• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/cpufeatures.h>
44#include <asm/alternative-asm.h>
45#include <asm/asm.h>
46#include <asm/smap.h>
47#include <asm/nospec-branch.h>
48
49	.section .entry.text, "ax"
50
51/*
52 * We use macros for low-level operations which need to be overridden
53 * for paravirtualization.  The following will never clobber any registers:
54 *   INTERRUPT_RETURN (aka. "iret")
55 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
56 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
57 *
58 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
59 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
60 * Allowing a register to be clobbered can shrink the paravirt replacement
61 * enough to patch inline, increasing performance.
62 */
63
64#ifdef CONFIG_PREEMPT
65# define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
66#else
67# define preempt_stop(clobbers)
68# define resume_kernel		restore_all
69#endif
70
71.macro TRACE_IRQS_IRET
72#ifdef CONFIG_TRACE_IRQFLAGS
73	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
74	jz	1f
75	TRACE_IRQS_ON
761:
77#endif
78.endm
79
80/*
81 * User gs save/restore
82 *
83 * %gs is used for userland TLS and kernel only uses it for stack
84 * canary which is required to be at %gs:20 by gcc.  Read the comment
85 * at the top of stackprotector.h for more info.
86 *
87 * Local labels 98 and 99 are used.
88 */
89#ifdef CONFIG_X86_32_LAZY_GS
90
91 /* unfortunately push/pop can't be no-op */
92.macro PUSH_GS
93	pushl	$0
94.endm
95.macro POP_GS pop=0
96	addl	$(4 + \pop), %esp
97.endm
98.macro POP_GS_EX
99.endm
100
101 /* all the rest are no-op */
102.macro PTGS_TO_GS
103.endm
104.macro PTGS_TO_GS_EX
105.endm
106.macro GS_TO_REG reg
107.endm
108.macro REG_TO_PTGS reg
109.endm
110.macro SET_KERNEL_GS reg
111.endm
112
113#else	/* CONFIG_X86_32_LAZY_GS */
114
115.macro PUSH_GS
116	pushl	%gs
117.endm
118
119.macro POP_GS pop=0
12098:	popl	%gs
121  .if \pop <> 0
122	add	$\pop, %esp
123  .endif
124.endm
125.macro POP_GS_EX
126.pushsection .fixup, "ax"
12799:	movl	$0, (%esp)
128	jmp	98b
129.popsection
130	_ASM_EXTABLE(98b, 99b)
131.endm
132
133.macro PTGS_TO_GS
13498:	mov	PT_GS(%esp), %gs
135.endm
136.macro PTGS_TO_GS_EX
137.pushsection .fixup, "ax"
13899:	movl	$0, PT_GS(%esp)
139	jmp	98b
140.popsection
141	_ASM_EXTABLE(98b, 99b)
142.endm
143
144.macro GS_TO_REG reg
145	movl	%gs, \reg
146.endm
147.macro REG_TO_PTGS reg
148	movl	\reg, PT_GS(%esp)
149.endm
150.macro SET_KERNEL_GS reg
151	movl	$(__KERNEL_STACK_CANARY), \reg
152	movl	\reg, %gs
153.endm
154
155#endif /* CONFIG_X86_32_LAZY_GS */
156
157.macro SAVE_ALL pt_regs_ax=%eax
158	cld
159	PUSH_GS
160	pushl	%fs
161	pushl	%es
162	pushl	%ds
163	pushl	\pt_regs_ax
164	pushl	%ebp
165	pushl	%edi
166	pushl	%esi
167	pushl	%edx
168	pushl	%ecx
169	pushl	%ebx
170	movl	$(__USER_DS), %edx
171	movl	%edx, %ds
172	movl	%edx, %es
173	movl	$(__KERNEL_PERCPU), %edx
174	movl	%edx, %fs
175	SET_KERNEL_GS %edx
176.endm
177
178.macro RESTORE_INT_REGS
179	popl	%ebx
180	popl	%ecx
181	popl	%edx
182	popl	%esi
183	popl	%edi
184	popl	%ebp
185	popl	%eax
186.endm
187
188.macro RESTORE_REGS pop=0
189	RESTORE_INT_REGS
1901:	popl	%ds
1912:	popl	%es
1923:	popl	%fs
193	POP_GS \pop
194.pushsection .fixup, "ax"
1954:	movl	$0, (%esp)
196	jmp	1b
1975:	movl	$0, (%esp)
198	jmp	2b
1996:	movl	$0, (%esp)
200	jmp	3b
201.popsection
202	_ASM_EXTABLE(1b, 4b)
203	_ASM_EXTABLE(2b, 5b)
204	_ASM_EXTABLE(3b, 6b)
205	POP_GS_EX
206.endm
207
208ENTRY(ret_from_fork)
209	pushl	%eax
210	call	schedule_tail
211	GET_THREAD_INFO(%ebp)
212	popl	%eax
213	pushl	$0x0202				# Reset kernel eflags
214	popfl
215
216	/* When we fork, we trace the syscall return in the child, too. */
217	movl    %esp, %eax
218	call    syscall_return_slowpath
219	jmp     restore_all
220END(ret_from_fork)
221
222ENTRY(ret_from_kernel_thread)
223	pushl	%eax
224	call	schedule_tail
225	GET_THREAD_INFO(%ebp)
226	popl	%eax
227	pushl	$0x0202				# Reset kernel eflags
228	popfl
229	movl	PT_EBP(%esp), %eax
230	movl	PT_EBX(%esp), %edx
231	CALL_NOSPEC %edx
232	movl	$0, PT_EAX(%esp)
233
234	/*
235	 * Kernel threads return to userspace as if returning from a syscall.
236	 * We should check whether anything actually uses this path and, if so,
237	 * consider switching it over to ret_from_fork.
238	 */
239	movl    %esp, %eax
240	call    syscall_return_slowpath
241	jmp     restore_all
242ENDPROC(ret_from_kernel_thread)
243
244/*
245 * Return to user mode is not as complex as all this looks,
246 * but we want the default path for a system call return to
247 * go as quickly as possible which is why some of this is
248 * less clear than it otherwise should be.
249 */
250
251	# userspace resumption stub bypassing syscall exit tracing
252	ALIGN
253ret_from_exception:
254	preempt_stop(CLBR_ANY)
255ret_from_intr:
256	GET_THREAD_INFO(%ebp)
257#ifdef CONFIG_VM86
258	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
259	movb	PT_CS(%esp), %al
260	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
261#else
262	/*
263	 * We can be coming here from child spawned by kernel_thread().
264	 */
265	movl	PT_CS(%esp), %eax
266	andl	$SEGMENT_RPL_MASK, %eax
267#endif
268	cmpl	$USER_RPL, %eax
269	jb	resume_kernel			# not returning to v8086 or userspace
270
271ENTRY(resume_userspace)
272	DISABLE_INTERRUPTS(CLBR_ANY)
273	TRACE_IRQS_OFF
274	movl	%esp, %eax
275	call	prepare_exit_to_usermode
276	jmp	restore_all
277END(ret_from_exception)
278
279#ifdef CONFIG_PREEMPT
280ENTRY(resume_kernel)
281	DISABLE_INTERRUPTS(CLBR_ANY)
282need_resched:
283	cmpl	$0, PER_CPU_VAR(__preempt_count)
284	jnz	restore_all
285	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
286	jz	restore_all
287	call	preempt_schedule_irq
288	jmp	need_resched
289END(resume_kernel)
290#endif
291
292	# SYSENTER  call handler stub
293ENTRY(entry_SYSENTER_32)
294	movl	TSS_sysenter_sp0(%esp), %esp
295sysenter_past_esp:
296	pushl	$__USER_DS		/* pt_regs->ss */
297	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
298	pushfl				/* pt_regs->flags (except IF = 0) */
299	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
300	pushl	$__USER_CS		/* pt_regs->cs */
301	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
302	pushl	%eax			/* pt_regs->orig_ax */
303	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
304
305	/*
306	 * User mode is traced as though IRQs are on, and SYSENTER
307	 * turned them off.
308	 */
309	TRACE_IRQS_OFF
310
311	movl	%esp, %eax
312	call	do_fast_syscall_32
313	/* XEN PV guests always use IRET path */
314	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
315		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
316
317/* Opportunistic SYSEXIT */
318	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
319	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
320	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
3211:	mov	PT_FS(%esp), %fs
322	PTGS_TO_GS
323	popl	%ebx			/* pt_regs->bx */
324	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
325	popl	%esi			/* pt_regs->si */
326	popl	%edi			/* pt_regs->di */
327	popl	%ebp			/* pt_regs->bp */
328	popl	%eax			/* pt_regs->ax */
329
330	/*
331	 * Return back to the vDSO, which will pop ecx and edx.
332	 * Don't bother with DS and ES (they already contain __USER_DS).
333	 */
334	sti
335	sysexit
336
337.pushsection .fixup, "ax"
3382:	movl	$0, PT_FS(%esp)
339	jmp	1b
340.popsection
341	_ASM_EXTABLE(1b, 2b)
342	PTGS_TO_GS_EX
343ENDPROC(entry_SYSENTER_32)
344
345	# system call handler stub
346ENTRY(entry_INT80_32)
347	ASM_CLAC
348	pushl	%eax			/* pt_regs->orig_ax */
349	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
350
351	/*
352	 * User mode is traced as though IRQs are on.  Unlike the 64-bit
353	 * case, INT80 is a trap gate on 32-bit kernels, so interrupts
354	 * are already on (unless user code is messing around with iopl).
355	 */
356
357	movl	%esp, %eax
358	call	do_syscall_32_irqs_on
359.Lsyscall_32_done:
360
361restore_all:
362	TRACE_IRQS_IRET
363restore_all_notrace:
364#ifdef CONFIG_X86_ESPFIX32
365	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
366	/*
367	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
368	 * are returning to the kernel.
369	 * See comments in process.c:copy_thread() for details.
370	 */
371	movb	PT_OLDSS(%esp), %ah
372	movb	PT_CS(%esp), %al
373	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
374	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
375	je ldt_ss				# returning to user-space with LDT SS
376#endif
377restore_nocheck:
378	RESTORE_REGS 4				# skip orig_eax/error_code
379irq_return:
380	INTERRUPT_RETURN
381.section .fixup, "ax"
382ENTRY(iret_exc	)
383	pushl	$0				# no error code
384	pushl	$do_iret_error
385	jmp	error_code
386.previous
387	_ASM_EXTABLE(irq_return, iret_exc)
388
389#ifdef CONFIG_X86_ESPFIX32
390ldt_ss:
391#ifdef CONFIG_PARAVIRT
392	/*
393	 * The kernel can't run on a non-flat stack if paravirt mode
394	 * is active.  Rather than try to fixup the high bits of
395	 * ESP, bypass this code entirely.  This may break DOSemu
396	 * and/or Wine support in a paravirt VM, although the option
397	 * is still available to implement the setting of the high
398	 * 16-bits in the INTERRUPT_RETURN paravirt-op.
399	 */
400	cmpl	$0, pv_info+PARAVIRT_enabled
401	jne	restore_nocheck
402#endif
403
404/*
405 * Setup and switch to ESPFIX stack
406 *
407 * We're returning to userspace with a 16 bit stack. The CPU will not
408 * restore the high word of ESP for us on executing iret... This is an
409 * "official" bug of all the x86-compatible CPUs, which we can work
410 * around to make dosemu and wine happy. We do this by preloading the
411 * high word of ESP with the high word of the userspace ESP while
412 * compensating for the offset by changing to the ESPFIX segment with
413 * a base address that matches for the difference.
414 */
415#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
416	mov	%esp, %edx			/* load kernel esp */
417	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
418	mov	%dx, %ax			/* eax: new kernel esp */
419	sub	%eax, %edx			/* offset (low word is 0) */
420	shr	$16, %edx
421	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
422	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
423	pushl	$__ESPFIX_SS
424	pushl	%eax				/* new kernel esp */
425	/*
426	 * Disable interrupts, but do not irqtrace this section: we
427	 * will soon execute iret and the tracer was already set to
428	 * the irqstate after the IRET:
429	 */
430	DISABLE_INTERRUPTS(CLBR_EAX)
431	lss	(%esp), %esp			/* switch to espfix segment */
432	jmp	restore_nocheck
433#endif
434ENDPROC(entry_INT80_32)
435
436.macro FIXUP_ESPFIX_STACK
437/*
438 * Switch back for ESPFIX stack to the normal zerobased stack
439 *
440 * We can't call C functions using the ESPFIX stack. This code reads
441 * the high word of the segment base from the GDT and swiches to the
442 * normal stack and adjusts ESP with the matching offset.
443 */
444#ifdef CONFIG_X86_ESPFIX32
445	/* fixup the stack */
446	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
447	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
448	shl	$16, %eax
449	addl	%esp, %eax			/* the adjusted stack pointer */
450	pushl	$__KERNEL_DS
451	pushl	%eax
452	lss	(%esp), %esp			/* switch to the normal stack segment */
453#endif
454.endm
455.macro UNWIND_ESPFIX_STACK
456#ifdef CONFIG_X86_ESPFIX32
457	movl	%ss, %eax
458	/* see if on espfix stack */
459	cmpw	$__ESPFIX_SS, %ax
460	jne	27f
461	movl	$__KERNEL_DS, %eax
462	movl	%eax, %ds
463	movl	%eax, %es
464	/* switch to normal stack */
465	FIXUP_ESPFIX_STACK
46627:
467#endif
468.endm
469
470/*
471 * Build the entry stubs with some assembler magic.
472 * We pack 1 stub into every 8-byte block.
473 */
474	.align 8
475ENTRY(irq_entries_start)
476    vector=FIRST_EXTERNAL_VECTOR
477    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
478	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
479    vector=vector+1
480	jmp	common_interrupt
481	.align	8
482    .endr
483END(irq_entries_start)
484
485/*
486 * the CPU automatically disables interrupts when executing an IRQ vector,
487 * so IRQ-flags tracing has to follow that:
488 */
489	.p2align CONFIG_X86_L1_CACHE_SHIFT
490common_interrupt:
491	ASM_CLAC
492	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
493	SAVE_ALL
494	TRACE_IRQS_OFF
495	movl	%esp, %eax
496	call	do_IRQ
497	jmp	ret_from_intr
498ENDPROC(common_interrupt)
499
500#define BUILD_INTERRUPT3(name, nr, fn)	\
501ENTRY(name)				\
502	ASM_CLAC;			\
503	pushl	$~(nr);			\
504	SAVE_ALL;			\
505	TRACE_IRQS_OFF			\
506	movl	%esp, %eax;		\
507	call	fn;			\
508	jmp	ret_from_intr;		\
509ENDPROC(name)
510
511
512#ifdef CONFIG_TRACING
513# define TRACE_BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
514#else
515# define TRACE_BUILD_INTERRUPT(name, nr)
516#endif
517
518#define BUILD_INTERRUPT(name, nr)		\
519	BUILD_INTERRUPT3(name, nr, smp_##name);	\
520	TRACE_BUILD_INTERRUPT(name, nr)
521
522/* The include is where all of the SMP etc. interrupts come from */
523#include <asm/entry_arch.h>
524
525ENTRY(coprocessor_error)
526	ASM_CLAC
527	pushl	$0
528	pushl	$do_coprocessor_error
529	jmp	error_code
530END(coprocessor_error)
531
532ENTRY(simd_coprocessor_error)
533	ASM_CLAC
534	pushl	$0
535#ifdef CONFIG_X86_INVD_BUG
536	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
537	ALTERNATIVE "pushl	$do_general_protection",	\
538		    "pushl	$do_simd_coprocessor_error",	\
539		    X86_FEATURE_XMM
540#else
541	pushl	$do_simd_coprocessor_error
542#endif
543	jmp	error_code
544END(simd_coprocessor_error)
545
546ENTRY(device_not_available)
547	ASM_CLAC
548	pushl	$-1				# mark this as an int
549	pushl	$do_device_not_available
550	jmp	error_code
551END(device_not_available)
552
553#ifdef CONFIG_PARAVIRT
554ENTRY(native_iret)
555	iret
556	_ASM_EXTABLE(native_iret, iret_exc)
557END(native_iret)
558#endif
559
560ENTRY(overflow)
561	ASM_CLAC
562	pushl	$0
563	pushl	$do_overflow
564	jmp	error_code
565END(overflow)
566
567ENTRY(bounds)
568	ASM_CLAC
569	pushl	$0
570	pushl	$do_bounds
571	jmp	error_code
572END(bounds)
573
574ENTRY(invalid_op)
575	ASM_CLAC
576	pushl	$0
577	pushl	$do_invalid_op
578	jmp	error_code
579END(invalid_op)
580
581ENTRY(coprocessor_segment_overrun)
582	ASM_CLAC
583	pushl	$0
584	pushl	$do_coprocessor_segment_overrun
585	jmp	error_code
586END(coprocessor_segment_overrun)
587
588ENTRY(invalid_TSS)
589	ASM_CLAC
590	pushl	$do_invalid_TSS
591	jmp	error_code
592END(invalid_TSS)
593
594ENTRY(segment_not_present)
595	ASM_CLAC
596	pushl	$do_segment_not_present
597	jmp	error_code
598END(segment_not_present)
599
600ENTRY(stack_segment)
601	ASM_CLAC
602	pushl	$do_stack_segment
603	jmp	error_code
604END(stack_segment)
605
606ENTRY(alignment_check)
607	ASM_CLAC
608	pushl	$do_alignment_check
609	jmp	error_code
610END(alignment_check)
611
612ENTRY(divide_error)
613	ASM_CLAC
614	pushl	$0				# no error code
615	pushl	$do_divide_error
616	jmp	error_code
617END(divide_error)
618
619#ifdef CONFIG_X86_MCE
620ENTRY(machine_check)
621	ASM_CLAC
622	pushl	$0
623	pushl	machine_check_vector
624	jmp	error_code
625END(machine_check)
626#endif
627
628ENTRY(spurious_interrupt_bug)
629	ASM_CLAC
630	pushl	$0
631	pushl	$do_spurious_interrupt_bug
632	jmp	error_code
633END(spurious_interrupt_bug)
634
635#ifdef CONFIG_XEN
636/*
637 * Xen doesn't set %esp to be precisely what the normal SYSENTER
638 * entry point expects, so fix it up before using the normal path.
639 */
640ENTRY(xen_sysenter_target)
641	addl	$5*4, %esp			/* remove xen-provided frame */
642	jmp	sysenter_past_esp
643
644ENTRY(xen_hypervisor_callback)
645	pushl	$-1				/* orig_ax = -1 => not a system call */
646	SAVE_ALL
647	TRACE_IRQS_OFF
648
649	/*
650	 * Check to see if we got the event in the critical
651	 * region in xen_iret_direct, after we've reenabled
652	 * events and checked for pending events.  This simulates
653	 * iret instruction's behaviour where it delivers a
654	 * pending interrupt when enabling interrupts:
655	 */
656	movl	PT_EIP(%esp), %eax
657	cmpl	$xen_iret_start_crit, %eax
658	jb	1f
659	cmpl	$xen_iret_end_crit, %eax
660	jae	1f
661
662	jmp	xen_iret_crit_fixup
663
664ENTRY(xen_do_upcall)
6651:	mov	%esp, %eax
666	call	xen_evtchn_do_upcall
667#ifndef CONFIG_PREEMPT
668	call	xen_maybe_preempt_hcall
669#endif
670	jmp	ret_from_intr
671ENDPROC(xen_hypervisor_callback)
672
673/*
674 * Hypervisor uses this for application faults while it executes.
675 * We get here for two reasons:
676 *  1. Fault while reloading DS, ES, FS or GS
677 *  2. Fault while executing IRET
678 * Category 1 we fix up by reattempting the load, and zeroing the segment
679 * register if the load fails.
680 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
681 * normal Linux return path in this case because if we use the IRET hypercall
682 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
683 * We distinguish between categories by maintaining a status value in EAX.
684 */
685ENTRY(xen_failsafe_callback)
686	pushl	%eax
687	movl	$1, %eax
6881:	mov	4(%esp), %ds
6892:	mov	8(%esp), %es
6903:	mov	12(%esp), %fs
6914:	mov	16(%esp), %gs
692	/* EAX == 0 => Category 1 (Bad segment)
693	   EAX != 0 => Category 2 (Bad IRET) */
694	testl	%eax, %eax
695	popl	%eax
696	lea	16(%esp), %esp
697	jz	5f
698	jmp	iret_exc
6995:	pushl	$-1				/* orig_ax = -1 => not a system call */
700	SAVE_ALL
701	jmp	ret_from_exception
702
703.section .fixup, "ax"
7046:	xorl	%eax, %eax
705	movl	%eax, 4(%esp)
706	jmp	1b
7077:	xorl	%eax, %eax
708	movl	%eax, 8(%esp)
709	jmp	2b
7108:	xorl	%eax, %eax
711	movl	%eax, 12(%esp)
712	jmp	3b
7139:	xorl	%eax, %eax
714	movl	%eax, 16(%esp)
715	jmp	4b
716.previous
717	_ASM_EXTABLE(1b, 6b)
718	_ASM_EXTABLE(2b, 7b)
719	_ASM_EXTABLE(3b, 8b)
720	_ASM_EXTABLE(4b, 9b)
721ENDPROC(xen_failsafe_callback)
722
723BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
724		xen_evtchn_do_upcall)
725
726#endif /* CONFIG_XEN */
727
728#if IS_ENABLED(CONFIG_HYPERV)
729
730BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
731	hyperv_vector_handler)
732
733#endif /* CONFIG_HYPERV */
734
735#ifdef CONFIG_FUNCTION_TRACER
736#ifdef CONFIG_DYNAMIC_FTRACE
737
738ENTRY(mcount)
739	ret
740END(mcount)
741
742ENTRY(ftrace_caller)
743	pushl	%eax
744	pushl	%ecx
745	pushl	%edx
746	pushl	$0				/* Pass NULL as regs pointer */
747	movl	4*4(%esp), %eax
748	movl	0x4(%ebp), %edx
749	movl	function_trace_op, %ecx
750	subl	$MCOUNT_INSN_SIZE, %eax
751
752.globl ftrace_call
753ftrace_call:
754	call	ftrace_stub
755
756	addl	$4, %esp			/* skip NULL pointer */
757	popl	%edx
758	popl	%ecx
759	popl	%eax
760ftrace_ret:
761#ifdef CONFIG_FUNCTION_GRAPH_TRACER
762.globl ftrace_graph_call
763ftrace_graph_call:
764	jmp	ftrace_stub
765#endif
766
767/* This is weak to keep gas from relaxing the jumps */
768WEAK(ftrace_stub)
769	ret
770END(ftrace_caller)
771
772ENTRY(ftrace_regs_caller)
773	pushf	/* push flags before compare (in cs location) */
774
775	/*
776	 * i386 does not save SS and ESP when coming from kernel.
777	 * Instead, to get sp, &regs->sp is used (see ptrace.h).
778	 * Unfortunately, that means eflags must be at the same location
779	 * as the current return ip is. We move the return ip into the
780	 * ip location, and move flags into the return ip location.
781	 */
782	pushl	4(%esp)				/* save return ip into ip slot */
783
784	pushl	$0				/* Load 0 into orig_ax */
785	pushl	%gs
786	pushl	%fs
787	pushl	%es
788	pushl	%ds
789	pushl	%eax
790	pushl	%ebp
791	pushl	%edi
792	pushl	%esi
793	pushl	%edx
794	pushl	%ecx
795	pushl	%ebx
796
797	movl	13*4(%esp), %eax		/* Get the saved flags */
798	movl	%eax, 14*4(%esp)		/* Move saved flags into regs->flags location */
799						/* clobbering return ip */
800	movl	$__KERNEL_CS, 13*4(%esp)
801
802	movl	12*4(%esp), %eax		/* Load ip (1st parameter) */
803	subl	$MCOUNT_INSN_SIZE, %eax		/* Adjust ip */
804	movl	0x4(%ebp), %edx			/* Load parent ip (2nd parameter) */
805	movl	function_trace_op, %ecx		/* Save ftrace_pos in 3rd parameter */
806	pushl	%esp				/* Save pt_regs as 4th parameter */
807
808GLOBAL(ftrace_regs_call)
809	call	ftrace_stub
810
811	addl	$4, %esp			/* Skip pt_regs */
812	movl	14*4(%esp), %eax		/* Move flags back into cs */
813	movl	%eax, 13*4(%esp)		/* Needed to keep addl	from modifying flags */
814	movl	12*4(%esp), %eax		/* Get return ip from regs->ip */
815	movl	%eax, 14*4(%esp)		/* Put return ip back for ret */
816
817	popl	%ebx
818	popl	%ecx
819	popl	%edx
820	popl	%esi
821	popl	%edi
822	popl	%ebp
823	popl	%eax
824	popl	%ds
825	popl	%es
826	popl	%fs
827	popl	%gs
828	addl	$8, %esp			/* Skip orig_ax and ip */
829	popf					/* Pop flags at end (no addl to corrupt flags) */
830	jmp	ftrace_ret
831
832	popf
833	jmp	ftrace_stub
834#else /* ! CONFIG_DYNAMIC_FTRACE */
835
836ENTRY(mcount)
837	cmpl	$__PAGE_OFFSET, %esp
838	jb	ftrace_stub			/* Paging not enabled yet? */
839
840	cmpl	$ftrace_stub, ftrace_trace_function
841	jnz	trace
842#ifdef CONFIG_FUNCTION_GRAPH_TRACER
843	cmpl	$ftrace_stub, ftrace_graph_return
844	jnz	ftrace_graph_caller
845
846	cmpl	$ftrace_graph_entry_stub, ftrace_graph_entry
847	jnz	ftrace_graph_caller
848#endif
849.globl ftrace_stub
850ftrace_stub:
851	ret
852
853	/* taken from glibc */
854trace:
855	pushl	%eax
856	pushl	%ecx
857	pushl	%edx
858	movl	0xc(%esp), %eax
859	movl	0x4(%ebp), %edx
860	subl	$MCOUNT_INSN_SIZE, %eax
861
862	movl    ftrace_trace_function, %ecx
863	CALL_NOSPEC %ecx
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_NOSPEC %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_NOSPEC %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	ASM_CLAC
1071	pushl	$do_general_protection
1072	jmp	error_code
1073END(general_protection)
1074
1075#ifdef CONFIG_KVM_GUEST
1076ENTRY(async_page_fault)
1077	ASM_CLAC
1078	pushl	$do_async_page_fault
1079	jmp	error_code
1080END(async_page_fault)
1081#endif
1082