypervisor uses this for application faults while it executes.
676 * We get here for two reasons:
677 *  1. Fault while reloading DS, ES, FS or GS
678 *  2. Fault while executing IRET
679 * Category 1 we fix up by reattempting the load, and zeroing the segment
680 * register if the load fails.
681 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
682 * normal Linux return path in this case because if we use the IRET hypercall
683 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
684 * We distinguish between categories by maintaining a status value in EAX.
685 */
686ENTRY(xen_failsafe_callback)
687	pushl	%eax
688	movl	$1, %eax
6891:	mov	4(%esp), %ds
6902:	mov	8(%esp), %es
6913:	mov	12(%esp), %fs
6924:	mov	16(%esp), %gs
693	/* EAX == 0 => Category 1 (Bad segment)
694	   EAX != 0 => Category 2 (Bad IRET) */
695	testl	%eax, %eax
696	popl	%eax
697	lea	16(%esp), %esp
698	jz	5f
699	jmp	iret_exc
7005:	pushl	$-1				/* orig_ax = -1 => not a system call */
701	SAVE_ALL
702	jmp	ret_from_exception
703
704.section .fixup, "ax"
7056:	xorl	%eax, %eax
706	movl	%eax, 4(%esp)
707	jmp	1b
7087:	xorl	%eax, %eax
709	movl	%eax, 8(%esp)
710	jmp	2b
7118:	xorl	%eax, %eax
712	movl	%eax, 12(%esp)
713	jmp	3b
7149:	xorl	%eax, %eax
715	movl	%eax, 16(%esp)
716	jmp	4b
717.previous
718	_ASM_EXTABLE(1b, 6b)
719	_ASM_EXTABLE(2b, 7b)
720	_ASM_EXTABLE(3b, 8b)
721	_ASM_EXTABLE(4b, 9b)
722ENDPROC(xen_failsafe_callback)
723
724BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
725		xen_evtchn_do_upcall)
726
727#endif /* CONFIG_XEN */
728
729#if IS_ENABLED(CONFIG_HYPERV)
730
731BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
732	hyperv_vector_handler)
733
734#endif /* CONFIG_HYPERV */
735
736#ifdef CONFIG_FUNCTION_TRACER
737#ifdef CONFIG_DYNAMIC_FTRACE
738
739ENTRY(mcount)
740	ret
741END(mcount)
742
743ENTRY(ftrace_caller)
744	pushl	%eax
745	pushl	%ecx
746	pushl	%edx
747	pushl	$0				/* Pass NULL as regs pointer */
748	movl	4*4(%esp), %eax
749	movl	0x4(%ebp), %edx
750	movl	function_trace_op, %ecx
751	subl	$MCOUNT_INSN_SIZE, %eax
752
753.globl ftrace_call
754ftrace_call:
755	call	ftrace_stub
756
757	addl	$4, %esp			/* skip NULL pointer */
758	popl	%edx
759	popl	%ecx
760	popl	%eax
761ftrace_ret:
762#ifdef CONFIG_FUNCTION_GRAPH_TRACER
763.globl ftrace_graph_call
764ftrace_graph_call:
765	jmp	ftrace_stub
766#endif
767
768.globl ftrace_stub
769ftrace_stub:
770	ret
771END(ftrace_caller)
772
773ENTRY(ftrace_regs_caller)
774	pushf	/* push flags before compare (in cs location) */
775
776	/*
777	 * i386 does not save SS and ESP when coming from kernel.
778	 * Instead, to get sp, &regs->sp is used (see ptrace.h).
779	 * Unfortunately, that means eflags must be at the same location
780	 * as the current return ip is. We move the return ip into the
781	 * ip location, and move flags into the return ip location.
782	 */
783	pushl	4(%esp)				/* save return ip into ip slot */
784
785	pushl	$0				/* Load 0 into orig_ax */
786	pushl	%gs
787	pushl	%fs
788	pushl	%es
789	pushl	%ds
790	pushl	%eax
791	pushl	%ebp
792	pushl	%edi
793	pushl	%esi
794	pushl	%edx
795	pushl	%ecx
796	pushl	%ebx
797
798	movl	13*4(%esp), %eax		/* Get the saved flags */
799	movl	%eax, 14*4(%esp)		/* Move saved flags into regs->flags location */
800						/* clobbering return ip */
801	movl	$__KERNEL_CS, 13*4(%esp)
802
803	movl	12*4(%esp), %eax		/* Load ip (1st parameter) */
804	subl	$MCOUNT_INSN_SIZE, %eax		/* Adjust ip */
805	movl	0x4(%ebp), %edx			/* Load parent ip (2nd parameter) */
806	movl	function_trace_op, %ecx		/* Save ftrace_pos in 3rd parameter */
807	pushl	%esp				/* Save pt_regs as 4th parameter */
808
809GLOBAL(ftrace_regs_call)
810	call	ftrace_stub
811
812	addl	$4, %esp			/* Skip pt_regs */
813	movl	14*4(%esp), %eax		/* Move flags back into cs */
814	movl	%eax, 13*4(%esp)		/* Needed to keep addl	from modifying flags */
815	movl	12*4(%esp), %eax		/* Get return ip from regs->ip */
816	movl	%eax, 14*4(%esp)		/* Put return ip back for ret */
817
818	popl	%ebx
819	popl	%ecx
820	popl	%edx
821	popl	%esi
822	popl	%edi
823	popl	%ebp
824	popl	%eax
825	popl	%ds
826	popl	%es
827	popl	%fs
828	popl	%gs
829	addl	$8, %esp			/* Skip orig_ax and ip */
830	popf					/* Pop flags at end (no addl to corrupt flags) */
831	jmp	ftrace_ret
832
833	popf
834	jmp	ftrace_stub
835#else /* ! CONFIG_DYNAMIC_FTRACE */
836
837ENTRY(mcount)
838	cmpl	$__PAGE_OFFSET, %esp
839	jb	ftrace_stub			/* Paging not enabled yet? */
840
841	cmpl	$ftrace_stub, ftrace_trace_function
842	jnz	trace
843#ifdef CONFIG_FUNCTION_GRAPH_TRACER
844	cmpl	$ftrace_stub, ftrace_graph_return
845	jnz	ftrace_graph_caller
846
847	cmpl	$ftrace_graph_entry_stub, ftrace_graph_entry
848	jnz	ftrace_graph_caller
849#endif
850.globl ftrace_stub
851ftrace_stub:
852	ret
853
854	/* taken from glibc */
855trace:
856	pushl	%eax
857	pushl	%ecx
858	pushl	%edx
859	movl	0xc(%esp), %eax
860	movl	0x4(%ebp), %edx
861	subl	$MCOUNT_INSN_SIZE, %eax
862
863	call	*ftrace_trace_function
864
865	popl	%edx
866	popl	%ecx
867	popl	%eax
868	jmp	ftrace_stub
869END(mcount)
870#endif /* CONFIG_DYNAMIC_FTRACE */
871#endif /* CONFIG_FUNCTION_TRACER */
872
873#ifdef CONFIG_FUNCTION_GRAPH_TRACER
874ENTRY(ftrace_graph_caller)
875	pushl	%eax
876	pushl	%ecx
877	pushl	%edx
878	movl	0xc(%esp), %eax
879	lea	0x4(%ebp), %edx
880	movl	(%ebp), %ecx
881	subl	$MCOUNT_INSN_SIZE, %eax
882	call	prepare_ftrace_return
883	popl	%edx
884	popl	%ecx
885	popl	%eax
886	ret
887END(ftrace_graph_caller)
888
889.globl return_to_handler
890return_to_handler:
891	pushl	%eax
892	pushl	%edx
893	movl	%ebp, %eax
894	call	ftrace_return_to_handler
895	movl	%eax, %ecx
896	popl	%edx
897	popl	%eax
898	jmp	*%ecx
899#endif
900
901#ifdef CONFIG_TRACING
902ENTRY(trace_page_fault)
903	ASM_CLAC
904	pushl	$trace_do_page_fault
905	jmp	error_code
906END(trace_page_fault)
907#endif
908
909ENTRY(page_fault)
910	ASM_CLAC
911	pushl	$do_page_fault
912	ALIGN
913error_code:
914	/* the function address is in %gs's slot on the stack */
915	pushl	%fs
916	pushl	%es
917	pushl	%ds
918	pushl	%eax
919	pushl	%ebp
920	pushl	%edi
921	pushl	%esi
922	pushl	%edx
923	pushl	%ecx
924	pushl	%ebx
925	cld
926	movl	$(__KERNEL_PERCPU), %ecx
927	movl	%ecx, %fs
928	UNWIND_ESPFIX_STACK
929	GS_TO_REG %ecx
930	movl	PT_GS(%esp), %edi		# get the function address
931	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
932	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
933	REG_TO_PTGS %ecx
934	SET_KERNEL_GS %ecx
935	movl	$(__USER_DS), %ecx
936	movl	%ecx, %ds
937	movl	%ecx, %es
938	TRACE_IRQS_OFF
939	movl	%esp, %eax			# pt_regs pointer
940	call	*%edi
941	jmp	ret_from_exception
942END(page_fault)
943
944/*
945 * Debug traps and NMI can happen at the one SYSENTER instruction
946 * that sets up the real kernel stack. Check here, since we can't
947 * allow the wrong stack to be used.
948 *
949 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
950 * already pushed 3 words if it hits on the sysenter instruction:
951 * eflags, cs and eip.
952 *
953 * We just load the right stack, and push the three (known) values
954 * by hand onto the new stack - while updating the return eip past
955 * the instruction that would have done it for sysenter.
956 */
957.macro FIX_STACK offset ok label
958	cmpw	$__KERNEL_CS, 4(%esp)
959	jne	\ok
960\label:
961	movl	TSS_sysenter_sp0 + \offset(%esp), %esp
962	pushfl
963	pushl	$__KERNEL_CS
964	pushl	$sysenter_past_esp
965.endm
966
967ENTRY(debug)
968	ASM_CLAC
969	cmpl	$entry_SYSENTER_32, (%esp)
970	jne	debug_stack_correct
971	FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
972debug_stack_correct:
973	pushl	$-1				# mark this as an int
974	SAVE_ALL
975	TRACE_IRQS_OFF
976	xorl	%edx, %edx			# error code 0
977	movl	%esp, %eax			# pt_regs pointer
978	call	do_debug
979	jmp	ret_from_exception
980END(debug)
981
982/*
983 * NMI is doubly nasty. It can happen _while_ we're handling
984 * a debug fault, and the debug fault hasn't yet been able to
985 * clear up the stack. So we first check whether we got  an
986 * NMI on the sysenter entry path, but after that we need to
987 * check whether we got an NMI on the debug path where the debug
988 * fault happened on the sysenter path.
989 */
990ENTRY(nmi)
991	ASM_CLAC
992#ifdef CONFIG_X86_ESPFIX32
993	pushl	%eax
994	movl	%ss, %eax
995	cmpw	$__ESPFIX_SS, %ax
996	popl	%eax
997	je	nmi_espfix_stack
998#endif
999	cmpl	$entry_SYSENTER_32, (%esp)
1000	je	nmi_stack_fixup
1001	pushl	%eax
1002	movl	%esp, %eax
1003	/*
1004	 * Do not access memory above the end of our stack page,
1005	 * it might not exist.
1006	 */
1007	andl	$(THREAD_SIZE-1), %eax
1008	cmpl	$(THREAD_SIZE-20), %eax
1009	popl	%eax
1010	jae	nmi_stack_correct
1011	cmpl	$entry_SYSENTER_32, 12(%esp)
1012	je	nmi_debug_stack_check
1013nmi_stack_correct:
1014	pushl	%eax
1015	SAVE_ALL
1016	xorl	%edx, %edx			# zero error code
1017	movl	%esp, %eax			# pt_regs pointer
1018	call	do_nmi
1019	jmp	restore_all_notrace
1020
1021nmi_stack_fixup:
1022	FIX_STACK 12, nmi_stack_correct, 1
1023	jmp	nmi_stack_correct
1024
1025nmi_debug_stack_check:
1026	cmpw	$__KERNEL_CS, 16(%esp)
1027	jne	nmi_stack_correct
1028	cmpl	$debug, (%esp)
1029	jb	nmi_stack_correct
1030	cmpl	$debug_esp_fix_insn, (%esp)
1031	ja	nmi_stack_correct
1032	FIX_STACK 24, nmi_stack_correct, 1
1033	jmp	nmi_stack_correct
1034
1035#ifdef CONFIG_X86_ESPFIX32
1036nmi_espfix_stack:
1037	/*
1038	 * create the pointer to lss back
1039	 */
1040	pushl	%ss
1041	pushl	%esp
1042	addl	$4, (%esp)
1043	/* copy the iret frame of 12 bytes */
1044	.rept 3
1045	pushl	16(%esp)
1046	.endr
1047	pushl	%eax
1048	SAVE_ALL
1049	FIXUP_ESPFIX_STACK			# %eax == %esp
1050	xorl	%edx, %edx			# zero error code
1051	call	do_nmi
1052	RESTORE_REGS
1053	lss	12+4(%esp), %esp		# back to espfix stack
1054	jmp	irq_return
1055#endif
1056END(nmi)
1057
1058ENTRY(int3)
1059	ASM_CLAC
1060	pushl	$-1				# mark this as an int
1061	SAVE_ALL
1062	TRACE_IRQS_OFF
1063	xorl	%edx, %edx			# zero error code
1064	movl	%esp, %eax			# pt_regs pointer
1065	call	do_int3
1066	jmp	ret_from_exception
1067END(int3)
1068
1069ENTRY(general_protection)
1070	pushl	$do_general_protection
1071	jmp	error_code
1072END(general_protection)
1073
1074#ifdef CONFIG_KVM_GUEST
1075ENTRY(async_page_fault)
1076	ASM_CLAC
1077	pushl	$do_async_page_fault
1078	jmp	error_code
1079END(async_page_fault)
1080#endif
1081