xref: /openbmc/linux/arch/x86/entry/entry_32.S (revision 4cff79e9)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 *  Copyright (C) 1991,1992  Linus Torvalds
4 *
5 * entry_32.S contains the system-call and low-level fault and trap handling routines.
6 *
7 * Stack layout while running C code:
8 *	ptrace needs to have all registers on the stack.
9 *	If the order here is changed, it needs to be
10 *	updated in fork.c:copy_process(), signal.c:do_signal(),
11 *	ptrace.c and ptrace.h
12 *
13 *	 0(%esp) - %ebx
14 *	 4(%esp) - %ecx
15 *	 8(%esp) - %edx
16 *	 C(%esp) - %esi
17 *	10(%esp) - %edi
18 *	14(%esp) - %ebp
19 *	18(%esp) - %eax
20 *	1C(%esp) - %ds
21 *	20(%esp) - %es
22 *	24(%esp) - %fs
23 *	28(%esp) - %gs		saved iff !CONFIG_X86_32_LAZY_GS
24 *	2C(%esp) - orig_eax
25 *	30(%esp) - %eip
26 *	34(%esp) - %cs
27 *	38(%esp) - %eflags
28 *	3C(%esp) - %oldesp
29 *	40(%esp) - %oldss
30 */
31
32#include <linux/linkage.h>
33#include <linux/err.h>
34#include <asm/thread_info.h>
35#include <asm/irqflags.h>
36#include <asm/errno.h>
37#include <asm/segment.h>
38#include <asm/smp.h>
39#include <asm/percpu.h>
40#include <asm/processor-flags.h>
41#include <asm/irq_vectors.h>
42#include <asm/cpufeatures.h>
43#include <asm/alternative-asm.h>
44#include <asm/asm.h>
45#include <asm/smap.h>
46#include <asm/frame.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/*
179 * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
180 * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
181 * is just clearing the MSB, which makes it an invalid stack address and is also
182 * a signal to the unwinder that it's a pt_regs pointer in disguise.
183 *
184 * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
185 * original rbp.
186 */
187.macro ENCODE_FRAME_POINTER
188#ifdef CONFIG_FRAME_POINTER
189	mov %esp, %ebp
190	andl $0x7fffffff, %ebp
191#endif
192.endm
193
194.macro RESTORE_INT_REGS
195	popl	%ebx
196	popl	%ecx
197	popl	%edx
198	popl	%esi
199	popl	%edi
200	popl	%ebp
201	popl	%eax
202.endm
203
204.macro RESTORE_REGS pop=0
205	RESTORE_INT_REGS
2061:	popl	%ds
2072:	popl	%es
2083:	popl	%fs
209	POP_GS \pop
210.pushsection .fixup, "ax"
2114:	movl	$0, (%esp)
212	jmp	1b
2135:	movl	$0, (%esp)
214	jmp	2b
2156:	movl	$0, (%esp)
216	jmp	3b
217.popsection
218	_ASM_EXTABLE(1b, 4b)
219	_ASM_EXTABLE(2b, 5b)
220	_ASM_EXTABLE(3b, 6b)
221	POP_GS_EX
222.endm
223
224/*
225 * %eax: prev task
226 * %edx: next task
227 */
228ENTRY(__switch_to_asm)
229	/*
230	 * Save callee-saved registers
231	 * This must match the order in struct inactive_task_frame
232	 */
233	pushl	%ebp
234	pushl	%ebx
235	pushl	%edi
236	pushl	%esi
237
238	/* switch stack */
239	movl	%esp, TASK_threadsp(%eax)
240	movl	TASK_threadsp(%edx), %esp
241
242#ifdef CONFIG_CC_STACKPROTECTOR
243	movl	TASK_stack_canary(%edx), %ebx
244	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
245#endif
246
247#ifdef CONFIG_RETPOLINE
248	/*
249	 * When switching from a shallower to a deeper call stack
250	 * the RSB may either underflow or use entries populated
251	 * with userspace addresses. On CPUs where those concerns
252	 * exist, overwrite the RSB with entries which capture
253	 * speculative execution to prevent attack.
254	 */
255	FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
256#endif
257
258	/* restore callee-saved registers */
259	popl	%esi
260	popl	%edi
261	popl	%ebx
262	popl	%ebp
263
264	jmp	__switch_to
265END(__switch_to_asm)
266
267/*
268 * The unwinder expects the last frame on the stack to always be at the same
269 * offset from the end of the page, which allows it to validate the stack.
270 * Calling schedule_tail() directly would break that convention because its an
271 * asmlinkage function so its argument has to be pushed on the stack.  This
272 * wrapper creates a proper "end of stack" frame header before the call.
273 */
274ENTRY(schedule_tail_wrapper)
275	FRAME_BEGIN
276
277	pushl	%eax
278	call	schedule_tail
279	popl	%eax
280
281	FRAME_END
282	ret
283ENDPROC(schedule_tail_wrapper)
284/*
285 * A newly forked process directly context switches into this address.
286 *
287 * eax: prev task we switched from
288 * ebx: kernel thread func (NULL for user thread)
289 * edi: kernel thread arg
290 */
291ENTRY(ret_from_fork)
292	call	schedule_tail_wrapper
293
294	testl	%ebx, %ebx
295	jnz	1f		/* kernel threads are uncommon */
296
2972:
298	/* When we fork, we trace the syscall return in the child, too. */
299	movl    %esp, %eax
300	call    syscall_return_slowpath
301	jmp     restore_all
302
303	/* kernel thread */
3041:	movl	%edi, %eax
305	CALL_NOSPEC %ebx
306	/*
307	 * A kernel thread is allowed to return here after successfully
308	 * calling do_execve().  Exit to userspace to complete the execve()
309	 * syscall.
310	 */
311	movl	$0, PT_EAX(%esp)
312	jmp	2b
313END(ret_from_fork)
314
315/*
316 * Return to user mode is not as complex as all this looks,
317 * but we want the default path for a system call return to
318 * go as quickly as possible which is why some of this is
319 * less clear than it otherwise should be.
320 */
321
322	# userspace resumption stub bypassing syscall exit tracing
323	ALIGN
324ret_from_exception:
325	preempt_stop(CLBR_ANY)
326ret_from_intr:
327#ifdef CONFIG_VM86
328	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
329	movb	PT_CS(%esp), %al
330	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
331#else
332	/*
333	 * We can be coming here from child spawned by kernel_thread().
334	 */
335	movl	PT_CS(%esp), %eax
336	andl	$SEGMENT_RPL_MASK, %eax
337#endif
338	cmpl	$USER_RPL, %eax
339	jb	resume_kernel			# not returning to v8086 or userspace
340
341ENTRY(resume_userspace)
342	DISABLE_INTERRUPTS(CLBR_ANY)
343	TRACE_IRQS_OFF
344	movl	%esp, %eax
345	call	prepare_exit_to_usermode
346	jmp	restore_all
347END(ret_from_exception)
348
349#ifdef CONFIG_PREEMPT
350ENTRY(resume_kernel)
351	DISABLE_INTERRUPTS(CLBR_ANY)
352.Lneed_resched:
353	cmpl	$0, PER_CPU_VAR(__preempt_count)
354	jnz	restore_all
355	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
356	jz	restore_all
357	call	preempt_schedule_irq
358	jmp	.Lneed_resched
359END(resume_kernel)
360#endif
361
362GLOBAL(__begin_SYSENTER_singlestep_region)
363/*
364 * All code from here through __end_SYSENTER_singlestep_region is subject
365 * to being single-stepped if a user program sets TF and executes SYSENTER.
366 * There is absolutely nothing that we can do to prevent this from happening
367 * (thanks Intel!).  To keep our handling of this situation as simple as
368 * possible, we handle TF just like AC and NT, except that our #DB handler
369 * will ignore all of the single-step traps generated in this range.
370 */
371
372#ifdef CONFIG_XEN
373/*
374 * Xen doesn't set %esp to be precisely what the normal SYSENTER
375 * entry point expects, so fix it up before using the normal path.
376 */
377ENTRY(xen_sysenter_target)
378	addl	$5*4, %esp			/* remove xen-provided frame */
379	jmp	.Lsysenter_past_esp
380#endif
381
382/*
383 * 32-bit SYSENTER entry.
384 *
385 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
386 * if X86_FEATURE_SEP is available.  This is the preferred system call
387 * entry on 32-bit systems.
388 *
389 * The SYSENTER instruction, in principle, should *only* occur in the
390 * vDSO.  In practice, a small number of Android devices were shipped
391 * with a copy of Bionic that inlined a SYSENTER instruction.  This
392 * never happened in any of Google's Bionic versions -- it only happened
393 * in a narrow range of Intel-provided versions.
394 *
395 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
396 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
397 * SYSENTER does not save anything on the stack,
398 * and does not save old EIP (!!!), ESP, or EFLAGS.
399 *
400 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
401 * user and/or vm86 state), we explicitly disable the SYSENTER
402 * instruction in vm86 mode by reprogramming the MSRs.
403 *
404 * Arguments:
405 * eax  system call number
406 * ebx  arg1
407 * ecx  arg2
408 * edx  arg3
409 * esi  arg4
410 * edi  arg5
411 * ebp  user stack
412 * 0(%ebp) arg6
413 */
414ENTRY(entry_SYSENTER_32)
415	movl	TSS_sysenter_sp0(%esp), %esp
416.Lsysenter_past_esp:
417	pushl	$__USER_DS		/* pt_regs->ss */
418	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
419	pushfl				/* pt_regs->flags (except IF = 0) */
420	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
421	pushl	$__USER_CS		/* pt_regs->cs */
422	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
423	pushl	%eax			/* pt_regs->orig_ax */
424	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
425
426	/*
427	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
428	 * and TF ourselves.  To save a few cycles, we can check whether
429	 * either was set instead of doing an unconditional popfq.
430	 * This needs to happen before enabling interrupts so that
431	 * we don't get preempted with NT set.
432	 *
433	 * If TF is set, we will single-step all the way to here -- do_debug
434	 * will ignore all the traps.  (Yes, this is slow, but so is
435	 * single-stepping in general.  This allows us to avoid having
436	 * a more complicated code to handle the case where a user program
437	 * forces us to single-step through the SYSENTER entry code.)
438	 *
439	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
440	 * out-of-line as an optimization: NT is unlikely to be set in the
441	 * majority of the cases and instead of polluting the I$ unnecessarily,
442	 * we're keeping that code behind a branch which will predict as
443	 * not-taken and therefore its instructions won't be fetched.
444	 */
445	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
446	jnz	.Lsysenter_fix_flags
447.Lsysenter_flags_fixed:
448
449	/*
450	 * User mode is traced as though IRQs are on, and SYSENTER
451	 * turned them off.
452	 */
453	TRACE_IRQS_OFF
454
455	movl	%esp, %eax
456	call	do_fast_syscall_32
457	/* XEN PV guests always use IRET path */
458	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
459		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
460
461/* Opportunistic SYSEXIT */
462	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
463	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
464	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
4651:	mov	PT_FS(%esp), %fs
466	PTGS_TO_GS
467	popl	%ebx			/* pt_regs->bx */
468	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
469	popl	%esi			/* pt_regs->si */
470	popl	%edi			/* pt_regs->di */
471	popl	%ebp			/* pt_regs->bp */
472	popl	%eax			/* pt_regs->ax */
473
474	/*
475	 * Restore all flags except IF. (We restore IF separately because
476	 * STI gives a one-instruction window in which we won't be interrupted,
477	 * whereas POPF does not.)
478	 */
479	addl	$PT_EFLAGS-PT_DS, %esp	/* point esp at pt_regs->flags */
480	btr	$X86_EFLAGS_IF_BIT, (%esp)
481	popfl
482
483	/*
484	 * Return back to the vDSO, which will pop ecx and edx.
485	 * Don't bother with DS and ES (they already contain __USER_DS).
486	 */
487	sti
488	sysexit
489
490.pushsection .fixup, "ax"
4912:	movl	$0, PT_FS(%esp)
492	jmp	1b
493.popsection
494	_ASM_EXTABLE(1b, 2b)
495	PTGS_TO_GS_EX
496
497.Lsysenter_fix_flags:
498	pushl	$X86_EFLAGS_FIXED
499	popfl
500	jmp	.Lsysenter_flags_fixed
501GLOBAL(__end_SYSENTER_singlestep_region)
502ENDPROC(entry_SYSENTER_32)
503
504/*
505 * 32-bit legacy system call entry.
506 *
507 * 32-bit x86 Linux system calls traditionally used the INT $0x80
508 * instruction.  INT $0x80 lands here.
509 *
510 * This entry point can be used by any 32-bit perform system calls.
511 * Instances of INT $0x80 can be found inline in various programs and
512 * libraries.  It is also used by the vDSO's __kernel_vsyscall
513 * fallback for hardware that doesn't support a faster entry method.
514 * Restarted 32-bit system calls also fall back to INT $0x80
515 * regardless of what instruction was originally used to do the system
516 * call.  (64-bit programs can use INT $0x80 as well, but they can
517 * only run on 64-bit kernels and therefore land in
518 * entry_INT80_compat.)
519 *
520 * This is considered a slow path.  It is not used by most libc
521 * implementations on modern hardware except during process startup.
522 *
523 * Arguments:
524 * eax  system call number
525 * ebx  arg1
526 * ecx  arg2
527 * edx  arg3
528 * esi  arg4
529 * edi  arg5
530 * ebp  arg6
531 */
532ENTRY(entry_INT80_32)
533	ASM_CLAC
534	pushl	%eax			/* pt_regs->orig_ax */
535	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
536
537	/*
538	 * User mode is traced as though IRQs are on, and the interrupt gate
539	 * turned them off.
540	 */
541	TRACE_IRQS_OFF
542
543	movl	%esp, %eax
544	call	do_int80_syscall_32
545.Lsyscall_32_done:
546
547restore_all:
548	TRACE_IRQS_IRET
549.Lrestore_all_notrace:
550#ifdef CONFIG_X86_ESPFIX32
551	ALTERNATIVE	"jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
552
553	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
554	/*
555	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
556	 * are returning to the kernel.
557	 * See comments in process.c:copy_thread() for details.
558	 */
559	movb	PT_OLDSS(%esp), %ah
560	movb	PT_CS(%esp), %al
561	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
562	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
563	je .Lldt_ss				# returning to user-space with LDT SS
564#endif
565.Lrestore_nocheck:
566	RESTORE_REGS 4				# skip orig_eax/error_code
567.Lirq_return:
568	/*
569	 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
570	 * when returning from IPI handler and when returning from
571	 * scheduler to user-space.
572	 */
573	INTERRUPT_RETURN
574
575.section .fixup, "ax"
576ENTRY(iret_exc	)
577	pushl	$0				# no error code
578	pushl	$do_iret_error
579	jmp	common_exception
580.previous
581	_ASM_EXTABLE(.Lirq_return, iret_exc)
582
583#ifdef CONFIG_X86_ESPFIX32
584.Lldt_ss:
585/*
586 * Setup and switch to ESPFIX stack
587 *
588 * We're returning to userspace with a 16 bit stack. The CPU will not
589 * restore the high word of ESP for us on executing iret... This is an
590 * "official" bug of all the x86-compatible CPUs, which we can work
591 * around to make dosemu and wine happy. We do this by preloading the
592 * high word of ESP with the high word of the userspace ESP while
593 * compensating for the offset by changing to the ESPFIX segment with
594 * a base address that matches for the difference.
595 */
596#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
597	mov	%esp, %edx			/* load kernel esp */
598	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
599	mov	%dx, %ax			/* eax: new kernel esp */
600	sub	%eax, %edx			/* offset (low word is 0) */
601	shr	$16, %edx
602	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
603	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
604	pushl	$__ESPFIX_SS
605	pushl	%eax				/* new kernel esp */
606	/*
607	 * Disable interrupts, but do not irqtrace this section: we
608	 * will soon execute iret and the tracer was already set to
609	 * the irqstate after the IRET:
610	 */
611	DISABLE_INTERRUPTS(CLBR_ANY)
612	lss	(%esp), %esp			/* switch to espfix segment */
613	jmp	.Lrestore_nocheck
614#endif
615ENDPROC(entry_INT80_32)
616
617.macro FIXUP_ESPFIX_STACK
618/*
619 * Switch back for ESPFIX stack to the normal zerobased stack
620 *
621 * We can't call C functions using the ESPFIX stack. This code reads
622 * the high word of the segment base from the GDT and swiches to the
623 * normal stack and adjusts ESP with the matching offset.
624 */
625#ifdef CONFIG_X86_ESPFIX32
626	/* fixup the stack */
627	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
628	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
629	shl	$16, %eax
630	addl	%esp, %eax			/* the adjusted stack pointer */
631	pushl	$__KERNEL_DS
632	pushl	%eax
633	lss	(%esp), %esp			/* switch to the normal stack segment */
634#endif
635.endm
636.macro UNWIND_ESPFIX_STACK
637#ifdef CONFIG_X86_ESPFIX32
638	movl	%ss, %eax
639	/* see if on espfix stack */
640	cmpw	$__ESPFIX_SS, %ax
641	jne	27f
642	movl	$__KERNEL_DS, %eax
643	movl	%eax, %ds
644	movl	%eax, %es
645	/* switch to normal stack */
646	FIXUP_ESPFIX_STACK
64727:
648#endif
649.endm
650
651/*
652 * Build the entry stubs with some assembler magic.
653 * We pack 1 stub into every 8-byte block.
654 */
655	.align 8
656ENTRY(irq_entries_start)
657    vector=FIRST_EXTERNAL_VECTOR
658    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
659	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
660    vector=vector+1
661	jmp	common_interrupt
662	.align	8
663    .endr
664END(irq_entries_start)
665
666/*
667 * the CPU automatically disables interrupts when executing an IRQ vector,
668 * so IRQ-flags tracing has to follow that:
669 */
670	.p2align CONFIG_X86_L1_CACHE_SHIFT
671common_interrupt:
672	ASM_CLAC
673	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
674	SAVE_ALL
675	ENCODE_FRAME_POINTER
676	TRACE_IRQS_OFF
677	movl	%esp, %eax
678	call	do_IRQ
679	jmp	ret_from_intr
680ENDPROC(common_interrupt)
681
682#define BUILD_INTERRUPT3(name, nr, fn)	\
683ENTRY(name)				\
684	ASM_CLAC;			\
685	pushl	$~(nr);			\
686	SAVE_ALL;			\
687	ENCODE_FRAME_POINTER;		\
688	TRACE_IRQS_OFF			\
689	movl	%esp, %eax;		\
690	call	fn;			\
691	jmp	ret_from_intr;		\
692ENDPROC(name)
693
694#define BUILD_INTERRUPT(name, nr)		\
695	BUILD_INTERRUPT3(name, nr, smp_##name);	\
696
697/* The include is where all of the SMP etc. interrupts come from */
698#include <asm/entry_arch.h>
699
700ENTRY(coprocessor_error)
701	ASM_CLAC
702	pushl	$0
703	pushl	$do_coprocessor_error
704	jmp	common_exception
705END(coprocessor_error)
706
707ENTRY(simd_coprocessor_error)
708	ASM_CLAC
709	pushl	$0
710#ifdef CONFIG_X86_INVD_BUG
711	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
712	ALTERNATIVE "pushl	$do_general_protection",	\
713		    "pushl	$do_simd_coprocessor_error",	\
714		    X86_FEATURE_XMM
715#else
716	pushl	$do_simd_coprocessor_error
717#endif
718	jmp	common_exception
719END(simd_coprocessor_error)
720
721ENTRY(device_not_available)
722	ASM_CLAC
723	pushl	$-1				# mark this as an int
724	pushl	$do_device_not_available
725	jmp	common_exception
726END(device_not_available)
727
728#ifdef CONFIG_PARAVIRT
729ENTRY(native_iret)
730	iret
731	_ASM_EXTABLE(native_iret, iret_exc)
732END(native_iret)
733#endif
734
735ENTRY(overflow)
736	ASM_CLAC
737	pushl	$0
738	pushl	$do_overflow
739	jmp	common_exception
740END(overflow)
741
742ENTRY(bounds)
743	ASM_CLAC
744	pushl	$0
745	pushl	$do_bounds
746	jmp	common_exception
747END(bounds)
748
749ENTRY(invalid_op)
750	ASM_CLAC
751	pushl	$0
752	pushl	$do_invalid_op
753	jmp	common_exception
754END(invalid_op)
755
756ENTRY(coprocessor_segment_overrun)
757	ASM_CLAC
758	pushl	$0
759	pushl	$do_coprocessor_segment_overrun
760	jmp	common_exception
761END(coprocessor_segment_overrun)
762
763ENTRY(invalid_TSS)
764	ASM_CLAC
765	pushl	$do_invalid_TSS
766	jmp	common_exception
767END(invalid_TSS)
768
769ENTRY(segment_not_present)
770	ASM_CLAC
771	pushl	$do_segment_not_present
772	jmp	common_exception
773END(segment_not_present)
774
775ENTRY(stack_segment)
776	ASM_CLAC
777	pushl	$do_stack_segment
778	jmp	common_exception
779END(stack_segment)
780
781ENTRY(alignment_check)
782	ASM_CLAC
783	pushl	$do_alignment_check
784	jmp	common_exception
785END(alignment_check)
786
787ENTRY(divide_error)
788	ASM_CLAC
789	pushl	$0				# no error code
790	pushl	$do_divide_error
791	jmp	common_exception
792END(divide_error)
793
794#ifdef CONFIG_X86_MCE
795ENTRY(machine_check)
796	ASM_CLAC
797	pushl	$0
798	pushl	machine_check_vector
799	jmp	common_exception
800END(machine_check)
801#endif
802
803ENTRY(spurious_interrupt_bug)
804	ASM_CLAC
805	pushl	$0
806	pushl	$do_spurious_interrupt_bug
807	jmp	common_exception
808END(spurious_interrupt_bug)
809
810#ifdef CONFIG_XEN
811ENTRY(xen_hypervisor_callback)
812	pushl	$-1				/* orig_ax = -1 => not a system call */
813	SAVE_ALL
814	ENCODE_FRAME_POINTER
815	TRACE_IRQS_OFF
816
817	/*
818	 * Check to see if we got the event in the critical
819	 * region in xen_iret_direct, after we've reenabled
820	 * events and checked for pending events.  This simulates
821	 * iret instruction's behaviour where it delivers a
822	 * pending interrupt when enabling interrupts:
823	 */
824	movl	PT_EIP(%esp), %eax
825	cmpl	$xen_iret_start_crit, %eax
826	jb	1f
827	cmpl	$xen_iret_end_crit, %eax
828	jae	1f
829
830	jmp	xen_iret_crit_fixup
831
832ENTRY(xen_do_upcall)
8331:	mov	%esp, %eax
834	call	xen_evtchn_do_upcall
835#ifndef CONFIG_PREEMPT
836	call	xen_maybe_preempt_hcall
837#endif
838	jmp	ret_from_intr
839ENDPROC(xen_hypervisor_callback)
840
841/*
842 * Hypervisor uses this for application faults while it executes.
843 * We get here for two reasons:
844 *  1. Fault while reloading DS, ES, FS or GS
845 *  2. Fault while executing IRET
846 * Category 1 we fix up by reattempting the load, and zeroing the segment
847 * register if the load fails.
848 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
849 * normal Linux return path in this case because if we use the IRET hypercall
850 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
851 * We distinguish between categories by maintaining a status value in EAX.
852 */
853ENTRY(xen_failsafe_callback)
854	pushl	%eax
855	movl	$1, %eax
8561:	mov	4(%esp), %ds
8572:	mov	8(%esp), %es
8583:	mov	12(%esp), %fs
8594:	mov	16(%esp), %gs
860	/* EAX == 0 => Category 1 (Bad segment)
861	   EAX != 0 => Category 2 (Bad IRET) */
862	testl	%eax, %eax
863	popl	%eax
864	lea	16(%esp), %esp
865	jz	5f
866	jmp	iret_exc
8675:	pushl	$-1				/* orig_ax = -1 => not a system call */
868	SAVE_ALL
869	ENCODE_FRAME_POINTER
870	jmp	ret_from_exception
871
872.section .fixup, "ax"
8736:	xorl	%eax, %eax
874	movl	%eax, 4(%esp)
875	jmp	1b
8767:	xorl	%eax, %eax
877	movl	%eax, 8(%esp)
878	jmp	2b
8798:	xorl	%eax, %eax
880	movl	%eax, 12(%esp)
881	jmp	3b
8829:	xorl	%eax, %eax
883	movl	%eax, 16(%esp)
884	jmp	4b
885.previous
886	_ASM_EXTABLE(1b, 6b)
887	_ASM_EXTABLE(2b, 7b)
888	_ASM_EXTABLE(3b, 8b)
889	_ASM_EXTABLE(4b, 9b)
890ENDPROC(xen_failsafe_callback)
891
892BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
893		 xen_evtchn_do_upcall)
894
895#endif /* CONFIG_XEN */
896
897#if IS_ENABLED(CONFIG_HYPERV)
898
899BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
900		 hyperv_vector_handler)
901
902BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
903		 hyperv_reenlightenment_intr)
904
905BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
906		 hv_stimer0_vector_handler)
907
908#endif /* CONFIG_HYPERV */
909
910ENTRY(page_fault)
911	ASM_CLAC
912	pushl	$do_page_fault
913	ALIGN
914	jmp common_exception
915END(page_fault)
916
917common_exception:
918	/* the function address is in %gs's slot on the stack */
919	pushl	%fs
920	pushl	%es
921	pushl	%ds
922	pushl	%eax
923	pushl	%ebp
924	pushl	%edi
925	pushl	%esi
926	pushl	%edx
927	pushl	%ecx
928	pushl	%ebx
929	ENCODE_FRAME_POINTER
930	cld
931	movl	$(__KERNEL_PERCPU), %ecx
932	movl	%ecx, %fs
933	UNWIND_ESPFIX_STACK
934	GS_TO_REG %ecx
935	movl	PT_GS(%esp), %edi		# get the function address
936	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
937	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
938	REG_TO_PTGS %ecx
939	SET_KERNEL_GS %ecx
940	movl	$(__USER_DS), %ecx
941	movl	%ecx, %ds
942	movl	%ecx, %es
943	TRACE_IRQS_OFF
944	movl	%esp, %eax			# pt_regs pointer
945	CALL_NOSPEC %edi
946	jmp	ret_from_exception
947END(common_exception)
948
949ENTRY(debug)
950	/*
951	 * #DB can happen at the first instruction of
952	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
953	 * happens, then we will be running on a very small stack.  We
954	 * need to detect this condition and switch to the thread
955	 * stack before calling any C code at all.
956	 *
957	 * If you edit this code, keep in mind that NMIs can happen in here.
958	 */
959	ASM_CLAC
960	pushl	$-1				# mark this as an int
961	SAVE_ALL
962	ENCODE_FRAME_POINTER
963	xorl	%edx, %edx			# error code 0
964	movl	%esp, %eax			# pt_regs pointer
965
966	/* Are we currently on the SYSENTER stack? */
967	movl	PER_CPU_VAR(cpu_entry_area), %ecx
968	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
969	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
970	cmpl	$SIZEOF_entry_stack, %ecx
971	jb	.Ldebug_from_sysenter_stack
972
973	TRACE_IRQS_OFF
974	call	do_debug
975	jmp	ret_from_exception
976
977.Ldebug_from_sysenter_stack:
978	/* We're on the SYSENTER stack.  Switch off. */
979	movl	%esp, %ebx
980	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
981	TRACE_IRQS_OFF
982	call	do_debug
983	movl	%ebx, %esp
984	jmp	ret_from_exception
985END(debug)
986
987/*
988 * NMI is doubly nasty.  It can happen on the first instruction of
989 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
990 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
991 * switched stacks.  We handle both conditions by simply checking whether we
992 * interrupted kernel code running on the SYSENTER stack.
993 */
994ENTRY(nmi)
995	ASM_CLAC
996#ifdef CONFIG_X86_ESPFIX32
997	pushl	%eax
998	movl	%ss, %eax
999	cmpw	$__ESPFIX_SS, %ax
1000	popl	%eax
1001	je	.Lnmi_espfix_stack
1002#endif
1003
1004	pushl	%eax				# pt_regs->orig_ax
1005	SAVE_ALL
1006	ENCODE_FRAME_POINTER
1007	xorl	%edx, %edx			# zero error code
1008	movl	%esp, %eax			# pt_regs pointer
1009
1010	/* Are we currently on the SYSENTER stack? */
1011	movl	PER_CPU_VAR(cpu_entry_area), %ecx
1012	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1013	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
1014	cmpl	$SIZEOF_entry_stack, %ecx
1015	jb	.Lnmi_from_sysenter_stack
1016
1017	/* Not on SYSENTER stack. */
1018	call	do_nmi
1019	jmp	.Lrestore_all_notrace
1020
1021.Lnmi_from_sysenter_stack:
1022	/*
1023	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1024	 * is using the thread stack right now, so it's safe for us to use it.
1025	 */
1026	movl	%esp, %ebx
1027	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1028	call	do_nmi
1029	movl	%ebx, %esp
1030	jmp	.Lrestore_all_notrace
1031
1032#ifdef CONFIG_X86_ESPFIX32
1033.Lnmi_espfix_stack:
1034	/*
1035	 * create the pointer to lss back
1036	 */
1037	pushl	%ss
1038	pushl	%esp
1039	addl	$4, (%esp)
1040	/* copy the iret frame of 12 bytes */
1041	.rept 3
1042	pushl	16(%esp)
1043	.endr
1044	pushl	%eax
1045	SAVE_ALL
1046	ENCODE_FRAME_POINTER
1047	FIXUP_ESPFIX_STACK			# %eax == %esp
1048	xorl	%edx, %edx			# zero error code
1049	call	do_nmi
1050	RESTORE_REGS
1051	lss	12+4(%esp), %esp		# back to espfix stack
1052	jmp	.Lirq_return
1053#endif
1054END(nmi)
1055
1056ENTRY(int3)
1057	ASM_CLAC
1058	pushl	$-1				# mark this as an int
1059	SAVE_ALL
1060	ENCODE_FRAME_POINTER
1061	TRACE_IRQS_OFF
1062	xorl	%edx, %edx			# zero error code
1063	movl	%esp, %eax			# pt_regs pointer
1064	call	do_int3
1065	jmp	ret_from_exception
1066END(int3)
1067
1068ENTRY(general_protection)
1069	pushl	$do_general_protection
1070	jmp	common_exception
1071END(general_protection)
1072
1073#ifdef CONFIG_KVM_GUEST
1074ENTRY(async_page_fault)
1075	ASM_CLAC
1076	pushl	$do_async_page_fault
1077	jmp	common_exception
1078END(async_page_fault)
1079#endif
1080
1081ENTRY(rewind_stack_do_exit)
1082	/* Prevent any naive code from trying to unwind to our caller. */
1083	xorl	%ebp, %ebp
1084
1085	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1086	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1087
1088	call	do_exit
10891:	jmp 1b
1090END(rewind_stack_do_exit)
1091