xref: /linux/arch/x86/entry/entry_32.S (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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
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/*
178 * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
179 * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
180 * is just clearing the MSB, which makes it an invalid stack address and is also
181 * a signal to the unwinder that it's a pt_regs pointer in disguise.
182 *
183 * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
184 * original rbp.
185 */
186.macro ENCODE_FRAME_POINTER
187#ifdef CONFIG_FRAME_POINTER
188	mov %esp, %ebp
189	andl $0x7fffffff, %ebp
190#endif
191.endm
192
193.macro RESTORE_INT_REGS
194	popl	%ebx
195	popl	%ecx
196	popl	%edx
197	popl	%esi
198	popl	%edi
199	popl	%ebp
200	popl	%eax
201.endm
202
203.macro RESTORE_REGS pop=0
204	RESTORE_INT_REGS
2051:	popl	%ds
2062:	popl	%es
2073:	popl	%fs
208	POP_GS \pop
209.pushsection .fixup, "ax"
2104:	movl	$0, (%esp)
211	jmp	1b
2125:	movl	$0, (%esp)
213	jmp	2b
2146:	movl	$0, (%esp)
215	jmp	3b
216.popsection
217	_ASM_EXTABLE(1b, 4b)
218	_ASM_EXTABLE(2b, 5b)
219	_ASM_EXTABLE(3b, 6b)
220	POP_GS_EX
221.endm
222
223/*
224 * %eax: prev task
225 * %edx: next task
226 */
227ENTRY(__switch_to_asm)
228	/*
229	 * Save callee-saved registers
230	 * This must match the order in struct inactive_task_frame
231	 */
232	pushl	%ebp
233	pushl	%ebx
234	pushl	%edi
235	pushl	%esi
236
237	/* switch stack */
238	movl	%esp, TASK_threadsp(%eax)
239	movl	TASK_threadsp(%edx), %esp
240
241#ifdef CONFIG_CC_STACKPROTECTOR
242	movl	TASK_stack_canary(%edx), %ebx
243	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
244#endif
245
246	/* restore callee-saved registers */
247	popl	%esi
248	popl	%edi
249	popl	%ebx
250	popl	%ebp
251
252	jmp	__switch_to
253END(__switch_to_asm)
254
255/*
256 * The unwinder expects the last frame on the stack to always be at the same
257 * offset from the end of the page, which allows it to validate the stack.
258 * Calling schedule_tail() directly would break that convention because its an
259 * asmlinkage function so its argument has to be pushed on the stack.  This
260 * wrapper creates a proper "end of stack" frame header before the call.
261 */
262ENTRY(schedule_tail_wrapper)
263	FRAME_BEGIN
264
265	pushl	%eax
266	call	schedule_tail
267	popl	%eax
268
269	FRAME_END
270	ret
271ENDPROC(schedule_tail_wrapper)
272/*
273 * A newly forked process directly context switches into this address.
274 *
275 * eax: prev task we switched from
276 * ebx: kernel thread func (NULL for user thread)
277 * edi: kernel thread arg
278 */
279ENTRY(ret_from_fork)
280	call	schedule_tail_wrapper
281
282	testl	%ebx, %ebx
283	jnz	1f		/* kernel threads are uncommon */
284
2852:
286	/* When we fork, we trace the syscall return in the child, too. */
287	movl    %esp, %eax
288	call    syscall_return_slowpath
289	jmp     restore_all
290
291	/* kernel thread */
2921:	movl	%edi, %eax
293	call	*%ebx
294	/*
295	 * A kernel thread is allowed to return here after successfully
296	 * calling do_execve().  Exit to userspace to complete the execve()
297	 * syscall.
298	 */
299	movl	$0, PT_EAX(%esp)
300	jmp	2b
301END(ret_from_fork)
302
303/*
304 * Return to user mode is not as complex as all this looks,
305 * but we want the default path for a system call return to
306 * go as quickly as possible which is why some of this is
307 * less clear than it otherwise should be.
308 */
309
310	# userspace resumption stub bypassing syscall exit tracing
311	ALIGN
312ret_from_exception:
313	preempt_stop(CLBR_ANY)
314ret_from_intr:
315#ifdef CONFIG_VM86
316	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
317	movb	PT_CS(%esp), %al
318	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
319#else
320	/*
321	 * We can be coming here from child spawned by kernel_thread().
322	 */
323	movl	PT_CS(%esp), %eax
324	andl	$SEGMENT_RPL_MASK, %eax
325#endif
326	cmpl	$USER_RPL, %eax
327	jb	resume_kernel			# not returning to v8086 or userspace
328
329ENTRY(resume_userspace)
330	DISABLE_INTERRUPTS(CLBR_ANY)
331	TRACE_IRQS_OFF
332	movl	%esp, %eax
333	call	prepare_exit_to_usermode
334	jmp	restore_all
335END(ret_from_exception)
336
337#ifdef CONFIG_PREEMPT
338ENTRY(resume_kernel)
339	DISABLE_INTERRUPTS(CLBR_ANY)
340.Lneed_resched:
341	cmpl	$0, PER_CPU_VAR(__preempt_count)
342	jnz	restore_all
343	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
344	jz	restore_all
345	call	preempt_schedule_irq
346	jmp	.Lneed_resched
347END(resume_kernel)
348#endif
349
350GLOBAL(__begin_SYSENTER_singlestep_region)
351/*
352 * All code from here through __end_SYSENTER_singlestep_region is subject
353 * to being single-stepped if a user program sets TF and executes SYSENTER.
354 * There is absolutely nothing that we can do to prevent this from happening
355 * (thanks Intel!).  To keep our handling of this situation as simple as
356 * possible, we handle TF just like AC and NT, except that our #DB handler
357 * will ignore all of the single-step traps generated in this range.
358 */
359
360#ifdef CONFIG_XEN
361/*
362 * Xen doesn't set %esp to be precisely what the normal SYSENTER
363 * entry point expects, so fix it up before using the normal path.
364 */
365ENTRY(xen_sysenter_target)
366	addl	$5*4, %esp			/* remove xen-provided frame */
367	jmp	.Lsysenter_past_esp
368#endif
369
370/*
371 * 32-bit SYSENTER entry.
372 *
373 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
374 * if X86_FEATURE_SEP is available.  This is the preferred system call
375 * entry on 32-bit systems.
376 *
377 * The SYSENTER instruction, in principle, should *only* occur in the
378 * vDSO.  In practice, a small number of Android devices were shipped
379 * with a copy of Bionic that inlined a SYSENTER instruction.  This
380 * never happened in any of Google's Bionic versions -- it only happened
381 * in a narrow range of Intel-provided versions.
382 *
383 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
384 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
385 * SYSENTER does not save anything on the stack,
386 * and does not save old EIP (!!!), ESP, or EFLAGS.
387 *
388 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
389 * user and/or vm86 state), we explicitly disable the SYSENTER
390 * instruction in vm86 mode by reprogramming the MSRs.
391 *
392 * Arguments:
393 * eax  system call number
394 * ebx  arg1
395 * ecx  arg2
396 * edx  arg3
397 * esi  arg4
398 * edi  arg5
399 * ebp  user stack
400 * 0(%ebp) arg6
401 */
402ENTRY(entry_SYSENTER_32)
403	movl	TSS_sysenter_sp0(%esp), %esp
404.Lsysenter_past_esp:
405	pushl	$__USER_DS		/* pt_regs->ss */
406	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
407	pushfl				/* pt_regs->flags (except IF = 0) */
408	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
409	pushl	$__USER_CS		/* pt_regs->cs */
410	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
411	pushl	%eax			/* pt_regs->orig_ax */
412	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
413
414	/*
415	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
416	 * and TF ourselves.  To save a few cycles, we can check whether
417	 * either was set instead of doing an unconditional popfq.
418	 * This needs to happen before enabling interrupts so that
419	 * we don't get preempted with NT set.
420	 *
421	 * If TF is set, we will single-step all the way to here -- do_debug
422	 * will ignore all the traps.  (Yes, this is slow, but so is
423	 * single-stepping in general.  This allows us to avoid having
424	 * a more complicated code to handle the case where a user program
425	 * forces us to single-step through the SYSENTER entry code.)
426	 *
427	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
428	 * out-of-line as an optimization: NT is unlikely to be set in the
429	 * majority of the cases and instead of polluting the I$ unnecessarily,
430	 * we're keeping that code behind a branch which will predict as
431	 * not-taken and therefore its instructions won't be fetched.
432	 */
433	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
434	jnz	.Lsysenter_fix_flags
435.Lsysenter_flags_fixed:
436
437	/*
438	 * User mode is traced as though IRQs are on, and SYSENTER
439	 * turned them off.
440	 */
441	TRACE_IRQS_OFF
442
443	movl	%esp, %eax
444	call	do_fast_syscall_32
445	/* XEN PV guests always use IRET path */
446	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
447		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
448
449/* Opportunistic SYSEXIT */
450	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
451	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
452	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
4531:	mov	PT_FS(%esp), %fs
454	PTGS_TO_GS
455	popl	%ebx			/* pt_regs->bx */
456	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
457	popl	%esi			/* pt_regs->si */
458	popl	%edi			/* pt_regs->di */
459	popl	%ebp			/* pt_regs->bp */
460	popl	%eax			/* pt_regs->ax */
461
462	/*
463	 * Restore all flags except IF. (We restore IF separately because
464	 * STI gives a one-instruction window in which we won't be interrupted,
465	 * whereas POPF does not.)
466	 */
467	addl	$PT_EFLAGS-PT_DS, %esp	/* point esp at pt_regs->flags */
468	btr	$X86_EFLAGS_IF_BIT, (%esp)
469	popfl
470
471	/*
472	 * Return back to the vDSO, which will pop ecx and edx.
473	 * Don't bother with DS and ES (they already contain __USER_DS).
474	 */
475	sti
476	sysexit
477
478.pushsection .fixup, "ax"
4792:	movl	$0, PT_FS(%esp)
480	jmp	1b
481.popsection
482	_ASM_EXTABLE(1b, 2b)
483	PTGS_TO_GS_EX
484
485.Lsysenter_fix_flags:
486	pushl	$X86_EFLAGS_FIXED
487	popfl
488	jmp	.Lsysenter_flags_fixed
489GLOBAL(__end_SYSENTER_singlestep_region)
490ENDPROC(entry_SYSENTER_32)
491
492/*
493 * 32-bit legacy system call entry.
494 *
495 * 32-bit x86 Linux system calls traditionally used the INT $0x80
496 * instruction.  INT $0x80 lands here.
497 *
498 * This entry point can be used by any 32-bit perform system calls.
499 * Instances of INT $0x80 can be found inline in various programs and
500 * libraries.  It is also used by the vDSO's __kernel_vsyscall
501 * fallback for hardware that doesn't support a faster entry method.
502 * Restarted 32-bit system calls also fall back to INT $0x80
503 * regardless of what instruction was originally used to do the system
504 * call.  (64-bit programs can use INT $0x80 as well, but they can
505 * only run on 64-bit kernels and therefore land in
506 * entry_INT80_compat.)
507 *
508 * This is considered a slow path.  It is not used by most libc
509 * implementations on modern hardware except during process startup.
510 *
511 * Arguments:
512 * eax  system call number
513 * ebx  arg1
514 * ecx  arg2
515 * edx  arg3
516 * esi  arg4
517 * edi  arg5
518 * ebp  arg6
519 */
520ENTRY(entry_INT80_32)
521	ASM_CLAC
522	pushl	%eax			/* pt_regs->orig_ax */
523	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
524
525	/*
526	 * User mode is traced as though IRQs are on, and the interrupt gate
527	 * turned them off.
528	 */
529	TRACE_IRQS_OFF
530
531	movl	%esp, %eax
532	call	do_int80_syscall_32
533.Lsyscall_32_done:
534
535restore_all:
536	TRACE_IRQS_IRET
537.Lrestore_all_notrace:
538#ifdef CONFIG_X86_ESPFIX32
539	ALTERNATIVE	"jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
540
541	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
542	/*
543	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
544	 * are returning to the kernel.
545	 * See comments in process.c:copy_thread() for details.
546	 */
547	movb	PT_OLDSS(%esp), %ah
548	movb	PT_CS(%esp), %al
549	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
550	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
551	je .Lldt_ss				# returning to user-space with LDT SS
552#endif
553.Lrestore_nocheck:
554	RESTORE_REGS 4				# skip orig_eax/error_code
555.Lirq_return:
556	INTERRUPT_RETURN
557
558.section .fixup, "ax"
559ENTRY(iret_exc	)
560	pushl	$0				# no error code
561	pushl	$do_iret_error
562	jmp	common_exception
563.previous
564	_ASM_EXTABLE(.Lirq_return, iret_exc)
565
566#ifdef CONFIG_X86_ESPFIX32
567.Lldt_ss:
568/*
569 * Setup and switch to ESPFIX stack
570 *
571 * We're returning to userspace with a 16 bit stack. The CPU will not
572 * restore the high word of ESP for us on executing iret... This is an
573 * "official" bug of all the x86-compatible CPUs, which we can work
574 * around to make dosemu and wine happy. We do this by preloading the
575 * high word of ESP with the high word of the userspace ESP while
576 * compensating for the offset by changing to the ESPFIX segment with
577 * a base address that matches for the difference.
578 */
579#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
580	mov	%esp, %edx			/* load kernel esp */
581	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
582	mov	%dx, %ax			/* eax: new kernel esp */
583	sub	%eax, %edx			/* offset (low word is 0) */
584	shr	$16, %edx
585	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
586	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
587	pushl	$__ESPFIX_SS
588	pushl	%eax				/* new kernel esp */
589	/*
590	 * Disable interrupts, but do not irqtrace this section: we
591	 * will soon execute iret and the tracer was already set to
592	 * the irqstate after the IRET:
593	 */
594	DISABLE_INTERRUPTS(CLBR_ANY)
595	lss	(%esp), %esp			/* switch to espfix segment */
596	jmp	.Lrestore_nocheck
597#endif
598ENDPROC(entry_INT80_32)
599
600.macro FIXUP_ESPFIX_STACK
601/*
602 * Switch back for ESPFIX stack to the normal zerobased stack
603 *
604 * We can't call C functions using the ESPFIX stack. This code reads
605 * the high word of the segment base from the GDT and swiches to the
606 * normal stack and adjusts ESP with the matching offset.
607 */
608#ifdef CONFIG_X86_ESPFIX32
609	/* fixup the stack */
610	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
611	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
612	shl	$16, %eax
613	addl	%esp, %eax			/* the adjusted stack pointer */
614	pushl	$__KERNEL_DS
615	pushl	%eax
616	lss	(%esp), %esp			/* switch to the normal stack segment */
617#endif
618.endm
619.macro UNWIND_ESPFIX_STACK
620#ifdef CONFIG_X86_ESPFIX32
621	movl	%ss, %eax
622	/* see if on espfix stack */
623	cmpw	$__ESPFIX_SS, %ax
624	jne	27f
625	movl	$__KERNEL_DS, %eax
626	movl	%eax, %ds
627	movl	%eax, %es
628	/* switch to normal stack */
629	FIXUP_ESPFIX_STACK
63027:
631#endif
632.endm
633
634/*
635 * Build the entry stubs with some assembler magic.
636 * We pack 1 stub into every 8-byte block.
637 */
638	.align 8
639ENTRY(irq_entries_start)
640    vector=FIRST_EXTERNAL_VECTOR
641    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
642	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
643    vector=vector+1
644	jmp	common_interrupt
645	.align	8
646    .endr
647END(irq_entries_start)
648
649/*
650 * the CPU automatically disables interrupts when executing an IRQ vector,
651 * so IRQ-flags tracing has to follow that:
652 */
653	.p2align CONFIG_X86_L1_CACHE_SHIFT
654common_interrupt:
655	ASM_CLAC
656	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
657	SAVE_ALL
658	ENCODE_FRAME_POINTER
659	TRACE_IRQS_OFF
660	movl	%esp, %eax
661	call	do_IRQ
662	jmp	ret_from_intr
663ENDPROC(common_interrupt)
664
665#define BUILD_INTERRUPT3(name, nr, fn)	\
666ENTRY(name)				\
667	ASM_CLAC;			\
668	pushl	$~(nr);			\
669	SAVE_ALL;			\
670	ENCODE_FRAME_POINTER;		\
671	TRACE_IRQS_OFF			\
672	movl	%esp, %eax;		\
673	call	fn;			\
674	jmp	ret_from_intr;		\
675ENDPROC(name)
676
677#define BUILD_INTERRUPT(name, nr)		\
678	BUILD_INTERRUPT3(name, nr, smp_##name);	\
679
680/* The include is where all of the SMP etc. interrupts come from */
681#include <asm/entry_arch.h>
682
683ENTRY(coprocessor_error)
684	ASM_CLAC
685	pushl	$0
686	pushl	$do_coprocessor_error
687	jmp	common_exception
688END(coprocessor_error)
689
690ENTRY(simd_coprocessor_error)
691	ASM_CLAC
692	pushl	$0
693#ifdef CONFIG_X86_INVD_BUG
694	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
695	ALTERNATIVE "pushl	$do_general_protection",	\
696		    "pushl	$do_simd_coprocessor_error",	\
697		    X86_FEATURE_XMM
698#else
699	pushl	$do_simd_coprocessor_error
700#endif
701	jmp	common_exception
702END(simd_coprocessor_error)
703
704ENTRY(device_not_available)
705	ASM_CLAC
706	pushl	$-1				# mark this as an int
707	pushl	$do_device_not_available
708	jmp	common_exception
709END(device_not_available)
710
711#ifdef CONFIG_PARAVIRT
712ENTRY(native_iret)
713	iret
714	_ASM_EXTABLE(native_iret, iret_exc)
715END(native_iret)
716#endif
717
718ENTRY(overflow)
719	ASM_CLAC
720	pushl	$0
721	pushl	$do_overflow
722	jmp	common_exception
723END(overflow)
724
725ENTRY(bounds)
726	ASM_CLAC
727	pushl	$0
728	pushl	$do_bounds
729	jmp	common_exception
730END(bounds)
731
732ENTRY(invalid_op)
733	ASM_CLAC
734	pushl	$0
735	pushl	$do_invalid_op
736	jmp	common_exception
737END(invalid_op)
738
739ENTRY(coprocessor_segment_overrun)
740	ASM_CLAC
741	pushl	$0
742	pushl	$do_coprocessor_segment_overrun
743	jmp	common_exception
744END(coprocessor_segment_overrun)
745
746ENTRY(invalid_TSS)
747	ASM_CLAC
748	pushl	$do_invalid_TSS
749	jmp	common_exception
750END(invalid_TSS)
751
752ENTRY(segment_not_present)
753	ASM_CLAC
754	pushl	$do_segment_not_present
755	jmp	common_exception
756END(segment_not_present)
757
758ENTRY(stack_segment)
759	ASM_CLAC
760	pushl	$do_stack_segment
761	jmp	common_exception
762END(stack_segment)
763
764ENTRY(alignment_check)
765	ASM_CLAC
766	pushl	$do_alignment_check
767	jmp	common_exception
768END(alignment_check)
769
770ENTRY(divide_error)
771	ASM_CLAC
772	pushl	$0				# no error code
773	pushl	$do_divide_error
774	jmp	common_exception
775END(divide_error)
776
777#ifdef CONFIG_X86_MCE
778ENTRY(machine_check)
779	ASM_CLAC
780	pushl	$0
781	pushl	machine_check_vector
782	jmp	common_exception
783END(machine_check)
784#endif
785
786ENTRY(spurious_interrupt_bug)
787	ASM_CLAC
788	pushl	$0
789	pushl	$do_spurious_interrupt_bug
790	jmp	common_exception
791END(spurious_interrupt_bug)
792
793#ifdef CONFIG_XEN
794ENTRY(xen_hypervisor_callback)
795	pushl	$-1				/* orig_ax = -1 => not a system call */
796	SAVE_ALL
797	ENCODE_FRAME_POINTER
798	TRACE_IRQS_OFF
799
800	/*
801	 * Check to see if we got the event in the critical
802	 * region in xen_iret_direct, after we've reenabled
803	 * events and checked for pending events.  This simulates
804	 * iret instruction's behaviour where it delivers a
805	 * pending interrupt when enabling interrupts:
806	 */
807	movl	PT_EIP(%esp), %eax
808	cmpl	$xen_iret_start_crit, %eax
809	jb	1f
810	cmpl	$xen_iret_end_crit, %eax
811	jae	1f
812
813	jmp	xen_iret_crit_fixup
814
815ENTRY(xen_do_upcall)
8161:	mov	%esp, %eax
817	call	xen_evtchn_do_upcall
818#ifndef CONFIG_PREEMPT
819	call	xen_maybe_preempt_hcall
820#endif
821	jmp	ret_from_intr
822ENDPROC(xen_hypervisor_callback)
823
824/*
825 * Hypervisor uses this for application faults while it executes.
826 * We get here for two reasons:
827 *  1. Fault while reloading DS, ES, FS or GS
828 *  2. Fault while executing IRET
829 * Category 1 we fix up by reattempting the load, and zeroing the segment
830 * register if the load fails.
831 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
832 * normal Linux return path in this case because if we use the IRET hypercall
833 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
834 * We distinguish between categories by maintaining a status value in EAX.
835 */
836ENTRY(xen_failsafe_callback)
837	pushl	%eax
838	movl	$1, %eax
8391:	mov	4(%esp), %ds
8402:	mov	8(%esp), %es
8413:	mov	12(%esp), %fs
8424:	mov	16(%esp), %gs
843	/* EAX == 0 => Category 1 (Bad segment)
844	   EAX != 0 => Category 2 (Bad IRET) */
845	testl	%eax, %eax
846	popl	%eax
847	lea	16(%esp), %esp
848	jz	5f
849	jmp	iret_exc
8505:	pushl	$-1				/* orig_ax = -1 => not a system call */
851	SAVE_ALL
852	ENCODE_FRAME_POINTER
853	jmp	ret_from_exception
854
855.section .fixup, "ax"
8566:	xorl	%eax, %eax
857	movl	%eax, 4(%esp)
858	jmp	1b
8597:	xorl	%eax, %eax
860	movl	%eax, 8(%esp)
861	jmp	2b
8628:	xorl	%eax, %eax
863	movl	%eax, 12(%esp)
864	jmp	3b
8659:	xorl	%eax, %eax
866	movl	%eax, 16(%esp)
867	jmp	4b
868.previous
869	_ASM_EXTABLE(1b, 6b)
870	_ASM_EXTABLE(2b, 7b)
871	_ASM_EXTABLE(3b, 8b)
872	_ASM_EXTABLE(4b, 9b)
873ENDPROC(xen_failsafe_callback)
874
875BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
876		 xen_evtchn_do_upcall)
877
878#endif /* CONFIG_XEN */
879
880#if IS_ENABLED(CONFIG_HYPERV)
881
882BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
883		 hyperv_vector_handler)
884
885#endif /* CONFIG_HYPERV */
886
887ENTRY(page_fault)
888	ASM_CLAC
889	pushl	$do_page_fault
890	ALIGN
891	jmp common_exception
892END(page_fault)
893
894common_exception:
895	/* the function address is in %gs's slot on the stack */
896	pushl	%fs
897	pushl	%es
898	pushl	%ds
899	pushl	%eax
900	pushl	%ebp
901	pushl	%edi
902	pushl	%esi
903	pushl	%edx
904	pushl	%ecx
905	pushl	%ebx
906	ENCODE_FRAME_POINTER
907	cld
908	movl	$(__KERNEL_PERCPU), %ecx
909	movl	%ecx, %fs
910	UNWIND_ESPFIX_STACK
911	GS_TO_REG %ecx
912	movl	PT_GS(%esp), %edi		# get the function address
913	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
914	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
915	REG_TO_PTGS %ecx
916	SET_KERNEL_GS %ecx
917	movl	$(__USER_DS), %ecx
918	movl	%ecx, %ds
919	movl	%ecx, %es
920	TRACE_IRQS_OFF
921	movl	%esp, %eax			# pt_regs pointer
922	call	*%edi
923	jmp	ret_from_exception
924END(common_exception)
925
926ENTRY(debug)
927	/*
928	 * #DB can happen at the first instruction of
929	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
930	 * happens, then we will be running on a very small stack.  We
931	 * need to detect this condition and switch to the thread
932	 * stack before calling any C code at all.
933	 *
934	 * If you edit this code, keep in mind that NMIs can happen in here.
935	 */
936	ASM_CLAC
937	pushl	$-1				# mark this as an int
938	SAVE_ALL
939	ENCODE_FRAME_POINTER
940	xorl	%edx, %edx			# error code 0
941	movl	%esp, %eax			# pt_regs pointer
942
943	/* Are we currently on the SYSENTER stack? */
944	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
945	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
946	cmpl	$SIZEOF_SYSENTER_stack, %ecx
947	jb	.Ldebug_from_sysenter_stack
948
949	TRACE_IRQS_OFF
950	call	do_debug
951	jmp	ret_from_exception
952
953.Ldebug_from_sysenter_stack:
954	/* We're on the SYSENTER stack.  Switch off. */
955	movl	%esp, %ebx
956	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
957	TRACE_IRQS_OFF
958	call	do_debug
959	movl	%ebx, %esp
960	jmp	ret_from_exception
961END(debug)
962
963/*
964 * NMI is doubly nasty.  It can happen on the first instruction of
965 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
966 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
967 * switched stacks.  We handle both conditions by simply checking whether we
968 * interrupted kernel code running on the SYSENTER stack.
969 */
970ENTRY(nmi)
971	ASM_CLAC
972#ifdef CONFIG_X86_ESPFIX32
973	pushl	%eax
974	movl	%ss, %eax
975	cmpw	$__ESPFIX_SS, %ax
976	popl	%eax
977	je	.Lnmi_espfix_stack
978#endif
979
980	pushl	%eax				# pt_regs->orig_ax
981	SAVE_ALL
982	ENCODE_FRAME_POINTER
983	xorl	%edx, %edx			# zero error code
984	movl	%esp, %eax			# pt_regs pointer
985
986	/* Are we currently on the SYSENTER stack? */
987	PER_CPU(cpu_tss + CPU_TSS_SYSENTER_stack + SIZEOF_SYSENTER_stack, %ecx)
988	subl	%eax, %ecx	/* ecx = (end of SYSENTER_stack) - esp */
989	cmpl	$SIZEOF_SYSENTER_stack, %ecx
990	jb	.Lnmi_from_sysenter_stack
991
992	/* Not on SYSENTER stack. */
993	call	do_nmi
994	jmp	.Lrestore_all_notrace
995
996.Lnmi_from_sysenter_stack:
997	/*
998	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
999	 * is using the thread stack right now, so it's safe for us to use it.
1000	 */
1001	movl	%esp, %ebx
1002	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1003	call	do_nmi
1004	movl	%ebx, %esp
1005	jmp	.Lrestore_all_notrace
1006
1007#ifdef CONFIG_X86_ESPFIX32
1008.Lnmi_espfix_stack:
1009	/*
1010	 * create the pointer to lss back
1011	 */
1012	pushl	%ss
1013	pushl	%esp
1014	addl	$4, (%esp)
1015	/* copy the iret frame of 12 bytes */
1016	.rept 3
1017	pushl	16(%esp)
1018	.endr
1019	pushl	%eax
1020	SAVE_ALL
1021	ENCODE_FRAME_POINTER
1022	FIXUP_ESPFIX_STACK			# %eax == %esp
1023	xorl	%edx, %edx			# zero error code
1024	call	do_nmi
1025	RESTORE_REGS
1026	lss	12+4(%esp), %esp		# back to espfix stack
1027	jmp	.Lirq_return
1028#endif
1029END(nmi)
1030
1031ENTRY(int3)
1032	ASM_CLAC
1033	pushl	$-1				# mark this as an int
1034	SAVE_ALL
1035	ENCODE_FRAME_POINTER
1036	TRACE_IRQS_OFF
1037	xorl	%edx, %edx			# zero error code
1038	movl	%esp, %eax			# pt_regs pointer
1039	call	do_int3
1040	jmp	ret_from_exception
1041END(int3)
1042
1043ENTRY(general_protection)
1044	pushl	$do_general_protection
1045	jmp	common_exception
1046END(general_protection)
1047
1048#ifdef CONFIG_KVM_GUEST
1049ENTRY(async_page_fault)
1050	ASM_CLAC
1051	pushl	$do_async_page_fault
1052	jmp	common_exception
1053END(async_page_fault)
1054#endif
1055
1056ENTRY(rewind_stack_do_exit)
1057	/* Prevent any naive code from trying to unwind to our caller. */
1058	xorl	%ebp, %ebp
1059
1060	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1061	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1062
1063	call	do_exit
10641:	jmp 1b
1065END(rewind_stack_do_exit)
1066