xref: /linux/arch/x86/entry/entry_32.S (revision 4f58e6dceb0e44ca8f21568ed81e1df24e55964c)
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
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
207/*
208 * %eax: prev task
209 * %edx: next task
210 */
211ENTRY(__switch_to_asm)
212	/*
213	 * Save callee-saved registers
214	 * This must match the order in struct inactive_task_frame
215	 */
216	pushl	%ebp
217	pushl	%ebx
218	pushl	%edi
219	pushl	%esi
220
221	/* switch stack */
222	movl	%esp, TASK_threadsp(%eax)
223	movl	TASK_threadsp(%edx), %esp
224
225#ifdef CONFIG_CC_STACKPROTECTOR
226	movl	TASK_stack_canary(%edx), %ebx
227	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
228#endif
229
230	/* restore callee-saved registers */
231	popl	%esi
232	popl	%edi
233	popl	%ebx
234	popl	%ebp
235
236	jmp	__switch_to
237END(__switch_to_asm)
238
239/*
240 * A newly forked process directly context switches into this address.
241 *
242 * eax: prev task we switched from
243 * ebx: kernel thread func (NULL for user thread)
244 * edi: kernel thread arg
245 */
246ENTRY(ret_from_fork)
247	pushl	%eax
248	call	schedule_tail
249	popl	%eax
250
251	testl	%ebx, %ebx
252	jnz	1f		/* kernel threads are uncommon */
253
2542:
255	/* When we fork, we trace the syscall return in the child, too. */
256	movl    %esp, %eax
257	call    syscall_return_slowpath
258	jmp     restore_all
259
260	/* kernel thread */
2611:	movl	%edi, %eax
262	call	*%ebx
263	/*
264	 * A kernel thread is allowed to return here after successfully
265	 * calling do_execve().  Exit to userspace to complete the execve()
266	 * syscall.
267	 */
268	movl	$0, PT_EAX(%esp)
269	jmp	2b
270END(ret_from_fork)
271
272/*
273 * Return to user mode is not as complex as all this looks,
274 * but we want the default path for a system call return to
275 * go as quickly as possible which is why some of this is
276 * less clear than it otherwise should be.
277 */
278
279	# userspace resumption stub bypassing syscall exit tracing
280	ALIGN
281ret_from_exception:
282	preempt_stop(CLBR_ANY)
283ret_from_intr:
284#ifdef CONFIG_VM86
285	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
286	movb	PT_CS(%esp), %al
287	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
288#else
289	/*
290	 * We can be coming here from child spawned by kernel_thread().
291	 */
292	movl	PT_CS(%esp), %eax
293	andl	$SEGMENT_RPL_MASK, %eax
294#endif
295	cmpl	$USER_RPL, %eax
296	jb	resume_kernel			# not returning to v8086 or userspace
297
298ENTRY(resume_userspace)
299	DISABLE_INTERRUPTS(CLBR_ANY)
300	TRACE_IRQS_OFF
301	movl	%esp, %eax
302	call	prepare_exit_to_usermode
303	jmp	restore_all
304END(ret_from_exception)
305
306#ifdef CONFIG_PREEMPT
307ENTRY(resume_kernel)
308	DISABLE_INTERRUPTS(CLBR_ANY)
309need_resched:
310	cmpl	$0, PER_CPU_VAR(__preempt_count)
311	jnz	restore_all
312	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
313	jz	restore_all
314	call	preempt_schedule_irq
315	jmp	need_resched
316END(resume_kernel)
317#endif
318
319GLOBAL(__begin_SYSENTER_singlestep_region)
320/*
321 * All code from here through __end_SYSENTER_singlestep_region is subject
322 * to being single-stepped if a user program sets TF and executes SYSENTER.
323 * There is absolutely nothing that we can do to prevent this from happening
324 * (thanks Intel!).  To keep our handling of this situation as simple as
325 * possible, we handle TF just like AC and NT, except that our #DB handler
326 * will ignore all of the single-step traps generated in this range.
327 */
328
329#ifdef CONFIG_XEN
330/*
331 * Xen doesn't set %esp to be precisely what the normal SYSENTER
332 * entry point expects, so fix it up before using the normal path.
333 */
334ENTRY(xen_sysenter_target)
335	addl	$5*4, %esp			/* remove xen-provided frame */
336	jmp	sysenter_past_esp
337#endif
338
339/*
340 * 32-bit SYSENTER entry.
341 *
342 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
343 * if X86_FEATURE_SEP is available.  This is the preferred system call
344 * entry on 32-bit systems.
345 *
346 * The SYSENTER instruction, in principle, should *only* occur in the
347 * vDSO.  In practice, a small number of Android devices were shipped
348 * with a copy of Bionic that inlined a SYSENTER instruction.  This
349 * never happened in any of Google's Bionic versions -- it only happened
350 * in a narrow range of Intel-provided versions.
351 *
352 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
353 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
354 * SYSENTER does not save anything on the stack,
355 * and does not save old EIP (!!!), ESP, or EFLAGS.
356 *
357 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
358 * user and/or vm86 state), we explicitly disable the SYSENTER
359 * instruction in vm86 mode by reprogramming the MSRs.
360 *
361 * Arguments:
362 * eax  system call number
363 * ebx  arg1
364 * ecx  arg2
365 * edx  arg3
366 * esi  arg4
367 * edi  arg5
368 * ebp  user stack
369 * 0(%ebp) arg6
370 */
371ENTRY(entry_SYSENTER_32)
372	movl	TSS_sysenter_sp0(%esp), %esp
373sysenter_past_esp:
374	pushl	$__USER_DS		/* pt_regs->ss */
375	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
376	pushfl				/* pt_regs->flags (except IF = 0) */
377	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
378	pushl	$__USER_CS		/* pt_regs->cs */
379	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
380	pushl	%eax			/* pt_regs->orig_ax */
381	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
382
383	/*
384	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
385	 * and TF ourselves.  To save a few cycles, we can check whether
386	 * either was set instead of doing an unconditional popfq.
387	 * This needs to happen before enabling interrupts so that
388	 * we don't get preempted with NT set.
389	 *
390	 * If TF is set, we will single-step all the way to here -- do_debug
391	 * will ignore all the traps.  (Yes, this is slow, but so is
392	 * single-stepping in general.  This allows us to avoid having
393	 * a more complicated code to handle the case where a user program
394	 * forces us to single-step through the SYSENTER entry code.)
395	 *
396	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
397	 * out-of-line as an optimization: NT is unlikely to be set in the
398	 * majority of the cases and instead of polluting the I$ unnecessarily,
399	 * we're keeping that code behind a branch which will predict as
400	 * not-taken and therefore its instructions won't be fetched.
401	 */
402	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
403	jnz	.Lsysenter_fix_flags
404.Lsysenter_flags_fixed:
405
406	/*
407	 * User mode is traced as though IRQs are on, and SYSENTER
408	 * turned them off.
409	 */
410	TRACE_IRQS_OFF
411
412	movl	%esp, %eax
413	call	do_fast_syscall_32
414	/* XEN PV guests always use IRET path */
415	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
416		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
417
418/* Opportunistic SYSEXIT */
419	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
420	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
421	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
4221:	mov	PT_FS(%esp), %fs
423	PTGS_TO_GS
424	popl	%ebx			/* pt_regs->bx */
425	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
426	popl	%esi			/* pt_regs->si */
427	popl	%edi			/* pt_regs->di */
428	popl	%ebp			/* pt_regs->bp */
429	popl	%eax			/* pt_regs->ax */
430
431	/*
432	 * Restore all flags except IF. (We restore IF separately because
433	 * STI gives a one-instruction window in which we won't be interrupted,
434	 * whereas POPF does not.)
435	 */
436	addl	$PT_EFLAGS-PT_DS, %esp	/* point esp at pt_regs->flags */
437	btr	$X86_EFLAGS_IF_BIT, (%esp)
438	popfl
439
440	/*
441	 * Return back to the vDSO, which will pop ecx and edx.
442	 * Don't bother with DS and ES (they already contain __USER_DS).
443	 */
444	sti
445	sysexit
446
447.pushsection .fixup, "ax"
4482:	movl	$0, PT_FS(%esp)
449	jmp	1b
450.popsection
451	_ASM_EXTABLE(1b, 2b)
452	PTGS_TO_GS_EX
453
454.Lsysenter_fix_flags:
455	pushl	$X86_EFLAGS_FIXED
456	popfl
457	jmp	.Lsysenter_flags_fixed
458GLOBAL(__end_SYSENTER_singlestep_region)
459ENDPROC(entry_SYSENTER_32)
460
461/*
462 * 32-bit legacy system call entry.
463 *
464 * 32-bit x86 Linux system calls traditionally used the INT $0x80
465 * instruction.  INT $0x80 lands here.
466 *
467 * This entry point can be used by any 32-bit perform system calls.
468 * Instances of INT $0x80 can be found inline in various programs and
469 * libraries.  It is also used by the vDSO's __kernel_vsyscall
470 * fallback for hardware that doesn't support a faster entry method.
471 * Restarted 32-bit system calls also fall back to INT $0x80
472 * regardless of what instruction was originally used to do the system
473 * call.  (64-bit programs can use INT $0x80 as well, but they can
474 * only run on 64-bit kernels and therefore land in
475 * entry_INT80_compat.)
476 *
477 * This is considered a slow path.  It is not used by most libc
478 * implementations on modern hardware except during process startup.
479 *
480 * Arguments:
481 * eax  system call number
482 * ebx  arg1
483 * ecx  arg2
484 * edx  arg3
485 * esi  arg4
486 * edi  arg5
487 * ebp  arg6
488 */
489ENTRY(entry_INT80_32)
490	ASM_CLAC
491	pushl	%eax			/* pt_regs->orig_ax */
492	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
493
494	/*
495	 * User mode is traced as though IRQs are on, and the interrupt gate
496	 * turned them off.
497	 */
498	TRACE_IRQS_OFF
499
500	movl	%esp, %eax
501	call	do_int80_syscall_32
502.Lsyscall_32_done:
503
504restore_all:
505	TRACE_IRQS_IRET
506restore_all_notrace:
507#ifdef CONFIG_X86_ESPFIX32
508	ALTERNATIVE	"jmp restore_nocheck", "", X86_BUG_ESPFIX
509
510	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
511	/*
512	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
513	 * are returning to the kernel.
514	 * See comments in process.c:copy_thread() for details.
515	 */
516	movb	PT_OLDSS(%esp), %ah
517	movb	PT_CS(%esp), %al
518	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
519	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
520	je ldt_ss				# returning to user-space with LDT SS
521#endif
522restore_nocheck:
523	RESTORE_REGS 4				# skip orig_eax/error_code
524irq_return:
525	INTERRUPT_RETURN
526.section .fixup, "ax"
527ENTRY(iret_exc	)
528	pushl	$0				# no error code
529	pushl	$do_iret_error
530	jmp	error_code
531.previous
532	_ASM_EXTABLE(irq_return, iret_exc)
533
534#ifdef CONFIG_X86_ESPFIX32
535ldt_ss:
536/*
537 * Setup and switch to ESPFIX stack
538 *
539 * We're returning to userspace with a 16 bit stack. The CPU will not
540 * restore the high word of ESP for us on executing iret... This is an
541 * "official" bug of all the x86-compatible CPUs, which we can work
542 * around to make dosemu and wine happy. We do this by preloading the
543 * high word of ESP with the high word of the userspace ESP while
544 * compensating for the offset by changing to the ESPFIX segment with
545 * a base address that matches for the difference.
546 */
547#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
548	mov	%esp, %edx			/* load kernel esp */
549	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
550	mov	%dx, %ax			/* eax: new kernel esp */
551	sub	%eax, %edx			/* offset (low word is 0) */
552	shr	$16, %edx
553	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
554	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
555	pushl	$__ESPFIX_SS
556	pushl	%eax				/* new kernel esp */
557	/*
558	 * Disable interrupts, but do not irqtrace this section: we
559	 * will soon execute iret and the tracer was already set to
560	 * the irqstate after the IRET:
561	 */
562	DISABLE_INTERRUPTS(CLBR_EAX)
563	lss	(%esp), %esp			/* switch to espfix segment */
564	jmp	restore_nocheck
565#endif
566ENDPROC(entry_INT80_32)
567
568.macro FIXUP_ESPFIX_STACK
569/*
570 * Switch back for ESPFIX stack to the normal zerobased stack
571 *
572 * We can't call C functions using the ESPFIX stack. This code reads
573 * the high word of the segment base from the GDT and swiches to the
574 * normal stack and adjusts ESP with the matching offset.
575 */
576#ifdef CONFIG_X86_ESPFIX32
577	/* fixup the stack */
578	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
579	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
580	shl	$16, %eax
581	addl	%esp, %eax			/* the adjusted stack pointer */
582	pushl	$__KERNEL_DS
583	pushl	%eax
584	lss	(%esp), %esp			/* switch to the normal stack segment */
585#endif
586.endm
587.macro UNWIND_ESPFIX_STACK
588#ifdef CONFIG_X86_ESPFIX32
589	movl	%ss, %eax
590	/* see if on espfix stack */
591	cmpw	$__ESPFIX_SS, %ax
592	jne	27f
593	movl	$__KERNEL_DS, %eax
594	movl	%eax, %ds
595	movl	%eax, %es
596	/* switch to normal stack */
597	FIXUP_ESPFIX_STACK
59827:
599#endif
600.endm
601
602/*
603 * Build the entry stubs with some assembler magic.
604 * We pack 1 stub into every 8-byte block.
605 */
606	.align 8
607ENTRY(irq_entries_start)
608    vector=FIRST_EXTERNAL_VECTOR
609    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
610	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
611    vector=vector+1
612	jmp	common_interrupt
613	.align	8
614    .endr
615END(irq_entries_start)
616
617/*
618 * the CPU automatically disables interrupts when executing an IRQ vector,
619 * so IRQ-flags tracing has to follow that:
620 */
621	.p2align CONFIG_X86_L1_CACHE_SHIFT
622common_interrupt:
623	ASM_CLAC
624	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
625	SAVE_ALL
626	TRACE_IRQS_OFF
627	movl	%esp, %eax
628	call	do_IRQ
629	jmp	ret_from_intr
630ENDPROC(common_interrupt)
631
632#define BUILD_INTERRUPT3(name, nr, fn)	\
633ENTRY(name)				\
634	ASM_CLAC;			\
635	pushl	$~(nr);			\
636	SAVE_ALL;			\
637	TRACE_IRQS_OFF			\
638	movl	%esp, %eax;		\
639	call	fn;			\
640	jmp	ret_from_intr;		\
641ENDPROC(name)
642
643
644#ifdef CONFIG_TRACING
645# define TRACE_BUILD_INTERRUPT(name, nr)	BUILD_INTERRUPT3(trace_##name, nr, smp_trace_##name)
646#else
647# define TRACE_BUILD_INTERRUPT(name, nr)
648#endif
649
650#define BUILD_INTERRUPT(name, nr)		\
651	BUILD_INTERRUPT3(name, nr, smp_##name);	\
652	TRACE_BUILD_INTERRUPT(name, nr)
653
654/* The include is where all of the SMP etc. interrupts come from */
655#include <asm/entry_arch.h>
656
657ENTRY(coprocessor_error)
658	ASM_CLAC
659	pushl	$0
660	pushl	$do_coprocessor_error
661	jmp	error_code
662END(coprocessor_error)
663
664ENTRY(simd_coprocessor_error)
665	ASM_CLAC
666	pushl	$0
667#ifdef CONFIG_X86_INVD_BUG
668	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
669	ALTERNATIVE "pushl	$do_general_protection",	\
670		    "pushl	$do_simd_coprocessor_error",	\
671		    X86_FEATURE_XMM
672#else
673	pushl	$do_simd_coprocessor_error
674#endif
675	jmp	error_code
676END(simd_coprocessor_error)
677
678ENTRY(device_not_available)
679	ASM_CLAC
680	pushl	$-1				# mark this as an int
681	pushl	$do_device_not_available
682	jmp	error_code
683END(device_not_available)
684
685#ifdef CONFIG_PARAVIRT
686ENTRY(native_iret)
687	iret
688	_ASM_EXTABLE(native_iret, iret_exc)
689END(native_iret)
690#endif
691
692ENTRY(overflow)
693	ASM_CLAC
694	pushl	$0
695	pushl	$do_overflow
696	jmp	error_code
697END(overflow)
698
699ENTRY(bounds)
700	ASM_CLAC
701	pushl	$0
702	pushl	$do_bounds
703	jmp	error_code
704END(bounds)
705
706ENTRY(invalid_op)
707	ASM_CLAC
708	pushl	$0
709	pushl	$do_invalid_op
710	jmp	error_code
711END(invalid_op)
712
713ENTRY(coprocessor_segment_overrun)
714	ASM_CLAC
715	pushl	$0
716	pushl	$do_coprocessor_segment_overrun
717	jmp	error_code
718END(coprocessor_segment_overrun)
719
720ENTRY(invalid_TSS)
721	ASM_CLAC
722	pushl	$do_invalid_TSS
723	jmp	error_code
724END(invalid_TSS)
725
726ENTRY(segment_not_present)
727	ASM_CLAC
728	pushl	$do_segment_not_present
729	jmp	error_code
730END(segment_not_present)
731
732ENTRY(stack_segment)
733	ASM_CLAC
734	pushl	$do_stack_segment
735	jmp	error_code
736END(stack_segment)
737
738ENTRY(alignment_check)
739	ASM_CLAC
740	pushl	$do_alignment_check
741	jmp	error_code
742END(alignment_check)
743
744ENTRY(divide_error)
745	ASM_CLAC
746	pushl	$0				# no error code
747	pushl	$do_divide_error
748	jmp	error_code
749END(divide_error)
750
751#ifdef CONFIG_X86_MCE
752ENTRY(machine_check)
753	ASM_CLAC
754	pushl	$0
755	pushl	machine_check_vector
756	jmp	error_code
757END(machine_check)
758#endif
759
760ENTRY(spurious_interrupt_bug)
761	ASM_CLAC
762	pushl	$0
763	pushl	$do_spurious_interrupt_bug
764	jmp	error_code
765END(spurious_interrupt_bug)
766
767#ifdef CONFIG_XEN
768ENTRY(xen_hypervisor_callback)
769	pushl	$-1				/* orig_ax = -1 => not a system call */
770	SAVE_ALL
771	TRACE_IRQS_OFF
772
773	/*
774	 * Check to see if we got the event in the critical
775	 * region in xen_iret_direct, after we've reenabled
776	 * events and checked for pending events.  This simulates
777	 * iret instruction's behaviour where it delivers a
778	 * pending interrupt when enabling interrupts:
779	 */
780	movl	PT_EIP(%esp), %eax
781	cmpl	$xen_iret_start_crit, %eax
782	jb	1f
783	cmpl	$xen_iret_end_crit, %eax
784	jae	1f
785
786	jmp	xen_iret_crit_fixup
787
788ENTRY(xen_do_upcall)
7891:	mov	%esp, %eax
790	call	xen_evtchn_do_upcall
791#ifndef CONFIG_PREEMPT
792	call	xen_maybe_preempt_hcall
793#endif
794	jmp	ret_from_intr
795ENDPROC(xen_hypervisor_callback)
796
797/*
798 * Hypervisor uses this for application faults while it executes.
799 * We get here for two reasons:
800 *  1. Fault while reloading DS, ES, FS or GS
801 *  2. Fault while executing IRET
802 * Category 1 we fix up by reattempting the load, and zeroing the segment
803 * register if the load fails.
804 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
805 * normal Linux return path in this case because if we use the IRET hypercall
806 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
807 * We distinguish between categories by maintaining a status value in EAX.
808 */
809ENTRY(xen_failsafe_callback)
810	pushl	%eax
811	movl	$1, %eax
8121:	mov	4(%esp), %ds
8132:	mov	8(%esp), %es
8143:	mov	12(%esp), %fs
8154:	mov	16(%esp), %gs
816	/* EAX == 0 => Category 1 (Bad segment)
817	   EAX != 0 => Category 2 (Bad IRET) */
818	testl	%eax, %eax
819	popl	%eax
820	lea	16(%esp), %esp
821	jz	5f
822	jmp	iret_exc
8235:	pushl	$-1				/* orig_ax = -1 => not a system call */
824	SAVE_ALL
825	jmp	ret_from_exception
826
827.section .fixup, "ax"
8286:	xorl	%eax, %eax
829	movl	%eax, 4(%esp)
830	jmp	1b
8317:	xorl	%eax, %eax
832	movl	%eax, 8(%esp)
833	jmp	2b
8348:	xorl	%eax, %eax
835	movl	%eax, 12(%esp)
836	jmp	3b
8379:	xorl	%eax, %eax
838	movl	%eax, 16(%esp)
839	jmp	4b
840.previous
841	_ASM_EXTABLE(1b, 6b)
842	_ASM_EXTABLE(2b, 7b)
843	_ASM_EXTABLE(3b, 8b)
844	_ASM_EXTABLE(4b, 9b)
845ENDPROC(xen_failsafe_callback)
846
847BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
848		xen_evtchn_do_upcall)
849
850#endif /* CONFIG_XEN */
851
852#if IS_ENABLED(CONFIG_HYPERV)
853
854BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
855	hyperv_vector_handler)
856
857#endif /* CONFIG_HYPERV */
858
859#ifdef CONFIG_FUNCTION_TRACER
860#ifdef CONFIG_DYNAMIC_FTRACE
861
862ENTRY(mcount)
863	ret
864END(mcount)
865
866ENTRY(ftrace_caller)
867	pushl	%eax
868	pushl	%ecx
869	pushl	%edx
870	pushl	$0				/* Pass NULL as regs pointer */
871	movl	4*4(%esp), %eax
872	movl	0x4(%ebp), %edx
873	movl	function_trace_op, %ecx
874	subl	$MCOUNT_INSN_SIZE, %eax
875
876.globl ftrace_call
877ftrace_call:
878	call	ftrace_stub
879
880	addl	$4, %esp			/* skip NULL pointer */
881	popl	%edx
882	popl	%ecx
883	popl	%eax
884ftrace_ret:
885#ifdef CONFIG_FUNCTION_GRAPH_TRACER
886.globl ftrace_graph_call
887ftrace_graph_call:
888	jmp	ftrace_stub
889#endif
890
891.globl ftrace_stub
892ftrace_stub:
893	ret
894END(ftrace_caller)
895
896ENTRY(ftrace_regs_caller)
897	pushf	/* push flags before compare (in cs location) */
898
899	/*
900	 * i386 does not save SS and ESP when coming from kernel.
901	 * Instead, to get sp, &regs->sp is used (see ptrace.h).
902	 * Unfortunately, that means eflags must be at the same location
903	 * as the current return ip is. We move the return ip into the
904	 * ip location, and move flags into the return ip location.
905	 */
906	pushl	4(%esp)				/* save return ip into ip slot */
907
908	pushl	$0				/* Load 0 into orig_ax */
909	pushl	%gs
910	pushl	%fs
911	pushl	%es
912	pushl	%ds
913	pushl	%eax
914	pushl	%ebp
915	pushl	%edi
916	pushl	%esi
917	pushl	%edx
918	pushl	%ecx
919	pushl	%ebx
920
921	movl	13*4(%esp), %eax		/* Get the saved flags */
922	movl	%eax, 14*4(%esp)		/* Move saved flags into regs->flags location */
923						/* clobbering return ip */
924	movl	$__KERNEL_CS, 13*4(%esp)
925
926	movl	12*4(%esp), %eax		/* Load ip (1st parameter) */
927	subl	$MCOUNT_INSN_SIZE, %eax		/* Adjust ip */
928	movl	0x4(%ebp), %edx			/* Load parent ip (2nd parameter) */
929	movl	function_trace_op, %ecx		/* Save ftrace_pos in 3rd parameter */
930	pushl	%esp				/* Save pt_regs as 4th parameter */
931
932GLOBAL(ftrace_regs_call)
933	call	ftrace_stub
934
935	addl	$4, %esp			/* Skip pt_regs */
936	movl	14*4(%esp), %eax		/* Move flags back into cs */
937	movl	%eax, 13*4(%esp)		/* Needed to keep addl	from modifying flags */
938	movl	12*4(%esp), %eax		/* Get return ip from regs->ip */
939	movl	%eax, 14*4(%esp)		/* Put return ip back for ret */
940
941	popl	%ebx
942	popl	%ecx
943	popl	%edx
944	popl	%esi
945	popl	%edi
946	popl	%ebp
947	popl	%eax
948	popl	%ds
949	popl	%es
950	popl	%fs
951	popl	%gs
952	addl	$8, %esp			/* Skip orig_ax and ip */
953	popf					/* Pop flags at end (no addl to corrupt flags) */
954	jmp	ftrace_ret
955
956	popf
957	jmp	ftrace_stub
958#else /* ! CONFIG_DYNAMIC_FTRACE */
959
960ENTRY(mcount)
961	cmpl	$__PAGE_OFFSET, %esp
962	jb	ftrace_stub			/* Paging not enabled yet? */
963
964	cmpl	$ftrace_stub, ftrace_trace_function
965	jnz	trace
966#ifdef CONFIG_FUNCTION_GRAPH_TRACER
967	cmpl	$ftrace_stub, ftrace_graph_return
968	jnz	ftrace_graph_caller
969
970	cmpl	$ftrace_graph_entry_stub, ftrace_graph_entry
971	jnz	ftrace_graph_caller
972#endif
973.globl ftrace_stub
974ftrace_stub:
975	ret
976
977	/* taken from glibc */
978trace:
979	pushl	%eax
980	pushl	%ecx
981	pushl	%edx
982	movl	0xc(%esp), %eax
983	movl	0x4(%ebp), %edx
984	subl	$MCOUNT_INSN_SIZE, %eax
985
986	call	*ftrace_trace_function
987
988	popl	%edx
989	popl	%ecx
990	popl	%eax
991	jmp	ftrace_stub
992END(mcount)
993#endif /* CONFIG_DYNAMIC_FTRACE */
994#endif /* CONFIG_FUNCTION_TRACER */
995
996#ifdef CONFIG_FUNCTION_GRAPH_TRACER
997ENTRY(ftrace_graph_caller)
998	pushl	%eax
999	pushl	%ecx
1000	pushl	%edx
1001	movl	0xc(%esp), %eax
1002	lea	0x4(%ebp), %edx
1003	movl	(%ebp), %ecx
1004	subl	$MCOUNT_INSN_SIZE, %eax
1005	call	prepare_ftrace_return
1006	popl	%edx
1007	popl	%ecx
1008	popl	%eax
1009	ret
1010END(ftrace_graph_caller)
1011
1012.globl return_to_handler
1013return_to_handler:
1014	pushl	%eax
1015	pushl	%edx
1016	movl	%ebp, %eax
1017	call	ftrace_return_to_handler
1018	movl	%eax, %ecx
1019	popl	%edx
1020	popl	%eax
1021	jmp	*%ecx
1022#endif
1023
1024#ifdef CONFIG_TRACING
1025ENTRY(trace_page_fault)
1026	ASM_CLAC
1027	pushl	$trace_do_page_fault
1028	jmp	error_code
1029END(trace_page_fault)
1030#endif
1031
1032ENTRY(page_fault)
1033	ASM_CLAC
1034	pushl	$do_page_fault
1035	ALIGN
1036error_code:
1037	/* the function address is in %gs's slot on the stack */
1038	pushl	%fs
1039	pushl	%es
1040	pushl	%ds
1041	pushl	%eax
1042	pushl	%ebp
1043	pushl	%edi
1044	pushl	%esi
1045	pushl	%edx
1046	pushl	%ecx
1047	pushl	%ebx
1048	cld
1049	movl	$(__KERNEL_PERCPU), %ecx
1050	movl	%ecx, %fs
1051	UNWIND_ESPFIX_STACK
1052	GS_TO_REG %ecx
1053	movl	PT_GS(%esp), %edi		# get the function address
1054	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
1055	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
1056	REG_TO_PTGS %ecx
1057	SET_KERNEL_GS %ecx
1058	movl	$(__USER_DS), %ecx
1059	movl	%ecx, %ds
1060	movl	%ecx, %es
1061	TRACE_IRQS_OFF
1062	movl	%esp, %eax			# pt_regs pointer
1063	call	*%edi
1064	jmp	ret_from_exception
1065END(page_fault)
1066
1067ENTRY(debug)
1068	/*
1069	 * #DB can happen at the first instruction of
1070	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
1071	 * happens, then we will be running on a very small stack.  We
1072	 * need to detect this condition and switch to the thread
1073	 * stack before calling any C code at all.
1074	 *
1075	 * If you edit this code, keep in mind that NMIs can happen in here.
1076	 */
1077	ASM_CLAC
1078	pushl	$-1				# mark this as an int
1079	SAVE_ALL
1080	xorl	%edx, %edx			# error code 0
1081	movl	%esp, %eax			# pt_regs pointer
1082
1083	/* Are we currently on the SYSENTER stack? */
1084	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1085	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
1086	cmpl	$SIZEOF_SYSENTER_stack, %ecx
1087	jb	.Ldebug_from_sysenter_stack
1088
1089	TRACE_IRQS_OFF
1090	call	do_debug
1091	jmp	ret_from_exception
1092
1093.Ldebug_from_sysenter_stack:
1094	/* We're on the SYSENTER stack.  Switch off. */
1095	movl	%esp, %ebp
1096	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1097	TRACE_IRQS_OFF
1098	call	do_debug
1099	movl	%ebp, %esp
1100	jmp	ret_from_exception
1101END(debug)
1102
1103/*
1104 * NMI is doubly nasty.  It can happen on the first instruction of
1105 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1106 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1107 * switched stacks.  We handle both conditions by simply checking whether we
1108 * interrupted kernel code running on the SYSENTER stack.
1109 */
1110ENTRY(nmi)
1111	ASM_CLAC
1112#ifdef CONFIG_X86_ESPFIX32
1113	pushl	%eax
1114	movl	%ss, %eax
1115	cmpw	$__ESPFIX_SS, %ax
1116	popl	%eax
1117	je	nmi_espfix_stack
1118#endif
1119
1120	pushl	%eax				# pt_regs->orig_ax
1121	SAVE_ALL
1122	xorl	%edx, %edx			# zero error code
1123	movl	%esp, %eax			# pt_regs pointer
1124
1125	/* Are we currently on the SYSENTER stack? */
1126	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
1127	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
1128	cmpl	$SIZEOF_SYSENTER_stack, %ecx
1129	jb	.Lnmi_from_sysenter_stack
1130
1131	/* Not on SYSENTER stack. */
1132	call	do_nmi
1133	jmp	restore_all_notrace
1134
1135.Lnmi_from_sysenter_stack:
1136	/*
1137	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1138	 * is using the thread stack right now, so it's safe for us to use it.
1139	 */
1140	movl	%esp, %ebp
1141	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1142	call	do_nmi
1143	movl	%ebp, %esp
1144	jmp	restore_all_notrace
1145
1146#ifdef CONFIG_X86_ESPFIX32
1147nmi_espfix_stack:
1148	/*
1149	 * create the pointer to lss back
1150	 */
1151	pushl	%ss
1152	pushl	%esp
1153	addl	$4, (%esp)
1154	/* copy the iret frame of 12 bytes */
1155	.rept 3
1156	pushl	16(%esp)
1157	.endr
1158	pushl	%eax
1159	SAVE_ALL
1160	FIXUP_ESPFIX_STACK			# %eax == %esp
1161	xorl	%edx, %edx			# zero error code
1162	call	do_nmi
1163	RESTORE_REGS
1164	lss	12+4(%esp), %esp		# back to espfix stack
1165	jmp	irq_return
1166#endif
1167END(nmi)
1168
1169ENTRY(int3)
1170	ASM_CLAC
1171	pushl	$-1				# mark this as an int
1172	SAVE_ALL
1173	TRACE_IRQS_OFF
1174	xorl	%edx, %edx			# zero error code
1175	movl	%esp, %eax			# pt_regs pointer
1176	call	do_int3
1177	jmp	ret_from_exception
1178END(int3)
1179
1180ENTRY(general_protection)
1181	pushl	$do_general_protection
1182	jmp	error_code
1183END(general_protection)
1184
1185#ifdef CONFIG_KVM_GUEST
1186ENTRY(async_page_fault)
1187	ASM_CLAC
1188	pushl	$do_async_page_fault
1189	jmp	error_code
1190END(async_page_fault)
1191#endif
1192
1193ENTRY(rewind_stack_do_exit)
1194	/* Prevent any naive code from trying to unwind to our caller. */
1195	xorl	%ebp, %ebp
1196
1197	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1198	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1199
1200	call	do_exit
12011:	jmp 1b
1202END(rewind_stack_do_exit)
1203