xref: /openbmc/linux/arch/x86/entry/entry_32.S (revision 1c2dd16a)
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/percpu.h>
39#include <asm/processor-flags.h>
40#include <asm/irq_vectors.h>
41#include <asm/cpufeatures.h>
42#include <asm/alternative-asm.h>
43#include <asm/asm.h>
44#include <asm/smap.h>
45#include <asm/frame.h>
46
47	.section .entry.text, "ax"
48
49/*
50 * We use macros for low-level operations which need to be overridden
51 * for paravirtualization.  The following will never clobber any registers:
52 *   INTERRUPT_RETURN (aka. "iret")
53 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
54 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
55 *
56 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
57 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
58 * Allowing a register to be clobbered can shrink the paravirt replacement
59 * enough to patch inline, increasing performance.
60 */
61
62#ifdef CONFIG_PREEMPT
63# define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
64#else
65# define preempt_stop(clobbers)
66# define resume_kernel		restore_all
67#endif
68
69.macro TRACE_IRQS_IRET
70#ifdef CONFIG_TRACE_IRQFLAGS
71	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
72	jz	1f
73	TRACE_IRQS_ON
741:
75#endif
76.endm
77
78/*
79 * User gs save/restore
80 *
81 * %gs is used for userland TLS and kernel only uses it for stack
82 * canary which is required to be at %gs:20 by gcc.  Read the comment
83 * at the top of stackprotector.h for more info.
84 *
85 * Local labels 98 and 99 are used.
86 */
87#ifdef CONFIG_X86_32_LAZY_GS
88
89 /* unfortunately push/pop can't be no-op */
90.macro PUSH_GS
91	pushl	$0
92.endm
93.macro POP_GS pop=0
94	addl	$(4 + \pop), %esp
95.endm
96.macro POP_GS_EX
97.endm
98
99 /* all the rest are no-op */
100.macro PTGS_TO_GS
101.endm
102.macro PTGS_TO_GS_EX
103.endm
104.macro GS_TO_REG reg
105.endm
106.macro REG_TO_PTGS reg
107.endm
108.macro SET_KERNEL_GS reg
109.endm
110
111#else	/* CONFIG_X86_32_LAZY_GS */
112
113.macro PUSH_GS
114	pushl	%gs
115.endm
116
117.macro POP_GS pop=0
11898:	popl	%gs
119  .if \pop <> 0
120	add	$\pop, %esp
121  .endif
122.endm
123.macro POP_GS_EX
124.pushsection .fixup, "ax"
12599:	movl	$0, (%esp)
126	jmp	98b
127.popsection
128	_ASM_EXTABLE(98b, 99b)
129.endm
130
131.macro PTGS_TO_GS
13298:	mov	PT_GS(%esp), %gs
133.endm
134.macro PTGS_TO_GS_EX
135.pushsection .fixup, "ax"
13699:	movl	$0, PT_GS(%esp)
137	jmp	98b
138.popsection
139	_ASM_EXTABLE(98b, 99b)
140.endm
141
142.macro GS_TO_REG reg
143	movl	%gs, \reg
144.endm
145.macro REG_TO_PTGS reg
146	movl	\reg, PT_GS(%esp)
147.endm
148.macro SET_KERNEL_GS reg
149	movl	$(__KERNEL_STACK_CANARY), \reg
150	movl	\reg, %gs
151.endm
152
153#endif /* CONFIG_X86_32_LAZY_GS */
154
155.macro SAVE_ALL pt_regs_ax=%eax
156	cld
157	PUSH_GS
158	pushl	%fs
159	pushl	%es
160	pushl	%ds
161	pushl	\pt_regs_ax
162	pushl	%ebp
163	pushl	%edi
164	pushl	%esi
165	pushl	%edx
166	pushl	%ecx
167	pushl	%ebx
168	movl	$(__USER_DS), %edx
169	movl	%edx, %ds
170	movl	%edx, %es
171	movl	$(__KERNEL_PERCPU), %edx
172	movl	%edx, %fs
173	SET_KERNEL_GS %edx
174.endm
175
176/*
177 * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
178 * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
179 * is just setting the LSB, which makes it an invalid stack address and is also
180 * a signal to the unwinder that it's a pt_regs pointer in disguise.
181 *
182 * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
183 * original rbp.
184 */
185.macro ENCODE_FRAME_POINTER
186#ifdef CONFIG_FRAME_POINTER
187	mov %esp, %ebp
188	orl $0x1, %ebp
189#endif
190.endm
191
192.macro RESTORE_INT_REGS
193	popl	%ebx
194	popl	%ecx
195	popl	%edx
196	popl	%esi
197	popl	%edi
198	popl	%ebp
199	popl	%eax
200.endm
201
202.macro RESTORE_REGS pop=0
203	RESTORE_INT_REGS
2041:	popl	%ds
2052:	popl	%es
2063:	popl	%fs
207	POP_GS \pop
208.pushsection .fixup, "ax"
2094:	movl	$0, (%esp)
210	jmp	1b
2115:	movl	$0, (%esp)
212	jmp	2b
2136:	movl	$0, (%esp)
214	jmp	3b
215.popsection
216	_ASM_EXTABLE(1b, 4b)
217	_ASM_EXTABLE(2b, 5b)
218	_ASM_EXTABLE(3b, 6b)
219	POP_GS_EX
220.endm
221
222/*
223 * %eax: prev task
224 * %edx: next task
225 */
226ENTRY(__switch_to_asm)
227	/*
228	 * Save callee-saved registers
229	 * This must match the order in struct inactive_task_frame
230	 */
231	pushl	%ebp
232	pushl	%ebx
233	pushl	%edi
234	pushl	%esi
235
236	/* switch stack */
237	movl	%esp, TASK_threadsp(%eax)
238	movl	TASK_threadsp(%edx), %esp
239
240#ifdef CONFIG_CC_STACKPROTECTOR
241	movl	TASK_stack_canary(%edx), %ebx
242	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
243#endif
244
245	/* restore callee-saved registers */
246	popl	%esi
247	popl	%edi
248	popl	%ebx
249	popl	%ebp
250
251	jmp	__switch_to
252END(__switch_to_asm)
253
254/*
255 * A newly forked process directly context switches into this address.
256 *
257 * eax: prev task we switched from
258 * ebx: kernel thread func (NULL for user thread)
259 * edi: kernel thread arg
260 */
261ENTRY(ret_from_fork)
262	FRAME_BEGIN		/* help unwinder find end of stack */
263
264	/*
265	 * schedule_tail() is asmlinkage so we have to put its 'prev' argument
266	 * on the stack.
267	 */
268	pushl	%eax
269	call	schedule_tail
270	popl	%eax
271
272	testl	%ebx, %ebx
273	jnz	1f		/* kernel threads are uncommon */
274
2752:
276	/* When we fork, we trace the syscall return in the child, too. */
277	leal	FRAME_OFFSET(%esp), %eax
278	call    syscall_return_slowpath
279	FRAME_END
280	jmp     restore_all
281
282	/* kernel thread */
2831:	movl	%edi, %eax
284	call	*%ebx
285	/*
286	 * A kernel thread is allowed to return here after successfully
287	 * calling do_execve().  Exit to userspace to complete the execve()
288	 * syscall.
289	 */
290	movl	$0, PT_EAX(%esp)
291	jmp	2b
292END(ret_from_fork)
293
294/*
295 * Return to user mode is not as complex as all this looks,
296 * but we want the default path for a system call return to
297 * go as quickly as possible which is why some of this is
298 * less clear than it otherwise should be.
299 */
300
301	# userspace resumption stub bypassing syscall exit tracing
302	ALIGN
303ret_from_exception:
304	preempt_stop(CLBR_ANY)
305ret_from_intr:
306#ifdef CONFIG_VM86
307	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
308	movb	PT_CS(%esp), %al
309	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
310#else
311	/*
312	 * We can be coming here from child spawned by kernel_thread().
313	 */
314	movl	PT_CS(%esp), %eax
315	andl	$SEGMENT_RPL_MASK, %eax
316#endif
317	cmpl	$USER_RPL, %eax
318	jb	resume_kernel			# not returning to v8086 or userspace
319
320ENTRY(resume_userspace)
321	DISABLE_INTERRUPTS(CLBR_ANY)
322	TRACE_IRQS_OFF
323	movl	%esp, %eax
324	call	prepare_exit_to_usermode
325	jmp	restore_all
326END(ret_from_exception)
327
328#ifdef CONFIG_PREEMPT
329ENTRY(resume_kernel)
330	DISABLE_INTERRUPTS(CLBR_ANY)
331.Lneed_resched:
332	cmpl	$0, PER_CPU_VAR(__preempt_count)
333	jnz	restore_all
334	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
335	jz	restore_all
336	call	preempt_schedule_irq
337	jmp	.Lneed_resched
338END(resume_kernel)
339#endif
340
341GLOBAL(__begin_SYSENTER_singlestep_region)
342/*
343 * All code from here through __end_SYSENTER_singlestep_region is subject
344 * to being single-stepped if a user program sets TF and executes SYSENTER.
345 * There is absolutely nothing that we can do to prevent this from happening
346 * (thanks Intel!).  To keep our handling of this situation as simple as
347 * possible, we handle TF just like AC and NT, except that our #DB handler
348 * will ignore all of the single-step traps generated in this range.
349 */
350
351#ifdef CONFIG_XEN
352/*
353 * Xen doesn't set %esp to be precisely what the normal SYSENTER
354 * entry point expects, so fix it up before using the normal path.
355 */
356ENTRY(xen_sysenter_target)
357	addl	$5*4, %esp			/* remove xen-provided frame */
358	jmp	.Lsysenter_past_esp
359#endif
360
361/*
362 * 32-bit SYSENTER entry.
363 *
364 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
365 * if X86_FEATURE_SEP is available.  This is the preferred system call
366 * entry on 32-bit systems.
367 *
368 * The SYSENTER instruction, in principle, should *only* occur in the
369 * vDSO.  In practice, a small number of Android devices were shipped
370 * with a copy of Bionic that inlined a SYSENTER instruction.  This
371 * never happened in any of Google's Bionic versions -- it only happened
372 * in a narrow range of Intel-provided versions.
373 *
374 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
375 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
376 * SYSENTER does not save anything on the stack,
377 * and does not save old EIP (!!!), ESP, or EFLAGS.
378 *
379 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
380 * user and/or vm86 state), we explicitly disable the SYSENTER
381 * instruction in vm86 mode by reprogramming the MSRs.
382 *
383 * Arguments:
384 * eax  system call number
385 * ebx  arg1
386 * ecx  arg2
387 * edx  arg3
388 * esi  arg4
389 * edi  arg5
390 * ebp  user stack
391 * 0(%ebp) arg6
392 */
393ENTRY(entry_SYSENTER_32)
394	movl	TSS_sysenter_sp0(%esp), %esp
395.Lsysenter_past_esp:
396	pushl	$__USER_DS		/* pt_regs->ss */
397	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
398	pushfl				/* pt_regs->flags (except IF = 0) */
399	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
400	pushl	$__USER_CS		/* pt_regs->cs */
401	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
402	pushl	%eax			/* pt_regs->orig_ax */
403	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
404
405	/*
406	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
407	 * and TF ourselves.  To save a few cycles, we can check whether
408	 * either was set instead of doing an unconditional popfq.
409	 * This needs to happen before enabling interrupts so that
410	 * we don't get preempted with NT set.
411	 *
412	 * If TF is set, we will single-step all the way to here -- do_debug
413	 * will ignore all the traps.  (Yes, this is slow, but so is
414	 * single-stepping in general.  This allows us to avoid having
415	 * a more complicated code to handle the case where a user program
416	 * forces us to single-step through the SYSENTER entry code.)
417	 *
418	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
419	 * out-of-line as an optimization: NT is unlikely to be set in the
420	 * majority of the cases and instead of polluting the I$ unnecessarily,
421	 * we're keeping that code behind a branch which will predict as
422	 * not-taken and therefore its instructions won't be fetched.
423	 */
424	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
425	jnz	.Lsysenter_fix_flags
426.Lsysenter_flags_fixed:
427
428	/*
429	 * User mode is traced as though IRQs are on, and SYSENTER
430	 * turned them off.
431	 */
432	TRACE_IRQS_OFF
433
434	movl	%esp, %eax
435	call	do_fast_syscall_32
436	/* XEN PV guests always use IRET path */
437	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
438		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
439
440/* Opportunistic SYSEXIT */
441	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
442	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
443	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
4441:	mov	PT_FS(%esp), %fs
445	PTGS_TO_GS
446	popl	%ebx			/* pt_regs->bx */
447	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
448	popl	%esi			/* pt_regs->si */
449	popl	%edi			/* pt_regs->di */
450	popl	%ebp			/* pt_regs->bp */
451	popl	%eax			/* pt_regs->ax */
452
453	/*
454	 * Restore all flags except IF. (We restore IF separately because
455	 * STI gives a one-instruction window in which we won't be interrupted,
456	 * whereas POPF does not.)
457	 */
458	addl	$PT_EFLAGS-PT_DS, %esp	/* point esp at pt_regs->flags */
459	btr	$X86_EFLAGS_IF_BIT, (%esp)
460	popfl
461
462	/*
463	 * Return back to the vDSO, which will pop ecx and edx.
464	 * Don't bother with DS and ES (they already contain __USER_DS).
465	 */
466	sti
467	sysexit
468
469.pushsection .fixup, "ax"
4702:	movl	$0, PT_FS(%esp)
471	jmp	1b
472.popsection
473	_ASM_EXTABLE(1b, 2b)
474	PTGS_TO_GS_EX
475
476.Lsysenter_fix_flags:
477	pushl	$X86_EFLAGS_FIXED
478	popfl
479	jmp	.Lsysenter_flags_fixed
480GLOBAL(__end_SYSENTER_singlestep_region)
481ENDPROC(entry_SYSENTER_32)
482
483/*
484 * 32-bit legacy system call entry.
485 *
486 * 32-bit x86 Linux system calls traditionally used the INT $0x80
487 * instruction.  INT $0x80 lands here.
488 *
489 * This entry point can be used by any 32-bit perform system calls.
490 * Instances of INT $0x80 can be found inline in various programs and
491 * libraries.  It is also used by the vDSO's __kernel_vsyscall
492 * fallback for hardware that doesn't support a faster entry method.
493 * Restarted 32-bit system calls also fall back to INT $0x80
494 * regardless of what instruction was originally used to do the system
495 * call.  (64-bit programs can use INT $0x80 as well, but they can
496 * only run on 64-bit kernels and therefore land in
497 * entry_INT80_compat.)
498 *
499 * This is considered a slow path.  It is not used by most libc
500 * implementations on modern hardware except during process startup.
501 *
502 * Arguments:
503 * eax  system call number
504 * ebx  arg1
505 * ecx  arg2
506 * edx  arg3
507 * esi  arg4
508 * edi  arg5
509 * ebp  arg6
510 */
511ENTRY(entry_INT80_32)
512	ASM_CLAC
513	pushl	%eax			/* pt_regs->orig_ax */
514	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
515
516	/*
517	 * User mode is traced as though IRQs are on, and the interrupt gate
518	 * turned them off.
519	 */
520	TRACE_IRQS_OFF
521
522	movl	%esp, %eax
523	call	do_int80_syscall_32
524.Lsyscall_32_done:
525
526restore_all:
527	TRACE_IRQS_IRET
528.Lrestore_all_notrace:
529#ifdef CONFIG_X86_ESPFIX32
530	ALTERNATIVE	"jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
531
532	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
533	/*
534	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
535	 * are returning to the kernel.
536	 * See comments in process.c:copy_thread() for details.
537	 */
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	je .Lldt_ss				# returning to user-space with LDT SS
543#endif
544.Lrestore_nocheck:
545	RESTORE_REGS 4				# skip orig_eax/error_code
546.Lirq_return:
547	INTERRUPT_RETURN
548
549.section .fixup, "ax"
550ENTRY(iret_exc	)
551	pushl	$0				# no error code
552	pushl	$do_iret_error
553	jmp	common_exception
554.previous
555	_ASM_EXTABLE(.Lirq_return, iret_exc)
556
557#ifdef CONFIG_X86_ESPFIX32
558.Lldt_ss:
559/*
560 * Setup and switch to ESPFIX stack
561 *
562 * We're returning to userspace with a 16 bit stack. The CPU will not
563 * restore the high word of ESP for us on executing iret... This is an
564 * "official" bug of all the x86-compatible CPUs, which we can work
565 * around to make dosemu and wine happy. We do this by preloading the
566 * high word of ESP with the high word of the userspace ESP while
567 * compensating for the offset by changing to the ESPFIX segment with
568 * a base address that matches for the difference.
569 */
570#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
571	mov	%esp, %edx			/* load kernel esp */
572	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
573	mov	%dx, %ax			/* eax: new kernel esp */
574	sub	%eax, %edx			/* offset (low word is 0) */
575	shr	$16, %edx
576	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
577	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
578	pushl	$__ESPFIX_SS
579	pushl	%eax				/* new kernel esp */
580	/*
581	 * Disable interrupts, but do not irqtrace this section: we
582	 * will soon execute iret and the tracer was already set to
583	 * the irqstate after the IRET:
584	 */
585	DISABLE_INTERRUPTS(CLBR_ANY)
586	lss	(%esp), %esp			/* switch to espfix segment */
587	jmp	.Lrestore_nocheck
588#endif
589ENDPROC(entry_INT80_32)
590
591.macro FIXUP_ESPFIX_STACK
592/*
593 * Switch back for ESPFIX stack to the normal zerobased stack
594 *
595 * We can't call C functions using the ESPFIX stack. This code reads
596 * the high word of the segment base from the GDT and swiches to the
597 * normal stack and adjusts ESP with the matching offset.
598 */
599#ifdef CONFIG_X86_ESPFIX32
600	/* fixup the stack */
601	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
602	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
603	shl	$16, %eax
604	addl	%esp, %eax			/* the adjusted stack pointer */
605	pushl	$__KERNEL_DS
606	pushl	%eax
607	lss	(%esp), %esp			/* switch to the normal stack segment */
608#endif
609.endm
610.macro UNWIND_ESPFIX_STACK
611#ifdef CONFIG_X86_ESPFIX32
612	movl	%ss, %eax
613	/* see if on espfix stack */
614	cmpw	$__ESPFIX_SS, %ax
615	jne	27f
616	movl	$__KERNEL_DS, %eax
617	movl	%eax, %ds
618	movl	%eax, %es
619	/* switch to normal stack */
620	FIXUP_ESPFIX_STACK
62127:
622#endif
623.endm
624
625/*
626 * Build the entry stubs with some assembler magic.
627 * We pack 1 stub into every 8-byte block.
628 */
629	.align 8
630ENTRY(irq_entries_start)
631    vector=FIRST_EXTERNAL_VECTOR
632    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
633	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
634    vector=vector+1
635	jmp	common_interrupt
636	.align	8
637    .endr
638END(irq_entries_start)
639
640/*
641 * the CPU automatically disables interrupts when executing an IRQ vector,
642 * so IRQ-flags tracing has to follow that:
643 */
644	.p2align CONFIG_X86_L1_CACHE_SHIFT
645common_interrupt:
646	ASM_CLAC
647	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
648	SAVE_ALL
649	ENCODE_FRAME_POINTER
650	TRACE_IRQS_OFF
651	movl	%esp, %eax
652	call	do_IRQ
653	jmp	ret_from_intr
654ENDPROC(common_interrupt)
655
656#define BUILD_INTERRUPT3(name, nr, fn)	\
657ENTRY(name)				\
658	ASM_CLAC;			\
659	pushl	$~(nr);			\
660	SAVE_ALL;			\
661	ENCODE_FRAME_POINTER;		\
662	TRACE_IRQS_OFF			\
663	movl	%esp, %eax;		\
664	call	fn;			\
665	jmp	ret_from_intr;		\
666ENDPROC(name)
667
668
669#ifdef CONFIG_TRACING
670# define TRACE_BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
671#else
672# define TRACE_BUILD_INTERRUPT(name, nr)
673#endif
674
675#define BUILD_INTERRUPT(name, nr)		\
676	BUILD_INTERRUPT3(name, nr, smp_##name);	\
677	TRACE_BUILD_INTERRUPT(name, nr)
678
679/* The include is where all of the SMP etc. interrupts come from */
680#include <asm/entry_arch.h>
681
682ENTRY(coprocessor_error)
683	ASM_CLAC
684	pushl	$0
685	pushl	$do_coprocessor_error
686	jmp	common_exception
687END(coprocessor_error)
688
689ENTRY(simd_coprocessor_error)
690	ASM_CLAC
691	pushl	$0
692#ifdef CONFIG_X86_INVD_BUG
693	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
694	ALTERNATIVE "pushl	$do_general_protection",	\
695		    "pushl	$do_simd_coprocessor_error",	\
696		    X86_FEATURE_XMM
697#else
698	pushl	$do_simd_coprocessor_error
699#endif
700	jmp	common_exception
701END(simd_coprocessor_error)
702
703ENTRY(device_not_available)
704	ASM_CLAC
705	pushl	$-1				# mark this as an int
706	pushl	$do_device_not_available
707	jmp	common_exception
708END(device_not_available)
709
710#ifdef CONFIG_PARAVIRT
711ENTRY(native_iret)
712	iret
713	_ASM_EXTABLE(native_iret, iret_exc)
714END(native_iret)
715#endif
716
717ENTRY(overflow)
718	ASM_CLAC
719	pushl	$0
720	pushl	$do_overflow
721	jmp	common_exception
722END(overflow)
723
724ENTRY(bounds)
725	ASM_CLAC
726	pushl	$0
727	pushl	$do_bounds
728	jmp	common_exception
729END(bounds)
730
731ENTRY(invalid_op)
732	ASM_CLAC
733	pushl	$0
734	pushl	$do_invalid_op
735	jmp	common_exception
736END(invalid_op)
737
738ENTRY(coprocessor_segment_overrun)
739	ASM_CLAC
740	pushl	$0
741	pushl	$do_coprocessor_segment_overrun
742	jmp	common_exception
743END(coprocessor_segment_overrun)
744
745ENTRY(invalid_TSS)
746	ASM_CLAC
747	pushl	$do_invalid_TSS
748	jmp	common_exception
749END(invalid_TSS)
750
751ENTRY(segment_not_present)
752	ASM_CLAC
753	pushl	$do_segment_not_present
754	jmp	common_exception
755END(segment_not_present)
756
757ENTRY(stack_segment)
758	ASM_CLAC
759	pushl	$do_stack_segment
760	jmp	common_exception
761END(stack_segment)
762
763ENTRY(alignment_check)
764	ASM_CLAC
765	pushl	$do_alignment_check
766	jmp	common_exception
767END(alignment_check)
768
769ENTRY(divide_error)
770	ASM_CLAC
771	pushl	$0				# no error code
772	pushl	$do_divide_error
773	jmp	common_exception
774END(divide_error)
775
776#ifdef CONFIG_X86_MCE
777ENTRY(machine_check)
778	ASM_CLAC
779	pushl	$0
780	pushl	machine_check_vector
781	jmp	common_exception
782END(machine_check)
783#endif
784
785ENTRY(spurious_interrupt_bug)
786	ASM_CLAC
787	pushl	$0
788	pushl	$do_spurious_interrupt_bug
789	jmp	common_exception
790END(spurious_interrupt_bug)
791
792#ifdef CONFIG_XEN
793ENTRY(xen_hypervisor_callback)
794	pushl	$-1				/* orig_ax = -1 => not a system call */
795	SAVE_ALL
796	ENCODE_FRAME_POINTER
797	TRACE_IRQS_OFF
798
799	/*
800	 * Check to see if we got the event in the critical
801	 * region in xen_iret_direct, after we've reenabled
802	 * events and checked for pending events.  This simulates
803	 * iret instruction's behaviour where it delivers a
804	 * pending interrupt when enabling interrupts:
805	 */
806	movl	PT_EIP(%esp), %eax
807	cmpl	$xen_iret_start_crit, %eax
808	jb	1f
809	cmpl	$xen_iret_end_crit, %eax
810	jae	1f
811
812	jmp	xen_iret_crit_fixup
813
814ENTRY(xen_do_upcall)
8151:	mov	%esp, %eax
816	call	xen_evtchn_do_upcall
817#ifndef CONFIG_PREEMPT
818	call	xen_maybe_preempt_hcall
819#endif
820	jmp	ret_from_intr
821ENDPROC(xen_hypervisor_callback)
822
823/*
824 * Hypervisor uses this for application faults while it executes.
825 * We get here for two reasons:
826 *  1. Fault while reloading DS, ES, FS or GS
827 *  2. Fault while executing IRET
828 * Category 1 we fix up by reattempting the load, and zeroing the segment
829 * register if the load fails.
830 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
831 * normal Linux return path in this case because if we use the IRET hypercall
832 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
833 * We distinguish between categories by maintaining a status value in EAX.
834 */
835ENTRY(xen_failsafe_callback)
836	pushl	%eax
837	movl	$1, %eax
8381:	mov	4(%esp), %ds
8392:	mov	8(%esp), %es
8403:	mov	12(%esp), %fs
8414:	mov	16(%esp), %gs
842	/* EAX == 0 => Category 1 (Bad segment)
843	   EAX != 0 => Category 2 (Bad IRET) */
844	testl	%eax, %eax
845	popl	%eax
846	lea	16(%esp), %esp
847	jz	5f
848	jmp	iret_exc
8495:	pushl	$-1				/* orig_ax = -1 => not a system call */
850	SAVE_ALL
851	ENCODE_FRAME_POINTER
852	jmp	ret_from_exception
853
854.section .fixup, "ax"
8556:	xorl	%eax, %eax
856	movl	%eax, 4(%esp)
857	jmp	1b
8587:	xorl	%eax, %eax
859	movl	%eax, 8(%esp)
860	jmp	2b
8618:	xorl	%eax, %eax
862	movl	%eax, 12(%esp)
863	jmp	3b
8649:	xorl	%eax, %eax
865	movl	%eax, 16(%esp)
866	jmp	4b
867.previous
868	_ASM_EXTABLE(1b, 6b)
869	_ASM_EXTABLE(2b, 7b)
870	_ASM_EXTABLE(3b, 8b)
871	_ASM_EXTABLE(4b, 9b)
872ENDPROC(xen_failsafe_callback)
873
874BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
875		xen_evtchn_do_upcall)
876
877#endif /* CONFIG_XEN */
878
879#if IS_ENABLED(CONFIG_HYPERV)
880
881BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
882	hyperv_vector_handler)
883
884#endif /* CONFIG_HYPERV */
885
886#ifdef CONFIG_TRACING
887ENTRY(trace_page_fault)
888	ASM_CLAC
889	pushl	$trace_do_page_fault
890	jmp	common_exception
891END(trace_page_fault)
892#endif
893
894ENTRY(page_fault)
895	ASM_CLAC
896	pushl	$do_page_fault
897	ALIGN
898	jmp common_exception
899END(page_fault)
900
901common_exception:
902	/* the function address is in %gs's slot on the stack */
903	pushl	%fs
904	pushl	%es
905	pushl	%ds
906	pushl	%eax
907	pushl	%ebp
908	pushl	%edi
909	pushl	%esi
910	pushl	%edx
911	pushl	%ecx
912	pushl	%ebx
913	ENCODE_FRAME_POINTER
914	cld
915	movl	$(__KERNEL_PERCPU), %ecx
916	movl	%ecx, %fs
917	UNWIND_ESPFIX_STACK
918	GS_TO_REG %ecx
919	movl	PT_GS(%esp), %edi		# get the function address
920	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
921	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
922	REG_TO_PTGS %ecx
923	SET_KERNEL_GS %ecx
924	movl	$(__USER_DS), %ecx
925	movl	%ecx, %ds
926	movl	%ecx, %es
927	TRACE_IRQS_OFF
928	movl	%esp, %eax			# pt_regs pointer
929	call	*%edi
930	jmp	ret_from_exception
931END(common_exception)
932
933ENTRY(debug)
934	/*
935	 * #DB can happen at the first instruction of
936	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
937	 * happens, then we will be running on a very small stack.  We
938	 * need to detect this condition and switch to the thread
939	 * stack before calling any C code at all.
940	 *
941	 * If you edit this code, keep in mind that NMIs can happen in here.
942	 */
943	ASM_CLAC
944	pushl	$-1				# mark this as an int
945	SAVE_ALL
946	ENCODE_FRAME_POINTER
947	xorl	%edx, %edx			# error code 0
948	movl	%esp, %eax			# pt_regs pointer
949
950	/* Are we currently on the SYSENTER stack? */
951	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
952	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
953	cmpl	$SIZEOF_SYSENTER_stack, %ecx
954	jb	.Ldebug_from_sysenter_stack
955
956	TRACE_IRQS_OFF
957	call	do_debug
958	jmp	ret_from_exception
959
960.Ldebug_from_sysenter_stack:
961	/* We're on the SYSENTER stack.  Switch off. */
962	movl	%esp, %ebx
963	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
964	TRACE_IRQS_OFF
965	call	do_debug
966	movl	%ebx, %esp
967	jmp	ret_from_exception
968END(debug)
969
970/*
971 * NMI is doubly nasty.  It can happen on the first instruction of
972 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
973 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
974 * switched stacks.  We handle both conditions by simply checking whether we
975 * interrupted kernel code running on the SYSENTER stack.
976 */
977ENTRY(nmi)
978	ASM_CLAC
979#ifdef CONFIG_X86_ESPFIX32
980	pushl	%eax
981	movl	%ss, %eax
982	cmpw	$__ESPFIX_SS, %ax
983	popl	%eax
984	je	.Lnmi_espfix_stack
985#endif
986
987	pushl	%eax				# pt_regs->orig_ax
988	SAVE_ALL
989	ENCODE_FRAME_POINTER
990	xorl	%edx, %edx			# zero error code
991	movl	%esp, %eax			# pt_regs pointer
992
993	/* Are we currently on the SYSENTER stack? */
994	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
995	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
996	cmpl	$SIZEOF_SYSENTER_stack, %ecx
997	jb	.Lnmi_from_sysenter_stack
998
999	/* Not on SYSENTER stack. */
1000	call	do_nmi
1001	jmp	.Lrestore_all_notrace
1002
1003.Lnmi_from_sysenter_stack:
1004	/*
1005	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1006	 * is using the thread stack right now, so it's safe for us to use it.
1007	 */
1008	movl	%esp, %ebx
1009	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1010	call	do_nmi
1011	movl	%ebx, %esp
1012	jmp	.Lrestore_all_notrace
1013
1014#ifdef CONFIG_X86_ESPFIX32
1015.Lnmi_espfix_stack:
1016	/*
1017	 * create the pointer to lss back
1018	 */
1019	pushl	%ss
1020	pushl	%esp
1021	addl	$4, (%esp)
1022	/* copy the iret frame of 12 bytes */
1023	.rept 3
1024	pushl	16(%esp)
1025	.endr
1026	pushl	%eax
1027	SAVE_ALL
1028	ENCODE_FRAME_POINTER
1029	FIXUP_ESPFIX_STACK			# %eax == %esp
1030	xorl	%edx, %edx			# zero error code
1031	call	do_nmi
1032	RESTORE_REGS
1033	lss	12+4(%esp), %esp		# back to espfix stack
1034	jmp	.Lirq_return
1035#endif
1036END(nmi)
1037
1038ENTRY(int3)
1039	ASM_CLAC
1040	pushl	$-1				# mark this as an int
1041	SAVE_ALL
1042	ENCODE_FRAME_POINTER
1043	TRACE_IRQS_OFF
1044	xorl	%edx, %edx			# zero error code
1045	movl	%esp, %eax			# pt_regs pointer
1046	call	do_int3
1047	jmp	ret_from_exception
1048END(int3)
1049
1050ENTRY(general_protection)
1051	pushl	$do_general_protection
1052	jmp	common_exception
1053END(general_protection)
1054
1055#ifdef CONFIG_KVM_GUEST
1056ENTRY(async_page_fault)
1057	ASM_CLAC
1058	pushl	$do_async_page_fault
1059	jmp	common_exception
1060END(async_page_fault)
1061#endif
1062
1063ENTRY(rewind_stack_do_exit)
1064	/* Prevent any naive code from trying to unwind to our caller. */
1065	xorl	%ebp, %ebp
1066
1067	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1068	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1069
1070	call	do_exit
10711:	jmp 1b
1072END(rewind_stack_do_exit)
1073