xref: /linux/arch/x86/entry/entry_32.S (revision 140eb5227767c6754742020a16d2691222b9c19b)
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	INTERRUPT_RETURN
569
570.section .fixup, "ax"
571ENTRY(iret_exc	)
572	pushl	$0				# no error code
573	pushl	$do_iret_error
574	jmp	common_exception
575.previous
576	_ASM_EXTABLE(.Lirq_return, iret_exc)
577
578#ifdef CONFIG_X86_ESPFIX32
579.Lldt_ss:
580/*
581 * Setup and switch to ESPFIX stack
582 *
583 * We're returning to userspace with a 16 bit stack. The CPU will not
584 * restore the high word of ESP for us on executing iret... This is an
585 * "official" bug of all the x86-compatible CPUs, which we can work
586 * around to make dosemu and wine happy. We do this by preloading the
587 * high word of ESP with the high word of the userspace ESP while
588 * compensating for the offset by changing to the ESPFIX segment with
589 * a base address that matches for the difference.
590 */
591#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
592	mov	%esp, %edx			/* load kernel esp */
593	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
594	mov	%dx, %ax			/* eax: new kernel esp */
595	sub	%eax, %edx			/* offset (low word is 0) */
596	shr	$16, %edx
597	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
598	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
599	pushl	$__ESPFIX_SS
600	pushl	%eax				/* new kernel esp */
601	/*
602	 * Disable interrupts, but do not irqtrace this section: we
603	 * will soon execute iret and the tracer was already set to
604	 * the irqstate after the IRET:
605	 */
606	DISABLE_INTERRUPTS(CLBR_ANY)
607	lss	(%esp), %esp			/* switch to espfix segment */
608	jmp	.Lrestore_nocheck
609#endif
610ENDPROC(entry_INT80_32)
611
612.macro FIXUP_ESPFIX_STACK
613/*
614 * Switch back for ESPFIX stack to the normal zerobased stack
615 *
616 * We can't call C functions using the ESPFIX stack. This code reads
617 * the high word of the segment base from the GDT and swiches to the
618 * normal stack and adjusts ESP with the matching offset.
619 */
620#ifdef CONFIG_X86_ESPFIX32
621	/* fixup the stack */
622	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
623	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
624	shl	$16, %eax
625	addl	%esp, %eax			/* the adjusted stack pointer */
626	pushl	$__KERNEL_DS
627	pushl	%eax
628	lss	(%esp), %esp			/* switch to the normal stack segment */
629#endif
630.endm
631.macro UNWIND_ESPFIX_STACK
632#ifdef CONFIG_X86_ESPFIX32
633	movl	%ss, %eax
634	/* see if on espfix stack */
635	cmpw	$__ESPFIX_SS, %ax
636	jne	27f
637	movl	$__KERNEL_DS, %eax
638	movl	%eax, %ds
639	movl	%eax, %es
640	/* switch to normal stack */
641	FIXUP_ESPFIX_STACK
64227:
643#endif
644.endm
645
646/*
647 * Build the entry stubs with some assembler magic.
648 * We pack 1 stub into every 8-byte block.
649 */
650	.align 8
651ENTRY(irq_entries_start)
652    vector=FIRST_EXTERNAL_VECTOR
653    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
654	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
655    vector=vector+1
656	jmp	common_interrupt
657	.align	8
658    .endr
659END(irq_entries_start)
660
661/*
662 * the CPU automatically disables interrupts when executing an IRQ vector,
663 * so IRQ-flags tracing has to follow that:
664 */
665	.p2align CONFIG_X86_L1_CACHE_SHIFT
666common_interrupt:
667	ASM_CLAC
668	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
669	SAVE_ALL
670	ENCODE_FRAME_POINTER
671	TRACE_IRQS_OFF
672	movl	%esp, %eax
673	call	do_IRQ
674	jmp	ret_from_intr
675ENDPROC(common_interrupt)
676
677#define BUILD_INTERRUPT3(name, nr, fn)	\
678ENTRY(name)				\
679	ASM_CLAC;			\
680	pushl	$~(nr);			\
681	SAVE_ALL;			\
682	ENCODE_FRAME_POINTER;		\
683	TRACE_IRQS_OFF			\
684	movl	%esp, %eax;		\
685	call	fn;			\
686	jmp	ret_from_intr;		\
687ENDPROC(name)
688
689#define BUILD_INTERRUPT(name, nr)		\
690	BUILD_INTERRUPT3(name, nr, smp_##name);	\
691
692/* The include is where all of the SMP etc. interrupts come from */
693#include <asm/entry_arch.h>
694
695ENTRY(coprocessor_error)
696	ASM_CLAC
697	pushl	$0
698	pushl	$do_coprocessor_error
699	jmp	common_exception
700END(coprocessor_error)
701
702ENTRY(simd_coprocessor_error)
703	ASM_CLAC
704	pushl	$0
705#ifdef CONFIG_X86_INVD_BUG
706	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
707	ALTERNATIVE "pushl	$do_general_protection",	\
708		    "pushl	$do_simd_coprocessor_error",	\
709		    X86_FEATURE_XMM
710#else
711	pushl	$do_simd_coprocessor_error
712#endif
713	jmp	common_exception
714END(simd_coprocessor_error)
715
716ENTRY(device_not_available)
717	ASM_CLAC
718	pushl	$-1				# mark this as an int
719	pushl	$do_device_not_available
720	jmp	common_exception
721END(device_not_available)
722
723#ifdef CONFIG_PARAVIRT
724ENTRY(native_iret)
725	iret
726	_ASM_EXTABLE(native_iret, iret_exc)
727END(native_iret)
728#endif
729
730ENTRY(overflow)
731	ASM_CLAC
732	pushl	$0
733	pushl	$do_overflow
734	jmp	common_exception
735END(overflow)
736
737ENTRY(bounds)
738	ASM_CLAC
739	pushl	$0
740	pushl	$do_bounds
741	jmp	common_exception
742END(bounds)
743
744ENTRY(invalid_op)
745	ASM_CLAC
746	pushl	$0
747	pushl	$do_invalid_op
748	jmp	common_exception
749END(invalid_op)
750
751ENTRY(coprocessor_segment_overrun)
752	ASM_CLAC
753	pushl	$0
754	pushl	$do_coprocessor_segment_overrun
755	jmp	common_exception
756END(coprocessor_segment_overrun)
757
758ENTRY(invalid_TSS)
759	ASM_CLAC
760	pushl	$do_invalid_TSS
761	jmp	common_exception
762END(invalid_TSS)
763
764ENTRY(segment_not_present)
765	ASM_CLAC
766	pushl	$do_segment_not_present
767	jmp	common_exception
768END(segment_not_present)
769
770ENTRY(stack_segment)
771	ASM_CLAC
772	pushl	$do_stack_segment
773	jmp	common_exception
774END(stack_segment)
775
776ENTRY(alignment_check)
777	ASM_CLAC
778	pushl	$do_alignment_check
779	jmp	common_exception
780END(alignment_check)
781
782ENTRY(divide_error)
783	ASM_CLAC
784	pushl	$0				# no error code
785	pushl	$do_divide_error
786	jmp	common_exception
787END(divide_error)
788
789#ifdef CONFIG_X86_MCE
790ENTRY(machine_check)
791	ASM_CLAC
792	pushl	$0
793	pushl	machine_check_vector
794	jmp	common_exception
795END(machine_check)
796#endif
797
798ENTRY(spurious_interrupt_bug)
799	ASM_CLAC
800	pushl	$0
801	pushl	$do_spurious_interrupt_bug
802	jmp	common_exception
803END(spurious_interrupt_bug)
804
805#ifdef CONFIG_XEN
806ENTRY(xen_hypervisor_callback)
807	pushl	$-1				/* orig_ax = -1 => not a system call */
808	SAVE_ALL
809	ENCODE_FRAME_POINTER
810	TRACE_IRQS_OFF
811
812	/*
813	 * Check to see if we got the event in the critical
814	 * region in xen_iret_direct, after we've reenabled
815	 * events and checked for pending events.  This simulates
816	 * iret instruction's behaviour where it delivers a
817	 * pending interrupt when enabling interrupts:
818	 */
819	movl	PT_EIP(%esp), %eax
820	cmpl	$xen_iret_start_crit, %eax
821	jb	1f
822	cmpl	$xen_iret_end_crit, %eax
823	jae	1f
824
825	jmp	xen_iret_crit_fixup
826
827ENTRY(xen_do_upcall)
8281:	mov	%esp, %eax
829	call	xen_evtchn_do_upcall
830#ifndef CONFIG_PREEMPT
831	call	xen_maybe_preempt_hcall
832#endif
833	jmp	ret_from_intr
834ENDPROC(xen_hypervisor_callback)
835
836/*
837 * Hypervisor uses this for application faults while it executes.
838 * We get here for two reasons:
839 *  1. Fault while reloading DS, ES, FS or GS
840 *  2. Fault while executing IRET
841 * Category 1 we fix up by reattempting the load, and zeroing the segment
842 * register if the load fails.
843 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
844 * normal Linux return path in this case because if we use the IRET hypercall
845 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
846 * We distinguish between categories by maintaining a status value in EAX.
847 */
848ENTRY(xen_failsafe_callback)
849	pushl	%eax
850	movl	$1, %eax
8511:	mov	4(%esp), %ds
8522:	mov	8(%esp), %es
8533:	mov	12(%esp), %fs
8544:	mov	16(%esp), %gs
855	/* EAX == 0 => Category 1 (Bad segment)
856	   EAX != 0 => Category 2 (Bad IRET) */
857	testl	%eax, %eax
858	popl	%eax
859	lea	16(%esp), %esp
860	jz	5f
861	jmp	iret_exc
8625:	pushl	$-1				/* orig_ax = -1 => not a system call */
863	SAVE_ALL
864	ENCODE_FRAME_POINTER
865	jmp	ret_from_exception
866
867.section .fixup, "ax"
8686:	xorl	%eax, %eax
869	movl	%eax, 4(%esp)
870	jmp	1b
8717:	xorl	%eax, %eax
872	movl	%eax, 8(%esp)
873	jmp	2b
8748:	xorl	%eax, %eax
875	movl	%eax, 12(%esp)
876	jmp	3b
8779:	xorl	%eax, %eax
878	movl	%eax, 16(%esp)
879	jmp	4b
880.previous
881	_ASM_EXTABLE(1b, 6b)
882	_ASM_EXTABLE(2b, 7b)
883	_ASM_EXTABLE(3b, 8b)
884	_ASM_EXTABLE(4b, 9b)
885ENDPROC(xen_failsafe_callback)
886
887BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
888		 xen_evtchn_do_upcall)
889
890#endif /* CONFIG_XEN */
891
892#if IS_ENABLED(CONFIG_HYPERV)
893
894BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
895		 hyperv_vector_handler)
896
897#endif /* CONFIG_HYPERV */
898
899ENTRY(page_fault)
900	ASM_CLAC
901	pushl	$do_page_fault
902	ALIGN
903	jmp common_exception
904END(page_fault)
905
906common_exception:
907	/* the function address is in %gs's slot on the stack */
908	pushl	%fs
909	pushl	%es
910	pushl	%ds
911	pushl	%eax
912	pushl	%ebp
913	pushl	%edi
914	pushl	%esi
915	pushl	%edx
916	pushl	%ecx
917	pushl	%ebx
918	ENCODE_FRAME_POINTER
919	cld
920	movl	$(__KERNEL_PERCPU), %ecx
921	movl	%ecx, %fs
922	UNWIND_ESPFIX_STACK
923	GS_TO_REG %ecx
924	movl	PT_GS(%esp), %edi		# get the function address
925	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
926	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
927	REG_TO_PTGS %ecx
928	SET_KERNEL_GS %ecx
929	movl	$(__USER_DS), %ecx
930	movl	%ecx, %ds
931	movl	%ecx, %es
932	TRACE_IRQS_OFF
933	movl	%esp, %eax			# pt_regs pointer
934	CALL_NOSPEC %edi
935	jmp	ret_from_exception
936END(common_exception)
937
938ENTRY(debug)
939	/*
940	 * #DB can happen at the first instruction of
941	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
942	 * happens, then we will be running on a very small stack.  We
943	 * need to detect this condition and switch to the thread
944	 * stack before calling any C code at all.
945	 *
946	 * If you edit this code, keep in mind that NMIs can happen in here.
947	 */
948	ASM_CLAC
949	pushl	$-1				# mark this as an int
950	SAVE_ALL
951	ENCODE_FRAME_POINTER
952	xorl	%edx, %edx			# error code 0
953	movl	%esp, %eax			# pt_regs pointer
954
955	/* Are we currently on the SYSENTER stack? */
956	movl	PER_CPU_VAR(cpu_entry_area), %ecx
957	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
958	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
959	cmpl	$SIZEOF_entry_stack, %ecx
960	jb	.Ldebug_from_sysenter_stack
961
962	TRACE_IRQS_OFF
963	call	do_debug
964	jmp	ret_from_exception
965
966.Ldebug_from_sysenter_stack:
967	/* We're on the SYSENTER stack.  Switch off. */
968	movl	%esp, %ebx
969	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
970	TRACE_IRQS_OFF
971	call	do_debug
972	movl	%ebx, %esp
973	jmp	ret_from_exception
974END(debug)
975
976/*
977 * NMI is doubly nasty.  It can happen on the first instruction of
978 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
979 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
980 * switched stacks.  We handle both conditions by simply checking whether we
981 * interrupted kernel code running on the SYSENTER stack.
982 */
983ENTRY(nmi)
984	ASM_CLAC
985#ifdef CONFIG_X86_ESPFIX32
986	pushl	%eax
987	movl	%ss, %eax
988	cmpw	$__ESPFIX_SS, %ax
989	popl	%eax
990	je	.Lnmi_espfix_stack
991#endif
992
993	pushl	%eax				# pt_regs->orig_ax
994	SAVE_ALL
995	ENCODE_FRAME_POINTER
996	xorl	%edx, %edx			# zero error code
997	movl	%esp, %eax			# pt_regs pointer
998
999	/* Are we currently on the SYSENTER stack? */
1000	movl	PER_CPU_VAR(cpu_entry_area), %ecx
1001	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1002	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
1003	cmpl	$SIZEOF_entry_stack, %ecx
1004	jb	.Lnmi_from_sysenter_stack
1005
1006	/* Not on SYSENTER stack. */
1007	call	do_nmi
1008	jmp	.Lrestore_all_notrace
1009
1010.Lnmi_from_sysenter_stack:
1011	/*
1012	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1013	 * is using the thread stack right now, so it's safe for us to use it.
1014	 */
1015	movl	%esp, %ebx
1016	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1017	call	do_nmi
1018	movl	%ebx, %esp
1019	jmp	.Lrestore_all_notrace
1020
1021#ifdef CONFIG_X86_ESPFIX32
1022.Lnmi_espfix_stack:
1023	/*
1024	 * create the pointer to lss back
1025	 */
1026	pushl	%ss
1027	pushl	%esp
1028	addl	$4, (%esp)
1029	/* copy the iret frame of 12 bytes */
1030	.rept 3
1031	pushl	16(%esp)
1032	.endr
1033	pushl	%eax
1034	SAVE_ALL
1035	ENCODE_FRAME_POINTER
1036	FIXUP_ESPFIX_STACK			# %eax == %esp
1037	xorl	%edx, %edx			# zero error code
1038	call	do_nmi
1039	RESTORE_REGS
1040	lss	12+4(%esp), %esp		# back to espfix stack
1041	jmp	.Lirq_return
1042#endif
1043END(nmi)
1044
1045ENTRY(int3)
1046	ASM_CLAC
1047	pushl	$-1				# mark this as an int
1048	SAVE_ALL
1049	ENCODE_FRAME_POINTER
1050	TRACE_IRQS_OFF
1051	xorl	%edx, %edx			# zero error code
1052	movl	%esp, %eax			# pt_regs pointer
1053	call	do_int3
1054	jmp	ret_from_exception
1055END(int3)
1056
1057ENTRY(general_protection)
1058	pushl	$do_general_protection
1059	jmp	common_exception
1060END(general_protection)
1061
1062#ifdef CONFIG_KVM_GUEST
1063ENTRY(async_page_fault)
1064	ASM_CLAC
1065	pushl	$do_async_page_fault
1066	jmp	common_exception
1067END(async_page_fault)
1068#endif
1069
1070ENTRY(rewind_stack_do_exit)
1071	/* Prevent any naive code from trying to unwind to our caller. */
1072	xorl	%ebp, %ebp
1073
1074	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1075	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1076
1077	call	do_exit
10781:	jmp 1b
1079END(rewind_stack_do_exit)
1080